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                 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr))
 128                         return vlan;
 129         }
 130         return NULL;
 131 }
 132
 133 static struct macvlan_source_entry *macvlan_hash_lookup_source(
 134         const struct macvlan_dev *vlan,
 135         const unsigned char *addr)
 136 {
 137         struct macvlan_source_entry *entry;
 138         u32 idx = macvlan_eth_hash(addr);
 139         struct hlist_head *h = &vlan->port->vlan_source_hash[idx];
 140
 141         hlist_for_each_entry_rcu(entry, h, hlist, lockdep_rtnl_is_held()) {
 142                 if (ether_addr_equal_64bits(entry->addr, addr) &&
 143                     entry->vlan == vlan)
 144                         return entry;
 145         }
 146         return NULL;
 147 }
 148
 149 static int macvlan_hash_add_source(struct macvlan_dev *vlan,
 150                                    const unsigned char *addr)
 151 {
 152         struct macvlan_port *port = vlan->port;
 153         struct macvlan_source_entry *entry;
 154         struct hlist_head *h;
 155
 156         entry = macvlan_hash_lookup_source(vlan, addr);
 157         if (entry)
 158                 return 0;
 159
 160         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 161         if (!entry)
 162                 return -ENOMEM;
 163
 164         ether_addr_copy(entry->addr, addr);
 165         entry->vlan = vlan;
 166         h = &port->vlan_source_hash[macvlan_eth_hash(addr)];
 167         hlist_add_head_rcu(&entry->hlist, h);
 168         vlan->macaddr_count++;
 169
 170         return 0;
 171 }
 172
 173 static void macvlan_hash_add(struct macvlan_dev *vlan)
 174 {
 175         struct macvlan_port *port = vlan->port;
 176         const unsigned char *addr = vlan->dev->dev_addr;
 177         u32 idx = macvlan_eth_hash(addr);
 178
 179         hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]);
 180 }
 181
 182 static void macvlan_hash_del_source(struct macvlan_source_entry *entry)
 183 {
 184         hlist_del_rcu(&entry->hlist);
 185         kfree_rcu(entry, rcu);
 186 }
 187
 188 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync)
 189 {
 190         hlist_del_rcu(&vlan->hlist);
 191         if (sync)
 192                 synchronize_rcu();
 193 }
 194
 195 static void macvlan_hash_change_addr(struct macvlan_dev *vlan,
 196                                         const unsigned char *addr)
 197 {
 198         macvlan_hash_del(vlan, true);
 199         /* Now that we are unhashed it is safe to change the device
 200          * address without confusing packet delivery.
 201          */
 202         memcpy(vlan->dev->dev_addr, addr, ETH_ALEN);
 203         macvlan_hash_add(vlan);
 204 }
 205
 206 static bool macvlan_addr_busy(const struct macvlan_port *port,
 207                               const unsigned char *addr)
 208 {
 209         /* Test to see if the specified address is
 210          * currently in use by the underlying device or
 211          * another macvlan.
 212          */
 213         if (!macvlan_passthru(port) && !macvlan_addr_change(port) &&
 214             ether_addr_equal_64bits(port->dev->dev_addr, addr))
 215                 return true;
 216
 217         if (macvlan_hash_lookup(port, addr))
 218                 return true;
 219
 220         return false;
 221 }
 222
 223
 224 static int macvlan_broadcast_one(struct sk_buff *skb,
 225                                  const struct macvlan_dev *vlan,
 226                                  const struct ethhdr *eth, bool local)
 227 {
 228         struct net_device *dev = vlan->dev;
 229
 230         if (local)
 231                 return __dev_forward_skb(dev, skb);
 232
 233         skb->dev = dev;
 234         if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 235                 skb->pkt_type = PACKET_BROADCAST;
 236         else
 237                 skb->pkt_type = PACKET_MULTICAST;
 238
 239         return 0;
 240 }
 241
 242 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan)
 243 {
 244         return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT);
 245 }
 246
 247
 248 static unsigned int mc_hash(const struct macvlan_dev *vlan,
 249                             const unsigned char *addr)
 250 {
 251         u32 val = __get_unaligned_cpu32(addr + 2);
 252
 253         val ^= macvlan_hash_mix(vlan);
 254         return hash_32(val, MACVLAN_MC_FILTER_BITS);
 255 }
 256
 257 static void macvlan_broadcast(struct sk_buff *skb,
 258                               const struct macvlan_port *port,
 259                               struct net_device *src,
 260                               enum macvlan_mode mode)
 261 {
 262         const struct ethhdr *eth = eth_hdr(skb);
 263         const struct macvlan_dev *vlan;
 264         struct sk_buff *nskb;
 265         unsigned int i;
 266         int err;
 267         unsigned int hash;
 268
 269         if (skb->protocol == htons(ETH_P_PAUSE))
 270                 return;
 271
 272         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 273                 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) {
 274                         if (vlan->dev == src || !(vlan->mode & mode))
 275                                 continue;
 276
 277                         hash = mc_hash(vlan, eth->h_dest);
 278                         if (!test_bit(hash, vlan->mc_filter))
 279                                 continue;
 280
 281                         err = NET_RX_DROP;
 282                         nskb = skb_clone(skb, GFP_ATOMIC);
 283                         if (likely(nskb))
 284                                 err = macvlan_broadcast_one(
 285                                         nskb, vlan, eth,
 286                                         mode == MACVLAN_MODE_BRIDGE) ?:
 287                                       netif_rx_ni(nskb);
 288                         macvlan_count_rx(vlan, skb->len + ETH_HLEN,
 289                                          err == NET_RX_SUCCESS, true);
 290                 }
 291         }
 292 }
 293
 294 static void macvlan_process_broadcast(struct work_struct *w)
 295 {
 296         struct macvlan_port *port = container_of(w, struct macvlan_port,
 297                                                  bc_work);
 298         struct sk_buff *skb;
 299         struct sk_buff_head list;
 300
 301         __skb_queue_head_init(&list);
 302
 303         spin_lock_bh(&port->bc_queue.lock);
 304         skb_queue_splice_tail_init(&port->bc_queue, &list);
 305         spin_unlock_bh(&port->bc_queue.lock);
 306
 307         while ((skb = __skb_dequeue(&list))) {
 308                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
 309
 310                 rcu_read_lock();
 311
 312                 if (!src)
 313                         /* frame comes from an external address */
 314                         macvlan_broadcast(skb, port, NULL,
 315                                           MACVLAN_MODE_PRIVATE |
 316                                           MACVLAN_MODE_VEPA    |
 317                                           MACVLAN_MODE_PASSTHRU|
 318                                           MACVLAN_MODE_BRIDGE);
 319                 else if (src->mode == MACVLAN_MODE_VEPA)
 320                         /* flood to everyone except source */
 321                         macvlan_broadcast(skb, port, src->dev,
 322                                           MACVLAN_MODE_VEPA |
 323                                           MACVLAN_MODE_BRIDGE);
 324                 else
 325                         /*
 326                          * flood only to VEPA ports, bridge ports
 327                          * already saw the frame on the way out.
