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_DEFAULT_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 bc_queue_len_used; 50 int bc_cutoff; 51 u32 flags; 52 int count; 53 struct hlist_head vlan_source_hash[MACVLAN_HASH_SIZE]; 54 DECLARE_BITMAP(bc_filter, MACVLAN_MC_FILTER_SZ); 55 DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ); 56 unsigned char perm_addr[ETH_ALEN]; 57 }; 58 59 struct macvlan_source_entry { 60 struct hlist_node hlist; 61 struct macvlan_dev *vlan; 62 unsigned char addr[6+2] __aligned(sizeof(u16)); 63 struct rcu_head rcu; 64 }; 65 66 struct macvlan_skb_cb { 67 const struct macvlan_dev *src; 68 }; 69 70 #define MACVLAN_SKB_CB(__skb) ((struct macvlan_skb_cb *)&((__skb)->cb[0])) 71 72 static void macvlan_port_destroy(struct net_device *dev); 73 static void update_port_bc_queue_len(struct macvlan_port *port); 74 75 static inline bool macvlan_passthru(const struct macvlan_port *port) 76 { 77 return port->flags & MACVLAN_F_PASSTHRU; 78 } 79 80 static inline void macvlan_set_passthru(struct macvlan_port *port) 81 { 82 port->flags |= MACVLAN_F_PASSTHRU; 83 } 84 85 static inline bool macvlan_addr_change(const struct macvlan_port *port) 86 { 87 return port->flags & MACVLAN_F_ADDRCHANGE; 88 } 89 90 static inline void macvlan_set_addr_change(struct macvlan_port *port) 91 { 92 port->flags |= MACVLAN_F_ADDRCHANGE; 93 } 94 95 static inline void macvlan_clear_addr_change(struct macvlan_port *port) 96 { 97 port->flags &= ~MACVLAN_F_ADDRCHANGE; 98 } 99 100 /* Hash Ethernet address */ 101 static u32 macvlan_eth_hash(const unsigned char *addr) 102 { 103 u64 value = get_unaligned((u64 *)addr); 104 105 /* only want 6 bytes */ 106 #ifdef __BIG_ENDIAN 107 value >>= 16; 108 #else 109 value <<= 16; 110 #endif 111 return hash_64(value, MACVLAN_HASH_BITS); 112 } 113 114 static struct macvlan_port *macvlan_port_get_rcu(const struct net_device *dev) 115 { 116 return rcu_dereference(dev->rx_handler_data); 117 } 118 119 static struct macvlan_port *macvlan_port_get_rtnl(const struct net_device *dev) 120 { 121 return rtnl_dereference(dev->rx_handler_data); 122 } 123 124 static struct macvlan_dev *macvlan_hash_lookup(const struct macvlan_port *port, 125 const unsigned char *addr) 126 { 127 struct macvlan_dev *vlan; 128 u32 idx = macvlan_eth_hash(addr); 129 130 hlist_for_each_entry_rcu(vlan, &port->vlan_hash[idx], hlist, 131 lockdep_rtnl_is_held()) { 132 if (ether_addr_equal_64bits(vlan->dev->dev_addr, addr)) 133 return vlan; 134 } 135 return NULL; 136 } 137 138 static struct macvlan_source_entry *macvlan_hash_lookup_source( 139 const struct macvlan_dev *vlan, 140 const unsigned char *addr) 141 { 142 struct macvlan_source_entry *entry; 143 u32 idx = macvlan_eth_hash(addr); 144 struct hlist_head *h = &vlan->port->vlan_source_hash[idx]; 145 146 hlist_for_each_entry_rcu(entry, h, hlist, lockdep_rtnl_is_held()) { 147 if (ether_addr_equal_64bits(entry->addr, addr) && 148 entry->vlan == vlan) 149 return entry; 150 } 151 return NULL; 152 } 153 154 static int macvlan_hash_add_source(struct macvlan_dev *vlan, 155 const unsigned char *addr) 156 { 157 struct macvlan_port *port = vlan->port; 158 struct macvlan_source_entry *entry; 159 struct hlist_head *h; 160 161 entry = macvlan_hash_lookup_source(vlan, addr); 162 if (entry) 163 return 0; 164 165 entry = kmalloc(sizeof(*entry), GFP_KERNEL); 166 if (!entry) 167 return -ENOMEM; 168 169 ether_addr_copy(entry->addr, addr); 170 entry->vlan = vlan; 171 h = &port->vlan_source_hash[macvlan_eth_hash(addr)]; 172 hlist_add_head_rcu(&entry->hlist, h); 173 vlan->macaddr_count++; 174 175 return 0; 176 } 177 178 static void macvlan_hash_add(struct macvlan_dev *vlan) 179 { 180 struct macvlan_port *port = vlan->port; 181 const unsigned char *addr = vlan->dev->dev_addr; 182 u32 idx = macvlan_eth_hash(addr); 183 184 hlist_add_head_rcu(&vlan->hlist, &port->vlan_hash[idx]); 185 } 186 187 static void macvlan_hash_del_source(struct macvlan_source_entry *entry) 188 { 189 hlist_del_rcu(&entry->hlist); 190 kfree_rcu(entry, rcu); 191 } 192 193 static void macvlan_hash_del(struct macvlan_dev *vlan, bool sync) 194 { 195 hlist_del_rcu(&vlan->hlist); 196 if (sync) 197 synchronize_rcu(); 198 } 199 200 static void macvlan_hash_change_addr(struct macvlan_dev *vlan, 201 const unsigned char *addr) 202 { 203 macvlan_hash_del(vlan, true); 204 /* Now that we are unhashed it is safe to change the device 205 * address without confusing packet delivery. 206 */ 207 eth_hw_addr_set(vlan->dev, addr); 208 macvlan_hash_add(vlan); 209 } 210 211 static bool macvlan_addr_busy(const struct macvlan_port *port, 212 const unsigned char *addr) 213 { 214 /* Test to see if the specified address is 215 * currently in use by the underlying device or 216 * another macvlan. 217 */ 218 if (!macvlan_passthru(port) && !macvlan_addr_change(port) && 219 ether_addr_equal_64bits(port->dev->dev_addr, addr)) 220 return true; 221 222 if (macvlan_hash_lookup(port, addr)) 223 return true; 224 225 return false; 226 } 227 228 229 static int macvlan_broadcast_one(struct sk_buff *skb, 230 const struct macvlan_dev *vlan, 231 const struct ethhdr *eth, bool local) 232 { 233 struct net_device *dev = vlan->dev; 234 235 if (local) 236 return __dev_forward_skb(dev, skb); 237 238 skb->dev = dev; 239 if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast)) 240 skb->pkt_type = PACKET_BROADCAST; 241 else 242 skb->pkt_type = PACKET_MULTICAST; 243 244 return 0; 245 } 246 247 static u32 macvlan_hash_mix(const struct macvlan_dev *vlan) 248 { 249 return (u32)(((unsigned long)vlan) >> L1_CACHE_SHIFT); 250 } 251 252 253 static unsigned int mc_hash(const struct macvlan_dev *vlan, 254 const unsigned char *addr) 255 { 256 u32 val = __get_unaligned_cpu32(addr + 2); 257 258 val ^= macvlan_hash_mix(vlan); 259 return hash_32(val, MACVLAN_MC_FILTER_BITS); 260 } 261 262 static void macvlan_broadcast(struct sk_buff *skb, 263 const struct macvlan_port *port, 264 struct net_device *src, 265 enum macvlan_mode mode) 266 { 267 const struct ethhdr *eth = eth_hdr(skb); 268 const struct macvlan_dev *vlan; 269 struct sk_buff *nskb; 270 unsigned int i; 271 int err; 272 unsigned int hash; 273 274 if (skb->protocol == htons(ETH_P_PAUSE)) 275 return; 276 277 hash_for_each_rcu(port->vlan_hash, i, vlan, hlist) { 278 if (vlan->dev == src || !