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/icmp.h> 45#include <linux/icmpv6.h> 46#include <linux/tcp.h> 47#include <linux/udp.h> 48#include <linux/slab.h> 49#include <linux/string.h> 50#include <linux/init.h> 51#include <linux/timer.h> 52#include <linux/socket.h> 53#include <linux/ctype.h> 54#include <linux/inet.h> 55#include <linux/bitops.h> 56#include <linux/io.h> 57#include <asm/dma.h> 58#include <linux/uaccess.h> 59#include <linux/errno.h> 60#include <linux/netdevice.h> 61#include <linux/inetdevice.h> 62#include <linux/igmp.h> 63#include <linux/etherdevice.h> 64#include <linux/skbuff.h> 65#include <net/sock.h> 66#include <linux/rtnetlink.h> 67#include <linux/smp.h> 68#include <linux/if_ether.h> 69#include <net/arp.h> 70#include <linux/mii.h> 71#include <linux/ethtool.h> 72#include <linux/if_vlan.h> 73#include <linux/if_bonding.h> 74#include <linux/jiffies.h> 75#include <linux/preempt.h> 76#include <net/route.h> 77#include <net/net_namespace.h> 78#include <net/netns/generic.h> 79#include <net/pkt_sched.h> 80#include <linux/rculist.h> 81#include <net/flow_dissector.h> 82#include <net/xfrm.h> 83#include <net/bonding.h> 84#include <net/bond_3ad.h> 85#include <net/bond_alb.h> 86 87#include "bonding_priv.h" 88 89/*---------------------------- Module parameters ----------------------------*/ 90 91/* monitor all links that often (in milliseconds). <=0 disables monitoring */ 92 93static int max_bonds = BOND_DEFAULT_MAX_BONDS; 94static int tx_queues = BOND_DEFAULT_TX_QUEUES; 95static int num_peer_notif = 1; 96static int miimon; 97static int updelay; 98static int downdelay; 99static int use_carrier = 1; 100static char *mode; 101static char *primary; 102static char *primary_reselect; 103static char *lacp_rate; 104static int min_links; 105static char *ad_select; 106static char *xmit_hash_policy; 107static int arp_interval; 108static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 109static char *arp_validate; 110static char *arp_all_targets; 111static char *fail_over_mac; 112static int all_slaves_active; 113static struct bond_params bonding_defaults; 114static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 115static int packets_per_slave = 1; 116static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 117 118module_param(max_bonds, int, 0); 119MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 120module_param(tx_queues, int, 0); 121MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 122module_param_named(num_grat_arp, num_peer_notif, int, 0644); 123MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 124 "failover event (alias of num_unsol_na)"); 125module_param_named(num_unsol_na, num_peer_notif, int, 0644); 126MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 127 "failover event (alias of num_grat_arp)"); 128module_param(miimon, int, 0); 129MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 130module_param(updelay, int, 0); 131MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 132module_param(downdelay, int, 0); 133MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 134 "in milliseconds"); 135module_param(use_carrier, int, 0); 136MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 137 "0 for off, 1 for on (default)"); 138module_param(mode, charp, 0); 139MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 140 "1 for active-backup, 2 for balance-xor, " 141 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 142 "6 for balance-alb"); 143module_param(primary, charp, 0); 144MODULE_PARM_DESC(primary, "Primary network device to use"); 145module_param(primary_reselect, charp, 0); 146MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 147 "once it comes up; " 148 "0 for always (default), " 149 "1 for only if speed of primary is " 150 "better, " 151 "2 for only on active slave " 152 "failure"); 153module_param(lacp_rate, charp, 0); 154MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 155 "0 for slow, 1 for fast"); 156module_param(ad_select, charp, 0); 157MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 158 "0 for stable (default), 1 for bandwidth, " 159 "2 for count"); 160module_param(min_links, int, 0); 161MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 162 163module_param(xmit_hash_policy, charp, 0); 164MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 165 "0 for layer 2 (default), 1 for layer 3+4, " 166 "2 for layer 2+3, 3 for encap layer 2+3, " 167 "4 for encap layer 3+4"); 168module_param(arp_interval, int, 0); 169MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 170module_param_array(arp_ip_target, charp, NULL, 0); 171MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 172module_param(arp_validate, charp, 0); 173MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 174 "0 for none (default), 1 for active, " 175 "2 for backup, 3 for all"); 176module_param(arp_all_targets, charp, 0); 177MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 178module_param(fail_over_mac, charp, 0); 179MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 180 "the same MAC; 0 for none (default), " 181 "1 for active, 2 for follow"); 182module_param(all_slaves_active, int, 0); 183MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 184 "by setting active flag for all slaves; " 185 "0 for never (default), 1 for always."); 186module_param(resend_igmp, int, 0); 187MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 188 "link failure"); 189module_param(packets_per_slave, int, 0); 190MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 191 "mode; 0 for a random slave, 1 packet per " 192 "slave (default), >1 packets per slave."); 193module_param(lp_interval, uint, 0); 194MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 195 "the bonding driver sends learning packets to " 196 "each slaves peer switch. The default is 1."); 197 198/*----------------------------- Global variables ----------------------------*/ 199 200#ifdef CONFIG_NET_POLL_CONTROLLER 201atomic_t netpoll_block_tx = ATOMIC_INIT(0); 202#endif 203 204unsigned int bond_net_id __read_mostly; 205 206static const struct flow_dissector_key flow_keys_bonding_keys[] = { 207 { 208 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 209 .offset = offsetof(struct flow_keys, control), 210 }, 211 { 212 .key_id = FLOW_DISSECTOR_KEY_BASIC, 213 .offset = offsetof(struct flow_keys, basic), 214 }, 215 { 216 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 217 .offset = offsetof(struct flow_keys, addrs.v4addrs), 218 }, 219 { 220 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 221 .offset = offsetof(struct flow_keys, addrs.v6addrs), 222 }, 223 { 224 .key_id = FLOW_DISSECTOR_KEY_TIPC, 225 .offset = offsetof(struct flow_keys, addrs.tipckey), 226 }, 227 { 228 .key_id = FLOW_DISSECTOR_KEY_PORTS, 229 .offset = offsetof(struct flow_keys, ports), 230 }, 231 { 232 .key_id = FLOW_DISSECTOR_KEY_ICMP, 233 .offset = offsetof(struct flow_keys, icmp), 234 }, 235 { 236 .key_id = FLOW_DISSECTOR_KEY_VLAN, 237 .offset = offsetof(struct flow_keys, vlan), 238 }, 239 { 240 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 241 .offset = offsetof(struct flow_keys, tags), 242 }, 243 { 244 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 245 .offset = offsetof(struct flow_keys, keyid), 246 }, 247}; 248 249static struct flow_dissector flow_keys_bonding __read_mostly; 250 251/*-------------------------- Forward declarations ---------------------------*/ 252 253static int bond_init(struct net_device *bond_dev); 254static void bond_uninit(struct net_device *bond_dev); 255static void bond_get_stats(struct net_device *bond_dev, 256 struct rtnl_link_stats64 *stats); 257static void bond_slave_arr_handler(struct work_struct *work); 258static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 259 int mod); 260static void bond_netdev_notify_work(struct work_struct *work); 261 262/*---------------------------- General routines -----------------------------*/ 263 264const char *bond_mode_name(int mode) 265{ 266 static const char *names[] = { 267 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 268 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 269 [BOND_MODE_XOR] = "load balancing (xor)", 270 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 271 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 272 [BOND_MODE_TLB] = "transmit load balancing", 273 [BOND_MODE_ALB] = "adaptive load balancing", 274 }; 275 276 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 277 return "unknown"; 278 279 return names[mode]; 280} 281 282/** 283 * bond_dev_queue_xmit - Prepare skb for xmit. 284 * 285 * @bond: bond device that got this skb for tx. 286 * @skb: hw accel VLAN tagged skb to transmit 287 * @slave_dev: slave that is supposed to xmit this skbuff 288 */ 289netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 290 struct net_device *slave_dev) 291{ 292 skb->dev = slave_dev; 293 294 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 295 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 296 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 297 298 if (unlikely(netpoll_tx_running(bond->dev))) 299 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 300 301 return dev_queue_xmit(skb); 302} 303 304/*---------------------------------- VLAN -----------------------------------*/ 305 306/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 307 * We don't protect the slave list iteration with a lock because: 308 * a. This operation is performed in IOCTL context, 309 * b. The operation is protected by the RTNL semaphore in the 8021q code, 310 * c. Holding a lock with BH disabled while directly calling a base driver 311 * entry point is generally a BAD idea. 312 * 313 * The design of synchronization/protection for this operation in the 8021q 314 * module is good for one or more VLAN devices over a single physical device 315 * and cannot be extended for a teaming solution like bonding, so there is a 316 * potential race condition here where a net device from the vlan group might 317 * be referenced (either by a base driver or the 8021q code) while it is being 318 * removed from the system. However, it turns out we're not making matters 319 * worse, and if it works for regular VLAN usage it will work here too. 320*/ 321 322/** 323 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 324 * @bond_dev: bonding net device that got called 325 * @proto: network protocol ID 326 * @vid: vlan id being added 327 */ 328static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 329 __be16 proto, u16 vid) 330{ 331 struct bonding *bond = netdev_priv(bond_dev); 332 struct slave *slave, *rollback_slave; 333 struct list_head *iter; 334 int res; 335 336 bond_for_each_slave(bond, slave, iter) { 337 res = vlan_vid_add(slave->dev, proto, vid); 338 if (res) 339 goto unwind; 340 } 341 342 return 0; 343 344unwind: 345 /* unwind to the slave that failed */ 346 bond_for_each_slave(bond, rollback_slave, iter) { 347 if (rollback_slave == slave) 348 break; 349 350 vlan_vid_del(rollback_slave->dev, proto, vid); 351 } 352 353 return res; 354} 355 356/** 357 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 358 * @bond_dev: bonding net device that got called 359 * @proto: network protocol ID 360 * @vid: vlan id being removed 361 */ 362static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 363 __be16 proto, u16 vid) 364{ 365 struct bonding *bond = netdev_priv(bond_dev); 366 struct list_head *iter; 367 struct slave *slave; 368 369 bond_for_each_slave(bond, slave, iter) 370 vlan_vid_del(slave->dev, proto, vid); 371 372 if (bond_is_lb(bond)) 373 bond_alb_clear_vlan(bond, vid); 374 375 return 0; 376} 377 378/*---------------------------------- XFRM -----------------------------------*/ 379 380#ifdef CONFIG_XFRM_OFFLOAD 381/** 382 * bond_ipsec_add_sa - program device with a security association 383 * @xs: pointer to transformer state struct 384 **/ 385static int bond_ipsec_add_sa(struct xfrm_state *xs) 386{ 387 struct net_device *bond_dev = xs->xso.dev; 388 struct bond_ipsec *ipsec; 389 struct bonding *bond; 390 struct slave *slave; 391 int err; 392 393 if (!bond_dev) 394 return -EINVAL; 395 396 rcu_read_lock(); 397 bond = netdev_priv(bond_dev); 398 slave = rcu_dereference(bond->curr_active_slave); 399 if (!slave) { 400 rcu_read_unlock(); 401 return -ENODEV; 402 } 403 404 if (!slave->dev->xfrmdev_ops || 405 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 406 netif_is_bond_master(slave->dev)) { 407 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n"); 408 rcu_read_unlock(); 409 return -EINVAL; 410 } 411 412 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC); 413 if (!ipsec) { 414 rcu_read_unlock(); 415 return -ENOMEM; 416 } 417 xs->xso.real_dev = slave->dev; 418 419 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs); 420 if (!err) { 421 ipsec->xs = xs; 422 INIT_LIST_HEAD(&ipsec->list); 423 spin_lock_bh(&bond->ipsec_lock); 424 list_add(&ipsec->list, &bond->ipsec_list); 425 spin_unlock_bh(&bond->ipsec_lock); 426 } else { 427 kfree(ipsec); 428 } 429 rcu_read_unlock(); 430 return err; 431} 432 433static void bond_ipsec_add_sa_all(struct bonding *bond) 434{ 435 struct net_device *bond_dev = bond->dev; 436 struct bond_ipsec *ipsec; 437 struct slave *slave; 438 439 rcu_read_lock(); 440 slave = rcu_dereference(bond->curr_active_slave); 441 if (!slave) 442 goto out; 443 444 if (!slave->dev->xfrmdev_ops || 445 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 446 netif_is_bond_master(slave->dev)) { 447 spin_lock_bh(&bond->ipsec_lock); 448 if (!list_empty(&bond->ipsec_list)) 449 slave_warn(bond_dev, slave->dev, 450 "%s: no slave xdo_dev_state_add\n", 451 __func__); 452 spin_unlock_bh(&bond->ipsec_lock); 453 goto out; 454 } 455 456 spin_lock_bh(&bond->ipsec_lock); 457 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 458 ipsec->xs->xso.real_dev = slave->dev; 459 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) { 460 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__); 461 ipsec->xs->xso.real_dev = NULL; 462 } 463 } 464 spin_unlock_bh(&bond->ipsec_lock); 465out: 466 rcu_read_unlock(); 467} 468 469/** 470 * bond_ipsec_del_sa - clear out this specific SA 471 * @xs: pointer to transformer state struct 472 **/ 473static void bond_ipsec_del_sa(struct xfrm_state *xs) 474{ 475 struct net_device *bond_dev = xs->xso.dev; 476 struct bond_ipsec *ipsec; 477 struct bonding *bond; 478 struct slave *slave; 479 480 if (!bond_dev) 481 return; 482 483 rcu_read_lock(); 484 bond = netdev_priv(bond_dev); 485 slave = rcu_dereference(bond->curr_active_slave); 486 487 if (!slave) 488 goto out; 489 490 if (!xs->xso.real_dev) 491 goto out; 492 493 WARN_ON(xs->xso.real_dev != slave->dev); 494 495 if (!slave->dev->xfrmdev_ops || 496 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 497 netif_is_bond_master(slave->dev)) { 498 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__); 499 goto out; 500 } 501 502 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs); 503out: 504 spin_lock_bh(&bond->ipsec_lock); 505 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 506 if (ipsec->xs == xs) { 507 list_del(&ipsec->list); 508 kfree(ipsec); 509 break; 510 } 511 } 512 spin_unlock_bh(&bond->ipsec_lock); 513 rcu_read_unlock(); 514} 515 516static void bond_ipsec_del_sa_all(struct bonding *bond) 517{ 518 struct net_device *bond_dev = bond->dev; 519 struct bond_ipsec *ipsec; 520 struct slave *slave; 521 522 rcu_read_lock(); 523 slave = rcu_dereference(bond->curr_active_slave); 524 if (!slave) { 525 rcu_read_unlock(); 526 return; 527 } 528 529 spin_lock_bh(&bond->ipsec_lock); 530 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 531 if (!ipsec->xs->xso.real_dev) 532 continue; 533 534 if (!slave->dev->xfrmdev_ops || 535 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 536 netif_is_bond_master(slave->dev)) { 537 slave_warn(bond_dev, slave->dev, 538 "%s: no slave xdo_dev_state_delete\n", 539 __func__); 540 } else { 541 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 542 } 543 } 544 spin_unlock_bh(&bond->ipsec_lock); 545 rcu_read_unlock(); 546} 547 548/** 549 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 550 * @skb: current data packet 551 * @xs: pointer to transformer state struct 552 **/ 553static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 554{ 555 struct net_device *bond_dev = xs->xso.dev; 556 struct net_device *real_dev; 557 struct slave *curr_active; 558 struct bonding *bond; 559 int err; 560 561 bond = netdev_priv(bond_dev); 562 rcu_read_lock(); 563 curr_active = rcu_dereference(bond->curr_active_slave); 564 if (!curr_active) 565 goto out; 566 real_dev = curr_active->dev; 567 568 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 569 err = false; 570 goto out; 571 } 572 573 if (!xs->xso.real_dev) { 574 err = false; 575 goto out; 576 } 577 578 if (!real_dev->xfrmdev_ops || 579 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 580 netif_is_bond_master(real_dev)) { 581 err = false; 582 goto out; 583 } 584 585 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 586out: 587 rcu_read_unlock(); 588 return err; 589} 590 591static const struct xfrmdev_ops bond_xfrmdev_ops = { 592 .xdo_dev_state_add = bond_ipsec_add_sa, 593 .xdo_dev_state_delete = bond_ipsec_del_sa, 594 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 595}; 596#endif /* CONFIG_XFRM_OFFLOAD */ 597 598/*------------------------------- Link status -------------------------------*/ 599 600/* Set the carrier state for the master according to the state of its 601 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 602 * do special 802.3ad magic. 603 * 604 * Returns zero if carrier state does not change, nonzero if it does. 605 */ 606int bond_set_carrier(struct bonding *bond) 607{ 608 struct list_head *iter; 609 struct slave *slave; 610 611 if (!bond_has_slaves(bond)) 612 goto down; 613 614 if (BOND_MODE(bond) == BOND_MODE_8023AD) 615 return bond_3ad_set_carrier(bond); 616 617 bond_for_each_slave(bond, slave, iter) { 618 if (slave->link == BOND_LINK_UP) { 619 if (!netif_carrier_ok(bond->dev)) { 620 netif_carrier_on(bond->dev); 621 return 1; 622 } 623 return 0; 624 } 625 } 626 627down: 628 if (netif_carrier_ok(bond->dev)) { 629 netif_carrier_off(bond->dev); 630 return 1; 631 } 632 return 0; 633} 634 635/* Get link speed and duplex from the slave's base driver 636 * using ethtool. If for some reason the call fails or the 637 * values are invalid, set speed and duplex to -1, 638 * and return. Return 1 if speed or duplex settings are 639 * UNKNOWN; 0 otherwise. 640 */ 641static int bond_update_speed_duplex(struct slave *slave) 642{ 643 struct net_device *slave_dev = slave->dev; 644 struct ethtool_link_ksettings ecmd; 645 int res; 646 647 slave->speed = SPEED_UNKNOWN; 648 slave->duplex = DUPLEX_UNKNOWN; 649 650 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 651 if (res < 0) 652 return 1; 653 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 654 return 1; 655 switch (ecmd.base.duplex) { 656 case DUPLEX_FULL: 657 case DUPLEX_HALF: 658 break; 659 default: 660 return 1; 661 } 662 663 slave->speed = ecmd.base.speed; 664 slave->duplex = ecmd.base.duplex; 665 666 return 0; 667} 668 669const char *bond_slave_link_status(s8 link) 670{ 671 switch (link) { 672 case BOND_LINK_UP: 673 return "up"; 674 case BOND_LINK_FAIL: 675 return "going down"; 676 case BOND_LINK_DOWN: 677 return "down"; 678 case BOND_LINK_BACK: 679 return "going back"; 680 default: 681 return "unknown"; 682 } 683} 684 685/* if <dev> supports MII link status reporting, check its link status. 686 * 687 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 688 * depending upon the setting of the use_carrier parameter. 689 * 690 * Return either BMSR_LSTATUS, meaning that the link is up (or we 691 * can't tell and just pretend it is), or 0, meaning that the link is 692 * down. 693 * 694 * If reporting is non-zero, instead of faking link up, return -1 if 695 * both ETHTOOL and MII ioctls fail (meaning the device does not 696 * support them). If use_carrier is set, return whatever it says. 697 * It'd be nice if there was a good way to tell if a driver supports 698 * netif_carrier, but there really isn't. 699 */ 700static int bond_check_dev_link(struct bonding *bond, 701 struct net_device *slave_dev, int reporting) 702{ 703 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 704 int (*ioctl)(struct net_device *, struct ifreq *, int); 705 struct ifreq ifr; 706 struct mii_ioctl_data *mii; 707 708 if (!reporting && !netif_running(slave_dev)) 709 return 0; 710 711 if (bond->params.use_carrier) 712 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 713 714 /* Try to get link status using Ethtool first. */ 715 if (slave_dev->ethtool_ops->get_link) 716 return slave_dev->ethtool_ops->get_link(slave_dev) ? 717 BMSR_LSTATUS : 0; 718 719 /* Ethtool can't be used, fallback to MII ioctls. */ 720 ioctl = slave_ops->ndo_do_ioctl; 721 if (ioctl) { 722 /* TODO: set pointer to correct ioctl on a per team member 723 * bases to make this more efficient. that is, once 724 * we determine the correct ioctl, we will always 725 * call it and not the others for that team 726 * member. 727 */ 728 729 /* We cannot assume that SIOCGMIIPHY will also read a 730 * register; not all network drivers (e.g., e100) 731 * support that. 732 */ 733 734 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 735 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 736 mii = if_mii(&ifr); 737 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 738 mii->reg_num = MII_BMSR; 739 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 740 return mii->val_out & BMSR_LSTATUS; 741 } 742 } 743 744 /* If reporting, report that either there's no dev->do_ioctl, 745 * or both SIOCGMIIREG and get_link failed (meaning that we 746 * cannot report link status). If not reporting, pretend 747 * we're ok. 748 */ 749 return reporting ? -1 : BMSR_LSTATUS; 750} 751 752/*----------------------------- Multicast list ------------------------------*/ 753 754/* Push the promiscuity flag down to appropriate slaves */ 755static int bond_set_promiscuity(struct bonding *bond, int inc) 756{ 757 struct list_head *iter; 758 int err = 0; 759 760 if (bond_uses_primary(bond)) { 761 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 762 763 if (curr_active) 764 err = dev_set_promiscuity(curr_active->dev, inc); 765 } else { 766 struct slave *slave; 767 768 bond_for_each_slave(bond, slave, iter) { 769 err = dev_set_promiscuity(slave->dev, inc); 770 if (err) 771 return err; 772 } 773 } 774 return err; 775} 776 777/* Push the allmulti flag down to all slaves */ 778static int bond_set_allmulti(struct bonding *bond, int inc) 779{ 780 struct list_head *iter; 781 int err = 0; 782 783 if (bond_uses_primary(bond)) { 784 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 785 786 if (curr_active) 787 err = dev_set_allmulti(curr_active->dev, inc); 788 } else { 789 struct slave *slave; 790 791 bond_for_each_slave(bond, slave, iter) { 792 err = dev_set_allmulti(slave->dev, inc); 793 if (err) 794 return err; 795 } 796 } 797 return err; 798} 799 800/* Retrieve the list of registered multicast addresses for the bonding 801 * device and retransmit an IGMP JOIN request to the current active 802 * slave. 803 */ 804static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 805{ 806 struct bonding *bond = container_of(work, struct bonding, 807 mcast_work.work); 808 809 if (!rtnl_trylock()) { 810 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 811 return; 812 } 813 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 814 815 if (bond->igmp_retrans > 1) { 816 bond->igmp_retrans--; 817 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 818 } 819 rtnl_unlock(); 820} 821 822/* Flush bond's hardware addresses from slave */ 823static void bond_hw_addr_flush(struct net_device *bond_dev, 824 struct net_device *slave_dev) 825{ 826 struct bonding *bond = netdev_priv(bond_dev); 827 828 dev_uc_unsync(slave_dev, bond_dev); 829 dev_mc_unsync(slave_dev, bond_dev); 830 831 if (BOND_MODE(bond) == BOND_MODE_8023AD) 832 dev_mc_del(slave_dev, lacpdu_mcast_addr); 833} 834 835/*--------------------------- Active slave change ---------------------------*/ 836 837/* Update the hardware address list and promisc/allmulti for the new and 838 * old active slaves (if any). Modes that are not using primary keep all 839 * slaves up date at all times; only the modes that use primary need to call 840 * this function to swap these settings during a failover. 841 */ 842static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 843 struct slave *old_active) 844{ 845 if (old_active) { 846 if (bond->dev->flags & IFF_PROMISC) 847 dev_set_promiscuity(old_active->dev, -1); 848 849 if (bond->dev->flags & IFF_ALLMULTI) 850 dev_set_allmulti(old_active->dev, -1); 851 852 if (bond->dev->flags & IFF_UP) 853 bond_hw_addr_flush(bond->dev, old_active->dev); 854 } 855 856 if (new_active) { 857 /* FIXME: Signal errors upstream. */ 858 if (bond->dev->flags & IFF_PROMISC) 859 dev_set_promiscuity(new_active->dev, 1); 860 861 if (bond->dev->flags & IFF_ALLMULTI) 862 dev_set_allmulti(new_active->dev, 1); 863 864 if (bond->dev->flags & IFF_UP) { 865 netif_addr_lock_bh(bond->dev); 866 dev_uc_sync(new_active->dev, bond->dev); 867 dev_mc_sync(new_active->dev, bond->dev); 868 netif_addr_unlock_bh(bond->dev); 869 } 870 } 871} 872 873/** 874 * bond_set_dev_addr - clone slave's address to bond 875 * @bond_dev: bond net device 876 * @slave_dev: slave net device 877 * 878 * Should be called with RTNL held. 879 */ 880static int bond_set_dev_addr(struct net_device *bond_dev, 881 struct net_device *slave_dev) 882{ 883 int err; 884 885 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 886 bond_dev, slave_dev, slave_dev->addr_len); 887 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 888 if (err) 889 return err; 890 891 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 892 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 893 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 894 return 0; 895} 896 897static struct slave *bond_get_old_active(struct bonding *bond, 898 struct slave *new_active) 899{ 900 struct slave *slave; 901 struct list_head *iter; 902 903 bond_for_each_slave(bond, slave, iter) { 904 if (slave == new_active) 905 continue; 906 907 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 908 return slave; 909 } 910 911 return NULL; 912} 913 914/* bond_do_fail_over_mac 915 * 916 * Perform special MAC address swapping for fail_over_mac settings 917 * 918 * Called with RTNL 919 */ 920static void bond_do_fail_over_mac(struct bonding *bond, 921 struct slave *new_active, 922 struct slave *old_active) 923{ 924 u8 tmp_mac[MAX_ADDR_LEN]; 925 struct sockaddr_storage ss; 926 int rv; 927 928 switch (bond->params.fail_over_mac) { 929 case BOND_FOM_ACTIVE: 930 if (new_active) { 931 rv = bond_set_dev_addr(bond->dev, new_active->dev); 932 if (rv) 933 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 934 -rv); 935 } 936 break; 937 case BOND_FOM_FOLLOW: 938 /* if new_active && old_active, swap them 939 * if just old_active, do nothing (going to no active slave) 940 * if just new_active, set new_active to bond's MAC 941 */ 942 if (!new_active) 943 return; 944 945 if (!old_active) 946 old_active = bond_get_old_active(bond, new_active); 947 948 if (old_active) { 949 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 950 new_active->dev->addr_len); 951 bond_hw_addr_copy(ss.