1 /* 2 * originally based on the dummy device. 3 * 4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 5 * Licensed under the GPL. Based on dummy.c, and eql.c devices. 6 * 7 * bonding.c: an Ethernet Bonding driver 8 * 9 * This is useful to talk to a Cisco EtherChannel compatible equipment: 10 * Cisco 5500 11 * Sun Trunking (Solaris) 12 * Alteon AceDirector Trunks 13 * Linux Bonding 14 * and probably many L2 switches ... 15 * 16 * How it works: 17 * ifconfig bond0 ipaddress netmask up 18 * will setup a network device, with an ip address. No mac address 19 * will be assigned at this time. The hw mac address will come from 20 * the first slave bonded to the channel. All slaves will then use 21 * this hw mac address. 22 * 23 * ifconfig bond0 down 24 * will release all slaves, marking them as down. 25 * 26 * ifenslave bond0 eth0 27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 28 * a: be used as initial mac address 29 * b: if a hw mac address already is there, eth0's hw mac address 30 * will then be set from bond0. 31 * 32 */ 33 34 #include <linux/kernel.h> 35 #include <linux/module.h> 36 #include <linux/types.h> 37 #include <linux/fcntl.h> 38 #include <linux/interrupt.h> 39 #include <linux/ptrace.h> 40 #include <linux/ioport.h> 41 #include <linux/in.h> 42 #include <net/ip.h> 43 #include <linux/ip.h> 44 #include <linux/tcp.h> 45 #include <linux/udp.h> 46 #include <linux/slab.h> 47 #include <linux/string.h> 48 #include <linux/init.h> 49 #include <linux/timer.h> 50 #include <linux/socket.h> 51 #include <linux/ctype.h> 52 #include <linux/inet.h> 53 #include <linux/bitops.h> 54 #include <linux/io.h> 55 #include <asm/dma.h> 56 #include <linux/uaccess.h> 57 #include <linux/errno.h> 58 #include <linux/netdevice.h> 59 #include <linux/inetdevice.h> 60 #include <linux/igmp.h> 61 #include <linux/etherdevice.h> 62 #include <linux/skbuff.h> 63 #include <net/sock.h> 64 #include <linux/rtnetlink.h> 65 #include <linux/smp.h> 66 #include <linux/if_ether.h> 67 #include <net/arp.h> 68 #include <linux/mii.h> 69 #include <linux/ethtool.h> 70 #include <linux/if_vlan.h> 71 #include <linux/if_bonding.h> 72 #include <linux/jiffies.h> 73 #include <linux/preempt.h> 74 #include <net/route.h> 75 #include <net/net_namespace.h> 76 #include <net/netns/generic.h> 77 #include <net/pkt_sched.h> 78 #include <linux/rculist.h> 79 #include <net/flow_keys.h> 80 #include <net/bonding.h> 81 #include <net/bond_3ad.h> 82 #include <net/bond_alb.h> 83 84 /*---------------------------- Module parameters ----------------------------*/ 85 86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 87 88 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 89 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 90 static int num_peer_notif = 1; 91 static int miimon; 92 static int updelay; 93 static int downdelay; 94 static int use_carrier = 1; 95 static char *mode; 96 static char *primary; 97 static char *primary_reselect; 98 static char *lacp_rate; 99 static int min_links; 100 static char *ad_select; 101 static char *xmit_hash_policy; 102 static int arp_interval; 103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 104 static char *arp_validate; 105 static char *arp_all_targets; 106 static char *fail_over_mac; 107 static int all_slaves_active; 108 static struct bond_params bonding_defaults; 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 110 static int packets_per_slave = 1; 111 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 112 113 module_param(max_bonds, int, 0); 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 115 module_param(tx_queues, int, 0); 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 117 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 119 "failover event (alias of num_unsol_na)"); 120 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 122 "failover event (alias of num_grat_arp)"); 123 module_param(miimon, int, 0); 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 125 module_param(updelay, int, 0); 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 127 module_param(downdelay, int, 0); 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 129 "in milliseconds"); 130 module_param(use_carrier, int, 0); 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 132 "0 for off, 1 for on (default)"); 133 module_param(mode, charp, 0); 134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 135 "1 for active-backup, 2 for balance-xor, " 136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 137 "6 for balance-alb"); 138 module_param(primary, charp, 0); 139 MODULE_PARM_DESC(primary, "Primary network device to use"); 140 module_param(primary_reselect, charp, 0); 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 142 "once it comes up; " 143 "0 for always (default), " 144 "1 for only if speed of primary is " 145 "better, " 146 "2 for only on active slave " 147 "failure"); 148 module_param(lacp_rate, charp, 0); 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 150 "0 for slow, 1 for fast"); 151 module_param(ad_select, charp, 0); 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; " 153 "0 for stable (default), 1 for bandwidth, " 154 "2 for count"); 155 module_param(min_links, int, 0); 156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 157 158 module_param(xmit_hash_policy, charp, 0); 159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; " 160 "0 for layer 2 (default), 1 for layer 3+4, " 161 "2 for layer 2+3, 3 for encap layer 2+3, " 162 "4 for encap layer 3+4"); 163 module_param(arp_interval, int, 0); 164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 165 module_param_array(arp_ip_target, charp, NULL, 0); 166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 167 module_param(arp_validate, charp, 0); 168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 169 "0 for none (default), 1 for active, " 170 "2 for backup, 3 for all"); 171 module_param(arp_all_targets, charp, 0); 172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 173 module_param(fail_over_mac, charp, 0); 174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 175 "the same MAC; 0 for none (default), " 176 "1 for active, 2 for follow"); 177 module_param(all_slaves_active, int, 0); 178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 179 "by setting active flag for all slaves; " 180 "0 for never (default), 1 for always."); 181 module_param(resend_igmp, int, 0); 182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 183 "link failure"); 184 module_param(packets_per_slave, int, 0); 185 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 186 "mode; 0 for a random slave, 1 packet per " 187 "slave (default), >1 packets per slave."); 188 module_param(lp_interval, uint, 0); 189 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 190 "the bonding driver sends learning packets to " 191 "each slaves peer switch. The default is 1."); 192 193 /*----------------------------- Global variables ----------------------------*/ 194 195 #ifdef CONFIG_NET_POLL_CONTROLLER 196 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 197 #endif 198 199 int bond_net_id __read_mostly; 200 201 static __be32 arp_target[BOND_MAX_ARP_TARGETS]; 202 static int arp_ip_count; 203 static int bond_mode = BOND_MODE_ROUNDROBIN; 204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 205 static int lacp_fast; 206 207 /*-------------------------- Forward declarations ---------------------------*/ 208 209 static int bond_init(struct net_device *bond_dev); 210 static void bond_uninit(struct net_device *bond_dev); 211 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev, 212 struct rtnl_link_stats64 *stats); 213 static void bond_slave_arr_handler(struct work_struct *work); 214 215 /*---------------------------- General routines -----------------------------*/ 216 217 const char *bond_mode_name(int mode) 218 { 219 static const char *names[] = { 220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 222 [BOND_MODE_XOR] = "load balancing (xor)", 223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 225 [BOND_MODE_TLB] = "transmit load balancing", 226 [BOND_MODE_ALB] = "adaptive load balancing", 227 }; 228 229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 230 return "unknown"; 231 232 return names[mode]; 233 } 234 235 /*---------------------------------- VLAN -----------------------------------*/ 236 237 /** 238 * bond_dev_queue_xmit - Prepare skb for xmit. 239 * 240 * @bond: bond device that got this skb for tx. 241 * @skb: hw accel VLAN tagged skb to transmit 242 * @slave_dev: slave that is supposed to xmit this skbuff 243 */ 244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 245 struct net_device *slave_dev) 246 { 247 skb->dev = slave_dev; 248 249 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 251 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping; 252 253 if (unlikely(netpoll_tx_running(bond->dev))) 254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 255 else 256 dev_queue_xmit(skb); 257 } 258 259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 260 * We don't protect the slave list iteration with a lock because: 261 * a. This operation is performed in IOCTL context, 262 * b. The operation is protected by the RTNL semaphore in the 8021q code, 263 * c. Holding a lock with BH disabled while directly calling a base driver 264 * entry point is generally a BAD idea. 265 * 266 * The design of synchronization/protection for this operation in the 8021q 267 * module is good for one or more VLAN devices over a single physical device 268 * and cannot be extended for a teaming solution like bonding, so there is a 269 * potential race condition here where a net device from the vlan group might 270 * be referenced (either by a base driver or the 8021q code) while it is being 271 * removed from the system. However, it turns out we're not making matters 272 * worse, and if it works for regular VLAN usage it will work here too. 273 */ 274 275 /** 276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 277 * @bond_dev: bonding net device that got called 278 * @vid: vlan id being added 279 */ 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 281 __be16 proto, u16 vid) 282 { 283 struct bonding *bond = netdev_priv(bond_dev); 284 struct slave *slave, *rollback_slave; 285 struct list_head *iter; 286 int res; 287 288 bond_for_each_slave(bond, slave, iter) { 289 res = vlan_vid_add(slave->dev, proto, vid); 290 if (res) 291 goto unwind; 292 } 293 294 return 0; 295 296 unwind: 297 /* unwind to the slave that failed */ 298 bond_for_each_slave(bond, rollback_slave, iter) { 299 if (rollback_slave == slave) 300 break; 301 302 vlan_vid_del(rollback_slave->dev, proto, vid); 303 } 304 305 return res; 306 } 307 308 /** 309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 310 * @bond_dev: bonding net device that got called 311 * @vid: vlan id being removed 312 */ 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 314 __be16 proto, u16 vid) 315 { 316 struct bonding *bond = netdev_priv(bond_dev); 317 struct list_head *iter; 318 struct slave *slave; 319 320 bond_for_each_slave(bond, slave, iter) 321 vlan_vid_del(slave->dev, proto, vid); 322 323 if (bond_is_lb(bond)) 324 bond_alb_clear_vlan(bond, vid); 325 326 return 0; 327 } 328 329 /*------------------------------- Link status -------------------------------*/ 330 331 /* Set the carrier state for the master according to the state of its 332 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 333 * do special 802.3ad magic. 334 * 335 * Returns zero if carrier state does not change, nonzero if it does. 336 */ 337 static int bond_set_carrier(struct bonding *bond) 338 { 339 struct list_head *iter; 340 struct slave *slave; 341 342 if (!bond_has_slaves(bond)) 343 goto down; 344 345 if (BOND_MODE(bond) == BOND_MODE_8023AD) 346 return bond_3ad_set_carrier(bond); 347 348 bond_for_each_slave(bond, slave, iter) { 349 if (slave->link == BOND_LINK_UP) { 350 if (!netif_carrier_ok(bond->dev)) { 351 netif_carrier_on(bond->dev); 352 return 1; 353 } 354 return 0; 355 } 356 } 357 358 down: 359 if (netif_carrier_ok(bond->dev)) { 360 netif_carrier_off(bond->dev); 361 return 1; 362 } 363 return 0; 364 } 365 366 /* Get link speed and duplex from the slave's base driver 367 * using ethtool. If for some reason the call fails or the 368 * values are invalid, set speed and duplex to -1, 369 * and return. 370 */ 371 static void bond_update_speed_duplex(struct slave *slave) 372 { 373 struct net_device *slave_dev = slave->dev; 374 struct ethtool_cmd ecmd; 375 u32 slave_speed; 376 int res; 377 378 slave->speed = SPEED_UNKNOWN; 379 slave->duplex = DUPLEX_UNKNOWN; 380 381 res = __ethtool_get_settings(slave_dev, &ecmd); 382 if (res < 0) 383 return; 384 385 slave_speed = ethtool_cmd_speed(&ecmd); 386 if (slave_speed == 0 || slave_speed == ((__u32) -1)) 387 return; 388 389 switch (ecmd.duplex) { 390 case DUPLEX_FULL: 391 case DUPLEX_HALF: 392 break; 393 default: 394 return; 395 } 396 397 slave->speed = slave_speed; 398 slave->duplex = ecmd.duplex; 399 400 return; 401 } 402 403 const char *bond_slave_link_status(s8 link) 404 { 405 switch (link) { 406 case BOND_LINK_UP: 407 return "up"; 408 case BOND_LINK_FAIL: 409 return "going down"; 410 case BOND_LINK_DOWN: 411 return "down"; 412 case BOND_LINK_BACK: 413 return "going back"; 414 default: 415 return "unknown"; 416 } 417 } 418 419 /* if <dev> supports MII link status reporting, check its link status. 420 * 421 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 422 * depending upon the setting of the use_carrier parameter. 423 * 424 * Return either BMSR_LSTATUS, meaning that the link is up (or we 425 * can't tell and just pretend it is), or 0, meaning that the link is 426 * down. 427 * 428 * If reporting is non-zero, instead of faking link up, return -1 if 429 * both ETHTOOL and MII ioctls fail (meaning the device does not 430 * support them). If use_carrier is set, return whatever it says. 431 * It'd be nice if there was a good way to tell if a driver supports 432 * netif_carrier, but there really isn't. 433 */ 434 static int bond_check_dev_link(struct bonding *bond, 435 struct net_device *slave_dev, int reporting) 436 { 437 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 438 int (*ioctl)(struct net_device *, struct ifreq *, int); 439 struct ifreq ifr; 440 struct mii_ioctl_data *mii; 441 442 if (!reporting && !netif_running(slave_dev)) 443 return 0; 444 445 if (bond->params.use_carrier) 446 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 447 448 /* Try to get link status using Ethtool first. */ 449 if (slave_dev->ethtool_ops->get_link) 450 return slave_dev->ethtool_ops->get_link(slave_dev) ? 451 BMSR_LSTATUS : 0; 452 453 /* Ethtool can't be used, fallback to MII ioctls. */ 454 ioctl = slave_ops->ndo_do_ioctl; 455 if (ioctl) { 456 /* TODO: set pointer to correct ioctl on a per team member 457 * bases to make this more efficient. that is, once 458 * we determine the correct ioctl, we will always 459 * call it and not the others for that team 460 * member. 461 */ 462 463 /* We cannot assume that SIOCGMIIPHY will also read a 464 * register; not all network drivers (e.g., e100) 465 * support that. 466 */ 467 468 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 469 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 470 mii = if_mii(&ifr); 471 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 472 mii->reg_num = MII_BMSR; 473 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) 474 return mii->val_out & BMSR_LSTATUS; 475 } 476 } 477 478 /* If reporting, report that either there's no dev->do_ioctl, 479 * or both SIOCGMIIREG and get_link failed (meaning that we 480 * cannot report link status). If not reporting, pretend 481 * we're ok. 482 */ 483 return reporting ? -1 : BMSR_LSTATUS; 484 } 485 486 /*----------------------------- Multicast list ------------------------------*/ 487 488 /* Push the promiscuity flag down to appropriate slaves */ 489 static int bond_set_promiscuity(struct bonding *bond, int inc) 490 { 491 struct list_head *iter; 492 int err = 0; 493 494 if (bond_uses_primary(bond)) { 495 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 496 497 if (curr_active) 498 err = dev_set_promiscuity(curr_active->dev, inc); 499 } else { 500 struct slave *slave; 501 502 bond_for_each_slave(bond, slave, iter) { 503 err = dev_set_promiscuity(slave->dev, inc); 504 if (err) 505 return err; 506 } 507 } 508 return err; 509 } 510 511 /* Push the allmulti flag down to all slaves */ 512 static int bond_set_allmulti(struct bonding *bond, int inc) 513 { 514 struct list_head *iter; 515 int err = 0; 516 517 if (bond_uses_primary(bond)) { 518 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 519 520 if (curr_active) 521 err = dev_set_allmulti(curr_active->dev, inc); 522 } else { 523 struct slave *slave; 524 525 bond_for_each_slave(bond, slave, iter) { 526 err = dev_set_allmulti(slave->dev, inc); 527 if (err) 528 return err; 529 } 530 } 531 return err; 532 } 533 534 /* Retrieve the list of registered multicast addresses for the bonding 535 * device and retransmit an IGMP JOIN request to the current active 536 * slave. 