drivers/net/ethernet/sfc/tc_encap_actions.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /****************************************************************************
   3  * Driver for Solarflare network controllers and boards
   4  * Copyright 2023, Advanced Micro Devices, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 as published
   8  * by the Free Software Foundation, incorporated herein by reference.
   9  */
  10
  11 #include "tc_encap_actions.h"
  12 #include "tc.h"
  13 #include "mae.h"
  14 #include <net/vxlan.h>
  15 #include <net/geneve.h>
  16 #include <net/netevent.h>
  17 #include <net/arp.h>
  18
  19 static const struct rhashtable_params efx_neigh_ht_params = {
  20         .key_len        = offsetof(struct efx_neigh_binder, ha),
  21         .key_offset     = 0,
  22         .head_offset    = offsetof(struct efx_neigh_binder, linkage),
  23 };
  24
  25 static const struct rhashtable_params efx_tc_encap_ht_params = {
  26         .key_len        = offsetofend(struct efx_tc_encap_action, key),
  27         .key_offset     = 0,
  28         .head_offset    = offsetof(struct efx_tc_encap_action, linkage),
  29 };
  30
  31 static void efx_tc_encap_free(void *ptr, void *__unused)
  32 {
  33         struct efx_tc_encap_action *enc = ptr;
  34
  35         WARN_ON(refcount_read(&enc->ref));
  36         kfree(enc);
  37 }
  38
  39 static void efx_neigh_free(void *ptr, void *__unused)
  40 {
  41         struct efx_neigh_binder *neigh = ptr;
  42
  43         WARN_ON(refcount_read(&neigh->ref));
  44         WARN_ON(!list_empty(&neigh->users));
  45         put_net_track(neigh->net, &neigh->ns_tracker);
  46         netdev_put(neigh->egdev, &neigh->dev_tracker);
  47         kfree(neigh);
  48 }
  49
  50 int efx_tc_init_encap_actions(struct efx_nic *efx)
  51 {
  52         int rc;
  53
  54         rc = rhashtable_init(&efx->tc->neigh_ht, &efx_neigh_ht_params);
  55         if (rc < 0)
  56                 goto fail_neigh_ht;
  57         rc = rhashtable_init(&efx->tc->encap_ht, &efx_tc_encap_ht_params);
  58         if (rc < 0)
  59                 goto fail_encap_ht;
  60         return 0;
  61 fail_encap_ht:
  62         rhashtable_destroy(&efx->tc->neigh_ht);
  63 fail_neigh_ht:
  64         return rc;
  65 }
  66
  67 /* Only call this in init failure teardown.
  68  * Normal exit should fini instead as there may be entries in the table.
  69  */
  70 void efx_tc_destroy_encap_actions(struct efx_nic *efx)
  71 {
  72         rhashtable_destroy(&efx->tc->encap_ht);
  73         rhashtable_destroy(&efx->tc->neigh_ht);
  74 }
  75
  76 void efx_tc_fini_encap_actions(struct efx_nic *efx)
  77 {
  78         rhashtable_free_and_destroy(&efx->tc->encap_ht, efx_tc_encap_free, NULL);
  79         rhashtable_free_and_destroy(&efx->tc->neigh_ht, efx_neigh_free, NULL);
  80 }
  81
  82 static void efx_neigh_update(struct work_struct *work);
  83
  84 static int efx_bind_neigh(struct efx_nic *efx,
  85                           struct efx_tc_encap_action *encap, struct net *net,
  86                           struct netlink_ext_ack *extack)
  87 {
  88         struct efx_neigh_binder *neigh, *old;
  89         struct flowi6 flow6 = {};
  90         struct flowi4 flow4 = {};
  91         int rc;
  92
  93         /* GCC stupidly thinks that only values explicitly listed in the enum
  94          * definition can _possibly_ be sensible case values, so without this
  95          * cast it complains about the IPv6 versions.