__data, 952 old_active->dev->dev_addr, 953 old_active->dev->addr_len); 954 ss.ss_family = new_active->dev->type; 955 } else { 956 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 957 bond->dev->addr_len); 958 ss.ss_family = bond->dev->type; 959 } 960 961 rv = dev_set_mac_address(new_active->dev, 962 (struct sockaddr *)&ss, NULL); 963 if (rv) { 964 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 965 -rv); 966 goto out; 967 } 968 969 if (!old_active) 970 goto out; 971 972 bond_hw_addr_copy(ss.__data, tmp_mac, 973 new_active->dev->addr_len); 974 ss.ss_family = old_active->dev->type; 975 976 rv = dev_set_mac_address(old_active->dev, 977 (struct sockaddr *)&ss, NULL); 978 if (rv) 979 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 980 -rv); 981out: 982 break; 983 default: 984 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 985 bond->params.fail_over_mac); 986 break; 987 } 988 989} 990 991static struct slave *bond_choose_primary_or_current(struct bonding *bond) 992{ 993 struct slave *prim = rtnl_dereference(bond->primary_slave); 994 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 995 996 if (!prim || prim->link != BOND_LINK_UP) { 997 if (!curr || curr->link != BOND_LINK_UP) 998 return NULL; 999 return curr; 1000 } 1001 1002 if (bond->force_primary) { 1003 bond->force_primary = false; 1004 return prim; 1005 } 1006 1007 if (!curr || curr->link != BOND_LINK_UP) 1008 return prim; 1009 1010 /* At this point, prim and curr are both up */ 1011 switch (bond->params.primary_reselect) { 1012 case BOND_PRI_RESELECT_ALWAYS: 1013 return prim; 1014 case BOND_PRI_RESELECT_BETTER: 1015 if (prim->speed < curr->speed) 1016 return curr; 1017 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1018 return curr; 1019 return prim; 1020 case BOND_PRI_RESELECT_FAILURE: 1021 return curr; 1022 default: 1023 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1024 bond->params.primary_reselect); 1025 return curr; 1026 } 1027} 1028 1029/** 1030 * bond_find_best_slave - select the best available slave to be the active one 1031 * @bond: our bonding struct 1032 */ 1033static struct slave *bond_find_best_slave(struct bonding *bond) 1034{ 1035 struct slave *slave, *bestslave = NULL; 1036 struct list_head *iter; 1037 int mintime = bond->params.updelay; 1038 1039 slave = bond_choose_primary_or_current(bond); 1040 if (slave) 1041 return slave; 1042 1043 bond_for_each_slave(bond, slave, iter) { 1044 if (slave->link == BOND_LINK_UP) 1045 return slave; 1046 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1047 slave->delay < mintime) { 1048 mintime = slave->delay; 1049 bestslave = slave; 1050 } 1051 } 1052 1053 return bestslave; 1054} 1055 1056static bool bond_should_notify_peers(struct bonding *bond) 1057{ 1058 struct slave *slave; 1059 1060 rcu_read_lock(); 1061 slave = rcu_dereference(bond->curr_active_slave); 1062 rcu_read_unlock(); 1063 1064 if (!slave || !bond->send_peer_notif || 1065 bond->send_peer_notif % 1066 max(1, bond->params.peer_notif_delay) != 0 || 1067 !netif_carrier_ok(bond->dev) || 1068 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1069 return false; 1070 1071 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1072 slave ? slave->dev->name : "NULL"); 1073 1074 return true; 1075} 1076 1077/** 1078 * change_active_interface - change the active slave into the specified one 1079 * @bond: our bonding struct 1080 * @new_active: the new slave to make the active one 1081 * 1082 * Set the new slave to the bond's settings and unset them on the old 1083 * curr_active_slave. 1084 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1085 * 1086 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1087 * because it is apparently the best available slave we have, even though its 1088 * updelay hasn't timed out yet. 1089 * 1090 * Caller must hold RTNL. 1091 */ 1092void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1093{ 1094 struct slave *old_active; 1095 1096 ASSERT_RTNL(); 1097 1098 old_active = rtnl_dereference(bond->curr_active_slave); 1099 1100 if (old_active == new_active) 1101 return; 1102 1103#ifdef CONFIG_XFRM_OFFLOAD 1104 bond_ipsec_del_sa_all(bond); 1105#endif /* CONFIG_XFRM_OFFLOAD */ 1106 1107 if (new_active) { 1108 new_active->last_link_up = jiffies; 1109 1110 if (new_active->link == BOND_LINK_BACK) { 1111 if (bond_uses_primary(bond)) { 1112 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1113 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1114 } 1115 1116 new_active->delay = 0; 1117 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1118 BOND_SLAVE_NOTIFY_NOW); 1119 1120 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1121 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1122 1123 if (bond_is_lb(bond)) 1124 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1125 } else { 1126 if (bond_uses_primary(bond)) { 1127 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1128 } 1129 } 1130 } 1131 1132 if (bond_uses_primary(bond)) 1133 bond_hw_addr_swap(bond, new_active, old_active); 1134 1135 if (bond_is_lb(bond)) { 1136 bond_alb_handle_active_change(bond, new_active); 1137 if (old_active) 1138 bond_set_slave_inactive_flags(old_active, 1139 BOND_SLAVE_NOTIFY_NOW); 1140 if (new_active) 1141 bond_set_slave_active_flags(new_active, 1142 BOND_SLAVE_NOTIFY_NOW); 1143 } else { 1144 rcu_assign_pointer(bond->curr_active_slave, new_active); 1145 } 1146 1147 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1148 if (old_active) 1149 bond_set_slave_inactive_flags(old_active, 1150 BOND_SLAVE_NOTIFY_NOW); 1151 1152 if (new_active) { 1153 bool should_notify_peers = false; 1154 1155 bond_set_slave_active_flags(new_active, 1156 BOND_SLAVE_NOTIFY_NOW); 1157 1158 if (bond->params.fail_over_mac) 1159 bond_do_fail_over_mac(bond, new_active, 1160 old_active); 1161 1162 if (netif_running(bond->dev)) { 1163 bond->send_peer_notif = 1164 bond->params.num_peer_notif * 1165 max(1, bond->params.peer_notif_delay); 1166 should_notify_peers = 1167 bond_should_notify_peers(bond); 1168 } 1169 1170 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1171 if (should_notify_peers) { 1172 bond->send_peer_notif--; 1173 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1174 bond->dev); 1175 } 1176 } 1177 } 1178 1179#ifdef CONFIG_XFRM_OFFLOAD 1180 bond_ipsec_add_sa_all(bond); 1181#endif /* CONFIG_XFRM_OFFLOAD */ 1182 1183 /* resend IGMP joins since active slave has changed or 1184 * all were sent on curr_active_slave. 1185 * resend only if bond is brought up with the affected 1186 * bonding modes and the retransmission is enabled 1187 */ 1188 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1189 ((bond_uses_primary(bond) && new_active) || 1190 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1191 bond->igmp_retrans = bond->params.resend_igmp; 1192 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1193 } 1194} 1195 1196/** 1197 * bond_select_active_slave - select a new active slave, if needed 1198 * @bond: our bonding struct 1199 * 1200 * This functions should be called when one of the following occurs: 1201 * - The old curr_active_slave has been released or lost its link. 1202 * - The primary_slave has got its link back. 1203 * - A slave has got its link back and there's no old curr_active_slave. 1204 * 1205 * Caller must hold RTNL. 1206 */ 1207void bond_select_active_slave(struct bonding *bond) 1208{ 1209 struct slave *best_slave; 1210 int rv; 1211 1212 ASSERT_RTNL(); 1213 1214 best_slave = bond_find_best_slave(bond); 1215 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1216 bond_change_active_slave(bond, best_slave); 1217 rv = bond_set_carrier(bond); 1218 if (!rv) 1219 return; 1220 1221 if (netif_carrier_ok(bond->dev)) 1222 netdev_info(bond->dev, "active interface up!\n"); 1223 else 1224 netdev_info(bond->dev, "now running without any active interface!\n"); 1225 } 1226} 1227 1228#ifdef CONFIG_NET_POLL_CONTROLLER 1229static inline int slave_enable_netpoll(struct slave *slave) 1230{ 1231 struct netpoll *np; 1232 int err = 0; 1233 1234 np = kzalloc(sizeof(*np), GFP_KERNEL); 1235 err = -ENOMEM; 1236 if (!np) 1237 goto out; 1238 1239 err = __netpoll_setup(np, slave->dev); 1240 if (err) { 1241 kfree(np); 1242 goto out; 1243 } 1244 slave->np = np; 1245out: 1246 return err; 1247} 1248static inline void slave_disable_netpoll(struct slave *slave) 1249{ 1250 struct netpoll *np = slave->np; 1251 1252 if (!np) 1253 return; 1254 1255 slave->np = NULL; 1256 1257 __netpoll_free(np); 1258} 1259 1260static void bond_poll_controller(struct net_device *bond_dev) 1261{ 1262 struct bonding *bond = netdev_priv(bond_dev); 1263 struct slave *slave = NULL; 1264 struct list_head *iter; 1265 struct ad_info ad_info; 1266 1267 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1268 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1269 return; 1270 1271 bond_for_each_slave_rcu(bond, slave, iter) { 1272 if (!bond_slave_is_up(slave)) 1273 continue; 1274 1275 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1276 struct aggregator *agg = 1277 SLAVE_AD_INFO(slave)->port.aggregator; 1278 1279 if (agg && 1280 agg->aggregator_identifier != ad_info.aggregator_id) 1281 continue; 1282 } 1283 1284 netpoll_poll_dev(slave->dev); 1285 } 1286} 1287 1288static void bond_netpoll_cleanup(struct net_device *bond_dev) 1289{ 1290 struct bonding *bond = netdev_priv(bond_dev); 1291 struct list_head *iter; 1292 struct slave *slave; 1293 1294 bond_for_each_slave(bond, slave, iter) 1295 if (bond_slave_is_up(slave)) 1296 slave_disable_netpoll(slave); 1297} 1298 1299static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1300{ 1301 struct bonding *bond = netdev_priv(dev); 1302 struct list_head *iter; 1303 struct slave *slave; 1304 int err = 0; 1305 1306 bond_for_each_slave(bond, slave, iter) { 1307 err = slave_enable_netpoll(slave); 1308 if (err) { 1309 bond_netpoll_cleanup(dev); 1310 break; 1311 } 1312 } 1313 return err; 1314} 1315#else 1316static inline int slave_enable_netpoll(struct slave *slave) 1317{ 1318 return 0; 1319} 1320static inline void slave_disable_netpoll(struct slave *slave) 1321{ 1322} 1323static void bond_netpoll_cleanup(struct net_device *bond_dev) 1324{ 1325} 1326#endif 1327 1328/*---------------------------------- IOCTL ----------------------------------*/ 1329 1330static netdev_features_t bond_fix_features(struct net_device *dev, 1331 netdev_features_t features) 1332{ 1333 struct bonding *bond = netdev_priv(dev); 1334 struct list_head *iter; 1335 netdev_features_t mask; 1336 struct slave *slave; 1337 1338 mask = features; 1339 1340 features &= ~NETIF_F_ONE_FOR_ALL; 1341 features |= NETIF_F_ALL_FOR_ALL; 1342 1343 bond_for_each_slave(bond, slave, iter) { 1344 features = netdev_increment_features(features, 1345 slave->dev->features, 1346 mask); 1347 } 1348 features = netdev_add_tso_features(features, mask); 1349 1350 return features; 1351} 1352 1353#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1354 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1355 NETIF_F_HIGHDMA | NETIF_F_LRO) 1356 1357#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1358 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1359 1360#define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1361 NETIF_F_ALL_TSO) 1362 1363 1364static void bond_compute_features(struct bonding *bond) 1365{ 1366 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1367 IFF_XMIT_DST_RELEASE_PERM; 1368 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1369 netdev_features_t enc_features = BOND_ENC_FEATURES; 1370#ifdef CONFIG_XFRM_OFFLOAD 1371 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1372#endif /* CONFIG_XFRM_OFFLOAD */ 1373 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1374 struct net_device *bond_dev = bond->dev; 1375 struct list_head *iter; 1376 struct slave *slave; 1377 unsigned short max_hard_header_len = ETH_HLEN; 1378 unsigned int gso_max_size = GSO_MAX_SIZE; 1379 u16 gso_max_segs = GSO_MAX_SEGS; 1380 1381 if (!bond_has_slaves(bond)) 1382 goto done; 1383 vlan_features &= NETIF_F_ALL_FOR_ALL; 1384 mpls_features &= NETIF_F_ALL_FOR_ALL; 1385 1386 bond_for_each_slave(bond, slave, iter) { 1387 vlan_features = netdev_increment_features(vlan_features, 1388 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1389 1390 enc_features = netdev_increment_features(enc_features, 1391 slave->dev->hw_enc_features, 1392 BOND_ENC_FEATURES); 1393 1394#ifdef CONFIG_XFRM_OFFLOAD 1395 xfrm_features = netdev_increment_features(xfrm_features, 1396 slave->dev->hw_enc_features, 1397 BOND_XFRM_FEATURES); 1398#endif /* CONFIG_XFRM_OFFLOAD */ 1399 1400 mpls_features = netdev_increment_features(mpls_features, 1401 slave->dev->mpls_features, 1402 BOND_MPLS_FEATURES); 1403 1404 dst_release_flag &= slave->dev->priv_flags; 1405 if (slave->dev->hard_header_len > max_hard_header_len) 1406 max_hard_header_len = slave->dev->hard_header_len; 1407 1408 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1409 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1410 } 1411 bond_dev->hard_header_len = max_hard_header_len; 1412 1413done: 1414 bond_dev->vlan_features = vlan_features; 1415 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1416 NETIF_F_HW_VLAN_CTAG_TX | 1417 NETIF_F_HW_VLAN_STAG_TX | 1418 NETIF_F_GSO_UDP_L4; 1419#ifdef CONFIG_XFRM_OFFLOAD 1420 bond_dev->hw_enc_features |= xfrm_features; 1421#endif /* CONFIG_XFRM_OFFLOAD */ 1422 bond_dev->mpls_features = mpls_features; 1423 bond_dev->gso_max_segs = gso_max_segs; 1424 netif_set_gso_max_size(bond_dev, gso_max_size); 1425 1426 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1427 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1428 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1429 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1430 1431 netdev_change_features(bond_dev); 1432} 1433 1434static void bond_setup_by_slave(struct net_device *bond_dev, 1435 struct net_device *slave_dev) 1436{ 1437 bool was_up = !!