537 */ 538 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 539 { 540 struct bonding *bond = container_of(work, struct bonding, 541 mcast_work.work); 542 543 if (!rtnl_trylock()) { 544 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 545 return; 546 } 547 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 548 549 if (bond->igmp_retrans > 1) { 550 bond->igmp_retrans--; 551 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 552 } 553 rtnl_unlock(); 554 } 555 556 /* Flush bond's hardware addresses from slave */ 557 static void bond_hw_addr_flush(struct net_device *bond_dev, 558 struct net_device *slave_dev) 559 { 560 struct bonding *bond = netdev_priv(bond_dev); 561 562 dev_uc_unsync(slave_dev, bond_dev); 563 dev_mc_unsync(slave_dev, bond_dev); 564 565 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 566 /* del lacpdu mc addr from mc list */ 567 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 568 569 dev_mc_del(slave_dev, lacpdu_multicast); 570 } 571 } 572 573 /*--------------------------- Active slave change ---------------------------*/ 574 575 /* Update the hardware address list and promisc/allmulti for the new and 576 * old active slaves (if any). Modes that are not using primary keep all 577 * slaves up date at all times; only the modes that use primary need to call 578 * this function to swap these settings during a failover. 579 */ 580 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 581 struct slave *old_active) 582 { 583 if (old_active) { 584 if (bond->dev->flags & IFF_PROMISC) 585 dev_set_promiscuity(old_active->dev, -1); 586 587 if (bond->dev->flags & IFF_ALLMULTI) 588 dev_set_allmulti(old_active->dev, -1); 589 590 bond_hw_addr_flush(bond->dev, old_active->dev); 591 } 592 593 if (new_active) { 594 /* FIXME: Signal errors upstream. */ 595 if (bond->dev->flags & IFF_PROMISC) 596 dev_set_promiscuity(new_active->dev, 1); 597 598 if (bond->dev->flags & IFF_ALLMULTI) 599 dev_set_allmulti(new_active->dev, 1); 600 601 netif_addr_lock_bh(bond->dev); 602 dev_uc_sync(new_active->dev, bond->dev); 603 dev_mc_sync(new_active->dev, bond->dev); 604 netif_addr_unlock_bh(bond->dev); 605 } 606 } 607 608 /** 609 * bond_set_dev_addr - clone slave's address to bond 610 * @bond_dev: bond net device 611 * @slave_dev: slave net device 612 * 613 * Should be called with RTNL held. 614 */ 615 static void bond_set_dev_addr(struct net_device *bond_dev, 616 struct net_device *slave_dev) 617 { 618 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 619 bond_dev, slave_dev, slave_dev->addr_len); 620 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 621 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 622 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 623 } 624 625 /* bond_do_fail_over_mac 626 * 627 * Perform special MAC address swapping for fail_over_mac settings 628 * 629 * Called with RTNL 630 */ 631 static void bond_do_fail_over_mac(struct bonding *bond, 632 struct slave *new_active, 633 struct slave *old_active) 634 { 635 u8 tmp_mac[ETH_ALEN]; 636 struct sockaddr saddr; 637 int rv; 638 639 switch (bond->params.fail_over_mac) { 640 case BOND_FOM_ACTIVE: 641 if (new_active) 642 bond_set_dev_addr(bond->dev, new_active->dev); 643 break; 644 case BOND_FOM_FOLLOW: 645 /* if new_active && old_active, swap them 646 * if just old_active, do nothing (going to no active slave) 647 * if just new_active, set new_active to bond's MAC 648 */ 649 if (!new_active) 650 return; 651 652 if (old_active) { 653 ether_addr_copy(tmp_mac, new_active->dev->dev_addr); 654 ether_addr_copy(saddr.sa_data, 655 old_active->dev->dev_addr); 656 saddr.sa_family = new_active->dev->type; 657 } else { 658 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr); 659 saddr.sa_family = bond->dev->type; 660 } 661 662 rv = dev_set_mac_address(new_active->dev, &saddr); 663 if (rv) { 664 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 665 -rv, new_active->dev->name); 666 goto out; 667 } 668 669 if (!old_active) 670 goto out; 671 672 ether_addr_copy(saddr.sa_data, tmp_mac); 673 saddr.sa_family = old_active->dev->type; 674 675 rv = dev_set_mac_address(old_active->dev, &saddr); 676 if (rv) 677 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 678 -rv, new_active->dev->name); 679 out: 680 break; 681 default: 682 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 683 bond->params.fail_over_mac); 684 break; 685 } 686 687 } 688 689 static bool bond_should_change_active(struct bonding *bond) 690 { 691 struct slave *prim = rtnl_dereference(bond->primary_slave); 692 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 693 694 if (!prim || !curr || curr->link != BOND_LINK_UP) 695 return true; 696 if (bond->force_primary) { 697 bond->force_primary = false; 698 return true; 699 } 700 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER && 701 (prim->speed < curr->speed || 702 (prim->speed == curr->speed && prim->duplex <= curr->duplex))) 703 return false; 704 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE) 705 return false; 706 return true; 707 } 708 709 /** 710 * find_best_interface - select the best available slave to be the active one 711 * @bond: our bonding struct 712 */ 713 static struct slave *bond_find_best_slave(struct bonding *bond) 714 { 715 struct slave *slave, *bestslave = NULL, *primary; 716 struct list_head *iter; 717 int mintime = bond->params.updelay; 718 719 primary = rtnl_dereference(bond->primary_slave); 720 if (primary && primary->link == BOND_LINK_UP && 721 bond_should_change_active(bond)) 722 return primary; 723 724 bond_for_each_slave(bond, slave, iter) { 725 if (slave->link == BOND_LINK_UP) 726 return slave; 727 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 728 slave->delay < mintime) { 729 mintime = slave->delay; 730 bestslave = slave; 731 } 732 } 733 734 return bestslave; 735 } 736 737 static bool bond_should_notify_peers(struct bonding *bond) 738 { 739 struct slave *slave; 740 741 rcu_read_lock(); 742 slave = rcu_dereference(bond->curr_active_slave); 743 rcu_read_unlock(); 744 745 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 746 slave ? slave->dev->name : "NULL"); 747 748 if (!slave || !bond->send_peer_notif || 749 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 750 return false; 751 752 return true; 753 } 754 755 /** 756 * change_active_interface - change the active slave into the specified one 757 * @bond: our bonding struct 758 * @new: the new slave to make the active one 759 * 760 * Set the new slave to the bond's settings and unset them on the old 761 * curr_active_slave. 762 * Setting include flags, mc-list, promiscuity, allmulti, etc. 763 * 764 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 765 * because it is apparently the best available slave we have, even though its 766 * updelay hasn't timed out yet. 767 * 768 * Caller must hold RTNL. 769 */ 770 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 771 { 772 struct slave *old_active; 773 774 ASSERT_RTNL(); 775 776 old_active = rtnl_dereference(bond->curr_active_slave); 777 778 if (old_active == new_active) 779 return; 780 781 if (new_active) { 782 new_active->last_link_up = jiffies; 783 784 if (new_active->link == BOND_LINK_BACK) { 785 if (bond_uses_primary(bond)) { 786 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n", 787 new_active->dev->name, 788 (bond->params.updelay - new_active->delay) * bond->params.miimon); 789 } 790 791 new_active->delay = 0; 792 new_active->link = BOND_LINK_UP; 793 794 if (BOND_MODE(bond) == BOND_MODE_8023AD) 795 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 796 797 if (bond_is_lb(bond)) 798 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 799 } else { 800 if (bond_uses_primary(bond)) { 801 netdev_info(bond->dev, "making interface %s the new active one\n", 802 new_active->dev->name); 803 } 804 } 805 } 806 807 if (bond_uses_primary(bond)) 808 bond_hw_addr_swap(bond, new_active, old_active); 809 810 if (bond_is_lb(bond)) { 811 bond_alb_handle_active_change(bond, new_active); 812 if (old_active) 813 bond_set_slave_inactive_flags(old_active, 814 BOND_SLAVE_NOTIFY_NOW); 815 if (new_active) 816 bond_set_slave_active_flags(new_active, 817 BOND_SLAVE_NOTIFY_NOW); 818 } else { 819 rcu_assign_pointer(bond->curr_active_slave, new_active); 820 } 821 822 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 823 if (old_active) 824 bond_set_slave_inactive_flags(old_active, 825 BOND_SLAVE_NOTIFY_NOW); 826 827 if (new_active) { 828 bool should_notify_peers = false; 829 830 bond_set_slave_active_flags(new_active, 831 BOND_SLAVE_NOTIFY_NOW); 832 833 if (bond->params.fail_over_mac) 834 bond_do_fail_over_mac(bond, new_active, 835 old_active); 836 837 if (netif_running(bond->dev)) { 838 bond->send_peer_notif = 839 bond->params.num_peer_notif; 840 should_notify_peers = 841 bond_should_notify_peers(bond); 842 } 843 844 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 845 if (should_notify_peers) 846 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 847 bond->dev); 848 } 849 } 850 851 /* resend IGMP joins since active slave has changed or 852 * all were sent on curr_active_slave. 853 * resend only if bond is brought up with the affected 854 * bonding modes and the retransmission is enabled 855 */ 856 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 857 ((bond_uses_primary(bond) && new_active) || 858 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 859 bond->igmp_retrans = bond->params.resend_igmp; 860 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 861 } 862 } 863 864 /** 865 * bond_select_active_slave - select a new active slave, if needed 866 * @bond: our bonding struct 867 * 868 * This functions should be called when one of the following occurs: 869 * - The old curr_active_slave has been released or lost its link. 870 * - The primary_slave has got its link back. 871 * - A slave has got its link back and there's no old curr_active_slave. 872 * 873 * Caller must hold RTNL. 874 */ 875 void bond_select_active_slave(struct bonding *bond) 876 { 877 struct slave *best_slave; 878 int rv; 879 880 ASSERT_RTNL(); 881 882 best_slave = bond_find_best_slave(bond); 883 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 884 bond_change_active_slave(bond, best_slave); 885 rv = bond_set_carrier(bond); 886 if (!rv) 887 return; 888 889 if (netif_carrier_ok(bond->dev)) { 890 netdev_info(bond->dev, "first active interface up!\n"); 891 } else { 892 netdev_info(bond->dev, "now running without any active interface!\n"); 893 } 894 } 895 } 896 897 #ifdef CONFIG_NET_POLL_CONTROLLER 898 static inline int slave_enable_netpoll(struct slave *slave) 899 { 900 struct netpoll *np; 901 int err = 0; 902 903 np = kzalloc(sizeof(*np), GFP_KERNEL); 904 err = -ENOMEM; 905 if (!np) 906 goto out; 907 908 err = __netpoll_setup(np, slave->dev); 909 if (err) { 910 kfree(np); 911 goto out; 912 } 913 slave->np = np; 914 out: 915 return err; 916 } 917 static inline void slave_disable_netpoll(struct slave *slave) 918 { 919 struct netpoll *np = slave->np; 920 921 if (!np) 922 return; 923 924 slave->np = NULL; 925 __netpoll_free_async(np); 926 } 927 928 static void bond_poll_controller(struct net_device *bond_dev) 929 { 930 } 931 932 static void bond_netpoll_cleanup(struct net_device *bond_dev) 933 { 934 struct bonding *bond = netdev_priv(bond_dev); 935 struct list_head *iter; 936 struct slave *slave; 937 938 bond_for_each_slave(bond, slave, iter) 939 if (bond_slave_is_up(slave)) 940 slave_disable_netpoll(slave); 941 } 942 943 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 944 { 945 struct bonding *bond = netdev_priv(dev); 946 struct list_head *iter; 947 struct slave *slave; 948 int err = 0; 949 950 bond_for_each_slave(bond, slave, iter) { 951 err = slave_enable_netpoll(slave); 952 if (err) { 953 bond_netpoll_cleanup(dev); 954 break; 955 } 956 } 957 return err; 958 } 959 #else 960 static inline int slave_enable_netpoll(struct slave *slave) 961 { 962 return 0; 963 } 964 static inline void slave_disable_netpoll(struct slave *slave) 965 { 966 } 967 static void bond_netpoll_cleanup(struct net_device *bond_dev) 968 { 969 } 970 #endif 971 972 /*---------------------------------- IOCTL ----------------------------------*/ 973 974 static netdev_features_t bond_fix_features(struct net_device *dev, 975 netdev_features_t features) 976 { 977 struct bonding *bond = netdev_priv(dev); 978 struct list_head *iter; 979 netdev_features_t mask; 980 struct slave *slave; 981 982 mask = features; 983 features &= ~NETIF_F_ONE_FOR_ALL; 984 features |= NETIF_F_ALL_FOR_ALL; 985 986 bond_for_each_slave(bond, slave, iter) { 987 features = netdev_increment_features(features, 988 slave->dev->features, 989 mask); 990 } 991 features = netdev_add_tso_features(features, mask); 992 993 return features; 994 } 995 996 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \ 997 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 998 NETIF_F_HIGHDMA | NETIF_F_LRO) 999 1000 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\ 1001 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL) 1002 1003 static void bond_compute_features(struct bonding *bond) 1004 { 1005 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1006 IFF_XMIT_DST_RELEASE_PERM; 1007 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1008 netdev_features_t enc_features = BOND_ENC_FEATURES; 1009 struct net_device *bond_dev = bond->dev; 1010 struct list_head *iter; 1011 struct slave *slave; 1012 unsigned short max_hard_header_len = ETH_HLEN; 1013 unsigned int gso_max_size = GSO_MAX_SIZE; 1014 u16 gso_max_segs = GSO_MAX_SEGS; 1015 1016 if (!bond_has_slaves(bond)) 1017 goto done; 1018 vlan_features &= NETIF_F_ALL_FOR_ALL; 1019 1020 bond_for_each_slave(bond, slave, iter) { 1021 vlan_features = netdev_increment_features(vlan_features, 1022 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1023 1024 enc_features = netdev_increment_features(enc_features, 1025 slave->dev->hw_enc_features, 1026 BOND_ENC_FEATURES); 1027 dst_release_flag &= slave->dev->priv_flags; 1028 if (slave->dev->hard_header_len > max_hard_header_len) 1029 max_hard_header_len = slave->dev->hard_header_len; 1030 1031 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1032 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1033 } 1034 1035 done: 1036 bond_dev->vlan_features = vlan_features; 1037 bond_dev->hw_enc_features = enc_features; 1038 bond_dev->hard_header_len = max_hard_header_len; 1039 bond_dev->gso_max_segs = gso_max_segs; 1040 netif_set_gso_max_size(bond_dev, gso_max_size); 1041 1042 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1043 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1044 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1045 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1046 1047 netdev_change_features(bond_dev); 1048 } 1049 1050 static void bond_setup_by_slave(struct net_device *bond_dev, 1051 struct net_device *slave_dev) 1052 { 1053 bond_dev->header_ops = slave_dev->header_ops; 1054 1055 bond_dev->type = slave_dev->type; 1056 bond_dev->hard_header_len = slave_dev->hard_header_len; 1057 bond_dev->addr_len = slave_dev->addr_len; 1058 1059 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1060 slave_dev->addr_len); 1061 } 1062 1063 /* On bonding slaves other than the currently active slave, suppress 1064 * duplicates except for alb non-mcast/bcast. 