 328                          */
 329                         macvlan_broadcast(skb, port, src->dev,
 330                                           MACVLAN_MODE_VEPA);
 331
 332                 rcu_read_unlock();
 333
 334                 if (src)
 335                         dev_put(src->dev);
 336                 consume_skb(skb);
 337
 338                 cond_resched();
 339         }
 340 }
 341
 342 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 343                                       const struct macvlan_dev *src,
 344                                       struct sk_buff *skb)
 345 {
 346         struct sk_buff *nskb;
 347         int err = -ENOMEM;
 348
 349         nskb = skb_clone(skb, GFP_ATOMIC);
 350         if (!nskb)
 351                 goto err;
 352
 353         MACVLAN_SKB_CB(nskb)->src = src;
 354
 355         spin_lock(&port->bc_queue.lock);
 356         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 357                 if (src)
 358                         dev_hold(src->dev);
 359                 __skb_queue_tail(&port->bc_queue, nskb);
 360                 err = 0;
 361         }
 362         spin_unlock(&port->bc_queue.lock);
 363
 364         schedule_work(&port->bc_work);
 365
 366         if (err)
 367                 goto free_nskb;
 368
 369         return;
 370
 371 free_nskb:
 372         kfree_skb(nskb);
 373 err:
 374         atomic_long_inc(&skb->dev->rx_dropped);
 375 }
 376
 377 static void macvlan_flush_sources(struct macvlan_port *port,
 378                                   struct macvlan_dev *vlan)
 379 {
 380         int i;
 381
 382         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 383                 struct hlist_node *h, *n;
 384
 385                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 386                         struct macvlan_source_entry *entry;
 387
 388                         entry = hlist_entry(h, struct macvlan_source_entry,
 389                                             hlist);
 390                         if (entry->vlan == vlan)
 391                                 macvlan_hash_del_source(entry);
 392                 }
 393         }
 394         vlan->macaddr_count = 0;
 395 }
 396
 397 static void macvlan_forward_source_one(struct sk_buff *skb,
 398                                        struct macvlan_dev *vlan)
 399 {
 400         struct sk_buff *nskb;
 401         struct net_device *dev;
 402         int len;
 403         int ret;
 404
 405         dev = vlan->dev;
 406         if (unlikely(!(dev->flags & IFF_UP)))
 407                 return;
 408
 409         nskb = skb_clone(skb, GFP_ATOMIC);
 410         if (!nskb)
 411                 return;
 412
 413         len = nskb->len + ETH_HLEN;
 414         nskb->dev = dev;
 415
 416         if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
 417                 nskb->pkt_type = PACKET_HOST;
 418
 419         ret = netif_rx(nskb);
 420         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 421 }
 422
 423 static void macvlan_forward_source(struct sk_buff *skb,
 424                                    struct macvlan_port *port,
 425                                    const unsigned char *addr)
 426 {
 427         struct macvlan_source_entry *entry;
 428         u32 idx = macvlan_eth_hash(addr);
 429         struct hlist_head *h = &port->vlan_source_hash[idx];
 430
 431         hlist_for_each_entry_rcu(entry, h, hlist) {
 432                 if (ether_addr_equal_64bits(entry->addr, addr))
 433                         macvlan_forward_source_one(skb, entry->vlan);
 434         }
 435 }
 436
 437 /* called under rcu_read_lock() from netif_receive_skb */
 438 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 439 {
 440         struct macvlan_port *port;
 441         struct sk_buff *skb = *pskb;
 442         const struct ethhdr *eth = eth_hdr(skb);
 443         const struct macvlan_dev *vlan;
 444         const struct macvlan_dev *src;
 445         struct net_device *dev;
 446         unsigned int len = 0;
 447         int ret;
 448         rx_handler_result_t handle_res;
 449
 450         /* Packets from dev_loopback_xmit() do not have L2 header, bail out */
 451         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 452                 return RX_HANDLER_PASS;
 453
 454         port = macvlan_port_get_rcu(skb->dev);
 455         if (is_multicast_ether_addr(eth->h_dest)) {
 456                 unsigned int hash;
 457
 458                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 459                 if (!skb)
 460                         return RX_HANDLER_CONSUMED;
 461                 *pskb = skb;
 462                 eth = eth_hdr(skb);
 463                 macvlan_forward_source(skb, port, eth->h_source);
 464                 src = macvlan_hash_lookup(port, eth->h_source);
 465                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 466                     src->mode != MACVLAN_MODE_BRIDGE) {
 467                         /* forward to original port. */
 468                         vlan = src;
 469                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 470                               netif_rx(skb);
 471                         handle_res = RX_HANDLER_CONSUMED;
 472                         goto out;
 473                 }
 474
 475                 hash = mc_hash(NULL, eth->h_dest);
 476                 if (test_bit(hash, port->mc_filter))
 477                         macvlan_broadcast_enqueue(port, src, skb);
 478
 479                 return RX_HANDLER_PASS;
 480         }
 481
 482         macvlan_forward_source(skb, port, eth->h_source);
 483         if (macvlan_passthru(port))
 484                 vlan = list_first_or_null_rcu(&port->vlans,
 485                                               struct macvlan_dev, list);
 486         else
 487                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 488         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 489                 return RX_HANDLER_PASS;
 490
 491         dev = vlan->dev;
 492         if (unlikely(!(dev->flags & IFF_UP))) {
