macvlan.c revision 8b4703e9bd1172a5f8244276ebb94302e6153e26
1/* 2 * Copyright (c) 2007 Patrick McHardy <kaber@trash.net> 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License as 6 * published by the Free Software Foundation; either version 2 of 7 * the License, or (at your option) any later version. 8 * 9 * The code this is based on carried the following copyright notice: 10 * --- 11 * (C) Copyright 2001-2006 12 * Alex Zeffertt, Cambridge Broadband Ltd, ajz@cambridgebroadband.com 13 * Re-worked by Ben Greear <greearb@candelatech.com> 14 * --- 15 */ 16#include <linux/kernel.h> 17#include <linux/types.h> 18#include <linux/module.h> 19#include <linux/init.h> 20#include <linux/errno.h> 21#include <linux/slab.h> 22#include <linux/string.h> 23#include <linux/rculist.h> 24#include <linux/notifier.h> 25#include <linux/netdevice.h> 26#include <linux/etherdevice.h> 27#include <linux/ethtool.h> 28#include <linux/if_arp.h> 29#include <linux/if_vlan.h> 30#include <linux/if_link.h> 31#include <linux/if_macvlan.h> 32#include <linux/hash.h> 33#include <net/rtnetlink.h> 34#include <net/xfrm.h> 35 36#define MACVLAN_HASH_SIZE (1 << BITS_PER_BYTE) 37 38struct macvlan_port { 39 struct net_device *dev; 40 struct hlist_head vlan_hash[MACVLAN_HASH_SIZE]; 41 struct list_head vlans; 42 struct rcu_head rcu; 43 bool passthru; 44 int count; 45}; 46 47static void macvlan_port_destroy(struct net_device *dev); 48 49static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev) 50{ 51 return rcu_dereference(dev->rx_handler_data); 52} 53 54static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev) 55{ 56 return rtnl_dereference(dev->rx_handler_data); 57} 58 59#define macvlan_port_exists(dev) (dev->priv_flags & IFF_MACVLAN_PORT) 60 61static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port, 62 const unsigned char *addr) 63{ 64 struct macvlan_dev *vlan; 65 66 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[addr[5]], hlist) { 67 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr)) 68 return vlan; 69 } 70 return NULL; 71} 72 73static void macvlan_hash_add(struct macvlan_dev *vlan) 74{ 75 struct macvlan_port *port = vlan->port; 76 const unsigned char *addr = vlan->dev->dev_addr; 77 78 hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[addr[5]]); 79} 80 81static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync) 82{ 83 hlist_del_rcu(&vlan->hlist); 84 if (sync) 85 synchronize_rcu(); 86} 87 88static void macvlan_hash_change_addr(struct macvlan_dev *vlan, 89 const unsigned char *addr) 90{ 91 macvlan_hash_del(vlan, true); 92 /* Now that we are unhashed it is safe to change the device 93 * address without confusing packet delivery. 94 */ 95 memcpy(vlan->dev->dev_addr, addr, ETH_ALEN); 96 macvlan_hash_add(vlan); 97} 98 99static int macvlan_addr_busy(const struct macvlan_port *port, 100 const unsigned char *addr) 101{ 102 /* Test to see if the specified multicast address is 103 * currently in use by the underlying device or 104 * another macvlan. 105 */ 106 if (ether_addr_equal_64bits(port->dev->dev_addr, addr)) 107 return 1; 108 109 if (macvlan_hash_lookup(port, addr)) 110 return 1; 111 112 return 0; 113} 114 115 116static int macvlan_broadcast_one(struct sk_buff *skb, 117 const struct macvlan_dev *vlan, 118 const struct ethhdr *eth, bool local) 119{ 120 struct net_device *dev = vlan->dev; 121 122 if (local) 123 return dev_forward_skb(dev, skb); 124 125 skb->dev = dev; 126 if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast)) 127 skb->pkt_type = PACKET_BROADCAST; 128 else 129 skb->pkt_type = PACKET_MULTICAST; 130 131 return netif_rx(skb); 132} 133 134static u32 macvlan_hash_mix(const struct macvlan_dev *vlan) 135{ 136 return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT); 137} 138 139 140static unsigned