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_dissector.h> 80 #include <net/switchdev.h> 81 #include <net/bonding.h> 82 #include <net/bond_3ad.h> 83 #include <net/bond_alb.h> 84 85 #include "bonding_priv.h" 86 87 /*---------------------------- Module parameters ----------------------------*/ 88 89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 90 91 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 92 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 93 static int num_peer_notif = 1; 94 static int miimon; 95 static int updelay; 96 static int downdelay; 97 static int use_carrier = 1; 98 static char *mode; 99 static char *primary; 100 static char *primary_reselect; 101 static char *lacp_rate; 102 static int min_links; 103 static char *ad_select; 104 static char *xmit_hash_policy; 105 static int arp_interval; 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 107 static char *arp_validate; 108 static char *arp_all_targets; 109 static char *fail_over_mac; 110 static int all_slaves_active; 111 static struct bond_params bonding_defaults; 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 113 static int packets_per_slave = 1; 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 115 116 module_param(max_bonds, int, 0); 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 118 module_param(tx_queues, int, 0); 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 122 "failover event (alias of num_unsol_na)"); 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 125 "failover event (alias of num_grat_arp)"); 126 module_param(miimon, int, 0); 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 128 module_param(updelay, int, 0); 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 130 module_param(downdelay, int, 0); 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 132 "in milliseconds"); 133 module_param(use_carrier, int, 0); 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 135 "0 for off, 1 for on (default)"); 136 module_param(mode, charp, 0); 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 138 "1 for active-backup, 2 for balance-xor, " 139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 140 "6 for balance-alb"); 141 module_param(primary, charp, 0); 142 MODULE_PARM_DESC(primary, "Primary network device to use"); 143 module_param(primary_reselect, charp, 0); 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 145 "once it comes up; " 146 "0 for always (default), " 147 "1 for only if speed of primary is " 148 "better, " 149 "2 for only on active slave " 150 "failure"); 151 module_param(lacp_rate, charp, 0); 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 153 "0 for slow, 1 for fast"); 154 module_param(ad_select, charp, 0); 155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 156 "0 for stable (default), 1 for bandwidth, " 157 "2 for count"); 158 module_param(min_links, int, 0); 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 160 161 module_param(xmit_hash_policy, charp, 0); 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; " 163 "0 for layer 2 (default), 1 for layer 3+4, " 164 "2 for layer 2+3, 3 for encap layer 2+3, " 165 "4 for encap layer 3+4"); 166 module_param(arp_interval, int, 0); 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 168 module_param_array(arp_ip_target, charp, NULL, 0); 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 170 module_param(arp_validate, charp, 0); 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 172 "0 for none (default), 1 for active, " 173 "2 for backup, 3 for all"); 174 module_param(arp_all_targets, charp, 0); 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 176 module_param(fail_over_mac, charp, 0); 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 178 "the same MAC; 0 for none (default), " 179 "1 for active, 2 for follow"); 180 module_param(all_slaves_active, int, 0); 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 182 "by setting active flag for all slaves; " 183 "0 for never (default), 1 for always."); 184 module_param(resend_igmp, int, 0); 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 186 "link failure"); 187 module_param(packets_per_slave, int, 0); 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 189 "mode; 0 for a random slave, 1 packet per " 190 "slave (default), >1 packets per slave."); 191 module_param(lp_interval, uint, 0); 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 193 "the bonding driver sends learning packets to " 194 "each slaves peer switch. The default is 1."); 195 196 /*----------------------------- Global variables ----------------------------*/ 197 198 #ifdef CONFIG_NET_POLL_CONTROLLER 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 200 #endif 201 202 unsigned int bond_net_id __read_mostly; 203 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS]; 205 static int arp_ip_count; 206 static int bond_mode = BOND_MODE_ROUNDROBIN; 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 208 static int lacp_fast; 209 210 /*-------------------------- Forward declarations ---------------------------*/ 211 212 static int bond_init(struct net_device *bond_dev); 213 static void bond_uninit(struct net_device *bond_dev); 214 static void bond_get_stats(struct net_device *bond_dev, 215 struct rtnl_link_stats64 *stats); 216 static void bond_slave_arr_handler(struct work_struct *work); 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 218 int mod); 219 220 /*---------------------------- General routines -----------------------------*/ 221 222 const char *bond_mode_name(int mode) 223 { 224 static const char *names[] = { 225 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 226 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 227 [BOND_MODE_XOR] = "load balancing (xor)", 228 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 229 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 230 [BOND_MODE_TLB] = "transmit load balancing", 231 [BOND_MODE_ALB] = "adaptive load balancing", 232 }; 233 234 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 235 return "unknown"; 236 237 return names[mode]; 238 } 239 240 /*---------------------------------- VLAN -----------------------------------*/ 241 242 /** 243 * bond_dev_queue_xmit - Prepare skb for xmit. 244 * 245 * @bond: bond device that got this skb for tx. 246 * @skb: hw accel VLAN tagged skb to transmit 247 * @slave_dev: slave that is supposed to xmit this skbuff 248 */ 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 250 struct net_device *slave_dev) 251 { 252 skb->dev = slave_dev; 253 254 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 255 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 256 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping; 257 258 if (unlikely(netpoll_tx_running(bond->dev))) 259 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 260 else 261 dev_queue_xmit(skb); 262 } 263 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 265 * We don't protect the slave list iteration with a lock because: 266 * a. This operation is performed in IOCTL context, 267 * b. The operation is protected by the RTNL semaphore in the 8021q code, 268 * c. Holding a lock with BH disabled while directly calling a base driver 269 * entry point is generally a BAD idea. 270 * 271 * The design of synchronization/protection for this operation in the 8021q 272 * module is good for one or more VLAN devices over a single physical device 273 * and cannot be extended for a teaming solution like bonding, so there is a 274 * potential race condition here where a net device from the vlan group might 275 * be referenced (either by a base driver or the 8021q code) while it is being 276 * removed from the system. However, it turns out we're not making matters 277 * worse, and if it works for regular VLAN usage it will work here too. 278 */ 279 280 /** 281 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 282 * @bond_dev: bonding net device that got called 283 * @vid: vlan id being added 284 */ 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 286 __be16 proto, u16 vid) 287 { 288 struct bonding *bond = netdev_priv(bond_dev); 289 struct slave *slave, *rollback_slave; 290 struct list_head *iter; 291 int res; 292 293 bond_for_each_slave(bond, slave, iter) { 294 res = vlan_vid_add(slave->dev, proto, vid); 295 if (res) 296 goto unwind; 297 } 298 299 return 0; 300 301 unwind: 302 /* unwind to the slave that failed */ 303 bond_for_each_slave(bond, rollback_slave, iter) { 304 if (rollback_slave == slave) 305 break; 306 307 vlan_vid_del(rollback_slave->dev, proto, vid); 308 } 309 310 return res; 311 } 312 313 /** 314 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 315 * @bond_dev: bonding net device that got called 316 * @vid: vlan id being removed 317 */ 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 319 __be16 proto, u16 vid) 320 { 321 struct bonding *bond = netdev_priv(bond_dev); 322 struct list_head *iter; 323 struct slave *slave; 324 325 bond_for_each_slave(bond, slave, iter) 326 vlan_vid_del(slave->dev, proto, vid); 327 328 if (bond_is_lb(bond)) 329 bond_alb_clear_vlan(bond, vid); 330 331 return 0; 332 } 333 334 /*------------------------------- Link status -------------------------------*/ 335 336 /* Set the carrier state for the master according to the state of its 337 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 338 * do special 802.3ad magic. 339 * 340 * Returns zero if carrier state does not change, nonzero if it does. 341 */ 342 int bond_set_carrier(struct bonding *bond) 343 { 344 struct list_head *iter; 345 struct slave *slave; 346 347 if (!bond_has_slaves(bond)) 348 goto down; 349 350 if (BOND_MODE(bond) == BOND_MODE_8023AD) 351 return bond_3ad_set_carrier(bond); 352 353 bond_for_each_slave(bond, slave, iter) { 354 if (slave->link == BOND_LINK_UP) { 355 if (!netif_carrier_ok(bond->dev)) { 356 netif_carrier_on(bond->dev); 357 return 1; 358 } 359 return 0; 360 } 361 } 362 363 down: 364 if (netif_carrier_ok(bond->dev)) { 365 netif_carrier_off(bond->dev); 366 return 1; 367 } 368 return 0; 369 } 370 371 /* Get link speed and duplex from the slave's base driver 372 * using ethtool. If for some reason the call fails or the 373 * values are invalid, set speed and duplex to -1, 374 * and return. 375 */ 376 static void bond_update_speed_duplex(struct slave *slave) 377 { 378 struct net_device *slave_dev = slave->dev; 379 struct ethtool_link_ksettings ecmd; 380 int res; 381 382 slave->speed = SPEED_UNKNOWN; 383 slave->duplex = DUPLEX_UNKNOWN; 384 385 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 386 if (res < 0) 387 return; 388 389 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 390 return; 391 392 switch (ecmd.base.duplex) { 393 case DUPLEX_FULL: 394 case DUPLEX_HALF: 395 break; 396 default: 397 return; 398 } 399 400 slave->speed = ecmd.base.speed; 401 slave->duplex = ecmd.base.duplex; 402 403 return; 404 } 405 406 const char *bond_slave_link_status(s8 link) 407 { 408 switch (link) { 409 case BOND_LINK_UP: 410 return "up"; 411 case BOND_LINK_FAIL: 412 return "going down"; 413 case BOND_LINK_DOWN: 414 return "down"; 415 case BOND_LINK_BACK: 416 return "going back"; 417 default: 418 return "unknown"; 419 } 420 } 421 422 /* if <dev> supports MII link status reporting, check its link status. 423 * 424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 425 * depending upon the setting of the use_carrier parameter. 426 * 427 * Return either BMSR_LSTATUS, meaning that the link is up (or we 428 * can't tell and just pretend it is), or 0, meaning that the link is 429 * down. 430 * 431 * If reporting is non-zero, instead of faking link up, return -1 if 432 * both ETHTOOL and MII ioctls fail (meaning the device does not 433 * support them). If use_carrier is set, return whatever it says. 434 * It'd be nice if there was a good way to tell if a driver supports 435 * netif_carrier, but there really isn't. 436 */ 437 static int bond_check_dev_link(struct bonding *bond, 438 struct net_device *slave_dev, int reporting) 439 { 440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 441 int (*ioctl)(struct net_device *, struct ifreq *, int); 442 struct ifreq ifr; 443 struct mii_ioctl_data *mii; 444 445 if (!reporting && !netif_running(slave_dev)) 446 return 0; 447 448 if (bond->params.use_carrier) 449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 450 451 /* Try to get link status using Ethtool first. */ 452 if (slave_dev->ethtool_ops->get_link) 453 return slave_dev->ethtool_ops->get_link(slave_dev) ? 454 BMSR_LSTATUS : 0; 455 456 /* Ethtool can't be used, fallback to MII ioctls. */ 457 ioctl = slave_ops->ndo_do_ioctl; 458 if (ioctl) { 459 /* TODO: set pointer to correct ioctl on a per team member 460 * bases to make this more efficient. that is, once 461 * we determine the correct ioctl, we will always 462 * call it and not the others for that team 463 * member. 464 */ 465 466 /* We cannot assume that SIOCGMIIPHY will also read a 467 * register; not all network drivers (e.g., e100) 468 * support that. 469 */ 470 471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 473 mii = if_mii(&ifr); 474 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 475 mii->reg_num = MII_BMSR; 476 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 477 return mii->val_out & BMSR_LSTATUS; 478 } 479 } 480 481 /* If reporting, report that either there's no dev->do_ioctl, 482 * or both SIOCGMIIREG and get_link failed (meaning that we 483 * cannot report link status). If not reporting, pretend 484 * we're ok. 485 */ 486 return reporting ? -1 : BMSR_LSTATUS; 487 } 488 489 /*----------------------------- Multicast list ------------------------------*/ 490 491 /* Push the promiscuity flag down to appropriate slaves */ 492 static int bond_set_promiscuity(struct bonding *bond, int inc) 493 { 494 struct list_head *iter; 495 int err = 0; 496 497 if (bond_uses_primary(bond)) { 498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 499 500 if (curr_active) 501 err = dev_set_promiscuity(curr_active->dev, inc); 502 } else { 503 struct slave *slave; 504 505 bond_for_each_slave(bond, slave, iter) { 506 err = dev_set_promiscuity(slave->dev, inc); 507 if (err) 508 return err; 509 } 510 } 511 return err; 512 } 513 514 /* Push the allmulti flag down to all slaves */ 515 static int bond_set_allmulti(struct bonding *bond, int inc) 516 { 517 struct list_head *iter; 518 int err = 0; 519 520 if (bond_uses_primary(bond)) { 521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 522 523 if (curr_active) 524 err = dev_set_allmulti(curr_active->dev, inc); 525 } else { 526 struct slave *slave; 527 528 bond_for_each_slave(bond, slave, iter) { 529 err = dev_set_allmulti(slave->dev, inc); 530 if (err) 531 return err; 532 } 533 } 534 return err; 535 } 536 537 /* Retrieve the list of registered multicast addresses for the bonding 538 * device and retransmit an IGMP JOIN request to the current active 539 * slave. 540 */ 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 542 { 543 struct bonding *bond = container_of(work, struct bonding, 544 mcast_work.work); 545 546 if (!rtnl_trylock()) { 547 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 548 return; 549 } 550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 551 552 if (bond->igmp_retrans > 1) { 553 bond->igmp_retrans--; 554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 555 } 556 rtnl_unlock(); 557 } 558 559 /* Flush bond's hardware addresses from slave */ 560 static void bond_hw_addr_flush(struct net_device *bond_dev, 561 struct net_device *slave_dev) 562 { 563 struct bonding *bond = netdev_priv(bond_dev); 564 565 dev_uc_unsync(slave_dev, bond_dev); 566 dev_mc_unsync(slave_dev, bond_dev); 567 568 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 569 /* del lacpdu mc addr from mc list */ 570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 571 572 dev_mc_del(slave_dev, lacpdu_multicast); 573 } 574 } 575 576 /*--------------------------- Active slave change ---------------------------*/ 577 578 /* Update the hardware address list and promisc/allmulti for the new and 579 * old active slaves (if any). Modes that are not using primary keep all 580 * slaves up date at all times; only the modes that use primary need to call 581 * this function to swap these settings during a failover. 