drivers/net/macvlan.c

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