(bond_dev->flags & IFF_UP); 1438 1439 dev_close(bond_dev); 1440 1441 bond_dev->header_ops = slave_dev->header_ops; 1442 1443 bond_dev->type = slave_dev->type; 1444 bond_dev->hard_header_len = slave_dev->hard_header_len; 1445 bond_dev->needed_headroom = slave_dev->needed_headroom; 1446 bond_dev->addr_len = slave_dev->addr_len; 1447 1448 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1449 slave_dev->addr_len); 1450 1451 if (slave_dev->flags & IFF_POINTOPOINT) { 1452 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1453 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1454 } 1455 if (was_up) 1456 dev_open(bond_dev, NULL); 1457} 1458 1459/* On bonding slaves other than the currently active slave, suppress 1460 * duplicates except for alb non-mcast/bcast. 1461 */ 1462static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1463 struct slave *slave, 1464 struct bonding *bond) 1465{ 1466 if (bond_is_slave_inactive(slave)) { 1467 if (BOND_MODE(bond) == BOND_MODE_ALB && 1468 skb->pkt_type != PACKET_BROADCAST && 1469 skb->pkt_type != PACKET_MULTICAST) 1470 return false; 1471 return true; 1472 } 1473 return false; 1474} 1475 1476static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1477{ 1478 struct sk_buff *skb = *pskb; 1479 struct slave *slave; 1480 struct bonding *bond; 1481 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1482 struct slave *); 1483 int ret = RX_HANDLER_ANOTHER; 1484 1485 skb = skb_share_check(skb, GFP_ATOMIC); 1486 if (unlikely(!skb)) 1487 return RX_HANDLER_CONSUMED; 1488 1489 *pskb = skb; 1490 1491 slave = bond_slave_get_rcu(skb->dev); 1492 bond = slave->bond; 1493 1494 recv_probe = READ_ONCE(bond->recv_probe); 1495 if (recv_probe) { 1496 ret = recv_probe(skb, bond, slave); 1497 if (ret == RX_HANDLER_CONSUMED) { 1498 consume_skb(skb); 1499 return ret; 1500 } 1501 } 1502 1503 /* 1504 * For packets determined by bond_should_deliver_exact_match() call to 1505 * be suppressed we want to make an exception for link-local packets. 1506 * This is necessary for e.g. LLDP daemons to be able to monitor 1507 * inactive slave links without being forced to bind to them 1508 * explicitly. 1509 * 1510 * At the same time, packets that are passed to the bonding master 1511 * (including link-local ones) can have their originating interface 1512 * determined via PACKET_ORIGDEV socket option. 1513 */ 1514 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1515 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1516 return RX_HANDLER_PASS; 1517 return RX_HANDLER_EXACT; 1518 } 1519 1520 skb->dev = bond->dev; 1521 1522 if (BOND_MODE(bond) == BOND_MODE_ALB && 1523 netif_is_bridge_port(bond->dev) && 1524 skb->pkt_type == PACKET_HOST) { 1525 1526 if (unlikely(skb_cow_head(skb, 1527 skb->data - skb_mac_header(skb)))) { 1528 kfree_skb(skb); 1529 return RX_HANDLER_CONSUMED; 1530 } 1531 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1532 bond->dev->addr_len); 1533 } 1534 1535 return ret; 1536} 1537 1538static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1539{ 1540 switch (BOND_MODE(bond)) { 1541 case BOND_MODE_ROUNDROBIN: 1542 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1543 case BOND_MODE_ACTIVEBACKUP: 1544 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1545 case BOND_MODE_BROADCAST: 1546 return NETDEV_LAG_TX_TYPE_BROADCAST; 1547 case BOND_MODE_XOR: 1548 case BOND_MODE_8023AD: 1549 return NETDEV_LAG_TX_TYPE_HASH; 1550 default: 1551 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1552 } 1553} 1554 1555static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1556 enum netdev_lag_tx_type type) 1557{ 1558 if (type != NETDEV_LAG_TX_TYPE_HASH) 1559 return NETDEV_LAG_HASH_NONE; 1560 1561 switch (bond->params.xmit_policy) { 1562 case BOND_XMIT_POLICY_LAYER2: 1563 return NETDEV_LAG_HASH_L2; 1564 case BOND_XMIT_POLICY_LAYER34: 1565 return NETDEV_LAG_HASH_L34; 1566 case BOND_XMIT_POLICY_LAYER23: 1567 return NETDEV_LAG_HASH_L23; 1568 case BOND_XMIT_POLICY_ENCAP23: 1569 return NETDEV_LAG_HASH_E23; 1570 case BOND_XMIT_POLICY_ENCAP34: 1571 return NETDEV_LAG_HASH_E34; 1572 default: 1573 return NETDEV_LAG_HASH_UNKNOWN; 1574 } 1575} 1576 1577static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1578 struct netlink_ext_ack *extack) 1579{ 1580 struct netdev_lag_upper_info lag_upper_info; 1581 enum netdev_lag_tx_type type; 1582 1583 type = bond_lag_tx_type(bond); 1584 lag_upper_info.tx_type = type; 1585 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1586 1587 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1588 &lag_upper_info, extack); 1589} 1590 1591static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1592{ 1593 netdev_upper_dev_unlink(slave->dev, bond->dev); 1594 slave->dev->flags &= ~IFF_SLAVE; 1595} 1596 1597static void slave_kobj_release(struct kobject *kobj) 1598{ 1599 struct slave *slave = to_slave(kobj); 1600 struct bonding *bond = bond_get_bond_by_slave(slave); 1601 1602 cancel_delayed_work_sync(&slave->notify_work); 1603 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1604 kfree(SLAVE_AD_INFO(slave)); 1605 1606 kfree(slave); 1607} 1608 1609static struct kobj_type slave_ktype = { 1610 .release = slave_kobj_release, 1611#ifdef CONFIG_SYSFS 1612 .sysfs_ops = &slave_sysfs_ops, 1613#endif 1614}; 1615 1616static int bond_kobj_init(struct slave *slave) 1617{ 1618 int err; 1619 1620 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1621 &(slave->dev->dev.kobj), "bonding_slave"); 1622 if (err) 1623 kobject_put(&slave->kobj); 1624 1625 return err; 1626} 1627 1628static struct slave *bond_alloc_slave(struct bonding *bond, 1629 struct net_device *slave_dev) 1630{ 1631 struct slave *slave = NULL; 1632 1633 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1634 if (!slave) 1635 return NULL; 1636 1637 slave->bond = bond; 1638 slave->dev = slave_dev; 1639 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1640 1641 if (bond_kobj_init(slave)) 1642 return NULL; 1643 1644 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1645 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1646 GFP_KERNEL); 1647 if (!SLAVE_AD_INFO(slave)) { 1648 kobject_put(&slave->kobj); 1649 return NULL; 1650 } 1651 } 1652 1653 return slave; 1654} 1655 1656static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1657{ 1658 info->bond_mode = BOND_MODE(bond); 1659 info->miimon = bond->params.miimon; 1660 info->num_slaves = bond->slave_cnt; 1661} 1662 1663static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1664{ 1665 strcpy(info->slave_name, slave->dev->name); 1666 info->link = slave->link; 1667 info->state = bond_slave_state(slave); 1668 info->link_failure_count = slave->link_failure_count; 1669} 1670 1671static void bond_netdev_notify_work(struct work_struct *_work) 1672{ 1673 struct slave *slave = container_of(_work, struct slave, 1674 notify_work.work); 1675 1676 if (rtnl_trylock()) { 1677 struct netdev_bonding_info binfo; 1678 1679 bond_fill_ifslave(slave, &binfo.slave); 1680 bond_fill_ifbond(slave->bond, &binfo.master); 1681 netdev_bonding_info_change(slave->dev, &binfo); 1682 rtnl_unlock(); 1683 } else { 1684 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1685 } 1686} 1687 1688void bond_queue_slave_event(struct slave *slave) 1689{ 1690 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1691} 1692 1693void bond_lower_state_changed(struct slave *slave) 1694{ 1695 struct netdev_lag_lower_state_info info; 1696 1697 info.link_up = slave->link == BOND_LINK_UP || 1698 slave->link == BOND_LINK_FAIL; 1699 info.tx_enabled = bond_is_active_slave(slave); 1700 netdev_lower_state_changed(slave->dev, &info); 1701} 1702 1703/* enslave device <slave> to bond device <master> */ 1704int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1705 struct netlink_ext_ack *extack) 1706{ 1707 struct bonding *bond = netdev_priv(bond_dev); 1708 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1709 struct slave *new_slave = NULL, *prev_slave; 1710 struct sockaddr_storage ss; 1711 int link_reporting; 1712 int res = 0, i; 1713 1714 if (!bond->params.use_carrier && 1715 slave_dev->ethtool_ops->get_link == NULL && 1716 slave_ops->ndo_do_ioctl == NULL) { 1717 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1718 } 1719 1720 /* already in-use? */ 1721 if (netdev_is_rx_handler_busy(slave_dev)) { 1722 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1723 slave_err(bond_dev, slave_dev, 1724 "Error: Device is in use and cannot be enslaved\n"); 1725 return -EBUSY; 1726 } 1727 1728 if (bond_dev == slave_dev) { 1729 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1730 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1731 return -EPERM; 1732 } 1733 1734 /* vlan challenged mutual exclusion */ 1735 /* no need to lock since we're protected by rtnl_lock */ 1736 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1737 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1738 if (vlan_uses_dev(bond_dev)) { 1739 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1740 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n"); 1741 return -EPERM; 1742 } else { 1743 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1744 } 1745 } else { 1746 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1747 } 1748 1749 if (slave_dev->features & NETIF_F_HW_ESP) 1750 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1751 1752 /* Old ifenslave binaries are no longer supported. These can 1753 * be identified with moderate accuracy by the state of the slave: 1754 * the current ifenslave will set the interface down prior to 1755 * enslaving it; the old ifenslave will not. 1756 */ 1757 if (slave_dev->flags & IFF_UP) { 1758 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1759 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n"); 1760 return -EPERM; 1761 } 1762 1763 /* set bonding device ether type by slave - bonding netdevices are 1764 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1765 * there is a need to override some of the type dependent attribs/funcs. 1766 * 1767 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1768 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1769 */ 1770 if (!bond_has_slaves(bond)) { 1771 if (bond_dev->type != slave_dev->type) { 1772 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1773 bond_dev->type, slave_dev->type); 1774 1775 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1776 bond_dev); 1777 res = notifier_to_errno(res); 1778 if (res) { 1779 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1780 return -EBUSY; 1781 } 1782 1783 /* Flush unicast and multicast addresses */ 1784 dev_uc_flush(bond_dev); 1785 dev_mc_flush(bond_dev); 1786 1787 if (slave_dev->type != ARPHRD_ETHER) 1788 bond_setup_by_slave(bond_dev, slave_dev); 1789 else { 1790 ether_setup(bond_dev); 1791 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1792 } 1793 1794 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1795 bond_dev); 1796 } 1797 } else if (bond_dev->type != slave_dev->type) { 1798 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1799 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n", 1800 slave_dev->type, bond_dev->type); 1801 return -EINVAL; 1802 } 1803 1804 if (slave_dev->type == ARPHRD_INFINIBAND && 1805 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1806 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1807 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n", 1808 slave_dev->type); 1809 res = -EOPNOTSUPP; 1810 goto err_undo_flags; 1811 } 1812 1813 if (!slave_ops->ndo_set_mac_address || 1814 slave_dev->type == ARPHRD_INFINIBAND) { 1815 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1816 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1817 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1818 if (!bond_has_slaves(bond)) { 1819 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1820 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1821 } else { 1822 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1823 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n"); 1824 res = -EOPNOTSUPP; 1825 goto err_undo_flags; 1826 } 1827 } 1828 } 1829 1830 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1831 1832 /* If this is the first slave, then we need to set the master's hardware 1833 * address to be the same as the slave's. 1834 */ 1835 if (!bond_has_slaves(bond) && 1836 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1837 res = bond_set_dev_addr(bond->dev, slave_dev); 1838 if (res) 1839 goto err_undo_flags; 1840 } 1841 1842 new_slave = bond_alloc_slave(bond, slave_dev); 1843 if (!new_slave) { 1844 res = -ENOMEM; 1845 goto err_undo_flags; 1846 } 1847 1848 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1849 * is set via sysfs or module option if desired. 