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/net_tstamp.h>
  28 #include <linux/ethtool.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_vlan.h>
  31 #include <linux/if_link.h>
  32 #include <linux/if_macvlan.h>
  33 #include <linux/hash.h>
  34 #include <linux/workqueue.h>
  35 #include <net/rtnetlink.h>
  36 #include <net/xfrm.h>
  37 #include <linux/netpoll.h>
  38 #include <linux/phy.h>
  39
  40 #define MACVLAN_HASH_BITS       8
  41 #define MACVLAN_HASH_SIZE       (1<<MACVLAN_HASH_BITS)
  42 #define MACVLAN_BC_QUEUE_LEN    1000
  43
  44 #define MACVLAN_F_PASSTHRU      1
  45 #define MACVLAN_F_ADDRCHANGE    2
  46
  47 struct macvlan_port {
  48         struct net_device       *dev;
  49         struct hlist_head       vlan_hash[MACVLAN_HASH_SIZE];
  50         struct list_head        vlans;
  51         struct sk_buff_head     bc_queue;
  52         struct work_struct      bc_work;
  53         u32                     flags;
  54         int                     count;
  55         struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
  56         DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
  57         unsigned char           perm_addr[ETH_ALEN];
  58 };
  59
  60 struct macvlan_source_entry {
  61         struct hlist_node       hlist;
  62         struct macvlan_dev      *vlan;
  63         unsigned char           addr[6+2] __aligned(sizeof(u16));
  64         struct rcu_head         rcu;
  65 };
  66
  67 struct macvlan_skb_cb {
  68         const struct macvlan_dev *src;
  69 };
  70
  71 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0]))
  72
  73 static void macvlan_port_destroy(struct net_device *dev);
  74
  75 static inline bool macvlan_passthru(const struct macvlan_port *port)
  76 {
  77         return port->flags & MACVLAN_F_PASSTHRU;
  78 }
  79
  80 static inline void macvlan_set_passthru(struct macvlan_port *port)
  81 {
  82         port->flags |= MACVLAN_F_PASSTHRU;
  83 }
  84
  85 static inline bool macvlan_addr_change(const struct macvlan_port *port)
  86 {
  87         return port->flags & MACVLAN_F_ADDRCHANGE;
  88 }
  89
  90 static inline void macvlan_set_addr_change(struct macvlan_port *port)
  91 {
  92         port->flags |= MACVLAN_F_ADDRCHANGE;
  93 }
  94
  95 static inline void macvlan_clear_addr_change(struct macvlan_port *port)
  96 {
  97         port->flags &= ~MACVLAN_F_ADDRCHANGE;
  98 }
  99
 100 /* Hash Ethernet address */
 101 static u32 macvlan_eth_hash(const unsigned char *addr)
 102 {
 103         u64 value = get_unaligned((u64 *)addr);
 104
 105         /* only want 6 bytes */
 106 #ifdef __BIG_ENDIAN
 107         value >>= 16;
 108 #else
 109         value <<= 16;
 110 #endif
 111         return hash_64(value, MACVLAN_HASH_BITS);
 112 }
 113
 114 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev)
 115 {
 116         return rcu_dereference(dev->rx_handler_data);
 117 }
 118
 119 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev)
 120 {
 121         return rtnl_dereference(dev->rx_handler_data);
 122 }
 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                 consume_skb(skb);
 341         }
 342 }
 343
 344 static void macvlan_broadcast_enqueue(struct macvlan_port *port,
 345                                       const struct macvlan_dev *src,
 346                                       struct sk_buff *skb)
 347 {
 348         struct sk_buff *nskb;
 349         int err = -ENOMEM;
 350
 351         nskb = skb_clone(skb, GFP_ATOMIC);
 352         if (!nskb)
 353                 goto err;
 354
 355         MACVLAN_SKB_CB(nskb)->src = src;
 356
 357         spin_lock(&port->bc_queue.lock);
 358         if (skb_queue_len(&port->bc_queue) < MACVLAN_BC_QUEUE_LEN) {
 359                 if (src)
 360                         dev_hold(src->dev);
 361                 __skb_queue_tail(&port->bc_queue, nskb);
 362                 err = 0;
 363         }
 364         spin_unlock(&port->bc_queue.lock);
 365
 366         if (err)
 367                 goto free_nskb;
 368
 369         schedule_work(&port->bc_work);
 370         return;
 371
 372 free_nskb:
 373         kfree_skb(nskb);
 374 err:
 375         atomic_long_inc(&skb->dev->rx_dropped);
 376 }
 377
 378 static void macvlan_flush_sources(struct macvlan_port *port,
 379                                   struct macvlan_dev *vlan)
 380 {
 381         int i;
 382
 383         for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
 384                 struct hlist_node *h, *n;
 385
 386                 hlist_for_each_safe(h, n, &port->vlan_source_hash[i]) {
 387                         struct macvlan_source_entry *entry;
 388
 389                         entry = hlist_entry(h, struct macvlan_source_entry,
 390                                             hlist);
 391                         if (entry->vlan == vlan)
 392                                 macvlan_hash_del_source(entry);
 393                 }
 394         }
 395         vlan->macaddr_count = 0;
 396 }
 397
 398 static void macvlan_forward_source_one(struct sk_buff *skb,
 399                                        struct macvlan_dev *vlan)
 400 {
 401         struct sk_buff *nskb;
 402         struct net_device *dev;
 403         int len;
 404         int ret;
 405
 406         dev = vlan->dev;
 407         if (unlikely(!(dev->flags & IFF_UP)))
 408                 return;
 409
 410         nskb = skb_clone(skb, GFP_ATOMIC);
 411         if (!nskb)
 412                 return;
 413
 414         len = nskb->len + ETH_HLEN;
 415         nskb->dev = dev;
 416
 417         if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr))
 418                 nskb->pkt_type = PACKET_HOST;
 419
 420         ret = netif_rx(nskb);
 421         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 422 }
 423
 424 static void macvlan_forward_source(struct sk_buff *skb,
 425                                    struct macvlan_port *port,
 426                                    const unsigned char *addr)
 427 {
 428         struct macvlan_source_entry *entry;
 429         u32 idx = macvlan_eth_hash(addr);
 430         struct hlist_head *h = &port->vlan_source_hash[idx];
 431
 432         hlist_for_each_entry_rcu(entry, h, hlist) {
 433                 if (ether_addr_equal_64bits(entry->addr, addr))
 434                         macvlan_forward_source_one(skb, entry->vlan);
 435         }
 436 }
 437
 438 /* called under rcu_read_lock() from netif_receive_skb */
 439 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 440 {
 441         struct macvlan_port *port;
 442         struct sk_buff *skb = *pskb;
 443         const struct ethhdr *eth = eth_hdr(skb);
 444         const struct macvlan_dev *vlan;
 445         const struct macvlan_dev *src;
 446         struct net_device *dev;
 447         unsigned int len = 0;
 448         int ret;
 449         rx_handler_result_t handle_res;
 450
 451         port = macvlan_port_get_rcu(skb->dev);