  96          */
  97         switch ((int)encap->type) {
  98         case EFX_ENCAP_TYPE_VXLAN:
  99         case EFX_ENCAP_TYPE_GENEVE:
 100                 flow4.flowi4_proto = IPPROTO_UDP;
 101                 flow4.fl4_dport = encap->key.tp_dst;
 102                 flow4.flowi4_tos = encap->key.tos;
 103                 flow4.daddr = encap->key.u.ipv4.dst;
 104                 flow4.saddr = encap->key.u.ipv4.src;
 105                 break;
 106         case EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6:
 107         case EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6:
 108                 flow6.flowi6_proto = IPPROTO_UDP;
 109                 flow6.fl6_dport = encap->key.tp_dst;
 110                 flow6.flowlabel = ip6_make_flowinfo(encap->key.tos,
 111                                                     encap->key.label);
 112                 flow6.daddr = encap->key.u.ipv6.dst;
 113                 flow6.saddr = encap->key.u.ipv6.src;
 114                 break;
 115         default:
 116                 NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported encap type %d",
 117                                        (int)encap->type);
 118                 return -EOPNOTSUPP;
 119         }
 120
 121         neigh = kzalloc(sizeof(*neigh), GFP_KERNEL_ACCOUNT);
 122         if (!neigh)
 123                 return -ENOMEM;
 124         neigh->net = get_net_track(net, &neigh->ns_tracker, GFP_KERNEL_ACCOUNT);
 125         neigh->dst_ip = flow4.daddr;
 126         neigh->dst_ip6 = flow6.daddr;
 127
 128         old = rhashtable_lookup_get_insert_fast(&efx->tc->neigh_ht,
 129                                                 &neigh->linkage,
 130                                                 efx_neigh_ht_params);
 131         if (old) {
 132                 /* don't need our new entry */
 133                 put_net_track(neigh->net, &neigh->ns_tracker);
 134                 kfree(neigh);
 135                 if (!refcount_inc_not_zero(&old->ref))
 136                         return -EAGAIN;
 137                 /* existing entry found, ref taken */
 138                 neigh = old;
 139         } else {
 140                 /* New entry.  We need to initiate a lookup */
 141                 struct neighbour *n;
 142                 struct rtable *rt;
 143
 144                 if (encap->type & EFX_ENCAP_FLAG_IPV6) {
 145 #if IS_ENABLED(CONFIG_IPV6)
 146                         struct dst_entry *dst;
 147
 148                         dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &flow6,
 149                                                               NULL);
 150                         rc = PTR_ERR_OR_ZERO(dst);
 151                         if (rc) {
 152                                 NL_SET_ERR_MSG_MOD(extack, "Failed to lookup route for IPv6 encap");
 153                                 goto out_free;
 154                         }
 155                         neigh->egdev = dst->dev;
 156                         netdev_hold(neigh->egdev, &neigh->dev_tracker,
 157                                     GFP_KERNEL_ACCOUNT);
 158                         neigh->ttl = ip6_dst_hoplimit(dst);
 159                         n = dst_neigh_lookup(dst, &flow6.daddr);
 160                         dst_release(dst);
 161 #else
 162                         /* We shouldn't ever get here, because if IPv6 isn't
 163                          * enabled how did someone create an IPv6 tunnel_key?
 164                          */
 165                         rc = -EOPNOTSUPP;
 166                         NL_SET_ERR_MSG_MOD(extack, "No IPv6 support (neigh bind)");
 167                         goto out_free;
 168 #endif
 169                 } else {
 170                         rt = ip_route_output_key(net, &flow4);
 171                         if (IS_ERR_OR_NULL(rt)) {
 172                                 rc = PTR_ERR_OR_ZERO(rt);
 173                                 if (!rc)
 174                                         rc = -EIO;
 175                                 NL_SET_ERR_MSG_MOD(extack, "Failed to lookup route for encap");