582 */ 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 584 struct slave *old_active) 585 { 586 if (old_active) { 587 if (bond->dev->flags & IFF_PROMISC) 588 dev_set_promiscuity(old_active->dev, -1); 589 590 if (bond->dev->flags & IFF_ALLMULTI) 591 dev_set_allmulti(old_active->dev, -1); 592 593 bond_hw_addr_flush(bond->dev, old_active->dev); 594 } 595 596 if (new_active) { 597 /* FIXME: Signal errors upstream. */ 598 if (bond->dev->flags & IFF_PROMISC) 599 dev_set_promiscuity(new_active->dev, 1); 600 601 if (bond->dev->flags & IFF_ALLMULTI) 602 dev_set_allmulti(new_active->dev, 1); 603 604 netif_addr_lock_bh(bond->dev); 605 dev_uc_sync(new_active->dev, bond->dev); 606 dev_mc_sync(new_active->dev, bond->dev); 607 netif_addr_unlock_bh(bond->dev); 608 } 609 } 610 611 /** 612 * bond_set_dev_addr - clone slave's address to bond 613 * @bond_dev: bond net device 614 * @slave_dev: slave net device 615 * 616 * Should be called with RTNL held. 617 */ 618 static void bond_set_dev_addr(struct net_device *bond_dev, 619 struct net_device *slave_dev) 620 { 621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n", 622 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len); 623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 624 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 626 } 627 628 static struct slave *bond_get_old_active(struct bonding *bond, 629 struct slave *new_active) 630 { 631 struct slave *slave; 632 struct list_head *iter; 633 634 bond_for_each_slave(bond, slave, iter) { 635 if (slave == new_active) 636 continue; 637 638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 639 return slave; 640 } 641 642 return NULL; 643 } 644 645 /* bond_do_fail_over_mac 646 * 647 * Perform special MAC address swapping for fail_over_mac settings 648 * 649 * Called with RTNL 650 */ 651 static void bond_do_fail_over_mac(struct bonding *bond, 652 struct slave *new_active, 653 struct slave *old_active) 654 { 655 u8 tmp_mac[ETH_ALEN]; 656 struct sockaddr saddr; 657 int rv; 658 659 switch (bond->params.fail_over_mac) { 660 case BOND_FOM_ACTIVE: 661 if (new_active) 662 bond_set_dev_addr(bond->dev, new_active->dev); 663 break; 664 case BOND_FOM_FOLLOW: 665 /* if new_active && old_active, swap them 666 * if just old_active, do nothing (going to no active slave) 667 * if just new_active, set new_active to bond's MAC 668 */ 669 if (!new_active) 670 return; 671 672 if (!old_active) 673 old_active = bond_get_old_active(bond, new_active); 674 675 if (old_active) { 676 ether_addr_copy(tmp_mac, new_active->dev->dev_addr); 677 ether_addr_copy(saddr.sa_data, 678 old_active->dev->dev_addr); 679 saddr.sa_family = new_active->dev->type; 680 } else { 681 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr); 682 saddr.sa_family = bond->dev->type; 683 } 684 685 rv = dev_set_mac_address(new_active->dev, &saddr); 686 if (rv) { 687 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 688 -rv, new_active->dev->name); 689 goto out; 690 } 691 692 if (!old_active) 693 goto out; 694 695 ether_addr_copy(saddr.sa_data, tmp_mac); 696 saddr.sa_family = old_active->dev->type; 697 698 rv = dev_set_mac_address(old_active->dev, &saddr); 699 if (rv) 700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 701 -rv, new_active->dev->name); 702 out: 703 break; 704 default: 705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 706 bond->params.fail_over_mac); 707 break; 708 } 709 710 } 711 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 713 { 714 struct slave *prim = rtnl_dereference(bond->primary_slave); 715 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 716 717 if (!prim || prim->link != BOND_LINK_UP) { 718 if (!curr || curr->link != BOND_LINK_UP) 719 return NULL; 720 return curr; 721 } 722 723 if (bond->force_primary) { 724 bond->force_primary = false; 725 return prim; 726 } 727 728 if (!curr || curr->link != BOND_LINK_UP) 729 return prim; 730 731 /* At this point, prim and curr are both up */ 732 switch (bond->params.primary_reselect) { 733 case BOND_PRI_RESELECT_ALWAYS: 734 return prim; 735 case BOND_PRI_RESELECT_BETTER: 736 if (prim->speed < curr->speed) 737 return curr; 738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 739 return curr; 740 return prim; 741 case BOND_PRI_RESELECT_FAILURE: 742 return curr; 743 default: 744 netdev_err(bond->dev, "impossible primary_reselect %d\n", 745 bond->params.primary_reselect); 746 return curr; 747 } 748 } 749 750 /** 751 * bond_find_best_slave - select the best available slave to be the active one 752 * @bond: our bonding struct 753 */ 754 static struct slave *bond_find_best_slave(struct bonding *bond) 755 { 756 struct slave *slave, *bestslave = NULL; 757 struct list_head *iter; 758 int mintime = bond->params.updelay; 759 760 slave = bond_choose_primary_or_current(bond); 761 if (slave) 762 return slave; 763 764 bond_for_each_slave(bond, slave, iter) { 765 if (slave->link == BOND_LINK_UP) 766 return slave; 767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 768 slave->delay < mintime) { 769 mintime = slave->delay; 770 bestslave = slave; 771 } 772 } 773 774 return bestslave; 775 } 776 777 static bool bond_should_notify_peers(struct bonding *bond) 778 { 779 struct slave *slave; 780 781 rcu_read_lock(); 782 slave = rcu_dereference(bond->curr_active_slave); 783 rcu_read_unlock(); 784 785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 786 slave ? slave->dev->name : "NULL"); 787 788 if (!slave || !bond->send_peer_notif || 789 !netif_carrier_ok(bond->dev) || 790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 791 return false; 792 793 return true; 794 } 795 796 /** 797 * change_active_interface - change the active slave into the specified one 798 * @bond: our bonding struct 799 * @new: the new slave to make the active one 800 * 801 * Set the new slave to the bond's settings and unset them on the old 802 * curr_active_slave. 803 * Setting include flags, mc-list, promiscuity, allmulti, etc. 804 * 805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 806 * because it is apparently the best available slave we have, even though its 807 * updelay hasn't timed out yet. 808 * 809 * Caller must hold RTNL. 810 */ 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 812 { 813 struct slave *old_active; 814 815 ASSERT_RTNL(); 816 817 old_active = rtnl_dereference(bond->curr_active_slave); 818 819 if (old_active == new_active) 820 return; 821 822 if (new_active) { 823 new_active->last_link_up = jiffies; 824 825 if (new_active->link == BOND_LINK_BACK) { 826 if (bond_uses_primary(bond)) { 827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n", 828 new_active->dev->name, 829 (bond->params.updelay - new_active->delay) * bond->params.miimon); 830 } 831 832 new_active->delay = 0; 833 bond_set_slave_link_state(new_active, BOND_LINK_UP, 834 BOND_SLAVE_NOTIFY_NOW); 835 836 if (BOND_MODE(bond) == BOND_MODE_8023AD) 837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 838 839 if (bond_is_lb(bond)) 840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 841 } else { 842 if (bond_uses_primary(bond)) { 843 netdev_info(bond->dev, "making interface %s the new active one\n", 844 new_active->dev->name); 845 } 846 } 847 } 848 849 if (bond_uses_primary(bond)) 850 bond_hw_addr_swap(bond, new_active, old_active); 851 852 if (bond_is_lb(bond)) { 853 bond_alb_handle_active_change(bond, new_active); 854 if (old_active) 855 bond_set_slave_inactive_flags(old_active, 856 BOND_SLAVE_NOTIFY_NOW); 857 if (new_active) 858 bond_set_slave_active_flags(new_active, 859 BOND_SLAVE_NOTIFY_NOW); 860 } else { 861 rcu_assign_pointer(bond->curr_active_slave, new_active); 862 } 863 864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 865 if (old_active) 866 bond_set_slave_inactive_flags(old_active, 867 BOND_SLAVE_NOTIFY_NOW); 868 869 if (new_active) { 870 bool should_notify_peers = false; 871 872 bond_set_slave_active_flags(new_active, 873 BOND_SLAVE_NOTIFY_NOW); 874 875 if (bond->params.fail_over_mac) 876 bond_do_fail_over_mac(bond, new_active, 877 old_active); 878 879 if (netif_running(bond->dev)) { 880 bond->send_peer_notif = 881 bond->params.num_peer_notif; 882 should_notify_peers = 883 bond_should_notify_peers(bond); 884 } 885 886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 887 if (should_notify_peers) 888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 889 bond->dev); 890 } 891 } 892 893 /* resend IGMP joins since active slave has changed or 894 * all were sent on curr_active_slave. 895 * resend only if bond is brought up with the affected 896 * bonding modes and the retransmission is enabled 897 */ 898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 899 ((bond_uses_primary(bond) && new_active) || 900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 901 bond->igmp_retrans = bond->params.resend_igmp; 902 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 903 } 904 } 905 906 /** 907 * bond_select_active_slave - select a new active slave, if needed 908 * @bond: our bonding struct 909 * 910 * This functions should be called when one of the following occurs: 911 * - The old curr_active_slave has been released or lost its link. 912 * - The primary_slave has got its link back. 913 * - A slave has got its link back and there's no old curr_active_slave. 914 * 915 * Caller must hold RTNL. 916 */ 917 void bond_select_active_slave(struct bonding *bond) 918 { 919 struct slave *best_slave; 920 int rv; 921 922 ASSERT_RTNL(); 923 924 best_slave = bond_find_best_slave(bond); 925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 926 bond_change_active_slave(bond, best_slave); 927 rv = bond_set_carrier(bond); 928 if (!rv) 929 return; 930 931 if (netif_carrier_ok(bond->dev)) 932 netdev_info(bond->dev, "first active interface up!\n"); 933 else 934 netdev_info(bond->dev, "now running without any active interface!\n"); 935 } 936 } 937 938 #ifdef CONFIG_NET_POLL_CONTROLLER 939 static inline int slave_enable_netpoll(struct slave *slave) 940 { 941 struct netpoll *np; 942 int err = 0; 943 944 np = kzalloc(sizeof(*np), GFP_KERNEL); 945 err = -ENOMEM; 946 if (!np) 947 goto out; 948 949 err = __netpoll_setup(np, slave->dev); 950 if (err) { 951 kfree(np); 952 goto out; 953 } 954 slave->np = np; 955 out: 956 return err; 957 } 958 static inline void slave_disable_netpoll(struct slave *slave) 959 { 960 struct netpoll *np = slave->np; 961 962 if (!np) 963 return; 964 965 slave->np = NULL; 966 __netpoll_free_async(np); 967 } 968 969 static void bond_poll_controller(struct net_device *bond_dev) 970 { 971 struct bonding *bond = netdev_priv(bond_dev); 972 struct slave *slave = NULL; 973 struct list_head *iter; 974 struct ad_info ad_info; 975 struct netpoll_info *ni; 976 const struct net_device_ops *ops; 977 978 if (BOND_MODE(bond) == BOND_MODE_8023AD) 979 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 980 return; 981 982 bond_for_each_slave_rcu(bond, slave, iter) { 983 ops = slave->dev->netdev_ops; 984 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller) 985 continue; 986 987 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 988 struct aggregator *agg = 989 SLAVE_AD_INFO(slave)->port.aggregator; 990 991 if (agg && 992 agg->aggregator_identifier != ad_info.aggregator_id) 993 continue; 994 } 995 996 ni = rcu_dereference_bh(slave->dev->npinfo); 997 if (down_trylock(&ni->dev_lock)) 998 continue; 999 ops->ndo_poll_controller(slave->dev); 1000 up(&ni->dev_lock); 1001 } 1002 } 1003 1004 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1005 { 1006 struct bonding *bond = netdev_priv(bond_dev); 1007 struct list_head *iter; 1008 struct slave *slave; 1009 1010 bond_for_each_slave(bond, slave, iter) 1011 if (bond_slave_is_up(slave)) 1012 slave_disable_netpoll(slave); 1013 } 1014 1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1016 { 1017 struct bonding *bond = netdev_priv(dev); 1018 struct list_head *iter; 1019 struct slave *slave; 1020 int err = 0; 1021 1022 bond_for_each_slave(bond, slave, iter) { 1023 err = slave_enable_netpoll(slave); 1024 if (err) { 1025 bond_netpoll_cleanup(dev); 1026 break; 1027 } 1028 } 1029 return err; 1030 } 1031 #else 1032 static inline int slave_enable_netpoll(struct slave *slave) 1033 { 1034 return 0; 1035 } 1036 static inline void slave_disable_netpoll(struct slave *slave) 1037 { 1038 } 1039 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1040 { 1041 } 1042 #endif 1043 1044 /*---------------------------------- IOCTL ----------------------------------*/ 1045 1046 static netdev_features_t bond_fix_features(struct net_device *dev, 1047 netdev_features_t features) 1048 { 1049 struct bonding *bond = netdev_priv(dev); 1050 struct list_head *iter; 1051 netdev_features_t mask; 1052 struct slave *slave; 1053 1054 mask = features; 1055 1056 features &= ~NETIF_F_ONE_FOR_ALL; 1057 features |= NETIF_F_ALL_FOR_ALL; 1058 1059 bond_for_each_slave(bond, slave, iter) { 1060 features = netdev_increment_features(features, 1061 slave->dev->features, 1062 mask); 1063 } 1064 features = netdev_add_tso_features(features, mask); 1065 1066 return features; 1067 } 1068 1069 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1070 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1071 NETIF_F_HIGHDMA | NETIF_F_LRO) 1072 1073 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1074 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1075 1076 static void bond_compute_features(struct bonding *bond) 1077 { 1078 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1079 IFF_XMIT_DST_RELEASE_PERM; 1080 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1081 netdev_features_t enc_features = BOND_ENC_FEATURES; 1082 struct net_device *bond_dev = bond->dev; 1083 struct list_head *iter; 1084 struct slave *slave; 1085 unsigned short max_hard_header_len = ETH_HLEN; 1086 unsigned int gso_max_size = GSO_MAX_SIZE; 1087 u16 gso_max_segs = GSO_MAX_SEGS; 1088 1089 if (!bond_has_slaves(bond)) 1090 goto done; 1091 vlan_features &= NETIF_F_ALL_FOR_ALL; 1092 1093 bond_for_each_slave(bond, slave, iter) { 1094 vlan_features = netdev_increment_features(vlan_features, 1095 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1096 1097 enc_features = netdev_increment_features(enc_features, 1098 slave->dev->hw_enc_features, 1099 BOND_ENC_FEATURES); 1100 dst_release_flag &= slave->dev->priv_flags; 1101 if (slave->dev->hard_header_len > max_hard_header_len) 1102 max_hard_header_len = slave->dev->hard_header_len; 1103 1104 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1105 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1106 } 1107 bond_dev->hard_header_len = max_hard_header_len; 1108 1109 done: 1110 bond_dev->vlan_features = vlan_features; 1111 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL; 1112 bond_dev->gso_max_segs = gso_max_segs; 1113 netif_set_gso_max_size(bond_dev, gso_max_size); 1114 1115 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1116 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1117 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1118 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1119 1120 netdev_change_features(bond_dev); 1121 } 1122 1123 static void bond_setup_by_slave(struct net_device *bond_dev, 1124 struct net_device *slave_dev) 1125 { 1126 bond_dev->header_ops = slave_dev->header_ops; 1127 1128 bond_dev->type = slave_dev->type; 1129 bond_dev->hard_header_len = slave_dev->hard_header_len; 1130 bond_dev->addr_len = slave_dev->addr_len; 1131 1132 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1133 slave_dev->addr_len); 1134 } 1135 1136 /* On bonding slaves other than the currently active slave, suppress 1137 * duplicates except for alb non-mcast/bcast. 1138 */ 1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1140 struct slave *slave, 1141 struct bonding *bond) 1142 { 1143 if (bond_is_slave_inactive(slave)) { 1144 if (BOND_MODE(bond) == BOND_MODE_ALB && 1145 skb->pkt_type != PACKET_BROADCAST && 1146 skb->pkt_type != PACKET_MULTICAST) 1147 return false; 1148 return true; 1149 } 1150 return false; 1151 } 1152 1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1154 { 1155 struct sk_buff *skb = *pskb; 1156 struct slave *slave; 1157 struct bonding *bond; 1158 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1159 struct slave *); 1160 int ret = RX_HANDLER_ANOTHER; 1161 1162 skb = skb_share_check(skb, GFP_ATOMIC); 1163 if (unlikely(!skb)) 1164 return RX_HANDLER_CONSUMED; 1165 1166 *pskb = skb; 1167 1168 slave = bond_slave_get_rcu(skb->dev); 1169 bond = slave->bond; 1170 1171 recv_probe = ACCESS_ONCE(bond->recv_probe); 1172 if (recv_probe) { 1173 ret = recv_probe(skb, bond, slave); 1174 if (ret == RX_HANDLER_CONSUMED) { 1175 consume_skb(skb); 1176 return ret; 1177 } 1178 } 1179 1180 if (bond_should_deliver_exact_match(skb, slave, bond)) 1181 return RX_HANDLER_EXACT; 1182 1183 skb->dev = bond->dev; 1184 1185 if (BOND_MODE(bond) == BOND_MODE_ALB && 1186 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1187 skb->pkt_type == PACKET_HOST) { 1188 1189 if (unlikely(skb_cow_head(skb, 1190 skb->data - skb_mac_header(skb)))) { 1191 kfree_skb(skb); 1192 return RX_HANDLER_CONSUMED; 1193 } 1194 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr); 1195 } 1196 1197 return ret; 1198 } 1199 1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1201 { 1202 switch (BOND_MODE(bond)) { 1203 case BOND_MODE_ROUNDROBIN: 1204 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1205 case BOND_MODE_ACTIVEBACKUP: 1206 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1207 case BOND_MODE_BROADCAST: 1208 return NETDEV_LAG_TX_TYPE_BROADCAST; 1209 case BOND_MODE_XOR: 1210 case BOND_MODE_8023AD: 1211 return NETDEV_LAG_TX_TYPE_HASH; 1212 default: 1213 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1214 } 1215 } 1216 1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave) 1218 { 1219 struct netdev_lag_upper_info lag_upper_info; 1220 int err; 1221 1222 lag_upper_info.tx_type = bond_lag_tx_type(bond); 1223 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1224 &lag_upper_info); 1225 if (err) 1226 return err; 1227 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL); 1228 return 0; 1229 } 1230 1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1232 { 1233 netdev_upper_dev_unlink(slave->dev, bond->dev); 1234 slave->dev->flags &= ~IFF_SLAVE; 1235 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL); 1236 } 1237 1238 static struct slave *bond_alloc_slave(struct bonding *bond) 1239 { 1240 struct slave *slave = NULL; 1241 1242 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1243 if (!slave) 1244 return NULL; 1245 1246 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1247 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1248 GFP_KERNEL); 1249 if (!SLAVE_AD_INFO(slave)) { 1250 kfree(slave); 1251 return NULL; 1252 } 1253 } 1254 return slave; 1255 } 1256 1257 static void bond_free_slave(struct slave *slave) 1258 { 1259 struct bonding *bond = bond_get_bond_by_slave(slave); 1260 1261 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1262 kfree(SLAVE_AD_INFO(slave)); 1263 1264 kfree(slave); 1265 } 1266 1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1268 { 1269 info->bond_mode = BOND_MODE(bond); 1270 info->miimon = bond->params.miimon; 1271 info->num_slaves = bond->slave_cnt; 1272 } 1273 1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1275 { 1276 strcpy(info->slave_name, slave->dev->name); 1277 info->link = slave->link; 1278 info->state = bond_slave_state(slave); 1279 info->link_failure_count = slave->link_failure_count; 1280 } 1281 1282 static void bond_netdev_notify(struct net_device *dev, 1283 struct netdev_bonding_info *info) 1284 { 1285 rtnl_lock(); 1286 netdev_bonding_info_change(dev, info); 1287 rtnl_unlock(); 1288 } 1289 1290 static void bond_netdev_notify_work(struct work_struct *_work) 1291 { 1292 struct netdev_notify_work *w = 1293 container_of(_work, struct netdev_notify_work, work.work); 1294 1295 bond_netdev_notify(w->dev, &w->bonding_info); 1296 dev_put(w->dev); 1297 kfree(w); 1298 } 1299 1300 void bond_queue_slave_event(struct slave *slave) 1301 { 1302 struct bonding *bond = slave->bond; 1303 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC); 1304 1305 if (!nnw) 1306 return; 1307 1308 dev_hold(slave->dev); 1309 nnw->dev = slave->dev; 1310 bond_fill_ifslave(slave, &nnw->bonding_info.slave); 1311 bond_fill_ifbond(bond, &nnw->bonding_info.master); 1312 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work); 1313 1314 queue_delayed_work(slave->bond->wq, &nnw->work, 0); 1315 } 1316 1317 void bond_lower_state_changed(struct slave *slave) 1318 { 1319 struct netdev_lag_lower_state_info info; 1320 1321 info.link_up = slave->link == BOND_LINK_UP || 1322 slave->link == BOND_LINK_FAIL; 1323 info.tx_enabled = bond_is_active_slave(slave); 1324 netdev_lower_state_changed(slave->dev, &info); 1325 } 1326 1327 /* enslave device <slave> to bond device <master> */ 1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev) 1329 { 1330 struct bonding *bond = netdev_priv(bond_dev); 1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1332 struct slave *new_slave = NULL, *prev_slave; 1333 struct sockaddr addr; 1334 int link_reporting; 1335 int res = 0, i; 1336 1337 if (!bond->params.use_carrier && 1338 slave_dev->ethtool_ops->get_link == NULL && 1339 slave_ops->ndo_do_ioctl == NULL) { 1340 netdev_warn(bond_dev, "no link monitoring support for %s\n", 1341 slave_dev->name); 1342 } 1343 1344 /* already in-use? */ 1345 if (netdev_is_rx_handler_busy(slave_dev)) { 1346 netdev_err(bond_dev, 1347 "Error: Device is in use and cannot be enslaved\n"); 1348 return -EBUSY; 1349 } 1350 1351 if (bond_dev == slave_dev) { 1352 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1353 return -EPERM; 1354 } 1355 1356 /* vlan challenged mutual exclusion */ 1357 /* no need to lock since we're protected by rtnl_lock */ 1358 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1359 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n", 1360 slave_dev->name); 1361 if (vlan_uses_dev(bond_dev)) { 1362 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n", 1363 slave_dev->name, bond_dev->name); 1364 return -EPERM; 1365 } else { 1366 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n", 1367 slave_dev->name, slave_dev->name, 1368 bond_dev->name); 1369 } 1370 } else { 1371 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n", 1372 slave_dev->name); 1373 } 1374 1375 /* Old ifenslave binaries are no longer supported. These can 1376 * be identified with moderate accuracy by the state of the slave: 1377 * the current ifenslave will set the interface down prior to 1378 * enslaving it; the old ifenslave will not. 1379 */ 1380 if (slave_dev->flags & IFF_UP) { 1381 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n", 1382 slave_dev->name); 1383 return -EPERM; 1384 } 1385 1386 /* set bonding device ether type by slave - bonding netdevices are 1387 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1388 * there is a need to override some of the type dependent attribs/funcs. 1389 * 1390 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1391 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1392 */ 1393 if (!bond_has_slaves(bond)) { 1394 if (bond_dev->type != slave_dev->type) { 1395 netdev_dbg(bond_dev, "change device type from %d to %d\n", 1396 bond_dev->type, slave_dev->type); 1397 1398 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1399 bond_dev); 1400 res = notifier_to_errno(res); 1401 if (res) { 1402 netdev_err(bond_dev, "refused to change device type\n"); 1403 return -EBUSY; 1404 } 1405 1406 /* Flush unicast and multicast addresses */ 1407 dev_uc_flush(bond_dev); 1408 dev_mc_flush(bond_dev); 1409 1410 if (slave_dev->type != ARPHRD_ETHER) 1411 bond_setup_by_slave(bond_dev, slave_dev); 1412 else { 1413 ether_setup(bond_dev); 1414 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1415 } 1416 1417 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1418 bond_dev); 1419 } 1420 } else if (bond_dev->type != slave_dev->type) { 1421 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n", 1422 slave_dev->name, slave_dev->type, bond_dev->type); 1423 return -EINVAL; 1424 } 1425 1426 if (slave_dev->type == ARPHRD_INFINIBAND && 1427 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1428 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n", 1429 slave_dev->type); 1430 res = -EOPNOTSUPP; 1431 goto err_undo_flags; 1432 } 1433 1434 if (!slave_ops->ndo_set_mac_address || 1435 slave_dev->type == ARPHRD_INFINIBAND) { 1436 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n"); 1437 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1438 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1439 if (!bond_has_slaves(bond)) { 1440 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1441 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1442 } else { 1443 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"); 1444 res = -EOPNOTSUPP; 1445 goto err_undo_flags; 1446 } 1447 } 1448 } 1449 1450 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1451 1452 /* If this is the first slave, then we need to set the master's hardware 1453 * address to be the same as the slave's. 1454 */ 1455 if (!bond_has_slaves(bond) && 1456 bond->dev->addr_assign_type == NET_ADDR_RANDOM) 1457 bond_set_dev_addr(bond->dev, slave_dev); 1458 1459 new_slave = bond_alloc_slave(bond); 1460 if (!new_slave) { 1461 res = -ENOMEM; 1462 goto err_undo_flags; 1463 } 1464 1465 new_slave->bond = bond; 1466 new_slave->dev = slave_dev; 1467 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1468 * is set via sysfs or module option if desired. 1469 */ 1470 new_slave->queue_id = 0; 1471 1472 /* Save slave's original mtu and then set it to match the bond */ 1473 new_slave->original_mtu = slave_dev->mtu; 1474 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1475 if (res) { 1476 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res); 1477 goto err_free; 1478 } 1479 1480 /* Save slave's original ("permanent") mac address for modes 1481 * that need it, and for restoring it upon release, and then 1482 * set it to the master's address 1483 */ 1484 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr); 1485 1486 if (!bond->params.fail_over_mac || 1487 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1488 /* Set slave to master's mac address. The application already 1489 * set the master's mac address to that of the first slave 1490 */ 1491 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len); 1492 addr.sa_family = slave_dev->type; 1493 res = dev_set_mac_address(slave_dev, &addr); 1494 if (res) { 1495 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res); 1496 goto err_restore_mtu; 1497 } 1498 } 1499 1500 /* set slave flag before open to prevent IPv6 addrconf */ 1501 slave_dev->flags |= IFF_SLAVE; 1502 1503 /* open the slave since the application closed it */ 1504 res = dev_open(slave_dev); 1505 if (res) { 1506 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name); 1507 goto err_restore_mac; 1508 } 1509 1510 slave_dev->priv_flags |= IFF_BONDING; 1511 /* initialize slave stats */ 1512 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1513 1514 if (bond_is_lb(bond)) { 1515 /* bond_alb_init_slave() must be called before all other stages since 1516 * it might fail and we do not want to have to undo everything 1517 */ 1518 res = bond_alb_init_slave(bond, new_slave); 1519 if (res) 1520 goto err_close; 1521 } 1522 1523 /* If the mode uses primary, then the following is handled by 1524 * bond_change_active_slave(). 1525 */ 1526 if (!bond_uses_primary(bond)) { 1527 /* set promiscuity level to new slave */ 1528 if (bond_dev->flags & IFF_PROMISC) { 1529 res = dev_set_promiscuity(slave_dev, 1); 1530 if (res) 1531 goto err_close; 1532 } 1533 1534 /* set allmulti level to new slave */ 1535 if (bond_dev->flags & IFF_ALLMULTI) { 1536 res = dev_set_allmulti(slave_dev, 1); 1537 if (res) 1538 goto err_close; 1539 } 1540 1541 netif_addr_lock_bh(bond_dev); 1542 1543 dev_mc_sync_multiple(slave_dev, bond_dev); 1544 dev_uc_sync_multiple(slave_dev, bond_dev); 1545 1546 netif_addr_unlock_bh(bond_dev); 1547 } 1548 1549 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1550 /* add lacpdu mc addr to mc list */ 1551 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1552 1553 dev_mc_add(slave_dev, lacpdu_multicast); 1554 } 1555 1556 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1557 if (res) { 1558 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n", 1559 slave_dev->name); 1560 goto err_close; 1561 } 1562 1563 prev_slave = bond_last_slave(bond); 1564 1565 new_slave->delay = 0; 1566 new_slave->link_failure_count = 0; 1567 1568 bond_update_speed_duplex(new_slave); 1569 1570 new_slave->last_rx = jiffies - 1571 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1572 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1573 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1574 1575 if (bond->params.miimon && !bond->params.use_carrier) { 1576 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1577 1578 if ((link_reporting == -1) && !bond->params.arp_interval) { 1579 /* miimon is set but a bonded network driver 1580 * does not support ETHTOOL/MII and 1581 * arp_interval is not set. Note: if 1582 * use_carrier is enabled, we will never go 1583 * here (because netif_carrier is always 1584 * supported); thus, we don't need to change 1585 * the messages for netif_carrier. 1586 */ 1587 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", 1588 slave_dev->name); 1589 } else if (link_reporting == -1) { 1590 /* unable get link status using mii/ethtool */ 1591 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", 1592 slave_dev->name); 1593 } 1594 } 1595 1596 /* check for initial state */ 1597 new_slave->link = BOND_LINK_NOCHANGE; 1598 if (bond->params.miimon) { 1599 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1600 if (bond->params.updelay) { 1601 bond_set_slave_link_state(new_slave, 1602 BOND_LINK_BACK, 1603 BOND_SLAVE_NOTIFY_NOW); 1604 new_slave->delay = bond->params.updelay; 1605 } else { 1606 bond_set_slave_link_state(new_slave, 1607 BOND_LINK_UP, 1608 BOND_SLAVE_NOTIFY_NOW); 1609 } 1610 } else { 1611 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1612 BOND_SLAVE_NOTIFY_NOW); 1613 } 1614 } else if (bond->params.arp_interval) { 1615 bond_set_slave_link_state(new_slave, 1616 (netif_carrier_ok(slave_dev) ? 1617 BOND_LINK_UP : BOND_LINK_DOWN), 1618 BOND_SLAVE_NOTIFY_NOW); 1619 } else { 1620 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1621 BOND_SLAVE_NOTIFY_NOW); 1622 } 1623 1624 if (new_slave->link != BOND_LINK_DOWN) 1625 new_slave->last_link_up = jiffies; 1626 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n", 1627 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1628 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1629 1630 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1631 /* if there is a primary slave, remember it */ 1632 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1633 rcu_assign_pointer(bond->primary_slave, new_slave); 1634 bond->force_primary = true; 1635 } 1636 } 1637 1638 switch (BOND_MODE(bond)) { 1639 case BOND_MODE_ACTIVEBACKUP: 1640 bond_set_slave_inactive_flags(new_slave, 1641 BOND_SLAVE_NOTIFY_NOW); 1642 break; 1643 case BOND_MODE_8023AD: 1644 /* in 802.3ad mode, the internal mechanism 1645 * will activate the slaves in the selected 1646 * aggregator 1647 */ 1648 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1649 /* if this is the first slave */ 1650 if (!prev_slave) { 1651 SLAVE_AD_INFO(new_slave)->id = 1; 1652 /* Initialize AD with the number of times that the AD timer is called in 1 second 1653 * can be called only after the mac address of the bond is set 1654 */ 1655 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1656 } else { 1657 SLAVE_AD_INFO(new_slave)->id = 1658 SLAVE_AD_INFO(prev_slave)->id + 1; 1659 } 1660 1661 bond_3ad_bind_slave(new_slave); 1662 break; 1663 case BOND_MODE_TLB: 1664 case BOND_MODE_ALB: 1665 bond_set_active_slave(new_slave); 1666 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1667 break; 1668 default: 1669 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n"); 1670 1671 /* always active in trunk mode */ 1672 bond_set_active_slave(new_slave); 1673 1674 /* In trunking mode there is little meaning to curr_active_slave 1675 * anyway (it holds no special properties of the bond device), 1676 * so we can change it without calling change_active_interface() 1677 */ 1678 if (!rcu_access_pointer(bond->curr_active_slave) && 1679 new_slave->link == BOND_LINK_UP) 1680 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1681 1682 break; 1683 } /* switch(bond_mode) */ 1684 1685 #ifdef CONFIG_NET_POLL_CONTROLLER 1686 slave_dev->npinfo = bond->dev->npinfo; 1687 if (slave_dev->npinfo) { 1688 if (slave_enable_netpoll(new_slave)) { 1689 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1690 res = -EBUSY; 1691 goto err_detach; 1692 } 1693 } 1694 #endif 1695 1696 if (!(bond_dev->features & NETIF_F_LRO)) 1697 dev_disable_lro(slave_dev); 1698 1699 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1700 new_slave); 1701 if (res) { 1702 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res); 1703 goto err_detach; 1704 } 1705 1706 res = bond_master_upper_dev_link(bond, new_slave); 1707 if (res) { 1708 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1709 goto err_unregister; 1710 } 1711 1712 res = bond_sysfs_slave_add(new_slave); 1713 if (res) { 1714 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1715 goto err_upper_unlink; 1716 } 1717 1718 bond->slave_cnt++; 1719 bond_compute_features(bond); 1720 bond_set_carrier(bond); 1721 1722 if (bond_uses_primary(bond)) { 1723 block_netpoll_tx(); 1724 bond_select_active_slave(bond); 1725 unblock_netpoll_tx(); 1726 } 1727 1728 if (bond_mode_uses_xmit_hash(bond)) 1729 bond_update_slave_arr(bond, NULL); 1730 1731 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n", 1732 slave_dev->name, 1733 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1734 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1735 1736 /* enslave is successful */ 1737 bond_queue_slave_event(new_slave); 1738 return 0; 1739 1740 /* Undo stages on error */ 1741 err_upper_unlink: 1742 bond_upper_dev_unlink(bond, new_slave); 1743 1744 err_unregister: 1745 netdev_rx_handler_unregister(slave_dev); 1746 1747 err_detach: 1748 if (!bond_uses_primary(bond)) 1749 bond_hw_addr_flush(bond_dev, slave_dev); 1750 1751 vlan_vids_del_by_dev(slave_dev, bond_dev); 1752 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1753 RCU_INIT_POINTER(bond->primary_slave, NULL); 1754 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1755 block_netpoll_tx(); 1756 bond_change_active_slave(bond, NULL); 1757 bond_select_active_slave(bond); 1758 unblock_netpoll_tx(); 1759 } 1760 /* either primary_slave or curr_active_slave might've changed */ 1761 synchronize_rcu(); 1762 slave_disable_netpoll(new_slave); 1763 1764 err_close: 1765 slave_dev->priv_flags &= ~IFF_BONDING; 1766 dev_close(slave_dev); 1767 1768 err_restore_mac: 1769 slave_dev->flags &= ~IFF_SLAVE; 1770 if (!bond->params.fail_over_mac || 1771 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1772 /* XXX TODO - fom follow mode needs to change master's 1773 * MAC if this slave's MAC is in use by the bond, or at 1774 * least print a warning. 1775 */ 1776 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr); 1777 addr.sa_family = slave_dev->type; 1778 dev_set_mac_address(slave_dev, &addr); 1779 } 1780 1781 err_restore_mtu: 1782 dev_set_mtu(slave_dev, new_slave->original_mtu); 1783 1784 err_free: 1785 bond_free_slave(new_slave); 1786 1787 err_undo_flags: 1788 /* Enslave of first slave has failed and we need to fix master's mac */ 1789 if (!bond_has_slaves(bond)) { 1790 if (ether_addr_equal_64bits(bond_dev->dev_addr, 1791 slave_dev->dev_addr)) 1792 eth_hw_addr_random(bond_dev); 1793 if (bond_dev->type != ARPHRD_ETHER) { 1794 dev_close(bond_dev); 1795 ether_setup(bond_dev); 1796 bond_dev->flags |= IFF_MASTER; 1797 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1798 } 1799 } 1800 1801 return res; 1802 } 1803 1804 /* Try to release the slave device <slave> from the bond device <master> 1805 * It is legal to access curr_active_slave without a lock because all the function 1806 * is RTNL-locked. If "all" is true it means that the function is being called 1807 * while destroying a bond interface and all slaves are being released. 1808 * 1809 * The rules for slave state should be: 1810 * for Active/Backup: 1811 * Active stays on all backups go down 1812 * for Bonded connections: 1813 * The first up interface should be left on and all others downed. 1814 */ 1815 static int __bond_release_one(struct net_device *bond_dev, 1816 struct net_device *slave_dev, 1817 bool all) 1818 { 1819 struct bonding *bond = netdev_priv(bond_dev); 1820 struct slave *slave, *oldcurrent; 1821 struct sockaddr addr; 1822 int old_flags = bond_dev->flags; 1823 netdev_features_t old_features = bond_dev->features; 1824 1825 /* slave is not a slave or master is not master of this slave */ 1826 if (!