 452         if (is_multicast_ether_addr(eth->h_dest)) {
 453                 unsigned int hash;
 454
 455                 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
 456                 if (!skb)
 457                         return RX_HANDLER_CONSUMED;
 458                 *pskb = skb;
 459                 eth = eth_hdr(skb);
 460                 macvlan_forward_source(skb, port, eth->h_source);
 461                 src = macvlan_hash_lookup(port, eth->h_source);
 462                 if (src && src->mode != MACVLAN_MODE_VEPA &&
 463                     src->mode != MACVLAN_MODE_BRIDGE) {
 464                         /* forward to original port. */
 465                         vlan = src;
 466                         ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?:
 467                               netif_rx(skb);
 468                         handle_res = RX_HANDLER_CONSUMED;
 469                         goto out;
 470                 }
 471
 472                 hash = mc_hash(NULL, eth->h_dest);
 473                 if (test_bit(hash, port->mc_filter))
 474                         macvlan_broadcast_enqueue(port, src, skb);
 475
 476                 return RX_HANDLER_PASS;
 477         }
 478
 479         macvlan_forward_source(skb, port, eth->h_source);
 480         if (macvlan_passthru(port))
 481                 vlan = list_first_or_null_rcu(&port->vlans,
 482                                               struct macvlan_dev, list);
 483         else
 484                 vlan = macvlan_hash_lookup(port, eth->h_dest);
 485         if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE)
 486                 return RX_HANDLER_PASS;
 487
 488         dev = vlan->dev;
 489         if (unlikely(!(dev->flags & IFF_UP))) {
 490                 kfree_skb(skb);
 491                 return RX_HANDLER_CONSUMED;
 492         }
 493         len = skb->len + ETH_HLEN;
 494         skb = skb_share_check(skb, GFP_ATOMIC);
 495         if (!skb) {
 496                 ret = NET_RX_DROP;
 497                 handle_res = RX_HANDLER_CONSUMED;
 498                 goto out;
 499         }
 500
 501         *pskb = skb;
 502         skb->dev = dev;
 503         skb->pkt_type = PACKET_HOST;
 504
 505         ret = NET_RX_SUCCESS;
 506         handle_res = RX_HANDLER_ANOTHER;
 507 out:
 508         macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false);
 509         return handle_res;
 510 }
 511
 512 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev)
 513 {
 514         const struct macvlan_dev *vlan = netdev_priv(dev);
 515         const struct macvlan_port *port = vlan->port;
 516         const struct macvlan_dev *dest;
 517
 518         if (vlan->mode == MACVLAN_MODE_BRIDGE) {
 519                 const struct ethhdr *eth = (void *)skb->data;
 520
 521                 /* send to other bridge ports directly */
 522                 if (is_multicast_ether_addr(eth->h_dest)) {
 523                         macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE);
 524                         goto xmit_world;
 525                 }
 526
 527                 dest = macvlan_hash_lookup(port, eth->h_dest);
 528                 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) {
 529                         /* send to lowerdev first for its network taps */
 530                         dev_forward_skb(vlan->lowerdev, skb);
 531
 532                         return NET_XMIT_SUCCESS;
 533                 }
 534         }
 535 xmit_world:
 536         skb->dev = vlan->lowerdev;
 537         return dev_queue_xmit_accel(skb,
 538                                     netdev_get_sb_channel(dev) ? dev : NULL);
 539 }
 540
 541 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb)
 542 {
 543 #ifdef CONFIG_NET_POLL_CONTROLLER
 544         if (vlan->netpoll)
 545                 netpoll_send_skb(vlan->netpoll, skb);
 546 #else
 547         BUG();
 548 #endif
 549         return NETDEV_TX_OK;
 550 }
 551
 552 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb,
 553                                       struct net_device *dev)
 554 {
 555         struct macvlan_dev *vlan = netdev_priv(dev);
 556         unsigned int len = skb->len;
 557         int ret;
 558
 559         if (unlikely(netpoll_tx_running(dev)))
 560                 return macvlan_netpoll_send_skb(vlan, skb);
 561
 562         ret = macvlan_queue_xmit(skb, dev);
 563
 564         if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) {
 565                 struct vlan_pcpu_stats *pcpu_stats;
 566
 567                 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats);
 568                 u64_stats_update_begin(&pcpu_stats->syncp);
 569                 pcpu_stats->tx_packets++;
 570                 pcpu_stats->tx_bytes += len;
 571                 u64_stats_update_end(&pcpu_stats->syncp);
 572         } else {
 573                 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 574         }
 575         return ret;
 576 }
 577
 578 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev,
 579                                unsigned short type, const void *daddr,
 580                                const void *saddr, unsigned len)
 581 {
 582         const struct macvlan_dev *vlan = netdev_priv(dev);
 583         struct net_device *lowerdev = vlan->lowerdev;
 584
 585         return dev_hard_header(skb, lowerdev, type, daddr,
 586                                saddr ? : dev->dev_addr, len);
 587 }
 588
 589 static const struct header_ops macvlan_hard_header_ops = {
 590         .create         = macvlan_hard_header,
 591         .parse          = eth_header_parse,
 592         .cache          = eth_header_cache,
 593         .cache_update   = eth_header_cache_update,
 594 };
 595
 596 static int macvlan_open(struct net_device *dev)
 597 {
 598         struct macvlan_dev *vlan = netdev_priv(dev);
 599         struct net_device *lowerdev = vlan->lowerdev;
 600         int err;
 601
 602         if (macvlan_passthru(vlan->port)) {
 603                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) {
 604                         err = dev_set_promiscuity(lowerdev, 1);
 605                         if (err < 0)
 606                                 goto out;
 607                 }
 608                 goto hash_add;
 609         }
 610
 611         err = -EADDRINUSE;
 612         if (macvlan_addr_busy(vlan->port, dev->dev_addr))
 613                 goto out;
 614
 615         /* Attempt to populate accel_priv which is used to offload the L2
 616          * forwarding requests for unicast packets.
 617          */
 618         if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD)
 619                 vlan->accel_priv =
 620                       lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev);
 621
 622         /* If earlier attempt to offload failed, or accel_priv is not
 623          * populated we must add the unicast address to the lower device.