int mc_hash(const struct macvlan_dev *vlan, 141 const unsigned char *addr) 142{ 143 u32 val = __get_unaligned_cpu32(addr + 2); 144 145 val ^= macvlan_hash_mix(vlan); 146 return hash_32(val, MACVLAN_MC_FILTER_BITS); 147} 148 149static void macvlan_broadcast(struct sk_buff *skb, 150 const struct macvlan_port *port, 151 struct net_device *src, 152 enum macvlan_mode mode) 153{ 154 const struct ethhdr *eth = eth_hdr(skb); 155 const struct macvlan_dev *vlan; 156 struct sk_buff *nskb; 157 unsigned int i; 158 int err; 159 unsigned int hash; 160 161 if (skb->protocol == htons(ETH_P_PAUSE)) 162 return; 163 164 for (i = 0; i < MACVLAN_HASH_SIZE; i++) { 165 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[i], hlist) { 166 if (vlan->dev == src || !(vlan->mode & mode)) 167 continue; 168 169 hash = mc_hash(vlan, eth->h_dest); 170 if (!test_bit(hash, vlan->mc_filter)) 171 continue; 172 173 err = NET_RX_DROP; 174 nskb = skb_clone(skb, GFP_ATOMIC); 175 if (likely(nskb)) 176 err = macvlan_broadcast_one( 177 nskb, vlan, eth, 178 mode == MACVLAN_MODE_BRIDGE); 179 macvlan_count_rx(vlan, skb->len + ETH_HLEN, 180 err == NET_RX_SUCCESS, 1); 181 } 182 } 183} 184 185/* called under rcu_read_lock() from netif_receive_skb */ 186static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb) 187{ 188 struct macvlan_port *port; 189 struct sk_buff *skb = *pskb; 190 const struct ethhdr *eth = eth_hdr(skb); 191 const struct macvlan_dev *vlan; 192 const struct macvlan_dev *src; 193 struct net_device *dev; 194 unsigned int len = 0; 195 int ret = NET_RX_DROP; 196 197 port = macvlan_port_get_rcu(skb->dev); 198 if (is_multicast_ether_addr(eth->h_dest)) { 199 skb = ip_check_defrag(skb, IP_DEFRAG_MACVLAN); 200 if (!skb) 201 return RX_HANDLER_CONSUMED; 202 eth = eth_hdr(skb); 203 src = macvlan_hash_lookup(port, eth->h_source); 204 if (!src) 205 /* frame comes from an external address */ 206 macvlan_broadcast(skb, port, NULL, 207 MACVLAN_MODE_PRIVATE | 208 MACVLAN_MODE_VEPA | 209 MACVLAN_MODE_PASSTHRU| 210 MACVLAN_MODE_BRIDGE); 211 else if (src->mode == MACVLAN_MODE_VEPA) 212 /* flood to everyone except source */ 213 macvlan_broadcast(skb, port, src->dev, 214 MACVLAN_MODE_VEPA | 215 MACVLAN_MODE_BRIDGE); 216 else if (src->mode == MACVLAN_MODE_BRIDGE) 217 /* 218 * flood only to VEPA ports, bridge ports 219 * already saw the frame on the way out. 220 */ 221 macvlan_broadcast(skb, port, src->dev, 222 MACVLAN_MODE_VEPA); 223 else { 224 /* forward to original port. */ 225 vlan = src; 226 ret = macvlan_broadcast_one(skb, vlan, eth, 0); 227 goto out; 228 } 229 230 return RX_HANDLER_PASS; 231 } 232 233 if (port->passthru) 234 vlan = list_first_or_null_rcu(&port->vlans, 235 struct macvlan_dev, list); 236 else 237 vlan = macvlan_hash_lookup(port, eth->h_dest); 238 if (vlan == NULL) 239 return RX_HANDLER_PASS; 240 241 dev = vlan->dev; 242 if (unlikely(!(dev->flags & IFF_UP))) { 243 kfree_skb(skb); 244 return RX_HANDLER_CONSUMED; 245 } 246 len = skb->len + ETH_HLEN; 247 skb = skb_share_check(skb, GFP_ATOMIC); 248 if (!skb) 249 goto out; 250 251 skb->dev = dev; 252 skb->pkt_type = PACKET_HOST; 253 254 ret = netif_rx(skb); 255 256out: 257 macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, 0); 258 return RX_HANDLER_CONSUMED; 259} 260 261static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev) 262{ 263 const struct macvlan_dev *vlan = netdev_priv(dev); 264 const struct macvlan_port *port = vlan->port; 265 const struct macvlan_dev *dest; 266 __u8 ip_summed = skb->ip_summed; 267 268 if (vlan->mode == MACVLAN_MODE_BRIDGE) { 269 const struct ethhdr *eth = (void *)skb->data; 270 skb->ip_summed = CHECKSUM_UNNECESSARY; 271 272 /* send to other bridge ports directly */ 273 if (is_multicast_ether_addr(eth->h_dest)) { 274 macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE); 