 493                 kfree_skb(skb);
 494                 return RX_HANDLER_CONSUMED;
 495         }
 496         len = skb->len + ETH_HLEN;
 497         skb = skb_share_check(skb, GFP_ATOMIC);
 498         if (!skb) {
 499                 ret = NET_RX_DROP;
 500                 handle_res = RX_HANDLER_CONSUMED;
 501                 goto out;
 502         }
 503
 504         *pskb = skb;
 505         skb->dev = dev;
 506         skb->pkt_type = PACKET_HOST;
 507
 508         ret = NET_RX_SUCCESS;
 509         handle_res = RX_HANDLER_ANOTHER;
 510 out:
 511         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 512         return handle_res;
 513 }
 514
 515 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 516 {
 517         const struct macvlan_dev *vlan = netdev_priv(dev);
 518         const struct macvlan_port *port = vlan->port;
 519         const struct macvlan_dev *dest;
 520
 521         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 522                 const struct ethhdr *eth = skb_eth_hdr(skb);
 523
 524                 /* send to other bridge ports directly */
 525                 if (is_multicast_ether_addr(eth->h_dest)) {
 526                         skb_reset_mac_header(skb);
 527                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 528                         goto xmit_world;
 529                 }
 530
 531                 dest = macvlan_hash_lookup(port, eth->h_dest);
 532                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 533                         /* send to lowerdev first for its network taps */
 534                         dev_forward_skb(vlan->lowerdev, skb);
 535
 536                         return NET_XMIT_SUCCESS;
 537                 }
 538         }
 539 xmit_world:
 540         skb->dev = vlan->lowerdev;
 541         return dev_queue_xmit_accel(skb,
 542                                     netdev_get_sb_channel(dev) ? dev : NULL);
 543 }
 544
 545 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 546 {
 547 #ifdef CONFIG_NET_POLL_CONTROLLER
 548         if (vlan->netpoll)
 549                 netpoll_send_skb(vlan->netpoll, skb);
 550 #else
 551         BUG();
 552 #endif
 553         return NETDEV_TX_OK;
 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 #define ALWAYS_ON_OFFLOADS \
 864         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
 865          NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
 866
 867 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
 868
 869 #define MACVLAN_FEATURES \
 870         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 871          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
 872          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 873          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 874
 875 #define MACVLAN_STATE_MASK \
 876         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 877
 878 static int macvlan_init(struct net_device *dev)
 879 {
 880         struct macvlan_dev *vlan = netdev_priv(dev);
 881         const struct net_device *lowerdev = vlan->lowerdev;
 882         struct macvlan_port *port = vlan->port;
 883
 884         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 885                                   (lowerdev->state & MACVLAN_STATE_MASK);
 886         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 887         dev->features           |= ALWAYS_ON_FEATURES;
 888         dev->hw_features        |= NETIF_F_LRO;
 889         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 890         dev->vlan_features      |= ALWAYS_ON_OFFLOADS;
 891         dev->hw_enc_features    |= dev->features;
 892         dev->gso_max_size       = lowerdev->gso_max_size;
 893         dev->gso_max_segs       = lowerdev->gso_max_segs;
 894         dev->hard_header_len    = lowerdev->hard_header_len;
 895
 896         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 897         if (!vlan->pcpu_stats)
 898                 return -ENOMEM;
 899
 900         port->count += 1;
 901
 902         return 0;
 903 }
 904
 905 static void macvlan_uninit(struct net_device *dev)
 906 {
 907         struct macvlan_dev *vlan = netdev_priv(dev);
 908         struct macvlan_port *port = vlan->port;
 909
 910         free_percpu(vlan->pcpu_stats);
 911
 912         macvlan_flush_sources(port, vlan);
 913         port->count -= 1;
 914         if (!port->count)
 915                 macvlan_port_destroy(port->dev);
 916 }
 917
 918 static void macvlan_dev_get_stats64(struct net_device *dev,
 919                                     struct rtnl_link_stats64 *stats)
 920 {
 921         struct macvlan_dev *vlan = netdev_priv(dev);
 922
 923         if (vlan->pcpu_stats) {
 924                 struct vlan_pcpu_stats *p;
 925                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 926                 u32 rx_errors = 0, tx_dropped = 0;
 927                 unsigned int start;
 928                 int i;
 929
 930                 for_each_possible_cpu(i) {
 931                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 932                         do {
 933                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 934                                 rx_packets      = p->rx_packets;
 935                                 rx_bytes        = p->rx_bytes;
 936                                 rx_multicast    = p->rx_multicast;
 937                                 tx_packets      = p->tx_packets;
 938                                 tx_bytes        = p->tx_bytes;
 939                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 940
 941                         stats->rx_packets       += rx_packets;
 942                         stats->rx_bytes         += rx_bytes;
 943                         stats->multicast        += rx_multicast;
 944                         stats->tx_packets       += tx_packets;
 945                         stats->tx_bytes         += tx_bytes;
 946                         /* rx_errors & tx_dropped are u32, updated
 947                          * without syncp protection.