 176                                 goto out_free;
 177                         }
 178                         neigh->egdev = rt->dst.dev;
 179                         netdev_hold(neigh->egdev, &neigh->dev_tracker,
 180                                     GFP_KERNEL_ACCOUNT);
 181                         neigh->ttl = ip4_dst_hoplimit(&rt->dst);
 182                         n = dst_neigh_lookup(&rt->dst, &flow4.daddr);
 183                         ip_rt_put(rt);
 184                 }
 185                 if (!n) {
 186                         rc = -ENETUNREACH;
 187                         NL_SET_ERR_MSG_MOD(extack, "Failed to lookup neighbour for encap");
 188                         netdev_put(neigh->egdev, &neigh->dev_tracker);
 189                         goto out_free;
 190                 }
 191                 refcount_set(&neigh->ref, 1);
 192                 INIT_LIST_HEAD(&neigh->users);
 193                 read_lock_bh(&n->lock);
 194                 ether_addr_copy(neigh->ha, n->ha);
 195                 neigh->n_valid = n->nud_state & NUD_VALID;
 196                 read_unlock_bh(&n->lock);
 197                 rwlock_init(&neigh->lock);
 198                 INIT_WORK(&neigh->work, efx_neigh_update);
 199                 neigh->efx = efx;
 200                 neigh->used = jiffies;
 201                 if (!neigh->n_valid)
 202                         /* Prod ARP to find us a neighbour */
 203                         neigh_event_send(n, NULL);
 204                 neigh_release(n);
 205         }
 206         /* Add us to this neigh */
 207         encap->neigh = neigh;
 208         list_add_tail(&encap->list, &neigh->users);
 209         return 0;
 210
 211 out_free:
 212         /* cleanup common to several error paths */
 213         rhashtable_remove_fast(&efx->tc->neigh_ht, &neigh->linkage,
 214                                efx_neigh_ht_params);
 215         synchronize_rcu();
 216         put_net_track(net, &neigh->ns_tracker);
 217         kfree(neigh);
 218         return rc;
 219 }
 220
 221 static void efx_free_neigh(struct efx_neigh_binder *neigh)
 222 {
 223         struct efx_nic *efx = neigh->efx;
 224
 225         rhashtable_remove_fast(&efx->tc->neigh_ht, &neigh->linkage,
 226                                efx_neigh_ht_params);
 227         synchronize_rcu();
 228         netdev_put(neigh->egdev, &neigh->dev_tracker);
 229         put_net_track(neigh->net, &neigh->ns_tracker);
 230         kfree(neigh);
 231 }
 232
 233 static void efx_release_neigh(struct efx_nic *efx,
 234                               struct efx_tc_encap_action *encap)
 235 {
 236         struct efx_neigh_binder *neigh = encap->neigh;
 237
 238         if (!neigh)
 239                 return;
 240         list_del(&encap->list);
 241         encap->neigh = NULL;
 242         if (!refcount_dec_and_test(&neigh->ref))
 243                 return; /* still in use */
 244         efx_free_neigh(neigh);
 245 }
 246
 247 static void efx_gen_tun_header_eth(struct efx_tc_encap_action *encap, u16 proto)
 248 {
 249         struct efx_neigh_binder *neigh = encap->neigh;
 250         struct ethhdr *eth;
 251
 252         encap->encap_hdr_len = sizeof(*eth);
 253         eth = (struct ethhdr *)encap->encap_hdr;
 254
 255         if (encap->neigh->n_valid)
 256                 ether_addr_copy(eth->h_dest, neigh->ha);
 257         else
 258                 eth_zero_addr(eth->h_dest);
 259         ether_addr_copy(eth->h_source, neigh->egdev->dev_addr);
 260         eth->h_proto = htons(proto);
 261 }
 262
 263 static void efx_gen_tun_header_ipv4(struct efx_tc_encap_action *encap, u8 ipproto, u8 len)
 264 {
 265         struct efx_neigh_binder *neigh = encap->neigh;
 266         struct ip_tunnel_key *key = &encap->key;
 267         struct iphdr *ip;
 268
 269         ip = (struct iphdr *)(encap->encap_hdr + encap->encap_hdr_len);
 270         encap->encap_hdr_len += sizeof(*ip);
 271
 272         ip->daddr = key->u.ipv4.dst;
 273         ip->saddr = key->u.ipv4.src;
 274         ip->ttl = neigh->ttl;
 275         ip->protocol = ipproto;