 624          */
 625         if (IS_ERR_OR_NULL(vlan->accel_priv)) {
 626                 vlan->accel_priv = NULL;
 627                 err = dev_uc_add(lowerdev, dev->dev_addr);
 628                 if (err < 0)
 629                         goto out;
 630         }
 631
 632         if (dev->flags & IFF_ALLMULTI) {
 633                 err = dev_set_allmulti(lowerdev, 1);
 634                 if (err < 0)
 635                         goto del_unicast;
 636         }
 637
 638         if (dev->flags & IFF_PROMISC) {
 639                 err = dev_set_promiscuity(lowerdev, 1);
 640                 if (err < 0)
 641                         goto clear_multi;
 642         }
 643
 644 hash_add:
 645         macvlan_hash_add(vlan);
 646         return 0;
 647
 648 clear_multi:
 649         if (dev->flags & IFF_ALLMULTI)
 650                 dev_set_allmulti(lowerdev, -1);
 651 del_unicast:
 652         if (vlan->accel_priv) {
 653                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 654                                                            vlan->accel_priv);
 655                 vlan->accel_priv = NULL;
 656         } else {
 657                 dev_uc_del(lowerdev, dev->dev_addr);
 658         }
 659 out:
 660         return err;
 661 }
 662
 663 static int macvlan_stop(struct net_device *dev)
 664 {
 665         struct macvlan_dev *vlan = netdev_priv(dev);
 666         struct net_device *lowerdev = vlan->lowerdev;
 667
 668         if (vlan->accel_priv) {
 669                 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev,
 670                                                            vlan->accel_priv);
 671                 vlan->accel_priv = NULL;
 672         }
 673
 674         dev_uc_unsync(lowerdev, dev);
 675         dev_mc_unsync(lowerdev, dev);
 676
 677         if (macvlan_passthru(vlan->port)) {
 678                 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC))
 679                         dev_set_promiscuity(lowerdev, -1);
 680                 goto hash_del;
 681         }
 682
 683         if (dev->flags & IFF_ALLMULTI)
 684                 dev_set_allmulti(lowerdev, -1);
 685
 686         if (dev->flags & IFF_PROMISC)
 687                 dev_set_promiscuity(lowerdev, -1);
 688
 689         dev_uc_del(lowerdev, dev->dev_addr);
 690
 691 hash_del:
 692         macvlan_hash_del(vlan, !dev->dismantle);
 693         return 0;
 694 }
 695
 696 static int macvlan_sync_address(struct net_device *dev, unsigned char *addr)
 697 {
 698         struct macvlan_dev *vlan = netdev_priv(dev);
 699         struct net_device *lowerdev = vlan->lowerdev;
 700         struct macvlan_port *port = vlan->port;
 701         int err;
 702
 703         if (!(dev->flags & IFF_UP)) {
 704                 /* Just copy in the new address */
 705                 ether_addr_copy(dev->dev_addr, addr);
 706         } else {
 707                 /* Rehash and update the device filters */
 708                 if (macvlan_addr_busy(vlan->port, addr))
 709                         return -EADDRINUSE;
 710
 711                 if (!macvlan_passthru(port)) {
 712                         err = dev_uc_add(lowerdev, addr);
 713                         if (err)
 714                                 return err;
 715
 716                         dev_uc_del(lowerdev, dev->dev_addr);
 717                 }
 718
 719                 macvlan_hash_change_addr(vlan, addr);
 720         }
 721         if (macvlan_passthru(port) && !macvlan_addr_change(port)) {
 722                 /* Since addr_change isn't set, we are here due to lower
 723                  * device change.  Save the lower-dev address so we can
 724                  * restore it later.
 725                  */
 726                 ether_addr_copy(vlan->port->perm_addr,
 727                                 lowerdev->dev_addr);
 728         }
 729         macvlan_clear_addr_change(port);
 730         return 0;
 731 }
 732
 733 static int macvlan_set_mac_address(struct net_device *dev, void *p)
 734 {
 735         struct macvlan_dev *vlan = netdev_priv(dev);
 736         struct sockaddr *addr = p;
 737
 738         if (!is_valid_ether_addr(addr->sa_data))
 739                 return -EADDRNOTAVAIL;
 740
 741         /* If the addresses are the same, this is a no-op */
 742         if (ether_addr_equal(dev->dev_addr, addr->sa_data))
 743                 return 0;
 744
 745         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
 746                 macvlan_set_addr_change(vlan->port);
 747                 return dev_set_mac_address(vlan->lowerdev, addr, NULL);
 748         }
 749
 750         if (macvlan_addr_busy(vlan->port, addr->sa_data))
 751                 return -EADDRINUSE;
 752
 753         return macvlan_sync_address(dev, addr->sa_data);
 754 }
 755
 756 static void macvlan_change_rx_flags(struct net_device *dev, int change)
 757 {
 758         struct macvlan_dev *vlan = netdev_priv(dev);
 759         struct net_device *lowerdev = vlan->lowerdev;
 760
 761         if (dev->flags & IFF_UP) {
 762                 if (change & IFF_ALLMULTI)
 763                         dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1);
 764                 if (change & IFF_PROMISC)
 765                         dev_set_promiscuity(lowerdev,
 766                                             dev->flags & IFF_PROMISC ? 1 : -1);
 767
 768         }
 769 }
 770
 771 static void macvlan_compute_filter(unsigned long *mc_filter,
 772                                    struct net_device *dev,
 773                                    struct macvlan_dev *vlan)
 774 {
 775         if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
 776                 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
 777         } else {
 778                 struct netdev_hw_addr *ha;
 779                 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
 780
 781                 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ);
 782                 netdev_for_each_mc_addr(ha, dev) {
 783                         __set_bit(mc_hash(vlan, ha->addr), filter);
 784                 }
 785
 786                 __set_bit(mc_hash(vlan, dev->broadcast), filter);
 787
 788                 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
 789         }
 790 }
 791
 792 static void macvlan_set_mac_lists(struct net_device *dev)
 793 {
 794         struct macvlan_dev *vlan = netdev_priv(dev);
 795
 796         macvlan_compute_filter(vlan->mc_filter, dev, vlan);
 797
 798         dev_uc_sync(vlan->lowerdev, dev);
 799         dev_mc_sync(vlan->lowerdev, dev);
 800
 801         /* This is slightly inaccurate as we're including the subscription
 802          * list of vlan->lowerdev too.
 803          *
 804          * Bug alert: This only works if everyone has the same broadcast
 805          * address as lowerdev.  As soon as someone changes theirs this
 806          * will break.
 807          *
 808          * However, this is already broken as when you change your broadcast
 809          * address we don't get called.
 810          *
 811          * The solution is to maintain a list of broadcast addresses like
 812          * we do for uc/mc, if you care.