(vlan->mode & mode)) 279 continue; 280 281 hash = mc_hash(vlan, eth->h_dest); 282 if (!test_bit(hash, vlan->mc_filter)) 283 continue; 284 285 err = NET_RX_DROP; 286 nskb = skb_clone(skb, GFP_ATOMIC); 287 if (likely(nskb)) 288 err = macvlan_broadcast_one(nskb, vlan, eth, 289 mode == MACVLAN_MODE_BRIDGE) ?: 290 netif_rx(nskb); 291 macvlan_count_rx(vlan, skb->len + ETH_HLEN, 292 err == NET_RX_SUCCESS, true); 293 } 294 } 295 296 static void macvlan_multicast_rx(const struct macvlan_port *port, 297 const struct macvlan_dev *src, 298 struct sk_buff *skb) 299 { 300 if (!src) 301 /* frame comes from an external address */ 302 macvlan_broadcast(skb, port, NULL, 303 MACVLAN_MODE_PRIVATE | 304 MACVLAN_MODE_VEPA | 305 MACVLAN_MODE_PASSTHRU| 306 MACVLAN_MODE_BRIDGE); 307 else if (src->mode == MACVLAN_MODE_VEPA) 308 /* flood to everyone except source */ 309 macvlan_broadcast(skb, port, src->dev, 310 MACVLAN_MODE_VEPA | 311 MACVLAN_MODE_BRIDGE); 312 else 313 /* 314 * flood only to VEPA ports, bridge ports 315 * already saw the frame on the way out. 316 */ 317 macvlan_broadcast(skb, port, src->dev, 318 MACVLAN_MODE_VEPA); 319 } 320 321 static void macvlan_process_broadcast(struct work_struct *w) 322 { 323 struct macvlan_port *port = container_of(w, struct macvlan_port, 324 bc_work); 325 struct sk_buff *skb; 326 struct sk_buff_head list; 327 328 __skb_queue_head_init(&list); 329 330 spin_lock_bh(&port->bc_queue.lock); 331 skb_queue_splice_tail_init(&port->bc_queue, &list); 332 spin_unlock_bh(&port->bc_queue.lock); 333 334 while ((skb = __skb_dequeue(&list))) { 335 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src; 336 337 rcu_read_lock(); 338 macvlan_multicast_rx(port, src, skb); 339 rcu_read_unlock(); 340 341 if (src) 342 dev_put(src->dev); 343 consume_skb(skb); 344 345 cond_resched(); 346 } 347 } 348 349 static void macvlan_broadcast_enqueue(struct macvlan_port *port, 350 const struct macvlan_dev *src, 351 struct sk_buff *skb) 352 { 353 struct sk_buff *nskb; 354 int err = -ENOMEM; 355 356 nskb = skb_clone(skb, GFP_ATOMIC); 357 if (!nskb) 358 goto err; 359 360 MACVLAN_SKB_CB(nskb)->src = src; 361 362 spin_lock(&port->bc_queue.lock); 363 if (skb_queue_len(&port->bc_queue) < port->bc_queue_len_used) { 364 if (src) 365 dev_hold(src->dev); 366 __skb_queue_tail(&port->bc_queue, nskb); 367 err = 0; 368 } 369 spin_unlock(&port->bc_queue.lock); 370 371 queue_work(system_unbound_wq, &port->bc_work); 372 373 if (err) 374 goto free_nskb; 375 376 return; 377 378 free_nskb: 379 kfree_skb(nskb); 380 err: 381 dev_core_stats_rx_dropped_inc(skb->dev); 382 } 383 384 static void macvlan_flush_sources(struct macvlan_port *port, 385 struct macvlan_dev *vlan) 386 { 387 struct macvlan_source_entry *entry; 388 struct hlist_node *next; 389 int i; 390 391 hash_for_each_safe(port->vlan_source_hash, i, next, entry, hlist) 392 if (entry->vlan == vlan) 393 macvlan_hash_del_source(entry); 394 395 vlan->macaddr_count = 0; 396 } 397 398 static void macvlan_forward_source_one(struct sk_buff *skb, 399 struct macvlan_dev *vlan) 400 { 401 struct sk_buff *nskb; 402 struct net_device *dev; 403 int len; 404 int ret; 405 406 dev = vlan->dev; 407 if (unlikely(!(dev->flags & IFF_UP))) 408 return; 409 410 nskb = skb_clone(skb, GFP_ATOMIC); 411 if (!nskb) 412 return; 413 414 len = nskb->len + ETH_HLEN; 415 nskb->dev = dev; 416 417 if (ether_addr_equal_64bits(eth_hdr(skb)->h_dest, dev->dev_addr)) 418 nskb->pkt_type = PACKET_HOST; 419 420 ret = __netif_rx(nskb); 421 macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false); 422 } 423 424 static bool macvlan_forward_source(struct sk_buff *skb, 425 struct macvlan_port *port, 426 const unsigned char *addr) 427 { 428 struct macvlan_source_entry *entry; 429 u32 idx = macvlan_eth_hash(addr); 430 struct hlist_head *h = &port->vlan_source_hash[idx]; 431 bool consume = false; 432 433 hlist_for_each_entry_rcu(entry, h, hlist) { 434 if (ether_addr_equal_64bits(entry->addr, addr)) { 435 if (entry->vlan->flags & MACVLAN_FLAG_NODST) 436 consume = true; 437 macvlan_forward_source_one(skb, entry->vlan); 438 } 439 } 440 441 return consume; 442 } 443 444 /* called under rcu_read_lock() from netif_receive_skb */ 445 static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb) 446 { 447 struct macvlan_port *port; 448 struct sk_buff *skb = *pskb; 449 const struct ethhdr *eth = eth_hdr(skb); 450 const struct macvlan_dev *vlan; 451 const struct macvlan_dev *src; 452 struct net_device *dev; 453 unsigned int len = 0; 454 int ret; 455 rx_handler_result_t handle_res; 456 457 /* Packets from dev_loopback_xmit() do not have L2 header, bail out */ 458 if (unlikely(skb->pkt_type == PACKET_LOOPBACK)) 459 return RX_HANDLER_PASS; 460 461 port = macvlan_port_get_rcu(skb->dev); 462 if (is_multicast_ether_addr(eth->h_dest)) { 463 unsigned int hash; 464 465 skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN); 466 if (!skb) 467 return RX_HANDLER_CONSUMED; 468 *pskb = skb; 469 eth = eth_hdr(skb); 470 if (macvlan_forward_source(skb, port, eth->h_source)) { 471 kfree_skb(skb); 472 return RX_HANDLER_CONSUMED; 473 } 474 src = macvlan_hash_lookup(port, eth->h_source); 475 if (src && src->mode != MACVLAN_MODE_VEPA && 476 src->mode != MACVLAN_MODE_BRIDGE) { 477 /* forward to original port. */ 478 vlan = src; 479 ret = macvlan_broadcast_one(skb, vlan, eth, 0) ?: 480 __netif_rx(skb); 481 handle_res = RX_HANDLER_CONSUMED; 482 goto out; 483 } 484 485 hash = mc_hash(NULL, eth->h_dest); 486 if (test_bit(hash, port->bc_filter)) 487 macvlan_broadcast_enqueue(port, src, skb); 488 else if (test_bit(hash, port->mc_filter)) 489 macvlan_multicast_rx(port, src, skb); 490 491 return RX_HANDLER_PASS; 492 } 493 494 if (macvlan_forward_source(skb, port, eth->h_source)) { 495 kfree_skb(skb); 496 return RX_HANDLER_CONSUMED; 497 } 498 if (macvlan_passthru(port)) 499 vlan = list_first_or_null_rcu(&port->vlans, 500 struct macvlan_dev, list); 501 else 502 vlan = macvlan_hash_lookup(port, eth->h_dest); 503 if (!vlan || vlan->mode == MACVLAN_MODE_SOURCE) 504 return RX_HANDLER_PASS; 505 506 dev = vlan->dev; 507 if (unlikely(!