 276         ip->version = 0x4;
 277         ip->ihl = 0x5;
 278         ip->tot_len = cpu_to_be16(ip->ihl * 4 + len);
 279         ip_send_check(ip);
 280 }
 281
 282 #ifdef CONFIG_IPV6
 283 static void efx_gen_tun_header_ipv6(struct efx_tc_encap_action *encap, u8 ipproto, u8 len)
 284 {
 285         struct efx_neigh_binder *neigh = encap->neigh;
 286         struct ip_tunnel_key *key = &encap->key;
 287         struct ipv6hdr *ip;
 288
 289         ip = (struct ipv6hdr *)(encap->encap_hdr + encap->encap_hdr_len);
 290         encap->encap_hdr_len += sizeof(*ip);
 291
 292         ip6_flow_hdr(ip, key->tos, key->label);
 293         ip->daddr = key->u.ipv6.dst;
 294         ip->saddr = key->u.ipv6.src;
 295         ip->hop_limit = neigh->ttl;
 296         ip->nexthdr = ipproto;
 297         ip->version = 0x6;
 298         ip->payload_len = cpu_to_be16(len);
 299 }
 300 #endif
 301
 302 static void efx_gen_tun_header_udp(struct efx_tc_encap_action *encap, u8 len)
 303 {
 304         struct ip_tunnel_key *key = &encap->key;
 305         struct udphdr *udp;
 306
 307         udp = (struct udphdr *)(encap->encap_hdr + encap->encap_hdr_len);
 308         encap->encap_hdr_len += sizeof(*udp);
 309
 310         udp->dest = key->tp_dst;
 311         udp->len = cpu_to_be16(sizeof(*udp) + len);
 312 }
 313
 314 static void efx_gen_tun_header_vxlan(struct efx_tc_encap_action *encap)
 315 {
 316         struct ip_tunnel_key *key = &encap->key;
 317         struct vxlanhdr *vxlan;
 318
 319         vxlan = (struct vxlanhdr *)(encap->encap_hdr + encap->encap_hdr_len);
 320         encap->encap_hdr_len += sizeof(*vxlan);
 321
 322         vxlan->vx_flags = VXLAN_HF_VNI;
 323         vxlan->vx_vni = vxlan_vni_field(tunnel_id_to_key32(key->tun_id));
 324 }
 325
 326 static void efx_gen_tun_header_geneve(struct efx_tc_encap_action *encap)
 327 {
 328         struct ip_tunnel_key *key = &encap->key;
 329         struct genevehdr *geneve;
 330         u32 vni;
 331
 332         geneve = (struct genevehdr *)(encap->encap_hdr + encap->encap_hdr_len);
 333         encap->encap_hdr_len += sizeof(*geneve);
 334
 335         geneve->proto_type = htons(ETH_P_TEB);
 336         /* convert tun_id to host-endian so we can use host arithmetic to
 337          * extract individual bytes.
 338          */
 339         vni = ntohl(tunnel_id_to_key32(key->tun_id));
 340         geneve->vni[0] = vni >> 16;
 341         geneve->vni[1] = vni >> 8;
 342         geneve->vni[2] = vni;
 343 }
 344
 345 #define vxlan_header_l4_len     (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
 346 #define vxlan4_header_len       (sizeof(struct ethhdr) + sizeof(struct iphdr) + vxlan_header_l4_len)
 347 static void efx_gen_vxlan_header_ipv4(struct efx_tc_encap_action *encap)
 348 {
 349         BUILD_BUG_ON(sizeof(encap->encap_hdr) < vxlan4_header_len);
 350         efx_gen_tun_header_eth(encap, ETH_P_IP);
 351         efx_gen_tun_header_ipv4(encap, IPPROTO_UDP, vxlan_header_l4_len);
 352         efx_gen_tun_header_udp(encap, sizeof(struct vxlanhdr));
 353         efx_gen_tun_header_vxlan(encap);
 354 }
 355
 356 #define geneve_header_l4_len    (sizeof(struct udphdr) + sizeof(struct genevehdr))
 357 #define geneve4_header_len      (sizeof(struct ethhdr) + sizeof(struct iphdr) + geneve_header_l4_len)
 358 static void efx_gen_geneve_header_ipv4(struct efx_tc_encap_action *encap)
 359 {
 360         BUILD_BUG_ON(sizeof(encap->encap_hdr) < geneve4_header_len);
 361         efx_gen_tun_header_eth(encap, ETH_P_IP);
 362         efx_gen_tun_header_ipv4(encap, IPPROTO_UDP, geneve_header_l4_len);
 363         efx_gen_tun_header_udp(encap, sizeof(struct genevehdr));
 364         efx_gen_tun_header_geneve(encap);
 365 }
 366
 367 #ifdef CONFIG_IPV6
 368 #define vxlan6_header_len       (sizeof(struct ethhdr) + sizeof(struct ipv6hdr) + vxlan_header_l4_len)
 369 static void efx_gen_vxlan_header_ipv6(struct efx_tc_encap_action *encap)
 370 {
 371         BUILD_BUG_ON(sizeof(encap->encap_hdr) < vxlan6_header_len);
 372         efx_gen_tun_header_eth(encap, ETH_P_IPV6);
 373         efx_gen_tun_header_ipv6(encap, IPPROTO_UDP, vxlan_header_l4_len);