275 goto xmit_world; 276 } 277 278 dest = macvlan_hash_lookup(port, eth->h_dest); 279 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) { 280 /* send to lowerdev first for its network taps */ 281 dev_forward_skb(vlan->lowerdev, skb); 282 283 return NET_XMIT_SUCCESS; 284 } 285 } 286 287xmit_world: 288 skb->ip_summed = ip_summed; 289 skb->dev = vlan->lowerdev; 290 return dev_queue_xmit(skb); 291} 292 293static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb, 294 struct net_device *dev) 295{ 296 unsigned int len = skb->len; 297 int ret; 298 const struct macvlan_dev *vlan = netdev_priv(dev); 299 300 if (vlan->fwd_priv) { 301 skb->dev = vlan->lowerdev; 302 ret = dev_queue_xmit_accel(skb, vlan->fwd_priv); 303 } else { 304 ret = macvlan_queue_xmit(skb, dev); 305 } 306 307 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 308 struct vlan_pcpu_stats *pcpu_stats; 309 310 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats); 311 u64_stats_update_begin(&pcpu_stats->syncp); 312 pcpu_stats->tx_packets++; 313 pcpu_stats->tx_bytes += len; 314 u64_stats_update_end(&pcpu_stats->syncp); 315 } else { 316 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 317 } 318 return ret; 319} 320 321static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev, 322 unsigned short type, const void *daddr, 323 const void *saddr, unsigned len) 324{ 325 const struct macvlan_dev *vlan = netdev_priv(dev); 326 struct net_device *lowerdev = vlan->lowerdev; 327 328 return dev_hard_header(skb, lowerdev, type, daddr, 329 saddr ? : dev->dev_addr, len); 330} 331 332static const struct header_ops macvlan_hard_header_ops = { 333 .create = macvlan_hard_header, 334 .rebuild = eth_rebuild_header, 335 .parse = eth_header_parse, 336 .cache = eth_header_cache, 337 .cache_update = eth_header_cache_update, 338}; 339 340static struct rtnl_link_ops macvlan_link_ops; 341 342static int macvlan_open(struct net_device *dev) 343{ 344 struct macvlan_dev *vlan = netdev_priv(dev); 345 struct net_device *lowerdev = vlan->lowerdev; 346 int err; 347 348 if (vlan->port->passthru) { 349 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) { 350 err = dev_set_promiscuity(lowerdev, 1); 351 if (err < 0) 352 goto out; 353 } 354 goto hash_add; 355 } 356 357 if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD && 358 dev->rtnl_link_ops == &macvlan_link_ops) { 359 vlan->fwd_priv = 360 lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev); 361 362 /* If we get a NULL pointer back, or if we get an error 363 * then we should just fall through to the non accelerated path 364 */ 365 if (IS_ERR_OR_NULL(vlan->fwd_priv)) { 366 vlan->fwd_priv = NULL; 367 } else 368 return 0; 369 } 370 371 err = -EBUSY; 372 if (macvlan_addr_busy(vlan->port, dev->dev_addr)) 373 goto out; 374 375 err = dev_uc_add(lowerdev, dev->dev_addr); 376 if (err < 0) 377 goto out; 378 if (dev->flags & IFF_ALLMULTI) { 379 err = dev_set_allmulti(lowerdev, 1); 380 if (err < 0) 381 goto del_unicast; 382 } 383 384hash_add: 385 macvlan_hash_add(vlan); 386 return 0; 387 388del_unicast: 389 dev_uc_del(lowerdev, dev->dev_addr); 390out: 391 if (vlan->fwd_priv) { 392 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev, 393 vlan->fwd_priv); 394 vlan->fwd_priv = NULL; 395 } 396 return err; 397} 398 399static int macvlan_stop(struct net_device *dev) 400{ 401 struct macvlan_dev *vlan = netdev_priv(dev); 402 struct net_device *lowerdev = vlan->lowerdev; 403 404 if (vlan->fwd_priv) { 405 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev, 406 vlan->fwd_priv); 407 vlan->fwd_priv = NULL; 408 return 0; 409 } 410 411 dev_uc_unsync(lowerdev, dev); 412 dev_mc_unsync(lowerdev, dev); 413 414 if (vlan->port->passthru) { 415 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) 416 dev_set_promiscuity(lowerdev, -1); 417 goto hash_del; 418 } 419 420 if (dev->flags & IFF_ALLMULTI) 421 dev_set_allmulti(lowerdev, -1); 422 423 dev_uc_del(lowerdev, dev->dev_addr); 424 425hash_del: 426 macvlan_hash_del(vlan, !dev->dismantle); 427 return 0; 428} 429 430static int macvlan_set_mac_address(struct net_device *dev, void *p) 431{ 432 struct macvlan_dev *vlan = netdev_priv(dev); 433 struct net_device *lowerdev = vlan->lowerdev; 434 struct sockaddr *addr = p; 435 int err; 436 437 if (!is_valid_ether_addr(addr->sa_data)) 438 return -EADDRNOTAVAIL; 439 440 if (!