(dev->flags & IFF_UP))) { 508 kfree_skb(skb); 509 return RX_HANDLER_CONSUMED; 510 } 511 len = skb->len + ETH_HLEN; 512 skb = skb_share_check(skb, GFP_ATOMIC); 513 if (!skb) { 514 ret = NET_RX_DROP; 515 handle_res = RX_HANDLER_CONSUMED; 516 goto out; 517 } 518 519 *pskb = skb; 520 skb->dev = dev; 521 skb->pkt_type = PACKET_HOST; 522 523 ret = NET_RX_SUCCESS; 524 handle_res = RX_HANDLER_ANOTHER; 525 out: 526 macvlan_count_rx(vlan, len, ret == NET_RX_SUCCESS, false); 527 return handle_res; 528 } 529 530 static int macvlan_queue_xmit(struct sk_buff *skb, struct net_device *dev) 531 { 532 const struct macvlan_dev *vlan = netdev_priv(dev); 533 const struct macvlan_port *port = vlan->port; 534 const struct macvlan_dev *dest; 535 536 if (vlan->mode == MACVLAN_MODE_BRIDGE) { 537 const struct ethhdr *eth = skb_eth_hdr(skb); 538 539 /* send to other bridge ports directly */ 540 if (is_multicast_ether_addr(eth->h_dest)) { 541 skb_reset_mac_header(skb); 542 macvlan_broadcast(skb, port, dev, MACVLAN_MODE_BRIDGE); 543 goto xmit_world; 544 } 545 546 dest = macvlan_hash_lookup(port, eth->h_dest); 547 if (dest && dest->mode == MACVLAN_MODE_BRIDGE) { 548 /* send to lowerdev first for its network taps */ 549 dev_forward_skb(vlan->lowerdev, skb); 550 551 return NET_XMIT_SUCCESS; 552 } 553 } 554 xmit_world: 555 skb->dev = vlan->lowerdev; 556 return dev_queue_xmit_accel(skb, 557 netdev_get_sb_channel(dev) ? dev : NULL); 558 } 559 560 static inline netdev_tx_t macvlan_netpoll_send_skb(struct macvlan_dev *vlan, struct sk_buff *skb) 561 { 562 #ifdef CONFIG_NET_POLL_CONTROLLER 563 return netpoll_send_skb(vlan->netpoll, skb); 564 #else 565 BUG(); 566 return NETDEV_TX_OK; 567 #endif 568 } 569 570 static netdev_tx_t macvlan_start_xmit(struct sk_buff *skb, 571 struct net_device *dev) 572 { 573 struct macvlan_dev *vlan = netdev_priv(dev); 574 unsigned int len = skb->len; 575 int ret; 576 577 if (unlikely(netpoll_tx_running(dev))) 578 return macvlan_netpoll_send_skb(vlan, skb); 579 580 ret = macvlan_queue_xmit(skb, dev); 581 582 if (likely(ret == NET_XMIT_SUCCESS || ret == NET_XMIT_CN)) { 583 struct vlan_pcpu_stats *pcpu_stats; 584 585 pcpu_stats = this_cpu_ptr(vlan->pcpu_stats); 586 u64_stats_update_begin(&pcpu_stats->syncp); 587 u64_stats_inc(&pcpu_stats->tx_packets); 588 u64_stats_add(&pcpu_stats->tx_bytes, len); 589 u64_stats_update_end(&pcpu_stats->syncp); 590 } else { 591 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 592 } 593 return ret; 594 } 595 596 static int macvlan_hard_header(struct sk_buff *skb, struct net_device *dev, 597 unsigned short type, const void *daddr, 598 const void *saddr, unsigned len) 599 { 600 const struct macvlan_dev *vlan = netdev_priv(dev); 601 struct net_device *lowerdev = vlan->lowerdev; 602 603 return dev_hard_header(skb, lowerdev, type, daddr, 604 saddr ? : dev->dev_addr, len); 605 } 606 607 static const struct header_ops macvlan_hard_header_ops = { 608 .create = macvlan_hard_header, 609 .parse = eth_header_parse, 610 .cache = eth_header_cache, 611 .cache_update = eth_header_cache_update, 612 .parse_protocol = eth_header_parse_protocol, 613 }; 614 615 static int macvlan_open(struct net_device *dev) 616 { 617 struct macvlan_dev *vlan = netdev_priv(dev); 618 struct net_device *lowerdev = vlan->lowerdev; 619 int err; 620 621 if (macvlan_passthru(vlan->port)) { 622 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) { 623 err = dev_set_promiscuity(lowerdev, 1); 624 if (err < 0) 625 goto out; 626 } 627 goto hash_add; 628 } 629 630 err = -EADDRINUSE; 631 if (macvlan_addr_busy(vlan->port, dev->dev_addr)) 632 goto out; 633 634 /* Attempt to populate accel_priv which is used to offload the L2 635 * forwarding requests for unicast packets. 636 */ 637 if (lowerdev->features & NETIF_F_HW_L2FW_DOFFLOAD) 638 vlan->accel_priv = 639 lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, dev); 640 641 /* If earlier attempt to offload failed, or accel_priv is not 642 * populated we must add the unicast address to the lower device. 643 */ 644 if (IS_ERR_OR_NULL(vlan->accel_priv)) { 645 vlan->accel_priv = NULL; 646 err = dev_uc_add(lowerdev, dev->dev_addr); 647 if (err < 0) 648 goto out; 649 } 650 651 if (dev->flags & IFF_ALLMULTI) { 652 err = dev_set_allmulti(lowerdev, 1); 653 if (err < 0) 654 goto del_unicast; 655 } 656 657 if (dev->flags & IFF_PROMISC) { 658 err = dev_set_promiscuity(lowerdev, 1); 659 if (err < 0) 660 goto clear_multi; 661 } 662 663 hash_add: 664 macvlan_hash_add(vlan); 665 return 0; 666 667 clear_multi: 668 if (dev->flags & IFF_ALLMULTI) 669 dev_set_allmulti(lowerdev, -1); 670 del_unicast: 671 if (vlan->accel_priv) { 672 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev, 673 vlan->accel_priv); 674 vlan->accel_priv = NULL; 675 } else { 676 dev_uc_del(lowerdev, dev->dev_addr); 677 } 678 out: 679 return err; 680 } 681 682 static int macvlan_stop(struct net_device *dev) 683 { 684 struct macvlan_dev *vlan = netdev_priv(dev); 685 struct net_device *lowerdev = vlan->lowerdev; 686 687 if (vlan->accel_priv) { 688 lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev, 689 vlan->accel_priv); 690 vlan->accel_priv = NULL; 691 } 692 693 dev_uc_unsync(lowerdev, dev); 694 dev_mc_unsync(lowerdev, dev); 695 696 if (macvlan_passthru(vlan->port)) { 697 if (!(vlan->flags & MACVLAN_FLAG_NOPROMISC)) 698 dev_set_promiscuity(lowerdev, -1); 699 goto hash_del; 700 } 701 702 if (dev->flags & IFF_ALLMULTI) 703 dev_set_allmulti(lowerdev, -1); 704 705 if (dev->flags & IFF_PROMISC) 706 dev_set_promiscuity(lowerdev, -1); 707 708 dev_uc_del(lowerdev, dev->dev_addr); 709 710 hash_del: 711 macvlan_hash_del(vlan, !dev->dismantle); 712 return 0; 713 } 714 715 static int macvlan_sync_address(struct net_device *dev, 716 const unsigned char *addr) 717 { 718 struct macvlan_dev *vlan = netdev_priv(dev); 719 struct net_device *lowerdev = vlan->lowerdev; 720 struct macvlan_port *port = vlan->port; 721 int err; 722 723 if (!(dev->flags & IFF_UP)) { 724 /* Just copy in the new address */ 725 eth_hw_addr_set(dev, addr); 726 } else { 727 /* Rehash and update the device filters */ 728 if (macvlan_addr_busy(vlan->port, addr)) 729 return -EADDRINUSE; 730 731 if (!macvlan_passthru(port)) { 732 err = dev_uc_add(lowerdev, addr); 733 if (err) 734 return err; 735 736 dev_uc_del(lowerdev, dev->dev_addr); 737 } 738 739 macvlan_hash_change_addr(vlan, addr); 740 } 741 if (macvlan_passthru(port) && !macvlan_addr_change(port)) { 742 /* Since addr_change isn't set, we are here due to lower 743 * device change. Save the lower-dev address so we can 744 * restore it later. 745 */ 746 ether_addr_copy(vlan->port->perm_addr, 747 lowerdev->dev_addr); 748 } 749 macvlan_clear_addr_change(port); 750 return 0; 751 } 752 753 static int macvlan_set_mac_address(struct net_device *dev, void *p) 754 { 755 struct macvlan_dev *vlan = netdev_priv(dev); 756 struct sockaddr *addr = p; 757 758 if (!is_valid_ether_addr(addr->sa_data)) 759 return -EADDRNOTAVAIL; 760 761 /* If the addresses are the same, this is a no-op */ 762 if (ether_addr_equal(dev->dev_addr, addr->sa_data)) 763 return 0; 764 765 if (vlan->mode == MACVLAN_MODE_PASSTHRU) { 766 macvlan_set_addr_change(vlan->port); 767 return dev_set_mac_address(vlan->lowerdev, addr, NULL); 768 } 769 770 if (macvlan_addr_busy(vlan->port, addr->sa_data)) 771 return -EADDRINUSE; 772 773 return macvlan_sync_address(dev, addr->sa_data); 774 } 775 776 static void macvlan_change_rx_flags(struct net_device *dev, int change) 777 { 778 struct macvlan_dev *vlan = netdev_priv(dev); 779 struct net_device *lowerdev = vlan->lowerdev; 780 781 if (dev->flags & IFF_UP) { 782 if (change & IFF_ALLMULTI) 783 dev_set_allmulti(lowerdev, dev->flags & IFF_ALLMULTI ? 