 374         efx_gen_tun_header_udp(encap, sizeof(struct vxlanhdr));
 375         efx_gen_tun_header_vxlan(encap);
 376 }
 377
 378 #define geneve6_header_len      (sizeof(struct ethhdr) + sizeof(struct ipv6hdr) + geneve_header_l4_len)
 379 static void efx_gen_geneve_header_ipv6(struct efx_tc_encap_action *encap)
 380 {
 381         BUILD_BUG_ON(sizeof(encap->encap_hdr) < geneve6_header_len);
 382         efx_gen_tun_header_eth(encap, ETH_P_IPV6);
 383         efx_gen_tun_header_ipv6(encap, IPPROTO_UDP, geneve_header_l4_len);
 384         efx_gen_tun_header_udp(encap, sizeof(struct genevehdr));
 385         efx_gen_tun_header_geneve(encap);
 386 }
 387 #endif
 388
 389 static void efx_gen_encap_header(struct efx_nic *efx,
 390                                  struct efx_tc_encap_action *encap)
 391 {
 392         encap->n_valid = encap->neigh->n_valid;
 393
 394         /* GCC stupidly thinks that only values explicitly listed in the enum
 395          * definition can _possibly_ be sensible case values, so without this
 396          * cast it complains about the IPv6 versions.
 397          */
 398         switch ((int)encap->type) {
 399         case EFX_ENCAP_TYPE_VXLAN:
 400                 efx_gen_vxlan_header_ipv4(encap);
 401                 break;
 402         case EFX_ENCAP_TYPE_GENEVE:
 403                 efx_gen_geneve_header_ipv4(encap);
 404                 break;
 405 #ifdef CONFIG_IPV6
 406         case EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6:
 407                 efx_gen_vxlan_header_ipv6(encap);
 408                 break;
 409         case EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6:
 410                 efx_gen_geneve_header_ipv6(encap);
 411                 break;
 412 #endif
 413         default:
 414                 /* unhandled encap type, can't happen */
 415                 if (net_ratelimit())
 416                         netif_err(efx, drv, efx->net_dev,
 417                                   "Bogus encap type %d, can't generate\n",
 418                                   encap->type);
 419
 420                 /* Use fallback action. */
 421                 encap->n_valid = false;
 422                 break;
 423         }
 424 }
 425
 426 static void efx_tc_update_encap(struct efx_nic *efx,
 427                                 struct efx_tc_encap_action *encap)
 428 {
 429         struct efx_tc_action_set_list *acts, *fallback;
 430         struct efx_tc_flow_rule *rule;
 431         struct efx_tc_action_set *act;
 432         int rc;
 433
 434         if (encap->n_valid) {
 435                 /* Make sure no rules are using this encap while we change it */
 436                 list_for_each_entry(act, &encap->users, encap_user) {
 437                         acts = act->user;
 438                         if (WARN_ON(!acts)) /* can't happen */
 439                                 continue;
 440                         rule = container_of(acts, struct efx_tc_flow_rule, acts);
 441                         if (rule->fallback)
 442                                 fallback = rule->fallback;
 443                         else /* fallback fallback: deliver to PF */
 444                                 fallback = &efx->tc->facts.pf;
 445                         rc = efx_mae_update_rule(efx, fallback->fw_id,
 446                                                  rule->fw_id);
 447                         if (rc)
 448                                 netif_err(efx, drv, efx->net_dev,
 449                                           "Failed to update (f) rule %08x rc %d\n",
 450                                           rule->fw_id, rc);
 451                         else
 452                                 netif_dbg(efx, drv, efx->net_dev, "Updated (f) rule %08x\n",
 453                                           rule->fw_id);
 454                 }
 455         }
 456
 457         /* Make sure we don't leak arbitrary bytes on the wire;
 458          * set an all-0s ethernet header.  A successful call to
 459          * efx_gen_encap_header() will overwrite this.