1065 */ 1066 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1067 struct slave *slave, 1068 struct bonding *bond) 1069 { 1070 if (bond_is_slave_inactive(slave)) { 1071 if (BOND_MODE(bond) == BOND_MODE_ALB && 1072 skb->pkt_type != PACKET_BROADCAST && 1073 skb->pkt_type != PACKET_MULTICAST) 1074 return false; 1075 return true; 1076 } 1077 return false; 1078 } 1079 1080 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1081 { 1082 struct sk_buff *skb = *pskb; 1083 struct slave *slave; 1084 struct bonding *bond; 1085 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1086 struct slave *); 1087 int ret = RX_HANDLER_ANOTHER; 1088 1089 skb = skb_share_check(skb, GFP_ATOMIC); 1090 if (unlikely(!skb)) 1091 return RX_HANDLER_CONSUMED; 1092 1093 *pskb = skb; 1094 1095 slave = bond_slave_get_rcu(skb->dev); 1096 bond = slave->bond; 1097 1098 recv_probe = ACCESS_ONCE(bond->recv_probe); 1099 if (recv_probe) { 1100 ret = recv_probe(skb, bond, slave); 1101 if (ret == RX_HANDLER_CONSUMED) { 1102 consume_skb(skb); 1103 return ret; 1104 } 1105 } 1106 1107 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1108 return RX_HANDLER_EXACT; 1109 } 1110 1111 skb->dev = bond->dev; 1112 1113 if (BOND_MODE(bond) == BOND_MODE_ALB && 1114 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1115 skb->pkt_type == PACKET_HOST) { 1116 1117 if (unlikely(skb_cow_head(skb, 1118 skb->data - skb_mac_header(skb)))) { 1119 kfree_skb(skb); 1120 return RX_HANDLER_CONSUMED; 1121 } 1122 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr); 1123 } 1124 1125 return ret; 1126 } 1127 1128 static int bond_master_upper_dev_link(struct net_device *bond_dev, 1129 struct net_device *slave_dev, 1130 struct slave *slave) 1131 { 1132 int err; 1133 1134 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave); 1135 if (err) 1136 return err; 1137 slave_dev->flags |= IFF_SLAVE; 1138 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL); 1139 return 0; 1140 } 1141 1142 static void bond_upper_dev_unlink(struct net_device *bond_dev, 1143 struct net_device *slave_dev) 1144 { 1145 netdev_upper_dev_unlink(slave_dev, bond_dev); 1146 slave_dev->flags &= ~IFF_SLAVE; 1147 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL); 1148 } 1149 1150 static struct slave *bond_alloc_slave(struct bonding *bond) 1151 { 1152 struct slave *slave = NULL; 1153 1154 slave = kzalloc(sizeof(struct slave), GFP_KERNEL); 1155 if (!slave) 1156 return NULL; 1157 1158 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1159 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1160 GFP_KERNEL); 1161 if (!SLAVE_AD_INFO(slave)) { 1162 kfree(slave); 1163 return NULL; 1164 } 1165 } 1166 return slave; 1167 } 1168 1169 static void bond_free_slave(struct slave *slave) 1170 { 1171 struct bonding *bond = bond_get_bond_by_slave(slave); 1172 1173 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1174 kfree(SLAVE_AD_INFO(slave)); 1175 1176 kfree(slave); 1177 } 1178 1179 /* enslave device <slave> to bond device <master> */ 1180 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev) 1181 { 1182 struct bonding *bond = netdev_priv(bond_dev); 1183 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1184 struct slave *new_slave = NULL, *prev_slave; 1185 struct sockaddr addr; 1186 int link_reporting; 1187 int res = 0, i; 1188 1189 if (!bond->params.use_carrier && 1190 slave_dev->ethtool_ops->get_link == NULL && 1191 slave_ops->ndo_do_ioctl == NULL) { 1192 netdev_warn(bond_dev, "no link monitoring support for %s\n", 1193 slave_dev->name); 1194 } 1195 1196 /* already enslaved */ 1197 if (slave_dev->flags & IFF_SLAVE) { 1198 netdev_dbg(bond_dev, "Error: Device was already enslaved\n"); 1199 return -EBUSY; 1200 } 1201 1202 if (bond_dev == slave_dev) { 1203 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1204 return -EPERM; 1205 } 1206 1207 /* vlan challenged mutual exclusion */ 1208 /* no need to lock since we're protected by rtnl_lock */ 1209 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1210 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n", 1211 slave_dev->name); 1212 if (vlan_uses_dev(bond_dev)) { 1213 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n", 1214 slave_dev->name, bond_dev->name); 1215 return -EPERM; 1216 } else { 1217 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n", 1218 slave_dev->name, slave_dev->name, 1219 bond_dev->name); 1220 } 1221 } else { 1222 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n", 1223 slave_dev->name); 1224 } 1225 1226 /* Old ifenslave binaries are no longer supported. These can 1227 * be identified with moderate accuracy by the state of the slave: 1228 * the current ifenslave will set the interface down prior to 1229 * enslaving it; the old ifenslave will not. 1230 */ 1231 if ((slave_dev->flags & IFF_UP)) { 1232 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n", 1233 slave_dev->name); 1234 res = -EPERM; 1235 goto err_undo_flags; 1236 } 1237 1238 /* set bonding device ether type by slave - bonding netdevices are 1239 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1240 * there is a need to override some of the type dependent attribs/funcs. 1241 * 1242 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1243 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1244 */ 1245 if (!bond_has_slaves(bond)) { 1246 if (bond_dev->type != slave_dev->type) { 1247 netdev_dbg(bond_dev, "change device type from %d to %d\n", 1248 bond_dev->type, slave_dev->type); 1249 1250 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1251 bond_dev); 1252 res = notifier_to_errno(res); 1253 if (res) { 1254 netdev_err(bond_dev, "refused to change device type\n"); 1255 res = -EBUSY; 1256 goto err_undo_flags; 1257 } 1258 1259 /* Flush unicast and multicast addresses */ 1260 dev_uc_flush(bond_dev); 1261 dev_mc_flush(bond_dev); 1262 1263 if (slave_dev->type != ARPHRD_ETHER) 1264 bond_setup_by_slave(bond_dev, slave_dev); 1265 else { 1266 ether_setup(bond_dev); 1267 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1268 } 1269 1270 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1271 bond_dev); 1272 } 1273 } else if (bond_dev->type != slave_dev->type) { 1274 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n", 1275 slave_dev->name, slave_dev->type, bond_dev->type); 1276 res = -EINVAL; 1277 goto err_undo_flags; 1278 } 1279 1280 if (slave_ops->ndo_set_mac_address == NULL) { 1281 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n"); 1282 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1283 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1284 if (!bond_has_slaves(bond)) { 1285 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1286 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1287 } else { 1288 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n"); 1289 res = -EOPNOTSUPP; 1290 goto err_undo_flags; 1291 } 1292 } 1293 } 1294 1295 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1296 1297 /* If this is the first slave, then we need to set the master's hardware 1298 * address to be the same as the slave's. 1299 */ 1300 if (!bond_has_slaves(bond) && 1301 bond->dev->addr_assign_type == NET_ADDR_RANDOM) 1302 bond_set_dev_addr(bond->dev, slave_dev); 1303 1304 new_slave = bond_alloc_slave(bond); 1305 if (!new_slave) { 1306 res = -ENOMEM; 1307 goto err_undo_flags; 1308 } 1309 1310 new_slave->bond = bond; 1311 new_slave->dev = slave_dev; 1312 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1313 * is set via sysfs or module option if desired. 1314 */ 1315 new_slave->queue_id = 0; 1316 1317 /* Save slave's original mtu and then set it to match the bond */ 1318 new_slave->original_mtu = slave_dev->mtu; 1319 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1320 if (res) { 1321 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res); 1322 goto err_free; 1323 } 1324 1325 /* Save slave's original ("permanent") mac address for modes 1326 * that need it, and for restoring it upon release, and then 1327 * set it to the master's address 1328 */ 1329 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr); 1330 1331 if (!bond->params.fail_over_mac || 1332 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1333 /* Set slave to master's mac address. The application already 1334 * set the master's mac address to that of the first slave 1335 */ 1336 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len); 1337 addr.sa_family = slave_dev->type; 1338 res = dev_set_mac_address(slave_dev, &addr); 1339 if (res) { 1340 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res); 1341 goto err_restore_mtu; 1342 } 1343 } 1344 1345 /* open the slave since the application closed it */ 1346 res = dev_open(slave_dev); 1347 if (res) { 1348 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name); 1349 goto err_restore_mac; 1350 } 1351 1352 slave_dev->priv_flags |= IFF_BONDING; 1353 /* initialize slave stats */ 1354 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1355 1356 if (bond_is_lb(bond)) { 1357 /* bond_alb_init_slave() must be called before all other stages since 1358 * it might fail and we do not want to have to undo everything 1359 */ 1360 res = bond_alb_init_slave(bond, new_slave); 1361 if (res) 1362 goto err_close; 1363 } 1364 1365 /* If the mode uses primary, then the following is handled by 1366 * bond_change_active_slave(). 1367 */ 1368 if (!bond_uses_primary(bond)) { 1369 /* set promiscuity level to new slave */ 1370 if (bond_dev->flags & IFF_PROMISC) { 1371 res = dev_set_promiscuity(slave_dev, 1); 1372 if (res) 1373 goto err_close; 1374 } 1375 1376 /* set allmulti level to new slave */ 1377 if (bond_dev->flags & IFF_ALLMULTI) { 1378 res = dev_set_allmulti(slave_dev, 1); 1379 if (res) 1380 goto err_close; 1381 } 1382 1383 netif_addr_lock_bh(bond_dev); 1384 1385 dev_mc_sync_multiple(slave_dev, bond_dev); 1386 dev_uc_sync_multiple(slave_dev, bond_dev); 1387 1388 netif_addr_unlock_bh(bond_dev); 1389 } 1390 1391 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1392 /* add lacpdu mc addr to mc list */ 1393 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1394 1395 dev_mc_add(slave_dev, lacpdu_multicast); 1396 } 1397 1398 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1399 if (res) { 1400 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n", 1401 slave_dev->name); 1402 goto err_close; 1403 } 1404 1405 prev_slave = bond_last_slave(bond); 1406 1407 new_slave->delay = 0; 1408 new_slave->link_failure_count = 0; 1409 1410 bond_update_speed_duplex(new_slave); 1411 1412 new_slave->last_rx = jiffies - 1413 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1414 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1415 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1416 1417 if (bond->params.miimon && !bond->params.use_carrier) { 1418 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1419 1420 if ((link_reporting == -1) && !bond->params.arp_interval) { 1421 /* miimon is set but a bonded network driver 1422 * does not support ETHTOOL/MII and 1423 * arp_interval is not set. Note: if 1424 * use_carrier is enabled, we will never go 1425 * here (because netif_carrier is always 1426 * supported); thus, we don't need to change 1427 * the messages for netif_carrier. 1428 */ 1429 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n", 1430 slave_dev->name); 1431 } else if (link_reporting == -1) { 1432 /* unable get link status using mii/ethtool */ 1433 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n", 1434 slave_dev->name); 1435 } 1436 } 1437 1438 /* check for initial state */ 1439 if (bond->params.miimon) { 1440 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1441 if (bond->params.updelay) { 1442 new_slave->link = BOND_LINK_BACK; 1443 new_slave->delay = bond->params.updelay; 1444 } else { 1445 new_slave->link = BOND_LINK_UP; 1446 } 1447 } else { 1448 new_slave->link = BOND_LINK_DOWN; 1449 } 1450 } else if (bond->params.arp_interval) { 1451 new_slave->link = (netif_carrier_ok(slave_dev) ? 1452 BOND_LINK_UP : BOND_LINK_DOWN); 1453 } else { 1454 new_slave->link = BOND_LINK_UP; 1455 } 1456 1457 if (new_slave->link != BOND_LINK_DOWN) 1458 new_slave->last_link_up = jiffies; 1459 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n", 1460 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1461 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1462 1463 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1464 /* if there is a primary slave, remember it */ 1465 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1466 rcu_assign_pointer(bond->primary_slave, new_slave); 1467 bond->force_primary = true; 1468 } 1469 } 1470 1471 switch (BOND_MODE(bond)) { 1472 case BOND_MODE_ACTIVEBACKUP: 1473 bond_set_slave_inactive_flags(new_slave, 1474 BOND_SLAVE_NOTIFY_NOW); 1475 break; 1476 case BOND_MODE_8023AD: 1477 /* in 802.3ad mode, the internal mechanism 1478 * will activate the slaves in the selected 1479 * aggregator 1480 */ 1481 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1482 /* if this is the first slave */ 1483 if (!prev_slave) { 1484 SLAVE_AD_INFO(new_slave)->id = 1; 1485 /* Initialize AD with the number of times that the AD timer is called in 1 second 1486 * can be called only after the mac address of the bond is set 1487 */ 1488 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1489 } else { 1490 SLAVE_AD_INFO(new_slave)->id = 1491 SLAVE_AD_INFO(prev_slave)->id + 1; 1492 } 1493 1494 bond_3ad_bind_slave(new_slave); 1495 break; 1496 case BOND_MODE_TLB: 1497 case BOND_MODE_ALB: 1498 bond_set_active_slave(new_slave); 1499 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1500 break; 1501 default: 1502 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n"); 1503 1504 /* always active in trunk mode */ 1505 bond_set_active_slave(new_slave); 1506 1507 /* In trunking mode there is little meaning to curr_active_slave 1508 * anyway (it holds no special properties of the bond device), 1509 * so we can change it without calling change_active_interface() 1510 */ 1511 if (!rcu_access_pointer(bond->curr_active_slave) && 1512 new_slave->link == BOND_LINK_UP) 1513 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1514 1515 break; 1516 } /* switch(bond_mode) */ 1517 1518 #ifdef CONFIG_NET_POLL_CONTROLLER 1519 slave_dev->npinfo = bond->dev->npinfo; 1520 if (slave_dev->npinfo) { 1521 if (slave_enable_netpoll(new_slave)) { 1522 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1523 res = -EBUSY; 1524 goto err_detach; 1525 } 1526 } 1527 #endif 1528 1529 if (!(bond_dev->features & NETIF_F_LRO)) 1530 dev_disable_lro(slave_dev); 1531 1532 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1533 new_slave); 1534 if (res) { 1535 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res); 1536 goto err_detach; 1537 } 1538 1539 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave); 1540 if (res) { 1541 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1542 goto err_unregister; 1543 } 1544 1545 res = bond_sysfs_slave_add(new_slave); 1546 if (res) { 1547 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1548 goto err_upper_unlink; 1549 } 1550 1551 bond->slave_cnt++; 1552 bond_compute_features(bond); 1553 bond_set_carrier(bond); 1554 1555 if (bond_uses_primary(bond)) { 1556 block_netpoll_tx(); 1557 bond_select_active_slave(bond); 1558 unblock_netpoll_tx(); 1559 } 1560 1561 if (bond_mode_uses_xmit_hash(bond)) 1562 bond_update_slave_arr(bond, NULL); 1563 1564 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n", 1565 slave_dev->name, 1566 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1567 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1568 1569 /* enslave is successful */ 1570 return 0; 1571 1572 /* Undo stages on error */ 1573 err_upper_unlink: 1574 bond_upper_dev_unlink(bond_dev, slave_dev); 1575 1576 err_unregister: 1577 netdev_rx_handler_unregister(slave_dev); 1578 1579 err_detach: 1580 if (!bond_uses_primary(bond)) 1581 bond_hw_addr_flush(bond_dev, slave_dev); 1582 1583 vlan_vids_del_by_dev(slave_dev, bond_dev); 1584 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1585 RCU_INIT_POINTER(bond->primary_slave, NULL); 1586 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1587 block_netpoll_tx(); 1588 bond_change_active_slave(bond, NULL); 1589 bond_select_active_slave(bond); 1590 unblock_netpoll_tx(); 1591 } 1592 /* either primary_slave or curr_active_slave might've changed */ 1593 synchronize_rcu(); 1594 slave_disable_netpoll(new_slave); 1595 1596 err_close: 1597 slave_dev->priv_flags &= ~IFF_BONDING; 1598 dev_close(slave_dev); 1599 1600 err_restore_mac: 1601 if (!bond->params.fail_over_mac || 1602 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1603 /* XXX TODO - fom follow mode needs to change master's 1604 * MAC if this slave's MAC is in use by the bond, or at 1605 * least print a warning. 1606 */ 1607 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr); 1608 addr.sa_family = slave_dev->type; 1609 dev_set_mac_address(slave_dev, &addr); 1610 } 1611 1612 err_restore_mtu: 1613 dev_set_mtu(slave_dev, new_slave->original_mtu); 1614 1615 err_free: 1616 bond_free_slave(new_slave); 1617 1618 err_undo_flags: 1619 /* Enslave of first slave has failed and we need to fix master's mac */ 1620 if (!bond_has_slaves(bond) && 1621 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr)) 1622 eth_hw_addr_random(bond_dev); 1623 1624 return res; 1625 } 1626 1627 /* Try to release the slave device <slave> from the bond device <master> 1628 * It is legal to access curr_active_slave without a lock because all the function 1629 * is RTNL-locked. If "all" is true it means that the function is being called 1630 * while destroying a bond interface and all slaves are being released. 1631 * 1632 * The rules for slave state should be: 1633 * for Active/Backup: 1634 * Active stays on all backups go down 1635 * for Bonded connections: 1636 * The first up interface should be left on and all others downed. 1637 */ 1638 static int __bond_release_one(struct net_device *bond_dev, 1639 struct net_device *slave_dev, 1640 bool all) 1641 { 1642 struct bonding *bond = netdev_priv(bond_dev); 1643 struct slave *slave, *oldcurrent; 1644 struct sockaddr addr; 1645 int old_flags = bond_dev->flags; 1646 netdev_features_t old_features = bond_dev->features; 1647 1648 /* slave is not a slave or master is not master of this slave */ 1649 if (!(slave_dev->flags & IFF_SLAVE) || 1650 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1651 netdev_err(bond_dev, "cannot release %s\n", 1652 slave_dev->name); 1653 return -EINVAL; 1654 } 1655 1656 block_netpoll_tx(); 1657 1658 slave = bond_get_slave_by_dev(bond, slave_dev); 1659 if (!slave) { 1660 /* not a slave of this bond */ 1661 netdev_info(bond_dev, "%s not enslaved\n", 1662 slave_dev->name); 1663 unblock_netpoll_tx(); 1664 return -EINVAL; 1665 } 1666 1667 bond_sysfs_slave_del(slave); 1668 1669 /* recompute stats just before removing the slave */ 1670 bond_get_stats(bond->dev, &bond->bond_stats); 1671 1672 bond_upper_dev_unlink(bond_dev, slave_dev); 1673 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1674 * for this slave anymore. 