1 : -1); 784 if (!macvlan_passthru(vlan->port) && change & IFF_PROMISC) 785 dev_set_promiscuity(lowerdev, 786 dev->flags & IFF_PROMISC ? 1 : -1); 787 788 } 789 } 790 791 static void macvlan_compute_filter(unsigned long *mc_filter, 792 struct net_device *dev, 793 struct macvlan_dev *vlan, int cutoff) 794 { 795 if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) { 796 bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ); 797 } else { 798 DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ); 799 struct netdev_hw_addr *ha; 800 801 bitmap_zero(filter, MACVLAN_MC_FILTER_SZ); 802 netdev_for_each_mc_addr(ha, dev) { 803 if (!vlan && ha->synced <= cutoff) 804 continue; 805 806 __set_bit(mc_hash(vlan, ha->addr), filter); 807 } 808 809 __set_bit(mc_hash(vlan, dev->broadcast), filter); 810 811 bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ); 812 } 813 } 814 815 static void macvlan_recompute_bc_filter(struct macvlan_dev *vlan) 816 { 817 if (vlan->port->bc_cutoff < 0) { 818 bitmap_zero(vlan->port->bc_filter, MACVLAN_MC_FILTER_SZ); 819 return; 820 } 821 822 macvlan_compute_filter(vlan->port->bc_filter, vlan->lowerdev, NULL, 823 vlan->port->bc_cutoff); 824 } 825 826 static void macvlan_set_mac_lists(struct net_device *dev) 827 { 828 struct macvlan_dev *vlan = netdev_priv(dev); 829 830 macvlan_compute_filter(vlan->mc_filter, dev, vlan, 0); 831 832 dev_uc_sync(vlan->lowerdev, dev); 833 dev_mc_sync(vlan->lowerdev, dev); 834 835 /* This is slightly inaccurate as we're including the subscription 836 * list of vlan->lowerdev too. 837 * 838 * Bug alert: This only works if everyone has the same broadcast 839 * address as lowerdev. As soon as someone changes theirs this 840 * will break. 841 * 842 * However, this is already broken as when you change your broadcast 843 * address we don't get called. 844 * 845 * The solution is to maintain a list of broadcast addresses like 846 * we do for uc/mc, if you care. 847 */ 848 macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL, 849 0); 850 macvlan_recompute_bc_filter(vlan); 851 } 852 853 static void update_port_bc_cutoff(struct macvlan_dev *vlan, int cutoff) 854 { 855 if (vlan->port->bc_cutoff == cutoff) 856 return; 857 858 vlan->port->bc_cutoff = cutoff; 859 macvlan_recompute_bc_filter(vlan); 860 } 861 862 static int macvlan_change_mtu(struct net_device *dev, int new_mtu) 863 { 864 struct macvlan_dev *vlan = netdev_priv(dev); 865 866 if (vlan->lowerdev->mtu < new_mtu) 867 return -EINVAL; 868 dev->mtu = new_mtu; 869 return 0; 870 } 871 872 static int macvlan_hwtstamp_get(struct net_device *dev, 873 struct kernel_hwtstamp_config *cfg) 874 { 875 struct net_device *real_dev = macvlan_dev_real_dev(dev); 876 877 return generic_hwtstamp_get_lower(real_dev, cfg); 878 } 879 880 static int macvlan_hwtstamp_set(struct net_device *dev, 881 struct kernel_hwtstamp_config *cfg, 882 struct netlink_ext_ack *extack) 883 { 884 struct net_device *real_dev = macvlan_dev_real_dev(dev); 885 886 if (!net_eq(dev_net(dev), &init_net)) 887 return -EOPNOTSUPP; 888 889 return generic_hwtstamp_set_lower(real_dev, cfg, extack); 890 } 891 892 /* 893 * macvlan network devices have devices nesting below it and are a special 894 * "super class" of normal network devices; split their locks off into a 895 * separate class since they always nest. 896 */ 897 static struct lock_class_key macvlan_netdev_addr_lock_key; 898 899 #define ALWAYS_ON_OFFLOADS \ 900 (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE | \ 901 NETIF_F_GSO_ROBUST | NETIF_F_GSO_ENCAP_ALL) 902 903 #define ALWAYS_ON_FEATURES (ALWAYS_ON_OFFLOADS | NETIF_F_LLTX) 904 905 #define MACVLAN_FEATURES \ 906 (NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST | \ 907 NETIF_F_GSO | NETIF_F_TSO | NETIF_F_LRO | \ 908 NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_GRO | NETIF_F_RXCSUM | \ 909 NETIF_F_HW_VLAN_CTAG_FILTER | NETIF_F_HW_VLAN_STAG_FILTER) 910 911 #define MACVLAN_STATE_MASK \ 912 ((1<<__LINK_STATE_NOCARRIER) | (1<<__LINK_STATE_DORMANT)) 913 914 static void macvlan_set_lockdep_class(struct net_device *dev) 915 { 916 netdev_lockdep_set_classes(dev); 917 lockdep_set_class(&dev->addr_list_lock, 918 &macvlan_netdev_addr_lock_key); 919 } 920 921 static int macvlan_init(struct net_device *dev) 922 { 923 struct macvlan_dev *vlan = netdev_priv(dev); 924 struct net_device *lowerdev = vlan->lowerdev; 925 struct macvlan_port *port = vlan->port; 926 927 dev->state = (dev->state & ~MACVLAN_STATE_MASK) | 928 (lowerdev->state & MACVLAN_STATE_MASK); 929 dev->features = lowerdev->features & MACVLAN_FEATURES; 930 dev->features |= ALWAYS_ON_FEATURES; 931 dev->hw_features |= NETIF_F_LRO; 932 dev->vlan_features = lowerdev->vlan_features & MACVLAN_FEATURES; 933 dev->vlan_features |= ALWAYS_ON_OFFLOADS; 934 dev->hw_enc_features |= dev->features; 935 netif_inherit_tso_max(dev, lowerdev); 936 dev->hard_header_len = lowerdev->hard_header_len; 937 macvlan_set_lockdep_class(dev); 938 939 vlan->pcpu_stats = netdev_alloc_pcpu_stats(struct vlan_pcpu_stats); 940 if (!vlan->pcpu_stats) 941 return -ENOMEM; 942 943 port->count += 1; 944 945 /* Get macvlan's reference to lowerdev */ 946 netdev_hold(lowerdev, &vlan->dev_tracker, GFP_KERNEL); 947 948 return 0; 949 } 950 951 static void macvlan_uninit(struct net_device *dev) 952 { 953 struct macvlan_dev *vlan = netdev_priv(dev); 954 struct macvlan_port *port = vlan->port; 955 956 free_percpu(vlan->pcpu_stats); 957 958 macvlan_flush_sources(port, vlan); 959 port->count -= 1; 960 if (!port->count) 961 macvlan_port_destroy(port->dev); 962 } 963 964 static void macvlan_dev_get_stats64(struct net_device *dev, 965 struct rtnl_link_stats64 *stats) 966 { 967 struct macvlan_dev *vlan = netdev_priv(dev); 968 969 if (vlan->pcpu_stats) { 970 struct vlan_pcpu_stats *p; 971 u64 rx_packets, rx_bytes, rx_multicast, tx_packets, tx_bytes; 972 u32 rx_errors = 0, tx_dropped = 0; 973 unsigned int start; 974 int i; 975 976 for_each_possible_cpu(i) { 977 p = per_cpu_ptr(vlan->pcpu_stats, i); 978 do { 979 start = u64_stats_fetch_begin(&p->syncp); 980 rx_packets = u64_stats_read(&p->rx_packets); 981 rx_bytes = u64_stats_read(&p->rx_bytes); 982 rx_multicast = u64_stats_read(&p->rx_multicast); 983 tx_packets = u64_stats_read(&p->tx_packets); 984 tx_bytes = u64_stats_read(&p->tx_bytes); 985 } while (u64_stats_fetch_retry(&p->syncp, start)); 986 987 stats->rx_packets += rx_packets; 988 stats->rx_bytes += rx_bytes; 989 stats->multicast += rx_multicast; 990 stats->tx_packets += tx_packets; 991 stats->tx_bytes += tx_bytes; 992 /* rx_errors & tx_dropped are u32, updated 993 * without syncp protection. 