 813          */
 814         macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 815 }
 816
 817 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)
 818 {
 819         struct macvlan_dev *vlan = netdev_priv(dev);
 820
 821         if (vlan->lowerdev->mtu < new_mtu)
 822                 return -EINVAL;
 823         dev->mtu = new_mtu;
 824         return 0;
 825 }
 826
 827 static int macvlan_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 828 {
 829         struct net_device *real_dev = macvlan_dev_real_dev(dev);
 830         const struct net_device_ops *ops = real_dev->netdev_ops;
 831         struct ifreq ifrr;
 832         int err = -EOPNOTSUPP;
 833
 834         strncpy(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
 835         ifrr.ifr_ifru = ifr->ifr_ifru;
 836
 837         switch (cmd) {
 838         case SIOCSHWTSTAMP:
 839                 if (!net_eq(dev_net(dev), &init_net))
 840                         break;
 841         case SIOCGHWTSTAMP:
 842                 if (netif_device_present(real_dev) && ops->ndo_do_ioctl)
 843                         err = ops->ndo_do_ioctl(real_dev, &ifrr, cmd);
 844                 break;
 845         }
 846
 847         if (!err)
 848                 ifr->ifr_ifru = ifrr.ifr_ifru;
 849
 850         return err;
 851 }
 852
 853 /*
 854  * macvlan network devices have devices nesting below it and are a special
 855  * "super class" of normal network devices; split their locks off into a
 856  * separate class since they always nest.
 857  */
 858 static struct lock_class_key macvlan_netdev_addr_lock_key;
 859
 860 #define ALWAYS_ON_OFFLOADS \
 861         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \
 862          NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL)
 863
 864 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX)
 865
 866 #define MACVLAN_FEATURES \
 867         (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \
 868          NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \
 869          NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \
 870          NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER)
 871
 872 #define MACVLAN_STATE_MASK \
 873         ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT))
 874
 875 static int macvlan_get_nest_level(struct net_device *dev)
 876 {
 877         return ((struct macvlan_dev *)netdev_priv(dev))->nest_level;
 878 }
 879
 880 static void macvlan_set_lockdep_class(struct net_device *dev)
 881 {
 882         netdev_lockdep_set_classes(dev);
 883         lockdep_set_class_and_subclass(&dev->addr_list_lock,
 884                                        &macvlan_netdev_addr_lock_key,
 885                                        macvlan_get_nest_level(dev));
 886 }
 887
 888 static int macvlan_init(struct net_device *dev)
 889 {
 890         struct macvlan_dev *vlan = netdev_priv(dev);
 891         const struct net_device *lowerdev = vlan->lowerdev;
 892         struct macvlan_port *port = vlan->port;
 893
 894         dev->state              = (dev->state & ~MACVLAN_STATE_MASK) |
 895                                   (lowerdev->state & MACVLAN_STATE_MASK);
 896         dev->features           = lowerdev->features & MACVLAN_FEATURES;
 897         dev->features           |= ALWAYS_ON_FEATURES;
 898         dev->hw_features        |= NETIF_F_LRO;
 899         dev->vlan_features      = lowerdev->vlan_features & MACVLAN_FEATURES;
 900         dev->vlan_features      |= ALWAYS_ON_OFFLOADS;
 901         dev->hw_enc_features    |= dev->features;
 902         dev->gso_max_size       = lowerdev->gso_max_size;
 903         dev->gso_max_segs       = lowerdev->gso_max_segs;
 904         dev->hard_header_len    = lowerdev->hard_header_len;
 905
 906         macvlan_set_lockdep_class(dev);
 907
 908         vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats);
 909         if (!vlan->pcpu_stats)
 910                 return -ENOMEM;
 911
 912         port->count += 1;
 913
 914         return 0;
 915 }
 916
 917 static void macvlan_uninit(struct net_device *dev)
 918 {
 919         struct macvlan_dev *vlan = netdev_priv(dev);
 920         struct macvlan_port *port = vlan->port;
 921
 922         free_percpu(vlan->pcpu_stats);
 923
 924         macvlan_flush_sources(port, vlan);
 925         port->count -= 1;
 926         if (!port->count)
 927                 macvlan_port_destroy(port->dev);
 928 }
 929
 930 static void macvlan_dev_get_stats64(struct net_device *dev,
 931                                     struct rtnl_link_stats64 *stats)
 932 {
 933         struct macvlan_dev *vlan = netdev_priv(dev);
 934
 935         if (vlan->pcpu_stats) {
 936                 struct vlan_pcpu_stats *p;
 937                 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes;
 938                 u32 rx_errors = 0, tx_dropped = 0;
 939                 unsigned int start;
 940                 int i;
 941
 942                 for_each_possible_cpu(i) {
 943                         p = per_cpu_ptr(vlan->pcpu_stats, i);
 944                         do {
 945                                 start = u64_stats_fetch_begin_irq(&p->syncp);
 946                                 rx_packets      = p->rx_packets;
 947                                 rx_bytes        = p->rx_bytes;
 948                                 rx_multicast    = p->rx_multicast;
 949                                 tx_packets      = p->tx_packets;
 950                                 tx_bytes        = p->tx_bytes;
 951                         } while (u64_stats_fetch_retry_irq(&p->syncp, start));
 952
 953                         stats->rx_packets       += rx_packets;
 954                         stats->rx_bytes         += rx_bytes;
 955                         stats->multicast        += rx_multicast;
 956                         stats->tx_packets       += tx_packets;
 957                         stats->tx_bytes         += tx_bytes;
 958                         /* rx_errors & tx_dropped are u32, updated
 959                          * without syncp protection.
 960                          */
 961                         rx_errors       += p->rx_errors;
 962                         tx_dropped      += p->tx_dropped;
 963                 }
 964                 stats->rx_errors        = rx_errors;
 965                 stats->rx_dropped       = rx_errors;
 966                 stats->tx_dropped       = tx_dropped;
 967         }
 968 }
 969
 970 static int macvlan_vlan_rx_add_vid(struct net_device *dev,
 971                                    __be16 proto, u16 vid)
 972 {
 973         struct macvlan_dev *vlan = netdev_priv(dev);
 974         struct net_device *lowerdev = vlan->lowerdev;
 975
 976         return vlan_vid_add(lowerdev, proto, vid);
 977 }
 978
 979 static int macvlan_vlan_rx_kill_vid(struct net_device *dev,
 980                                     __be16 proto, u16 vid)
 981 {
 982         struct macvlan_dev *vlan = netdev_priv(dev);
 983         struct net_device *lowerdev = vlan->lowerdev;
 984
 985         vlan_vid_del(lowerdev, proto, vid);
 986         return 0;
 987 }
 988
 989 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 990                            struct net_device *dev,
 991                            const unsigned char *addr, u16 vid,
 992                            u16 flags,
 993                            struct netlink_ext_ack *extack)
 994 {
 995         struct macvlan_dev *vlan = netdev_priv(dev);
 996         int err = -EINVAL;
 997
 998         /* Support unicast filter only on passthru devices.