1675 */ 1676 netdev_rx_handler_unregister(slave_dev); 1677 1678 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1679 bond_3ad_unbind_slave(slave); 1680 1681 if (bond_mode_uses_xmit_hash(bond)) 1682 bond_update_slave_arr(bond, slave); 1683 1684 netdev_info(bond_dev, "Releasing %s interface %s\n", 1685 bond_is_active_slave(slave) ? "active" : "backup", 1686 slave_dev->name); 1687 1688 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1689 1690 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1691 1692 if (!all && (!bond->params.fail_over_mac || 1693 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1694 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1695 bond_has_slaves(bond)) 1696 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n", 1697 slave_dev->name, slave->perm_hwaddr, 1698 bond_dev->name, slave_dev->name); 1699 } 1700 1701 if (rtnl_dereference(bond->primary_slave) == slave) 1702 RCU_INIT_POINTER(bond->primary_slave, NULL); 1703 1704 if (oldcurrent == slave) 1705 bond_change_active_slave(bond, NULL); 1706 1707 if (bond_is_lb(bond)) { 1708 /* Must be called only after the slave has been 1709 * detached from the list and the curr_active_slave 1710 * has been cleared (if our_slave == old_current), 1711 * but before a new active slave is selected. 1712 */ 1713 bond_alb_deinit_slave(bond, slave); 1714 } 1715 1716 if (all) { 1717 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1718 } else if (oldcurrent == slave) { 1719 /* Note that we hold RTNL over this sequence, so there 1720 * is no concern that another slave add/remove event 1721 * will interfere. 1722 */ 1723 bond_select_active_slave(bond); 1724 } 1725 1726 if (!bond_has_slaves(bond)) { 1727 bond_set_carrier(bond); 1728 eth_hw_addr_random(bond_dev); 1729 } 1730 1731 unblock_netpoll_tx(); 1732 synchronize_rcu(); 1733 bond->slave_cnt--; 1734 1735 if (!bond_has_slaves(bond)) { 1736 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 1737 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 1738 } 1739 1740 bond_compute_features(bond); 1741 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 1742 (old_features & NETIF_F_VLAN_CHALLENGED)) 1743 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n", 1744 slave_dev->name, bond_dev->name); 1745 1746 vlan_vids_del_by_dev(slave_dev, bond_dev); 1747 1748 /* If the mode uses primary, then this case was handled above by 1749 * bond_change_active_slave(..., NULL) 1750 */ 1751 if (!bond_uses_primary(bond)) { 1752 /* unset promiscuity level from slave 1753 * NOTE: The NETDEV_CHANGEADDR call above may change the value 1754 * of the IFF_PROMISC flag in the bond_dev, but we need the 1755 * value of that flag before that change, as that was the value 1756 * when this slave was attached, so we cache at the start of the 1757 * function and use it here. Same goes for ALLMULTI below 1758 */ 1759 if (old_flags & IFF_PROMISC) 1760 dev_set_promiscuity(slave_dev, -1); 1761 1762 /* unset allmulti level from slave */ 1763 if (old_flags & IFF_ALLMULTI) 1764 dev_set_allmulti(slave_dev, -1); 1765 1766 bond_hw_addr_flush(bond_dev, slave_dev); 1767 } 1768 1769 slave_disable_netpoll(slave); 1770 1771 /* close slave before restoring its mac address */ 1772 dev_close(slave_dev); 1773 1774 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 1775 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1776 /* restore original ("permanent") mac address */ 1777 ether_addr_copy(addr.sa_data, slave->perm_hwaddr); 1778 addr.sa_family = slave_dev->type; 1779 dev_set_mac_address(slave_dev, &addr); 1780 } 1781 1782 dev_set_mtu(slave_dev, slave->original_mtu); 1783 1784 slave_dev->priv_flags &= ~IFF_BONDING; 1785 1786 bond_free_slave(slave); 1787 1788 return 0; 1789 } 1790 1791 /* A wrapper used because of ndo_del_link */ 1792 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 1793 { 1794 return __bond_release_one(bond_dev, slave_dev, false); 1795 } 1796 1797 /* First release a slave and then destroy the bond if no more slaves are left. 1798 * Must be under rtnl_lock when this function is called. 1799 */ 1800 static int bond_release_and_destroy(struct net_device *bond_dev, 1801 struct net_device *slave_dev) 1802 { 1803 struct bonding *bond = netdev_priv(bond_dev); 1804 int ret; 1805 1806 ret = bond_release(bond_dev, slave_dev); 1807 if (ret == 0 && !bond_has_slaves(bond)) { 1808 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 1809 netdev_info(bond_dev, "Destroying bond %s\n", 1810 bond_dev->name); 1811 unregister_netdevice(bond_dev); 1812 } 1813 return ret; 1814 } 1815 1816 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info) 1817 { 1818 struct bonding *bond = netdev_priv(bond_dev); 1819 1820 info->bond_mode = BOND_MODE(bond); 1821 info->miimon = bond->params.miimon; 1822 info->num_slaves = bond->slave_cnt; 1823 1824 return 0; 1825 } 1826 1827 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 1828 { 1829 struct bonding *bond = netdev_priv(bond_dev); 1830 struct list_head *iter; 1831 int i = 0, res = -ENODEV; 1832 struct slave *slave; 1833 1834 bond_for_each_slave(bond, slave, iter) { 1835 if (i++ == (int)info->slave_id) { 1836 res = 0; 1837 strcpy(info->slave_name, slave->dev->name); 1838 info->link = slave->link; 1839 info->state = bond_slave_state(slave); 1840 info->link_failure_count = slave->link_failure_count; 1841 break; 1842 } 1843 } 1844 1845 return res; 1846 } 1847 1848 /*-------------------------------- Monitoring -------------------------------*/ 1849 1850 /* called with rcu_read_lock() */ 1851 static int bond_miimon_inspect(struct bonding *bond) 1852 { 1853 int link_state, commit = 0; 1854 struct list_head *iter; 1855 struct slave *slave; 1856 bool ignore_updelay; 1857 1858 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 1859 1860 bond_for_each_slave_rcu(bond, slave, iter) { 1861 slave->new_link = BOND_LINK_NOCHANGE; 1862 1863 link_state = bond_check_dev_link(bond, slave->dev, 0); 1864 1865 switch (slave->link) { 1866 case BOND_LINK_UP: 1867 if (link_state) 1868 continue; 1869 1870 slave->link = BOND_LINK_FAIL; 1871 slave->delay = bond->params.downdelay; 1872 if (slave->delay) { 1873 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n", 1874 (BOND_MODE(bond) == 1875 BOND_MODE_ACTIVEBACKUP) ? 1876 (bond_is_active_slave(slave) ? 1877 "active " : "backup ") : "", 1878 slave->dev->name, 1879 bond->params.downdelay * bond->params.miimon); 1880 } 1881 /*FALLTHRU*/ 1882 case BOND_LINK_FAIL: 1883 if (link_state) { 1884 /* recovered before downdelay expired */ 1885 slave->link = BOND_LINK_UP; 1886 slave->last_link_up = jiffies; 1887 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n", 1888 (bond->params.downdelay - slave->delay) * 1889 bond->params.miimon, 1890 slave->dev->name); 1891 continue; 1892 } 1893 1894 if (slave->delay <= 0) { 1895 slave->new_link = BOND_LINK_DOWN; 1896 commit++; 1897 continue; 1898 } 1899 1900 slave->delay--; 1901 break; 1902 1903 case BOND_LINK_DOWN: 1904 if (!link_state) 1905 continue; 1906 1907 slave->link = BOND_LINK_BACK; 1908 slave->delay = bond->params.updelay; 1909 1910 if (slave->delay) { 1911 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n", 1912 slave->dev->name, 1913 ignore_updelay ? 0 : 1914 bond->params.updelay * 1915 bond->params.miimon); 1916 } 1917 /*FALLTHRU*/ 1918 case BOND_LINK_BACK: 1919 if (!link_state) { 1920 slave->link = BOND_LINK_DOWN; 1921 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n", 1922 (bond->params.updelay - slave->delay) * 1923 bond->params.miimon, 1924 slave->dev->name); 1925 1926 continue; 1927 } 1928 1929 if (ignore_updelay) 1930 slave->delay = 0; 1931 1932 if (slave->delay <= 0) { 1933 slave->new_link = BOND_LINK_UP; 1934 commit++; 1935 ignore_updelay = false; 1936 continue; 1937 } 1938 1939 slave->delay--; 1940 break; 1941 } 1942 } 1943 1944 return commit; 1945 } 1946 1947 static void bond_miimon_commit(struct bonding *bond) 1948 { 1949 struct list_head *iter; 1950 struct slave *slave, *primary; 1951 1952 bond_for_each_slave(bond, slave, iter) { 1953 switch (slave->new_link) { 1954 case BOND_LINK_NOCHANGE: 1955 continue; 1956 1957 case BOND_LINK_UP: 1958 slave->link = BOND_LINK_UP; 1959 slave->last_link_up = jiffies; 1960 1961 primary = rtnl_dereference(bond->primary_slave); 1962 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1963 /* prevent it from being the active one */ 1964 bond_set_backup_slave(slave); 1965 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1966 /* make it immediately active */ 1967 bond_set_active_slave(slave); 1968 } else if (slave != primary) { 1969 /* prevent it from being the active one */ 1970 bond_set_backup_slave(slave); 1971 } 1972 1973 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n", 1974 slave->dev->name, 1975 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 1976 slave->duplex ? "full" : "half"); 1977 1978 /* notify ad that the link status has changed */ 1979 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1980 bond_3ad_handle_link_change(slave, BOND_LINK_UP); 1981 1982 if (bond_is_lb(bond)) 1983 bond_alb_handle_link_change(bond, slave, 1984 BOND_LINK_UP); 1985 1986 if (BOND_MODE(bond) == BOND_MODE_XOR) 1987 bond_update_slave_arr(bond, NULL); 1988 1989 if (!bond->curr_active_slave || slave == primary) 1990 goto do_failover; 1991 1992 continue; 1993 1994 case BOND_LINK_DOWN: 1995 if (slave->link_failure_count < UINT_MAX) 1996 slave->link_failure_count++; 1997 1998 slave->link = BOND_LINK_DOWN; 1999 2000 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2001 BOND_MODE(bond) == BOND_MODE_8023AD) 2002 bond_set_slave_inactive_flags(slave, 2003 BOND_SLAVE_NOTIFY_NOW); 2004 2005 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2006 slave->dev->name); 2007 2008 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2009 bond_3ad_handle_link_change(slave, 2010 BOND_LINK_DOWN); 2011 2012 if (bond_is_lb(bond)) 2013 bond_alb_handle_link_change(bond, slave, 2014 BOND_LINK_DOWN); 2015 2016 if (BOND_MODE(bond) == BOND_MODE_XOR) 2017 bond_update_slave_arr(bond, NULL); 2018 2019 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2020 goto do_failover; 2021 2022 continue; 2023 2024 default: 2025 netdev_err(bond->dev, "invalid new link %d on slave %s\n", 2026 slave->new_link, slave->dev->name); 2027 slave->new_link = BOND_LINK_NOCHANGE; 2028 2029 continue; 2030 } 2031 2032 do_failover: 2033 block_netpoll_tx(); 2034 bond_select_active_slave(bond); 2035 unblock_netpoll_tx(); 2036 } 2037 2038 bond_set_carrier(bond); 2039 } 2040 2041 /* bond_mii_monitor 2042 * 2043 * Really a wrapper that splits the mii monitor into two phases: an 2044 * inspection, then (if inspection indicates something needs to be done) 2045 * an acquisition of appropriate locks followed by a commit phase to 2046 * implement whatever link state changes are indicated. 2047 */ 2048 static void bond_mii_monitor(struct work_struct *work) 2049 { 2050 struct bonding *bond = container_of(work, struct bonding, 2051 mii_work.work); 2052 bool should_notify_peers = false; 2053 unsigned long delay; 2054 2055 delay = msecs_to_jiffies(bond->params.miimon); 2056 2057 if (!bond_has_slaves(bond)) 2058 goto re_arm; 2059 2060 rcu_read_lock(); 2061 2062 should_notify_peers = bond_should_notify_peers(bond); 2063 2064 if (bond_miimon_inspect(bond)) { 2065 rcu_read_unlock(); 2066 2067 /* Race avoidance with bond_close cancel of workqueue */ 2068 if (!rtnl_trylock()) { 2069 delay = 1; 2070 should_notify_peers = false; 2071 goto re_arm; 2072 } 2073 2074 bond_miimon_commit(bond); 2075 2076 rtnl_unlock(); /* might sleep, hold no other locks */ 2077 } else 2078 rcu_read_unlock(); 2079 2080 re_arm: 2081 if (bond->params.miimon) 2082 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2083 2084 if (should_notify_peers) { 2085 if (!rtnl_trylock()) 2086 return; 2087 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2088 rtnl_unlock(); 2089 } 2090 } 2091 2092 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2093 { 2094 struct net_device *upper; 2095 struct list_head *iter; 2096 bool ret = false; 2097 2098 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2099 return true; 2100 2101 rcu_read_lock(); 2102 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) { 2103 if (ip == bond_confirm_addr(upper, 0, ip)) { 2104 ret = true; 2105 break; 2106 } 2107 } 2108 rcu_read_unlock(); 2109 2110 return ret; 2111 } 2112 2113 /* We go to the (large) trouble of VLAN tagging ARP frames because 2114 * switches in VLAN mode (especially if ports are configured as 2115 * "native" to a VLAN) might not pass non-tagged frames. 2116 */ 2117 static void bond_arp_send(struct net_device *slave_dev, int arp_op, 2118 __be32 dest_ip, __be32 src_ip, 2119 struct bond_vlan_tag *tags) 2120 { 2121 struct sk_buff *skb; 2122 struct bond_vlan_tag *outer_tag = tags; 2123 2124 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n", 2125 arp_op, slave_dev->name, &dest_ip, &src_ip); 2126 2127 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2128 NULL, slave_dev->dev_addr, NULL); 2129 2130 if (!skb) { 2131 net_err_ratelimited("ARP packet allocation failed\n"); 2132 return; 2133 } 2134 2135 if (!tags || tags->vlan_proto == VLAN_N_VID) 2136 goto xmit; 2137 2138 tags++; 2139 2140 /* Go through all the tags backwards and add them to the packet */ 2141 while (tags->vlan_proto != VLAN_N_VID) { 2142 if (!tags->vlan_id) { 2143 tags++; 2144 continue; 2145 } 2146 2147 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n", 2148 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2149 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2150 tags->vlan_id); 2151 if (!skb) { 2152 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2153 return; 2154 } 2155 2156 tags++; 2157 } 2158 /* Set the outer tag */ 2159 if (outer_tag->vlan_id) { 2160 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n", 2161 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2162 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2163 outer_tag->vlan_id); 2164 } 2165 2166 xmit: 2167 arp_xmit(skb); 2168 } 2169 2170 /* Validate the device path between the @start_dev and the @end_dev. 2171 * The path is valid if the @end_dev is reachable through device 2172 * stacking. 2173 * When the path is validated, collect any vlan information in the 2174 * path. 2175 */ 2176 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2177 struct net_device *end_dev, 2178 int level) 2179 { 2180 struct bond_vlan_tag *tags; 2181 struct net_device *upper; 2182 struct list_head *iter; 2183 2184 if (start_dev == end_dev) { 2185 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC); 2186 if (!tags) 2187 return ERR_PTR(-ENOMEM); 2188 tags[level].vlan_proto = VLAN_N_VID; 2189 return tags; 2190 } 2191 2192 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2193 tags = bond_verify_device_path(upper, end_dev, level + 1); 2194 if (IS_ERR_OR_NULL(tags)) { 2195 if (IS_ERR(tags)) 2196 return tags; 2197 continue; 2198 } 2199 if (is_vlan_dev(upper)) { 2200 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2201 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2202 } 2203 2204 return tags; 2205 } 2206 2207 return NULL; 2208 } 2209 2210 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2211 { 2212 struct rtable *rt; 2213 struct bond_vlan_tag *tags; 2214 __be32 *targets = bond->params.arp_targets, addr; 2215 int i; 2216 2217 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2218 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]); 2219 tags = NULL; 2220 2221 /* Find out through which dev should the packet go */ 2222 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2223 RTO_ONLINK, 0); 2224 if (IS_ERR(rt)) { 2225 /* there's no route to target - try to send arp 2226 * probe to generate any traffic (arp_validate=0) 2227 */ 2228 if (bond->params.arp_validate) 2229 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2230 bond->dev->name, 2231 &targets[i]); 2232 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2233 0, tags); 2234 continue; 2235 } 2236 2237 /* bond device itself */ 2238 if (rt->dst.dev == bond->dev) 2239 goto found; 2240 2241 rcu_read_lock(); 2242 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2243 rcu_read_unlock(); 2244 2245 if (!IS_ERR_OR_NULL(tags)) 2246 goto found; 2247 2248 /* Not our device - skip */ 2249 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2250 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2251 2252 ip_rt_put(rt); 2253 continue; 2254 2255 found: 2256 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2257 ip_rt_put(rt); 2258 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2259 addr, tags); 2260 kfree(tags); 2261 } 2262 } 2263 2264 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2265 { 2266 int i; 2267 2268 if (!sip || !bond_has_this_ip(bond, tip)) { 2269 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n", 2270 &sip, &tip); 2271 return; 2272 } 2273 2274 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2275 if (i == -1) { 2276 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n", 2277 &sip); 2278 return; 2279 } 2280 slave->last_rx = jiffies; 2281 slave->target_last_arp_rx[i] = jiffies; 2282 } 2283 2284 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2285 struct slave *slave) 2286 { 2287 struct arphdr *arp = (struct arphdr *)skb->data; 2288 struct slave *curr_active_slave; 2289 unsigned char *arp_ptr; 2290 __be32 sip, tip; 2291 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2292 2293 if (!slave_do_arp_validate(bond, slave)) { 2294 if ((slave_do_arp_validate_only(bond) && is_arp) || 2295 !slave_do_arp_validate_only(bond)) 2296 slave->last_rx = jiffies; 2297 return RX_HANDLER_ANOTHER; 2298 } else if (!is_arp) { 2299 return RX_HANDLER_ANOTHER; 2300 } 2301 2302 alen = arp_hdr_len(bond->dev); 2303 2304 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n", 2305 skb->dev->name); 2306 2307 if (alen > skb_headlen(skb)) { 2308 arp = kmalloc(alen, GFP_ATOMIC); 2309 if (!arp) 2310 goto out_unlock; 2311 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2312 goto out_unlock; 2313 } 2314 2315 if (arp->ar_hln != bond->dev->addr_len || 2316 skb->pkt_type == PACKET_OTHERHOST || 2317 skb->pkt_type == PACKET_LOOPBACK || 2318 arp->ar_hrd != htons(ARPHRD_ETHER) || 2319 arp->ar_pro != htons(ETH_P_IP) || 2320 arp->ar_pln != 4) 2321 goto out_unlock; 2322 2323 arp_ptr = (unsigned char *)(arp + 1); 2324 arp_ptr += bond->dev->addr_len; 2325 memcpy(&sip, arp_ptr, 4); 2326 arp_ptr += 4 + bond->dev->addr_len; 2327 memcpy(&tip, arp_ptr, 4); 2328 2329 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2330 slave->dev->name, bond_slave_state(slave), 2331 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2332 &sip, &tip); 2333 2334 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2335 2336 /* Backup slaves won't see the ARP reply, but do come through 2337 * here for each ARP probe (so we swap the sip/tip to validate 2338 * the probe). In a "redundant switch, common router" type of 2339 * configuration, the ARP probe will (hopefully) travel from 2340 * the active, through one switch, the router, then the other 2341 * switch before reaching the backup. 