(slave_dev->flags & IFF_SLAVE) || 1827 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1828 netdev_dbg(bond_dev, "cannot release %s\n", 1829 slave_dev->name); 1830 return -EINVAL; 1831 } 1832 1833 block_netpoll_tx(); 1834 1835 slave = bond_get_slave_by_dev(bond, slave_dev); 1836 if (!slave) { 1837 /* not a slave of this bond */ 1838 netdev_info(bond_dev, "%s not enslaved\n", 1839 slave_dev->name); 1840 unblock_netpoll_tx(); 1841 return -EINVAL; 1842 } 1843 1844 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 1845 1846 bond_sysfs_slave_del(slave); 1847 1848 /* recompute stats just before removing the slave */ 1849 bond_get_stats(bond->dev, &bond->bond_stats); 1850 1851 bond_upper_dev_unlink(bond, slave); 1852 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1853 * for this slave anymore. 1854 */ 1855 netdev_rx_handler_unregister(slave_dev); 1856 1857 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1858 bond_3ad_unbind_slave(slave); 1859 1860 if (bond_mode_uses_xmit_hash(bond)) 1861 bond_update_slave_arr(bond, slave); 1862 1863 netdev_info(bond_dev, "Releasing %s interface %s\n", 1864 bond_is_active_slave(slave) ? "active" : "backup", 1865 slave_dev->name); 1866 1867 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1868 1869 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1870 1871 if (!all && (!bond->params.fail_over_mac || 1872 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1873 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1874 bond_has_slaves(bond)) 1875 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", 1876 slave_dev->name, slave->perm_hwaddr, 1877 bond_dev->name, slave_dev->name); 1878 } 1879 1880 if (rtnl_dereference(bond->primary_slave) == slave) 1881 RCU_INIT_POINTER(bond->primary_slave, NULL); 1882 1883 if (oldcurrent == slave) 1884 bond_change_active_slave(bond, NULL); 1885 1886 if (bond_is_lb(bond)) { 1887 /* Must be called only after the slave has been 1888 * detached from the list and the curr_active_slave 1889 * has been cleared (if our_slave == old_current), 1890 * but before a new active slave is selected. 1891 */ 1892 bond_alb_deinit_slave(bond, slave); 1893 } 1894 1895 if (all) { 1896 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1897 } else if (oldcurrent == slave) { 1898 /* Note that we hold RTNL over this sequence, so there 1899 * is no concern that another slave add/remove event 1900 * will interfere. 1901 */ 1902 bond_select_active_slave(bond); 1903 } 1904 1905 if (!bond_has_slaves(bond)) { 1906 bond_set_carrier(bond); 1907 eth_hw_addr_random(bond_dev); 1908 } 1909 1910 unblock_netpoll_tx(); 1911 synchronize_rcu(); 1912 bond->slave_cnt--; 1913 1914 if (!bond_has_slaves(bond)) { 1915 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 1916 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 1917 } 1918 1919 bond_compute_features(bond); 1920 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 1921 (old_features & NETIF_F_VLAN_CHALLENGED)) 1922 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n", 1923 slave_dev->name, bond_dev->name); 1924 1925 vlan_vids_del_by_dev(slave_dev, bond_dev); 1926 1927 /* If the mode uses primary, then this case was handled above by 1928 * bond_change_active_slave(..., NULL) 1929 */ 1930 if (!bond_uses_primary(bond)) { 1931 /* unset promiscuity level from slave 1932 * NOTE: The NETDEV_CHANGEADDR call above may change the value 1933 * of the IFF_PROMISC flag in the bond_dev, but we need the 1934 * value of that flag before that change, as that was the value 1935 * when this slave was attached, so we cache at the start of the 1936 * function and use it here. Same goes for ALLMULTI below 1937 */ 1938 if (old_flags & IFF_PROMISC) 1939 dev_set_promiscuity(slave_dev, -1); 1940 1941 /* unset allmulti level from slave */ 1942 if (old_flags & IFF_ALLMULTI) 1943 dev_set_allmulti(slave_dev, -1); 1944 1945 bond_hw_addr_flush(bond_dev, slave_dev); 1946 } 1947 1948 slave_disable_netpoll(slave); 1949 1950 /* close slave before restoring its mac address */ 1951 dev_close(slave_dev); 1952 1953 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 1954 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1955 /* restore original ("permanent") mac address */ 1956 ether_addr_copy(addr.sa_data, slave->perm_hwaddr); 1957 addr.sa_family = slave_dev->type; 1958 dev_set_mac_address(slave_dev, &addr); 1959 } 1960 1961 dev_set_mtu(slave_dev, slave->original_mtu); 1962 1963 slave_dev->priv_flags &= ~IFF_BONDING; 1964 1965 bond_free_slave(slave); 1966 1967 return 0; 1968 } 1969 1970 /* A wrapper used because of ndo_del_link */ 1971 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 1972 { 1973 return __bond_release_one(bond_dev, slave_dev, false); 1974 } 1975 1976 /* First release a slave and then destroy the bond if no more slaves are left. 1977 * Must be under rtnl_lock when this function is called. 1978 */ 1979 static int bond_release_and_destroy(struct net_device *bond_dev, 1980 struct net_device *slave_dev) 1981 { 1982 struct bonding *bond = netdev_priv(bond_dev); 1983 int ret; 1984 1985 ret = bond_release(bond_dev, slave_dev); 1986 if (ret == 0 && !bond_has_slaves(bond)) { 1987 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 1988 netdev_info(bond_dev, "Destroying bond %s\n", 1989 bond_dev->name); 1990 bond_remove_proc_entry(bond); 1991 unregister_netdevice(bond_dev); 1992 } 1993 return ret; 1994 } 1995 1996 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 1997 { 1998 struct bonding *bond = netdev_priv(bond_dev); 1999 bond_fill_ifbond(bond, info); 2000 } 2001 2002 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2003 { 2004 struct bonding *bond = netdev_priv(bond_dev); 2005 struct list_head *iter; 2006 int i = 0, res = -ENODEV; 2007 struct slave *slave; 2008 2009 bond_for_each_slave(bond, slave, iter) { 2010 if (i++ == (int)info->slave_id) { 2011 res = 0; 2012 bond_fill_ifslave(slave, info); 2013 break; 2014 } 2015 } 2016 2017 return res; 2018 } 2019 2020 /*-------------------------------- Monitoring -------------------------------*/ 2021 2022 /* called with rcu_read_lock() */ 2023 static int bond_miimon_inspect(struct bonding *bond) 2024 { 2025 int link_state, commit = 0; 2026 struct list_head *iter; 2027 struct slave *slave; 2028 bool ignore_updelay; 2029 2030 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2031 2032 bond_for_each_slave_rcu(bond, slave, iter) { 2033 slave->new_link = BOND_LINK_NOCHANGE; 2034 2035 link_state = bond_check_dev_link(bond, slave->dev, 0); 2036 2037 switch (slave->link) { 2038 case BOND_LINK_UP: 2039 if (link_state) 2040 continue; 2041 2042 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 2043 BOND_SLAVE_NOTIFY_LATER); 2044 slave->delay = bond->params.downdelay; 2045 if (slave->delay) { 2046 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n", 2047 (BOND_MODE(bond) == 2048 BOND_MODE_ACTIVEBACKUP) ? 2049 (bond_is_active_slave(slave) ? 2050 "active " : "backup ") : "", 2051 slave->dev->name, 2052 bond->params.downdelay * bond->params.miimon); 2053 } 2054 /*FALLTHRU*/ 2055 case BOND_LINK_FAIL: 2056 if (link_state) { 2057 /* recovered before downdelay expired */ 2058 bond_set_slave_link_state(slave, BOND_LINK_UP, 2059 BOND_SLAVE_NOTIFY_LATER); 2060 slave->last_link_up = jiffies; 2061 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n", 2062 (bond->params.downdelay - slave->delay) * 2063 bond->params.miimon, 2064 slave->dev->name); 2065 continue; 2066 } 2067 2068 if (slave->delay <= 0) { 2069 slave->new_link = BOND_LINK_DOWN; 2070 commit++; 2071 continue; 2072 } 2073 2074 slave->delay--; 2075 break; 2076 2077 case BOND_LINK_DOWN: 2078 if (!link_state) 2079 continue; 2080 2081 bond_set_slave_link_state(slave, BOND_LINK_BACK, 2082 BOND_SLAVE_NOTIFY_LATER); 2083 slave->delay = bond->params.updelay; 2084 2085 if (slave->delay) { 2086 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n", 2087 slave->dev->name, 2088 ignore_updelay ? 0 : 2089 bond->params.updelay * 2090 bond->params.miimon); 2091 } 2092 /*FALLTHRU*/ 2093 case BOND_LINK_BACK: 2094 if (!link_state) { 2095 bond_set_slave_link_state(slave, 2096 BOND_LINK_DOWN, 2097 BOND_SLAVE_NOTIFY_LATER); 2098 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n", 2099 (bond->params.updelay - slave->delay) * 2100 bond->params.miimon, 2101 slave->dev->name); 2102 2103 continue; 2104 } 2105 2106 if (ignore_updelay) 2107 slave->delay = 0; 2108 2109 if (slave->delay <= 0) { 2110 slave->new_link = BOND_LINK_UP; 2111 commit++; 2112 ignore_updelay = false; 2113 continue; 2114 } 2115 2116 slave->delay--; 2117 break; 2118 } 2119 } 2120 2121 return commit; 2122 } 2123 2124 static void bond_miimon_commit(struct bonding *bond) 2125 { 2126 struct list_head *iter; 2127 struct slave *slave, *primary; 2128 2129 bond_for_each_slave(bond, slave, iter) { 2130 switch (slave->new_link) { 2131 case BOND_LINK_NOCHANGE: 2132 continue; 2133 2134 case BOND_LINK_UP: 2135 bond_update_speed_duplex(slave); 2136 bond_set_slave_link_state(slave, BOND_LINK_UP, 2137 BOND_SLAVE_NOTIFY_NOW); 2138 slave->last_link_up = jiffies; 2139 2140 primary = rtnl_dereference(bond->primary_slave); 2141 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2142 /* prevent it from being the active one */ 2143 bond_set_backup_slave(slave); 2144 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2145 /* make it immediately active */ 2146 bond_set_active_slave(slave); 2147 } else if (slave != primary) { 2148 /* prevent it from being the active one */ 2149 bond_set_backup_slave(slave); 2150 } 2151 2152 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n", 2153 slave->dev->name, 2154 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2155 slave->duplex ? "full" : "half"); 2156 2157 /* notify ad that the link status has changed */ 2158 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2159 bond_3ad_handle_link_change(slave, BOND_LINK_UP); 2160 2161 if (bond_is_lb(bond)) 2162 bond_alb_handle_link_change(bond, slave, 2163 BOND_LINK_UP); 2164 2165 if (BOND_MODE(bond) == BOND_MODE_XOR) 2166 bond_update_slave_arr(bond, NULL); 2167 2168 if (!bond->curr_active_slave || slave == primary) 2169 goto do_failover; 2170 2171 continue; 2172 2173 case BOND_LINK_DOWN: 2174 if (slave->link_failure_count < UINT_MAX) 2175 slave->link_failure_count++; 2176 2177 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2178 BOND_SLAVE_NOTIFY_NOW); 2179 2180 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2181 BOND_MODE(bond) == BOND_MODE_8023AD) 2182 bond_set_slave_inactive_flags(slave, 2183 BOND_SLAVE_NOTIFY_NOW); 2184 2185 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2186 slave->dev->name); 2187 2188 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2189 bond_3ad_handle_link_change(slave, 2190 BOND_LINK_DOWN); 2191 2192 if (bond_is_lb(bond)) 2193 bond_alb_handle_link_change(bond, slave, 2194 BOND_LINK_DOWN); 2195 2196 if (BOND_MODE(bond) == BOND_MODE_XOR) 2197 bond_update_slave_arr(bond, NULL); 2198 2199 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2200 goto do_failover; 2201 2202 continue; 2203 2204 default: 2205 netdev_err(bond->dev, "invalid new link %d on slave %s\n", 2206 slave->new_link, slave->dev->name); 2207 slave->new_link = BOND_LINK_NOCHANGE; 2208 2209 continue; 2210 } 2211 2212 do_failover: 2213 block_netpoll_tx(); 2214 bond_select_active_slave(bond); 2215 unblock_netpoll_tx(); 2216 } 2217 2218 bond_set_carrier(bond); 2219 } 2220 2221 /* bond_mii_monitor 2222 * 2223 * Really a wrapper that splits the mii monitor into two phases: an 2224 * inspection, then (if inspection indicates something needs to be done) 2225 * an acquisition of appropriate locks followed by a commit phase to 2226 * implement whatever link state changes are indicated. 2227 */ 2228 static void bond_mii_monitor(struct work_struct *work) 2229 { 2230 struct bonding *bond = container_of(work, struct bonding, 2231 mii_work.work); 2232 bool should_notify_peers = false; 2233 unsigned long delay; 2234 2235 delay = msecs_to_jiffies(bond->params.miimon); 2236 2237 if (!bond_has_slaves(bond)) 2238 goto re_arm; 2239 2240 rcu_read_lock(); 2241 2242 should_notify_peers = bond_should_notify_peers(bond); 2243 2244 if (bond_miimon_inspect(bond)) { 2245 rcu_read_unlock(); 2246 2247 /* Race avoidance with bond_close cancel of workqueue */ 2248 if (!rtnl_trylock()) { 2249 delay = 1; 2250 should_notify_peers = false; 2251 goto re_arm; 2252 } 2253 2254 bond_miimon_commit(bond); 2255 2256 rtnl_unlock(); /* might sleep, hold no other locks */ 2257 } else 2258 rcu_read_unlock(); 2259 2260 re_arm: 2261 if (bond->params.miimon) 2262 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2263 2264 if (should_notify_peers) { 2265 if (!rtnl_trylock()) 2266 return; 2267 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2268 rtnl_unlock(); 2269 } 2270 } 2271 2272 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2273 { 2274 __be32 ip = *((__be32 *)data); 2275 2276 return ip == bond_confirm_addr(upper, 0, ip); 2277 } 2278 2279 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2280 { 2281 bool ret = false; 2282 2283 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2284 return true; 2285 2286 rcu_read_lock(); 2287 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2288 ret = true; 2289 rcu_read_unlock(); 2290 2291 return ret; 2292 } 2293 2294 /* We go to the (large) trouble of VLAN tagging ARP frames because 2295 * switches in VLAN mode (especially if ports are configured as 2296 * "native" to a VLAN) might not pass non-tagged frames. 2297 */ 2298 static void bond_arp_send(struct net_device *slave_dev, int arp_op, 2299 __be32 dest_ip, __be32 src_ip, 2300 struct bond_vlan_tag *tags) 2301 { 2302 struct sk_buff *skb; 2303 struct bond_vlan_tag *outer_tag = tags; 2304 2305 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n", 2306 arp_op, slave_dev->name, &dest_ip, &src_ip); 2307 2308 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2309 NULL, slave_dev->dev_addr, NULL); 2310 2311 if (!skb) { 2312 net_err_ratelimited("ARP packet allocation failed\n"); 2313 return; 2314 } 2315 2316 if (!tags || tags->vlan_proto == VLAN_N_VID) 2317 goto xmit; 2318 2319 tags++; 2320 2321 /* Go through all the tags backwards and add them to the packet */ 2322 while (tags->vlan_proto != VLAN_N_VID) { 2323 if (!tags->vlan_id) { 2324 tags++; 2325 continue; 2326 } 2327 2328 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n", 2329 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2330 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2331 tags->vlan_id); 2332 if (!skb) { 2333 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2334 return; 2335 } 2336 2337 tags++; 2338 } 2339 /* Set the outer tag */ 2340 if (outer_tag->vlan_id) { 2341 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n", 2342 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2343 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2344 outer_tag->vlan_id); 2345 } 2346 2347 xmit: 2348 arp_xmit(skb); 2349 } 2350 2351 /* Validate the device path between the @start_dev and the @end_dev. 2352 * The path is valid if the @end_dev is reachable through device 2353 * stacking. 