drivers/net/macvlan.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (c) 2007 Patrick McHardy <kaber@trash.net>
   4  *
   5  * The code this is based on carried the following copyright notice:
   6  * ---
   7  * (C) Copyright 2001-2006
   8  * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com
   9  * Re-worked by Ben Greear <greearb@candelatech.com>
  10  * ---
  11  */
  12 #include <linux/kernel.h>
  13 #include <linux/types.h>
  14 #include <linux/module.h>
  15 #include <linux/init.h>
  16 #include <linux/errno.h>
  17 #include <linux/slab.h>
  18 #include <linux/string.h>
  19 #include <linux/rculist.h>
  20 #include <linux/notifier.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/net_tstamp.h>
  24 #include <linux/ethtool.h>
  25 #include <linux/if_arp.h>
  26 #include <linux/if_vlan.h>
  27 #include <linux/if_link.h>
  28 #include <linux/if_macvlan.h>
  29 #include <linux/hash.h>
  30 #include <linux/workqueue.h>
  31 #include <net/rtnetlink.h>
  32 #include <net/xfrm.h>
  33 #include <linux/netpoll.h>
  34 #include <linux/phy.h>
  35
  36 #define MACVLAN_HASH_BITS       8
  37 #define MACVLAN_HASH_SIZE       (1<<MACVLAN_HASH_BITS)
  38 #define MACVLAN_BC_QUEUE_LEN    1000
  39
  40 #define MACVLAN_F_PASSTHRU      1
  41 #define MACVLAN_F_ADDRCHANGE    2
  42
  43 struct macvlan_port {
  44         struct net_device       *dev;
  45         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  46         struct list_head        vlans;
  47         struct sk_buff_head     bc_queue;
  48         struct work_struct      bc_work;
  49         u32                     flags;
  50         int                     count;
  51         struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
  52         DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
  53         unsigned char           perm_addr[ETH_ALEN];
  54 };
  55
  56 struct macvlan_source_entry {
  57         struct hlist_node       hlist;
  58         struct macvlan_dev      *vlan;
  59         unsigned char           addr[6+2] __aligned(sizeof(u16));
  60         struct rcu_head         rcu;
  61 };
  62
  63 struct macvlan_skb_cb {
  64         const struct macvlan_dev *src;
  65 };
  66
  67 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  68
  69 static void macvlan_port_destroy(struct net_device *dev);
  70
  71 static inline bool macvlan_passthru(const struct macvlan_port *port)
  72 {
  73         return port->flags & MACVLAN_F_PASSTHRU;
  74 }
  75
  76 static inline void macvlan_set_passthru(struct macvlan_port *port)
  77 {
  78         port->flags |= MACVLAN_F_PASSTHRU;
  79 }
  80
  81 static inline bool macvlan_addr_change(const struct macvlan_port *port)
  82 {
  83         return port->flags & MACVLAN_F_ADDRCHANGE;
  84 }
  85
  86 static inline void macvlan_set_addr_change(struct macvlan_port *port)
  87 {
  88         port->flags |= MACVLAN_F_ADDRCHANGE;
  89 }
  90
  91 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
  92 {
  93         port->flags &= ~MACVLAN_F_ADDRCHANGE;
  94 }
  95
  96 /* Hash Ethernet address */
  97 static u32 macvlan_eth_hash(const unsigned char *addr)
  98 {
  99         u64 value = get_unaligned((u64 *)addr);
 100
 101         /* only want 6 bytes */
 102 #ifdef __BIG_ENDIAN
 103         value >>= 16;
 104 #else
 105         value <<= 16;
 106 #endif
 107         return hash_64(value, MACVLAN_HASH_BITS);
 108 }
 109
 110 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
 111 {
 112         return rcu_dereference(dev->rx_handler_data);
 113 }
 114
 115 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
 116 {
 117         return rtnl_dereference(dev->rx_handler_data);
 118 }
 119
 120 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port,
 121                                                const unsigned char *addr)
 122 {
 123         struct macvlan_dev *vlan;
 124         u32 idx = macvlan_eth_hash(addr);
 125
 126         hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist,
 127                                  lockdep_rtnl_is_held()) {
 128                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 129                         return vlan;
 130         }
 131         return NULL;
 132 }
 133
 134 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 135         const struct macvlan_dev *vlan,
 136         const unsigned char *addr)
 137 {
 138         struct macvlan_source_entry *entry;
 139         u32 idx = macvlan_eth_hash(addr);
 140         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 141
 142         hlist_for_each_entry_rcu(entry, h, hlist) {
 143                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 144                     entry->vlan == vlan)
 145                         return entry;
 146         }
 147         return NULL;
 148 }
 149
 150 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 151                                    const unsigned char *addr)
 152 {
 153         struct macvlan_port *port = vlan->port;
 154         struct macvlan_source_entry *entry;
 155         struct hlist_head *h;
 156
 157         entry = macvlan_hash_lookup_source(vlan, addr);
 158         if (entry)
 159                 return 0;
 160
 161         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 162         if (!entry)
 163                 return -ENOMEM;
 164
 165         ether_addr_copy(entry->addr, addr);
 166         entry->vlan = vlan;
 167         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 168         hlist_add_head_rcu(&entry->hlist, h);
 169         vlan->macaddr_count++;
 170
 171         return 0;
 172 }
 173
 174 static void macvlan_hash_add(struct macvlan_dev *vlan)
 175 {
 176         struct macvlan_port *port = vlan->port;
 177         const unsigned char *addr = vlan->dev->dev_addr;
 178         u32 idx = macvlan_eth_hash(addr);
 179
 180         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 181 }
 182
 183 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 184 {
 185         hlist_del_rcu(&entry->hlist);
 186         kfree_rcu(entry, rcu);
 187 }
 188
 189 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 190 {
 191         hlist_del_rcu(&vlan->hlist);
 192         if (sync)
 193                 synchronize_rcu();
 194 }
 195
 196 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 197                                         const unsigned char *addr)
 198 {
 199         macvlan_hash_del(vlan, true);
 200         /* Now that we are unhashed it is safe to change the device
 201          * address without confusing packet delivery.
 202          */
 203         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 204         macvlan_hash_add(vlan);
 205 }
 206
 207 static bool macvlan_addr_busy(const struct macvlan_port *port,
 208                               const unsigned char *addr)
 209 {
 210         /* Test to see if the specified address is
 211          * currently in use by the underlying device or
 212          * another macvlan.