2342 * 2343 * We 'trust' the arp requests if there is an active slave and 2344 * it received valid arp reply(s) after it became active. This 2345 * is done to avoid endless looping when we can't reach the 2346 * arp_ip_target and fool ourselves with our own arp requests. 2347 */ 2348 2349 if (bond_is_active_slave(slave)) 2350 bond_validate_arp(bond, slave, sip, tip); 2351 else if (curr_active_slave && 2352 time_after(slave_last_rx(bond, curr_active_slave), 2353 curr_active_slave->last_link_up)) 2354 bond_validate_arp(bond, slave, tip, sip); 2355 2356 out_unlock: 2357 if (arp != (struct arphdr *)skb->data) 2358 kfree(arp); 2359 return RX_HANDLER_ANOTHER; 2360 } 2361 2362 /* function to verify if we're in the arp_interval timeslice, returns true if 2363 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2364 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2365 */ 2366 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2367 int mod) 2368 { 2369 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2370 2371 return time_in_range(jiffies, 2372 last_act - delta_in_ticks, 2373 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2374 } 2375 2376 /* This function is called regularly to monitor each slave's link 2377 * ensuring that traffic is being sent and received when arp monitoring 2378 * is used in load-balancing mode. if the adapter has been dormant, then an 2379 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2380 * arp monitoring in active backup mode. 2381 */ 2382 static void bond_loadbalance_arp_mon(struct work_struct *work) 2383 { 2384 struct bonding *bond = container_of(work, struct bonding, 2385 arp_work.work); 2386 struct slave *slave, *oldcurrent; 2387 struct list_head *iter; 2388 int do_failover = 0, slave_state_changed = 0; 2389 2390 if (!bond_has_slaves(bond)) 2391 goto re_arm; 2392 2393 rcu_read_lock(); 2394 2395 oldcurrent = rcu_dereference(bond->curr_active_slave); 2396 /* see if any of the previous devices are up now (i.e. they have 2397 * xmt and rcv traffic). the curr_active_slave does not come into 2398 * the picture unless it is null. also, slave->last_link_up is not 2399 * needed here because we send an arp on each slave and give a slave 2400 * as long as it needs to get the tx/rx within the delta. 2401 * TODO: what about up/down delay in arp mode? it wasn't here before 2402 * so it can wait 2403 */ 2404 bond_for_each_slave_rcu(bond, slave, iter) { 2405 unsigned long trans_start = dev_trans_start(slave->dev); 2406 2407 if (slave->link != BOND_LINK_UP) { 2408 if (bond_time_in_interval(bond, trans_start, 1) && 2409 bond_time_in_interval(bond, slave->last_rx, 1)) { 2410 2411 slave->link = BOND_LINK_UP; 2412 slave_state_changed = 1; 2413 2414 /* primary_slave has no meaning in round-robin 2415 * mode. the window of a slave being up and 2416 * curr_active_slave being null after enslaving 2417 * is closed. 2418 */ 2419 if (!oldcurrent) { 2420 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2421 slave->dev->name); 2422 do_failover = 1; 2423 } else { 2424 netdev_info(bond->dev, "interface %s is now up\n", 2425 slave->dev->name); 2426 } 2427 } 2428 } else { 2429 /* slave->link == BOND_LINK_UP */ 2430 2431 /* not all switches will respond to an arp request 2432 * when the source ip is 0, so don't take the link down 2433 * if we don't know our ip yet 2434 */ 2435 if (!bond_time_in_interval(bond, trans_start, 2) || 2436 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2437 2438 slave->link = BOND_LINK_DOWN; 2439 slave_state_changed = 1; 2440 2441 if (slave->link_failure_count < UINT_MAX) 2442 slave->link_failure_count++; 2443 2444 netdev_info(bond->dev, "interface %s is now down\n", 2445 slave->dev->name); 2446 2447 if (slave == oldcurrent) 2448 do_failover = 1; 2449 } 2450 } 2451 2452 /* note: if switch is in round-robin mode, all links 2453 * must tx arp to ensure all links rx an arp - otherwise 2454 * links may oscillate or not come up at all; if switch is 2455 * in something like xor mode, there is nothing we can 2456 * do - all replies will be rx'ed on same link causing slaves 2457 * to be unstable during low/no traffic periods 2458 */ 2459 if (bond_slave_is_up(slave)) 2460 bond_arp_send_all(bond, slave); 2461 } 2462 2463 rcu_read_unlock(); 2464 2465 if (do_failover || slave_state_changed) { 2466 if (!rtnl_trylock()) 2467 goto re_arm; 2468 2469 if (slave_state_changed) { 2470 bond_slave_state_change(bond); 2471 if (BOND_MODE(bond) == BOND_MODE_XOR) 2472 bond_update_slave_arr(bond, NULL); 2473 } 2474 if (do_failover) { 2475 block_netpoll_tx(); 2476 bond_select_active_slave(bond); 2477 unblock_netpoll_tx(); 2478 } 2479 rtnl_unlock(); 2480 } 2481 2482 re_arm: 2483 if (bond->params.arp_interval) 2484 queue_delayed_work(bond->wq, &bond->arp_work, 2485 msecs_to_jiffies(bond->params.arp_interval)); 2486 } 2487 2488 /* Called to inspect slaves for active-backup mode ARP monitor link state 2489 * changes. Sets new_link in slaves to specify what action should take 2490 * place for the slave. Returns 0 if no changes are found, >0 if changes 2491 * to link states must be committed. 2492 * 2493 * Called with rcu_read_lock held. 2494 */ 2495 static int bond_ab_arp_inspect(struct bonding *bond) 2496 { 2497 unsigned long trans_start, last_rx; 2498 struct list_head *iter; 2499 struct slave *slave; 2500 int commit = 0; 2501 2502 bond_for_each_slave_rcu(bond, slave, iter) { 2503 slave->new_link = BOND_LINK_NOCHANGE; 2504 last_rx = slave_last_rx(bond, slave); 2505 2506 if (slave->link != BOND_LINK_UP) { 2507 if (bond_time_in_interval(bond, last_rx, 1)) { 2508 slave->new_link = BOND_LINK_UP; 2509 commit++; 2510 } 2511 continue; 2512 } 2513 2514 /* Give slaves 2*delta after being enslaved or made 2515 * active. This avoids bouncing, as the last receive 2516 * times need a full ARP monitor cycle to be updated. 2517 */ 2518 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2519 continue; 2520 2521 /* Backup slave is down if: 2522 * - No current_arp_slave AND 2523 * - more than 3*delta since last receive AND 2524 * - the bond has an IP address 2525 * 2526 * Note: a non-null current_arp_slave indicates 2527 * the curr_active_slave went down and we are 2528 * searching for a new one; under this condition 2529 * we only take the curr_active_slave down - this 2530 * gives each slave a chance to tx/rx traffic 2531 * before being taken out 2532 */ 2533 if (!bond_is_active_slave(slave) && 2534 !rcu_access_pointer(bond->current_arp_slave) && 2535 !bond_time_in_interval(bond, last_rx, 3)) { 2536 slave->new_link = BOND_LINK_DOWN; 2537 commit++; 2538 } 2539 2540 /* Active slave is down if: 2541 * - more than 2*delta since transmitting OR 2542 * - (more than 2*delta since receive AND 2543 * the bond has an IP address) 2544 */ 2545 trans_start = dev_trans_start(slave->dev); 2546 if (bond_is_active_slave(slave) && 2547 (!bond_time_in_interval(bond, trans_start, 2) || 2548 !bond_time_in_interval(bond, last_rx, 2))) { 2549 slave->new_link = BOND_LINK_DOWN; 2550 commit++; 2551 } 2552 } 2553 2554 return commit; 2555 } 2556 2557 /* Called to commit link state changes noted by inspection step of 2558 * active-backup mode ARP monitor. 2559 * 2560 * Called with RTNL hold. 2561 */ 2562 static void bond_ab_arp_commit(struct bonding *bond) 2563 { 2564 unsigned long trans_start; 2565 struct list_head *iter; 2566 struct slave *slave; 2567 2568 bond_for_each_slave(bond, slave, iter) { 2569 switch (slave->new_link) { 2570 case BOND_LINK_NOCHANGE: 2571 continue; 2572 2573 case BOND_LINK_UP: 2574 trans_start = dev_trans_start(slave->dev); 2575 if (rtnl_dereference(bond->curr_active_slave) != slave || 2576 (!rtnl_dereference(bond->curr_active_slave) && 2577 bond_time_in_interval(bond, trans_start, 1))) { 2578 struct slave *current_arp_slave; 2579 2580 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2581 slave->link = BOND_LINK_UP; 2582 if (current_arp_slave) { 2583 bond_set_slave_inactive_flags( 2584 current_arp_slave, 2585 BOND_SLAVE_NOTIFY_NOW); 2586 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2587 } 2588 2589 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2590 slave->dev->name); 2591 2592 if (!rtnl_dereference(bond->curr_active_slave) || 2593 slave == rtnl_dereference(bond->primary_slave)) 2594 goto do_failover; 2595 2596 } 2597 2598 continue; 2599 2600 case BOND_LINK_DOWN: 2601 if (slave->link_failure_count < UINT_MAX) 2602 slave->link_failure_count++; 2603 2604 slave->link = BOND_LINK_DOWN; 2605 bond_set_slave_inactive_flags(slave, 2606 BOND_SLAVE_NOTIFY_NOW); 2607 2608 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2609 slave->dev->name); 2610 2611 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2612 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2613 goto do_failover; 2614 } 2615 2616 continue; 2617 2618 default: 2619 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n", 2620 slave->new_link, slave->dev->name); 2621 continue; 2622 } 2623 2624 do_failover: 2625 block_netpoll_tx(); 2626 bond_select_active_slave(bond); 2627 unblock_netpoll_tx(); 2628 } 2629 2630 bond_set_carrier(bond); 2631 } 2632 2633 /* Send ARP probes for active-backup mode ARP monitor. 2634 * 2635 * Called with rcu_read_lock held. 2636 */ 2637 static bool bond_ab_arp_probe(struct bonding *bond) 2638 { 2639 struct slave *slave, *before = NULL, *new_slave = NULL, 2640 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2641 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2642 struct list_head *iter; 2643 bool found = false; 2644 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2645 2646 if (curr_arp_slave && curr_active_slave) 2647 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2648 curr_arp_slave->dev->name, 2649 curr_active_slave->dev->name); 2650 2651 if (curr_active_slave) { 2652 bond_arp_send_all(bond, curr_active_slave); 2653 return should_notify_rtnl; 2654 } 2655 2656 /* if we don't have a curr_active_slave, search for the next available 2657 * backup slave from the current_arp_slave and make it the candidate 2658 * for becoming the curr_active_slave 2659 */ 2660 2661 if (!curr_arp_slave) { 2662 curr_arp_slave = bond_first_slave_rcu(bond); 2663 if (!curr_arp_slave) 2664 return should_notify_rtnl; 2665 } 2666 2667 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2668 2669 bond_for_each_slave_rcu(bond, slave, iter) { 2670 if (!found && !before && bond_slave_is_up(slave)) 2671 before = slave; 2672 2673 if (found && !new_slave && bond_slave_is_up(slave)) 2674 new_slave = slave; 2675 /* if the link state is up at this point, we 2676 * mark it down - this can happen if we have 2677 * simultaneous link failures and 2678 * reselect_active_interface doesn't make this 2679 * one the current slave so it is still marked 2680 * up when it is actually down 2681 */ 2682 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 2683 slave->link = BOND_LINK_DOWN; 2684 if (slave->link_failure_count < UINT_MAX) 2685 slave->link_failure_count++; 2686 2687 bond_set_slave_inactive_flags(slave, 2688 BOND_SLAVE_NOTIFY_LATER); 2689 2690 netdev_info(bond->dev, "backup interface %s is now down\n", 2691 slave->dev->name); 2692 } 2693 if (slave == curr_arp_slave) 2694 found = true; 2695 } 2696 2697 if (!new_slave && before) 2698 new_slave = before; 2699 2700 if (!new_slave) 2701 goto check_state; 2702 2703 new_slave->link = BOND_LINK_BACK; 2704 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 2705 bond_arp_send_all(bond, new_slave); 2706 new_slave->last_link_up = jiffies; 2707 rcu_assign_pointer(bond->current_arp_slave, new_slave); 2708 2709 check_state: 2710 bond_for_each_slave_rcu(bond, slave, iter) { 2711 if (slave->should_notify) { 2712 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 2713 break; 2714 } 2715 } 2716 return should_notify_rtnl; 2717 } 2718 2719 static void bond_activebackup_arp_mon(struct work_struct *work) 2720 { 2721 struct bonding *bond = container_of(work, struct bonding, 2722 arp_work.work); 2723 bool should_notify_peers = false; 2724 bool should_notify_rtnl = false; 2725 int delta_in_ticks; 2726 2727 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2728 2729 if (!bond_has_slaves(bond)) 2730 goto re_arm; 2731 2732 rcu_read_lock(); 2733 2734 should_notify_peers = bond_should_notify_peers(bond); 2735 2736 if (bond_ab_arp_inspect(bond)) { 2737 rcu_read_unlock(); 2738 2739 /* Race avoidance with bond_close flush of workqueue */ 2740 if (!rtnl_trylock()) { 2741 delta_in_ticks = 1; 2742 should_notify_peers = false; 2743 goto re_arm; 2744 } 2745 2746 bond_ab_arp_commit(bond); 2747 2748 rtnl_unlock(); 2749 rcu_read_lock(); 2750 } 2751 2752 should_notify_rtnl = bond_ab_arp_probe(bond); 2753 rcu_read_unlock(); 2754 2755 re_arm: 2756 if (bond->params.arp_interval) 2757 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 2758 2759 if (should_notify_peers || should_notify_rtnl) { 2760 if (!rtnl_trylock()) 2761 return; 2762 2763 if (should_notify_peers) 2764 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 2765 bond->dev); 2766 if (should_notify_rtnl) 2767 bond_slave_state_notify(bond); 2768 2769 rtnl_unlock(); 2770 } 2771 } 2772 2773 /*-------------------------- netdev event handling --------------------------*/ 2774 2775 /* Change device name */ 2776 static int bond_event_changename(struct bonding *bond) 2777 { 2778 bond_remove_proc_entry(bond); 2779 bond_create_proc_entry(bond); 2780 2781 bond_debug_reregister(bond); 2782 2783 return NOTIFY_DONE; 2784 } 2785 2786 static int bond_master_netdev_event(unsigned long event, 2787 struct net_device *bond_dev) 2788 { 2789 struct bonding *event_bond = netdev_priv(bond_dev); 2790 2791 switch (event) { 2792 case NETDEV_CHANGENAME: 2793 return bond_event_changename(event_bond); 2794 case NETDEV_UNREGISTER: 2795 bond_remove_proc_entry(event_bond); 2796 break; 2797 case NETDEV_REGISTER: 2798 bond_create_proc_entry(event_bond); 2799 break; 2800 case NETDEV_NOTIFY_PEERS: 2801 if (event_bond->send_peer_notif) 2802 event_bond->send_peer_notif--; 2803 break; 2804 default: 2805 break; 2806 } 2807 2808 return NOTIFY_DONE; 2809 } 2810 2811 static int bond_slave_netdev_event(unsigned long event, 2812 struct net_device *slave_dev) 2813 { 2814 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 2815 struct bonding *bond; 2816 struct net_device *bond_dev; 2817 u32 old_speed; 2818 u8 old_duplex; 2819 2820 /* A netdev event can be generated while enslaving a device 2821 * before netdev_rx_handler_register is called in which case 2822 * slave will be NULL 2823 */ 2824 if (!slave) 2825 return NOTIFY_DONE; 2826 bond_dev = slave->bond->dev; 2827 bond = slave->bond; 2828 primary = rtnl_dereference(bond->primary_slave); 2829 2830 switch (event) { 2831 case NETDEV_UNREGISTER: 2832 if (bond_dev->type != ARPHRD_ETHER) 2833 bond_release_and_destroy(bond_dev, slave_dev); 2834 else 2835 bond_release(bond_dev, slave_dev); 2836 break; 2837 case NETDEV_UP: 2838 case NETDEV_CHANGE: 2839 old_speed = slave->speed; 2840 old_duplex = slave->duplex; 2841 2842 bond_update_speed_duplex(slave); 2843 2844 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2845 if (old_speed != slave->speed) 2846 bond_3ad_adapter_speed_changed(slave); 2847 if (old_duplex != slave->duplex) 2848 bond_3ad_adapter_duplex_changed(slave); 2849 } 2850 /* Refresh slave-array if applicable! 