 999          * Multicast filter should be allowed on all devices.
1000          */
1001         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1002                 return -EOPNOTSUPP;
1003
1004         if (flags & NLM_F_REPLACE)
1005                 return -EOPNOTSUPP;
1006
1007         if (is_unicast_ether_addr(addr))
1008                 err = dev_uc_add_excl(dev, addr);
1009         else if (is_multicast_ether_addr(addr))
1010                 err = dev_mc_add_excl(dev, addr);
1011
1012         return err;
1013 }
1014
1015 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[],
1016                            struct net_device *dev,
1017                            const unsigned char *addr, u16 vid)
1018 {
1019         struct macvlan_dev *vlan = netdev_priv(dev);
1020         int err = -EINVAL;
1021
1022         /* Support unicast filter only on passthru devices.
1023          * Multicast filter should be allowed on all devices.
1024          */
1025         if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr))
1026                 return -EOPNOTSUPP;
1027
1028         if (is_unicast_ether_addr(addr))
1029                 err = dev_uc_del(dev, addr);
1030         else if (is_multicast_ether_addr(addr))
1031                 err = dev_mc_del(dev, addr);
1032
1033         return err;
1034 }
1035
1036 static void macvlan_ethtool_get_drvinfo(struct net_device *dev,
1037                                         struct ethtool_drvinfo *drvinfo)
1038 {
1039         strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver));
1040         strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version));
1041 }
1042
1043 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,
1044                                               struct ethtool_link_ksettings *cmd)
1045 {
1046         const struct macvlan_dev *vlan = netdev_priv(dev);
1047
1048         return __ethtool_get_link_ksettings(vlan->lowerdev, cmd);
1049 }
1050
1051 static int macvlan_ethtool_get_ts_info(struct net_device *dev,
1052                                        struct ethtool_ts_info *info)
1053 {
1054         struct net_device *real_dev = macvlan_dev_real_dev(dev);
1055         const struct ethtool_ops *ops = real_dev->ethtool_ops;
1056         struct phy_device *phydev = real_dev->phydev;
1057
1058         if (phydev && phydev->drv && phydev->drv->ts_info) {
1059                  return phydev->drv->ts_info(phydev, info);
1060         } else if (ops->get_ts_info) {
1061                 return ops->get_ts_info(real_dev, info);
1062         } else {
1063                 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
1064                         SOF_TIMESTAMPING_SOFTWARE;
1065                 info->phc_index = -1;
1066         }
1067
1068         return 0;
1069 }
1070
1071 static netdev_features_t macvlan_fix_features(struct net_device *dev,
1072                                               netdev_features_t features)
1073 {
1074         struct macvlan_dev *vlan = netdev_priv(dev);
1075         netdev_features_t lowerdev_features = vlan->lowerdev->features;
1076         netdev_features_t mask;
1077
1078         features |= NETIF_F_ALL_FOR_ALL;
1079         features &= (vlan->set_features | ~MACVLAN_FEATURES);
1080         mask = features;
1081
1082         lowerdev_features &= (features | ~NETIF_F_LRO);
1083         features = netdev_increment_features(lowerdev_features, features, mask);
1084         features |= ALWAYS_ON_FEATURES;
1085         features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES);
1086
1087         return features;
1088 }
1089
1090 #ifdef CONFIG_NET_POLL_CONTROLLER
1091 static void macvlan_dev_poll_controller(struct net_device *dev)
1092 {
1093         return;
1094 }
1095
1096 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo)
1097 {
1098         struct macvlan_dev *vlan = netdev_priv(dev);
1099         struct net_device *real_dev = vlan->lowerdev;
1100         struct netpoll *netpoll;
1101         int err = 0;
1102
1103         netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL);
1104         err = -ENOMEM;
1105         if (!netpoll)
1106                 goto out;
1107
1108         err = __netpoll_setup(netpoll, real_dev);
1109         if (err) {
1110                 kfree(netpoll);
1111                 goto out;
1112         }
1113
1114         vlan->netpoll = netpoll;
1115
1116 out:
1117         return err;
1118 }
1119
1120 static void macvlan_dev_netpoll_cleanup(struct net_device *dev)
1121 {
1122         struct macvlan_dev *vlan = netdev_priv(dev);
1123         struct netpoll *netpoll = vlan->netpoll;
1124
1125         if (!netpoll)
1126                 return;
1127
1128         vlan->netpoll = NULL;
1129
1130         __netpoll_free(netpoll);
1131 }
1132 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1133
1134 static int macvlan_dev_get_iflink(const struct net_device *dev)
1135 {
1136         struct macvlan_dev *vlan = netdev_priv(dev);
1137
1138         return vlan->lowerdev->ifindex;
1139 }
1140
1141 static const struct ethtool_ops macvlan_ethtool_ops = {
1142         .get_link               = ethtool_op_get_link,
1143         .get_link_ksettings     = macvlan_ethtool_get_link_ksettings,
1144         .get_drvinfo            = macvlan_ethtool_get_drvinfo,
1145         .get_ts_info            = macvlan_ethtool_get_ts_info,
1146 };
1147
1148 static const struct net_device_ops macvlan_netdev_ops = {
1149         .ndo_init               = macvlan_init,
1150         .ndo_uninit             = macvlan_uninit,
1151         .ndo_open               = macvlan_open,
1152         .ndo_stop               = macvlan_stop,
1153         .ndo_start_xmit         = macvlan_start_xmit,
1154         .ndo_change_mtu         = macvlan_change_mtu,
1155         .ndo_do_ioctl           = macvlan_do_ioctl,
1156         .ndo_fix_features       = macvlan_fix_features,
1157         .ndo_change_rx_flags    = macvlan_change_rx_flags,
1158         .ndo_set_mac_address    = macvlan_set_mac_address,
1159         .ndo_set_rx_mode        = macvlan_set_mac_lists,
1160         .ndo_get_stats64        = macvlan_dev_get_stats64,
1161         .ndo_validate_addr      = eth_validate_addr,
1162         .ndo_vlan_rx_add_vid    = macvlan_vlan_rx_add_vid,
1163         .ndo_vlan_rx_kill_vid   = macvlan_vlan_rx_kill_vid,
1164         .ndo_fdb_add            = macvlan_fdb_add,