 948                          */
 949                         rx_errors       += p->rx_errors;
 950                         tx_dropped      += p->tx_dropped;
 951                 }
 952                 stats->rx_errors        = rx_errors;
 953                 stats->rx_dropped       = rx_errors;
 954                 stats->tx_dropped       = tx_dropped;
 955         }
 956 }
 957
 958 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 959                                    __be16 proto, u16 vid)
 960 {
 961         struct macvlan_dev *vlan = netdev_priv(dev);
 962         struct net_device *lowerdev = vlan->lowerdev;
 963
 964         return vlan_vid_add(lowerdev, proto, vid);
 965 }
 966
 967 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 968                                     __be16 proto, u16 vid)
 969 {
 970         struct macvlan_dev *vlan = netdev_priv(dev);
 971         struct net_device *lowerdev = vlan->lowerdev;
 972
 973         vlan_vid_del(lowerdev, proto, vid);
 974         return 0;
 975 }
 976
 977 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 978                            struct net_device *dev,
 979                            const unsigned char *addr, u16 vid,
 980                            u16 flags,
 981                            struct netlink_ext_ack *extack)
 982 {
 983         struct macvlan_dev *vlan = netdev_priv(dev);
 984         int err = -EINVAL;
 985
 986         /* Support unicast filter only on passthru devices.
 987          * Multicast filter should be allowed on all devices.
 988          */
 989         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
 990                 return -EOPNOTSUPP;
 991
 992         if (flags & NLM_F_REPLACE)
 993                 return -EOPNOTSUPP;
 994
 995         if (is_unicast_ether_addr(addr))
 996                 err = dev_uc_add_excl(dev, addr);
 997         else if (is_multicast_ether_addr(addr))
 998                 err = dev_mc_add_excl(dev, addr);
 999
1000         return err;
1001 }
1002
1003 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1004                            struct net_device *dev,
1005                            const unsigned char *addr, u16 vid)
1006 {
1007         struct macvlan_dev *vlan = netdev_priv(dev);
1008         int err = -EINVAL;
1009
1010         /* Support unicast filter only on passthru devices.
1011          * Multicast filter should be allowed on all devices.
1012          */
1013         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1014                 return -EOPNOTSUPP;
1015
1016         if (is_unicast_ether_addr(addr))
1017                 err = dev_uc_del(dev, addr);
1018         else if (is_multicast_ether_addr(addr))
1019                 err = dev_mc_del(dev, addr);
1020
1021         return err;
1022 }
1023
1024 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1025                                         struct ethtool_drvinfo *drvinfo)
1026 {
1027         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1028         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1029 }
1030
1031 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1032                                               struct ethtool_link_ksettings *cmd)
1033 {
1034         const struct macvlan_dev *vlan = netdev_priv(dev);
1035
1036         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1037 }
1038
1039 static int macvlan_ethtool_get_ts_info(struct net_device *dev,
1040                                        struct ethtool_ts_info *info)
1041 {
1042         struct net_device *real_dev = macvlan_dev_real_dev(dev);
1043         const struct ethtool_ops *ops = real_dev->ethtool_ops;
1044         struct phy_device *phydev = real_dev->phydev;
1045
1046         if (phydev && phydev->drv && phydev->drv->ts_info) {
1047                  return phydev->drv->ts_info(phydev, info);
1048         } else if (ops->get_ts_info) {
1049                 return ops->get_ts_info(real_dev, info);
1050         } else {
1051                 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
1052                         SOF_TIMESTAMPING_SOFTWARE;
1053                 info->phc_index = -1;
1054         }
1055
1056         return 0;
1057 }
1058
1059 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1060                                               netdev_features_t features)
1061 {
1062         struct macvlan_dev *vlan = netdev_priv(dev);
1063         netdev_features_t lowerdev_features = vlan->lowerdev->features;
1064         netdev_features_t mask;
1065
1066         features |= NETIF_F_ALL_FOR_ALL;
1067         features &= (vlan->set_features | ~MACVLAN_FEATURES);
1068         mask = features;
1069
1070         lowerdev_features &= (features | ~NETIF_F_LRO);
1071         features = netdev_increment_features(lowerdev_features, features, mask);
1072         features |= ALWAYS_ON_FEATURES;
1073         features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1074
1075         return features;
1076 }
1077
1078 #ifdef CONFIG_NET_POLL_CONTROLLER
1079 static void macvlan_dev_poll_controller(struct net_device *dev)
1080 {
1081         return;
1082 }
1083
1084 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1085 {
1086         struct macvlan_dev *vlan = netdev_priv(dev);
1087         struct net_device *real_dev = vlan->lowerdev;
1088         struct netpoll *netpoll;
1089         int err = 0;
1090
1091         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1092         err = -ENOMEM;
1093         if (!netpoll)
1094                 goto out;
1095
1096         err = __netpoll_setup(netpoll, real_dev);
1097         if (err) {
1098                 kfree(netpoll);