 460          */
 461         memset(encap->encap_hdr, 0, sizeof(encap->encap_hdr));
 462         encap->encap_hdr_len = ETH_HLEN;
 463
 464         if (encap->neigh) {
 465                 read_lock_bh(&encap->neigh->lock);
 466                 efx_gen_encap_header(efx, encap);
 467                 read_unlock_bh(&encap->neigh->lock);
 468         } else {
 469                 encap->n_valid = false;
 470         }
 471
 472         rc = efx_mae_update_encap_md(efx, encap);
 473         if (rc) {
 474                 netif_err(efx, drv, efx->net_dev,
 475                           "Failed to update encap hdr %08x rc %d\n",
 476                           encap->fw_id, rc);
 477                 return;
 478         }
 479         netif_dbg(efx, drv, efx->net_dev, "Updated encap hdr %08x\n",
 480                   encap->fw_id);
 481         if (!encap->n_valid)
 482                 return;
 483         /* Update rule users: use the action if they are now ready */
 484         list_for_each_entry(act, &encap->users, encap_user) {
 485                 acts = act->user;
 486                 if (WARN_ON(!acts)) /* can't happen */
 487                         continue;
 488                 rule = container_of(acts, struct efx_tc_flow_rule, acts);
 489                 if (!efx_tc_check_ready(efx, rule))
 490                         continue;
 491                 rc = efx_mae_update_rule(efx, acts->fw_id, rule->fw_id);
 492                 if (rc)
 493                         netif_err(efx, drv, efx->net_dev,
 494                                   "Failed to update rule %08x rc %d\n",
 495                                   rule->fw_id, rc);
 496                 else
 497                         netif_dbg(efx, drv, efx->net_dev, "Updated rule %08x\n",
 498                                   rule->fw_id);
 499         }
 500 }
 501
 502 static void efx_neigh_update(struct work_struct *work)
 503 {
 504         struct efx_neigh_binder *neigh = container_of(work, struct efx_neigh_binder, work);
 505         struct efx_tc_encap_action *encap;
 506         struct efx_nic *efx = neigh->efx;
 507
 508         mutex_lock(&efx->tc->mutex);
 509         list_for_each_entry(encap, &neigh->users, list)
 510                 efx_tc_update_encap(neigh->efx, encap);
 511         /* release ref taken in efx_neigh_event() */
 512         if (refcount_dec_and_test(&neigh->ref))
 513                 efx_free_neigh(neigh);
 514         mutex_unlock(&efx->tc->mutex);
 515 }
 516
 517 static int efx_neigh_event(struct efx_nic *efx, struct neighbour *n)
 518 {
 519         struct efx_neigh_binder keys = {NULL}, *neigh;
 520         bool n_valid, ipv6 = false;
 521         char ha[ETH_ALEN];
 522         size_t keysize;
 523
 524         if (WARN_ON(!efx->tc))
 525                 return NOTIFY_DONE;
 526
 527         if (n->tbl == &arp_tbl) {
 528                 keysize = sizeof(keys.dst_ip);
 529 #if IS_ENABLED(CONFIG_IPV6)
 530         } else if (n->tbl == ipv6_stub->nd_tbl) {
 531                 ipv6 = true;
 532                 keysize = sizeof(keys.dst_ip6);
 533 #endif
 534         } else {
 535                 return NOTIFY_DONE;
 536         }
 537         if (!n->parms) {
 538                 netif_warn(efx, drv, efx->net_dev, "neigh_event with no parms!\n");
 539                 return NOTIFY_DONE;
 540         }
 541         keys.net = read_pnet(&n->parms->net);
 542         if (n->tbl->key_len != keysize) {
 543                 netif_warn(efx, drv, efx->net_dev, "neigh_event with bad key_len %u\n",
 544                            n->tbl->key_len);
 545                 return NOTIFY_DONE;
 546         }
 547         read_lock_bh(&n->lock); /* Get a consistent view */
 548         memcpy(ha, n->ha, ETH_ALEN);
 549         n_valid = (n->nud_state & NUD_VALID) && !n->dead;
 550         read_unlock_bh(&n->lock);
 551         if (ipv6)
 552                 memcpy(&keys.dst_ip6, n->primary_key, n->tbl->key_len);
 553         else
 554                 memcpy(&keys.dst_ip, n->primary_key, n->tbl->key_len);
 555         rcu_read_lock();