2851 * If the setup does not use miimon or arpmon (mode-specific!), 2852 * then these events will not cause the slave-array to be 2853 * refreshed. This will cause xmit to use a slave that is not 2854 * usable. Avoid such situation by refeshing the array at these 2855 * events. If these (miimon/arpmon) parameters are configured 2856 * then array gets refreshed twice and that should be fine! 2857 */ 2858 if (bond_mode_uses_xmit_hash(bond)) 2859 bond_update_slave_arr(bond, NULL); 2860 break; 2861 case NETDEV_DOWN: 2862 if (bond_mode_uses_xmit_hash(bond)) 2863 bond_update_slave_arr(bond, NULL); 2864 break; 2865 case NETDEV_CHANGEMTU: 2866 /* TODO: Should slaves be allowed to 2867 * independently alter their MTU? For 2868 * an active-backup bond, slaves need 2869 * not be the same type of device, so 2870 * MTUs may vary. For other modes, 2871 * slaves arguably should have the 2872 * same MTUs. To do this, we'd need to 2873 * take over the slave's change_mtu 2874 * function for the duration of their 2875 * servitude. 2876 */ 2877 break; 2878 case NETDEV_CHANGENAME: 2879 /* we don't care if we don't have primary set */ 2880 if (!bond_uses_primary(bond) || 2881 !bond->params.primary[0]) 2882 break; 2883 2884 if (slave == primary) { 2885 /* slave's name changed - he's no longer primary */ 2886 RCU_INIT_POINTER(bond->primary_slave, NULL); 2887 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 2888 /* we have a new primary slave */ 2889 rcu_assign_pointer(bond->primary_slave, slave); 2890 } else { /* we didn't change primary - exit */ 2891 break; 2892 } 2893 2894 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 2895 primary ? slave_dev->name : "none"); 2896 2897 block_netpoll_tx(); 2898 bond_select_active_slave(bond); 2899 unblock_netpoll_tx(); 2900 break; 2901 case NETDEV_FEAT_CHANGE: 2902 bond_compute_features(bond); 2903 break; 2904 case NETDEV_RESEND_IGMP: 2905 /* Propagate to master device */ 2906 call_netdevice_notifiers(event, slave->bond->dev); 2907 break; 2908 default: 2909 break; 2910 } 2911 2912 return NOTIFY_DONE; 2913 } 2914 2915 /* bond_netdev_event: handle netdev notifier chain events. 2916 * 2917 * This function receives events for the netdev chain. The caller (an 2918 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 2919 * locks for us to safely manipulate the slave devices (RTNL lock, 2920 * dev_probe_lock). 2921 */ 2922 static int bond_netdev_event(struct notifier_block *this, 2923 unsigned long event, void *ptr) 2924 { 2925 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 2926 2927 netdev_dbg(event_dev, "event: %lx\n", event); 2928 2929 if (!(event_dev->priv_flags & IFF_BONDING)) 2930 return NOTIFY_DONE; 2931 2932 if (event_dev->flags & IFF_MASTER) { 2933 netdev_dbg(event_dev, "IFF_MASTER\n"); 2934 return bond_master_netdev_event(event, event_dev); 2935 } 2936 2937 if (event_dev->flags & IFF_SLAVE) { 2938 netdev_dbg(event_dev, "IFF_SLAVE\n"); 2939 return bond_slave_netdev_event(event, event_dev); 2940 } 2941 2942 return NOTIFY_DONE; 2943 } 2944 2945 static struct notifier_block bond_netdev_notifier = { 2946 .notifier_call = bond_netdev_event, 2947 }; 2948 2949 /*---------------------------- Hashing Policies -----------------------------*/ 2950 2951 /* L2 hash helper */ 2952 static inline u32 bond_eth_hash(struct sk_buff *skb) 2953 { 2954 struct ethhdr *ep, hdr_tmp; 2955 2956 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 2957 if (ep) 2958 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 2959 return 0; 2960 } 2961 2962 /* Extract the appropriate headers based on bond's xmit policy */ 2963 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 2964 struct flow_keys *fk) 2965 { 2966 const struct ipv6hdr *iph6; 2967 const struct iphdr *iph; 2968 int noff, proto = -1; 2969 2970 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) 2971 return skb_flow_dissect(skb, fk); 2972 2973 fk->ports = 0; 2974 noff = skb_network_offset(skb); 2975 if (skb->protocol == htons(ETH_P_IP)) { 2976 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 2977 return false; 2978 iph = ip_hdr(skb); 2979 fk->src = iph->saddr; 2980 fk->dst = iph->daddr; 2981 noff += iph->ihl << 2; 2982 if (!ip_is_fragment(iph)) 2983 proto = iph->protocol; 2984 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2985 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6)))) 2986 return false; 2987 iph6 = ipv6_hdr(skb); 2988 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr); 2989 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr); 2990 noff += sizeof(*iph6); 2991 proto = iph6->nexthdr; 2992 } else { 2993 return false; 2994 } 2995 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) 2996 fk->ports = skb_flow_get_ports(skb, noff, proto); 2997 2998 return true; 2999 } 3000 3001 /** 3002 * bond_xmit_hash - generate a hash value based on the xmit policy 3003 * @bond: bonding device 3004 * @skb: buffer to use for headers 3005 * 3006 * This function will extract the necessary headers from the skb buffer and use 3007 * them to generate a hash based on the xmit_policy set in the bonding device 3008 */ 3009 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3010 { 3011 struct flow_keys flow; 3012 u32 hash; 3013 3014 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3015 !bond_flow_dissect(bond, skb, &flow)) 3016 return bond_eth_hash(skb); 3017 3018 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3019 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) 3020 hash = bond_eth_hash(skb); 3021 else 3022 hash = (__force u32)flow.ports; 3023 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; 3024 hash ^= (hash >> 16); 3025 hash ^= (hash >> 8); 3026 3027 return hash; 3028 } 3029 3030 /*-------------------------- Device entry points ----------------------------*/ 3031 3032 static void bond_work_init_all(struct bonding *bond) 3033 { 3034 INIT_DELAYED_WORK(&bond->mcast_work, 3035 bond_resend_igmp_join_requests_delayed); 3036 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3037 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3038 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3039 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon); 3040 else 3041 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon); 3042 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3043 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3044 } 3045 3046 static void bond_work_cancel_all(struct bonding *bond) 3047 { 3048 cancel_delayed_work_sync(&bond->mii_work); 3049 cancel_delayed_work_sync(&bond->arp_work); 3050 cancel_delayed_work_sync(&bond->alb_work); 3051 cancel_delayed_work_sync(&bond->ad_work); 3052 cancel_delayed_work_sync(&bond->mcast_work); 3053 cancel_delayed_work_sync(&bond->slave_arr_work); 3054 } 3055 3056 static int bond_open(struct net_device *bond_dev) 3057 { 3058 struct bonding *bond = netdev_priv(bond_dev); 3059 struct list_head *iter; 3060 struct slave *slave; 3061 3062 /* reset slave->backup and slave->inactive */ 3063 if (bond_has_slaves(bond)) { 3064 bond_for_each_slave(bond, slave, iter) { 3065 if (bond_uses_primary(bond) && 3066 slave != rcu_access_pointer(bond->curr_active_slave)) { 3067 bond_set_slave_inactive_flags(slave, 3068 BOND_SLAVE_NOTIFY_NOW); 3069 } else { 3070 bond_set_slave_active_flags(slave, 3071 BOND_SLAVE_NOTIFY_NOW); 3072 } 3073 } 3074 } 3075 3076 bond_work_init_all(bond); 3077 3078 if (bond_is_lb(bond)) { 3079 /* bond_alb_initialize must be called before the timer 3080 * is started. 3081 */ 3082 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3083 return -ENOMEM; 3084 if (bond->params.tlb_dynamic_lb) 3085 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3086 } 3087 3088 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3089 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3090 3091 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3092 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3093 bond->recv_probe = bond_arp_rcv; 3094 } 3095 3096 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3097 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3098 /* register to receive LACPDUs */ 3099 bond->recv_probe = bond_3ad_lacpdu_recv; 3100 bond_3ad_initiate_agg_selection(bond, 1); 3101 } 3102 3103 if (bond_mode_uses_xmit_hash(bond)) 3104 bond_update_slave_arr(bond, NULL); 3105 3106 return 0; 3107 } 3108 3109 static int bond_close(struct net_device *bond_dev) 3110 { 3111 struct bonding *bond = netdev_priv(bond_dev); 3112 3113 bond_work_cancel_all(bond); 3114 bond->send_peer_notif = 0; 3115 if (bond_is_lb(bond)) 3116 bond_alb_deinitialize(bond); 3117 bond->recv_probe = NULL; 3118 3119 return 0; 3120 } 3121 3122 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev, 3123 struct rtnl_link_stats64 *stats) 3124 { 3125 struct bonding *bond = netdev_priv(bond_dev); 3126 struct rtnl_link_stats64 temp; 3127 struct list_head *iter; 3128 struct slave *slave; 3129 3130 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3131 3132 bond_for_each_slave(bond, slave, iter) { 3133 const struct rtnl_link_stats64 *sstats = 3134 dev_get_stats(slave->dev, &temp); 3135 struct rtnl_link_stats64 *pstats = &slave->slave_stats; 3136 3137 stats->rx_packets += sstats->rx_packets - pstats->rx_packets; 3138 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes; 3139 stats->rx_errors += sstats->rx_errors - pstats->rx_errors; 3140 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped; 3141 3142 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;; 3143 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes; 3144 stats->tx_errors += sstats->tx_errors - pstats->tx_errors; 3145 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped; 3146 3147 stats->multicast += sstats->multicast - pstats->multicast; 3148 stats->collisions += sstats->collisions - pstats->collisions; 3149 3150 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors; 3151 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors; 3152 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors; 3153 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors; 3154 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors; 3155 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors; 3156 3157 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors; 3158 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors; 3159 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors; 3160 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors; 3161 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors; 3162 3163 /* save off the slave stats for the next run */ 3164 memcpy(pstats, sstats, sizeof(*sstats)); 3165 } 3166 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3167 3168 return stats; 3169 } 3170 3171 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3172 { 3173 struct bonding *bond = netdev_priv(bond_dev); 3174 struct net_device *slave_dev = NULL; 3175 struct ifbond k_binfo; 3176 struct ifbond __user *u_binfo = NULL; 3177 struct ifslave k_sinfo; 3178 struct ifslave __user *u_sinfo = NULL; 3179 struct mii_ioctl_data *mii = NULL; 3180 struct bond_opt_value newval; 3181 struct net *net; 3182 int res = 0; 3183 3184 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3185 3186 switch (cmd) { 3187 case SIOCGMIIPHY: 3188 mii = if_mii(ifr); 3189 if (!mii) 3190 return -EINVAL; 3191 3192 mii->phy_id = 0; 3193 /* Fall Through */ 3194 case SIOCGMIIREG: 3195 /* We do this again just in case we were called by SIOCGMIIREG 3196 * instead of SIOCGMIIPHY. 3197 */ 3198 mii = if_mii(ifr); 3199 if (!mii) 3200 return -EINVAL; 3201 3202 if (mii->reg_num == 1) { 3203 mii->val_out = 0; 3204 if (netif_carrier_ok(bond->dev)) 3205 mii->val_out = BMSR_LSTATUS; 3206 } 3207 3208 return 0; 3209 case BOND_INFO_QUERY_OLD: 3210 case SIOCBONDINFOQUERY: 3211 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3212 3213 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3214 return -EFAULT; 3215 3216 res = bond_info_query(bond_dev, &k_binfo); 3217 if (res == 0 && 3218 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3219 return -EFAULT; 3220 3221 return res; 3222 case BOND_SLAVE_INFO_QUERY_OLD: 3223 case SIOCBONDSLAVEINFOQUERY: 3224 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3225 3226 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3227 return -EFAULT; 3228 3229 res = bond_slave_info_query(bond_dev, &k_sinfo); 3230 if (res == 0 && 3231 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3232 return -EFAULT; 3233 3234 return res; 3235 default: 3236 break; 3237 } 3238 3239 net = dev_net(bond_dev); 3240 3241 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3242 return -EPERM; 3243 3244 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3245 3246 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev); 3247 3248 if (!slave_dev) 3249 return -ENODEV; 3250 3251 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name); 3252 switch (cmd) { 3253 case BOND_ENSLAVE_OLD: 3254 case SIOCBONDENSLAVE: 3255 res = bond_enslave(bond_dev, slave_dev); 3256 break; 3257 case BOND_RELEASE_OLD: 3258 case SIOCBONDRELEASE: 3259 res = bond_release(bond_dev, slave_dev); 3260 break; 3261 case BOND_SETHWADDR_OLD: 3262 case SIOCBONDSETHWADDR: 3263 bond_set_dev_addr(bond_dev, slave_dev); 3264 res = 0; 3265 break; 3266 case BOND_CHANGE_ACTIVE_OLD: 3267 case SIOCBONDCHANGEACTIVE: 3268 bond_opt_initstr(&newval, slave_dev->name); 3269 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval); 3270 break; 3271 default: 3272 res = -EOPNOTSUPP; 3273 } 3274 3275 return res; 3276 } 3277 3278 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3279 { 3280 struct bonding *bond = netdev_priv(bond_dev); 3281 3282 if (change & IFF_PROMISC) 3283 bond_set_promiscuity(bond, 3284 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3285 3286 if (change & IFF_ALLMULTI) 3287 bond_set_allmulti(bond, 3288 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3289 } 3290 3291 static void bond_set_rx_mode(struct net_device *bond_dev) 3292 { 3293 struct bonding *bond = netdev_priv(bond_dev); 3294 struct list_head *iter; 3295 struct slave *slave; 3296 3297 rcu_read_lock(); 3298 if (bond_uses_primary(bond)) { 3299 slave = rcu_dereference(bond->curr_active_slave); 3300 if (slave) { 3301 dev_uc_sync(slave->dev, bond_dev); 3302 dev_mc_sync(slave->dev, bond_dev); 3303 } 3304 } else { 3305 bond_for_each_slave_rcu(bond, slave, iter) { 3306 dev_uc_sync_multiple(slave->dev, bond_dev); 3307 dev_mc_sync_multiple(slave->dev, bond_dev); 3308 } 3309 } 3310 rcu_read_unlock(); 3311 } 3312 3313 static int bond_neigh_init(struct neighbour *n) 3314 { 3315 struct bonding *bond = netdev_priv(n->dev); 3316 const struct net_device_ops *slave_ops; 3317 struct neigh_parms parms; 3318 struct slave *slave; 3319 int ret; 3320 3321 slave = bond_first_slave(bond); 3322 if (!slave) 3323 return 0; 3324 slave_ops = slave->dev->netdev_ops; 3325 if (!slave_ops->ndo_neigh_setup) 3326 return 0; 3327 3328 parms.neigh_setup = NULL; 3329 parms.neigh_cleanup = NULL; 3330 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3331 if (ret) 3332 return ret; 3333 3334 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called 3335 * after the last slave has been detached. Assumes that all slaves 3336 * utilize the same neigh_cleanup (true at this writing as only user 3337 * is ipoib). 3338 */ 3339 n->parms->neigh_cleanup = parms.neigh_cleanup; 3340 3341 if (!parms.neigh_setup) 3342 return 0; 3343 3344 return parms.neigh_setup(n); 3345 } 3346 3347 /* The bonding ndo_neigh_setup is called at init time beofre any 3348 * slave exists. So we must declare proxy setup function which will 3349 * be used at run time to resolve the actual slave neigh param setup. 3350 * 3351 * It's also called by master devices (such as vlans) to setup their 3352 * underlying devices. In that case - do nothing, we're already set up from 3353 * our init. 3354 */ 3355 static int bond_neigh_setup(struct net_device *dev, 3356 struct neigh_parms *parms) 3357 { 3358 /* modify only our neigh_parms */ 3359 if (parms->dev == dev) 3360 parms->neigh_setup = bond_neigh_init; 3361 3362 return 0; 3363 } 3364 3365 /* Change the MTU of all of a master's slaves to match the master */ 3366 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3367 { 3368 struct bonding *bond = netdev_priv(bond_dev); 3369 struct slave *slave, *rollback_slave; 3370 struct list_head *iter; 3371 int res = 0; 3372 3373 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3374 3375 bond_for_each_slave(bond, slave, iter) { 3376 netdev_dbg(bond_dev, "s %p c_m %p\n", 3377 slave, slave->dev->netdev_ops->ndo_change_mtu); 3378 3379 res = dev_set_mtu(slave->dev, new_mtu); 3380 3381 if (res) { 3382 /* If we failed to set the slave's mtu to the new value 3383 * we must abort the operation even in ACTIVE_BACKUP 3384 * mode, because if we allow the backup slaves to have 3385 * different mtu values than the active slave we'll 3386 * need to change their mtu when doing a failover. That 3387 * means changing their mtu from timer context, which 3388 * is probably not a good idea. 3389 */ 3390 netdev_dbg(bond_dev, "err %d %s\n", res, 3391 slave->dev->name); 3392 goto unwind; 3393 } 3394 } 3395 3396 bond_dev->mtu = new_mtu; 3397 3398 return 0; 3399 3400 unwind: 3401 /* unwind from head to the slave that failed */ 3402 bond_for_each_slave(bond, rollback_slave, iter) { 3403 int tmp_res; 3404 3405 if (rollback_slave == slave) 3406 break; 3407 3408 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3409 if (tmp_res) { 3410 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3411 tmp_res, rollback_slave->dev->name); 3412 } 3413 } 3414 3415 return res; 3416 } 3417 3418 /* Change HW address 3419 * 3420 * Note that many devices must be down to change the HW address, and 3421 * downing the master releases all slaves. We can make bonds full of 3422 * bonding devices to test this, however. 