 213          */
 214         if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
 215             ether_addr_equal_64bits(port->dev->dev_addr, addr))
 216                 return true;
 217
 218         if (macvlan_hash_lookup(port, addr))
 219                 return true;
 220
 221         return false;
 222 }
 223
 224
 225 static int macvlan_broadcast_one(struct sk_buff *skb,
 226                                  const struct macvlan_dev *vlan,
 227                                  const struct ethhdr *eth, bool local)
 228 {
 229         struct net_device *dev = vlan->dev;
 230
 231         if (local)
 232                 return __dev_forward_skb(dev, skb);
 233
 234         skb->dev = dev;
 235         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 236                 skb->pkt_type = PACKET_BROADCAST;
 237         else
 238                 skb->pkt_type = PACKET_MULTICAST;
 239
 240         return 0;
 241 }
 242
 243 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 244 {
 245         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 246 }
 247
 248
 249 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 250                             const unsigned char *addr)
 251 {
 252         u32 val = __get_unaligned_cpu32(addr + 2);
 253
 254         val ^= macvlan_hash_mix(vlan);
 255         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 256 }
 257
 258 static void macvlan_broadcast(struct sk_buff *skb,
 259                               const struct macvlan_port *port,
 260                               struct net_device *src,
 261                               enum macvlan_mode mode)
 262 {
 263         const struct ethhdr *eth = eth_hdr(skb);
 264         const struct macvlan_dev *vlan;
 265         struct sk_buff *nskb;
 266         unsigned int i;
 267         int err;
 268         unsigned int hash;
 269
 270         if (skb->protocol == htons(ETH_P_PAUSE))
 271                 return;
 272
 273         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 274                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 275                         if (vlan->dev == src || !(vlan->mode & mode))
 276                                 continue;
 277
 278                         hash = mc_hash(vlan, eth->h_dest);
 279                         if (!test_bit(hash, vlan->mc_filter))
 280                                 continue;
 281
 282                         err = NET_RX_DROP;
 283                         nskb = skb_clone(skb, GFP_ATOMIC);
 284                         if (likely(nskb))
 285                                 err = macvlan_broadcast_one(
 286                                         nskb, vlan, eth,
 287                                         mode == MACVLAN_MODE_BRIDGE) ?:
 288                                       netif_rx_ni(nskb);
 289                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 290                                          err == NET_RX_SUCCESS, true);
 291                 }
 292         }
 293 }
 294
 295 static void macvlan_process_broadcast(struct work_struct *w)
 296 {
 297         struct macvlan_port *port = container_of(w, struct macvlan_port,
 298                                                  bc_work);
 299         struct sk_buff *skb;
 300         struct sk_buff_head list;
 301
 302         __skb_queue_head_init(&list);
 303
 304         spin_lock_bh(&port->bc_queue.lock);
 305         skb_queue_splice_tail_init(&port->bc_queue, &list);
 306         spin_unlock_bh(&port->bc_queue.lock);
 307
 308         while ((skb = __skb_dequeue(&list))) {
 309                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 310
 311                 rcu_read_lock();
 312
 313                 if (!src)
 314                         /* frame comes from an external address */
 315                         macvlan_broadcast(skb, port, NULL,
 316                                           MACVLAN_MODE_PRIVATE |
 317                                           MACVLAN_MODE_VEPA    |
 318                                           MACVLAN_MODE_PASSTHRU|
 319                                           MACVLAN_MODE_BRIDGE);
 320                 else if (src->mode == MACVLAN_MODE_VEPA)
 321                         /* flood to everyone except source */
 322                         macvlan_broadcast(skb, port, src->dev,
 323                                           MACVLAN_MODE_VEPA |
 324                                           MACVLAN_MODE_BRIDGE);
 325                 else
 326                         /*
 327                          * flood only to VEPA ports, bridge ports
 328                          * already saw the frame on the way out.
 329                          */
 330                         macvlan_broadcast(skb, port, src->dev,
 331                                           MACVLAN_MODE_VEPA);
 332
 333                 rcu_read_unlock();
 334
 335                 if (src)
 336                         dev_put(src->dev);
 337                 consume_skb(skb);
 338
 339                 cond_resched();
 340         }
 341 }
 342
 343 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 344                                       const struct macvlan_dev *src,
 345                                       struct sk_buff *skb)
 346 {
 347         struct sk_buff *nskb;
 348         int err = -ENOMEM;
 349
 350         nskb = skb_clone(skb, GFP_ATOMIC);
 351         if (!nskb)
 352                 goto err;
 353
 354         MACVLAN_SKB_CB(nskb)->src = src;
 355
 356         spin_lock(&port->bc_queue.lock);
 357         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 358                 if (src)
 359                         dev_hold(src->dev);
 360                 __skb_queue_tail(&port->bc_queue, nskb);
 361                 err = 0;
 362         }
 363         spin_unlock(&port->bc_queue.lock);
 364
 365         schedule_work(&port->bc_work);
 366
 367         if (err)
 368                 goto free_nskb;
 369
 370         return;
 371
 372 free_nskb:
 373         kfree_skb(nskb);
 374 err:
 375         atomic_long_inc(&skb->dev->rx_dropped);
 376 }
 377
 378 static void macvlan_flush_sources(struct macvlan_port *port,
 379                                   struct macvlan_dev *vlan)
 380 {
 381         int i;
 382
 383         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 384                 struct hlist_node *h, *n;
 385
 386                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 387                         struct macvlan_source_entry *entry;
 388
 389                         entry = hlist_entry(h, struct macvlan_source_entry,
 390                                             hlist);
 391                         if (entry->vlan == vlan)
 392                                 macvlan_hash_del_source(entry);
 393                 }
 394         }
 395         vlan->macaddr_count = 0;
 396 }
 397
 398 static void macvlan_forward_source_one(struct sk_buff *skb,
 399                                        struct macvlan_dev *vlan)
 400 {
 401         struct sk_buff *nskb;
 402         struct net_device *dev;
 403         int len;
 404         int ret;
 405
 406         dev = vlan->dev;
 407         if (unlikely(!(dev->flags & IFF_UP)))
 408                 return;
 409
 410         nskb = skb_clone(skb, GFP_ATOMIC);
 411         if (!nskb)
 412                 return;
 413
 414         len = nskb->len + ETH_HLEN;
 415         nskb->dev = dev;
 416
 417         if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
 418                 nskb->pkt_type = PACKET_HOST;
 419
 420         ret = netif_rx(nskb);
 421         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 422 }
 423
 424 static void macvlan_forward_source(struct sk_buff *skb,
 425                                    struct macvlan_port *port,
 426                                    const unsigned char *addr)
 427 {
 428         struct macvlan_source_entry *entry;
 429         u32 idx = macvlan_eth_hash(addr);