1099                 goto out;
1100         }
1101
1102         vlan->netpoll = netpoll;
1103
1104 out:
1105         return err;
1106 }
1107
1108 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1109 {
1110         struct macvlan_dev *vlan = netdev_priv(dev);
1111         struct netpoll *netpoll = vlan->netpoll;
1112
1113         if (!netpoll)
1114                 return;
1115
1116         vlan->netpoll = NULL;
1117
1118         __netpoll_free(netpoll);
1119 }
1120 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1121
1122 static int macvlan_dev_get_iflink(const struct net_device *dev)
1123 {
1124         struct macvlan_dev *vlan = netdev_priv(dev);
1125
1126         return vlan->lowerdev->ifindex;
1127 }
1128
1129 static const struct ethtool_ops macvlan_ethtool_ops = {
1130         .get_link               = ethtool_op_get_link,
1131         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1132         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1133         .get_ts_info            = macvlan_ethtool_get_ts_info,
1134 };
1135
1136 static const struct net_device_ops macvlan_netdev_ops = {
1137         .ndo_init               = macvlan_init,
1138         .ndo_uninit             = macvlan_uninit,
1139         .ndo_open               = macvlan_open,
1140         .ndo_stop               = macvlan_stop,
1141         .ndo_start_xmit         = macvlan_start_xmit,
1142         .ndo_change_mtu         = macvlan_change_mtu,
1143         .ndo_do_ioctl           = macvlan_do_ioctl,
1144         .ndo_fix_features       = macvlan_fix_features,
1145         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1146         .ndo_set_mac_address    = macvlan_set_mac_address,
1147         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1148         .ndo_get_stats64        = macvlan_dev_get_stats64,
1149         .ndo_validate_addr      = eth_validate_addr,
1150         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1151         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1152         .ndo_fdb_add            = macvlan_fdb_add,
1153         .ndo_fdb_del            = macvlan_fdb_del,
1154         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1155 #ifdef CONFIG_NET_POLL_CONTROLLER
1156         .ndo_poll_controller    = macvlan_dev_poll_controller,
1157         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1158         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1159 #endif
1160         .ndo_get_iflink         = macvlan_dev_get_iflink,
1161         .ndo_features_check     = passthru_features_check,
1162         .ndo_change_proto_down  = dev_change_proto_down_generic,
1163 };
1164
1165 void macvlan_common_setup(struct net_device *dev)
1166 {
1167         ether_setup(dev);
1168
1169         /* ether_setup() has set dev->min_mtu to ETH_MIN_MTU. */
1170         dev->max_mtu            = ETH_MAX_MTU;
1171         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1172         netif_keep_dst(dev);
1173         dev->priv_flags        |= IFF_UNICAST_FLT;
1174         dev->netdev_ops         = &macvlan_netdev_ops;
1175         dev->needs_free_netdev  = true;
1176         dev->header_ops         = &macvlan_hard_header_ops;
1177         dev->ethtool_ops        = &macvlan_ethtool_ops;
1178 }
1179 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1180
1181 static void macvlan_setup(struct net_device *dev)
1182 {
1183         macvlan_common_setup(dev);
1184         dev->priv_flags |= IFF_NO_QUEUE;
1185 }
1186
1187 static int macvlan_port_create(struct net_device *dev)
1188 {
1189         struct macvlan_port *port;
1190         unsigned int i;
1191         int err;
1192
1193         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1194                 return -EINVAL;
1195
1196         if (netdev_is_rx_handler_busy(dev))
1197                 return -EBUSY;
1198
1199         port = kzalloc(sizeof(*port), GFP_KERNEL);
1200         if (port == NULL)
1201                 return -ENOMEM;
1202
1203         port->dev = dev;
1204         ether_addr_copy(port->perm_addr, dev->dev_addr);
1205         INIT_LIST_HEAD(&port->vlans);
1206         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1207                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1208         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1209                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1210
1211         skb_queue_head_init(&port->bc_queue);
1212         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1213
1214         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1215         if (err)
1216                 kfree(port);
1217         else
1218                 dev->priv_flags |= IFF_MACVLAN_PORT;
1219         return err;
1220 }
1221
1222 static void macvlan_port_destroy(struct net_device *dev)
1223 {
1224         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1225         struct sk_buff *skb;
1226
1227         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1228         netdev_rx_handler_unregister(dev);
1229
1230         /* After this point, no packet can schedule bc_work anymore,
1231          * but we need to cancel it and purge left skbs if any.
1232          */
1233         cancel_work_sync(&port->bc_work);
1234
1235         while ((skb = __skb_dequeue(&port->bc_queue))) {
1236                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1237
1238                 if (src)
1239                         dev_put(src->dev);
1240
1241                 kfree_skb(skb);
1242         }
1243
1244         /* If the lower device address has been changed by passthru
1245          * macvlan, put it back.