2354 * When the path is validated, collect any vlan information in the 2355 * path. 2356 */ 2357 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2358 struct net_device *end_dev, 2359 int level) 2360 { 2361 struct bond_vlan_tag *tags; 2362 struct net_device *upper; 2363 struct list_head *iter; 2364 2365 if (start_dev == end_dev) { 2366 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC); 2367 if (!tags) 2368 return ERR_PTR(-ENOMEM); 2369 tags[level].vlan_proto = VLAN_N_VID; 2370 return tags; 2371 } 2372 2373 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2374 tags = bond_verify_device_path(upper, end_dev, level + 1); 2375 if (IS_ERR_OR_NULL(tags)) { 2376 if (IS_ERR(tags)) 2377 return tags; 2378 continue; 2379 } 2380 if (is_vlan_dev(upper)) { 2381 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2382 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2383 } 2384 2385 return tags; 2386 } 2387 2388 return NULL; 2389 } 2390 2391 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2392 { 2393 struct rtable *rt; 2394 struct bond_vlan_tag *tags; 2395 __be32 *targets = bond->params.arp_targets, addr; 2396 int i; 2397 2398 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2399 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]); 2400 tags = NULL; 2401 2402 /* Find out through which dev should the packet go */ 2403 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2404 RTO_ONLINK, 0); 2405 if (IS_ERR(rt)) { 2406 /* there's no route to target - try to send arp 2407 * probe to generate any traffic (arp_validate=0) 2408 */ 2409 if (bond->params.arp_validate) 2410 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2411 bond->dev->name, 2412 &targets[i]); 2413 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2414 0, tags); 2415 continue; 2416 } 2417 2418 /* bond device itself */ 2419 if (rt->dst.dev == bond->dev) 2420 goto found; 2421 2422 rcu_read_lock(); 2423 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2424 rcu_read_unlock(); 2425 2426 if (!IS_ERR_OR_NULL(tags)) 2427 goto found; 2428 2429 /* Not our device - skip */ 2430 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2431 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2432 2433 ip_rt_put(rt); 2434 continue; 2435 2436 found: 2437 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2438 ip_rt_put(rt); 2439 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2440 addr, tags); 2441 kfree(tags); 2442 } 2443 } 2444 2445 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2446 { 2447 int i; 2448 2449 if (!sip || !bond_has_this_ip(bond, tip)) { 2450 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n", 2451 &sip, &tip); 2452 return; 2453 } 2454 2455 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2456 if (i == -1) { 2457 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n", 2458 &sip); 2459 return; 2460 } 2461 slave->last_rx = jiffies; 2462 slave->target_last_arp_rx[i] = jiffies; 2463 } 2464 2465 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2466 struct slave *slave) 2467 { 2468 struct arphdr *arp = (struct arphdr *)skb->data; 2469 struct slave *curr_active_slave, *curr_arp_slave; 2470 unsigned char *arp_ptr; 2471 __be32 sip, tip; 2472 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2473 2474 if (!slave_do_arp_validate(bond, slave)) { 2475 if ((slave_do_arp_validate_only(bond) && is_arp) || 2476 !slave_do_arp_validate_only(bond)) 2477 slave->last_rx = jiffies; 2478 return RX_HANDLER_ANOTHER; 2479 } else if (!is_arp) { 2480 return RX_HANDLER_ANOTHER; 2481 } 2482 2483 alen = arp_hdr_len(bond->dev); 2484 2485 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n", 2486 skb->dev->name); 2487 2488 if (alen > skb_headlen(skb)) { 2489 arp = kmalloc(alen, GFP_ATOMIC); 2490 if (!arp) 2491 goto out_unlock; 2492 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2493 goto out_unlock; 2494 } 2495 2496 if (arp->ar_hln != bond->dev->addr_len || 2497 skb->pkt_type == PACKET_OTHERHOST || 2498 skb->pkt_type == PACKET_LOOPBACK || 2499 arp->ar_hrd != htons(ARPHRD_ETHER) || 2500 arp->ar_pro != htons(ETH_P_IP) || 2501 arp->ar_pln != 4) 2502 goto out_unlock; 2503 2504 arp_ptr = (unsigned char *)(arp + 1); 2505 arp_ptr += bond->dev->addr_len; 2506 memcpy(&sip, arp_ptr, 4); 2507 arp_ptr += 4 + bond->dev->addr_len; 2508 memcpy(&tip, arp_ptr, 4); 2509 2510 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2511 slave->dev->name, bond_slave_state(slave), 2512 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2513 &sip, &tip); 2514 2515 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2516 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2517 2518 /* We 'trust' the received ARP enough to validate it if: 2519 * 2520 * (a) the slave receiving the ARP is active (which includes the 2521 * current ARP slave, if any), or 2522 * 2523 * (b) the receiving slave isn't active, but there is a currently 2524 * active slave and it received valid arp reply(s) after it became 2525 * the currently active slave, or 2526 * 2527 * (c) there is an ARP slave that sent an ARP during the prior ARP 2528 * interval, and we receive an ARP reply on any slave. We accept 2529 * these because switch FDB update delays may deliver the ARP 2530 * reply to a slave other than the sender of the ARP request. 2531 * 2532 * Note: for (b), backup slaves are receiving the broadcast ARP 2533 * request, not a reply. This request passes from the sending 2534 * slave through the L2 switch(es) to the receiving slave. Since 2535 * this is checking the request, sip/tip are swapped for 2536 * validation. 2537 * 2538 * This is done to avoid endless looping when we can't reach the 2539 * arp_ip_target and fool ourselves with our own arp requests. 2540 */ 2541 if (bond_is_active_slave(slave)) 2542 bond_validate_arp(bond, slave, sip, tip); 2543 else if (curr_active_slave && 2544 time_after(slave_last_rx(bond, curr_active_slave), 2545 curr_active_slave->last_link_up)) 2546 bond_validate_arp(bond, slave, tip, sip); 2547 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2548 bond_time_in_interval(bond, 2549 dev_trans_start(curr_arp_slave->dev), 1)) 2550 bond_validate_arp(bond, slave, sip, tip); 2551 2552 out_unlock: 2553 if (arp != (struct arphdr *)skb->data) 2554 kfree(arp); 2555 return RX_HANDLER_ANOTHER; 2556 } 2557 2558 /* function to verify if we're in the arp_interval timeslice, returns true if 2559 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2560 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2561 */ 2562 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2563 int mod) 2564 { 2565 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2566 2567 return time_in_range(jiffies, 2568 last_act - delta_in_ticks, 2569 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2570 } 2571 2572 /* This function is called regularly to monitor each slave's link 2573 * ensuring that traffic is being sent and received when arp monitoring 2574 * is used in load-balancing mode. if the adapter has been dormant, then an 2575 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2576 * arp monitoring in active backup mode. 2577 */ 2578 static void bond_loadbalance_arp_mon(struct work_struct *work) 2579 { 2580 struct bonding *bond = container_of(work, struct bonding, 2581 arp_work.work); 2582 struct slave *slave, *oldcurrent; 2583 struct list_head *iter; 2584 int do_failover = 0, slave_state_changed = 0; 2585 2586 if (!bond_has_slaves(bond)) 2587 goto re_arm; 2588 2589 rcu_read_lock(); 2590 2591 oldcurrent = rcu_dereference(bond->curr_active_slave); 2592 /* see if any of the previous devices are up now (i.e. they have 2593 * xmt and rcv traffic). the curr_active_slave does not come into 2594 * the picture unless it is null. also, slave->last_link_up is not 2595 * needed here because we send an arp on each slave and give a slave 2596 * as long as it needs to get the tx/rx within the delta. 2597 * TODO: what about up/down delay in arp mode? it wasn't here before 2598 * so it can wait 2599 */ 2600 bond_for_each_slave_rcu(bond, slave, iter) { 2601 unsigned long trans_start = dev_trans_start(slave->dev); 2602 2603 if (slave->link != BOND_LINK_UP) { 2604 if (bond_time_in_interval(bond, trans_start, 1) && 2605 bond_time_in_interval(bond, slave->last_rx, 1)) { 2606 2607 slave->link = BOND_LINK_UP; 2608 slave_state_changed = 1; 2609 2610 /* primary_slave has no meaning in round-robin 2611 * mode. the window of a slave being up and 2612 * curr_active_slave being null after enslaving 2613 * is closed. 2614 */ 2615 if (!oldcurrent) { 2616 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2617 slave->dev->name); 2618 do_failover = 1; 2619 } else { 2620 netdev_info(bond->dev, "interface %s is now up\n", 2621 slave->dev->name); 2622 } 2623 } 2624 } else { 2625 /* slave->link == BOND_LINK_UP */ 2626 2627 /* not all switches will respond to an arp request 2628 * when the source ip is 0, so don't take the link down 2629 * if we don't know our ip yet 2630 */ 2631 if (!bond_time_in_interval(bond, trans_start, 2) || 2632 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2633 2634 slave->link = BOND_LINK_DOWN; 2635 slave_state_changed = 1; 2636 2637 if (slave->link_failure_count < UINT_MAX) 2638 slave->link_failure_count++; 2639 2640 netdev_info(bond->dev, "interface %s is now down\n", 2641 slave->dev->name); 2642 2643 if (slave == oldcurrent) 2644 do_failover = 1; 2645 } 2646 } 2647 2648 /* note: if switch is in round-robin mode, all links 2649 * must tx arp to ensure all links rx an arp - otherwise 2650 * links may oscillate or not come up at all; if switch is 2651 * in something like xor mode, there is nothing we can 2652 * do - all replies will be rx'ed on same link causing slaves 2653 * to be unstable during low/no traffic periods 2654 */ 2655 if (bond_slave_is_up(slave)) 2656 bond_arp_send_all(bond, slave); 2657 } 2658 2659 rcu_read_unlock(); 2660 2661 if (do_failover || slave_state_changed) { 2662 if (!rtnl_trylock()) 2663 goto re_arm; 2664 2665 if (slave_state_changed) { 2666 bond_slave_state_change(bond); 2667 if (BOND_MODE(bond) == BOND_MODE_XOR) 2668 bond_update_slave_arr(bond, NULL); 2669 } 2670 if (do_failover) { 2671 block_netpoll_tx(); 2672 bond_select_active_slave(bond); 2673 unblock_netpoll_tx(); 2674 } 2675 rtnl_unlock(); 2676 } 2677 2678 re_arm: 2679 if (bond->params.arp_interval) 2680 queue_delayed_work(bond->wq, &bond->arp_work, 2681 msecs_to_jiffies(bond->params.arp_interval)); 2682 } 2683 2684 /* Called to inspect slaves for active-backup mode ARP monitor link state 2685 * changes. Sets new_link in slaves to specify what action should take 2686 * place for the slave. Returns 0 if no changes are found, >0 if changes 2687 * to link states must be committed. 2688 * 2689 * Called with rcu_read_lock held. 2690 */ 2691 static int bond_ab_arp_inspect(struct bonding *bond) 2692 { 2693 unsigned long trans_start, last_rx; 2694 struct list_head *iter; 2695 struct slave *slave; 2696 int commit = 0; 2697 2698 bond_for_each_slave_rcu(bond, slave, iter) { 2699 slave->new_link = BOND_LINK_NOCHANGE; 2700 last_rx = slave_last_rx(bond, slave); 2701 2702 if (slave->link != BOND_LINK_UP) { 2703 if (bond_time_in_interval(bond, last_rx, 1)) { 2704 slave->new_link = BOND_LINK_UP; 2705 commit++; 2706 } 2707 continue; 2708 } 2709 2710 /* Give slaves 2*delta after being enslaved or made 2711 * active. This avoids bouncing, as the last receive 2712 * times need a full ARP monitor cycle to be updated. 2713 */ 2714 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2715 continue; 2716 2717 /* Backup slave is down if: 2718 * - No current_arp_slave AND 2719 * - more than 3*delta since last receive AND 2720 * - the bond has an IP address 2721 * 2722 * Note: a non-null current_arp_slave indicates 2723 * the curr_active_slave went down and we are 2724 * searching for a new one; under this condition 2725 * we only take the curr_active_slave down - this 2726 * gives each slave a chance to tx/rx traffic 2727 * before being taken out 2728 */ 2729 if (!bond_is_active_slave(slave) && 2730 !rcu_access_pointer(bond->current_arp_slave) && 2731 !bond_time_in_interval(bond, last_rx, 3)) { 2732 slave->new_link = BOND_LINK_DOWN; 2733 commit++; 2734 } 2735 2736 /* Active slave is down if: 2737 * - more than 2*delta since transmitting OR 2738 * - (more than 2*delta since receive AND 2739 * the bond has an IP address) 2740 */ 2741 trans_start = dev_trans_start(slave->dev); 2742 if (bond_is_active_slave(slave) && 2743 (!bond_time_in_interval(bond, trans_start, 2) || 2744 !bond_time_in_interval(bond, last_rx, 2))) { 2745 slave->new_link = BOND_LINK_DOWN; 2746 commit++; 2747 } 2748 } 2749 2750 return commit; 2751 } 2752 2753 /* Called to commit link state changes noted by inspection step of 2754 * active-backup mode ARP monitor. 2755 * 2756 * Called with RTNL hold. 2757 */ 2758 static void bond_ab_arp_commit(struct bonding *bond) 2759 { 2760 unsigned long trans_start; 2761 struct list_head *iter; 2762 struct slave *slave; 2763 2764 bond_for_each_slave(bond, slave, iter) { 2765 switch (slave->new_link) { 2766 case BOND_LINK_NOCHANGE: 2767 continue; 2768 2769 case BOND_LINK_UP: 2770 trans_start = dev_trans_start(slave->dev); 2771 if (rtnl_dereference(bond->curr_active_slave) != slave || 2772 (!rtnl_dereference(bond->curr_active_slave) && 2773 bond_time_in_interval(bond, trans_start, 1))) { 2774 struct slave *current_arp_slave; 2775 2776 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2777 bond_set_slave_link_state(slave, BOND_LINK_UP, 2778 BOND_SLAVE_NOTIFY_NOW); 2779 if (current_arp_slave) { 2780 bond_set_slave_inactive_flags( 2781 current_arp_slave, 2782 BOND_SLAVE_NOTIFY_NOW); 2783 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2784 } 2785 2786 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2787 slave->dev->name); 2788 2789 if (!rtnl_dereference(bond->curr_active_slave) || 2790 slave == rtnl_dereference(bond->primary_slave)) 2791 goto do_failover; 2792 2793 } 2794 2795 continue; 2796 2797 case BOND_LINK_DOWN: 2798 if (slave->link_failure_count < UINT_MAX) 2799 slave->link_failure_count++; 2800 2801 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2802 BOND_SLAVE_NOTIFY_NOW); 2803 bond_set_slave_inactive_flags(slave, 2804 BOND_SLAVE_NOTIFY_NOW); 2805 2806 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2807 slave->dev->name); 2808 2809 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2810 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2811 goto do_failover; 2812 } 2813 2814 continue; 2815 2816 default: 2817 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n", 2818 slave->new_link, slave->dev->name); 2819 continue; 2820 } 2821 2822 do_failover: 2823 block_netpoll_tx(); 2824 bond_select_active_slave(bond); 2825 unblock_netpoll_tx(); 2826 } 2827 2828 bond_set_carrier(bond); 2829 } 2830 2831 /* Send ARP probes for active-backup mode ARP monitor. 2832 * 2833 * Called with rcu_read_lock held. 2834 */ 2835 static bool bond_ab_arp_probe(struct bonding *bond) 2836 { 2837 struct slave *slave, *before = NULL, *new_slave = NULL, 2838 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2839 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2840 struct list_head *iter; 2841 bool found = false; 2842 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2843 2844 if (curr_arp_slave && curr_active_slave) 2845 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2846 curr_arp_slave->dev->name, 2847 curr_active_slave->dev->name); 2848 2849 if (curr_active_slave) { 2850 bond_arp_send_all(bond, curr_active_slave); 2851 return should_notify_rtnl; 2852 } 2853 2854 /* if we don't have a curr_active_slave, search for the next available 2855 * backup slave from the current_arp_slave and make it the candidate 2856 * for becoming the curr_active_slave 2857 */ 2858 2859 if (!curr_arp_slave) { 2860 curr_arp_slave = bond_first_slave_rcu(bond); 2861 if (!curr_arp_slave) 2862 return should_notify_rtnl; 2863 } 2864 2865 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2866 2867 bond_for_each_slave_rcu(bond, slave, iter) { 2868 if (!found && !before && bond_slave_is_up(slave)) 2869 before = slave; 2870 2871 if (found && !new_slave && bond_slave_is_up(slave)) 2872 new_slave = slave; 2873 /* if the link state is up at this point, we 2874 * mark it down - this can happen if we have 2875 * simultaneous link failures and 2876 * reselect_active_interface doesn't make this 2877 * one the current slave so it is still marked 2878 * up when it is actually down 2879 */ 2880 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 2881 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2882 BOND_SLAVE_NOTIFY_LATER); 2883 if (slave->link_failure_count < UINT_MAX) 2884 slave->link_failure_count++; 2885 2886 bond_set_slave_inactive_flags(slave, 2887 BOND_SLAVE_NOTIFY_LATER); 2888 2889 netdev_info(bond->dev, "backup interface %s is now down\n", 2890 slave->dev->name); 2891 } 2892 if (slave == curr_arp_slave) 2893 found = true; 2894 } 2895 2896 if (!new_slave && before) 2897 new_slave = before; 2898 2899 if (!new_slave) 2900 goto check_state; 2901 2902 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 2903 BOND_SLAVE_NOTIFY_LATER); 2904 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 2905 bond_arp_send_all(bond, new_slave); 2906 new_slave->last_link_up = jiffies; 2907 rcu_assign_pointer(bond->current_arp_slave, new_slave); 2908 2909 check_state: 2910 bond_for_each_slave_rcu(bond, slave, iter) { 2911 if (slave->should_notify || slave->should_notify_link) { 2912 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 2913 break; 2914 } 2915 } 2916 return should_notify_rtnl; 2917 } 2918 2919 static void bond_activebackup_arp_mon(struct work_struct *work) 2920 { 2921 struct bonding *bond = container_of(work, struct bonding, 2922 arp_work.work); 2923 bool should_notify_peers = false; 2924 bool should_notify_rtnl = false; 2925 int delta_in_ticks; 2926 2927 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2928 2929 if (!bond_has_slaves(bond)) 2930 goto re_arm; 2931 2932 rcu_read_lock(); 2933 2934 should_notify_peers = bond_should_notify_peers(bond); 2935 2936 if (bond_ab_arp_inspect(bond)) { 2937 rcu_read_unlock(); 2938 2939 /* Race avoidance with bond_close flush of workqueue */ 2940 if (!rtnl_trylock()) { 2941 delta_in_ticks = 1; 2942 should_notify_peers = false; 2943 goto re_arm; 2944 } 2945 2946 bond_ab_arp_commit(bond); 2947 2948 rtnl_unlock(); 2949 rcu_read_lock(); 2950 } 2951 2952 should_notify_rtnl = bond_ab_arp_probe(bond); 2953 rcu_read_unlock(); 2954 2955 re_arm: 2956 if (bond->params.arp_interval) 2957 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 2958 2959 if (should_notify_peers || should_notify_rtnl) { 2960 if (!rtnl_trylock()) 2961 return; 2962 2963 if (should_notify_peers) 2964 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 2965 bond->dev); 2966 if (should_notify_rtnl) { 2967 bond_slave_state_notify(bond); 2968 bond_slave_link_notify(bond); 2969 } 2970 2971 rtnl_unlock(); 2972 } 2973 } 2974 2975 /*-------------------------- netdev event handling --------------------------*/ 2976 2977 /* Change device name */ 2978 static int bond_event_changename(struct bonding *bond) 2979 { 2980 bond_remove_proc_entry(bond); 2981 bond_create_proc_entry(bond); 2982 2983 bond_debug_reregister(bond); 2984 2985 return NOTIFY_DONE; 2986 } 2987 2988 static int bond_master_netdev_event(unsigned long event, 2989 struct net_device *bond_dev) 2990 { 2991 struct bonding *event_bond = netdev_priv(bond_dev); 2992 2993 switch (event) { 2994 case NETDEV_CHANGENAME: 2995 return bond_event_changename(event_bond); 2996 case NETDEV_UNREGISTER: 2997 bond_remove_proc_entry(event_bond); 2998 break; 2999 case NETDEV_REGISTER: 3000 bond_create_proc_entry(event_bond); 3001 break; 3002 case NETDEV_NOTIFY_PEERS: 3003 if (event_bond->send_peer_notif) 3004 event_bond->send_peer_notif--; 3005 break; 3006 default: 3007 break; 3008 } 3009 3010 return NOTIFY_DONE; 3011 } 3012 3013 static int bond_slave_netdev_event(unsigned long event, 3014 struct net_device *slave_dev) 3015 { 3016 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3017 struct bonding *bond; 3018 struct net_device *bond_dev; 3019 3020 /* A netdev event can be generated while enslaving a device 3021 * before netdev_rx_handler_register is called in which case 3022 * slave will be NULL 3023 */ 3024 if (!slave) 3025 return NOTIFY_DONE; 3026 bond_dev = slave->bond->dev; 3027 bond = slave->bond; 3028 primary = rtnl_dereference(bond->primary_slave); 3029 3030 switch (event) { 3031 case NETDEV_UNREGISTER: 3032 if (bond_dev->type != ARPHRD_ETHER) 3033 bond_release_and_destroy(bond_dev, slave_dev); 3034 else 3035 bond_release(bond_dev, slave_dev); 3036 break; 3037 case NETDEV_UP: 3038 case NETDEV_CHANGE: 3039 bond_update_speed_duplex(slave); 3040 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3041 bond_3ad_adapter_speed_duplex_changed(slave); 3042 /* Fallthrough */ 3043 case NETDEV_DOWN: 3044 /* Refresh slave-array if applicable! 