 430         struct hlist_head *h = &port->vlan_source_hash[idx];
 431
 432         hlist_for_each_entry_rcu(entry, h, hlist) {
 433                 if (ether_addr_equal_64bits(entry->addr, addr))
 434                         macvlan_forward_source_one(skb, entry->vlan);
 435         }
 436 }
 437
 438 /* called under rcu_read_lock() from netif_receive_skb */
 439 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 440 {
 441         struct macvlan_port *port;
 442         struct sk_buff *skb = *pskb;
 443         const struct ethhdr *eth = eth_hdr(skb);
 444         const struct macvlan_dev *vlan;
 445         const struct macvlan_dev *src;
 446         struct net_device *dev;
 447         unsigned int len = 0;
 448         int ret;
 449         rx_handler_result_t handle_res;
 450
 451         /* Packets from dev_loopback_xmit() do not have L2 header, bail out */
 452         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 453                 return RX_HANDLER_PASS;
 454
 455         port = macvlan_port_get_rcu(skb->dev);
 456         if (is_multicast_ether_addr(eth->h_dest)) {
 457                 unsigned int hash;
 458
 459                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 460                 if (!skb)
 461                         return RX_HANDLER_CONSUMED;
 462                 *pskb = skb;
 463                 eth = eth_hdr(skb);
 464                 macvlan_forward_source(skb, port, eth->h_source);
 465                 src = macvlan_hash_lookup(port, eth->h_source);
 466                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 467                     src->mode != MACVLAN_MODE_BRIDGE) {
 468                         /* forward to original port. */
 469                         vlan = src;
 470                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 471                               netif_rx(skb);
 472                         handle_res = RX_HANDLER_CONSUMED;
 473                         goto out;
 474                 }
 475
 476                 hash = mc_hash(NULL, eth->h_dest);
 477                 if (test_bit(hash, port->mc_filter))
 478                         macvlan_broadcast_enqueue(port, src, skb);
 479
 480                 return RX_HANDLER_PASS;
 481         }
 482
 483         macvlan_forward_source(skb, port, eth->h_source);
 484         if (macvlan_passthru(port))
 485                 vlan = list_first_or_null_rcu(&port->vlans,
 486                                               struct macvlan_dev, list);
 487         else
 488                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 489         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 490                 return RX_HANDLER_PASS;
 491
 492         dev = vlan->dev;
 493         if (unlikely(!(dev->flags & IFF_UP))) {
 494                 kfree_skb(skb);
 495                 return RX_HANDLER_CONSUMED;
 496         }
 497         len = skb->len + ETH_HLEN;
 498         skb = skb_share_check(skb, GFP_ATOMIC);
 499         if (!skb) {
 500                 ret = NET_RX_DROP;
 501                 handle_res = RX_HANDLER_CONSUMED;
 502                 goto out;
 503         }
 504
 505         *pskb = skb;
 506         skb->dev = dev;
 507         skb->pkt_type = PACKET_HOST;
 508
 509         ret = NET_RX_SUCCESS;
 510         handle_res = RX_HANDLER_ANOTHER;
 511 out:
 512         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 513         return handle_res;
 514 }
 515
 516 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 517 {
 518         const struct macvlan_dev *vlan = netdev_priv(dev);
 519         const struct macvlan_port *port = vlan->port;
 520         const struct macvlan_dev *dest;
 521
 522         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 523                 const struct ethhdr *eth = skb_eth_hdr(skb);
 524
 525                 /* send to other bridge ports directly */
 526                 if (is_multicast_ether_addr(eth->h_dest)) {
 527                         skb_reset_mac_header(skb);
 528                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 529                         goto xmit_world;
 530                 }
 531
 532                 dest = macvlan_hash_lookup(port, eth->h_dest);
 533                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 534                         /* send to lowerdev first for its network taps */
 535                         dev_forward_skb(vlan->lowerdev, skb);
 536
 537                         return NET_XMIT_SUCCESS;
 538                 }
 539         }
 540 xmit_world:
 541         skb->dev = vlan->lowerdev;
 542         return dev_queue_xmit_accel(skb,
 543                                     netdev_get_sb_channel(dev) ? dev : NULL);
 544 }
 545
 546 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 547 {
 548 #ifdef CONFIG_NET_POLL_CONTROLLER
 549         return netpoll_send_skb(vlan->netpoll, skb);
 550 #else
 551         BUG();
 552         return NETDEV_TX_OK;
 553 #endif
 554 }
 555
 556 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 557                                       struct net_device *dev)
 558 {
 559         struct macvlan_dev *vlan = netdev_priv(dev);
 560         unsigned int len = skb->len;
 561         int ret;
 562
 563         if (unlikely(netpoll_tx_running(dev)))
 564                 return macvlan_netpoll_send_skb(vlan, skb);
 565
 566         ret = macvlan_queue_xmit(skb, dev);
 567
 568         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 569                 struct vlan_pcpu_stats *pcpu_stats;
 570
 571                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 572                 u64_stats_update_begin(&pcpu_stats->syncp);
 573                 pcpu_stats->tx_packets++;
 574                 pcpu_stats->tx_bytes += len;
 575                 u64_stats_update_end(&pcpu_stats->syncp);
 576         } else {
 577                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 578         }
 579         return ret;
 580 }
 581
 582 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 583                                unsigned short type, const void *daddr,
 584                                const void *saddr, unsigned len)
 585 {
 586         const struct macvlan_dev *vlan = netdev_priv(dev);
 587         struct net_device *lowerdev = vlan->lowerdev;
 588
 589         return dev_hard_header(skb, lowerdev, type, daddr,
 590                                saddr ? : dev->dev_addr, len);
 591 }
 592
 593 static const struct header_ops macvlan_hard_header_ops = {
 594         .create         = macvlan_hard_header,
 595         .parse          = eth_header_parse,
 596         .cache          = eth_header_cache,
 597         .cache_update   = eth_header_cache_update,
 598 };
 599
 600 static int macvlan_open(struct net_device *dev)
 601 {
 602         struct macvlan_dev *vlan = netdev_priv(dev);
 603         struct net_device *lowerdev = vlan->lowerdev;
 604         int err;
 605
 606         if (macvlan_passthru(vlan->port)) {
 607                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 608                         err = dev_set_promiscuity(lowerdev, 1);
 609                         if (err < 0)
 610                                 goto out;
 611                 }
 612                 goto hash_add;
 613         }
 614
 615         err = -EADDRINUSE;
 616         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 617                 goto out;
 618
 619         /* Attempt to populate accel_priv which is used to offload the L2
 620          * forwarding requests for unicast packets.
 621          */
 622         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
 623                 vlan->accel_priv =
 624                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 625
 626         /* If earlier attempt to offload failed, or accel_priv is not
 627          * populated we must add the unicast address to the lower device.