1246          */
1247         if (macvlan_passthru(port) &&
1248             !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1249                 struct sockaddr sa;
1250
1251                 sa.sa_family = port->dev->type;
1252                 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1253                 dev_set_mac_address(port->dev, &sa, NULL);
1254         }
1255
1256         kfree(port);
1257 }
1258
1259 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1260                             struct netlink_ext_ack *extack)
1261 {
1262         struct nlattr *nla, *head;
1263         int rem, len;
1264
1265         if (tb[IFLA_ADDRESS]) {
1266                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1267                         return -EINVAL;
1268                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1269                         return -EADDRNOTAVAIL;
1270         }
1271
1272         if (!data)
1273                 return 0;
1274
1275         if (data[IFLA_MACVLAN_FLAGS] &&
1276             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1277                 return -EINVAL;
1278
1279         if (data[IFLA_MACVLAN_MODE]) {
1280                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1281                 case MACVLAN_MODE_PRIVATE:
1282                 case MACVLAN_MODE_VEPA:
1283                 case MACVLAN_MODE_BRIDGE:
1284                 case MACVLAN_MODE_PASSTHRU:
1285                 case MACVLAN_MODE_SOURCE:
1286                         break;
1287                 default:
1288                         return -EINVAL;
1289                 }
1290         }
1291
1292         if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1293                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1294                 case MACVLAN_MACADDR_ADD:
1295                 case MACVLAN_MACADDR_DEL:
1296                 case MACVLAN_MACADDR_FLUSH:
1297                 case MACVLAN_MACADDR_SET:
1298                         break;
1299                 default:
1300                         return -EINVAL;
1301                 }
1302         }
1303
1304         if (data[IFLA_MACVLAN_MACADDR]) {
1305                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1306                         return -EINVAL;
1307
1308                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1309                         return -EADDRNOTAVAIL;
1310         }
1311
1312         if (data[IFLA_MACVLAN_MACADDR_DATA]) {
1313                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1314                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1315
1316                 nla_for_each_attr(nla, head, len, rem) {
1317                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1318                             nla_len(nla) != ETH_ALEN)
1319                                 return -EINVAL;
1320
1321                         if (!is_valid_ether_addr(nla_data(nla)))
1322                                 return -EADDRNOTAVAIL;
1323                 }
1324         }
1325
1326         if (data[IFLA_MACVLAN_MACADDR_COUNT])
1327                 return -EINVAL;
1328
1329         return 0;
1330 }
1331
1332 /**
1333  * reconfigure list of remote source mac address
1334  * (only for macvlan devices in source mode)
1335  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1336  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1337  */
1338 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1339                                       struct nlattr *data[])
1340 {
1341         char *addr = NULL;
1342         int ret, rem, len;
1343         struct nlattr *nla, *head;
1344         struct macvlan_source_entry *entry;
1345
1346         if (data[IFLA_MACVLAN_MACADDR])
1347                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1348
1349         if (mode == MACVLAN_MACADDR_ADD) {
1350                 if (!addr)
1351                         return -EINVAL;
1352
1353                 return macvlan_hash_add_source(vlan, addr);
1354
1355         } else if (mode == MACVLAN_MACADDR_DEL) {
1356                 if (!addr)
1357                         return -EINVAL;
1358
1359                 entry = macvlan_hash_lookup_source(vlan, addr);
1360                 if (entry) {
1361                         macvlan_hash_del_source(entry);
1362                         vlan->macaddr_count--;
1363                 }
1364         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1365                 macvlan_flush_sources(vlan->port, vlan);
1366         } else if (mode == MACVLAN_MACADDR_SET) {
1367                 macvlan_flush_sources(vlan->port, vlan);
1368
1369                 if (addr) {
1370                         ret = macvlan_hash_add_source(vlan, addr);
1371                         if (ret)
1372                                 return ret;
1373                 }
1374
1375                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1376                         return 0;
1377
1378                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1379                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1380
1381                 nla_for_each_attr(nla, head, len, rem) {
1382                         addr = nla_data(nla);
1383                         ret = macvlan_hash_add_source(vlan, addr);
1384                         if (ret)
1385                                 return ret;
1386                 }
1387         } else {
1388                 return -EINVAL;
1389         }
1390
1391         return 0;
1392 }
1393
1394 int macvlan_common_newlink(struct net *src_net, struct net_device *dev,
1395                            struct nlattr *tb[], struct nlattr *data[],
1396                            struct netlink_ext_ack *extack)
1397 {
1398         struct macvlan_dev *vlan = netdev_priv(dev);
1399         struct macvlan_port *port;
1400         struct net_device *lowerdev;
1401         int err;
1402         int macmode;
1403         bool create = false;
1404
1405         if (!tb[IFLA_LINK])
1406                 return -EINVAL;
1407
1408         lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK]));
1409         if (lowerdev == NULL)
1410                 return -ENODEV;
1411
1412         /* When creating macvlans or macvtaps on top of other macvlans - use
1413          * the real device as the lowerdev.
1414          */
1415         if (netif_is_macvlan(lowerdev))
1416                 lowerdev = macvlan_dev_real_dev(lowerdev);
1417
1418         if (!tb[IFLA_MTU])
1419                 dev->mtu = lowerdev->mtu;
1420         else if (dev->mtu > lowerdev->mtu)
1421                 return -EINVAL;
1422
1423         /* MTU range: 68 - lowerdev->max_mtu */
1424         dev->min_mtu = ETH_MIN_MTU;
1425         dev->max_mtu = lowerdev->max_mtu;
1426
1427         if (!tb[IFLA_ADDRESS])
1428                 eth_hw_addr_random(dev);
1429
1430         if (!netif_is_macvlan_port(lowerdev)) {
1431                 err = macvlan_port_create(lowerdev);
1432                 if (err < 0)
1433                         return err;
1434                 create = true;
1435         }
1436         port = macvlan_port_get_rtnl(lowerdev);
1437
1438         /* Only 1 macvlan device can be created in passthru mode */
1439         if (macvlan_passthru(port)) {
1440                 /* The macvlan port must be not created this time,
1441                  * still goto destroy_macvlan_port for readability.