994 */ 995 rx_errors += READ_ONCE(p->rx_errors); 996 tx_dropped += READ_ONCE(p->tx_dropped); 997 } 998 stats->rx_errors = rx_errors; 999 stats->rx_dropped = rx_errors; 1000 stats->tx_dropped = tx_dropped; 1001 } 1002 } 1003 1004 static int macvlan_vlan_rx_add_vid(struct net_device *dev, 1005 __be16 proto, u16 vid) 1006 { 1007 struct macvlan_dev *vlan = netdev_priv(dev); 1008 struct net_device *lowerdev = vlan->lowerdev; 1009 1010 return vlan_vid_add(lowerdev, proto, vid); 1011 } 1012 1013 static int macvlan_vlan_rx_kill_vid(struct net_device *dev, 1014 __be16 proto, u16 vid) 1015 { 1016 struct macvlan_dev *vlan = netdev_priv(dev); 1017 struct net_device *lowerdev = vlan->lowerdev; 1018 1019 vlan_vid_del(lowerdev, proto, vid); 1020 return 0; 1021 } 1022 1023 static int macvlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 1024 struct net_device *dev, 1025 const unsigned char *addr, u16 vid, 1026 u16 flags, 1027 struct netlink_ext_ack *extack) 1028 { 1029 struct macvlan_dev *vlan = netdev_priv(dev); 1030 int err = -EINVAL; 1031 1032 /* Support unicast filter only on passthru devices. 1033 * Multicast filter should be allowed on all devices. 1034 */ 1035 if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr)) 1036 return -EOPNOTSUPP; 1037 1038 if (flags & NLM_F_REPLACE) 1039 return -EOPNOTSUPP; 1040 1041 if (is_unicast_ether_addr(addr)) 1042 err = dev_uc_add_excl(dev, addr); 1043 else if (is_multicast_ether_addr(addr)) 1044 err = dev_mc_add_excl(dev, addr); 1045 1046 return err; 1047 } 1048 1049 static int macvlan_fdb_del(struct ndmsg *ndm, struct nlattr *tb[], 1050 struct net_device *dev, 1051 const unsigned char *addr, u16 vid, 1052 struct netlink_ext_ack *extack) 1053 { 1054 struct macvlan_dev *vlan = netdev_priv(dev); 1055 int err = -EINVAL; 1056 1057 /* Support unicast filter only on passthru devices. 1058 * Multicast filter should be allowed on all devices. 1059 */ 1060 if (!macvlan_passthru(vlan->port) && is_unicast_ether_addr(addr)) 1061 return -EOPNOTSUPP; 1062 1063 if (is_unicast_ether_addr(addr)) 1064 err = dev_uc_del(dev, addr); 1065 else if (is_multicast_ether_addr(addr)) 1066 err = dev_mc_del(dev, addr); 1067 1068 return err; 1069 } 1070 1071 static void macvlan_ethtool_get_drvinfo(struct net_device *dev, 1072 struct ethtool_drvinfo *drvinfo) 1073 { 1074 strscpy(drvinfo->driver, "macvlan", sizeof(drvinfo->driver)); 1075 strscpy(drvinfo->version, "0.1", sizeof(drvinfo->version)); 1076 } 1077 1078 static int macvlan_ethtool_get_link_ksettings(struct net_device *dev, 1079 struct ethtool_link_ksettings *cmd) 1080 { 1081 const struct macvlan_dev *vlan = netdev_priv(dev); 1082 1083 return __ethtool_get_link_ksettings(vlan->lowerdev, cmd); 1084 } 1085 1086 static int macvlan_ethtool_get_ts_info(struct net_device *dev, 1087 struct ethtool_ts_info *info) 1088 { 1089 struct net_device *real_dev = macvlan_dev_real_dev(dev); 1090 1091 return ethtool_get_ts_info_by_layer(real_dev, info); 1092 } 1093 1094 static netdev_features_t macvlan_fix_features(struct net_device *dev, 1095 netdev_features_t features) 1096 { 1097 struct macvlan_dev *vlan = netdev_priv(dev); 1098 netdev_features_t lowerdev_features = vlan->lowerdev->features; 1099 netdev_features_t mask; 1100 1101 features |= NETIF_F_ALL_FOR_ALL; 1102 features &= (vlan->set_features | ~MACVLAN_FEATURES); 1103 mask = features; 1104 1105 lowerdev_features &= (features | ~NETIF_F_LRO); 1106 features = netdev_increment_features(lowerdev_features, features, mask); 1107 features |= ALWAYS_ON_FEATURES; 1108 features &= (ALWAYS_ON_FEATURES | MACVLAN_FEATURES); 1109 1110 return features; 1111 } 1112 1113 #ifdef CONFIG_NET_POLL_CONTROLLER 1114 static void macvlan_dev_poll_controller(struct net_device *dev) 1115 { 1116 return; 1117 } 1118 1119 static int macvlan_dev_netpoll_setup(struct net_device *dev, struct netpoll_info *npinfo) 1120 { 1121 struct macvlan_dev *vlan = netdev_priv(dev); 1122 struct net_device *real_dev = vlan->lowerdev; 1123 struct netpoll *netpoll; 1124 int err; 1125 1126 netpoll = kzalloc(sizeof(*netpoll), GFP_KERNEL); 1127 err = -ENOMEM; 1128 if (!netpoll) 1129 goto out; 1130 1131 err = __netpoll_setup(netpoll, real_dev); 1132 if (err) { 1133 kfree(netpoll); 1134 goto out; 1135 } 1136 1137 vlan->netpoll = netpoll; 1138 1139 out: 1140 return err; 1141 } 1142 1143 static void macvlan_dev_netpoll_cleanup(struct net_device *dev) 1144 { 1145 struct macvlan_dev *vlan = netdev_priv(dev); 1146 struct netpoll *netpoll = vlan->netpoll; 1147 1148 if (!netpoll) 1149 return; 1150 1151 vlan->netpoll = NULL; 1152 1153 __netpoll_free(netpoll); 1154 } 1155 #endif /* CONFIG_NET_POLL_CONTROLLER */ 1156 1157 static int macvlan_dev_get_iflink(const struct net_device *dev) 1158 { 1159 struct macvlan_dev *vlan = netdev_priv(dev); 1160 1161 return READ_ONCE(vlan->lowerdev->ifindex); 1162 } 1163 1164 static const struct ethtool_ops macvlan_ethtool_ops = { 1165 .get_link = ethtool_op_get_link, 1166 .get_link_ksettings = macvlan_ethtool_get_link_ksettings, 1167 .get_drvinfo = macvlan_ethtool_get_drvinfo, 1168 .get_ts_info = macvlan_ethtool_get_ts_info, 1169 }; 1170 1171 static const struct net_device_ops macvlan_netdev_ops = { 1172 .ndo_init = macvlan_init, 1173 .ndo_uninit = macvlan_uninit, 1174 .ndo_open = macvlan_open, 1175 .ndo_stop = macvlan_stop, 1176 .ndo_start_xmit = macvlan_start_xmit, 1177 .ndo_change_mtu = macvlan_change_mtu, 1178 .ndo_fix_features = macvlan_fix_features, 1179 .ndo_change_rx_flags = macvlan_change_rx_flags, 1180 .ndo_set_mac_address = macvlan_set_mac_address, 1181 .ndo_set_rx_mode = macvlan_set_mac_lists, 1182 .ndo_get_stats64 = macvlan_dev_get_stats64, 1183 .ndo_validate_addr = eth_validate_addr, 1184 .ndo_vlan_rx_add_vid = macvlan_vlan_rx_add_vid, 1185 .ndo_vlan_rx_kill_vid = macvlan_vlan_rx_kill_vid, 1186 .ndo_fdb_add = macvlan_fdb_add, 1187 .ndo_fdb_del = macvlan_fdb_del, 1188 .ndo_fdb_dump = ndo_dflt_fdb_dump, 1189 #ifdef CONFIG_NET_POLL_CONTROLLER 1190 .ndo_poll_controller = macvlan_dev_poll_controller, 1191 .ndo_netpoll_setup = macvlan_dev_netpoll_setup, 1192 .ndo_netpoll_cleanup = macvlan_dev_netpoll_cleanup, 1193 #endif 1194 .ndo_get_iflink = macvlan_dev_get_iflink, 1195 .ndo_features_check = passthru_features_check, 1196 .ndo_hwtstamp_get = macvlan_hwtstamp_get, 1197 .ndo_hwtstamp_set = macvlan_hwtstamp_set, 1198 }; 1199 1200 static void macvlan_dev_free(struct net_device *dev) 1201 { 1202 struct macvlan_dev *vlan = netdev_priv(dev); 1203 1204 /* Get rid of the macvlan's reference to lowerdev */ 1205 netdev_put(vlan->lowerdev, &vlan->dev_tracker); 1206 } 1207 1208 void macvlan_common_setup(struct net_device *dev) 1209 { 1210 ether_setup(dev); 1211 1212 /* ether_setup() has set dev->min_mtu to