3045 * If the setup does not use miimon or arpmon (mode-specific!), 3046 * then these events will not cause the slave-array to be 3047 * refreshed. This will cause xmit to use a slave that is not 3048 * usable. Avoid such situation by refeshing the array at these 3049 * events. If these (miimon/arpmon) parameters are configured 3050 * then array gets refreshed twice and that should be fine! 3051 */ 3052 if (bond_mode_uses_xmit_hash(bond)) 3053 bond_update_slave_arr(bond, NULL); 3054 break; 3055 case NETDEV_CHANGEMTU: 3056 /* TODO: Should slaves be allowed to 3057 * independently alter their MTU? For 3058 * an active-backup bond, slaves need 3059 * not be the same type of device, so 3060 * MTUs may vary. For other modes, 3061 * slaves arguably should have the 3062 * same MTUs. To do this, we'd need to 3063 * take over the slave's change_mtu 3064 * function for the duration of their 3065 * servitude. 3066 */ 3067 break; 3068 case NETDEV_CHANGENAME: 3069 /* we don't care if we don't have primary set */ 3070 if (!bond_uses_primary(bond) || 3071 !bond->params.primary[0]) 3072 break; 3073 3074 if (slave == primary) { 3075 /* slave's name changed - he's no longer primary */ 3076 RCU_INIT_POINTER(bond->primary_slave, NULL); 3077 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3078 /* we have a new primary slave */ 3079 rcu_assign_pointer(bond->primary_slave, slave); 3080 } else { /* we didn't change primary - exit */ 3081 break; 3082 } 3083 3084 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3085 primary ? slave_dev->name : "none"); 3086 3087 block_netpoll_tx(); 3088 bond_select_active_slave(bond); 3089 unblock_netpoll_tx(); 3090 break; 3091 case NETDEV_FEAT_CHANGE: 3092 bond_compute_features(bond); 3093 break; 3094 case NETDEV_RESEND_IGMP: 3095 /* Propagate to master device */ 3096 call_netdevice_notifiers(event, slave->bond->dev); 3097 break; 3098 default: 3099 break; 3100 } 3101 3102 return NOTIFY_DONE; 3103 } 3104 3105 /* bond_netdev_event: handle netdev notifier chain events. 3106 * 3107 * This function receives events for the netdev chain. The caller (an 3108 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3109 * locks for us to safely manipulate the slave devices (RTNL lock, 3110 * dev_probe_lock). 3111 */ 3112 static int bond_netdev_event(struct notifier_block *this, 3113 unsigned long event, void *ptr) 3114 { 3115 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3116 3117 netdev_dbg(event_dev, "event: %lx\n", event); 3118 3119 if (!(event_dev->priv_flags & IFF_BONDING)) 3120 return NOTIFY_DONE; 3121 3122 if (event_dev->flags & IFF_MASTER) { 3123 netdev_dbg(event_dev, "IFF_MASTER\n"); 3124 return bond_master_netdev_event(event, event_dev); 3125 } 3126 3127 if (event_dev->flags & IFF_SLAVE) { 3128 netdev_dbg(event_dev, "IFF_SLAVE\n"); 3129 return bond_slave_netdev_event(event, event_dev); 3130 } 3131 3132 return NOTIFY_DONE; 3133 } 3134 3135 static struct notifier_block bond_netdev_notifier = { 3136 .notifier_call = bond_netdev_event, 3137 }; 3138 3139 /*---------------------------- Hashing Policies -----------------------------*/ 3140 3141 /* L2 hash helper */ 3142 static inline u32 bond_eth_hash(struct sk_buff *skb) 3143 { 3144 struct ethhdr *ep, hdr_tmp; 3145 3146 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3147 if (ep) 3148 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3149 return 0; 3150 } 3151 3152 /* Extract the appropriate headers based on bond's xmit policy */ 3153 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3154 struct flow_keys *fk) 3155 { 3156 const struct ipv6hdr *iph6; 3157 const struct iphdr *iph; 3158 int noff, proto = -1; 3159 3160 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) 3161 return skb_flow_dissect_flow_keys(skb, fk, 0); 3162 3163 fk->ports.ports = 0; 3164 noff = skb_network_offset(skb); 3165 if (skb->protocol == htons(ETH_P_IP)) { 3166 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 3167 return false; 3168 iph = ip_hdr(skb); 3169 iph_to_flow_copy_v4addrs(fk, iph); 3170 noff += iph->ihl << 2; 3171 if (!ip_is_fragment(iph)) 3172 proto = iph->protocol; 3173 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3174 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6)))) 3175 return false; 3176 iph6 = ipv6_hdr(skb); 3177 iph_to_flow_copy_v6addrs(fk, iph6); 3178 noff += sizeof(*iph6); 3179 proto = iph6->nexthdr; 3180 } else { 3181 return false; 3182 } 3183 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) 3184 fk->ports.ports = skb_flow_get_ports(skb, noff, proto); 3185 3186 return true; 3187 } 3188 3189 /** 3190 * bond_xmit_hash - generate a hash value based on the xmit policy 3191 * @bond: bonding device 3192 * @skb: buffer to use for headers 3193 * 3194 * This function will extract the necessary headers from the skb buffer and use 3195 * them to generate a hash based on the xmit_policy set in the bonding device 3196 */ 3197 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3198 { 3199 struct flow_keys flow; 3200 u32 hash; 3201 3202 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3203 skb->l4_hash) 3204 return skb->hash; 3205 3206 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3207 !bond_flow_dissect(bond, skb, &flow)) 3208 return bond_eth_hash(skb); 3209 3210 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3211 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) 3212 hash = bond_eth_hash(skb); 3213 else 3214 hash = (__force u32)flow.ports.ports; 3215 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3216 (__force u32)flow_get_u32_src(&flow); 3217 hash ^= (hash >> 16); 3218 hash ^= (hash >> 8); 3219 3220 return hash; 3221 } 3222 3223 /*-------------------------- Device entry points ----------------------------*/ 3224 3225 static void bond_work_init_all(struct bonding *bond) 3226 { 3227 INIT_DELAYED_WORK(&bond->mcast_work, 3228 bond_resend_igmp_join_requests_delayed); 3229 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3230 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3231 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3232 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon); 3233 else 3234 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon); 3235 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3236 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3237 } 3238 3239 static void bond_work_cancel_all(struct bonding *bond) 3240 { 3241 cancel_delayed_work_sync(&bond->mii_work); 3242 cancel_delayed_work_sync(&bond->arp_work); 3243 cancel_delayed_work_sync(&bond->alb_work); 3244 cancel_delayed_work_sync(&bond->ad_work); 3245 cancel_delayed_work_sync(&bond->mcast_work); 3246 cancel_delayed_work_sync(&bond->slave_arr_work); 3247 } 3248 3249 static int bond_open(struct net_device *bond_dev) 3250 { 3251 struct bonding *bond = netdev_priv(bond_dev); 3252 struct list_head *iter; 3253 struct slave *slave; 3254 3255 /* reset slave->backup and slave->inactive */ 3256 if (bond_has_slaves(bond)) { 3257 bond_for_each_slave(bond, slave, iter) { 3258 if (bond_uses_primary(bond) && 3259 slave != rcu_access_pointer(bond->curr_active_slave)) { 3260 bond_set_slave_inactive_flags(slave, 3261 BOND_SLAVE_NOTIFY_NOW); 3262 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3263 bond_set_slave_active_flags(slave, 3264 BOND_SLAVE_NOTIFY_NOW); 3265 } 3266 } 3267 } 3268 3269 bond_work_init_all(bond); 3270 3271 if (bond_is_lb(bond)) { 3272 /* bond_alb_initialize must be called before the timer 3273 * is started. 3274 */ 3275 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3276 return -ENOMEM; 3277 if (bond->params.tlb_dynamic_lb) 3278 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3279 } 3280 3281 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3282 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3283 3284 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3285 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3286 bond->recv_probe = bond_arp_rcv; 3287 } 3288 3289 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3290 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3291 /* register to receive LACPDUs */ 3292 bond->recv_probe = bond_3ad_lacpdu_recv; 3293 bond_3ad_initiate_agg_selection(bond, 1); 3294 } 3295 3296 if (bond_mode_uses_xmit_hash(bond)) 3297 bond_update_slave_arr(bond, NULL); 3298 3299 return 0; 3300 } 3301 3302 static int bond_close(struct net_device *bond_dev) 3303 { 3304 struct bonding *bond = netdev_priv(bond_dev); 3305 3306 bond_work_cancel_all(bond); 3307 bond->send_peer_notif = 0; 3308 if (bond_is_lb(bond)) 3309 bond_alb_deinitialize(bond); 3310 bond->recv_probe = NULL; 3311 3312 return 0; 3313 } 3314 3315 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3316 * that some drivers can provide 32bit values only. 3317 */ 3318 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3319 const struct rtnl_link_stats64 *_new, 3320 const struct rtnl_link_stats64 *_old) 3321 { 3322 const u64 *new = (const u64 *)_new; 3323 const u64 *old = (const u64 *)_old; 3324 u64 *res = (u64 *)_res; 3325 int i; 3326 3327 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3328 u64 nv = new[i]; 3329 u64 ov = old[i]; 3330 3331 /* detects if this particular field is 32bit only */ 3332 if (((nv | ov) >> 32) == 0) 3333 res[i] += (u32)nv - (u32)ov; 3334 else 3335 res[i] += nv - ov; 3336 } 3337 } 3338 3339 static void bond_get_stats(struct net_device *bond_dev, 3340 struct rtnl_link_stats64 *stats) 3341 { 3342 struct bonding *bond = netdev_priv(bond_dev); 3343 struct rtnl_link_stats64 temp; 3344 struct list_head *iter; 3345 struct slave *slave; 3346 3347 spin_lock(&bond->stats_lock); 3348 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3349 3350 rcu_read_lock(); 3351 bond_for_each_slave_rcu(bond, slave, iter) { 3352 const struct rtnl_link_stats64 *new = 3353 dev_get_stats(slave->dev, &temp); 3354 3355 bond_fold_stats(stats, new, &slave->slave_stats); 3356 3357 /* save off the slave stats for the next run */ 3358 memcpy(&slave->slave_stats, new, sizeof(*new)); 3359 } 3360 rcu_read_unlock(); 3361 3362 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3363 spin_unlock(&bond->stats_lock); 3364 } 3365 3366 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3367 { 3368 struct bonding *bond = netdev_priv(bond_dev); 3369 struct net_device *slave_dev = NULL; 3370 struct ifbond k_binfo; 3371 struct ifbond __user *u_binfo = NULL; 3372 struct ifslave k_sinfo; 3373 struct ifslave __user *u_sinfo = NULL; 3374 struct mii_ioctl_data *mii = NULL; 3375 struct bond_opt_value newval; 3376 struct net *net; 3377 int res = 0; 3378 3379 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3380 3381 switch (cmd) { 3382 case SIOCGMIIPHY: 3383 mii = if_mii(ifr); 3384 if (!mii) 3385 return -EINVAL; 3386 3387 mii->phy_id = 0; 3388 /* Fall Through */ 3389 case SIOCGMIIREG: 3390 /* We do this again just in case we were called by SIOCGMIIREG 3391 * instead of SIOCGMIIPHY. 3392 */ 3393 mii = if_mii(ifr); 3394 if (!mii) 3395 return -EINVAL; 3396 3397 if (mii->reg_num == 1) { 3398 mii->val_out = 0; 3399 if (netif_carrier_ok(bond->dev)) 3400 mii->val_out = BMSR_LSTATUS; 3401 } 3402 3403 return 0; 3404 case BOND_INFO_QUERY_OLD: 3405 case SIOCBONDINFOQUERY: 3406 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3407 3408 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3409 return -EFAULT; 3410 3411 bond_info_query(bond_dev, &k_binfo); 3412 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3413 return -EFAULT; 3414 3415 return 0; 3416 case BOND_SLAVE_INFO_QUERY_OLD: 3417 case SIOCBONDSLAVEINFOQUERY: 3418 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3419 3420 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3421 return -EFAULT; 3422 3423 res = bond_slave_info_query(bond_dev, &k_sinfo); 3424 if (res == 0 && 3425 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3426 return -EFAULT; 3427 3428 return res; 3429 default: 3430 break; 3431 } 3432 3433 net = dev_net(bond_dev); 3434 3435 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3436 return -EPERM; 3437 3438 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3439 3440 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev); 3441 3442 if (!slave_dev) 3443 return -ENODEV; 3444 3445 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name); 3446 switch (cmd) { 3447 case BOND_ENSLAVE_OLD: 3448 case SIOCBONDENSLAVE: 3449 res = bond_enslave(bond_dev, slave_dev); 3450 break; 3451 case BOND_RELEASE_OLD: 3452 case SIOCBONDRELEASE: 3453 res = bond_release(bond_dev, slave_dev); 3454 break; 3455 case BOND_SETHWADDR_OLD: 3456 case SIOCBONDSETHWADDR: 3457 bond_set_dev_addr(bond_dev, slave_dev); 3458 res = 0; 3459 break; 3460 case BOND_CHANGE_ACTIVE_OLD: 3461 case SIOCBONDCHANGEACTIVE: 3462 bond_opt_initstr(&newval, slave_dev->name); 3463 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval); 3464 break; 3465 default: 3466 res = -EOPNOTSUPP; 3467 } 3468 3469 return res; 3470 } 3471 3472 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3473 { 3474 struct bonding *bond = netdev_priv(bond_dev); 3475 3476 if (change & IFF_PROMISC) 3477 bond_set_promiscuity(bond, 3478 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3479 3480 if (change & IFF_ALLMULTI) 3481 bond_set_allmulti(bond, 3482 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3483 } 3484 3485 static void bond_set_rx_mode(struct net_device *bond_dev) 3486 { 3487 struct bonding *bond = netdev_priv(bond_dev); 3488 struct list_head *iter; 3489 struct slave *slave; 3490 3491 rcu_read_lock(); 3492 if (bond_uses_primary(bond)) { 3493 slave = rcu_dereference(bond->curr_active_slave); 3494 if (slave) { 3495 dev_uc_sync(slave->dev, bond_dev); 3496 dev_mc_sync(slave->dev, bond_dev); 3497 } 3498 } else { 3499 bond_for_each_slave_rcu(bond, slave, iter) { 3500 dev_uc_sync_multiple(slave->dev, bond_dev); 3501 dev_mc_sync_multiple(slave->dev, bond_dev); 3502 } 3503 } 3504 rcu_read_unlock(); 3505 } 3506 3507 static int bond_neigh_init(struct neighbour *n) 3508 { 3509 struct bonding *bond = netdev_priv(n->dev); 3510 const struct net_device_ops *slave_ops; 3511 struct neigh_parms parms; 3512 struct slave *slave; 3513 int ret; 3514 3515 slave = bond_first_slave(bond); 3516 if (!slave) 3517 return 0; 3518 slave_ops = slave->dev->netdev_ops; 3519 if (!slave_ops->ndo_neigh_setup) 3520 return 0; 3521 3522 parms.neigh_setup = NULL; 3523 parms.neigh_cleanup = NULL; 3524 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3525 if (ret) 3526 return ret; 3527 3528 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called 3529 * after the last slave has been detached. Assumes that all slaves 3530 * utilize the same neigh_cleanup (true at this writing as only user 3531 * is ipoib). 3532 */ 3533 n->parms->neigh_cleanup = parms.neigh_cleanup; 3534 3535 if (!parms.neigh_setup) 3536 return 0; 3537 3538 return parms.neigh_setup(n); 3539 } 3540 3541 /* The bonding ndo_neigh_setup is called at init time beofre any 3542 * slave exists. So we must declare proxy setup function which will 3543 * be used at run time to resolve the actual slave neigh param setup. 3544 * 3545 * It's also called by master devices (such as vlans) to setup their 3546 * underlying devices. In that case - do nothing, we're already set up from 3547 * our init. 3548 */ 3549 static int bond_neigh_setup(struct net_device *dev, 3550 struct neigh_parms *parms) 3551 { 3552 /* modify only our neigh_parms */ 3553 if (parms->dev == dev) 3554 parms->neigh_setup = bond_neigh_init; 3555 3556 return 0; 3557 } 3558 3559 /* Change the MTU of all of a master's slaves to match the master */ 3560 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3561 { 3562 struct bonding *bond = netdev_priv(bond_dev); 3563 struct slave *slave, *rollback_slave; 3564 struct list_head *iter; 3565 int res = 0; 3566 3567 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3568 3569 bond_for_each_slave(bond, slave, iter) { 3570 netdev_dbg(bond_dev, "s %p c_m %p\n", 3571 slave, slave->dev->netdev_ops->ndo_change_mtu); 3572 3573 res = dev_set_mtu(slave->dev, new_mtu); 3574 3575 if (res) { 3576 /* If we failed to set the slave's mtu to the new value 3577 * we must abort the operation even in ACTIVE_BACKUP 3578 * mode, because if we allow the backup slaves to have 3579 * different mtu values than the active slave we'll 3580 * need to change their mtu when doing a failover. That 3581 * means changing their mtu from timer context, which 3582 * is probably not a good idea. 3583 */ 3584 netdev_dbg(bond_dev, "err %d %s\n", res, 3585 slave->dev->name); 3586 goto unwind; 3587 } 3588 } 3589 3590 bond_dev->mtu = new_mtu; 3591 3592 return 0; 3593 3594 unwind: 3595 /* unwind from head to the slave that failed */ 3596 bond_for_each_slave(bond, rollback_slave, iter) { 3597 int tmp_res; 3598 3599 if (rollback_slave == slave) 3600 break; 3601 3602 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3603 if (tmp_res) { 3604 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3605 tmp_res, rollback_slave->dev->name); 3606 } 3607 } 3608 3609 return res; 3610 } 3611 3612 /* Change HW address 3613 * 3614 * Note that many devices must be down to change the HW address, and 3615 * downing the master releases all slaves. We can make bonds full of 3616 * bonding devices to test this, however. 