3423 */ 3424 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3425 { 3426 struct bonding *bond = netdev_priv(bond_dev); 3427 struct slave *slave, *rollback_slave; 3428 struct sockaddr *sa = addr, tmp_sa; 3429 struct list_head *iter; 3430 int res = 0; 3431 3432 if (BOND_MODE(bond) == BOND_MODE_ALB) 3433 return bond_alb_set_mac_address(bond_dev, addr); 3434 3435 3436 netdev_dbg(bond_dev, "bond=%p\n", bond); 3437 3438 /* If fail_over_mac is enabled, do nothing and return success. 3439 * Returning an error causes ifenslave to fail. 3440 */ 3441 if (bond->params.fail_over_mac && 3442 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3443 return 0; 3444 3445 if (!is_valid_ether_addr(sa->sa_data)) 3446 return -EADDRNOTAVAIL; 3447 3448 bond_for_each_slave(bond, slave, iter) { 3449 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name); 3450 res = dev_set_mac_address(slave->dev, addr); 3451 if (res) { 3452 /* TODO: consider downing the slave 3453 * and retry ? 3454 * User should expect communications 3455 * breakage anyway until ARP finish 3456 * updating, so... 3457 */ 3458 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name); 3459 goto unwind; 3460 } 3461 } 3462 3463 /* success */ 3464 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len); 3465 return 0; 3466 3467 unwind: 3468 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len); 3469 tmp_sa.sa_family = bond_dev->type; 3470 3471 /* unwind from head to the slave that failed */ 3472 bond_for_each_slave(bond, rollback_slave, iter) { 3473 int tmp_res; 3474 3475 if (rollback_slave == slave) 3476 break; 3477 3478 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa); 3479 if (tmp_res) { 3480 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3481 tmp_res, rollback_slave->dev->name); 3482 } 3483 } 3484 3485 return res; 3486 } 3487 3488 /** 3489 * bond_xmit_slave_id - transmit skb through slave with slave_id 3490 * @bond: bonding device that is transmitting 3491 * @skb: buffer to transmit 3492 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3493 * 3494 * This function tries to transmit through slave with slave_id but in case 3495 * it fails, it tries to find the first available slave for transmission. 3496 * The skb is consumed in all cases, thus the function is void. 3497 */ 3498 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3499 { 3500 struct list_head *iter; 3501 struct slave *slave; 3502 int i = slave_id; 3503 3504 /* Here we start from the slave with slave_id */ 3505 bond_for_each_slave_rcu(bond, slave, iter) { 3506 if (--i < 0) { 3507 if (bond_slave_can_tx(slave)) { 3508 bond_dev_queue_xmit(bond, skb, slave->dev); 3509 return; 3510 } 3511 } 3512 } 3513 3514 /* Here we start from the first slave up to slave_id */ 3515 i = slave_id; 3516 bond_for_each_slave_rcu(bond, slave, iter) { 3517 if (--i < 0) 3518 break; 3519 if (bond_slave_can_tx(slave)) { 3520 bond_dev_queue_xmit(bond, skb, slave->dev); 3521 return; 3522 } 3523 } 3524 /* no slave that can tx has been found */ 3525 bond_tx_drop(bond->dev, skb); 3526 } 3527 3528 /** 3529 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3530 * @bond: bonding device to use 3531 * 3532 * Based on the value of the bonding device's packets_per_slave parameter 3533 * this function generates a slave id, which is usually used as the next 3534 * slave to transmit through. 3535 */ 3536 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3537 { 3538 u32 slave_id; 3539 struct reciprocal_value reciprocal_packets_per_slave; 3540 int packets_per_slave = bond->params.packets_per_slave; 3541 3542 switch (packets_per_slave) { 3543 case 0: 3544 slave_id = prandom_u32(); 3545 break; 3546 case 1: 3547 slave_id = bond->rr_tx_counter; 3548 break; 3549 default: 3550 reciprocal_packets_per_slave = 3551 bond->params.reciprocal_packets_per_slave; 3552 slave_id = reciprocal_divide(bond->rr_tx_counter, 3553 reciprocal_packets_per_slave); 3554 break; 3555 } 3556 bond->rr_tx_counter++; 3557 3558 return slave_id; 3559 } 3560 3561 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev) 3562 { 3563 struct bonding *bond = netdev_priv(bond_dev); 3564 struct iphdr *iph = ip_hdr(skb); 3565 struct slave *slave; 3566 u32 slave_id; 3567 3568 /* Start with the curr_active_slave that joined the bond as the 3569 * default for sending IGMP traffic. For failover purposes one 3570 * needs to maintain some consistency for the interface that will 3571 * send the join/membership reports. The curr_active_slave found 3572 * will send all of this type of traffic. 3573 */ 3574 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) { 3575 slave = rcu_dereference(bond->curr_active_slave); 3576 if (slave) 3577 bond_dev_queue_xmit(bond, skb, slave->dev); 3578 else 3579 bond_xmit_slave_id(bond, skb, 0); 3580 } else { 3581 int slave_cnt = ACCESS_ONCE(bond->slave_cnt); 3582 3583 if (likely(slave_cnt)) { 3584 slave_id = bond_rr_gen_slave_id(bond); 3585 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3586 } else { 3587 bond_tx_drop(bond_dev, skb); 3588 } 3589 } 3590 3591 return NETDEV_TX_OK; 3592 } 3593 3594 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3595 * the bond has a usable interface. 3596 */ 3597 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev) 3598 { 3599 struct bonding *bond = netdev_priv(bond_dev); 3600 struct slave *slave; 3601 3602 slave = rcu_dereference(bond->curr_active_slave); 3603 if (slave) 3604 bond_dev_queue_xmit(bond, skb, slave->dev); 3605 else 3606 bond_tx_drop(bond_dev, skb); 3607 3608 return NETDEV_TX_OK; 3609 } 3610 3611 /* Use this to update slave_array when (a) it's not appropriate to update 3612 * slave_array right away (note that update_slave_array() may sleep) 3613 * and / or (b) RTNL is not held. 3614 */ 3615 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 3616 { 3617 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 3618 } 3619 3620 /* Slave array work handler. Holds only RTNL */ 3621 static void bond_slave_arr_handler(struct work_struct *work) 3622 { 3623 struct bonding *bond = container_of(work, struct bonding, 3624 slave_arr_work.work); 3625 int ret; 3626 3627 if (!rtnl_trylock()) 3628 goto err; 3629 3630 ret = bond_update_slave_arr(bond, NULL); 3631 rtnl_unlock(); 3632 if (ret) { 3633 pr_warn_ratelimited("Failed to update slave array from WT\n"); 3634 goto err; 3635 } 3636 return; 3637 3638 err: 3639 bond_slave_arr_work_rearm(bond, 1); 3640 } 3641 3642 /* Build the usable slaves array in control path for modes that use xmit-hash 3643 * to determine the slave interface - 3644 * (a) BOND_MODE_8023AD 3645 * (b) BOND_MODE_XOR 3646 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0 3647 * 3648 * The caller is expected to hold RTNL only and NO other lock! 3649 */ 3650 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 3651 { 3652 struct slave *slave; 3653 struct list_head *iter; 3654 struct bond_up_slave *new_arr, *old_arr; 3655 int slaves_in_agg; 3656 int agg_id = 0; 3657 int ret = 0; 3658 3659 #ifdef CONFIG_LOCKDEP 3660 WARN_ON(lockdep_is_held(&bond->mode_lock)); 3661 #endif 3662 3663 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 3664 GFP_KERNEL); 3665 if (!new_arr) { 3666 ret = -ENOMEM; 3667 pr_err("Failed to build slave-array.\n"); 3668 goto out; 3669 } 3670 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3671 struct ad_info ad_info; 3672 3673 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 3674 pr_debug("bond_3ad_get_active_agg_info failed\n"); 3675 kfree_rcu(new_arr, rcu); 3676 /* No active aggragator means it's not safe to use 3677 * the previous array. 3678 */ 3679 old_arr = rtnl_dereference(bond->slave_arr); 3680 if (old_arr) { 3681 RCU_INIT_POINTER(bond->slave_arr, NULL); 3682 kfree_rcu(old_arr, rcu); 3683 } 3684 goto out; 3685 } 3686 slaves_in_agg = ad_info.ports; 3687 agg_id = ad_info.aggregator_id; 3688 } 3689 bond_for_each_slave(bond, slave, iter) { 3690 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3691 struct aggregator *agg; 3692 3693 agg = SLAVE_AD_INFO(slave)->port.aggregator; 3694 if (!agg || agg->aggregator_identifier != agg_id) 3695 continue; 3696 } 3697 if (!bond_slave_can_tx(slave)) 3698 continue; 3699 if (skipslave == slave) 3700 continue; 3701 new_arr->arr[new_arr->count++] = slave; 3702 } 3703 3704 old_arr = rtnl_dereference(bond->slave_arr); 3705 rcu_assign_pointer(bond->slave_arr, new_arr); 3706 if (old_arr) 3707 kfree_rcu(old_arr, rcu); 3708 out: 3709 if (ret != 0 && skipslave) { 3710 int idx; 3711 3712 /* Rare situation where caller has asked to skip a specific 3713 * slave but allocation failed (most likely!). BTW this is 3714 * only possible when the call is initiated from 3715 * __bond_release_one(). In this situation; overwrite the 3716 * skipslave entry in the array with the last entry from the 3717 * array to avoid a situation where the xmit path may choose 3718 * this to-be-skipped slave to send a packet out. 3719 */ 3720 old_arr = rtnl_dereference(bond->slave_arr); 3721 for (idx = 0; idx < old_arr->count; idx++) { 3722 if (skipslave == old_arr->arr[idx]) { 3723 old_arr->arr[idx] = 3724 old_arr->arr[old_arr->count-1]; 3725 old_arr->count--; 3726 break; 3727 } 3728 } 3729 } 3730 return ret; 3731 } 3732 3733 /* Use this Xmit function for 3AD as well as XOR modes. The current 3734 * usable slave array is formed in the control path. The xmit function 3735 * just calculates hash and sends the packet out. 3736 */ 3737 int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev) 3738 { 3739 struct bonding *bond = netdev_priv(dev); 3740 struct slave *slave; 3741 struct bond_up_slave *slaves; 3742 unsigned int count; 3743 3744 slaves = rcu_dereference(bond->slave_arr); 3745 count = slaves ? ACCESS_ONCE(slaves->count) : 0; 3746 if (likely(count)) { 3747 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 3748 bond_dev_queue_xmit(bond, skb, slave->dev); 3749 } else { 3750 bond_tx_drop(dev, skb); 3751 } 3752 3753 return NETDEV_TX_OK; 3754 } 3755 3756 /* in broadcast mode, we send everything to all usable interfaces. */ 3757 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev) 3758 { 3759 struct bonding *bond = netdev_priv(bond_dev); 3760 struct slave *slave = NULL; 3761 struct list_head *iter; 3762 3763 bond_for_each_slave_rcu(bond, slave, iter) { 3764 if (bond_is_last_slave(bond, slave)) 3765 break; 3766 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3767 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 3768 3769 if (!skb2) { 3770 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 3771 bond_dev->name, __func__); 3772 continue; 3773 } 3774 bond_dev_queue_xmit(bond, skb2, slave->dev); 3775 } 3776 } 3777 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 3778 bond_dev_queue_xmit(bond, skb, slave->dev); 3779 else 3780 bond_tx_drop(bond_dev, skb); 3781 3782 return NETDEV_TX_OK; 3783 } 3784 3785 /*------------------------- Device initialization ---------------------------*/ 3786 3787 /* Lookup the slave that corresponds to a qid */ 3788 static inline int bond_slave_override(struct bonding *bond, 3789 struct sk_buff *skb) 3790 { 3791 struct slave *slave = NULL; 3792 struct list_head *iter; 3793 3794 if (!skb->queue_mapping) 3795 return 1; 3796 3797 /* Find out if any slaves have the same mapping as this skb. */ 3798 bond_for_each_slave_rcu(bond, slave, iter) { 3799 if (slave->queue_id == skb->queue_mapping) { 3800 if (bond_slave_can_tx(slave)) { 3801 bond_dev_queue_xmit(bond, skb, slave->dev); 3802 return 0; 3803 } 3804 /* If the slave isn't UP, use default transmit policy. */ 3805 break; 3806 } 3807 } 3808 3809 return 1; 3810 } 3811 3812 3813 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 3814 void *accel_priv, select_queue_fallback_t fallback) 3815 { 3816 /* This helper function exists to help dev_pick_tx get the correct 3817 * destination queue. Using a helper function skips a call to 3818 * skb_tx_hash and will put the skbs in the queue we expect on their 3819 * way down to the bonding driver. 3820 */ 3821 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 3822 3823 /* Save the original txq to restore before passing to the driver */ 3824 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping; 3825 3826 if (unlikely(txq >= dev->real_num_tx_queues)) { 3827 do { 3828 txq -= dev->real_num_tx_queues; 3829 } while (txq >= dev->real_num_tx_queues); 3830 } 3831 return txq; 3832 } 3833 3834 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 3835 { 3836 struct bonding *bond = netdev_priv(dev); 3837 3838 if (bond_should_override_tx_queue(bond) && 3839 !bond_slave_override(bond, skb)) 3840 return NETDEV_TX_OK; 3841 3842 switch (BOND_MODE(bond)) { 3843 case BOND_MODE_ROUNDROBIN: 3844 return bond_xmit_roundrobin(skb, dev); 3845 case BOND_MODE_ACTIVEBACKUP: 3846 return bond_xmit_activebackup(skb, dev); 3847 case BOND_MODE_8023AD: 3848 case BOND_MODE_XOR: 3849 return bond_3ad_xor_xmit(skb, dev); 3850 case BOND_MODE_BROADCAST: 3851 return bond_xmit_broadcast(skb, dev); 3852 case BOND_MODE_ALB: 3853 return bond_alb_xmit(skb, dev); 3854 case BOND_MODE_TLB: 3855 return bond_tlb_xmit(skb, dev); 3856 default: 3857 /* Should never happen, mode already checked */ 3858 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 3859 WARN_ON_ONCE(1); 3860 bond_tx_drop(dev, skb); 3861 return NETDEV_TX_OK; 3862 } 3863 } 3864 3865 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 3866 { 3867 struct bonding *bond = netdev_priv(dev); 3868 netdev_tx_t ret = NETDEV_TX_OK; 3869 3870 /* If we risk deadlock from transmitting this in the 3871 * netpoll path, tell netpoll to queue the frame for later tx 3872 */ 3873 if (unlikely(is_netpoll_tx_blocked(dev))) 3874 return NETDEV_TX_BUSY; 3875 3876 rcu_read_lock(); 3877 if (bond_has_slaves(bond)) 3878 ret = __bond_start_xmit(skb, dev); 3879 else 3880 bond_tx_drop(dev, skb); 3881 rcu_read_unlock(); 3882 3883 return ret; 3884 } 3885 3886 static int bond_ethtool_get_settings(struct net_device *bond_dev, 3887 struct ethtool_cmd *ecmd) 3888 { 3889 struct bonding *bond = netdev_priv(bond_dev); 3890 unsigned long speed = 0; 3891 struct list_head *iter; 3892 struct slave *slave; 3893 3894 ecmd->duplex = DUPLEX_UNKNOWN; 3895 ecmd->port = PORT_OTHER; 3896 3897 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 3898 * do not need to check mode. Though link speed might not represent 3899 * the true receive or transmit bandwidth (not all modes are symmetric) 3900 * this is an accurate maximum. 3901 */ 3902 bond_for_each_slave(bond, slave, iter) { 3903 if (bond_slave_can_tx(slave)) { 3904 if (slave->speed != SPEED_UNKNOWN) 3905 speed += slave->speed; 3906 if (ecmd->duplex == DUPLEX_UNKNOWN && 3907 slave->duplex != DUPLEX_UNKNOWN) 3908 ecmd->duplex = slave->duplex; 3909 } 3910 } 3911 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN); 3912 3913 return 0; 3914 } 3915 3916 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 3917 struct ethtool_drvinfo *drvinfo) 3918 { 3919 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 3920 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 3921 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 3922 BOND_ABI_VERSION); 3923 } 3924 3925 static const struct ethtool_ops bond_ethtool_ops = { 3926 .get_drvinfo = bond_ethtool_get_drvinfo, 3927 .get_settings = bond_ethtool_get_settings, 3928 .get_link = ethtool_op_get_link, 3929 }; 3930 3931 static const struct net_device_ops bond_netdev_ops = { 3932 .ndo_init = bond_init, 3933 .ndo_uninit = bond_uninit, 3934 .ndo_open = bond_open, 3935 .ndo_stop = bond_close, 3936 .ndo_start_xmit = bond_start_xmit, 3937 .ndo_select_queue = bond_select_queue, 3938 .ndo_get_stats64 = bond_get_stats, 3939 .ndo_do_ioctl = bond_do_ioctl, 3940 .ndo_change_rx_flags = bond_change_rx_flags, 3941 .ndo_set_rx_mode = bond_set_rx_mode, 3942 .ndo_change_mtu = bond_change_mtu, 3943 .ndo_set_mac_address = bond_set_mac_address, 3944 .ndo_neigh_setup = bond_neigh_setup, 3945 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 3946 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 3947 #ifdef CONFIG_NET_POLL_CONTROLLER 3948 .ndo_netpoll_setup = bond_netpoll_setup, 3949 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 3950 .ndo_poll_controller = bond_poll_controller, 3951 #endif 3952 .ndo_add_slave = bond_enslave, 3953 .ndo_del_slave = bond_release, 3954 .ndo_fix_features = bond_fix_features, 3955 }; 3956 3957 static const struct device_type bond_type = { 3958 .name = "bond", 3959 }; 3960 3961 static void bond_destructor(struct net_device *bond_dev) 3962 { 3963 struct bonding *bond = netdev_priv(bond_dev); 3964 if (bond->wq) 3965 destroy_workqueue(bond->wq); 3966 free_netdev(bond_dev); 3967 } 3968 3969 void bond_setup(struct net_device *bond_dev) 3970 { 3971 struct bonding *bond = netdev_priv(bond_dev); 3972 3973 spin_lock_init(&bond->mode_lock); 3974 bond->params = bonding_defaults; 3975 3976 /* Initialize pointers */ 3977 bond->dev = bond_dev; 3978 3979 /* Initialize the device entry points */ 3980 ether_setup(bond_dev); 3981 bond_dev->netdev_ops = &bond_netdev_ops; 3982 bond_dev->ethtool_ops = &bond_ethtool_ops; 3983 3984 bond_dev->destructor = bond_destructor; 3985 3986 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 3987 3988 /* Initialize the device options */ 3989 bond_dev->tx_queue_len = 0; 3990 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST; 3991 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT; 3992 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 3993 3994 /* don't acquire bond device's netif_tx_lock when transmitting */ 3995 bond_dev->features |= NETIF_F_LLTX; 3996 3997 /* By default, we declare the bond to be fully 3998 * VLAN hardware accelerated capable. Special 3999 * care is taken in the various xmit functions 4000 * when there are slaves that are not hw accel 4001 * capable 4002 */ 4003 4004 /* Don't allow bond devices to change network namespaces. */ 4005 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4006 4007 bond_dev->hw_features = BOND_VLAN_FEATURES | 4008 NETIF_F_HW_VLAN_CTAG_TX | 4009 NETIF_F_HW_VLAN_CTAG_RX | 4010 NETIF_F_HW_VLAN_CTAG_FILTER; 4011 4012 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM); 4013 bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL; 4014 bond_dev->features |= bond_dev->hw_features; 4015 } 4016 4017 /* Destroy a bonding device. 