(dev->flags & IFF_UP)) { 441 /* Just copy in the new address */ 442 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN); 443 } else { 444 /* Rehash and update the device filters */ 445 if (macvlan_addr_busy(vlan->port, addr->sa_data)) 446 return -EBUSY; 447 448 err = dev_uc_add(lowerdev, addr->sa_data); 449 if (err) 450 return err; 451 452 dev_uc_del(lowerdev, dev->dev_addr); 453 454 macvlan_hash_change_addr(vlan, addr->sa_data); 455 } 456 return 0; 457} 458 459static void macvlan_change_rx_flags(struct net_device *dev, int change) 460{ 461 struct macvlan_dev *vlan = netdev_priv(dev); 462 struct net_device *lowerdev = vlan->lowerdev; 463 464 if (change & IFF_ALLMULTI) 465 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1); 466} 467 468static void macvlan_set_mac_lists(struct net_device *dev) 469{ 470 struct macvlan_dev *vlan = netdev_priv(dev); 471 472 if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) { 473 bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ); 474 } else { 475 struct netdev_hw_addr *ha; 476 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ); 477 478 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ); 479 netdev_for_each_mc_addr(ha, dev) { 480 __set_bit(mc_hash(vlan, ha->addr), filter); 481 } 482 483 __set_bit(mc_hash(vlan, dev->broadcast), filter); 484 485 bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ); 486 } 487 dev_uc_sync(vlan->lowerdev, dev); 488 dev_mc_sync(vlan->lowerdev, dev); 489} 490 491static int macvlan_change_mtu(struct net_device *dev, int new_mtu) 492{ 493 struct macvlan_dev *vlan = netdev_priv(dev); 494 495 if (new_mtu < 68 || vlan->lowerdev->mtu < new_mtu) 496 return -EINVAL; 497 dev->mtu = new_mtu; 498 return 0; 499} 500 501/* 502 * macvlan network devices have devices nesting below it and are a special 503 * "super class" of normal network devices; split their locks off into a 504 * separate class since they always nest. 505 */ 506static struct lock_class_key macvlan_netdev_xmit_lock_key; 507static struct lock_class_key macvlan_netdev_addr_lock_key; 508 509#define ALWAYS_ON_FEATURES \ 510 (NETIF_F_SG | NETIF_F_GEN_CSUM | NETIF_F_GSO_SOFTWARE | NETIF_F_LLTX) 511 512#define MACVLAN_FEATURES \ 513 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \ 514 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_UFO | NETIF_F_GSO_ROBUST | \ 515 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \ 516 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER) 517 518#define MACVLAN_STATE_MASK \ 519 ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT)) 520 521static void macvlan_set_lockdep_class_one(struct net_device *dev, 522 struct netdev_queue *txq, 523 void *_unused) 524{ 525 lockdep_set_class(&txq->_xmit_lock, 526 &macvlan_netdev_xmit_lock_key); 527} 528 529static void macvlan_set_lockdep_class(struct net_device *dev) 530{ 531 lockdep_set_class(&dev->addr_list_lock, 532 &macvlan_netdev_addr_lock_key); 533 netdev_for_each_tx_queue(dev, macvlan_set_lockdep_class_one, NULL); 534} 535 536static int macvlan_init(struct net_device *dev) 537{ 538 struct macvlan_dev *vlan = netdev_priv(dev); 539 const struct net_device *lowerdev = vlan->lowerdev; 540 int i; 541 542 dev->state = (dev->state & ~MACVLAN_STATE_MASK) | 543 (lowerdev->state & MACVLAN_STATE_MASK); 544 dev->features = lowerdev->features & MACVLAN_FEATURES; 545 dev->features |= ALWAYS_ON_FEATURES; 546 dev->gso_max_size = lowerdev->gso_max_size; 547 dev->iflink = lowerdev->ifindex; 548 dev->hard_header_len = lowerdev->hard_header_len; 549 550 macvlan_set_lockdep_class(dev); 551 552 vlan->pcpu_stats = alloc_percpu(struct vlan_pcpu_stats); 553 if (!vlan->pcpu_stats) 554 return -ENOMEM; 555 556 for_each_possible_cpu(i) { 557 struct vlan_pcpu_stats *mvlstats; 558 mvlstats = per_cpu_ptr(vlan->pcpu_stats, i); 559 u64_stats_init(&mvlstats->syncp); 560 } 561 562 return 0; 563} 564 565static void macvlan_uninit(struct net_device *dev) 566{ 567 struct macvlan_dev *vlan = netdev_priv(dev); 568 struct macvlan_port *port = vlan->port; 569 570 free_percpu(vlan->pcpu_stats); 571 572 port->count -= 1; 573 if (!port->count) 574 macvlan_port_destroy(port->dev); 575} 576 577static struct rtnl_link_stats64 *macvlan_dev_get_stats64(struct net_device *dev, 578 struct rtnl_link_stats64 *stats) 579{ 580 struct macvlan_dev *vlan = netdev_priv(dev); 581 582 if (vlan->pcpu_stats) { 583 struct vlan_pcpu_stats *p; 584 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes; 585 u32 rx_errors = 0, tx_dropped = 0; 586 unsigned int start; 587 int i; 588 589 for_each_possible_cpu(i) { 590 p = per_cpu_ptr(vlan->pcpu_stats, i); 591 do { 592 start = u64_stats_fetch_begin_bh(&p->syncp); 593 rx_packets = p->rx_packets; 594 rx_bytes = p->rx_bytes; 595 rx_multicast = p->rx_multicast; 596 tx_packets = p->tx_packets; 597 tx_bytes = p->tx_bytes; 598 } while (u64_stats_fetch_retry_bh(&p->syncp, start)); 599 600 stats->rx_packets += rx_packets; 601 stats->rx_bytes += rx_bytes; 602 stats->multicast += rx_multicast; 603 stats->tx_packets += tx_packets; 604 stats->tx_bytes += tx_bytes; 605 /* rx_errors & tx_dropped are u32, updated 606 * without syncp protection. 607 */ 608 rx_errors += p->rx_errors; 609 tx_dropped += p->tx_dropped; 610 } 611 stats->rx_errors = rx_errors; 612 stats->rx_dropped = rx_errors; 613 stats->tx_dropped = tx_dropped; 614 } 615 return stats; 616} 617 618static int macvlan_vlan_rx_add_vid(struct net_device *dev, 619 __be16 proto, u16 vid) 620{ 621 struct macvlan_dev *vlan = netdev_priv(dev); 622 struct net_device *lowerdev = vlan->lowerdev; 623 624 return vlan_vid_add(lowerdev, proto, vid); 625} 626 627static int macvlan_vlan_rx_kill_vid(struct net_device *dev, 628 __be16 proto, u16 vid) 629{ 630 struct macvlan_dev *vlan = netdev_priv(dev); 631 struct net_device *lowerdev = vlan->lowerdev; 632 633 vlan_vid_del(lowerdev, proto, vid); 634 return 0; 635} 636 637static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 638 struct net_device *dev, 639 const unsigned char *addr, 640 u16 flags) 641{ 642 struct macvlan_dev *vlan = netdev_priv(dev); 643 int err = -EINVAL; 644 645 if (!vlan->port->passthru) 646 return -EOPNOTSUPP; 647 648 if (flags & NLM_F_REPLACE) 649 return -EOPNOTSUPP; 650 651 if (is_unicast_ether_addr(addr)) 652 err = dev_uc_add_excl(dev, addr); 653 else if (is_multicast_ether_addr(addr)) 654 err = dev_mc_add_excl(dev, addr); 655 656 return err; 657} 658 659static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[], 660 struct net_device *dev, 661 const unsigned char *addr) 662{ 663 struct macvlan_dev *vlan = netdev_priv(dev); 664 int err = -EINVAL; 665 666 if (!vlan->port->passthru) 667 return -EOPNOTSUPP; 668 669 if (is_unicast_ether_addr(addr)) 670 err = dev_uc_del(dev, addr); 671 else if (is_multicast_ether_addr(addr)) 672 err = dev_mc_del(dev, addr); 673 674 return err; 675} 676 677static void macvlan_ethtool_get_drvinfo(struct net_device *dev, 678 struct ethtool_drvinfo *drvinfo) 679{ 680 strlcpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver)); 