1442                  */
1443                 err = -EINVAL;
1444                 goto destroy_macvlan_port;
1445         }
1446
1447         vlan->lowerdev = lowerdev;
1448         vlan->dev      = dev;
1449         vlan->port     = port;
1450         vlan->set_features = MACVLAN_FEATURES;
1451
1452         vlan->mode     = MACVLAN_MODE_VEPA;
1453         if (data && data[IFLA_MACVLAN_MODE])
1454                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1455
1456         if (data && data[IFLA_MACVLAN_FLAGS])
1457                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1458
1459         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1460                 if (port->count) {
1461                         err = -EINVAL;
1462                         goto destroy_macvlan_port;
1463                 }
1464                 macvlan_set_passthru(port);
1465                 eth_hw_addr_inherit(dev, lowerdev);
1466         }
1467
1468         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1469                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1470                         err = -EINVAL;
1471                         goto destroy_macvlan_port;
1472                 }
1473                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1474                 err = macvlan_changelink_sources(vlan, macmode, data);
1475                 if (err)
1476                         goto destroy_macvlan_port;
1477         }
1478
1479         err = register_netdevice(dev);
1480         if (err < 0)
1481                 goto destroy_macvlan_port;
1482
1483         dev->priv_flags |= IFF_MACVLAN;
1484         err = netdev_upper_dev_link(lowerdev, dev, extack);
1485         if (err)
1486                 goto unregister_netdev;
1487
1488         list_add_tail_rcu(&vlan->list, &port->vlans);
1489         netif_stacked_transfer_operstate(lowerdev, dev);
1490         linkwatch_fire_event(dev);
1491
1492         return 0;
1493
1494 unregister_netdev:
1495         /* macvlan_uninit would free the macvlan port */
1496         unregister_netdevice(dev);
1497         return err;
1498 destroy_macvlan_port:
1499         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1500          * so we destroy the macvlan port only when it's valid.
1501          */
1502         if (create && macvlan_port_get_rtnl(lowerdev)) {
1503                 macvlan_flush_sources(port, vlan);
1504                 macvlan_port_destroy(port->dev);
1505         }
1506         return err;
1507 }
1508 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1509
1510 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1511                            struct nlattr *tb[], struct nlattr *data[],
1512                            struct netlink_ext_ack *extack)
1513 {
1514         return macvlan_common_newlink(src_net, dev, tb, data, extack);
1515 }
1516
1517 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1518 {
1519         struct macvlan_dev *vlan = netdev_priv(dev);
1520
1521         if (vlan->mode == MACVLAN_MODE_SOURCE)
1522                 macvlan_flush_sources(vlan->port, vlan);
1523         list_del_rcu(&vlan->list);
1524         unregister_netdevice_queue(dev, head);
1525         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1526 }
1527 EXPORT_SYMBOL_GPL(macvlan_dellink);
1528
1529 static int macvlan_changelink(struct net_device *dev,
1530                               struct nlattr *tb[], struct nlattr *data[],
1531                               struct netlink_ext_ack *extack)
1532 {
1533         struct macvlan_dev *vlan = netdev_priv(dev);
1534         enum macvlan_mode mode;
1535         bool set_mode = false;
1536         enum macvlan_macaddr_mode macmode;
1537         int ret;
1538
1539         /* Validate mode, but don't set yet: setting flags may fail. */
1540         if (data && data[IFLA_MACVLAN_MODE]) {
1541                 set_mode = true;
1542                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1543                 /* Passthrough mode can't be set or cleared dynamically */
1544                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1545                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1546                         return -EINVAL;
1547                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1548                     vlan->mode != mode)
1549                         macvlan_flush_sources(vlan->port, vlan);
1550         }
1551
1552         if (data && data[IFLA_MACVLAN_FLAGS]) {
1553                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1554                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1555                 if (macvlan_passthru(vlan->port) && promisc) {
1556                         int err;
1557
1558                         if (flags & MACVLAN_FLAG_NOPROMISC)
1559                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1560                         else
1561                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1562                         if (err < 0)
1563                                 return err;
1564                 }
1565                 vlan->flags = flags;
1566         }
1567         if (set_mode)
1568                 vlan->mode = mode;
1569         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1570                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1571                         return -EINVAL;
1572                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1573                 ret = macvlan_changelink_sources(vlan, macmode, data);
1574                 if (ret)
1575                         return ret;
1576         }
1577         return 0;
1578 }
1579
1580 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1581 {
1582         if (vlan->macaddr_count == 0)
1583                 return 0;
1584         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1585                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1586 }
1587
1588 static size_t macvlan_get_size(const struct net_device *dev)
1589 {
1590         struct macvlan_dev *vlan = netdev_priv(dev);
1591
1592         return (0
1593                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1594                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1595                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1596                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1597                 );
1598 }
1599
1600 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1601                                      const struct macvlan_dev *vlan,
1602                                      const int i)
1603 {
1604         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1605         struct macvlan_source_entry *entry;
1606
1607         hlist_for_each_entry_rcu(entry, h, hlist, lockdep_rtnl_is_held()) {
1608                 if (entry->vlan != vlan)