 556         neigh = rhashtable_lookup_fast(&efx->tc->neigh_ht, &keys,
 557                                        efx_neigh_ht_params);
 558         if (!neigh || neigh->dying)
 559                 /* We're not interested in this neighbour */
 560                 goto done;
 561         write_lock_bh(&neigh->lock);
 562         if (n_valid == neigh->n_valid && !memcmp(ha, neigh->ha, ETH_ALEN)) {
 563                 write_unlock_bh(&neigh->lock);
 564                 /* Nothing has changed; no work to do */
 565                 goto done;
 566         }
 567         neigh->n_valid = n_valid;
 568         memcpy(neigh->ha, ha, ETH_ALEN);
 569         write_unlock_bh(&neigh->lock);
 570         if (refcount_inc_not_zero(&neigh->ref)) {
 571                 rcu_read_unlock();
 572                 if (!schedule_work(&neigh->work))
 573                         /* failed to schedule, release the ref we just took */
 574                         if (refcount_dec_and_test(&neigh->ref))
 575                                 efx_free_neigh(neigh);
 576         } else {
 577 done:
 578                 rcu_read_unlock();
 579         }
 580         return NOTIFY_DONE;
 581 }
 582
 583 bool efx_tc_check_ready(struct efx_nic *efx, struct efx_tc_flow_rule *rule)
 584 {
 585         struct efx_tc_action_set *act;
 586
 587         /* Encap actions can only be offloaded if they have valid
 588          * neighbour info for the outer Ethernet header.
 589          */
 590         list_for_each_entry(act, &rule->acts.list, list)
 591                 if (act->encap_md && !act->encap_md->n_valid)
 592                         return false;
 593         return true;
 594 }
 595
 596 struct efx_tc_encap_action *efx_tc_flower_create_encap_md(
 597                         struct efx_nic *efx, const struct ip_tunnel_info *info,
 598                         struct net_device *egdev, struct netlink_ext_ack *extack)
 599 {
 600         enum efx_encap_type type = efx_tc_indr_netdev_type(egdev);
 601         struct efx_tc_encap_action *encap, *old;
 602         struct efx_rep *to_efv;
 603         s64 rc;
 604
 605         if (type == EFX_ENCAP_TYPE_NONE) {
 606                 /* dest is not an encap device */
 607                 NL_SET_ERR_MSG_MOD(extack, "Not a (supported) tunnel device but tunnel_key is set");
 608                 return ERR_PTR(-EOPNOTSUPP);
 609         }
 610         rc = efx_mae_check_encap_type_supported(efx, type);
 611         if (rc < 0) {
 612                 NL_SET_ERR_MSG_MOD(extack, "Firmware reports no support for this tunnel type");
 613                 return ERR_PTR(rc);
 614         }
 615         /* No support yet for Geneve options */
 616         if (info->options_len) {
 617                 NL_SET_ERR_MSG_MOD(extack, "Unsupported tunnel options");
 618                 return ERR_PTR(-EOPNOTSUPP);
 619         }
 620         switch (info->mode) {
 621         case IP_TUNNEL_INFO_TX:
 622                 break;
 623         case IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6:
 624                 type |= EFX_ENCAP_FLAG_IPV6;
 625                 break;
 626         default:
 627                 NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported tunnel mode %u",
 628                                        info->mode);
 629                 return ERR_PTR(-EOPNOTSUPP);
 630         }
 631         encap = kzalloc(sizeof(*encap), GFP_KERNEL_ACCOUNT);
 632         if (!encap)
 633                 return ERR_PTR(-ENOMEM);
 634         encap->type = type;
 635         encap->key = info->key;
 636         INIT_LIST_HEAD(&encap->users);
 637         old = rhashtable_lookup_get_insert_fast(&efx->tc->encap_ht,
 638                                                 &encap->linkage,
 639                                                 efx_tc_encap_ht_params);
 640         if (old) {
 641                 /* don't need our new entry */
 642                 kfree(encap);
 643                 if (!refcount_inc_not_zero(&old->ref))
 644                         return ERR_PTR(-EAGAIN);