3617 */ 3618 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3619 { 3620 struct bonding *bond = netdev_priv(bond_dev); 3621 struct slave *slave, *rollback_slave; 3622 struct sockaddr *sa = addr, tmp_sa; 3623 struct list_head *iter; 3624 int res = 0; 3625 3626 if (BOND_MODE(bond) == BOND_MODE_ALB) 3627 return bond_alb_set_mac_address(bond_dev, addr); 3628 3629 3630 netdev_dbg(bond_dev, "bond=%p\n", bond); 3631 3632 /* If fail_over_mac is enabled, do nothing and return success. 3633 * Returning an error causes ifenslave to fail. 3634 */ 3635 if (bond->params.fail_over_mac && 3636 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3637 return 0; 3638 3639 if (!is_valid_ether_addr(sa->sa_data)) 3640 return -EADDRNOTAVAIL; 3641 3642 bond_for_each_slave(bond, slave, iter) { 3643 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name); 3644 res = dev_set_mac_address(slave->dev, addr); 3645 if (res) { 3646 /* TODO: consider downing the slave 3647 * and retry ? 3648 * User should expect communications 3649 * breakage anyway until ARP finish 3650 * updating, so... 3651 */ 3652 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name); 3653 goto unwind; 3654 } 3655 } 3656 3657 /* success */ 3658 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len); 3659 return 0; 3660 3661 unwind: 3662 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len); 3663 tmp_sa.sa_family = bond_dev->type; 3664 3665 /* unwind from head to the slave that failed */ 3666 bond_for_each_slave(bond, rollback_slave, iter) { 3667 int tmp_res; 3668 3669 if (rollback_slave == slave) 3670 break; 3671 3672 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa); 3673 if (tmp_res) { 3674 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3675 tmp_res, rollback_slave->dev->name); 3676 } 3677 } 3678 3679 return res; 3680 } 3681 3682 /** 3683 * bond_xmit_slave_id - transmit skb through slave with slave_id 3684 * @bond: bonding device that is transmitting 3685 * @skb: buffer to transmit 3686 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3687 * 3688 * This function tries to transmit through slave with slave_id but in case 3689 * it fails, it tries to find the first available slave for transmission. 3690 * The skb is consumed in all cases, thus the function is void. 3691 */ 3692 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3693 { 3694 struct list_head *iter; 3695 struct slave *slave; 3696 int i = slave_id; 3697 3698 /* Here we start from the slave with slave_id */ 3699 bond_for_each_slave_rcu(bond, slave, iter) { 3700 if (--i < 0) { 3701 if (bond_slave_can_tx(slave)) { 3702 bond_dev_queue_xmit(bond, skb, slave->dev); 3703 return; 3704 } 3705 } 3706 } 3707 3708 /* Here we start from the first slave up to slave_id */ 3709 i = slave_id; 3710 bond_for_each_slave_rcu(bond, slave, iter) { 3711 if (--i < 0) 3712 break; 3713 if (bond_slave_can_tx(slave)) { 3714 bond_dev_queue_xmit(bond, skb, slave->dev); 3715 return; 3716 } 3717 } 3718 /* no slave that can tx has been found */ 3719 bond_tx_drop(bond->dev, skb); 3720 } 3721 3722 /** 3723 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3724 * @bond: bonding device to use 3725 * 3726 * Based on the value of the bonding device's packets_per_slave parameter 3727 * this function generates a slave id, which is usually used as the next 3728 * slave to transmit through. 3729 */ 3730 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3731 { 3732 u32 slave_id; 3733 struct reciprocal_value reciprocal_packets_per_slave; 3734 int packets_per_slave = bond->params.packets_per_slave; 3735 3736 switch (packets_per_slave) { 3737 case 0: 3738 slave_id = prandom_u32(); 3739 break; 3740 case 1: 3741 slave_id = bond->rr_tx_counter; 3742 break; 3743 default: 3744 reciprocal_packets_per_slave = 3745 bond->params.reciprocal_packets_per_slave; 3746 slave_id = reciprocal_divide(bond->rr_tx_counter, 3747 reciprocal_packets_per_slave); 3748 break; 3749 } 3750 bond->rr_tx_counter++; 3751 3752 return slave_id; 3753 } 3754 3755 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev) 3756 { 3757 struct bonding *bond = netdev_priv(bond_dev); 3758 struct iphdr *iph = ip_hdr(skb); 3759 struct slave *slave; 3760 u32 slave_id; 3761 3762 /* Start with the curr_active_slave that joined the bond as the 3763 * default for sending IGMP traffic. For failover purposes one 3764 * needs to maintain some consistency for the interface that will 3765 * send the join/membership reports. The curr_active_slave found 3766 * will send all of this type of traffic. 3767 */ 3768 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) { 3769 slave = rcu_dereference(bond->curr_active_slave); 3770 if (slave) 3771 bond_dev_queue_xmit(bond, skb, slave->dev); 3772 else 3773 bond_xmit_slave_id(bond, skb, 0); 3774 } else { 3775 int slave_cnt = ACCESS_ONCE(bond->slave_cnt); 3776 3777 if (likely(slave_cnt)) { 3778 slave_id = bond_rr_gen_slave_id(bond); 3779 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3780 } else { 3781 bond_tx_drop(bond_dev, skb); 3782 } 3783 } 3784 3785 return NETDEV_TX_OK; 3786 } 3787 3788 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3789 * the bond has a usable interface. 3790 */ 3791 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev) 3792 { 3793 struct bonding *bond = netdev_priv(bond_dev); 3794 struct slave *slave; 3795 3796 slave = rcu_dereference(bond->curr_active_slave); 3797 if (slave) 3798 bond_dev_queue_xmit(bond, skb, slave->dev); 3799 else 3800 bond_tx_drop(bond_dev, skb); 3801 3802 return NETDEV_TX_OK; 3803 } 3804 3805 /* Use this to update slave_array when (a) it's not appropriate to update 3806 * slave_array right away (note that update_slave_array() may sleep) 3807 * and / or (b) RTNL is not held. 3808 */ 3809 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 3810 { 3811 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 3812 } 3813 3814 /* Slave array work handler. Holds only RTNL */ 3815 static void bond_slave_arr_handler(struct work_struct *work) 3816 { 3817 struct bonding *bond = container_of(work, struct bonding, 3818 slave_arr_work.work); 3819 int ret; 3820 3821 if (!rtnl_trylock()) 3822 goto err; 3823 3824 ret = bond_update_slave_arr(bond, NULL); 3825 rtnl_unlock(); 3826 if (ret) { 3827 pr_warn_ratelimited("Failed to update slave array from WT\n"); 3828 goto err; 3829 } 3830 return; 3831 3832 err: 3833 bond_slave_arr_work_rearm(bond, 1); 3834 } 3835 3836 /* Build the usable slaves array in control path for modes that use xmit-hash 3837 * to determine the slave interface - 3838 * (a) BOND_MODE_8023AD 3839 * (b) BOND_MODE_XOR 3840 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0 3841 * 3842 * The caller is expected to hold RTNL only and NO other lock! 3843 */ 3844 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 3845 { 3846 struct slave *slave; 3847 struct list_head *iter; 3848 struct bond_up_slave *new_arr, *old_arr; 3849 int agg_id = 0; 3850 int ret = 0; 3851 3852 #ifdef CONFIG_LOCKDEP 3853 WARN_ON(lockdep_is_held(&bond->mode_lock)); 3854 #endif 3855 3856 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 3857 GFP_KERNEL); 3858 if (!new_arr) { 3859 ret = -ENOMEM; 3860 pr_err("Failed to build slave-array.\n"); 3861 goto out; 3862 } 3863 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3864 struct ad_info ad_info; 3865 3866 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 3867 pr_debug("bond_3ad_get_active_agg_info failed\n"); 3868 kfree_rcu(new_arr, rcu); 3869 /* No active aggragator means it's not safe to use 3870 * the previous array. 3871 */ 3872 old_arr = rtnl_dereference(bond->slave_arr); 3873 if (old_arr) { 3874 RCU_INIT_POINTER(bond->slave_arr, NULL); 3875 kfree_rcu(old_arr, rcu); 3876 } 3877 goto out; 3878 } 3879 agg_id = ad_info.aggregator_id; 3880 } 3881 bond_for_each_slave(bond, slave, iter) { 3882 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3883 struct aggregator *agg; 3884 3885 agg = SLAVE_AD_INFO(slave)->port.aggregator; 3886 if (!agg || agg->aggregator_identifier != agg_id) 3887 continue; 3888 } 3889 if (!bond_slave_can_tx(slave)) 3890 continue; 3891 if (skipslave == slave) 3892 continue; 3893 new_arr->arr[new_arr->count++] = slave; 3894 } 3895 3896 old_arr = rtnl_dereference(bond->slave_arr); 3897 rcu_assign_pointer(bond->slave_arr, new_arr); 3898 if (old_arr) 3899 kfree_rcu(old_arr, rcu); 3900 out: 3901 if (ret != 0 && skipslave) { 3902 int idx; 3903 3904 /* Rare situation where caller has asked to skip a specific 3905 * slave but allocation failed (most likely!). BTW this is 3906 * only possible when the call is initiated from 3907 * __bond_release_one(). In this situation; overwrite the 3908 * skipslave entry in the array with the last entry from the 3909 * array to avoid a situation where the xmit path may choose 3910 * this to-be-skipped slave to send a packet out. 3911 */ 3912 old_arr = rtnl_dereference(bond->slave_arr); 3913 for (idx = 0; idx < old_arr->count; idx++) { 3914 if (skipslave == old_arr->arr[idx]) { 3915 old_arr->arr[idx] = 3916 old_arr->arr[old_arr->count-1]; 3917 old_arr->count--; 3918 break; 3919 } 3920 } 3921 } 3922 return ret; 3923 } 3924 3925 /* Use this Xmit function for 3AD as well as XOR modes. The current 3926 * usable slave array is formed in the control path. The xmit function 3927 * just calculates hash and sends the packet out. 3928 */ 3929 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev) 3930 { 3931 struct bonding *bond = netdev_priv(dev); 3932 struct slave *slave; 3933 struct bond_up_slave *slaves; 3934 unsigned int count; 3935 3936 slaves = rcu_dereference(bond->slave_arr); 3937 count = slaves ? ACCESS_ONCE(slaves->count) : 0; 3938 if (likely(count)) { 3939 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 3940 bond_dev_queue_xmit(bond, skb, slave->dev); 3941 } else { 3942 bond_tx_drop(dev, skb); 3943 } 3944 3945 return NETDEV_TX_OK; 3946 } 3947 3948 /* in broadcast mode, we send everything to all usable interfaces. */ 3949 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev) 3950 { 3951 struct bonding *bond = netdev_priv(bond_dev); 3952 struct slave *slave = NULL; 3953 struct list_head *iter; 3954 3955 bond_for_each_slave_rcu(bond, slave, iter) { 3956 if (bond_is_last_slave(bond, slave)) 3957 break; 3958 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3959 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 3960 3961 if (!skb2) { 3962 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 3963 bond_dev->name, __func__); 3964 continue; 3965 } 3966 bond_dev_queue_xmit(bond, skb2, slave->dev); 3967 } 3968 } 3969 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 3970 bond_dev_queue_xmit(bond, skb, slave->dev); 3971 else 3972 bond_tx_drop(bond_dev, skb); 3973 3974 return NETDEV_TX_OK; 3975 } 3976 3977 /*------------------------- Device initialization ---------------------------*/ 3978 3979 /* Lookup the slave that corresponds to a qid */ 3980 static inline int bond_slave_override(struct bonding *bond, 3981 struct sk_buff *skb) 3982 { 3983 struct slave *slave = NULL; 3984 struct list_head *iter; 3985 3986 if (!skb->queue_mapping) 3987 return 1; 3988 3989 /* Find out if any slaves have the same mapping as this skb. */ 3990 bond_for_each_slave_rcu(bond, slave, iter) { 3991 if (slave->queue_id == skb->queue_mapping) { 3992 if (bond_slave_is_up(slave) && 3993 slave->link == BOND_LINK_UP) { 3994 bond_dev_queue_xmit(bond, skb, slave->dev); 3995 return 0; 3996 } 3997 /* If the slave isn't UP, use default transmit policy. */ 3998 break; 3999 } 4000 } 4001 4002 return 1; 4003 } 4004 4005 4006 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4007 void *accel_priv, select_queue_fallback_t fallback) 4008 { 4009 /* This helper function exists to help dev_pick_tx get the correct 4010 * destination queue. Using a helper function skips a call to 4011 * skb_tx_hash and will put the skbs in the queue we expect on their 4012 * way down to the bonding driver. 4013 */ 4014 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4015 4016 /* Save the original txq to restore before passing to the driver */ 4017 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping; 4018 4019 if (unlikely(txq >= dev->real_num_tx_queues)) { 4020 do { 4021 txq -= dev->real_num_tx_queues; 4022 } while (txq >= dev->real_num_tx_queues); 4023 } 4024 return txq; 4025 } 4026 4027 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4028 { 4029 struct bonding *bond = netdev_priv(dev); 4030 4031 if (bond_should_override_tx_queue(bond) && 4032 !bond_slave_override(bond, skb)) 4033 return NETDEV_TX_OK; 4034 4035 switch (BOND_MODE(bond)) { 4036 case BOND_MODE_ROUNDROBIN: 4037 return bond_xmit_roundrobin(skb, dev); 4038 case BOND_MODE_ACTIVEBACKUP: 4039 return bond_xmit_activebackup(skb, dev); 4040 case BOND_MODE_8023AD: 4041 case BOND_MODE_XOR: 4042 return bond_3ad_xor_xmit(skb, dev); 4043 case BOND_MODE_BROADCAST: 4044 return bond_xmit_broadcast(skb, dev); 4045 case BOND_MODE_ALB: 4046 return bond_alb_xmit(skb, dev); 4047 case BOND_MODE_TLB: 4048 return bond_tlb_xmit(skb, dev); 4049 default: 4050 /* Should never happen, mode already checked */ 4051 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4052 WARN_ON_ONCE(1); 4053 bond_tx_drop(dev, skb); 4054 return NETDEV_TX_OK; 4055 } 4056 } 4057 4058 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4059 { 4060 struct bonding *bond = netdev_priv(dev); 4061 netdev_tx_t ret = NETDEV_TX_OK; 4062 4063 /* If we risk deadlock from transmitting this in the 4064 * netpoll path, tell netpoll to queue the frame for later tx 4065 */ 4066 if (unlikely(is_netpoll_tx_blocked(dev))) 4067 return NETDEV_TX_BUSY; 4068 4069 rcu_read_lock(); 4070 if (bond_has_slaves(bond)) 4071 ret = __bond_start_xmit(skb, dev); 4072 else 4073 bond_tx_drop(dev, skb); 4074 rcu_read_unlock(); 4075 4076 return ret; 4077 } 4078 4079 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4080 struct ethtool_link_ksettings *cmd) 4081 { 4082 struct bonding *bond = netdev_priv(bond_dev); 4083 unsigned long speed = 0; 4084 struct list_head *iter; 4085 struct slave *slave; 4086 4087 cmd->base.duplex = DUPLEX_UNKNOWN; 4088 cmd->base.port = PORT_OTHER; 4089 4090 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4091 * do not need to check mode. Though link speed might not represent 4092 * the true receive or transmit bandwidth (not all modes are symmetric) 4093 * this is an accurate maximum. 4094 */ 4095 bond_for_each_slave(bond, slave, iter) { 4096 if (bond_slave_can_tx(slave)) { 4097 if (slave->speed != SPEED_UNKNOWN) 4098 speed += slave->speed; 4099 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4100 slave->duplex != DUPLEX_UNKNOWN) 4101 cmd->base.duplex = slave->duplex; 4102 } 4103 } 4104 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4105 4106 return 0; 4107 } 4108 4109 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4110 struct ethtool_drvinfo *drvinfo) 4111 { 4112 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4113 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 4114 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4115 BOND_ABI_VERSION); 4116 } 4117 4118 static const struct ethtool_ops bond_ethtool_ops = { 4119 .get_drvinfo = bond_ethtool_get_drvinfo, 4120 .get_link = ethtool_op_get_link, 4121 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4122 }; 4123 4124 static const struct net_device_ops bond_netdev_ops = { 4125 .ndo_init = bond_init, 4126 .ndo_uninit = bond_uninit, 4127 .ndo_open = bond_open, 4128 .ndo_stop = bond_close, 4129 .ndo_start_xmit = bond_start_xmit, 4130 .ndo_select_queue = bond_select_queue, 4131 .ndo_get_stats64 = bond_get_stats, 4132 .ndo_do_ioctl = bond_do_ioctl, 4133 .ndo_change_rx_flags = bond_change_rx_flags, 4134 .ndo_set_rx_mode = bond_set_rx_mode, 4135 .ndo_change_mtu = bond_change_mtu, 4136 .ndo_set_mac_address = bond_set_mac_address, 4137 .ndo_neigh_setup = bond_neigh_setup, 4138 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4139 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4140 #ifdef CONFIG_NET_POLL_CONTROLLER 4141 .ndo_netpoll_setup = bond_netpoll_setup, 4142 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4143 .ndo_poll_controller = bond_poll_controller, 4144 #endif 4145 .ndo_add_slave = bond_enslave, 4146 .ndo_del_slave = bond_release, 4147 .ndo_fix_features = bond_fix_features, 4148 .ndo_bridge_setlink = switchdev_port_bridge_setlink, 4149 .ndo_bridge_getlink = switchdev_port_bridge_getlink, 4150 .ndo_bridge_dellink = switchdev_port_bridge_dellink, 4151 .ndo_fdb_add = switchdev_port_fdb_add, 4152 .ndo_fdb_del = switchdev_port_fdb_del, 4153 .ndo_fdb_dump = switchdev_port_fdb_dump, 4154 .ndo_features_check = passthru_features_check, 4155 }; 4156 4157 static const struct device_type bond_type = { 4158 .name = "bond", 4159 }; 4160 4161 static void bond_destructor(struct net_device *bond_dev) 4162 { 4163 struct bonding *bond = netdev_priv(bond_dev); 4164 if (bond->wq) 4165 destroy_workqueue(bond->wq); 4166 free_netdev(bond_dev); 4167 } 4168 4169 void bond_setup(struct net_device *bond_dev) 4170 { 4171 struct bonding *bond = netdev_priv(bond_dev); 4172 4173 spin_lock_init(&bond->mode_lock); 4174 spin_lock_init(&bond->stats_lock); 4175 bond->params = bonding_defaults; 4176 4177 /* Initialize pointers */ 4178 bond->dev = bond_dev; 4179 4180 /* Initialize the device entry points */ 4181 ether_setup(bond_dev); 4182 bond_dev->max_mtu = ETH_MAX_MTU; 4183 bond_dev->netdev_ops = &bond_netdev_ops; 4184 bond_dev->ethtool_ops = &bond_ethtool_ops; 4185 4186 bond_dev->destructor = bond_destructor; 4187 4188 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4189 4190 /* Initialize the device options */ 4191 bond_dev->flags |= IFF_MASTER; 4192 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4193 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4194 4195 /* don't acquire bond device's netif_tx_lock when transmitting */ 4196 bond_dev->features |= NETIF_F_LLTX; 4197 4198 /* By default, we declare the bond to be fully 4199 * VLAN hardware accelerated capable. Special 4200 * care is taken in the various xmit functions 4201 * when there are slaves that are not hw accel 4202 * capable 4203 */ 4204 4205 /* Don't allow bond devices to change network namespaces. */ 4206 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4207 4208 bond_dev->hw_features = BOND_VLAN_FEATURES | 4209 NETIF_F_HW_VLAN_CTAG_TX | 4210 NETIF_F_HW_VLAN_CTAG_RX | 4211 NETIF_F_HW_VLAN_CTAG_FILTER; 4212 4213 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 4214 bond_dev->features |= bond_dev->hw_features; 4215 } 4216 4217 /* Destroy a bonding device. 