1165         .ndo_fdb_del            = macvlan_fdb_del,
1166         .ndo_fdb_dump           = ndo_dflt_fdb_dump,
1167         .ndo_get_lock_subclass  = macvlan_get_nest_level,
1168 #ifdef CONFIG_NET_POLL_CONTROLLER
1169         .ndo_poll_controller    = macvlan_dev_poll_controller,
1170         .ndo_netpoll_setup      = macvlan_dev_netpoll_setup,
1171         .ndo_netpoll_cleanup    = macvlan_dev_netpoll_cleanup,
1172 #endif
1173         .ndo_get_iflink         = macvlan_dev_get_iflink,
1174         .ndo_features_check     = passthru_features_check,
1175         .ndo_change_proto_down  = dev_change_proto_down_generic,
1176 };
1177
1178 void macvlan_common_setup(struct net_device *dev)
1179 {
1180         ether_setup(dev);
1181
1182         dev->min_mtu            = 0;
1183         dev->max_mtu            = ETH_MAX_MTU;
1184         dev->priv_flags        &= ~IFF_TX_SKB_SHARING;
1185         netif_keep_dst(dev);
1186         dev->priv_flags        |= IFF_UNICAST_FLT;
1187         dev->netdev_ops         = &macvlan_netdev_ops;
1188         dev->needs_free_netdev  = true;
1189         dev->header_ops         = &macvlan_hard_header_ops;
1190         dev->ethtool_ops        = &macvlan_ethtool_ops;
1191 }
1192 EXPORT_SYMBOL_GPL(macvlan_common_setup);
1193
1194 static void macvlan_setup(struct net_device *dev)
1195 {
1196         macvlan_common_setup(dev);
1197         dev->priv_flags |= IFF_NO_QUEUE;
1198 }
1199
1200 static int macvlan_port_create(struct net_device *dev)
1201 {
1202         struct macvlan_port *port;
1203         unsigned int i;
1204         int err;
1205
1206         if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK)
1207                 return -EINVAL;
1208
1209         if (netdev_is_rx_handler_busy(dev))
1210                 return -EBUSY;
1211
1212         port = kzalloc(sizeof(*port), GFP_KERNEL);
1213         if (port == NULL)
1214                 return -ENOMEM;
1215
1216         port->dev = dev;
1217         ether_addr_copy(port->perm_addr, dev->dev_addr);
1218         INIT_LIST_HEAD(&port->vlans);
1219         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1220                 INIT_HLIST_HEAD(&port->vlan_hash[i]);
1221         for (i = 0; i < MACVLAN_HASH_SIZE; i++)
1222                 INIT_HLIST_HEAD(&port->vlan_source_hash[i]);
1223
1224         skb_queue_head_init(&port->bc_queue);
1225         INIT_WORK(&port->bc_work, macvlan_process_broadcast);
1226
1227         err = netdev_rx_handler_register(dev, macvlan_handle_frame, port);
1228         if (err)
1229                 kfree(port);
1230         else
1231                 dev->priv_flags |= IFF_MACVLAN_PORT;
1232         return err;
1233 }
1234
1235 static void macvlan_port_destroy(struct net_device *dev)
1236 {
1237         struct macvlan_port *port = macvlan_port_get_rtnl(dev);
1238         struct sk_buff *skb;
1239
1240         dev->priv_flags &= ~IFF_MACVLAN_PORT;
1241         netdev_rx_handler_unregister(dev);
1242
1243         /* After this point, no packet can schedule bc_work anymore,
1244          * but we need to cancel it and purge left skbs if any.
1245          */
1246         cancel_work_sync(&port->bc_work);
1247
1248         while ((skb = __skb_dequeue(&port->bc_queue))) {
1249                 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src;
1250
1251                 if (src)
1252                         dev_put(src->dev);
1253
1254                 kfree_skb(skb);
1255         }
1256
1257         /* If the lower device address has been changed by passthru
1258          * macvlan, put it back.
1259          */
1260         if (macvlan_passthru(port) &&
1261             !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) {
1262                 struct sockaddr sa;
1263
1264                 sa.sa_family = port->dev->type;
1265                 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len);
1266                 dev_set_mac_address(port->dev, &sa, NULL);
1267         }
1268
1269         kfree(port);
1270 }
1271
1272 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[],
1273                             struct netlink_ext_ack *extack)
1274 {
1275         if (tb[IFLA_ADDRESS]) {
1276                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1277                         return -EINVAL;
1278                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1279                         return -EADDRNOTAVAIL;
1280         }
1281
1282         if (!data)
1283                 return 0;
1284
1285         if (data[IFLA_MACVLAN_FLAGS] &&
1286             nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC)
1287                 return -EINVAL;
1288
1289         if (data[IFLA_MACVLAN_MODE]) {
1290                 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) {
1291                 case MACVLAN_MODE_PRIVATE:
1292                 case MACVLAN_MODE_VEPA:
1293                 case MACVLAN_MODE_BRIDGE:
1294                 case MACVLAN_MODE_PASSTHRU:
1295                 case MACVLAN_MODE_SOURCE:
1296                         break;
1297                 default:
1298                         return -EINVAL;
1299                 }
1300         }
1301
1302         if (data[IFLA_MACVLAN_MACADDR_MODE]) {
1303                 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) {
1304                 case MACVLAN_MACADDR_ADD:
1305                 case MACVLAN_MACADDR_DEL:
1306                 case MACVLAN_MACADDR_FLUSH:
1307                 case MACVLAN_MACADDR_SET:
1308                         break;
1309                 default:
1310                         return -EINVAL;
1311                 }
1312         }
1313
1314         if (data[IFLA_MACVLAN_MACADDR]) {
1315                 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN)
1316                         return -EINVAL;
1317
1318                 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR])))
1319                         return -EADDRNOTAVAIL;
1320         }
1321
1322         if (data[IFLA_MACVLAN_MACADDR_COUNT])
1323                 return -EINVAL;
1324
1325         return 0;
1326 }
1327
1328 /**
1329  * reconfigure list of remote source mac address
1330  * (only for macvlan devices in source mode)
1331  * Note regarding alignment: all netlink data is aligned to 4 Byte, which
1332  * suffices for both ether_addr_copy and ether_addr_equal_64bits usage.