1850 */ 1851 new_slave->queue_id = 0; 1852 1853 /* Save slave's original mtu and then set it to match the bond */ 1854 new_slave->original_mtu = slave_dev->mtu; 1855 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1856 if (res) { 1857 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1858 goto err_free; 1859 } 1860 1861 /* Save slave's original ("permanent") mac address for modes 1862 * that need it, and for restoring it upon release, and then 1863 * set it to the master's address 1864 */ 1865 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1866 slave_dev->addr_len); 1867 1868 if (!bond->params.fail_over_mac || 1869 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1870 /* Set slave to master's mac address. The application already 1871 * set the master's mac address to that of the first slave 1872 */ 1873 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1874 ss.ss_family = slave_dev->type; 1875 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1876 extack); 1877 if (res) { 1878 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1879 goto err_restore_mtu; 1880 } 1881 } 1882 1883 /* set slave flag before open to prevent IPv6 addrconf */ 1884 slave_dev->flags |= IFF_SLAVE; 1885 1886 /* open the slave since the application closed it */ 1887 res = dev_open(slave_dev, extack); 1888 if (res) { 1889 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1890 goto err_restore_mac; 1891 } 1892 1893 slave_dev->priv_flags |= IFF_BONDING; 1894 /* initialize slave stats */ 1895 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1896 1897 if (bond_is_lb(bond)) { 1898 /* bond_alb_init_slave() must be called before all other stages since 1899 * it might fail and we do not want to have to undo everything 1900 */ 1901 res = bond_alb_init_slave(bond, new_slave); 1902 if (res) 1903 goto err_close; 1904 } 1905 1906 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1907 if (res) { 1908 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1909 goto err_close; 1910 } 1911 1912 prev_slave = bond_last_slave(bond); 1913 1914 new_slave->delay = 0; 1915 new_slave->link_failure_count = 0; 1916 1917 if (bond_update_speed_duplex(new_slave) && 1918 bond_needs_speed_duplex(bond)) 1919 new_slave->link = BOND_LINK_DOWN; 1920 1921 new_slave->last_rx = jiffies - 1922 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1923 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1924 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1925 1926 if (bond->params.miimon && !bond->params.use_carrier) { 1927 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1928 1929 if ((link_reporting == -1) && !bond->params.arp_interval) { 1930 /* miimon is set but a bonded network driver 1931 * does not support ETHTOOL/MII and 1932 * arp_interval is not set. Note: if 1933 * use_carrier is enabled, we will never go 1934 * here (because netif_carrier is always 1935 * supported); thus, we don't need to change 1936 * the messages for netif_carrier. 1937 */ 1938 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n"); 1939 } else if (link_reporting == -1) { 1940 /* unable get link status using mii/ethtool */ 1941 slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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"); 1942 } 1943 } 1944 1945 /* check for initial state */ 1946 new_slave->link = BOND_LINK_NOCHANGE; 1947 if (bond->params.miimon) { 1948 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1949 if (bond->params.updelay) { 1950 bond_set_slave_link_state(new_slave, 1951 BOND_LINK_BACK, 1952 BOND_SLAVE_NOTIFY_NOW); 1953 new_slave->delay = bond->params.updelay; 1954 } else { 1955 bond_set_slave_link_state(new_slave, 1956 BOND_LINK_UP, 1957 BOND_SLAVE_NOTIFY_NOW); 1958 } 1959 } else { 1960 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1961 BOND_SLAVE_NOTIFY_NOW); 1962 } 1963 } else if (bond->params.arp_interval) { 1964 bond_set_slave_link_state(new_slave, 1965 (netif_carrier_ok(slave_dev) ? 1966 BOND_LINK_UP : BOND_LINK_DOWN), 1967 BOND_SLAVE_NOTIFY_NOW); 1968 } else { 1969 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1970 BOND_SLAVE_NOTIFY_NOW); 1971 } 1972 1973 if (new_slave->link != BOND_LINK_DOWN) 1974 new_slave->last_link_up = jiffies; 1975 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 1976 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1977 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1978 1979 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1980 /* if there is a primary slave, remember it */ 1981 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1982 rcu_assign_pointer(bond->primary_slave, new_slave); 1983 bond->force_primary = true; 1984 } 1985 } 1986 1987 switch (BOND_MODE(bond)) { 1988 case BOND_MODE_ACTIVEBACKUP: 1989 bond_set_slave_inactive_flags(new_slave, 1990 BOND_SLAVE_NOTIFY_NOW); 1991 break; 1992 case BOND_MODE_8023AD: 1993 /* in 802.3ad mode, the internal mechanism 1994 * will activate the slaves in the selected 1995 * aggregator 1996 */ 1997 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1998 /* if this is the first slave */ 1999 if (!prev_slave) { 2000 SLAVE_AD_INFO(new_slave)->id = 1; 2001 /* Initialize AD with the number of times that the AD timer is called in 1 second 2002 * can be called only after the mac address of the bond is set 2003 */ 2004 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 2005 } else { 2006 SLAVE_AD_INFO(new_slave)->id = 2007 SLAVE_AD_INFO(prev_slave)->id + 1; 2008 } 2009 2010 bond_3ad_bind_slave(new_slave); 2011 break; 2012 case BOND_MODE_TLB: 2013 case BOND_MODE_ALB: 2014 bond_set_active_slave(new_slave); 2015 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2016 break; 2017 default: 2018 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2019 2020 /* always active in trunk mode */ 2021 bond_set_active_slave(new_slave); 2022 2023 /* In trunking mode there is little meaning to curr_active_slave 2024 * anyway (it holds no special properties of the bond device), 2025 * so we can change it without calling change_active_interface() 2026 */ 2027 if (!rcu_access_pointer(bond->curr_active_slave) && 2028 new_slave->link == BOND_LINK_UP) 2029 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2030 2031 break; 2032 } /* switch(bond_mode) */ 2033 2034#ifdef CONFIG_NET_POLL_CONTROLLER 2035 if (bond->dev->npinfo) { 2036 if (slave_enable_netpoll(new_slave)) { 2037 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2038 res = -EBUSY; 2039 goto err_detach; 2040 } 2041 } 2042#endif 2043 2044 if (!(bond_dev->features & NETIF_F_LRO)) 2045 dev_disable_lro(slave_dev); 2046 2047 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2048 new_slave); 2049 if (res) { 2050 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2051 goto err_detach; 2052 } 2053 2054 res = bond_master_upper_dev_link(bond, new_slave, extack); 2055 if (res) { 2056 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2057 goto err_unregister; 2058 } 2059 2060 res = bond_sysfs_slave_add(new_slave); 2061 if (res) { 2062 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2063 goto err_upper_unlink; 2064 } 2065 2066 /* If the mode uses primary, then the following is handled by 2067 * bond_change_active_slave(). 2068 */ 2069 if (!bond_uses_primary(bond)) { 2070 /* set promiscuity level to new slave */ 2071 if (bond_dev->flags & IFF_PROMISC) { 2072 res = dev_set_promiscuity(slave_dev, 1); 2073 if (res) 2074 goto err_sysfs_del; 2075 } 2076 2077 /* set allmulti level to new slave */ 2078 if (bond_dev->flags & IFF_ALLMULTI) { 2079 res = dev_set_allmulti(slave_dev, 1); 2080 if (res) { 2081 if (bond_dev->flags & IFF_PROMISC) 2082 dev_set_promiscuity(slave_dev, -1); 2083 goto err_sysfs_del; 2084 } 2085 } 2086 2087 if (bond_dev->flags & IFF_UP) { 2088 netif_addr_lock_bh(bond_dev); 2089 dev_mc_sync_multiple(slave_dev, bond_dev); 2090 dev_uc_sync_multiple(slave_dev, bond_dev); 2091 netif_addr_unlock_bh(bond_dev); 2092 2093 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2094 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2095 } 2096 } 2097 2098 bond->slave_cnt++; 2099 bond_compute_features(bond); 2100 bond_set_carrier(bond); 2101 2102 if (bond_uses_primary(bond)) { 2103 block_netpoll_tx(); 2104 bond_select_active_slave(bond); 2105 unblock_netpoll_tx(); 2106 } 2107 2108 if (bond_mode_can_use_xmit_hash(bond)) 2109 bond_update_slave_arr(bond, NULL); 2110 2111 2112 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2113 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2114 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2115 2116 /* enslave is successful */ 2117 bond_queue_slave_event(new_slave); 2118 return 0; 2119 2120/* Undo stages on error */ 2121err_sysfs_del: 2122 bond_sysfs_slave_del(new_slave); 2123 2124err_upper_unlink: 2125 bond_upper_dev_unlink(bond, new_slave); 2126 2127err_unregister: 2128 netdev_rx_handler_unregister(slave_dev); 2129 2130err_detach: 2131 vlan_vids_del_by_dev(slave_dev, bond_dev); 2132 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2133 RCU_INIT_POINTER(bond->primary_slave, NULL); 2134 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2135 block_netpoll_tx(); 2136 bond_change_active_slave(bond, NULL); 2137 bond_select_active_slave(bond); 2138 unblock_netpoll_tx(); 2139 } 2140 /* either primary_slave or curr_active_slave might've changed */ 2141 synchronize_rcu(); 2142 slave_disable_netpoll(new_slave); 2143 2144err_close: 2145 if (!netif_is_bond_master(slave_dev)) 2146 slave_dev->priv_flags &= ~IFF_BONDING; 2147 dev_close(slave_dev); 2148 2149err_restore_mac: 2150 slave_dev->flags &= ~IFF_SLAVE; 2151 if (!bond->params.fail_over_mac || 2152 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2153 /* XXX TODO - fom follow mode needs to change master's 2154 * MAC if this slave's MAC is in use by the bond, or at 2155 * least print a warning. 2156 */ 2157 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2158 new_slave->dev->addr_len); 2159 ss.ss_family = slave_dev->type; 2160 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2161 } 2162 2163err_restore_mtu: 2164 dev_set_mtu(slave_dev, new_slave->original_mtu); 2165 2166err_free: 2167 kobject_put(&new_slave->kobj); 2168 2169err_undo_flags: 2170 /* Enslave of first slave has failed and we need to fix master's mac */ 2171 if (!bond_has_slaves(bond)) { 2172 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2173 slave_dev->dev_addr)) 2174 eth_hw_addr_random(bond_dev); 2175 if (bond_dev->type != ARPHRD_ETHER) { 2176 dev_close(bond_dev); 2177 ether_setup(bond_dev); 2178 bond_dev->flags |= IFF_MASTER; 2179 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2180 } 2181 } 2182 2183 return res; 2184} 2185 2186/* Try to release the slave device <slave> from the bond device <master> 2187 * It is legal to access curr_active_slave without a lock because all the function 2188 * is RTNL-locked. If "all" is true it means that the function is being called 2189 * while destroying a bond interface and all slaves are being released. 2190 * 2191 * The rules for slave state should be: 2192 * for Active/Backup: 2193 * Active stays on all backups go down 2194 * for Bonded connections: 2195 * The first up interface should be left on and all others downed. 2196 */ 2197static int __bond_release_one(struct net_device *bond_dev, 2198 struct net_device *slave_dev, 2199 bool all, bool unregister) 2200{ 2201 struct bonding *bond = netdev_priv(bond_dev); 2202 struct slave *slave, *oldcurrent; 2203 struct sockaddr_storage ss; 2204 int old_flags = bond_dev->flags; 2205 netdev_features_t old_features = bond_dev->features; 2206 2207 /* slave is not a slave or master is not master of this slave */ 2208 if (!(slave_dev->flags & IFF_SLAVE) || 2209 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2210 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2211 return -EINVAL; 2212 } 2213 2214 block_netpoll_tx(); 2215 2216 slave = bond_get_slave_by_dev(bond, slave_dev); 2217 if (!slave) { 2218 /* not a slave of this bond */ 2219 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2220 unblock_netpoll_tx(); 2221 return -EINVAL; 2222 } 2223 2224 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2225 2226 bond_sysfs_slave_del(slave); 2227 2228 /* recompute stats just before removing the slave */ 2229 bond_get_stats(bond->dev, &bond->bond_stats); 2230 2231 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2232 * for this slave anymore. 