 628          */
 629         if (IS_ERR_OR_NULL(vlan->accel_priv)) {
 630                 vlan->accel_priv = NULL;
 631                 err = dev_uc_add(lowerdev, dev->dev_addr);
 632                 if (err < 0)
 633                         goto out;
 634         }
 635
 636         if (dev->flags & IFF_ALLMULTI) {
 637                 err = dev_set_allmulti(lowerdev, 1);
 638                 if (err < 0)
 639                         goto del_unicast;
 640         }
 641
 642         if (dev->flags & IFF_PROMISC) {
 643                 err = dev_set_promiscuity(lowerdev, 1);
 644                 if (err < 0)
 645                         goto clear_multi;
 646         }
 647
 648 hash_add:
 649         macvlan_hash_add(vlan);
 650         return 0;
 651
 652 clear_multi:
 653         if (dev->flags & IFF_ALLMULTI)
 654                 dev_set_allmulti(lowerdev, -1);
 655 del_unicast:
 656         if (vlan->accel_priv) {
 657                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 658                                                            vlan->accel_priv);
 659                 vlan->accel_priv = NULL;
 660         } else {
 661                 dev_uc_del(lowerdev, dev->dev_addr);
 662         }
 663 out:
 664         return err;
 665 }
 666
 667 static int macvlan_stop(struct net_device *dev)
 668 {
 669         struct macvlan_dev *vlan = netdev_priv(dev);
 670         struct net_device *lowerdev = vlan->lowerdev;
 671
 672         if (vlan->accel_priv) {
 673                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 674                                                            vlan->accel_priv);
 675                 vlan->accel_priv = NULL;
 676         }
 677
 678         dev_uc_unsync(lowerdev, dev);
 679         dev_mc_unsync(lowerdev, dev);
 680
 681         if (macvlan_passthru(vlan->port)) {
 682                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 683                         dev_set_promiscuity(lowerdev, -1);
 684                 goto hash_del;
 685         }
 686
 687         if (dev->flags & IFF_ALLMULTI)
 688                 dev_set_allmulti(lowerdev, -1);
 689
 690         if (dev->flags & IFF_PROMISC)
 691                 dev_set_promiscuity(lowerdev, -1);
 692
 693         dev_uc_del(lowerdev, dev->dev_addr);
 694
 695 hash_del:
 696         macvlan_hash_del(vlan, !dev->dismantle);
 697         return 0;
 698 }
 699
 700 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 701 {
 702         struct macvlan_dev *vlan = netdev_priv(dev);
 703         struct net_device *lowerdev = vlan->lowerdev;
 704         struct macvlan_port *port = vlan->port;
 705         int err;
 706
 707         if (!(dev->flags & IFF_UP)) {
 708                 /* Just copy in the new address */
 709                 ether_addr_copy(dev->dev_addr, addr);
 710         } else {
 711                 /* Rehash and update the device filters */
 712                 if (macvlan_addr_busy(vlan->port, addr))
 713                         return -EADDRINUSE;
 714
 715                 if (!macvlan_passthru(port)) {
 716                         err = dev_uc_add(lowerdev, addr);
 717                         if (err)
 718                                 return err;
 719
 720                         dev_uc_del(lowerdev, dev->dev_addr);
 721                 }
 722
 723                 macvlan_hash_change_addr(vlan, addr);
 724         }
 725         if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
 726                 /* Since addr_change isn't set, we are here due to lower
 727                  * device change.  Save the lower-dev address so we can
 728                  * restore it later.
 729                  */
 730                 ether_addr_copy(vlan->port->perm_addr,
 731                                 lowerdev->dev_addr);
 732         }
 733         macvlan_clear_addr_change(port);
 734         return 0;
 735 }
 736
 737 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 738 {
 739         struct macvlan_dev *vlan = netdev_priv(dev);
 740         struct sockaddr *addr = p;
 741
 742         if (!is_valid_ether_addr(addr->sa_data))
 743                 return -EADDRNOTAVAIL;
 744
 745         /* If the addresses are the same, this is a no-op */
 746         if (ether_addr_equal(dev->dev_addr, addr->sa_data))
 747                 return 0;
 748
 749         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 750                 macvlan_set_addr_change(vlan->port);
 751                 return dev_set_mac_address(vlan->lowerdev, addr, NULL);
 752         }
 753
 754         if (macvlan_addr_busy(vlan->port, addr->sa_data))
 755                 return -EADDRINUSE;
 756
 757         return macvlan_sync_address(dev, addr->sa_data);
 758 }
 759
 760 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 761 {
 762         struct macvlan_dev *vlan = netdev_priv(dev);
 763         struct net_device *lowerdev = vlan->lowerdev;
 764
 765         if (dev->flags & IFF_UP) {
 766                 if (change & IFF_ALLMULTI)
 767                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 768                 if (change & IFF_PROMISC)
 769                         dev_set_promiscuity(lowerdev,
 770                                             dev->flags & IFF_PROMISC ? 1 : -1);
 771
 772         }
 773 }
 774
 775 static void macvlan_compute_filter(unsigned long *mc_filter,
 776                                    struct net_device *dev,
 777                                    struct macvlan_dev *vlan)
 778 {
 779         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 780                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 781         } else {
 782                 struct netdev_hw_addr *ha;
 783                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 784
 785                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 786                 netdev_for_each_mc_addr(ha, dev) {
 787                         __set_bit(mc_hash(vlan, ha->addr), filter);
 788                 }
 789
 790                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 791
 792                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 793         }
 794 }
 795
 796 static void macvlan_set_mac_lists(struct net_device *dev)
 797 {
 798         struct macvlan_dev *vlan = netdev_priv(dev);
 799
 800         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 801
 802         dev_uc_sync(vlan->lowerdev, dev);
 803         dev_mc_sync(vlan->lowerdev, dev);
 804
 805         /* This is slightly inaccurate as we're including the subscription
 806          * list of vlan->lowerdev too.
 807          *
 808          * Bug alert: This only works if everyone has the same broadcast
 809          * address as lowerdev.  As soon as someone changes theirs this
 810          * will break.
 811          *
 812          * However, this is already broken as when you change your broadcast
 813          * address we don't get called.
 814          *
 815          * The solution is to maintain a list of broadcast addresses like
 816          * we do for uc/mc, if you care.
 817          */
 818         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 819 }
 820
 821 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 822 {
 823         struct macvlan_dev *vlan = netdev_priv(dev);
 824
 825         if (vlan->lowerdev->mtu < new_mtu)
 826                 return -EINVAL;
 827         dev->mtu = new_mtu;
 828         return 0;
 829 }
 830
 831 static int macvlan_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 832 {
 833         struct net_device *real_dev = macvlan_dev_real_dev(dev);
 834         const struct net_device_ops *ops = real_dev->netdev_ops;
 835         struct ifreq ifrr;
 836         int err = -EOPNOTSUPP;
 837
 838         strscpy(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
 839         ifrr.ifr_ifru = ifr->ifr_ifru;
 840
 841         switch (cmd) {
 842         case SIOCSHWTSTAMP:
 843                 if (!net_eq(dev_net(dev), &init_net))
 844                         break;
 845                 /* fall through */
 846         case SIOCGHWTSTAMP:
 847                 if (netif_device_present(real_dev) && ops->ndo_do_ioctl)
 848                         err = ops->ndo_do_ioctl(real_dev, &ifrr, cmd);
 849                 break;
 850         }
 851
 852         if (!err)
 853                 ifr->ifr_ifru = ifrr.ifr_ifru;
 854
 855         return err;
 856 }
 857
 858 /*
 859  * macvlan network devices have devices nesting below it and are a special
 860  * "super class" of normal network devices; split their locks off into a
 861  * separate class since they always nest.
