drivers/net/macvlan.c

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