ETH_MIN_MTU. */ 1213 dev->max_mtu = ETH_MAX_MTU; 1214 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1215 netif_keep_dst(dev); 1216 dev->priv_flags |= IFF_UNICAST_FLT | IFF_CHANGE_PROTO_DOWN; 1217 dev->netdev_ops = &macvlan_netdev_ops; 1218 dev->needs_free_netdev = true; 1219 dev->priv_destructor = macvlan_dev_free; 1220 dev->header_ops = &macvlan_hard_header_ops; 1221 dev->ethtool_ops = &macvlan_ethtool_ops; 1222 } 1223 EXPORT_SYMBOL_GPL(macvlan_common_setup); 1224 1225 static void macvlan_setup(struct net_device *dev) 1226 { 1227 macvlan_common_setup(dev); 1228 dev->priv_flags |= IFF_NO_QUEUE; 1229 } 1230 1231 static int macvlan_port_create(struct net_device *dev) 1232 { 1233 struct macvlan_port *port; 1234 unsigned int i; 1235 int err; 1236 1237 if (dev->type != ARPHRD_ETHER || dev->flags & IFF_LOOPBACK) 1238 return -EINVAL; 1239 1240 if (netdev_is_rx_handler_busy(dev)) 1241 return -EBUSY; 1242 1243 port = kzalloc(sizeof(*port), GFP_KERNEL); 1244 if (port == NULL) 1245 return -ENOMEM; 1246 1247 port->dev = dev; 1248 ether_addr_copy(port->perm_addr, dev->dev_addr); 1249 INIT_LIST_HEAD(&port->vlans); 1250 for (i = 0; i < MACVLAN_HASH_SIZE; i++) 1251 INIT_HLIST_HEAD(&port->vlan_hash[i]); 1252 for (i = 0; i < MACVLAN_HASH_SIZE; i++) 1253 INIT_HLIST_HEAD(&port->vlan_source_hash[i]); 1254 1255 port->bc_queue_len_used = 0; 1256 port->bc_cutoff = 1; 1257 skb_queue_head_init(&port->bc_queue); 1258 INIT_WORK(&port->bc_work, macvlan_process_broadcast); 1259 1260 err = netdev_rx_handler_register(dev, macvlan_handle_frame, port); 1261 if (err) 1262 kfree(port); 1263 else 1264 dev->priv_flags |= IFF_MACVLAN_PORT; 1265 return err; 1266 } 1267 1268 static void macvlan_port_destroy(struct net_device *dev) 1269 { 1270 struct macvlan_port *port = macvlan_port_get_rtnl(dev); 1271 struct sk_buff *skb; 1272 1273 dev->priv_flags &= ~IFF_MACVLAN_PORT; 1274 netdev_rx_handler_unregister(dev); 1275 1276 /* After this point, no packet can schedule bc_work anymore, 1277 * but we need to cancel it and purge left skbs if any. 1278 */ 1279 cancel_work_sync(&port->bc_work); 1280 1281 while ((skb = __skb_dequeue(&port->bc_queue))) { 1282 const struct macvlan_dev *src = MACVLAN_SKB_CB(skb)->src; 1283 1284 if (src) 1285 dev_put(src->dev); 1286 1287 kfree_skb(skb); 1288 } 1289 1290 /* If the lower device address has been changed by passthru 1291 * macvlan, put it back. 1292 */ 1293 if (macvlan_passthru(port) && 1294 !ether_addr_equal(port->dev->dev_addr, port->perm_addr)) { 1295 struct sockaddr sa; 1296 1297 sa.sa_family = port->dev->type; 1298 memcpy(&sa.sa_data, port->perm_addr, port->dev->addr_len); 1299 dev_set_mac_address(port->dev, &sa, NULL); 1300 } 1301 1302 kfree(port); 1303 } 1304 1305 static int macvlan_validate(struct nlattr *tb[], struct nlattr *data[], 1306 struct netlink_ext_ack *extack) 1307 { 1308 struct nlattr *nla, *head; 1309 int rem, len; 1310 1311 if (tb[IFLA_ADDRESS]) { 1312 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1313 return -EINVAL; 1314 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1315 return -EADDRNOTAVAIL; 1316 } 1317 1318 if (!data) 1319 return 0; 1320 1321 if (data[IFLA_MACVLAN_FLAGS] && 1322 nla_get_u16(data[IFLA_MACVLAN_FLAGS]) & ~(MACVLAN_FLAG_NOPROMISC | 1323 MACVLAN_FLAG_NODST)) 1324 return -EINVAL; 1325 1326 if (data[IFLA_MACVLAN_MODE]) { 1327 switch (nla_get_u32(data[IFLA_MACVLAN_MODE])) { 1328 case MACVLAN_MODE_PRIVATE: 1329 case MACVLAN_MODE_VEPA: 1330 case MACVLAN_MODE_BRIDGE: 1331 case MACVLAN_MODE_PASSTHRU: 1332 case MACVLAN_MODE_SOURCE: 1333 break; 1334 default: 1335 return -EINVAL; 1336 } 1337 } 1338 1339 if (data[IFLA_MACVLAN_MACADDR_MODE]) { 1340 switch (nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE])) { 1341 case MACVLAN_MACADDR_ADD: 1342 case MACVLAN_MACADDR_DEL: 1343 case MACVLAN_MACADDR_FLUSH: 1344 case MACVLAN_MACADDR_SET: 1345 break; 1346 default: 1347 return -EINVAL; 1348 } 1349 } 1350 1351 if (data[IFLA_MACVLAN_MACADDR]) { 1352 if (nla_len(data[IFLA_MACVLAN_MACADDR]) != ETH_ALEN) 1353 return -EINVAL; 1354 1355 if (!is_valid_ether_addr(nla_data(data[IFLA_MACVLAN_MACADDR]))) 1356 return -EADDRNOTAVAIL; 1357 } 1358 1359 if (data[IFLA_MACVLAN_MACADDR_DATA]) { 1360 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]); 1361 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]); 1362 1363 nla_for_each_attr(nla, head, len, rem) { 1364 if (nla_type(nla) != IFLA_MACVLAN_MACADDR || 1365 nla_len(nla) != ETH_ALEN) 1366 return -EINVAL; 1367 1368 if (!is_valid_ether_addr(nla_data(nla))) 1369 return -EADDRNOTAVAIL; 1370 } 1371 } 1372 1373 if (data[IFLA_MACVLAN_MACADDR_COUNT]) 1374 return -EINVAL; 1375 1376 return 0; 1377 } 1378 1379 /* 1380 * reconfigure list of remote source mac address 1381 * (only for macvlan devices in source mode) 1382 * Note regarding alignment: all netlink data is aligned to 4 Byte, which 1383 * suffices for both ether_addr_copy and ether_addr_equal_64bits usage. 1384 */ 1385 static int macvlan_changelink_sources(struct macvlan_dev *vlan, u32 mode, 1386 struct nlattr *data[]) 1387 { 1388 char *addr = NULL; 1389 int ret, rem, len; 1390 struct nlattr *nla, *head; 1391 struct macvlan_source_entry *entry; 1392 1393 if (data[IFLA_MACVLAN_MACADDR]) 1394 addr = nla_data(data[IFLA_MACVLAN_MACADDR]); 1395 1396 if (mode == MACVLAN_MACADDR_ADD) { 1397 if (!addr) 1398 return -EINVAL; 1399 1400 return macvlan_hash_add_source(vlan, addr); 1401 1402 } else if (mode == MACVLAN_MACADDR_DEL) { 1403 if (!addr) 1404 return -EINVAL; 1405 1406 entry = macvlan_hash_lookup_source(vlan, addr); 1407 if (entry) { 1408 macvlan_hash_del_source(entry); 1409 vlan->macaddr_count--; 1410 } 1411 } else if (mode == MACVLAN_MACADDR_FLUSH) { 1412 macvlan_flush_sources(vlan->port, vlan); 1413 } else if (mode == MACVLAN_MACADDR_SET) { 1414 macvlan_flush_sources(vlan->port, vlan); 1415 1416 if (addr) { 1417 ret = macvlan_hash_add_source(vlan, addr); 1418 if (ret) 1419 return ret; 1420 } 1421 1422 if (!data[IFLA_MACVLAN_MACADDR_DATA]) 1423 return 0; 1424 1425 head = nla_data(data[IFLA_MACVLAN_MACADDR_DATA]); 1426 len = nla_len(data[IFLA_MACVLAN_MACADDR_DATA]); 1427 1428 nla_for_each_attr(nla, head, len, rem) { 1429 addr = nla_data(nla); 1430 ret = macvlan_hash_add_source(vlan, addr); 1431 if (ret) 1432 return ret; 1433 } 1434 } else { 1435 return -EINVAL; 1436 } 1437 1438 return 0; 1439 } 1440 1441 int macvlan_common_newlink(struct net *src_net, struct net_device *dev, 1442 struct nlattr *tb[], struct nlattr *data[], 1443 struct netlink_ext_ack *extack) 1444 { 1445 struct macvlan_dev *vlan = netdev_priv(dev); 1446 struct macvlan_port *port; 1447 struct net_device *lowerdev; 1448 int err; 1449 int macmode; 1450 bool create = false; 1451 1452 if (!tb[IFLA_LINK]) 1453 return -EINVAL; 1454 1455 lowerdev = __dev_get_by_index(src_net, nla_get_u32(tb[IFLA_LINK])); 1456 if (lowerdev == NULL) 1457 return -ENODEV; 1458 1459 /* When creating macvlans or macvtaps on top of other macvlans - use 1460 * the real device as the lowerdev. 