 862  */
 863 static struct lock_class_key macvlan_netdev_addr_lock_key;
 864
 865 #define ALWAYS_ON_OFFLOADS \
 866         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
 867          NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
 868
 869 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
 870
 871 #define MACVLAN_FEATURES \
 872         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 873          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
 874          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 875          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 876
 877 #define MACVLAN_STATE_MASK \
 878         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 879
 880 static void macvlan_set_lockdep_class(struct net_device *dev)
 881 {
 882         netdev_lockdep_set_classes(dev);
 883         lockdep_set_class(&dev->addr_list_lock,
 884                           &macvlan_netdev_addr_lock_key);
 885 }
 886
 887 static int macvlan_init(struct net_device *dev)
 888 {
 889         struct macvlan_dev *vlan = netdev_priv(dev);
 890         const struct net_device *lowerdev = vlan->lowerdev;
 891         struct macvlan_port *port = vlan->port;
 892
 893         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 894                                   (lowerdev->state & MACVLAN_STATE_MASK);
 895         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 896         dev->features           |= ALWAYS_ON_FEATURES;
 897         dev->hw_features        |= NETIF_F_LRO;
 898         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 899         dev->vlan_features      |= ALWAYS_ON_OFFLOADS;
 900         dev->hw_enc_features    |= dev->features;
 901         dev->gso_max_size       = lowerdev->gso_max_size;
 902         dev->gso_max_segs       = lowerdev->gso_max_segs;
 903         dev->hard_header_len    = lowerdev->hard_header_len;
 904         macvlan_set_lockdep_class(dev);
 905
 906         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 907         if (!vlan->pcpu_stats)
 908                 return -ENOMEM;
 909
 910         port->count += 1;
 911
 912         return 0;
 913 }
 914
 915 static void macvlan_uninit(struct net_device *dev)
 916 {
 917         struct macvlan_dev *vlan = netdev_priv(dev);
 918         struct macvlan_port *port = vlan->port;
 919
 920         free_percpu(vlan->pcpu_stats);
 921
 922         macvlan_flush_sources(port, vlan);
 923         port->count -= 1;
 924         if (!port->count)
 925                 macvlan_port_destroy(port->dev);
 926 }
 927
 928 static void macvlan_dev_get_stats64(struct net_device *dev,
 929                                     struct rtnl_link_stats64 *stats)
 930 {
 931         struct macvlan_dev *vlan = netdev_priv(dev);
 932
 933         if (vlan->pcpu_stats) {
 934                 struct vlan_pcpu_stats *p;
 935                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 936                 u32 rx_errors = 0, tx_dropped = 0;
 937                 unsigned int start;
 938                 int i;
 939
 940                 for_each_possible_cpu(i) {
 941                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 942                         do {
 943                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 944                                 rx_packets      = p->rx_packets;
 945                                 rx_bytes        = p->rx_bytes;
 946                                 rx_multicast    = p->rx_multicast;
 947                                 tx_packets      = p->tx_packets;
 948                                 tx_bytes        = p->tx_bytes;
 949                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 950
 951                         stats->rx_packets       += rx_packets;
 952                         stats->rx_bytes         += rx_bytes;
 953                         stats->multicast        += rx_multicast;
 954                         stats->tx_packets       += tx_packets;
 955                         stats->tx_bytes         += tx_bytes;
 956                         /* rx_errors & tx_dropped are u32, updated
 957                          * without syncp protection.
 958                          */
 959                         rx_errors       += p->rx_errors;
 960                         tx_dropped      += p->tx_dropped;
 961                 }
 962                 stats->rx_errors        = rx_errors;
 963                 stats->rx_dropped       = rx_errors;
 964                 stats->tx_dropped       = tx_dropped;
 965         }
 966 }
 967
 968 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 969                                    __be16 proto, u16 vid)
 970 {
 971         struct macvlan_dev *vlan = netdev_priv(dev);
 972         struct net_device *lowerdev = vlan->lowerdev;
 973
 974         return vlan_vid_add(lowerdev, proto, vid);
 975 }
 976
 977 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 978                                     __be16 proto, u16 vid)
 979 {
 980         struct macvlan_dev *vlan = netdev_priv(dev);
 981         struct net_device *lowerdev = vlan->lowerdev;
 982
 983         vlan_vid_del(lowerdev, proto, vid);
 984         return 0;
 985 }
 986
 987 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 988                            struct net_device *dev,
 989                            const unsigned char *addr, u16 vid,
 990                            u16 flags,
 991                            struct netlink_ext_ack *extack)
 992 {
 993         struct macvlan_dev *vlan = netdev_priv(dev);
 994         int err = -EINVAL;
 995
 996         /* Support unicast filter only on passthru devices.
 997          * Multicast filter should be allowed on all devices.
 998          */
 999         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1000                 return -EOPNOTSUPP;
1001
1002         if (flags & NLM_F_REPLACE)
1003                 return -EOPNOTSUPP;
1004
1005         if (is_unicast_ether_addr(addr))
1006                 err = dev_uc_add_excl(dev, addr);
1007         else if (is_multicast_ether_addr(addr))
1008                 err = dev_mc_add_excl(dev, addr);
1009
1010         return err;
1011 }
1012
1013 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1014                            struct net_device *dev,
1015                            const unsigned char *addr, u16 vid)
1016 {
1017         struct macvlan_dev *vlan = netdev_priv(dev);
1018         int err = -EINVAL;
1019
1020         /* Support unicast filter only on passthru devices.
1021          * Multicast filter should be allowed on all devices.
1022          */
1023         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1024                 return -EOPNOTSUPP;
1025
1026         if (is_unicast_ether_addr(addr))
1027                 err = dev_uc_del(dev, addr);
1028         else if (is_multicast_ether_addr(addr))
1029                 err = dev_mc_del(dev, addr);
1030
1031         return err;
1032 }
1033
1034 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1035                                         struct ethtool_drvinfo *drvinfo)
1036 {
1037         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1038         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1039 }
1040
1041 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1042                                               struct ethtool_link_ksettings *cmd)
1043 {
1044         const struct macvlan_dev *vlan = netdev_priv(dev);
1045
1046         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1047 }
1048
1049 static int macvlan_ethtool_get_ts_info(struct net_device *dev,
1050                                        struct ethtool_ts_info *info)
1051 {