2233 */ 2234 netdev_rx_handler_unregister(slave_dev); 2235 2236 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2237 bond_3ad_unbind_slave(slave); 2238 2239 bond_upper_dev_unlink(bond, slave); 2240 2241 if (bond_mode_can_use_xmit_hash(bond)) 2242 bond_update_slave_arr(bond, slave); 2243 2244 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2245 bond_is_active_slave(slave) ? "active" : "backup"); 2246 2247 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2248 2249 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2250 2251 if (!all && (!bond->params.fail_over_mac || 2252 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2253 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2254 bond_has_slaves(bond)) 2255 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n", 2256 slave->perm_hwaddr); 2257 } 2258 2259 if (rtnl_dereference(bond->primary_slave) == slave) 2260 RCU_INIT_POINTER(bond->primary_slave, NULL); 2261 2262 if (oldcurrent == slave) 2263 bond_change_active_slave(bond, NULL); 2264 2265 if (bond_is_lb(bond)) { 2266 /* Must be called only after the slave has been 2267 * detached from the list and the curr_active_slave 2268 * has been cleared (if our_slave == old_current), 2269 * but before a new active slave is selected. 2270 */ 2271 bond_alb_deinit_slave(bond, slave); 2272 } 2273 2274 if (all) { 2275 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2276 } else if (oldcurrent == slave) { 2277 /* Note that we hold RTNL over this sequence, so there 2278 * is no concern that another slave add/remove event 2279 * will interfere. 2280 */ 2281 bond_select_active_slave(bond); 2282 } 2283 2284 bond_set_carrier(bond); 2285 if (!bond_has_slaves(bond)) 2286 eth_hw_addr_random(bond_dev); 2287 2288 unblock_netpoll_tx(); 2289 synchronize_rcu(); 2290 bond->slave_cnt--; 2291 2292 if (!bond_has_slaves(bond)) { 2293 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2294 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2295 } 2296 2297 bond_compute_features(bond); 2298 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2299 (old_features & NETIF_F_VLAN_CHALLENGED)) 2300 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2301 2302 vlan_vids_del_by_dev(slave_dev, bond_dev); 2303 2304 /* If the mode uses primary, then this case was handled above by 2305 * bond_change_active_slave(..., NULL) 2306 */ 2307 if (!bond_uses_primary(bond)) { 2308 /* unset promiscuity level from slave 2309 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2310 * of the IFF_PROMISC flag in the bond_dev, but we need the 2311 * value of that flag before that change, as that was the value 2312 * when this slave was attached, so we cache at the start of the 2313 * function and use it here. Same goes for ALLMULTI below 2314 */ 2315 if (old_flags & IFF_PROMISC) 2316 dev_set_promiscuity(slave_dev, -1); 2317 2318 /* unset allmulti level from slave */ 2319 if (old_flags & IFF_ALLMULTI) 2320 dev_set_allmulti(slave_dev, -1); 2321 2322 if (old_flags & IFF_UP) 2323 bond_hw_addr_flush(bond_dev, slave_dev); 2324 } 2325 2326 slave_disable_netpoll(slave); 2327 2328 /* close slave before restoring its mac address */ 2329 dev_close(slave_dev); 2330 2331 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2332 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2333 /* restore original ("permanent") mac address */ 2334 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2335 slave->dev->addr_len); 2336 ss.ss_family = slave_dev->type; 2337 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2338 } 2339 2340 if (unregister) 2341 __dev_set_mtu(slave_dev, slave->original_mtu); 2342 else 2343 dev_set_mtu(slave_dev, slave->original_mtu); 2344 2345 if (!netif_is_bond_master(slave_dev)) 2346 slave_dev->priv_flags &= ~IFF_BONDING; 2347 2348 kobject_put(&slave->kobj); 2349 2350 return 0; 2351} 2352 2353/* A wrapper used because of ndo_del_link */ 2354int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2355{ 2356 return __bond_release_one(bond_dev, slave_dev, false, false); 2357} 2358 2359/* First release a slave and then destroy the bond if no more slaves are left. 2360 * Must be under rtnl_lock when this function is called. 2361 */ 2362static int bond_release_and_destroy(struct net_device *bond_dev, 2363 struct net_device *slave_dev) 2364{ 2365 struct bonding *bond = netdev_priv(bond_dev); 2366 int ret; 2367 2368 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2369 if (ret == 0 && !bond_has_slaves(bond) && 2370 bond_dev->reg_state != NETREG_UNREGISTERING) { 2371 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2372 netdev_info(bond_dev, "Destroying bond\n"); 2373 bond_remove_proc_entry(bond); 2374 unregister_netdevice(bond_dev); 2375 } 2376 return ret; 2377} 2378 2379static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2380{ 2381 struct bonding *bond = netdev_priv(bond_dev); 2382 bond_fill_ifbond(bond, info); 2383} 2384 2385static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2386{ 2387 struct bonding *bond = netdev_priv(bond_dev); 2388 struct list_head *iter; 2389 int i = 0, res = -ENODEV; 2390 struct slave *slave; 2391 2392 bond_for_each_slave(bond, slave, iter) { 2393 if (i++ == (int)info->slave_id) { 2394 res = 0; 2395 bond_fill_ifslave(slave, info); 2396 break; 2397 } 2398 } 2399 2400 return res; 2401} 2402 2403/*-------------------------------- Monitoring -------------------------------*/ 2404 2405/* called with rcu_read_lock() */ 2406static int bond_miimon_inspect(struct bonding *bond) 2407{ 2408 bool ignore_updelay = false; 2409 int link_state, commit = 0; 2410 struct list_head *iter; 2411 struct slave *slave; 2412 2413 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2414 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2415 } else { 2416 struct bond_up_slave *usable_slaves; 2417 2418 usable_slaves = rcu_dereference(bond->usable_slaves); 2419 2420 if (usable_slaves && usable_slaves->count == 0) 2421 ignore_updelay = true; 2422 } 2423 2424 bond_for_each_slave_rcu(bond, slave, iter) { 2425 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2426 2427 link_state = bond_check_dev_link(bond, slave->dev, 0); 2428 2429 switch (slave->link) { 2430 case BOND_LINK_UP: 2431 if (link_state) 2432 continue; 2433 2434 bond_propose_link_state(slave, BOND_LINK_FAIL); 2435 commit++; 2436 slave->delay = bond->params.downdelay; 2437 if (slave->delay) { 2438 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2439 (BOND_MODE(bond) == 2440 BOND_MODE_ACTIVEBACKUP) ? 2441 (bond_is_active_slave(slave) ? 2442 "active " : "backup ") : "", 2443 bond->params.downdelay * bond->params.miimon); 2444 } 2445 fallthrough; 2446 case BOND_LINK_FAIL: 2447 if (link_state) { 2448 /* recovered before downdelay expired */ 2449 bond_propose_link_state(slave, BOND_LINK_UP); 2450 slave->last_link_up = jiffies; 2451 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2452 (bond->params.downdelay - slave->delay) * 2453 bond->params.miimon); 2454 commit++; 2455 continue; 2456 } 2457 2458 if (slave->delay <= 0) { 2459 bond_propose_link_state(slave, BOND_LINK_DOWN); 2460 commit++; 2461 continue; 2462 } 2463 2464 slave->delay--; 2465 break; 2466 2467 case BOND_LINK_DOWN: 2468 if (!link_state) 2469 continue; 2470 2471 bond_propose_link_state(slave, BOND_LINK_BACK); 2472 commit++; 2473 slave->delay = bond->params.updelay; 2474 2475 if (slave->delay) { 2476 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2477 ignore_updelay ? 0 : 2478 bond->params.updelay * 2479 bond->params.miimon); 2480 } 2481 fallthrough; 2482 case BOND_LINK_BACK: 2483 if (!link_state) { 2484 bond_propose_link_state(slave, BOND_LINK_DOWN); 2485 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2486 (bond->params.updelay - slave->delay) * 2487 bond->params.miimon); 2488 commit++; 2489 continue; 2490 } 2491 2492 if (ignore_updelay) 2493 slave->delay = 0; 2494 2495 if (slave->delay <= 0) { 2496 bond_propose_link_state(slave, BOND_LINK_UP); 2497 commit++; 2498 ignore_updelay = false; 2499 continue; 2500 } 2501 2502 slave->delay--; 2503 break; 2504 } 2505 } 2506 2507 return commit; 2508} 2509 2510static void bond_miimon_link_change(struct bonding *bond, 2511 struct slave *slave, 2512 char link) 2513{ 2514 switch (BOND_MODE(bond)) { 2515 case BOND_MODE_8023AD: 2516 bond_3ad_handle_link_change(slave, link); 2517 break; 2518 case BOND_MODE_TLB: 2519 case BOND_MODE_ALB: 2520 bond_alb_handle_link_change(bond, slave, link); 2521 break; 2522 case BOND_MODE_XOR: 2523 bond_update_slave_arr(bond, NULL); 2524 break; 2525 } 2526} 2527 2528static void bond_miimon_commit(struct bonding *bond) 2529{ 2530 struct list_head *iter; 2531 struct slave *slave, *primary; 2532 2533 bond_for_each_slave(bond, slave, iter) { 2534 switch (slave->link_new_state) { 2535 case BOND_LINK_NOCHANGE: 2536 /* For 802.3ad mode, check current slave speed and 2537 * duplex again in case its port was disabled after 2538 * invalid speed/duplex reporting but recovered before 2539 * link monitoring could make a decision on the actual 2540 * link status 2541 */ 2542 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2543 slave->link == BOND_LINK_UP) 2544 bond_3ad_adapter_speed_duplex_changed(slave); 2545 continue; 2546 2547 case BOND_LINK_UP: 2548 if (bond_update_speed_duplex(slave) && 2549 bond_needs_speed_duplex(bond)) { 2550 slave->link = BOND_LINK_DOWN; 2551 if (net_ratelimit()) 2552 slave_warn(bond->dev, slave->dev, 2553 "failed to get link speed/duplex\n"); 2554 continue; 2555 } 2556 bond_set_slave_link_state(slave, BOND_LINK_UP, 2557 BOND_SLAVE_NOTIFY_NOW); 2558 slave->last_link_up = jiffies; 2559 2560 primary = rtnl_dereference(bond->primary_slave); 2561 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2562 /* prevent it from being the active one */ 2563 bond_set_backup_slave(slave); 2564 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2565 /* make it immediately active */ 2566 bond_set_active_slave(slave); 2567 } 2568 2569 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2570 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2571 slave->duplex ? "full" : "half"); 2572 2573 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2574 2575 if (!bond->curr_active_slave || slave == primary) 2576 goto do_failover; 2577 2578 continue; 2579 2580 case BOND_LINK_DOWN: 2581 if (slave->link_failure_count < UINT_MAX) 2582 slave->link_failure_count++; 2583 2584 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2585 BOND_SLAVE_NOTIFY_NOW); 2586 2587 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2588 BOND_MODE(bond) == BOND_MODE_8023AD) 2589 bond_set_slave_inactive_flags(slave, 2590 BOND_SLAVE_NOTIFY_NOW); 2591 2592 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2593 2594 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2595 2596 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2597 goto do_failover; 2598 2599 continue; 2600 2601 default: 2602 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2603 slave->link_new_state); 2604 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2605 2606 continue; 2607 } 2608 2609do_failover: 2610 block_netpoll_tx(); 2611 bond_select_active_slave(bond); 2612 unblock_netpoll_tx(); 2613 } 2614 2615 bond_set_carrier(bond); 2616} 2617 2618/* bond_mii_monitor 2619 * 2620 * Really a wrapper that splits the mii monitor into two phases: an 2621 * inspection, then (if inspection indicates something needs to be done) 2622 * an acquisition of appropriate locks followed by a commit phase to 2623 * implement whatever link state changes are indicated. 2624 */ 2625static void bond_mii_monitor(struct work_struct *work) 2626{ 2627 struct bonding *bond = container_of(work, struct bonding, 2628 mii_work.work); 2629 bool should_notify_peers = false; 2630 bool commit; 2631 unsigned long delay; 2632 struct slave *slave; 2633 struct list_head *iter; 2634 2635 delay = msecs_to_jiffies(bond->params.miimon); 2636 2637 if (!bond_has_slaves(bond)) 2638 goto re_arm; 2639 2640 rcu_read_lock(); 2641 should_notify_peers = bond_should_notify_peers(bond); 2642 commit = !!bond_miimon_inspect(bond); 2643 if (bond->send_peer_notif) { 2644 rcu_read_unlock(); 2645 if (rtnl_trylock()) { 2646 bond->send_peer_notif--; 2647 rtnl_unlock(); 2648 } 2649 } else { 2650 rcu_read_unlock(); 2651 } 2652 2653 if (commit) { 2654 /* Race avoidance with bond_close cancel of workqueue */ 2655 if (!rtnl_trylock()) { 2656 delay = 1; 2657 should_notify_peers = false; 2658 goto re_arm; 2659 } 2660 2661 bond_for_each_slave(bond, slave, iter) { 2662 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2663 } 2664 bond_miimon_commit(bond); 2665 2666 rtnl_unlock(); /* might sleep, hold no other locks */ 2667 } 2668 2669re_arm: 2670 if (bond->params.miimon) 2671 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2672 2673 if (should_notify_peers) { 2674 if (!rtnl_trylock()) 2675 return; 2676 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2677 rtnl_unlock(); 2678 } 2679} 2680 2681static int bond_upper_dev_walk(struct net_device *upper, 2682 struct netdev_nested_priv *priv) 2683{ 2684 __be32 ip = *(__be32 *)priv->data; 2685 2686 return ip == bond_confirm_addr(upper, 0, ip); 2687} 2688 2689static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2690{ 2691 struct netdev_nested_priv priv = { 2692 .data = (void *)&ip, 2693 }; 2694 bool ret = false; 2695 2696 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2697 return true; 2698 2699 rcu_read_lock(); 2700 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2701 ret = true; 2702 rcu_read_unlock(); 2703 2704 return ret; 2705} 2706 2707/* We go to the (large) trouble of VLAN tagging ARP frames because 2708 * switches in VLAN mode (especially if ports are configured as 2709 * "native" to a VLAN) might not pass non-tagged frames. 