1333  */
1334 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode,
1335                                       struct nlattr *data[])
1336 {
1337         char *addr = NULL;
1338         int ret, rem, len;
1339         struct nlattr *nla, *head;
1340         struct macvlan_source_entry *entry;
1341
1342         if (data[IFLA_MACVLAN_MACADDR])
1343                 addr = nla_data(data[IFLA_MACVLAN_MACADDR]);
1344
1345         if (mode == MACVLAN_MACADDR_ADD) {
1346                 if (!addr)
1347                         return -EINVAL;
1348
1349                 return macvlan_hash_add_source(vlan, addr);
1350
1351         } else if (mode == MACVLAN_MACADDR_DEL) {
1352                 if (!addr)
1353                         return -EINVAL;
1354
1355                 entry = macvlan_hash_lookup_source(vlan, addr);
1356                 if (entry) {
1357                         macvlan_hash_del_source(entry);
1358                         vlan->macaddr_count--;
1359                 }
1360         } else if (mode == MACVLAN_MACADDR_FLUSH) {
1361                 macvlan_flush_sources(vlan->port, vlan);
1362         } else if (mode == MACVLAN_MACADDR_SET) {
1363                 macvlan_flush_sources(vlan->port, vlan);
1364
1365                 if (addr) {
1366                         ret = macvlan_hash_add_source(vlan, addr);
1367                         if (ret)
1368                                 return ret;
1369                 }
1370
1371                 if (!data || !data[IFLA_MACVLAN_MACADDR_DATA])
1372                         return 0;
1373
1374                 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]);
1375                 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]);
1376
1377                 nla_for_each_attr(nla, head, len, rem) {
1378                         if (nla_type(nla) != IFLA_MACVLAN_MACADDR ||
1379                             nla_len(nla) != ETH_ALEN)
1380                                 continue;
1381
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         vlan->nest_level = dev_get_nest_level(lowerdev) + 1;
1452
1453         vlan->mode     = MACVLAN_MODE_VEPA;
1454         if (data && data[IFLA_MACVLAN_MODE])
1455                 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1456
1457         if (data && data[IFLA_MACVLAN_FLAGS])
1458                 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1459
1460         if (vlan->mode == MACVLAN_MODE_PASSTHRU) {
1461                 if (port->count) {
1462                         err = -EINVAL;
1463                         goto destroy_macvlan_port;
1464                 }
1465                 macvlan_set_passthru(port);
1466                 eth_hw_addr_inherit(dev, lowerdev);
1467         }
1468
1469         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1470                 if (vlan->mode != MACVLAN_MODE_SOURCE) {
1471                         err = -EINVAL;
1472                         goto destroy_macvlan_port;
1473                 }
1474                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1475                 err = macvlan_changelink_sources(vlan, macmode, data);
1476                 if (err)
1477                         goto destroy_macvlan_port;
1478         }
1479
1480         err = register_netdevice(dev);
1481         if (err < 0)
1482                 goto destroy_macvlan_port;
1483
1484         dev->priv_flags |= IFF_MACVLAN;
1485         err = netdev_upper_dev_link(lowerdev, dev, extack);
1486         if (err)
1487                 goto unregister_netdev;
1488
1489         list_add_tail_rcu(&vlan->list, &port->vlans);
1490         netif_stacked_transfer_operstate(lowerdev, dev);
1491         linkwatch_fire_event(dev);
1492
1493         return 0;
1494
1495 unregister_netdev:
1496         /* macvlan_uninit would free the macvlan port */
1497         unregister_netdevice(dev);
1498         return err;
1499 destroy_macvlan_port:
1500         /* the macvlan port may be freed by macvlan_uninit when fail to register.
1501          * so we destroy the macvlan port only when it's valid.
1502          */
1503         if (create && macvlan_port_get_rtnl(lowerdev))
1504                 macvlan_port_destroy(port->dev);
1505         return err;
1506 }
1507 EXPORT_SYMBOL_GPL(macvlan_common_newlink);
1508
1509 static int macvlan_newlink(struct net *src_net, struct net_device *dev,
1510                            struct nlattr *tb[], struct nlattr *data[],
1511                            struct netlink_ext_ack *extack)
1512 {
1513         return macvlan_common_newlink(src_net, dev, tb, data, extack);
1514 }
1515
1516 void macvlan_dellink(struct net_device *dev, struct list_head *head)
1517 {
1518         struct macvlan_dev *vlan = netdev_priv(dev);
1519
1520         if (vlan->mode == MACVLAN_MODE_SOURCE)
1521                 macvlan_flush_sources(vlan->port, vlan);
1522         list_del_rcu(&vlan->list);
1523         unregister_netdevice_queue(dev, head);
1524         netdev_upper_dev_unlink(vlan->lowerdev, dev);
1525 }
1526 EXPORT_SYMBOL_GPL(macvlan_dellink);
1527
1528 static int macvlan_changelink(struct net_device *dev,
1529                               struct nlattr *tb[], struct nlattr *data[],
1530                               struct netlink_ext_ack *extack)
1531 {
1532         struct macvlan_dev *vlan = netdev_priv(dev);
1533         enum macvlan_mode mode;
1534         bool set_mode = false;
1535         enum macvlan_macaddr_mode macmode;
1536         int ret;
1537
1538         /* Validate mode, but don't set yet: setting flags may fail. */
1539         if (data && data[IFLA_MACVLAN_MODE]) {
1540                 set_mode = true;
1541                 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]);
1542                 /* Passthrough mode can't be set or cleared dynamically */
1543                 if ((mode == MACVLAN_MODE_PASSTHRU) !=
1544                     (vlan->mode == MACVLAN_MODE_PASSTHRU))
1545                         return -EINVAL;
1546                 if (vlan->mode == MACVLAN_MODE_SOURCE &&
1547                     vlan->mode != mode)
1548                         macvlan_flush_sources(vlan->port, vlan);
1549         }
1550
1551         if (data && data[IFLA_MACVLAN_FLAGS]) {
1552                 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]);
1553                 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC;
1554                 if (macvlan_passthru(vlan->port) && promisc) {
1555                         int err;
1556
1557                         if (flags & MACVLAN_FLAG_NOPROMISC)
1558                                 err = dev_set_promiscuity(vlan->lowerdev, -1);
1559                         else
1560                                 err = dev_set_promiscuity(vlan->lowerdev, 1);
1561                         if (err < 0)
1562                                 return err;
1563                 }
1564                 vlan->flags = flags;
1565         }
1566         if (set_mode)
1567                 vlan->mode = mode;
1568         if (data && data[IFLA_MACVLAN_MACADDR_MODE]) {
1569                 if (vlan->mode != MACVLAN_MODE_SOURCE)
1570                         return -EINVAL;
1571                 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]);
1572                 ret = macvlan_changelink_sources(vlan, macmode, data);
1573                 if (ret)
1574                         return ret;
1575         }
1576         return 0;
1577 }
1578
1579 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan)
1580 {
1581         if (vlan->macaddr_count == 0)
1582                 return 0;
1583         return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */
1584                 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN);
1585 }
1586
1587 static size_t macvlan_get_size(const struct net_device *dev)
1588 {