1052         struct net_device *real_dev = macvlan_dev_real_dev(dev);
1053         const struct ethtool_ops *ops = real_dev->ethtool_ops;
1054         struct phy_device *phydev = real_dev->phydev;
1055
1056         if (phy_has_tsinfo(phydev)) {
1057                 return phy_ts_info(phydev, info);
1058         } else if (ops->get_ts_info) {
1059                 return ops->get_ts_info(real_dev, info);
1060         } else {
1061                 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
1062                         SOF_TIMESTAMPING_SOFTWARE;
1063                 info->phc_index = -1;
1064         }
1065
1066         return 0;
1067 }
1068
1069 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1070                                               netdev_features_t features)
1071 {
1072         struct macvlan_dev *vlan = netdev_priv(dev);
1073         netdev_features_t lowerdev_features = vlan->lowerdev->features;
1074         netdev_features_t mask;
1075
1076         features |= NETIF_F_ALL_FOR_ALL;
1077         features &= (vlan->set_features | ~MACVLAN_FEATURES);
1078         mask = features;
1079
1080         lowerdev_features &= (features | ~NETIF_F_LRO);
1081         features = netdev_increment_features(lowerdev_features, features, mask);
1082         features |= ALWAYS_ON_FEATURES;
1083         features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1084
1085         return features;
1086 }
1087
1088 #ifdef CONFIG_NET_POLL_CONTROLLER
1089 static void macvlan_dev_poll_controller(struct net_device *dev)
1090 {
1091         return;
1092 }
1093
1094 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1095 {
1096         struct macvlan_dev *vlan = netdev_priv(dev);
1097         struct net_device *real_dev = vlan->lowerdev;
1098         struct netpoll *netpoll;
1099         int err = 0;
1100
1101         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1102         err = -ENOMEM;
1103         if (!netpoll)
1104                 goto out;
1105
1106         err = __netpoll_setup(netpoll, real_dev);
1107         if (err) {
1108                 kfree(netpoll);
1109                 goto out;
1110         }
1111
1112         vlan->netpoll = netpoll;
1113
1114 out:
1115         return err;
1116 }
1117
1118 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1119 {
1120         struct macvlan_dev *vlan = netdev_priv(dev);
1121         struct netpoll *netpoll = vlan->netpoll;
1122
1123         if (!netpoll)
1124                 return;
1125
1126         vlan->netpoll = NULL;
1127
1128         __netpoll_free(netpoll);
1129 }
1130 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1131
1132 static int macvlan_dev_get_iflink(const struct net_device *dev)
1133 {
1134         struct macvlan_dev *vlan = netdev_priv(dev);
1135
1136         return vlan->lowerdev->ifindex;
1137 }
1138
1139 static const struct ethtool_ops macvlan_ethtool_ops = {
1140         .get_link               = ethtool_op_get_link,
1141         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1142         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1143         .get_ts_info            = macvlan_ethtool_get_ts_info,
1144 };
1145
1146 static const struct net_device_ops macvlan_netdev_ops = {
1147         .ndo_init               = macvlan_init,
1148         .ndo_uninit             = macvlan_uninit,
1149         .ndo_open               = macvlan_open,
1150         .ndo_stop               = macvlan_stop,
1151         .ndo_start_xmit         = macvlan_start_xmit,
1152         .ndo_change_mtu         = macvlan_change_mtu,
1153         .ndo_do_ioctl           = macvlan_do_ioctl,
1154         .ndo_fix_features       = macvlan_fix_features,
1155         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1156         .ndo_set_mac_address    = macvlan_set_mac_address,
1157         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1158         .ndo_get_stats64        = macvlan_dev_get_stats64,
1159         .ndo_validate_addr      = eth_validate_addr,
1160         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1161         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1162         .ndo_fdb_add            = macvlan_fdb_add,
1163         .ndo_fdb_del            = macvlan_fdb_del,
1164         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1165 #ifdef CONFIG_NET_POLL_CONTROLLER
1166         .ndo_poll_controller    = macvlan_dev_poll_controller,
1167         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1168         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1169 #endif
1170         .ndo_get_iflink         = macvlan_dev_get_iflink,
1171         .ndo_features_check     = passthru_features_check,
1172         .ndo_change_proto_down  = dev_change_proto_down_generic,
1173 };
1174
1175 void macvlan_common_setup(struct net_device *dev)
1176 {
1177         ether_setup(dev);
1178
1179         dev->min_mtu            = 0;
1180         dev->max_mtu            = ETH_MAX_MTU;
1181         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1182         netif_keep_dst(dev);
1183         dev->priv_flags        |= IFF_UNICAST_FLT;
1184         dev->netdev_ops         = &macvlan_netdev_ops;
1185         dev->needs_free_netdev  = true;
1186         dev->header_ops         = &macvlan_hard_header_ops;
1187         dev->ethtool_ops        = &macvlan_ethtool_ops;
1188 }
1189 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1190
1191 static void macvlan_setup(struct net_device *dev)
1192 {
1193         macvlan_common_setup(dev);
1194         dev->priv_flags |= IFF_NO_QUEUE;
1195 }
1196
1197 static int macvlan_port_create(struct net_device *dev)
1198 {
1199         struct macvlan_port *port;
1200         unsigned int i;
1201         int err;
1202
1203         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1204                 return -EINVAL;
1205
1206         if (netdev_is_rx_handler_busy(dev))
1207                 return -EBUSY;
1208
1209         port = kzalloc(sizeof(*port), GFP_KERNEL);
1210         if (port == NULL)
1211                 return -ENOMEM;
1212
1213         port->dev = dev;
1214         ether_addr_copy(port->perm_addr, dev->dev_addr);
1215         INIT_LIST_HEAD(&port->vlans);
1216         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1217                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1218         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1219                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1220
1221         skb_queue_head_init(&port->bc_queue);
1222         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1223
1224         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1225         if (err)
1226                 kfree(port);
1227         else
1228                 dev->priv_flags |= IFF_MACVLAN_PORT;
1229         return err;
1230 }
1231
1232 static void macvlan_port_destroy(struct net_device *dev)
1233 {
1234         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1235         struct sk_buff *skb;
1236
1237         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1238         netdev_rx_handler_unregister(dev);
1239
1240         /* After this point, no packet can schedule bc_work anymore,
1241          * but we need to cancel it and purge left skbs if any.
1242          */
1243         cancel_work_sync(&port->bc_work);
1244
1245         while ((skb = __skb_dequeue(&port->bc_queue))) {
1246                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1247
1248                 if (src)
1249                         dev_put(src->dev);
1250
1251                 kfree_skb(skb);
1252         }
1253
1254         /* If the lower device address has been changed by passthru
1255          * macvlan, put it back.