2710 */ 2711static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2712 __be32 src_ip, struct bond_vlan_tag *tags) 2713{ 2714 struct sk_buff *skb; 2715 struct bond_vlan_tag *outer_tag = tags; 2716 struct net_device *slave_dev = slave->dev; 2717 struct net_device *bond_dev = slave->bond->dev; 2718 2719 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2720 arp_op, &dest_ip, &src_ip); 2721 2722 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2723 NULL, slave_dev->dev_addr, NULL); 2724 2725 if (!skb) { 2726 net_err_ratelimited("ARP packet allocation failed\n"); 2727 return; 2728 } 2729 2730 if (!tags || tags->vlan_proto == VLAN_N_VID) 2731 goto xmit; 2732 2733 tags++; 2734 2735 /* Go through all the tags backwards and add them to the packet */ 2736 while (tags->vlan_proto != VLAN_N_VID) { 2737 if (!tags->vlan_id) { 2738 tags++; 2739 continue; 2740 } 2741 2742 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2743 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2744 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2745 tags->vlan_id); 2746 if (!skb) { 2747 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2748 return; 2749 } 2750 2751 tags++; 2752 } 2753 /* Set the outer tag */ 2754 if (outer_tag->vlan_id) { 2755 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2756 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2757 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2758 outer_tag->vlan_id); 2759 } 2760 2761xmit: 2762 arp_xmit(skb); 2763} 2764 2765/* Validate the device path between the @start_dev and the @end_dev. 2766 * The path is valid if the @end_dev is reachable through device 2767 * stacking. 2768 * When the path is validated, collect any vlan information in the 2769 * path. 2770 */ 2771struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2772 struct net_device *end_dev, 2773 int level) 2774{ 2775 struct bond_vlan_tag *tags; 2776 struct net_device *upper; 2777 struct list_head *iter; 2778 2779 if (start_dev == end_dev) { 2780 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2781 if (!tags) 2782 return ERR_PTR(-ENOMEM); 2783 tags[level].vlan_proto = VLAN_N_VID; 2784 return tags; 2785 } 2786 2787 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2788 tags = bond_verify_device_path(upper, end_dev, level + 1); 2789 if (IS_ERR_OR_NULL(tags)) { 2790 if (IS_ERR(tags)) 2791 return tags; 2792 continue; 2793 } 2794 if (is_vlan_dev(upper)) { 2795 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2796 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2797 } 2798 2799 return tags; 2800 } 2801 2802 return NULL; 2803} 2804 2805static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2806{ 2807 struct rtable *rt; 2808 struct bond_vlan_tag *tags; 2809 __be32 *targets = bond->params.arp_targets, addr; 2810 int i; 2811 2812 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2813 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2814 __func__, &targets[i]); 2815 tags = NULL; 2816 2817 /* Find out through which dev should the packet go */ 2818 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2819 RTO_ONLINK, 0); 2820 if (IS_ERR(rt)) { 2821 /* there's no route to target - try to send arp 2822 * probe to generate any traffic (arp_validate=0) 2823 */ 2824 if (bond->params.arp_validate) 2825 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2826 bond->dev->name, 2827 &targets[i]); 2828 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2829 0, tags); 2830 continue; 2831 } 2832 2833 /* bond device itself */ 2834 if (rt->dst.dev == bond->dev) 2835 goto found; 2836 2837 rcu_read_lock(); 2838 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2839 rcu_read_unlock(); 2840 2841 if (!IS_ERR_OR_NULL(tags)) 2842 goto found; 2843 2844 /* Not our device - skip */ 2845 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2846 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2847 2848 ip_rt_put(rt); 2849 continue; 2850 2851found: 2852 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2853 ip_rt_put(rt); 2854 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2855 kfree(tags); 2856 } 2857} 2858 2859static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2860{ 2861 int i; 2862 2863 if (!sip || !bond_has_this_ip(bond, tip)) { 2864 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2865 __func__, &sip, &tip); 2866 return; 2867 } 2868 2869 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2870 if (i == -1) { 2871 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2872 __func__, &sip); 2873 return; 2874 } 2875 slave->last_rx = jiffies; 2876 slave->target_last_arp_rx[i] = jiffies; 2877} 2878 2879int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2880 struct slave *slave) 2881{ 2882 struct arphdr *arp = (struct arphdr *)skb->data; 2883 struct slave *curr_active_slave, *curr_arp_slave; 2884 unsigned char *arp_ptr; 2885 __be32 sip, tip; 2886 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2887 unsigned int alen; 2888 2889 if (!slave_do_arp_validate(bond, slave)) { 2890 if ((slave_do_arp_validate_only(bond) && is_arp) || 2891 !slave_do_arp_validate_only(bond)) 2892 slave->last_rx = jiffies; 2893 return RX_HANDLER_ANOTHER; 2894 } else if (!is_arp) { 2895 return RX_HANDLER_ANOTHER; 2896 } 2897 2898 alen = arp_hdr_len(bond->dev); 2899 2900 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 2901 __func__, skb->dev->name); 2902 2903 if (alen > skb_headlen(skb)) { 2904 arp = kmalloc(alen, GFP_ATOMIC); 2905 if (!arp) 2906 goto out_unlock; 2907 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2908 goto out_unlock; 2909 } 2910 2911 if (arp->ar_hln != bond->dev->addr_len || 2912 skb->pkt_type == PACKET_OTHERHOST || 2913 skb->pkt_type == PACKET_LOOPBACK || 2914 arp->ar_hrd != htons(ARPHRD_ETHER) || 2915 arp->ar_pro != htons(ETH_P_IP) || 2916 arp->ar_pln != 4) 2917 goto out_unlock; 2918 2919 arp_ptr = (unsigned char *)(arp + 1); 2920 arp_ptr += bond->dev->addr_len; 2921 memcpy(&sip, arp_ptr, 4); 2922 arp_ptr += 4 + bond->dev->addr_len; 2923 memcpy(&tip, arp_ptr, 4); 2924 2925 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2926 __func__, slave->dev->name, bond_slave_state(slave), 2927 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2928 &sip, &tip); 2929 2930 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2931 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2932 2933 /* We 'trust' the received ARP enough to validate it if: 2934 * 2935 * (a) the slave receiving the ARP is active (which includes the 2936 * current ARP slave, if any), or 2937 * 2938 * (b) the receiving slave isn't active, but there is a currently 2939 * active slave and it received valid arp reply(s) after it became 2940 * the currently active slave, or 2941 * 2942 * (c) there is an ARP slave that sent an ARP during the prior ARP 2943 * interval, and we receive an ARP reply on any slave. We accept 2944 * these because switch FDB update delays may deliver the ARP 2945 * reply to a slave other than the sender of the ARP request. 2946 * 2947 * Note: for (b), backup slaves are receiving the broadcast ARP 2948 * request, not a reply. This request passes from the sending 2949 * slave through the L2 switch(es) to the receiving slave. Since 2950 * this is checking the request, sip/tip are swapped for 2951 * validation. 2952 * 2953 * This is done to avoid endless looping when we can't reach the 2954 * arp_ip_target and fool ourselves with our own arp requests. 2955 */ 2956 if (bond_is_active_slave(slave)) 2957 bond_validate_arp(bond, slave, sip, tip); 2958 else if (curr_active_slave && 2959 time_after(slave_last_rx(bond, curr_active_slave), 2960 curr_active_slave->last_link_up)) 2961 bond_validate_arp(bond, slave, tip, sip); 2962 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2963 bond_time_in_interval(bond, 2964 dev_trans_start(curr_arp_slave->dev), 1)) 2965 bond_validate_arp(bond, slave, sip, tip); 2966 2967out_unlock: 2968 if (arp != (struct arphdr *)skb->data) 2969 kfree(arp); 2970 return RX_HANDLER_ANOTHER; 2971} 2972 2973/* function to verify if we're in the arp_interval timeslice, returns true if 2974 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2975 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2976 */ 2977static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2978 int mod) 2979{ 2980 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2981 2982 return time_in_range(jiffies, 2983 last_act - delta_in_ticks, 2984 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2985} 2986 2987/* This function is called regularly to monitor each slave's link 2988 * ensuring that traffic is being sent and received when arp monitoring 2989 * is used in load-balancing mode. if the adapter has been dormant, then an 2990 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2991 * arp monitoring in active backup mode. 2992 */ 2993static void bond_loadbalance_arp_mon(struct bonding *bond) 2994{ 2995 struct slave *slave, *oldcurrent; 2996 struct list_head *iter; 2997 int do_failover = 0, slave_state_changed = 0; 2998 2999 if (!bond_has_slaves(bond)) 3000 goto re_arm; 3001 3002 rcu_read_lock(); 3003 3004 oldcurrent = rcu_dereference(bond->curr_active_slave); 3005 /* see if any of the previous devices are up now (i.e. they have 3006 * xmt and rcv traffic). the curr_active_slave does not come into 3007 * the picture unless it is null. also, slave->last_link_up is not 3008 * needed here because we send an arp on each slave and give a slave 3009 * as long as it needs to get the tx/rx within the delta. 3010 * TODO: what about up/down delay in arp mode? it wasn't here before 3011 * so it can wait 3012 */ 3013 bond_for_each_slave_rcu(bond, slave, iter) { 3014 unsigned long trans_start = dev_trans_start(slave->dev); 3015 3016 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3017 3018 if (slave->link != BOND_LINK_UP) { 3019 if (bond_time_in_interval(bond, trans_start, 1) && 3020 bond_time_in_interval(bond, slave->last_rx, 1)) { 3021 3022 bond_propose_link_state(slave, BOND_LINK_UP); 3023 slave_state_changed = 1; 3024 3025 /* primary_slave has no meaning in round-robin 3026 * mode. the window of a slave being up and 3027 * curr_active_slave being null after enslaving 3028 * is closed. 3029 */ 3030 if (!oldcurrent) { 3031 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3032 do_failover = 1; 3033 } else { 3034 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3035 } 3036 } 3037 } else { 3038 /* slave->link == BOND_LINK_UP */ 3039 3040 /* not all switches will respond to an arp request 3041 * when the source ip is 0, so don't take the link down 3042 * if we don't know our ip yet 3043 */ 3044 if (!bond_time_in_interval(bond, trans_start, 2) || 3045 !bond_time_in_interval(bond, slave->last_rx, 2)) { 3046 3047 bond_propose_link_state(slave, BOND_LINK_DOWN); 3048 slave_state_changed = 1; 3049 3050 if (slave->link_failure_count < UINT_MAX) 3051 slave->link_failure_count++; 3052 3053 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3054 3055 if (slave == oldcurrent) 3056 do_failover = 1; 3057 } 3058 } 3059 3060 /* note: if switch is in round-robin mode, all links 3061 * must tx arp to ensure all links rx an arp - otherwise 3062 * links may oscillate or not come up at all; if switch is 3063 * in something like xor mode, there is nothing we can 3064 * do - all replies will be rx'ed on same link causing slaves 3065 * to be unstable during low/no traffic periods 3066 */ 3067 if (bond_slave_is_up(slave)) 3068 bond_arp_send_all(bond, slave); 3069 } 3070 3071 rcu_read_unlock(); 3072 3073 if (do_failover || slave_state_changed) { 3074 if (!rtnl_trylock()) 3075 goto re_arm; 3076 3077 bond_for_each_slave(bond, slave, iter) { 3078 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3079 slave->link = slave->link_new_state; 3080 } 3081 3082 if (slave_state_changed) { 3083 bond_slave_state_change(bond); 3084 if (BOND_MODE(bond) == BOND_MODE_XOR) 3085 bond_update_slave_arr(bond, NULL); 3086 } 3087 if (do_failover) { 3088 block_netpoll_tx(); 3089 bond_select_active_slave(bond); 3090 unblock_netpoll_tx(); 3091 } 3092 rtnl_unlock(); 3093 } 3094 3095re_arm: 3096 if (bond->params.arp_interval) 3097 queue_delayed_work(bond->wq, &bond->arp_work, 3098 msecs_to_jiffies(bond->params.arp_interval)); 3099} 3100 3101/* Called to inspect slaves for active-backup mode ARP monitor link state 3102 * changes. Sets proposed link state in slaves to specify what action 3103 * should take place for the slave. Returns 0 if no changes are found, >0 3104 * if changes to link states must be committed. 