1589         struct macvlan_dev *vlan = netdev_priv(dev);
1590
1591         return (0
1592                 + nla_total_size(4) /* IFLA_MACVLAN_MODE */
1593                 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */
1594                 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */
1595                 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */
1596                 );
1597 }
1598
1599 static int macvlan_fill_info_macaddr(struct sk_buff *skb,
1600                                      const struct macvlan_dev *vlan,
1601                                      const int i)
1602 {
1603         struct hlist_head *h = &vlan->port->vlan_source_hash[i];
1604         struct macvlan_source_entry *entry;
1605
1606         hlist_for_each_entry_rcu(entry, h, hlist) {
1607                 if (entry->vlan != vlan)
1608                         continue;
1609                 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr))
1610                         return 1;
1611         }
1612         return 0;
1613 }
1614
1615 static int macvlan_fill_info(struct sk_buff *skb,
1616                                 const struct net_device *dev)
1617 {
1618         struct macvlan_dev *vlan = netdev_priv(dev);
1619         int i;
1620         struct nlattr *nest;
1621
1622         if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode))
1623                 goto nla_put_failure;
1624         if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags))
1625                 goto nla_put_failure;
1626         if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count))
1627                 goto nla_put_failure;
1628         if (vlan->macaddr_count > 0) {
1629                 nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA);
1630                 if (nest == NULL)
1631                         goto nla_put_failure;
1632
1633                 for (i = 0; i < MACVLAN_HASH_SIZE; i++) {
1634                         if (macvlan_fill_info_macaddr(skb, vlan, i))
1635                                 goto nla_put_failure;
1636                 }
1637                 nla_nest_end(skb, nest);
1638         }
1639         return 0;
1640
1641 nla_put_failure:
1642         return -EMSGSIZE;
1643 }
1644
1645 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = {
1646         [IFLA_MACVLAN_MODE]  = { .type = NLA_U32 },
1647         [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 },
1648         [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 },
1649         [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
1650         [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED },
1651         [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 },
1652 };
1653
1654 int macvlan_link_register(struct rtnl_link_ops *ops)
1655 {
1656         /* common fields */
1657         ops->validate           = macvlan_validate;
1658         ops->maxtype            = IFLA_MACVLAN_MAX;
1659         ops->policy             = macvlan_policy;
1660         ops->changelink         = macvlan_changelink;
1661         ops->get_size           = macvlan_get_size;
1662         ops->fill_info          = macvlan_fill_info;
1663
1664         return rtnl_link_register(ops);
1665 };
1666 EXPORT_SYMBOL_GPL(macvlan_link_register);
1667
1668 static struct net *macvlan_get_link_net(const struct net_device *dev)
1669 {
1670         return dev_net(macvlan_dev_real_dev(dev));
1671 }
1672
1673 static struct rtnl_link_ops macvlan_link_ops = {
1674         .kind           = "macvlan",
1675         .setup          = macvlan_setup,
1676         .newlink        = macvlan_newlink,
1677         .dellink        = macvlan_dellink,
1678         .get_link_net   = macvlan_get_link_net,
1679         .priv_size      = sizeof(struct macvlan_dev),
1680 };
1681
1682 static int macvlan_device_event(struct notifier_block *unused,
1683                                 unsigned long event, void *ptr)
1684 {
1685         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1686         struct macvlan_dev *vlan, *next;
1687         struct macvlan_port *port;
1688         LIST_HEAD(list_kill);
1689
1690         if (!netif_is_macvlan_port(dev))
1691                 return NOTIFY_DONE;
1692
1693         port = macvlan_port_get_rtnl(dev);
1694
1695         switch (event) {
1696         case NETDEV_UP:
1697         case NETDEV_DOWN:
1698         case NETDEV_CHANGE:
1699                 list_for_each_entry(vlan, &port->vlans, list)
1700                         netif_stacked_transfer_operstate(vlan->lowerdev,
1701                                                          vlan->dev);
1702                 break;
1703         case NETDEV_FEAT_CHANGE:
1704                 list_for_each_entry(vlan, &port->vlans, list) {
1705                         vlan->dev->gso_max_size = dev->gso_max_size;
1706                         vlan->dev->gso_max_segs = dev->gso_max_segs;
1707                         netdev_update_features(vlan->dev);
1708                 }
1709                 break;
1710         case NETDEV_CHANGEMTU:
1711                 list_for_each_entry(vlan, &port->vlans, list) {
1712                         if (vlan->dev->mtu <= dev->mtu)
1713                                 continue;
1714                         dev_set_mtu(vlan->dev, dev->mtu);
1715                 }
1716                 break;
1717         case NETDEV_CHANGEADDR:
1718                 if (!macvlan_passthru(port))
1719                         return NOTIFY_DONE;
1720
1721                 vlan = list_first_entry_or_null(&port->vlans,
1722                                                 struct macvlan_dev,
1723                                                 list);
1724
1725                 if (macvlan_sync_address(vlan->dev, dev->dev_addr))
1726                         return NOTIFY_BAD;
1727
1728                 break;
1729         case NETDEV_UNREGISTER:
1730                 /* twiddle thumbs on netns device moves */
1731                 if (dev->reg_state != NETREG_UNREGISTERING)
1732                         break;
1733
1734                 list_for_each_entry_safe(vlan, next, &port->vlans, list)
1735                         vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill);
1736                 unregister_netdevice_many(&list_kill);
1737                 break;
1738         case NETDEV_PRE_TYPE_CHANGE:
1739                 /* Forbid underlaying device to change its type. */
1740                 return NOTIFY_BAD;
1741
1742         case NETDEV_NOTIFY_PEERS:
1743         case NETDEV_BONDING_FAILOVER:
1744         case NETDEV_RESEND_IGMP:
1745                 /* Propagate to all vlans */
1746                 list_for_each_entry(vlan, &port->vlans, list)
1747                         call_netdevice_notifiers(event, vlan->dev);
1748         }
1749         return NOTIFY_DONE;
1750 }
1751
1752 static struct notifier_block macvlan_notifier_block __read_mostly = {
1753         .notifier_call  = macvlan_device_event,
1754 };
1755
1756 static int __init macvlan_init_module(void)
1757 {
1758         int err;
1759
1760         register_netdevice_notifier(&macvlan_notifier_block);
1761
1762         err = macvlan_link_register(&macvlan_link_ops);
1763         if (err < 0)
1764                 goto err1;
1765         return 0;
1766 err1:
1767         unregister_netdevice_notifier(&macvlan_notifier_block);
1768         return err;
1769 }
1770
1771 static void __exit macvlan_cleanup_module(void)
1772 {
1773         rtnl_link_unregister(&macvlan_link_ops);
1774         unregister_netdevice_notifier(&macvlan_notifier_block);
1775 }
1776
1777 module_init(macvlan_init_module);
1778 module_exit(macvlan_cleanup_module);
1779
1780 MODULE_LICENSE("GPL");
1781 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1782 MODULE_DESCRIPTION("Driver for MAC address based VLANs");
1783 MODULE_ALIAS_RTNL_LINK("macvlan");