1461 */ 1462 if (netif_is_macvlan(lowerdev)) 1463 lowerdev = macvlan_dev_real_dev(lowerdev); 1464 1465 if (!tb[IFLA_MTU]) 1466 dev->mtu = lowerdev->mtu; 1467 else if (dev->mtu > lowerdev->mtu) 1468 return -EINVAL; 1469 1470 /* MTU range: 68 - lowerdev->max_mtu */ 1471 dev->min_mtu = ETH_MIN_MTU; 1472 dev->max_mtu = lowerdev->max_mtu; 1473 1474 if (!tb[IFLA_ADDRESS]) 1475 eth_hw_addr_random(dev); 1476 1477 if (!netif_is_macvlan_port(lowerdev)) { 1478 err = macvlan_port_create(lowerdev); 1479 if (err < 0) 1480 return err; 1481 create = true; 1482 } 1483 port = macvlan_port_get_rtnl(lowerdev); 1484 1485 /* Only 1 macvlan device can be created in passthru mode */ 1486 if (macvlan_passthru(port)) { 1487 /* The macvlan port must be not created this time, 1488 * still goto destroy_macvlan_port for readability. 1489 */ 1490 err = -EINVAL; 1491 goto destroy_macvlan_port; 1492 } 1493 1494 vlan->lowerdev = lowerdev; 1495 vlan->dev = dev; 1496 vlan->port = port; 1497 vlan->set_features = MACVLAN_FEATURES; 1498 1499 vlan->mode = MACVLAN_MODE_VEPA; 1500 if (data && data[IFLA_MACVLAN_MODE]) 1501 vlan->mode = nla_get_u32(data[IFLA_MACVLAN_MODE]); 1502 1503 if (data && data[IFLA_MACVLAN_FLAGS]) 1504 vlan->flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]); 1505 1506 if (vlan->mode == MACVLAN_MODE_PASSTHRU) { 1507 if (port->count) { 1508 err = -EINVAL; 1509 goto destroy_macvlan_port; 1510 } 1511 macvlan_set_passthru(port); 1512 eth_hw_addr_inherit(dev, lowerdev); 1513 } 1514 1515 if (data && data[IFLA_MACVLAN_MACADDR_MODE]) { 1516 if (vlan->mode != MACVLAN_MODE_SOURCE) { 1517 err = -EINVAL; 1518 goto destroy_macvlan_port; 1519 } 1520 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]); 1521 err = macvlan_changelink_sources(vlan, macmode, data); 1522 if (err) 1523 goto destroy_macvlan_port; 1524 } 1525 1526 vlan->bc_queue_len_req = MACVLAN_DEFAULT_BC_QUEUE_LEN; 1527 if (data && data[IFLA_MACVLAN_BC_QUEUE_LEN]) 1528 vlan->bc_queue_len_req = nla_get_u32(data[IFLA_MACVLAN_BC_QUEUE_LEN]); 1529 1530 if (data && data[IFLA_MACVLAN_BC_CUTOFF]) 1531 update_port_bc_cutoff( 1532 vlan, nla_get_s32(data[IFLA_MACVLAN_BC_CUTOFF])); 1533 1534 err = register_netdevice(dev); 1535 if (err < 0) 1536 goto destroy_macvlan_port; 1537 1538 dev->priv_flags |= IFF_MACVLAN; 1539 err = netdev_upper_dev_link(lowerdev, dev, extack); 1540 if (err) 1541 goto unregister_netdev; 1542 1543 list_add_tail_rcu(&vlan->list, &port->vlans); 1544 update_port_bc_queue_len(vlan->port); 1545 netif_stacked_transfer_operstate(lowerdev, dev); 1546 linkwatch_fire_event(dev); 1547 1548 return 0; 1549 1550 unregister_netdev: 1551 /* macvlan_uninit would free the macvlan port */ 1552 unregister_netdevice(dev); 1553 return err; 1554 destroy_macvlan_port: 1555 /* the macvlan port may be freed by macvlan_uninit when fail to register. 1556 * so we destroy the macvlan port only when it's valid. 1557 */ 1558 if (create && macvlan_port_get_rtnl(lowerdev)) { 1559 macvlan_flush_sources(port, vlan); 1560 macvlan_port_destroy(port->dev); 1561 } 1562 return err; 1563 } 1564 EXPORT_SYMBOL_GPL(macvlan_common_newlink); 1565 1566 static int macvlan_newlink(struct net *src_net, struct net_device *dev, 1567 struct nlattr *tb[], struct nlattr *data[], 1568 struct netlink_ext_ack *extack) 1569 { 1570 return macvlan_common_newlink(src_net, dev, tb, data, extack); 1571 } 1572 1573 void macvlan_dellink(struct net_device *dev, struct list_head *head) 1574 { 1575 struct macvlan_dev *vlan = netdev_priv(dev); 1576 1577 if (vlan->mode == MACVLAN_MODE_SOURCE) 1578 macvlan_flush_sources(vlan->port, vlan); 1579 list_del_rcu(&vlan->list); 1580 update_port_bc_queue_len(vlan->port); 1581 unregister_netdevice_queue(dev, head); 1582 netdev_upper_dev_unlink(vlan->lowerdev, dev); 1583 } 1584 EXPORT_SYMBOL_GPL(macvlan_dellink); 1585 1586 static int macvlan_changelink(struct net_device *dev, 1587 struct nlattr *tb[], struct nlattr *data[], 1588 struct netlink_ext_ack *extack) 1589 { 1590 struct macvlan_dev *vlan = netdev_priv(dev); 1591 enum macvlan_mode mode; 1592 bool set_mode = false; 1593 enum macvlan_macaddr_mode macmode; 1594 int ret; 1595 1596 /* Validate mode, but don't set yet: setting flags may fail. */ 1597 if (data && data[IFLA_MACVLAN_MODE]) { 1598 set_mode = true; 1599 mode = nla_get_u32(data[IFLA_MACVLAN_MODE]); 1600 /* Passthrough mode can't be set or cleared dynamically */ 1601 if ((mode == MACVLAN_MODE_PASSTHRU) != 1602 (vlan->mode == MACVLAN_MODE_PASSTHRU)) 1603 return -EINVAL; 1604 if (vlan->mode == MACVLAN_MODE_SOURCE && 1605 vlan->mode != mode) 1606 macvlan_flush_sources(vlan->port, vlan); 1607 } 1608 1609 if (data && data[IFLA_MACVLAN_FLAGS]) { 1610 __u16 flags = nla_get_u16(data[IFLA_MACVLAN_FLAGS]); 1611 bool promisc = (flags ^ vlan->flags) & MACVLAN_FLAG_NOPROMISC; 1612 if (macvlan_passthru(vlan->port) && promisc) { 1613 int err; 1614 1615 if (flags & MACVLAN_FLAG_NOPROMISC) 1616 err = dev_set_promiscuity(vlan->lowerdev, -1); 1617 else 1618 err = dev_set_promiscuity(vlan->lowerdev, 1); 1619 if (err < 0) 1620 return err; 1621 } 1622 vlan->flags = flags; 1623 } 1624 1625 if (data && data[IFLA_MACVLAN_BC_QUEUE_LEN]) { 1626 vlan->bc_queue_len_req = nla_get_u32(data[IFLA_MACVLAN_BC_QUEUE_LEN]); 1627 update_port_bc_queue_len(vlan->port); 1628 } 1629 1630 if (data && data[IFLA_MACVLAN_BC_CUTOFF]) 1631 update_port_bc_cutoff( 1632 vlan, nla_get_s32(data[IFLA_MACVLAN_BC_CUTOFF])); 1633 1634 if (set_mode) 1635 vlan->mode = mode; 1636 if (data && data[IFLA_MACVLAN_MACADDR_MODE]) { 1637 if (vlan->mode != MACVLAN_MODE_SOURCE) 1638 return -EINVAL; 1639 macmode = nla_get_u32(data[IFLA_MACVLAN_MACADDR_MODE]); 1640 ret = macvlan_changelink_sources(vlan, macmode, data); 1641 if (ret) 1642 return ret; 1643 } 1644 return 0; 1645 } 1646 1647 static size_t macvlan_get_size_mac(const struct macvlan_dev *vlan) 1648 { 1649 if (vlan->macaddr_count == 0) 1650 return 0; 1651 return nla_total_size(0) /* IFLA_MACVLAN_MACADDR_DATA */ 1652 + vlan->macaddr_count * nla_total_size(sizeof(u8) * ETH_ALEN); 1653 } 1654 1655 static size_t macvlan_get_size(const struct net_device *dev) 1656 { 1657 struct macvlan_dev *vlan = netdev_priv(dev); 1658 1659 return (0 1660 + nla_total_size(4) /* IFLA_MACVLAN_MODE */ 1661 + nla_total_size(2) /* IFLA_MACVLAN_FLAGS */ 1662 + nla_total_size(4) /* IFLA_MACVLAN_MACADDR_COUNT */ 1663 + macvlan_get_size_mac(vlan) /* IFLA_MACVLAN_MACADDR */ 1664 + nla_total_size(4) /* IFLA_MACVLAN_BC_QUEUE_LEN */ 1665 + nla_total_size(4) /* IFLA_MACVLAN_BC_QUEUE_LEN_USED */ 1666 ); 1667 } 1668 1669 static int macvlan_fill_info_macaddr(struct sk_buff *skb, 1670 const struct macvlan_dev *vlan, 1671 const int i) 1672 { 1673 struct hlist_head *h = &vlan->port->vlan_source_hash[i]; 1674 struct macvlan_source_entry *entry; 1675 1676 hlist_for_each_entry_rcu(entry, h, hlist, lockdep_rtnl_is_held()) { 1677 if (entry->vlan != vlan) 1678 continue; 1679 if (nla_put(skb, IFLA_MACVLAN_MACADDR, ETH_ALEN, entry->addr)) 1680 return 1; 1681 } 1682 return 0; 1683 } 1684 1685 static int macvlan_fill_info(struct sk_buff *skb, 1686 const struct net_device *dev) 1687 { 1688 struct macvlan_dev *vlan = netdev_priv(dev); 1689 struct macvlan_port *port = vlan->port; 1690 int i; 1691 struct nlattr *nest; 1692 1693 if (nla_put_u32(skb, IFLA_MACVLAN_MODE, vlan->mode)) 1694 goto nla_put_failure; 1695 if (nla_put_u16(skb, IFLA_MACVLAN_FLAGS, vlan->flags)) 1696 goto nla_put_failure; 1697 if (nla_put_u32(skb, IFLA_MACVLAN_MACADDR_COUNT, vlan->macaddr_count)) 1698 goto nla_put_failure; 1699 if (vlan->macaddr_count > 0) { 1700 nest = nla_nest_start_noflag(skb, IFLA_MACVLAN_MACADDR_DATA); 1701 if (nest == NULL) 1702 goto nla_put_failure; 1703 1704 for (i = 0; i < MACVLAN_HASH_SIZE; i++) { 1705 if (macvlan_fill_info_macaddr(skb, vlan, i)) 1706 goto nla_put_failure; 1707 } 1708 nla_nest_end(skb, nest); 1709 } 1710 if (nla_put_u32(skb, IFLA_MACVLAN_BC_QUEUE_LEN, vlan->bc_queue_len_req)) 1711 goto nla_put_failure; 1712 if (nla_put_u32(skb, IFLA_MACVLAN_BC_QUEUE_LEN_USED, port->bc_queue_len_used)) 1713 goto nla_put_failure; 1714 if (port->bc_cutoff != 1 && 1715 nla_put_s32(skb, IFLA_MACVLAN_BC_CUTOFF, port->bc_cutoff)) 1716 goto nla_put_failure; 1717 return 0; 1718 1719 nla_put_failure: 1720 return -EMSGSIZE; 1721 } 1722 1723 static const struct nla_policy macvlan_policy[IFLA_MACVLAN_MAX + 1] = { 1724 [IFLA_MACVLAN_MODE] = { .type = NLA_U32 }, 1725 [IFLA_MACVLAN_FLAGS] = { .type = NLA_U16 }, 1726 [IFLA_MACVLAN_MACADDR_MODE] = { .type = NLA_U32 }, 1727 [IFLA_MACVLAN_MACADDR] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN }, 1728 [IFLA_MACVLAN_MACADDR_DATA] = { .type = NLA_NESTED }, 1729 [IFLA_MACVLAN_MACADDR_COUNT] = { .type = NLA_U32 }, 1730 [IFLA_MACVLAN_BC_QUEUE_LEN] = { .type = NLA_U32 }, 1731 [IFLA_MACVLAN_BC_QUEUE_LEN_USED] = { .type = NLA_REJECT }, 1732 [IFLA_MACVLAN_BC_CUTOFF] = { .type = NLA_S32 }, 1733 }; 1734 1735 int macvlan_link_register(struct rtnl_link_ops *ops) 1736 { 1737 /* common fields */ 1738 ops->validate = macvlan_validate; 1739 ops->maxtype = IFLA_MACVLAN_MAX; 1740 ops->policy = macvlan_policy; 1741 ops->changelink = macvlan_changelink; 1742 ops->get_size = macvlan_get_size; 1743 ops->fill_info = macvlan_fill_info; 1744 1745 return rtnl_link_register(ops); 1746 }; 1747 EXPORT_SYMBOL_GPL(macvlan_link_register); 1748 1749 static struct net *macvlan_get_link_net(const struct net_device *dev) 1750 { 1751 return dev_net(macvlan_dev_real_dev(dev)); 1752 } 1753 1754 static struct rtnl_link_ops macvlan_link_ops = { 1755 .kind = "macvlan", 1756 .setup = macvlan_setup, 1757 .newlink = macvlan_newlink, 1758 .dellink = macvlan_dellink, 1759 .get_link_net = macvlan_get_link_net, 1760 .priv_size = sizeof(struct macvlan_dev), 1761 }; 1762 1763 static void update_port_bc_queue_len(struct macvlan_port *port) 1764 { 1765 u32 max_bc_queue_len_req = 0; 1766 struct macvlan_dev *vlan; 1767 1768 list_for_each_entry(vlan, &port->vlans, list) { 1769 if (vlan->bc_queue_len_req > max_bc_queue_len_req) 1770 max_bc_queue_len_req = vlan->bc_queue_len_req; 1771 } 1772 port->bc_queue_len_used = max_bc_queue_len_req; 1773 } 1774 1775 static int macvlan_device_event(struct notifier_block *unused, 1776 unsigned long event, void *ptr) 1777 { 1778 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1779 struct macvlan_dev *vlan, *next; 1780 struct macvlan_port *port; 1781 LIST_HEAD(list_kill); 1782 1783 if (!netif_is_macvlan_port(dev)) 1784 return NOTIFY_DONE; 1785 1786 port = macvlan_port_get_rtnl(dev); 1787 1788 switch (event) { 1789 case NETDEV_UP: 1790 case NETDEV_DOWN: 1791 case NETDEV_CHANGE: 1792 list_for_each_entry(vlan, &port->vlans, list) 1793 netif_stacked_transfer_operstate(vlan->lowerdev, 1794 vlan->dev); 1795 break; 1796 case NETDEV_FEAT_CHANGE: 1797 list_for_each_entry(vlan, &port->vlans, list) { 1798 netif_inherit_tso_max(vlan->dev, dev); 1799 netdev_update_features(vlan->dev); 1800 } 1801 break; 1802 case NETDEV_CHANGEMTU: 1803 list_for_each_entry(vlan, &port->vlans, list) { 1804 if (vlan->dev->mtu <= dev->mtu) 1805 continue; 1806 dev_set_mtu(vlan->dev, dev->mtu); 1807 } 1808 break; 1809 case NETDEV_CHANGEADDR: 1810 if (!macvlan_passthru(port)) 1811 return NOTIFY_DONE; 1812 1813 vlan = list_first_entry_or_null(&port->vlans, 1814 struct macvlan_dev, 1815 list); 1816 1817 if (vlan && macvlan_sync_address(vlan->dev, dev->dev_addr)) 1818 return NOTIFY_BAD; 1819 1820 break; 1821 case NETDEV_UNREGISTER: 1822 /* twiddle thumbs on netns device moves */ 1823 if (dev->reg_state != NETREG_UNREGISTERING) 1824 break; 1825 1826 list_for_each_entry_safe(vlan, next, &port->vlans, list) 1827 vlan->dev->rtnl_link_ops->dellink(vlan->dev, &list_kill); 1828 unregister_netdevice_many(&list_kill); 1829 break; 1830 case NETDEV_PRE_TYPE_CHANGE: 1831 /* Forbid underlying device to change its type. */ 1832 return NOTIFY_BAD; 1833 1834 case NETDEV_NOTIFY_PEERS: 1835 case NETDEV_BONDING_FAILOVER: 1836 case NETDEV_RESEND_IGMP: 1837 /* Propagate to all vlans */ 1838 list_for_each_entry(vlan, &port->vlans, list) 1839 call_netdevice_notifiers(event, vlan->dev); 1840 } 1841 return NOTIFY_DONE; 1842 } 1843 1844 static struct notifier_block macvlan_notifier_block __read_mostly = { 1845 .notifier_call = macvlan_device_event, 1846 }; 1847 1848 static int __init macvlan_init_module(void) 1849 { 1850 int err; 1851 1852 register_netdevice_notifier(&macvlan_notifier_block); 1853 1854 err = macvlan_link_register(&macvlan_link_ops); 1855 if (err < 0) 1856 goto err1; 1857 return 0; 1858 err1: 1859 unregister_netdevice_notifier(&macvlan_notifier_block); 1860 return err; 1861 } 1862 1863 static void __exit macvlan_cleanup_module(void) 1864 { 1865 rtnl_link_unregister(&macvlan_link_ops); 1866 unregister_netdevice_notifier(&macvlan_notifier_block); 1867 } 1868 1869 module_init(macvlan_init_module); 1870 module_exit(macvlan_cleanup_module); 1871 1872 MODULE_LICENSE("GPL"); 1873 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>"); 1874 MODULE_DESCRIPTION("Driver for MAC address based VLANs"); 1875 MODULE_ALIAS_RTNL_LINK("macvlan"); 1876