1256          */
1257         if (macvlan_passthru(port) &&
1258             !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1259                 struct sockaddr sa;
1260
1261                 sa.sa_family = port->dev->type;
1262                 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1263                 dev_set_mac_address(port->dev, &sa, NULL);
1264         }
1265
1266         kfree(port);
1267 }
1268
1269 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1270                             struct netlink_ext_ack *extack)
1271 {
1272         if (tb[IFLA_ADDRESS]) {
1273                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1274                         return -EINVAL;
1275                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1276                         return -EADDRNOTAVAIL;
1277         }
1278
1279         if (!data)
1280                 return 0;
1281
1282         if (data[IFLA_MACVLAN_FLAGS] &&
1283             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1284                 return -EINVAL;
1285
1286         if (data[IFLA_MACVLAN_MODE]) {
1287                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1288                 case MACVLAN_MODE_PRIVATE:
1289                 case MACVLAN_MODE_VEPA:
1290                 case MACVLAN_MODE_BRIDGE:
1291                 case MACVLAN_MODE_PASSTHRU:
1292                 case MACVLAN_MODE_SOURCE:
1293                         break;
1294                 default:
1295                         return -EINVAL;
1296                 }
1297         }
1298
1299         if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1300                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1301                 case MACVLAN_MACADDR_ADD:
1302                 case MACVLAN_MACADDR_DEL:
1303                 case MACVLAN_MACADDR_FLUSH:
1304                 case MACVLAN_MACADDR_SET:
1305                         break;
1306                 default:
1307                         return -EINVAL;
1308                 }
1309         }
1310
1311         if (data[IFLA_MACVLAN_MACADDR]) {
1312                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1313                         return -EINVAL;
1314
1315                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1316                         return -EADDRNOTAVAIL;
1317         }
1318
1319         if (data[IFLA_MACVLAN_MACADDR_COUNT])
1320                 return -EINVAL;
1321
1322         return 0;
1323 }
1324
1325 /**
1326  * reconfigure list of remote source mac address
1327  * (only for macvlan devices in source mode)
1328  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1329  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1330  */
1331 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1332                                       struct nlattr *data[])
1333 {
1334         char *addr = NULL;
1335         int ret, rem, len;
1336         struct nlattr *nla, *head;
1337         struct macvlan_source_entry *entry;
1338
1339         if (data[IFLA_MACVLAN_MACADDR])
1340                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1341
1342         if (mode == MACVLAN_MACADDR_ADD) {
1343                 if (!addr)
1344                         return -EINVAL;
1345
1346                 return macvlan_hash_add_source(vlan, addr);
1347
1348         } else if (mode == MACVLAN_MACADDR_DEL) {
1349                 if (!addr)
1350                         return -EINVAL;
1351
1352                 entry = macvlan_hash_lookup_source(vlan, addr);
1353                 if (entry) {
1354                         macvlan_hash_del_source(entry);
1355                         vlan->macaddr_count--;
1356                 }
1357         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1358                 macvlan_flush_sources(vlan->port, vlan);
1359         } else if (mode == MACVLAN_MACADDR_SET) {
1360                 macvlan_flush_sources(vlan->port, vlan);
1361
1362                 if (addr) {
1363                         ret = macvlan_hash_add_source(vlan, addr);
1364                         if (ret)
1365                                 return ret;
1366                 }
1367
1368                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1369                         return 0;
1370
1371                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1372                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1373
1374                 nla_for_each_attr(nla, head, len, rem) {
1375                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1376                             nla_len(nla) != ETH_ALEN)
1377                                 continue;
1378
1379                         addr = nla_data(nla);
1380                         ret = macvlan_hash_add_source(vlan, addr);
1381                         if (ret)
1382                                 return ret;
1383                 }
1384         } else {
1385                 return -EINVAL;
1386         }
1387
1388         return 0;
1389 }
1390
1391 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1392                            struct nlattr *tb[], struct nlattr *data[],
1393                            struct netlink_ext_ack *extack)
1394 {
1395         struct macvlan_dev *vlan = netdev_priv(dev);
1396         struct macvlan_port *port;
1397         struct net_device *lowerdev;
1398         int err;
1399         int macmode;
1400         bool create = false;
1401
1402         if (!tb[IFLA_LINK])
1403                 return -EINVAL;
1404
1405         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1406         if (lowerdev == NULL)
1407                 return -ENODEV;
1408
1409         /* When creating macvlans or macvtaps on top of other macvlans - use
1410          * the real device as the lowerdev.
1411          */
1412         if (netif_is_macvlan(lowerdev))
1413                 lowerdev = macvlan_dev_real_dev(lowerdev);
1414
1415         if (!tb[IFLA_MTU])
1416                 dev->mtu = lowerdev->mtu;
1417         else if (dev->mtu > lowerdev->mtu)
1418                 return -EINVAL;
1419
1420         /* MTU range: 68 - lowerdev->max_mtu */
1421         dev->min_mtu = ETH_MIN_MTU;
1422         dev->max_mtu = lowerdev->max_mtu;
1423
1424         if (!tb[IFLA_ADDRESS])
1425                 eth_hw_addr_random(dev);
1426
1427         if (!netif_is_macvlan_port(lowerdev)) {
1428                 err = macvlan_port_create(lowerdev);
1429                 if (err < 0)
1430                         return err;
1431                 create = true;
1432         }
1433         port = macvlan_port_get_rtnl(lowerdev);
1434
1435         /* Only 1 macvlan device can be created in passthru mode */
1436         if (macvlan_passthru(port)) {
1437                 /* The macvlan port must be not created this time,
1438                  * still goto destroy_macvlan_port for readability.
1439                  */
1440                 err = -EINVAL;
1441                 goto destroy_macvlan_port;
1442         }
1443
1444         vlan->lowerdev = lowerdev;
1445         vlan->dev      = dev;
1446         vlan->port     = port;
1447         vlan->set_features = MACVLAN_FEATURES;
1448
1449         vlan->mode     = MACVLAN_MODE_VEPA;
1450         if (data && data[IFLA_MACVLAN_MODE])
1451                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1452
1453         if (data && data[IFLA_MACVLAN_FLAGS])
1454                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1455
1456         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1457                 if (port->count) {
1458                         err = -EINVAL;
1459                         goto destroy_macvlan_port;
1460                 }
1461                 macvlan_set_passthru(port);
1462                 eth_hw_addr_inherit(dev, lowerdev);
1463         }
1464
1465         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1466                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1467                         err = -EINVAL;
1468                         goto destroy_macvlan_port;
1469                 }
1470                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1471                 err = macvlan_changelink_sources(vlan, macmode, data);
1472                 if (err)
1473                         goto destroy_macvlan_port;
1474         }
1475
1476         err = register_netdevice(dev);
1477         if (err < 0)
1478                 goto destroy_macvlan_port;
1479
1480         dev->priv_flags |= IFF_MACVLAN;
1481         err = netdev_upper_dev_link(lowerdev, dev, extack);
1482         if (err)
1483                 goto unregister_netdev;
1484
1485         list_add_tail_rcu(&vlan->list, &port->vlans);
1486         netif_stacked_transfer_operstate(lowerdev, dev);
1487         linkwatch_fire_event(dev);
1488
1489         return 0;
1490
1491 unregister_netdev:
1492         /* macvlan_uninit would free the macvlan port */
1493         unregister_netdevice(dev);
1494         return err;
1495 destroy_macvlan_port:
1496         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1497          * so we destroy the macvlan port only when it's valid.