 645                 /* existing entry found, ref taken */
 646                 return old;
 647         }
 648
 649         rc = efx_bind_neigh(efx, encap, dev_net(egdev), extack);
 650         if (rc < 0)
 651                 goto out_remove;
 652         to_efv = efx_tc_flower_lookup_efv(efx, encap->neigh->egdev);
 653         if (IS_ERR(to_efv)) {
 654                 /* neigh->egdev isn't ours */
 655                 NL_SET_ERR_MSG_MOD(extack, "Tunnel egress device not on switch");
 656                 rc = PTR_ERR(to_efv);
 657                 goto out_release;
 658         }
 659         rc = efx_tc_flower_external_mport(efx, to_efv);
 660         if (rc < 0) {
 661                 NL_SET_ERR_MSG_MOD(extack, "Failed to identify tunnel egress m-port");
 662                 goto out_release;
 663         }
 664         encap->dest_mport = rc;
 665         read_lock_bh(&encap->neigh->lock);
 666         efx_gen_encap_header(efx, encap);
 667         read_unlock_bh(&encap->neigh->lock);
 668
 669         rc = efx_mae_allocate_encap_md(efx, encap);
 670         if (rc < 0) {
 671                 NL_SET_ERR_MSG_MOD(extack, "Failed to write tunnel header to hw");
 672                 goto out_release;
 673         }
 674
 675         /* ref and return */
 676         refcount_set(&encap->ref, 1);
 677         return encap;
 678 out_release:
 679         efx_release_neigh(efx, encap);
 680 out_remove:
 681         rhashtable_remove_fast(&efx->tc->encap_ht, &encap->linkage,
 682                                efx_tc_encap_ht_params);
 683         kfree(encap);
 684         return ERR_PTR(rc);
 685 }
 686
 687 void efx_tc_flower_release_encap_md(struct efx_nic *efx,
 688                                     struct efx_tc_encap_action *encap)
 689 {
 690         if (!refcount_dec_and_test(&encap->ref))
 691                 return; /* still in use */
 692         efx_release_neigh(efx, encap);
 693         rhashtable_remove_fast(&efx->tc->encap_ht, &encap->linkage,
 694                                efx_tc_encap_ht_params);
 695         efx_mae_free_encap_md(efx, encap);
 696         kfree(encap);
 697 }
 698
 699 static void efx_tc_remove_neigh_users(struct efx_nic *efx, struct efx_neigh_binder *neigh)
 700 {
 701         struct efx_tc_encap_action *encap, *next;
 702
 703         list_for_each_entry_safe(encap, next, &neigh->users, list) {
 704                 /* Should cause neigh usage count to fall to zero, freeing it */
 705                 efx_release_neigh(efx, encap);
 706                 /* The encap has lost its neigh, so it's now unready */
 707                 efx_tc_update_encap(efx, encap);
 708         }
 709 }
 710
 711 void efx_tc_unregister_egdev(struct efx_nic *efx, struct net_device *net_dev)
 712 {
 713         struct efx_neigh_binder *neigh;
 714         struct rhashtable_iter walk;
 715
 716         mutex_lock(&efx->tc->mutex);
 717         rhashtable_walk_enter(&efx->tc->neigh_ht, &walk);
 718         rhashtable_walk_start(&walk);
 719         while ((neigh = rhashtable_walk_next(&walk)) != NULL) {
 720                 if (IS_ERR(neigh))
 721                         continue;
 722                 if (neigh->egdev != net_dev)
 723                         continue;
 724                 neigh->dying = true;
 725                 rhashtable_walk_stop(&walk);
 726                 synchronize_rcu(); /* Make sure any updates see dying flag */
 727                 efx_tc_remove_neigh_users(efx, neigh); /* might sleep */
 728                 rhashtable_walk_start(&walk);
 729         }
 730         rhashtable_walk_stop(&walk);
 731         rhashtable_walk_exit(&walk);
 732         mutex_unlock(&efx->tc->mutex);
 733 }
 734
 735 int efx_tc_netevent_event(struct efx_nic *efx, unsigned long event,
 736                           void *ptr)
 737 {
 738         if (efx->type->is_vf)
 739                 return NOTIFY_DONE;
 740
 741         switch (event) {
 742         case NETEVENT_NEIGH_UPDATE:
 743                 return efx_neigh_event(efx, ptr);
 744         default:
 745                 return NOTIFY_DONE;
 746         }
 747 }