4218 * Must be under rtnl_lock when this function is called. 4219 */ 4220 static void bond_uninit(struct net_device *bond_dev) 4221 { 4222 struct bonding *bond = netdev_priv(bond_dev); 4223 struct list_head *iter; 4224 struct slave *slave; 4225 struct bond_up_slave *arr; 4226 4227 bond_netpoll_cleanup(bond_dev); 4228 4229 /* Release the bonded slaves */ 4230 bond_for_each_slave(bond, slave, iter) 4231 __bond_release_one(bond_dev, slave->dev, true); 4232 netdev_info(bond_dev, "Released all slaves\n"); 4233 4234 arr = rtnl_dereference(bond->slave_arr); 4235 if (arr) { 4236 RCU_INIT_POINTER(bond->slave_arr, NULL); 4237 kfree_rcu(arr, rcu); 4238 } 4239 4240 list_del(&bond->bond_list); 4241 4242 bond_debug_unregister(bond); 4243 } 4244 4245 /*------------------------- Module initialization ---------------------------*/ 4246 4247 static int bond_check_params(struct bond_params *params) 4248 { 4249 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4250 struct bond_opt_value newval; 4251 const struct bond_opt_value *valptr; 4252 int arp_all_targets_value; 4253 u16 ad_actor_sys_prio = 0; 4254 u16 ad_user_port_key = 0; 4255 4256 /* Convert string parameters. */ 4257 if (mode) { 4258 bond_opt_initstr(&newval, mode); 4259 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4260 if (!valptr) { 4261 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4262 return -EINVAL; 4263 } 4264 bond_mode = valptr->value; 4265 } 4266 4267 if (xmit_hash_policy) { 4268 if ((bond_mode != BOND_MODE_XOR) && 4269 (bond_mode != BOND_MODE_8023AD) && 4270 (bond_mode != BOND_MODE_TLB)) { 4271 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4272 bond_mode_name(bond_mode)); 4273 } else { 4274 bond_opt_initstr(&newval, xmit_hash_policy); 4275 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4276 &newval); 4277 if (!valptr) { 4278 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4279 xmit_hash_policy); 4280 return -EINVAL; 4281 } 4282 xmit_hashtype = valptr->value; 4283 } 4284 } 4285 4286 if (lacp_rate) { 4287 if (bond_mode != BOND_MODE_8023AD) { 4288 pr_info("lacp_rate param is irrelevant in mode %s\n", 4289 bond_mode_name(bond_mode)); 4290 } else { 4291 bond_opt_initstr(&newval, lacp_rate); 4292 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4293 &newval); 4294 if (!valptr) { 4295 pr_err("Error: Invalid lacp rate \"%s\"\n", 4296 lacp_rate); 4297 return -EINVAL; 4298 } 4299 lacp_fast = valptr->value; 4300 } 4301 } 4302 4303 if (ad_select) { 4304 bond_opt_initstr(&newval, ad_select); 4305 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4306 &newval); 4307 if (!valptr) { 4308 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4309 return -EINVAL; 4310 } 4311 params->ad_select = valptr->value; 4312 if (bond_mode != BOND_MODE_8023AD) 4313 pr_warn("ad_select param only affects 802.3ad mode\n"); 4314 } else { 4315 params->ad_select = BOND_AD_STABLE; 4316 } 4317 4318 if (max_bonds < 0) { 4319 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4320 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4321 max_bonds = BOND_DEFAULT_MAX_BONDS; 4322 } 4323 4324 if (miimon < 0) { 4325 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4326 miimon, INT_MAX); 4327 miimon = 0; 4328 } 4329 4330 if (updelay < 0) { 4331 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4332 updelay, INT_MAX); 4333 updelay = 0; 4334 } 4335 4336 if (downdelay < 0) { 4337 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4338 downdelay, INT_MAX); 4339 downdelay = 0; 4340 } 4341 4342 if ((use_carrier != 0) && (use_carrier != 1)) { 4343 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4344 use_carrier); 4345 use_carrier = 1; 4346 } 4347 4348 if (num_peer_notif < 0 || num_peer_notif > 255) { 4349 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4350 num_peer_notif); 4351 num_peer_notif = 1; 4352 } 4353 4354 /* reset values for 802.3ad/TLB/ALB */ 4355 if (!bond_mode_uses_arp(bond_mode)) { 4356 if (!miimon) { 4357 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"); 4358 pr_warn("Forcing miimon to 100msec\n"); 4359 miimon = BOND_DEFAULT_MIIMON; 4360 } 4361 } 4362 4363 if (tx_queues < 1 || tx_queues > 255) { 4364 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4365 tx_queues, BOND_DEFAULT_TX_QUEUES); 4366 tx_queues = BOND_DEFAULT_TX_QUEUES; 4367 } 4368 4369 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4370 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4371 all_slaves_active); 4372 all_slaves_active = 0; 4373 } 4374 4375 if (resend_igmp < 0 || resend_igmp > 255) { 4376 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4377 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4378 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4379 } 4380 4381 bond_opt_initval(&newval, packets_per_slave); 4382 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4383 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4384 packets_per_slave, USHRT_MAX); 4385 packets_per_slave = 1; 4386 } 4387 4388 if (bond_mode == BOND_MODE_ALB) { 4389 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", 4390 updelay); 4391 } 4392 4393 if (!miimon) { 4394 if (updelay || downdelay) { 4395 /* just warn the user the up/down delay will have 4396 * no effect since miimon is zero... 4397 */ 4398 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", 4399 updelay, downdelay); 4400 } 4401 } else { 4402 /* don't allow arp monitoring */ 4403 if (arp_interval) { 4404 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4405 miimon, arp_interval); 4406 arp_interval = 0; 4407 } 4408 4409 if ((updelay % miimon) != 0) { 4410 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4411 updelay, miimon, (updelay / miimon) * miimon); 4412 } 4413 4414 updelay /= miimon; 4415 4416 if ((downdelay % miimon) != 0) { 4417 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4418 downdelay, miimon, 4419 (downdelay / miimon) * miimon); 4420 } 4421 4422 downdelay /= miimon; 4423 } 4424 4425 if (arp_interval < 0) { 4426 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4427 arp_interval, INT_MAX); 4428 arp_interval = 0; 4429 } 4430 4431 for (arp_ip_count = 0, i = 0; 4432 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4433 __be32 ip; 4434 4435 /* not a complete check, but good enough to catch mistakes */ 4436 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4437 !bond_is_ip_target_ok(ip)) { 4438 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4439 arp_ip_target[i]); 4440 arp_interval = 0; 4441 } else { 4442 if (bond_get_targets_ip(arp_target, ip) == -1) 4443 arp_target[arp_ip_count++] = ip; 4444 else 4445 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4446 &ip); 4447 } 4448 } 4449 4450 if (arp_interval && !arp_ip_count) { 4451 /* don't allow arping if no arp_ip_target given... */ 4452 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4453 arp_interval); 4454 arp_interval = 0; 4455 } 4456 4457 if (arp_validate) { 4458 if (!arp_interval) { 4459 pr_err("arp_validate requires arp_interval\n"); 4460 return -EINVAL; 4461 } 4462 4463 bond_opt_initstr(&newval, arp_validate); 4464 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4465 &newval); 4466 if (!valptr) { 4467 pr_err("Error: invalid arp_validate \"%s\"\n", 4468 arp_validate); 4469 return -EINVAL; 4470 } 4471 arp_validate_value = valptr->value; 4472 } else { 4473 arp_validate_value = 0; 4474 } 4475 4476 arp_all_targets_value = 0; 4477 if (arp_all_targets) { 4478 bond_opt_initstr(&newval, arp_all_targets); 4479 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4480 &newval); 4481 if (!valptr) { 4482 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4483 arp_all_targets); 4484 arp_all_targets_value = 0; 4485 } else { 4486 arp_all_targets_value = valptr->value; 4487 } 4488 } 4489 4490 if (miimon) { 4491 pr_info("MII link monitoring set to %d ms\n", miimon); 4492 } else if (arp_interval) { 4493 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4494 arp_validate_value); 4495 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4496 arp_interval, valptr->string, arp_ip_count); 4497 4498 for (i = 0; i < arp_ip_count; i++) 4499 pr_cont(" %s", arp_ip_target[i]); 4500 4501 pr_cont("\n"); 4502 4503 } else if (max_bonds) { 4504 /* miimon and arp_interval not set, we need one so things 4505 * work as expected, see bonding.txt for details 4506 */ 4507 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"); 4508 } 4509 4510 if (primary && !bond_mode_uses_primary(bond_mode)) { 4511 /* currently, using a primary only makes sense 4512 * in active backup, TLB or ALB modes 4513 */ 4514 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4515 primary, bond_mode_name(bond_mode)); 4516 primary = NULL; 4517 } 4518 4519 if (primary && primary_reselect) { 4520 bond_opt_initstr(&newval, primary_reselect); 4521 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4522 &newval); 4523 if (!valptr) { 4524 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4525 primary_reselect); 4526 return -EINVAL; 4527 } 4528 primary_reselect_value = valptr->value; 4529 } else { 4530 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4531 } 4532 4533 if (fail_over_mac) { 4534 bond_opt_initstr(&newval, fail_over_mac); 4535 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4536 &newval); 4537 if (!valptr) { 4538 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4539 fail_over_mac); 4540 return -EINVAL; 4541 } 4542 fail_over_mac_value = valptr->value; 4543 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4544 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4545 } else { 4546 fail_over_mac_value = BOND_FOM_NONE; 4547 } 4548 4549 bond_opt_initstr(&newval, "default"); 4550 valptr = bond_opt_parse( 4551 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 4552 &newval); 4553 if (!valptr) { 4554 pr_err("Error: No ad_actor_sys_prio default value"); 4555 return -EINVAL; 4556 } 4557 ad_actor_sys_prio = valptr->value; 4558 4559 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 4560 &newval); 4561 if (!valptr) { 4562 pr_err("Error: No ad_user_port_key default value"); 4563 return -EINVAL; 4564 } 4565 ad_user_port_key = valptr->value; 4566 4567 if (lp_interval == 0) { 4568 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4569 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4570 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4571 } 4572 4573 /* fill params struct with the proper values */ 4574 params->mode = bond_mode; 4575 params->xmit_policy = xmit_hashtype; 4576 params->miimon = miimon; 4577 params->num_peer_notif = num_peer_notif; 4578 params->arp_interval = arp_interval; 4579 params->arp_validate = arp_validate_value; 4580 params->arp_all_targets = arp_all_targets_value; 4581 params->updelay = updelay; 4582 params->downdelay = downdelay; 4583 params->use_carrier = use_carrier; 4584 params->lacp_fast = lacp_fast; 4585 params->primary[0] = 0; 4586 params->primary_reselect = primary_reselect_value; 4587 params->fail_over_mac = fail_over_mac_value; 4588 params->tx_queues = tx_queues; 4589 params->all_slaves_active = all_slaves_active; 4590 params->resend_igmp = resend_igmp; 4591 params->min_links = min_links; 4592 params->lp_interval = lp_interval; 4593 params->packets_per_slave = packets_per_slave; 4594 params->tlb_dynamic_lb = 1; /* Default value */ 4595 params->ad_actor_sys_prio = ad_actor_sys_prio; 4596 eth_zero_addr(params->ad_actor_system); 4597 params->ad_user_port_key = ad_user_port_key; 4598 if (packets_per_slave > 0) { 4599 params->reciprocal_packets_per_slave = 4600 reciprocal_value(packets_per_slave); 4601 } else { 4602 /* reciprocal_packets_per_slave is unused if 4603 * packets_per_slave is 0 or 1, just initialize it 4604 */ 4605 params->reciprocal_packets_per_slave = 4606 (struct reciprocal_value) { 0 }; 4607 } 4608 4609 if (primary) { 4610 strncpy(params->primary, primary, IFNAMSIZ); 4611 params->primary[IFNAMSIZ - 1] = 0; 4612 } 4613 4614 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4615 4616 return 0; 4617 } 4618 4619 /* Called from registration process */ 4620 static int bond_init(struct net_device *bond_dev) 4621 { 4622 struct bonding *bond = netdev_priv(bond_dev); 4623 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4624 4625 netdev_dbg(bond_dev, "Begin bond_init\n"); 4626 4627 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 4628 if (!bond->wq) 4629 return -ENOMEM; 4630 4631 netdev_lockdep_set_classes(bond_dev); 4632 4633 list_add_tail(&bond->bond_list, &bn->dev_list); 4634 4635 bond_prepare_sysfs_group(bond); 4636 4637 bond_debug_register(bond); 4638 4639 /* Ensure valid dev_addr */ 4640 if (is_zero_ether_addr(bond_dev->dev_addr) && 4641 bond_dev->addr_assign_type == NET_ADDR_PERM) 4642 eth_hw_addr_random(bond_dev); 4643 4644 return 0; 4645 } 4646 4647 unsigned int bond_get_num_tx_queues(void) 4648 { 4649 return tx_queues; 4650 } 4651 4652 /* Create a new bond based on the specified name and bonding parameters. 4653 * If name is NULL, obtain a suitable "bond%d" name for us. 4654 * Caller must NOT hold rtnl_lock; we need to release it here before we 4655 * set up our sysfs entries. 4656 */ 4657 int bond_create(struct net *net, const char *name) 4658 { 4659 struct net_device *bond_dev; 4660 struct bonding *bond; 4661 struct alb_bond_info *bond_info; 4662 int res; 4663 4664 rtnl_lock(); 4665 4666 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4667 name ? name : "bond%d", NET_NAME_UNKNOWN, 4668 bond_setup, tx_queues); 4669 if (!bond_dev) { 4670 pr_err("%s: eek! can't alloc netdev!\n", name); 4671 rtnl_unlock(); 4672 return -ENOMEM; 4673 } 4674 4675 /* 4676 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 4677 * It is set to 0 by default which is wrong. 4678 */ 4679 bond = netdev_priv(bond_dev); 4680 bond_info = &(BOND_ALB_INFO(bond)); 4681 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 4682 4683 dev_net_set(bond_dev, net); 4684 bond_dev->rtnl_link_ops = &bond_link_ops; 4685 4686 res = register_netdevice(bond_dev); 4687 4688 netif_carrier_off(bond_dev); 4689 4690 rtnl_unlock(); 4691 if (res < 0) 4692 bond_destructor(bond_dev); 4693 return res; 4694 } 4695 4696 static int __net_init bond_net_init(struct net *net) 4697 { 4698 struct bond_net *bn = net_generic(net, bond_net_id); 4699 4700 bn->net = net; 4701 INIT_LIST_HEAD(&bn->dev_list); 4702 4703 bond_create_proc_dir(bn); 4704 bond_create_sysfs(bn); 4705 4706 return 0; 4707 } 4708 4709 static void __net_exit bond_net_exit(struct net *net) 4710 { 4711 struct bond_net *bn = net_generic(net, bond_net_id); 4712 struct bonding *bond, *tmp_bond; 4713 LIST_HEAD(list); 4714 4715 bond_destroy_sysfs(bn); 4716 4717 /* Kill off any bonds created after unregistering bond rtnl ops */ 4718 rtnl_lock(); 4719 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4720 unregister_netdevice_queue(bond->dev, &list); 4721 unregister_netdevice_many(&list); 4722 rtnl_unlock(); 4723 4724 bond_destroy_proc_dir(bn); 4725 } 4726 4727 static struct pernet_operations bond_net_ops = { 4728 .init = bond_net_init, 4729 .exit = bond_net_exit, 4730 .id = &bond_net_id, 4731 .size = sizeof(struct bond_net), 4732 }; 4733 4734 static int __init bonding_init(void) 4735 { 4736 int i; 4737 int res; 4738 4739 pr_info("%s", bond_version); 4740 4741 res = bond_check_params(&bonding_defaults); 4742 if (res) 4743 goto out; 4744 4745 res = register_pernet_subsys(&bond_net_ops); 4746 if (res) 4747 goto out; 4748 4749 res = bond_netlink_init(); 4750 if (res) 4751 goto err_link; 4752 4753 bond_create_debugfs(); 4754 4755 for (i = 0; i < max_bonds; i++) { 4756 res = bond_create(&init_net, NULL); 4757 if (res) 4758 goto err; 4759 } 4760 4761 register_netdevice_notifier(&bond_netdev_notifier); 4762 out: 4763 return res; 4764 err: 4765 bond_destroy_debugfs(); 4766 bond_netlink_fini(); 4767 err_link: 4768 unregister_pernet_subsys(&bond_net_ops); 4769 goto out; 4770 4771 } 4772 4773 static void __exit bonding_exit(void) 4774 { 4775 unregister_netdevice_notifier(&bond_netdev_notifier); 4776 4777 bond_destroy_debugfs(); 4778 4779 bond_netlink_fini(); 4780 unregister_pernet_subsys(&bond_net_ops); 4781 4782 #ifdef CONFIG_NET_POLL_CONTROLLER 4783 /* Make sure we don't have an imbalance on our netpoll blocking */ 4784 WARN_ON(atomic_read(&netpoll_block_tx)); 4785 #endif 4786 } 4787 4788 module_init(bonding_init); 4789 module_exit(bonding_exit); 4790 MODULE_LICENSE("GPL"); 4791 MODULE_VERSION(DRV_VERSION); 4792 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 4793 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 4794