1498          */
1499         if (create && macvlan_port_get_rtnl(lowerdev))
1500                 macvlan_port_destroy(port->dev);
1501         return err;
1502 }
1503 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1504
1505 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1506                            struct nlattr *tb[], struct nlattr *data[],
1507                            struct netlink_ext_ack *extack)
1508 {
1509         return macvlan_common_newlink(src_net, dev, tb, data, extack);
1510 }
1511
1512 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1513 {
1514         struct macvlan_dev *vlan = netdev_priv(dev);
1515
1516         if (vlan->mode == MACVLAN_MODE_SOURCE)
1517                 macvlan_flush_sources(vlan->port, vlan);
1518         list_del_rcu(&vlan->list);
1519         unregister_netdevice_queue(dev, head);
1520         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1521 }
1522 EXPORT_SYMBOL_GPL(macvlan_dellink);
1523
1524 static int macvlan_changelink(struct net_device *dev,
1525                               struct nlattr *tb[], struct nlattr *data[],
1526                               struct netlink_ext_ack *extack)
1527 {
1528         struct macvlan_dev *vlan = netdev_priv(dev);
1529         enum macvlan_mode mode;
1530         bool set_mode = false;
1531         enum macvlan_macaddr_mode macmode;
1532         int ret;
1533
1534         /* Validate mode, but don't set yet: setting flags may fail. */
1535         if (data && data[IFLA_MACVLAN_MODE]) {
1536                 set_mode = true;
1537                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1538                 /* Passthrough mode can't be set or cleared dynamically */
1539                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1540                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1541                         return -EINVAL;
1542                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1543                     vlan->mode != mode)
1544                         macvlan_flush_sources(vlan->port, vlan);
1545         }
1546
1547         if (data && data[IFLA_MACVLAN_FLAGS]) {
1548                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1549                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1550                 if (macvlan_passthru(vlan->port) && promisc) {
1551                         int err;
1552
1553                         if (flags & MACVLAN_FLAG_NOPROMISC)
1554                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1555                         else
1556                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1557                         if (err < 0)
1558                                 return err;
1559                 }
1560                 vlan->flags = flags;
1561         }
1562         if (set_mode)
1563                 vlan->mode = mode;
1564         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1565                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1566                         return -EINVAL;
1567                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1568                 ret = macvlan_changelink_sources(vlan, macmode, data);
1569                 if (ret)
1570                         return ret;
1571         }
1572         return 0;
1573 }
1574
1575 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1576 {
1577         if (vlan->macaddr_count == 0)
1578                 return 0;
1579         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1580                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1581 }
1582
1583 static size_t macvlan_get_size(const struct net_device *dev)
1584 {
1585         struct macvlan_dev *vlan = netdev_priv(dev);
1586
1587         return (0
1588                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1589                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1590                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1591                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1592                 );
1593 }
1594
1595 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1596                                      const struct macvlan_dev *vlan,
1597                                      const int i)
1598 {
1599         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1600         struct macvlan_source_entry *entry;
1601
1602         hlist_for_each_entry_rcu(entry, h, hlist) {
1603                 if (entry->vlan != vlan)
1604                         continue;
1605                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1606                         return 1;
1607         }
1608         return 0;
1609 }
1610
1611 static int macvlan_fill_info(struct sk_buff *skb,
1612                                 const struct net_device *dev)
1613 {
1614         struct macvlan_dev *vlan = netdev_priv(dev);
1615         int i;
1616         struct nlattr *nest;
1617
1618         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1619                 goto nla_put_failure;
1620         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1621                 goto nla_put_failure;
1622         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1623                 goto nla_put_failure;
1624         if (vlan->macaddr_count > 0) {
1625                 nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA);
1626                 if (nest == NULL)
1627                         goto nla_put_failure;
1628
1629                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1630                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1631                                 goto nla_put_failure;
1632                 }
1633                 nla_nest_end(skb, nest);
1634         }
1635         return 0;
1636
1637 nla_put_failure:
1638         return -EMSGSIZE;
1639 }
1640
1641 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1642         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1643         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1644         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1645         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1646         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1647         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1648 };
1649
1650 int macvlan_link_register(struct rtnl_link_ops *ops)
1651 {
1652         /* common fields */
1653         ops->validate           = macvlan_validate;
1654         ops->maxtype            = IFLA_MACVLAN_MAX;
1655         ops->policy             = macvlan_policy;
1656         ops->changelink         = macvlan_changelink;
1657         ops->get_size           = macvlan_get_size;
1658         ops->fill_info          = macvlan_fill_info;
1659
1660         return rtnl_link_register(ops);
1661 };
1662 EXPORT_SYMBOL_GPL(macvlan_link_register);
1663
1664 static struct net *macvlan_get_link_net(const struct net_device *dev)
1665 {
1666         return dev_net(macvlan_dev_real_dev(dev));
1667 }
1668
1669 static struct rtnl_link_ops macvlan_link_ops = {
1670         .kind           = "macvlan",
1671         .setup          = macvlan_setup,
1672         .newlink        = macvlan_newlink,
1673         .dellink        = macvlan_dellink,
1674         .get_link_net   = macvlan_get_link_net,
1675         .priv_size      = sizeof(struct macvlan_dev),
1676 };
1677
1678 static int macvlan_device_event(struct notifier_block *unused,
1679                                 unsigned long event, void *ptr)
1680 {
1681         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1682         struct macvlan_dev *vlan, *next;
1683         struct macvlan_port *port;
1684         LIST_HEAD(list_kill);
1685
1686         if (!netif_is_macvlan_port(dev))
1687                 return NOTIFY_DONE;
1688
1689         port = macvlan_port_get_rtnl(dev);
1690
1691         switch (event) {
1692         case NETDEV_UP:
1693         case NETDEV_DOWN:
1694         case NETDEV_CHANGE:
1695                 list_for_each_entry(vlan, &port->vlans, list)
1696                         netif_stacked_transfer_operstate(vlan->lowerdev,
1697                                                          vlan->dev);
1698                 break;
1699         case NETDEV_FEAT_CHANGE:
1700                 list_for_each_entry(vlan, &port->vlans, list) {
1701                         vlan->dev->gso_max_size = dev->gso_max_size;
1702                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1703                         netdev_update_features(vlan->dev);
1704                 }
1705                 break;
1706         case NETDEV_CHANGEMTU:
1707                 list_for_each_entry(vlan, &port->vlans, list) {
1708                         if (vlan->dev->mtu <= dev->mtu)
1709                                 continue;
1710                         dev_set_mtu(vlan->dev, dev->mtu);
1711                 }
1712                 break;
1713         case NETDEV_CHANGEADDR:
1714                 if (!macvlan_passthru(port))
1715                         return NOTIFY_DONE;
1716
1717                 vlan = list_first_entry_or_null(&port->vlans,
1718                                                 struct macvlan_dev,
1719                                                 list);
1720
1721                 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1722                         return NOTIFY_BAD;
1723
1724                 break;
1725         case NETDEV_UNREGISTER:
1726                 /* twiddle thumbs on netns device moves */
1727                 if (dev->reg_state != NETREG_UNREGISTERING)
1728                         break;
1729
1730                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1731                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1732                 unregister_netdevice_many(&list_kill);
1733                 break;
1734         case NETDEV_PRE_TYPE_CHANGE:
1735                 /* Forbid underlaying device to change its type. */
1736                 return NOTIFY_BAD;
1737
1738         case NETDEV_NOTIFY_PEERS:
1739         case NETDEV_BONDING_FAILOVER:
1740         case NETDEV_RESEND_IGMP:
1741                 /* Propagate to all vlans */
1742                 list_for_each_entry(vlan, &port->vlans, list)
1743                         call_netdevice_notifiers(event, vlan->dev);
1744         }
1745         return NOTIFY_DONE;
1746 }
1747
1748 static struct notifier_block macvlan_notifier_block __read_mostly = {
1749         .notifier_call  = macvlan_device_event,
1750 };
1751
1752 static int __init macvlan_init_module(void)
1753 {
1754         int err;
1755
1756         register_netdevice_notifier(&macvlan_notifier_block);
1757
1758         err = macvlan_link_register(&macvlan_link_ops);
1759         if (err < 0)
1760                 goto err1;
1761         return 0;
1762 err1:
1763         unregister_netdevice_notifier(&macvlan_notifier_block);
1764         return err;
1765 }
1766
1767 static void __exit macvlan_cleanup_module(void)
1768 {
1769         rtnl_link_unregister(&macvlan_link_ops);
1770         unregister_netdevice_notifier(&macvlan_notifier_block);
1771 }
1772
1773 module_init(macvlan_init_module);
1774 module_exit(macvlan_cleanup_module);
1775
1776 MODULE_LICENSE("GPL");
1777 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1778 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1779 MODULE_ALIAS_RTNL_LINK("macvlan");