4018 * Must be under rtnl_lock when this function is called. 4019 */ 4020 static void bond_uninit(struct net_device *bond_dev) 4021 { 4022 struct bonding *bond = netdev_priv(bond_dev); 4023 struct list_head *iter; 4024 struct slave *slave; 4025 struct bond_up_slave *arr; 4026 4027 bond_netpoll_cleanup(bond_dev); 4028 4029 /* Release the bonded slaves */ 4030 bond_for_each_slave(bond, slave, iter) 4031 __bond_release_one(bond_dev, slave->dev, true); 4032 netdev_info(bond_dev, "Released all slaves\n"); 4033 4034 arr = rtnl_dereference(bond->slave_arr); 4035 if (arr) { 4036 RCU_INIT_POINTER(bond->slave_arr, NULL); 4037 kfree_rcu(arr, rcu); 4038 } 4039 4040 list_del(&bond->bond_list); 4041 4042 bond_debug_unregister(bond); 4043 } 4044 4045 /*------------------------- Module initialization ---------------------------*/ 4046 4047 static int bond_check_params(struct bond_params *params) 4048 { 4049 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4050 struct bond_opt_value newval; 4051 const struct bond_opt_value *valptr; 4052 int arp_all_targets_value; 4053 4054 /* Convert string parameters. */ 4055 if (mode) { 4056 bond_opt_initstr(&newval, mode); 4057 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4058 if (!valptr) { 4059 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4060 return -EINVAL; 4061 } 4062 bond_mode = valptr->value; 4063 } 4064 4065 if (xmit_hash_policy) { 4066 if ((bond_mode != BOND_MODE_XOR) && 4067 (bond_mode != BOND_MODE_8023AD) && 4068 (bond_mode != BOND_MODE_TLB)) { 4069 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4070 bond_mode_name(bond_mode)); 4071 } else { 4072 bond_opt_initstr(&newval, xmit_hash_policy); 4073 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4074 &newval); 4075 if (!valptr) { 4076 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4077 xmit_hash_policy); 4078 return -EINVAL; 4079 } 4080 xmit_hashtype = valptr->value; 4081 } 4082 } 4083 4084 if (lacp_rate) { 4085 if (bond_mode != BOND_MODE_8023AD) { 4086 pr_info("lacp_rate param is irrelevant in mode %s\n", 4087 bond_mode_name(bond_mode)); 4088 } else { 4089 bond_opt_initstr(&newval, lacp_rate); 4090 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4091 &newval); 4092 if (!valptr) { 4093 pr_err("Error: Invalid lacp rate \"%s\"\n", 4094 lacp_rate); 4095 return -EINVAL; 4096 } 4097 lacp_fast = valptr->value; 4098 } 4099 } 4100 4101 if (ad_select) { 4102 bond_opt_initstr(&newval, ad_select); 4103 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4104 &newval); 4105 if (!valptr) { 4106 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4107 return -EINVAL; 4108 } 4109 params->ad_select = valptr->value; 4110 if (bond_mode != BOND_MODE_8023AD) 4111 pr_warn("ad_select param only affects 802.3ad mode\n"); 4112 } else { 4113 params->ad_select = BOND_AD_STABLE; 4114 } 4115 4116 if (max_bonds < 0) { 4117 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4118 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4119 max_bonds = BOND_DEFAULT_MAX_BONDS; 4120 } 4121 4122 if (miimon < 0) { 4123 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4124 miimon, INT_MAX); 4125 miimon = 0; 4126 } 4127 4128 if (updelay < 0) { 4129 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4130 updelay, INT_MAX); 4131 updelay = 0; 4132 } 4133 4134 if (downdelay < 0) { 4135 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4136 downdelay, INT_MAX); 4137 downdelay = 0; 4138 } 4139 4140 if ((use_carrier != 0) && (use_carrier != 1)) { 4141 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4142 use_carrier); 4143 use_carrier = 1; 4144 } 4145 4146 if (num_peer_notif < 0 || num_peer_notif > 255) { 4147 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4148 num_peer_notif); 4149 num_peer_notif = 1; 4150 } 4151 4152 /* reset values for 802.3ad/TLB/ALB */ 4153 if (!bond_mode_uses_arp(bond_mode)) { 4154 if (!miimon) { 4155 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n"); 4156 pr_warn("Forcing miimon to 100msec\n"); 4157 miimon = BOND_DEFAULT_MIIMON; 4158 } 4159 } 4160 4161 if (tx_queues < 1 || tx_queues > 255) { 4162 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4163 tx_queues, BOND_DEFAULT_TX_QUEUES); 4164 tx_queues = BOND_DEFAULT_TX_QUEUES; 4165 } 4166 4167 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4168 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4169 all_slaves_active); 4170 all_slaves_active = 0; 4171 } 4172 4173 if (resend_igmp < 0 || resend_igmp > 255) { 4174 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4175 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4176 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4177 } 4178 4179 bond_opt_initval(&newval, packets_per_slave); 4180 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4181 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4182 packets_per_slave, USHRT_MAX); 4183 packets_per_slave = 1; 4184 } 4185 4186 if (bond_mode == BOND_MODE_ALB) { 4187 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n", 4188 updelay); 4189 } 4190 4191 if (!miimon) { 4192 if (updelay || downdelay) { 4193 /* just warn the user the up/down delay will have 4194 * no effect since miimon is zero... 4195 */ 4196 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n", 4197 updelay, downdelay); 4198 } 4199 } else { 4200 /* don't allow arp monitoring */ 4201 if (arp_interval) { 4202 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4203 miimon, arp_interval); 4204 arp_interval = 0; 4205 } 4206 4207 if ((updelay % miimon) != 0) { 4208 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4209 updelay, miimon, (updelay / miimon) * miimon); 4210 } 4211 4212 updelay /= miimon; 4213 4214 if ((downdelay % miimon) != 0) { 4215 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4216 downdelay, miimon, 4217 (downdelay / miimon) * miimon); 4218 } 4219 4220 downdelay /= miimon; 4221 } 4222 4223 if (arp_interval < 0) { 4224 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4225 arp_interval, INT_MAX); 4226 arp_interval = 0; 4227 } 4228 4229 for (arp_ip_count = 0, i = 0; 4230 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4231 __be32 ip; 4232 4233 /* not a complete check, but good enough to catch mistakes */ 4234 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4235 !bond_is_ip_target_ok(ip)) { 4236 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4237 arp_ip_target[i]); 4238 arp_interval = 0; 4239 } else { 4240 if (bond_get_targets_ip(arp_target, ip) == -1) 4241 arp_target[arp_ip_count++] = ip; 4242 else 4243 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4244 &ip); 4245 } 4246 } 4247 4248 if (arp_interval && !arp_ip_count) { 4249 /* don't allow arping if no arp_ip_target given... */ 4250 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4251 arp_interval); 4252 arp_interval = 0; 4253 } 4254 4255 if (arp_validate) { 4256 if (!arp_interval) { 4257 pr_err("arp_validate requires arp_interval\n"); 4258 return -EINVAL; 4259 } 4260 4261 bond_opt_initstr(&newval, arp_validate); 4262 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4263 &newval); 4264 if (!valptr) { 4265 pr_err("Error: invalid arp_validate \"%s\"\n", 4266 arp_validate); 4267 return -EINVAL; 4268 } 4269 arp_validate_value = valptr->value; 4270 } else { 4271 arp_validate_value = 0; 4272 } 4273 4274 arp_all_targets_value = 0; 4275 if (arp_all_targets) { 4276 bond_opt_initstr(&newval, arp_all_targets); 4277 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4278 &newval); 4279 if (!valptr) { 4280 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4281 arp_all_targets); 4282 arp_all_targets_value = 0; 4283 } else { 4284 arp_all_targets_value = valptr->value; 4285 } 4286 } 4287 4288 if (miimon) { 4289 pr_info("MII link monitoring set to %d ms\n", miimon); 4290 } else if (arp_interval) { 4291 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4292 arp_validate_value); 4293 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4294 arp_interval, valptr->string, arp_ip_count); 4295 4296 for (i = 0; i < arp_ip_count; i++) 4297 pr_cont(" %s", arp_ip_target[i]); 4298 4299 pr_cont("\n"); 4300 4301 } else if (max_bonds) { 4302 /* miimon and arp_interval not set, we need one so things 4303 * work as expected, see bonding.txt for details 4304 */ 4305 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n"); 4306 } 4307 4308 if (primary && !bond_mode_uses_primary(bond_mode)) { 4309 /* currently, using a primary only makes sense 4310 * in active backup, TLB or ALB modes 4311 */ 4312 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4313 primary, bond_mode_name(bond_mode)); 4314 primary = NULL; 4315 } 4316 4317 if (primary && primary_reselect) { 4318 bond_opt_initstr(&newval, primary_reselect); 4319 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4320 &newval); 4321 if (!valptr) { 4322 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4323 primary_reselect); 4324 return -EINVAL; 4325 } 4326 primary_reselect_value = valptr->value; 4327 } else { 4328 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4329 } 4330 4331 if (fail_over_mac) { 4332 bond_opt_initstr(&newval, fail_over_mac); 4333 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4334 &newval); 4335 if (!valptr) { 4336 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4337 fail_over_mac); 4338 return -EINVAL; 4339 } 4340 fail_over_mac_value = valptr->value; 4341 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4342 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4343 } else { 4344 fail_over_mac_value = BOND_FOM_NONE; 4345 } 4346 4347 if (lp_interval == 0) { 4348 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4349 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4350 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4351 } 4352 4353 /* fill params struct with the proper values */ 4354 params->mode = bond_mode; 4355 params->xmit_policy = xmit_hashtype; 4356 params->miimon = miimon; 4357 params->num_peer_notif = num_peer_notif; 4358 params->arp_interval = arp_interval; 4359 params->arp_validate = arp_validate_value; 4360 params->arp_all_targets = arp_all_targets_value; 4361 params->updelay = updelay; 4362 params->downdelay = downdelay; 4363 params->use_carrier = use_carrier; 4364 params->lacp_fast = lacp_fast; 4365 params->primary[0] = 0; 4366 params->primary_reselect = primary_reselect_value; 4367 params->fail_over_mac = fail_over_mac_value; 4368 params->tx_queues = tx_queues; 4369 params->all_slaves_active = all_slaves_active; 4370 params->resend_igmp = resend_igmp; 4371 params->min_links = min_links; 4372 params->lp_interval = lp_interval; 4373 params->packets_per_slave = packets_per_slave; 4374 params->tlb_dynamic_lb = 1; /* Default value */ 4375 if (packets_per_slave > 0) { 4376 params->reciprocal_packets_per_slave = 4377 reciprocal_value(packets_per_slave); 4378 } else { 4379 /* reciprocal_packets_per_slave is unused if 4380 * packets_per_slave is 0 or 1, just initialize it 4381 */ 4382 params->reciprocal_packets_per_slave = 4383 (struct reciprocal_value) { 0 }; 4384 } 4385 4386 if (primary) { 4387 strncpy(params->primary, primary, IFNAMSIZ); 4388 params->primary[IFNAMSIZ - 1] = 0; 4389 } 4390 4391 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4392 4393 return 0; 4394 } 4395 4396 static struct lock_class_key bonding_netdev_xmit_lock_key; 4397 static struct lock_class_key bonding_netdev_addr_lock_key; 4398 static struct lock_class_key bonding_tx_busylock_key; 4399 4400 static void bond_set_lockdep_class_one(struct net_device *dev, 4401 struct netdev_queue *txq, 4402 void *_unused) 4403 { 4404 lockdep_set_class(&txq->_xmit_lock, 4405 &bonding_netdev_xmit_lock_key); 4406 } 4407 4408 static void bond_set_lockdep_class(struct net_device *dev) 4409 { 4410 lockdep_set_class(&dev->addr_list_lock, 4411 &bonding_netdev_addr_lock_key); 4412 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL); 4413 dev->qdisc_tx_busylock = &bonding_tx_busylock_key; 4414 } 4415 4416 /* Called from registration process */ 4417 static int bond_init(struct net_device *bond_dev) 4418 { 4419 struct bonding *bond = netdev_priv(bond_dev); 4420 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4421 4422 netdev_dbg(bond_dev, "Begin bond_init\n"); 4423 4424 bond->wq = create_singlethread_workqueue(bond_dev->name); 4425 if (!bond->wq) 4426 return -ENOMEM; 4427 4428 bond_set_lockdep_class(bond_dev); 4429 4430 list_add_tail(&bond->bond_list, &bn->dev_list); 4431 4432 bond_prepare_sysfs_group(bond); 4433 4434 bond_debug_register(bond); 4435 4436 /* Ensure valid dev_addr */ 4437 if (is_zero_ether_addr(bond_dev->dev_addr) && 4438 bond_dev->addr_assign_type == NET_ADDR_PERM) 4439 eth_hw_addr_random(bond_dev); 4440 4441 return 0; 4442 } 4443 4444 unsigned int bond_get_num_tx_queues(void) 4445 { 4446 return tx_queues; 4447 } 4448 4449 /* Create a new bond based on the specified name and bonding parameters. 4450 * If name is NULL, obtain a suitable "bond%d" name for us. 4451 * Caller must NOT hold rtnl_lock; we need to release it here before we 4452 * set up our sysfs entries. 4453 */ 4454 int bond_create(struct net *net, const char *name) 4455 { 4456 struct net_device *bond_dev; 4457 int res; 4458 4459 rtnl_lock(); 4460 4461 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4462 name ? name : "bond%d", NET_NAME_UNKNOWN, 4463 bond_setup, tx_queues); 4464 if (!bond_dev) { 4465 pr_err("%s: eek! can't alloc netdev!\n", name); 4466 rtnl_unlock(); 4467 return -ENOMEM; 4468 } 4469 4470 dev_net_set(bond_dev, net); 4471 bond_dev->rtnl_link_ops = &bond_link_ops; 4472 4473 res = register_netdevice(bond_dev); 4474 4475 netif_carrier_off(bond_dev); 4476 4477 rtnl_unlock(); 4478 if (res < 0) 4479 bond_destructor(bond_dev); 4480 return res; 4481 } 4482 4483 static int __net_init bond_net_init(struct net *net) 4484 { 4485 struct bond_net *bn = net_generic(net, bond_net_id); 4486 4487 bn->net = net; 4488 INIT_LIST_HEAD(&bn->dev_list); 4489 4490 bond_create_proc_dir(bn); 4491 bond_create_sysfs(bn); 4492 4493 return 0; 4494 } 4495 4496 static void __net_exit bond_net_exit(struct net *net) 4497 { 4498 struct bond_net *bn = net_generic(net, bond_net_id); 4499 struct bonding *bond, *tmp_bond; 4500 LIST_HEAD(list); 4501 4502 bond_destroy_sysfs(bn); 4503 4504 /* Kill off any bonds created after unregistering bond rtnl ops */ 4505 rtnl_lock(); 4506 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4507 unregister_netdevice_queue(bond->dev, &list); 4508 unregister_netdevice_many(&list); 4509 rtnl_unlock(); 4510 4511 bond_destroy_proc_dir(bn); 4512 } 4513 4514 static struct pernet_operations bond_net_ops = { 4515 .init = bond_net_init, 4516 .exit = bond_net_exit, 4517 .id = &bond_net_id, 4518 .size = sizeof(struct bond_net), 4519 }; 4520 4521 static int __init bonding_init(void) 4522 { 4523 int i; 4524 int res; 4525 4526 pr_info("%s", bond_version); 4527 4528 res = bond_check_params(&bonding_defaults); 4529 if (res) 4530 goto out; 4531 4532 res = register_pernet_subsys(&bond_net_ops); 4533 if (res) 4534 goto out; 4535 4536 res = bond_netlink_init(); 4537 if (res) 4538 goto err_link; 4539 4540 bond_create_debugfs(); 4541 4542 for (i = 0; i < max_bonds; i++) { 4543 res = bond_create(&init_net, NULL); 4544 if (res) 4545 goto err; 4546 } 4547 4548 register_netdevice_notifier(&bond_netdev_notifier); 4549 out: 4550 return res; 4551 err: 4552 bond_destroy_debugfs(); 4553 bond_netlink_fini(); 4554 err_link: 4555 unregister_pernet_subsys(&bond_net_ops); 4556 goto out; 4557 4558 } 4559 4560 static void __exit bonding_exit(void) 4561 { 4562 unregister_netdevice_notifier(&bond_netdev_notifier); 4563 4564 bond_destroy_debugfs(); 4565 4566 bond_netlink_fini(); 4567 unregister_pernet_subsys(&bond_net_ops); 4568 4569 #ifdef CONFIG_NET_POLL_CONTROLLER 4570 /* Make sure we don't have an imbalance on our netpoll blocking */ 4571 WARN_ON(atomic_read(&netpoll_block_tx)); 4572 #endif 4573 } 4574 4575 module_init(bonding_init); 4576 module_exit(bonding_exit); 4577 MODULE_LICENSE("GPL"); 4578 MODULE_VERSION(DRV_VERSION); 4579 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 4580 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 4581