681 strlcpy(drvinfo->version, "0.1", sizeof(drvinfo->version)); 682} 683 684static int macvlan_ethtool_get_settings(struct net_device *dev, 685 struct ethtool_cmd *cmd) 686{ 687 const struct macvlan_dev *vlan = netdev_priv(dev); 688 689 return __ethtool_get_settings(vlan->lowerdev, cmd); 690} 691 692static netdev_features_t macvlan_fix_features(struct net_device *dev, 693 netdev_features_t features) 694{ 695 struct macvlan_dev *vlan = netdev_priv(dev); 696 netdev_features_t mask; 697 698 features |= NETIF_F_ALL_FOR_ALL; 699 features &= (vlan->set_features | ~MACVLAN_FEATURES); 700 mask = features; 701 702 features = netdev_increment_features(vlan->lowerdev->features, 703 features, 704 mask); 705 features |= ALWAYS_ON_FEATURES; 706 707 return features; 708} 709 710static const struct ethtool_ops macvlan_ethtool_ops = { 711 .get_link = ethtool_op_get_link, 712 .get_settings = macvlan_ethtool_get_settings, 713 .get_drvinfo = macvlan_ethtool_get_drvinfo, 714}; 715 716static const struct net_device_ops macvlan_netdev_ops = { 717 .ndo_init = macvlan_init, 718 .ndo_uninit = macvlan_uninit, 719 .ndo_open = macvlan_open, 720 .ndo_stop = macvlan_stop, 721 .ndo_start_xmit = macvlan_start_xmit, 722 .ndo_change_mtu = macvlan_change_mtu, 723 .ndo_fix_features = macvlan_fix_features, 724 .ndo_change_rx_flags = macvlan_change_rx_flags, 725 .ndo_set_mac_address = macvlan_set_mac_address, 726 .ndo_set_rx_mode = macvlan_set_mac_lists, 727 .ndo_get_stats64 = macvlan_dev_get_stats64, 728 .ndo_validate_addr = eth_validate_addr, 729 .ndo_vlan_rx_add_vid = macvlan_vlan_rx_add_vid, 730 .ndo_vlan_rx_kill_vid = macvlan_vlan_rx_kill_vid, 731 .ndo_fdb_add = macvlan_fdb_add, 732 .ndo_fdb_del = macvlan_fdb_del, 733 .ndo_fdb_dump = ndo_dflt_fdb_dump, 734}; 735 736void macvlan_common_setup(struct net_device *dev) 737{ 738 ether_setup(dev); 739 740 dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 741 dev->priv_flags |= IFF_UNICAST_FLT; 742 dev->netdev_ops = &macvlan_netdev_ops; 743 dev->destructor = free_netdev; 744 dev->header_ops = &macvlan_hard_header_ops; 745 dev->ethtool_ops = &macvlan_ethtool_ops; 746} 747EXPORT_SYMBOL_GPL(macvlan_common_setup); 748 749static void macvlan_setup(struct net_device *dev) 750{ 751 macvlan_common_setup(dev); 752 dev->tx_queue_len = 0; 753} 754 755static int macvlan_port_create(struct net_device *dev) 756{ 757 struct macvlan_port *port; 758 unsigned int i; 759 int err; 760 761 if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) 762 return -EINVAL; 763 764 port = kzalloc(sizeof(*port), GFP_KERNEL); 765 if (port == NULL) 766 return -ENOMEM; 767 768 port->passthru = false; 769 port->dev = dev; 770 INIT_LIST_HEAD(&port->vlans); 771 for (i = 0; i < MACVLAN_HASH_SIZE; i++) 772 INIT_HLIST_HEAD(&port->vlan_hash[i]); 773 774 err = netdev_rx_handler_register(dev, macvlan_handle_frame, port); 775 if (err) 776 kfree(port); 777 else 778 dev->priv_flags |= IFF_MACVLAN_PORT; 779 return err; 780} 781 782static void macvlan_port_destroy(struct net_device *dev) 783{ 784 struct macvlan_port *port = macvlan_port_get_rtnl(dev); 785 786 dev->priv_flags &= ~IFF_MACVLAN_PORT; 787 netdev_rx_handler_unregister(dev); 788 kfree_rcu(port, rcu); 789} 790 791static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[]) 792{ 793 if (tb[IFLA_ADDRESS]) { 794 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 795 return -EINVAL; 796 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 797 return -EADDRNOTAVAIL; 798 } 799 800 if (data && data[IFLA_MACVLAN_FLAGS] && 801 nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~MACVLAN_FLAG_NOPROMISC) 802 return -EINVAL; 803 804 if (data && data[IFLA_MACVLAN_MODE]) { 805 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) { 806 case MACVLAN_MODE_PRIVATE: 807 case MACVLAN_MODE_VEPA: 