1609                         continue;
1610                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1611                         return 1;
1612         }
1613         return 0;
1614 }
1615
1616 static int macvlan_fill_info(struct sk_buff *skb,
1617                                 const struct net_device *dev)
1618 {
1619         struct macvlan_dev *vlan = netdev_priv(dev);
1620         int i;
1621         struct nlattr *nest;
1622
1623         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1624                 goto nla_put_failure;
1625         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1626                 goto nla_put_failure;
1627         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1628                 goto nla_put_failure;
1629         if (vlan->macaddr_count > 0) {
1630                 nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA);
1631                 if (nest == NULL)
1632                         goto nla_put_failure;
1633
1634                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1635                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1636                                 goto nla_put_failure;
1637                 }
1638                 nla_nest_end(skb, nest);
1639         }
1640         return 0;
1641
1642 nla_put_failure:
1643         return -EMSGSIZE;
1644 }
1645
1646 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1647         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1648         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1649         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1650         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1651         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1652         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1653 };
1654
1655 int macvlan_link_register(struct rtnl_link_ops *ops)
1656 {
1657         /* common fields */
1658         ops->validate           = macvlan_validate;
1659         ops->maxtype            = IFLA_MACVLAN_MAX;
1660         ops->policy             = macvlan_policy;
1661         ops->changelink         = macvlan_changelink;
1662         ops->get_size           = macvlan_get_size;
1663         ops->fill_info          = macvlan_fill_info;
1664
1665         return rtnl_link_register(ops);
1666 };
1667 EXPORT_SYMBOL_GPL(macvlan_link_register);
1668
1669 static struct net *macvlan_get_link_net(const struct net_device *dev)
1670 {
1671         return dev_net(macvlan_dev_real_dev(dev));
1672 }
1673
1674 static struct rtnl_link_ops macvlan_link_ops = {
1675         .kind           = "macvlan",
1676         .setup          = macvlan_setup,
1677         .newlink        = macvlan_newlink,
1678         .dellink        = macvlan_dellink,
1679         .get_link_net   = macvlan_get_link_net,
1680         .priv_size      = sizeof(struct macvlan_dev),
1681 };
1682
1683 static int macvlan_device_event(struct notifier_block *unused,
1684                                 unsigned long event, void *ptr)
1685 {
1686         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1687         struct macvlan_dev *vlan, *next;
1688         struct macvlan_port *port;
1689         LIST_HEAD(list_kill);
1690
1691         if (!netif_is_macvlan_port(dev))
1692                 return NOTIFY_DONE;
1693
1694         port = macvlan_port_get_rtnl(dev);
1695
1696         switch (event) {
1697         case NETDEV_UP:
1698         case NETDEV_DOWN:
1699         case NETDEV_CHANGE:
1700                 list_for_each_entry(vlan, &port->vlans, list)
1701                         netif_stacked_transfer_operstate(vlan->lowerdev,
1702                                                          vlan->dev);
1703                 break;
1704         case NETDEV_FEAT_CHANGE:
1705                 list_for_each_entry(vlan, &port->vlans, list) {
1706                         vlan->dev->gso_max_size = dev->gso_max_size;
1707                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1708                         netdev_update_features(vlan->dev);
1709                 }
1710                 break;
1711         case NETDEV_CHANGEMTU:
1712                 list_for_each_entry(vlan, &port->vlans, list) {
1713                         if (vlan->dev->mtu <= dev->mtu)
1714                                 continue;
1715                         dev_set_mtu(vlan->dev, dev->mtu);
1716                 }
1717                 break;
1718         case NETDEV_CHANGEADDR:
1719                 if (!macvlan_passthru(port))
1720                         return NOTIFY_DONE;
1721
1722                 vlan = list_first_entry_or_null(&port->vlans,
1723                                                 struct macvlan_dev,
1724                                                 list);
1725
1726                 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr))
1727                         return NOTIFY_BAD;
1728
1729                 break;
1730         case NETDEV_UNREGISTER:
1731                 /* twiddle thumbs on netns device moves */
1732                 if (dev->reg_state != NETREG_UNREGISTERING)
1733                         break;
1734
1735                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1736                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1737                 unregister_netdevice_many(&list_kill);
1738                 break;
1739         case NETDEV_PRE_TYPE_CHANGE:
1740                 /* Forbid underlaying device to change its type. */
1741                 return NOTIFY_BAD;
1742
1743         case NETDEV_NOTIFY_PEERS:
1744         case NETDEV_BONDING_FAILOVER:
1745         case NETDEV_RESEND_IGMP:
1746                 /* Propagate to all vlans */
1747                 list_for_each_entry(vlan, &port->vlans, list)
1748                         call_netdevice_notifiers(event, vlan->dev);
1749         }
1750         return NOTIFY_DONE;
1751 }
1752
1753 static struct notifier_block macvlan_notifier_block __read_mostly = {
1754         .notifier_call  = macvlan_device_event,
1755 };
1756
1757 static int __init macvlan_init_module(void)
1758 {
1759         int err;
1760
1761         register_netdevice_notifier(&macvlan_notifier_block);
1762
1763         err = macvlan_link_register(&macvlan_link_ops);
1764         if (err < 0)
1765                 goto err1;
1766         return 0;
1767 err1:
1768         unregister_netdevice_notifier(&macvlan_notifier_block);
1769         return err;
1770 }
1771
1772 static void __exit macvlan_cleanup_module(void)
1773 {
1774         rtnl_link_unregister(&macvlan_link_ops);
1775         unregister_netdevice_notifier(&macvlan_notifier_block);
1776 }
1777
1778 module_init(macvlan_init_module);
1779 module_exit(macvlan_cleanup_module);
1780
1781 MODULE_LICENSE("GPL");
1782 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1783 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1784 MODULE_ALIAS_RTNL_LINK("macvlan");