808 case MACVLAN_MODE_BRIDGE: 809 case MACVLAN_MODE_PASSTHRU: 810 break; 811 default: 812 return -EINVAL; 813 } 814 } 815 return 0; 816} 817 818int macvlan_common_newlink(struct net *src_net, struct net_device *dev, 819 struct nlattr *tb[], struct nlattr *data[]) 820{ 821 struct macvlan_dev *vlan = netdev_priv(dev); 822 struct macvlan_port *port; 823 struct net_device *lowerdev; 824 int err; 825 826 if (!tb[IFLA_LINK]) 827 return -EINVAL; 828 829 lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK])); 830 if (lowerdev == NULL) 831 return -ENODEV; 832 833 /* When creating macvlans or macvtaps on top of other macvlans - use 834 * the real device as the lowerdev. 835 */ 836 if (netif_is_macvlan(lowerdev)) 837 lowerdev = macvlan_dev_real_dev(lowerdev); 838 839 if (!tb[IFLA_MTU]) 840 dev->mtu = lowerdev->mtu; 841 else if (dev->mtu > lowerdev->mtu) 842 return -EINVAL; 843 844 if (!tb[IFLA_ADDRESS]) 845 eth_hw_addr_random(dev); 846 847 if (!macvlan_port_exists(lowerdev)) { 848 err = macvlan_port_create(lowerdev); 849 if (err < 0) 850 return err; 851 } 852 port = macvlan_port_get_rtnl(lowerdev); 853 854 /* Only 1 macvlan device can be created in passthru mode */ 855 if (port->passthru) 856 return -EINVAL; 857 858 vlan->lowerdev = lowerdev; 859 vlan->dev = dev; 860 vlan->port = port; 861 vlan->set_features = MACVLAN_FEATURES; 862 863 vlan->mode = MACVLAN_MODE_VEPA; 864 if (data && data[IFLA_MACVLAN_MODE]) 865 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]); 866 867 if (data && data[IFLA_MACVLAN_FLAGS]) 868 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]); 869 870 if (vlan->mode == MACVLAN_MODE_PASSTHRU) { 871 if (port->count) 872 return -EINVAL; 873 port->passthru = true; 874 eth_hw_addr_inherit(dev, lowerdev); 875 } 876 877 port->count += 1; 878 err = register_netdevice(dev); 879 if (err < 0) 880 goto destroy_port; 881 882 dev->priv_flags |= IFF_MACVLAN; 883 err = netdev_upper_dev_link(lowerdev, dev); 884 if (err) 885 goto unregister_netdev; 886 887 list_add_tail_rcu(&vlan->list, &port->vlans); 888 netif_stacked_transfer_operstate(lowerdev, dev); 889 890 return 0; 891 892unregister_netdev: 893 unregister_netdevice(dev); 894destroy_port: 895 port->count -= 1; 896 if (!port->count) 897 macvlan_port_destroy(lowerdev); 898 899 return err; 900} 901EXPORT_SYMBOL_GPL(macvlan_common_newlink); 902 903static int macvlan_newlink(struct net *src_net, struct net_device *dev, 904 struct nlattr *tb[], struct nlattr *data[]) 905{ 906 return macvlan_common_newlink(src_net, dev, tb, data); 907} 908 909void macvlan_dellink(struct net_device *dev, struct list_head *head) 910{ 911 struct macvlan_dev *vlan = netdev_priv(dev); 912 913 list_del_rcu(&vlan->list); 914 unregister_netdevice_queue(dev, head); 915 netdev_upper_dev_unlink(vlan->lowerdev, dev); 916} 917EXPORT_SYMBOL_GPL(macvlan_dellink); 918 919static int macvlan_changelink(struct net_device *dev, 920 struct nlattr *tb[], struct nlattr *data[]) 921{ 922 struct macvlan_dev *vlan = netdev_priv(dev); 923 enum macvlan_mode mode; 924 bool set_mode = false; 925 926 /* Validate mode, but don't set yet: setting flags may fail. */ 927 if (data && data[IFLA_MACVLAN_MODE]) { 928 set_mode = true; 929 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]); 930 /* Passthrough mode can't be set or cleared dynamically */ 931 if ((mode == MACVLAN_MODE_PASSTHRU) != 932 (vlan->mode == MACVLAN_MODE_PASSTHRU)) 933 return -EINVAL; 934 } 935 936 if (data && data[IFLA_MACVLAN_FLAGS]) { 937 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]); 938 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC; 939 if (vlan->port->passthru && promisc) { 940 int err; 941 942 if (flags & MACVLAN_FLAG_NOPROMISC) 943 err = dev_set_promiscuity(vlan->lowerdev, -1); 944 else 945 err = dev_set_promiscuity(vlan->lowerdev, 1); 946 if (err < 0) 947 return err; 948 } 949 vlan->flags = flags; 950 } 951 if (set_mode) 952 vlan->mode = mode; 953 return 0; 954} 955 956static size_t macvlan_get_size(const struct net_device *dev) 957{ 958 return (0 959 + nla_total_size(4) /* IFLA_MACVLAN_MODE */ 960 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */ 961 ); 962} 963 964static int macvlan_fill_info(struct sk_buff *skb, 965 const struct net_device *dev) 966{ 967 struct macvlan_dev *vlan = netdev_priv(dev); 968 969 if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode)) 970 goto nla_put_failure; 971 if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags)) 972 goto nla_put_failure; 973 return 0; 974 975nla_put_failure: 976 return -EMSGSIZE; 977} 978 979static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = { 980 [IFLA_MACVLAN_MODE] = { .type = NLA_U32 }, 981 [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 }, 982}; 983 984int macvlan_link_register(struct rtnl_link_ops *ops) 985{ 986 /* common fields */ 987 ops->priv_size = sizeof(struct macvlan_dev); 988 ops->validate = macvlan_validate; 989 ops->maxtype = IFLA_MACVLAN_MAX; 990 ops->policy = macvlan_policy; 991 ops->changelink = macvlan_changelink; 992 ops->get_size = macvlan_get_size; 993 ops->fill_info = macvlan_fill_info; 994 995 return rtnl_link_register(ops); 996}; 997EXPORT_SYMBOL_GPL(macvlan_link_register); 998 999static struct rtnl_link_ops macvlan_link_ops = { 1000 .kind = "macvlan", 1001 .setup = macvlan_setup, 1002 .newlink = macvlan_newlink, 1003 .dellink = macvlan_dellink, 1004}; 1005 1006static int macvlan_device_event(struct notifier_block *unused, 1007 unsigned long event, void *ptr) 1008{ 1009 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1010 struct macvlan_dev *vlan, *next; 1011 struct macvlan_port *port; 1012 LIST_HEAD(list_kill); 1013 1014 if (!macvlan_port_exists(dev)) 1015 return NOTIFY_DONE; 1016 1017 port = macvlan_port_get_rtnl(dev); 1018 1019 switch (event) { 1020 case NETDEV_CHANGE: 1021 list_for_each_entry(vlan, &port->vlans, list) 1022 netif_stacked_transfer_operstate(vlan->lowerdev, 1023 vlan->dev); 1024 break; 1025 case NETDEV_FEAT_CHANGE: 1026 list_for_each_entry(vlan, &port->vlans, list) { 1027 vlan->dev->gso_max_size = dev->gso_max_size; 1028 netdev_update_features(vlan->dev); 1029 } 1030 break; 1031 case NETDEV_UNREGISTER: 1032 /* twiddle thumbs on netns device moves */ 1033 if (dev->reg_state != NETREG_UNREGISTERING) 1034 break; 1035 1036 list_for_each_entry_safe(vlan, next, &port->vlans, list) 1037 vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill); 1038 unregister_netdevice_many(&list_kill); 1039 list_del(&list_kill); 1040 break; 1041 case NETDEV_PRE_TYPE_CHANGE: 1042 /* Forbid underlaying device to change its type. */ 1043 return NOTIFY_BAD; 1044 } 1045 return NOTIFY_DONE; 1046} 1047 1048static struct notifier_block macvlan_notifier_block __read_mostly = { 1049 .notifier_call = macvlan_device_event, 1050}; 1051 1052static int __init macvlan_init_module(void) 1053{ 1054 int err; 1055 1056 register_netdevice_notifier(&macvlan_notifier_block); 1057 1058 err = macvlan_link_register(&macvlan_link_ops); 1059 if (err < 0) 1060 goto err1; 1061 return 0; 1062err1: 1063 unregister_netdevice_notifier(&macvlan_notifier_block); 1064 return err; 1065} 1066 1067static void __exit macvlan_cleanup_module(void) 1068{ 1069 rtnl_link_unregister(&macvlan_link_ops); 1070 unregister_netdevice_notifier(&macvlan_notifier_block); 1071} 1072 1073module_init(macvlan_init_module); 1074module_exit(macvlan_cleanup_module); 1075 1076MODULE_LICENSE("GPL"); 1077MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); 1078MODULE_DESCRIPTION("Driver for MAC address based VLANs"); 1079MODULE_ALIAS_RTNL_LINK("macvlan"); 1080