3105 * 3106 * Called with rcu_read_lock held. 3107 */ 3108static int bond_ab_arp_inspect(struct bonding *bond) 3109{ 3110 unsigned long trans_start, last_rx; 3111 struct list_head *iter; 3112 struct slave *slave; 3113 int commit = 0; 3114 3115 bond_for_each_slave_rcu(bond, slave, iter) { 3116 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3117 last_rx = slave_last_rx(bond, slave); 3118 3119 if (slave->link != BOND_LINK_UP) { 3120 if (bond_time_in_interval(bond, last_rx, 1)) { 3121 bond_propose_link_state(slave, BOND_LINK_UP); 3122 commit++; 3123 } else if (slave->link == BOND_LINK_BACK) { 3124 bond_propose_link_state(slave, BOND_LINK_FAIL); 3125 commit++; 3126 } 3127 continue; 3128 } 3129 3130 /* Give slaves 2*delta after being enslaved or made 3131 * active. This avoids bouncing, as the last receive 3132 * times need a full ARP monitor cycle to be updated. 3133 */ 3134 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3135 continue; 3136 3137 /* Backup slave is down if: 3138 * - No current_arp_slave AND 3139 * - more than 3*delta since last receive AND 3140 * - the bond has an IP address 3141 * 3142 * Note: a non-null current_arp_slave indicates 3143 * the curr_active_slave went down and we are 3144 * searching for a new one; under this condition 3145 * we only take the curr_active_slave down - this 3146 * gives each slave a chance to tx/rx traffic 3147 * before being taken out 3148 */ 3149 if (!bond_is_active_slave(slave) && 3150 !rcu_access_pointer(bond->current_arp_slave) && 3151 !bond_time_in_interval(bond, last_rx, 3)) { 3152 bond_propose_link_state(slave, BOND_LINK_DOWN); 3153 commit++; 3154 } 3155 3156 /* Active slave is down if: 3157 * - more than 2*delta since transmitting OR 3158 * - (more than 2*delta since receive AND 3159 * the bond has an IP address) 3160 */ 3161 trans_start = dev_trans_start(slave->dev); 3162 if (bond_is_active_slave(slave) && 3163 (!bond_time_in_interval(bond, trans_start, 2) || 3164 !bond_time_in_interval(bond, last_rx, 2))) { 3165 bond_propose_link_state(slave, BOND_LINK_DOWN); 3166 commit++; 3167 } 3168 } 3169 3170 return commit; 3171} 3172 3173/* Called to commit link state changes noted by inspection step of 3174 * active-backup mode ARP monitor. 3175 * 3176 * Called with RTNL hold. 3177 */ 3178static void bond_ab_arp_commit(struct bonding *bond) 3179{ 3180 unsigned long trans_start; 3181 struct list_head *iter; 3182 struct slave *slave; 3183 3184 bond_for_each_slave(bond, slave, iter) { 3185 switch (slave->link_new_state) { 3186 case BOND_LINK_NOCHANGE: 3187 continue; 3188 3189 case BOND_LINK_UP: 3190 trans_start = dev_trans_start(slave->dev); 3191 if (rtnl_dereference(bond->curr_active_slave) != slave || 3192 (!rtnl_dereference(bond->curr_active_slave) && 3193 bond_time_in_interval(bond, trans_start, 1))) { 3194 struct slave *current_arp_slave; 3195 3196 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3197 bond_set_slave_link_state(slave, BOND_LINK_UP, 3198 BOND_SLAVE_NOTIFY_NOW); 3199 if (current_arp_slave) { 3200 bond_set_slave_inactive_flags( 3201 current_arp_slave, 3202 BOND_SLAVE_NOTIFY_NOW); 3203 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3204 } 3205 3206 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3207 3208 if (!rtnl_dereference(bond->curr_active_slave) || 3209 slave == rtnl_dereference(bond->primary_slave)) 3210 goto do_failover; 3211 3212 } 3213 3214 continue; 3215 3216 case BOND_LINK_DOWN: 3217 if (slave->link_failure_count < UINT_MAX) 3218 slave->link_failure_count++; 3219 3220 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3221 BOND_SLAVE_NOTIFY_NOW); 3222 bond_set_slave_inactive_flags(slave, 3223 BOND_SLAVE_NOTIFY_NOW); 3224 3225 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3226 3227 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3228 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3229 goto do_failover; 3230 } 3231 3232 continue; 3233 3234 case BOND_LINK_FAIL: 3235 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3236 BOND_SLAVE_NOTIFY_NOW); 3237 bond_set_slave_inactive_flags(slave, 3238 BOND_SLAVE_NOTIFY_NOW); 3239 3240 /* A slave has just been enslaved and has become 3241 * the current active slave. 3242 */ 3243 if (rtnl_dereference(bond->curr_active_slave)) 3244 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3245 continue; 3246 3247 default: 3248 slave_err(bond->dev, slave->dev, 3249 "impossible: link_new_state %d on slave\n", 3250 slave->link_new_state); 3251 continue; 3252 } 3253 3254do_failover: 3255 block_netpoll_tx(); 3256 bond_select_active_slave(bond); 3257 unblock_netpoll_tx(); 3258 } 3259 3260 bond_set_carrier(bond); 3261} 3262 3263/* Send ARP probes for active-backup mode ARP monitor. 3264 * 3265 * Called with rcu_read_lock held. 3266 */ 3267static bool bond_ab_arp_probe(struct bonding *bond) 3268{ 3269 struct slave *slave, *before = NULL, *new_slave = NULL, 3270 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3271 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3272 struct list_head *iter; 3273 bool found = false; 3274 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3275 3276 if (curr_arp_slave && curr_active_slave) 3277 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3278 curr_arp_slave->dev->name, 3279 curr_active_slave->dev->name); 3280 3281 if (curr_active_slave) { 3282 bond_arp_send_all(bond, curr_active_slave); 3283 return should_notify_rtnl; 3284 } 3285 3286 /* if we don't have a curr_active_slave, search for the next available 3287 * backup slave from the current_arp_slave and make it the candidate 3288 * for becoming the curr_active_slave 3289 */ 3290 3291 if (!curr_arp_slave) { 3292 curr_arp_slave = bond_first_slave_rcu(bond); 3293 if (!curr_arp_slave) 3294 return should_notify_rtnl; 3295 } 3296 3297 bond_for_each_slave_rcu(bond, slave, iter) { 3298 if (!found && !before && bond_slave_is_up(slave)) 3299 before = slave; 3300 3301 if (found && !new_slave && bond_slave_is_up(slave)) 3302 new_slave = slave; 3303 /* if the link state is up at this point, we 3304 * mark it down - this can happen if we have 3305 * simultaneous link failures and 3306 * reselect_active_interface doesn't make this 3307 * one the current slave so it is still marked 3308 * up when it is actually down 3309 */ 3310 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3311 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3312 BOND_SLAVE_NOTIFY_LATER); 3313 if (slave->link_failure_count < UINT_MAX) 3314 slave->link_failure_count++; 3315 3316 bond_set_slave_inactive_flags(slave, 3317 BOND_SLAVE_NOTIFY_LATER); 3318 3319 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3320 } 3321 if (slave == curr_arp_slave) 3322 found = true; 3323 } 3324 3325 if (!new_slave && before) 3326 new_slave = before; 3327 3328 if (!new_slave) 3329 goto check_state; 3330 3331 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3332 BOND_SLAVE_NOTIFY_LATER); 3333 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3334 bond_arp_send_all(bond, new_slave); 3335 new_slave->last_link_up = jiffies; 3336 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3337 3338check_state: 3339 bond_for_each_slave_rcu(bond, slave, iter) { 3340 if (slave->should_notify || slave->should_notify_link) { 3341 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3342 break; 3343 } 3344 } 3345 return should_notify_rtnl; 3346} 3347 3348static void bond_activebackup_arp_mon(struct bonding *bond) 3349{ 3350 bool should_notify_peers = false; 3351 bool should_notify_rtnl = false; 3352 int delta_in_ticks; 3353 3354 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3355 3356 if (!bond_has_slaves(bond)) 3357 goto re_arm; 3358 3359 rcu_read_lock(); 3360 3361 should_notify_peers = bond_should_notify_peers(bond); 3362 3363 if (bond_ab_arp_inspect(bond)) { 3364 rcu_read_unlock(); 3365 3366 /* Race avoidance with bond_close flush of workqueue */ 3367 if (!rtnl_trylock()) { 3368 delta_in_ticks = 1; 3369 should_notify_peers = false; 3370 goto re_arm; 3371 } 3372 3373 bond_ab_arp_commit(bond); 3374 3375 rtnl_unlock(); 3376 rcu_read_lock(); 3377 } 3378 3379 should_notify_rtnl = bond_ab_arp_probe(bond); 3380 rcu_read_unlock(); 3381 3382re_arm: 3383 if (bond->params.arp_interval) 3384 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3385 3386 if (should_notify_peers || should_notify_rtnl) { 3387 if (!rtnl_trylock()) 3388 return; 3389 3390 if (should_notify_peers) { 3391 bond->send_peer_notif--; 3392 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3393 bond->dev); 3394 } 3395 if (should_notify_rtnl) { 3396 bond_slave_state_notify(bond); 3397 bond_slave_link_notify(bond); 3398 } 3399 3400 rtnl_unlock(); 3401 } 3402} 3403 3404static void bond_arp_monitor(struct work_struct *work) 3405{ 3406 struct bonding *bond = container_of(work, struct bonding, 3407 arp_work.work); 3408 3409 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3410 bond_activebackup_arp_mon(bond); 3411 else 3412 bond_loadbalance_arp_mon(bond); 3413} 3414 3415/*-------------------------- netdev event handling --------------------------*/ 3416 3417/* Change device name */ 3418static int bond_event_changename(struct bonding *bond) 3419{ 3420 bond_remove_proc_entry(bond); 3421 bond_create_proc_entry(bond); 3422 3423 bond_debug_reregister(bond); 3424 3425 return NOTIFY_DONE; 3426} 3427 3428static int bond_master_netdev_event(unsigned long event, 3429 struct net_device *bond_dev) 3430{ 3431 struct bonding *event_bond = netdev_priv(bond_dev); 3432 3433 netdev_dbg(bond_dev, "%s called\n", __func__); 3434 3435 switch (event) { 3436 case NETDEV_CHANGENAME: 3437 return bond_event_changename(event_bond); 3438 case NETDEV_UNREGISTER: 3439 bond_remove_proc_entry(event_bond); 3440#ifdef CONFIG_XFRM_OFFLOAD 3441 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3442#endif /* CONFIG_XFRM_OFFLOAD */ 3443 break; 3444 case NETDEV_REGISTER: 3445 bond_create_proc_entry(event_bond); 3446 break; 3447 default: 3448 break; 3449 } 3450 3451 return NOTIFY_DONE; 3452} 3453 3454static int bond_slave_netdev_event(unsigned long event, 3455 struct net_device *slave_dev) 3456{ 3457 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3458 struct bonding *bond; 3459 struct net_device *bond_dev; 3460 3461 /* A netdev event can be generated while enslaving a device 3462 * before netdev_rx_handler_register is called in which case 3463 * slave will be NULL 3464 */ 3465 if (!slave) { 3466 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3467 return NOTIFY_DONE; 3468 } 3469 3470 bond_dev = slave->bond->dev; 3471 bond = slave->bond; 3472 primary = rtnl_dereference(bond->primary_slave); 3473 3474 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3475 3476 switch (event) { 3477 case NETDEV_UNREGISTER: 3478 if (bond_dev->type != ARPHRD_ETHER) 3479 bond_release_and_destroy(bond_dev, slave_dev); 3480 else 3481 __bond_release_one(bond_dev, slave_dev, false, true); 3482 break; 3483 case NETDEV_UP: 3484 case NETDEV_CHANGE: 3485 /* For 802.3ad mode only: 3486 * Getting invalid Speed/Duplex values here will put slave 3487 * in weird state. Mark it as link-fail if the link was 3488 * previously up or link-down if it hasn't yet come up, and 3489 * let link-monitoring (miimon) set it right when correct 3490 * speeds/duplex are available. 3491 */ 3492 if (bond_update_speed_duplex(slave) && 3493 BOND_MODE(bond) == BOND_MODE_8023AD) { 3494 if (slave->last_link_up) 3495 slave->link = BOND_LINK_FAIL; 3496 else 3497 slave->link = BOND_LINK_DOWN; 3498 } 3499 3500 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3501 bond_3ad_adapter_speed_duplex_changed(slave); 3502 fallthrough; 3503 case NETDEV_DOWN: 3504 /* Refresh slave-array if applicable! 3505 * If the setup does not use miimon or arpmon (mode-specific!), 3506 * then these events will not cause the slave-array to be 3507 * refreshed. This will cause xmit to use a slave that is not 3508 * usable. Avoid such situation by refeshing the array at these 3509 * events. If these (miimon/arpmon) parameters are configured 3510 * then array gets refreshed twice and that should be fine! 3511 */ 3512 if (bond_mode_can_use_xmit_hash(bond)) 3513 bond_update_slave_arr(bond, NULL); 3514 break; 3515 case NETDEV_CHANGEMTU: 3516 /* TODO: Should slaves be allowed to 3517 * independently alter their MTU? For 3518 * an active-backup bond, slaves need 3519 * not be the same type of device, so 3520 * MTUs may vary. For other modes, 3521 * slaves arguably should have the 3522 * same MTUs. To do this, we'd need to 3523 * take over the slave's change_mtu 3524 * function for the duration of their 3525 * servitude. 3526 */ 3527 break; 3528 case NETDEV_CHANGENAME: 3529 /* we don't care if we don't have primary set */ 3530 if (!bond_uses_primary(bond) || 3531 !bond->params.primary[0]) 3532 break; 3533 3534 if (slave == primary) { 3535 /* slave's name changed - he's no longer primary */ 3536 RCU_INIT_POINTER(bond->primary_slave, NULL); 3537 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3538 /* we have a new primary slave */ 3539 rcu_assign_pointer(bond->primary_slave, slave); 3540 } else { /* we didn't change primary - exit */ 3541 break; 3542 } 3543 3544 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3545 primary ? slave_dev->name : "none"); 3546 3547 block_netpoll_tx(); 3548 bond_select_active_slave(bond); 3549 unblock_netpoll_tx(); 3550 break; 3551 case NETDEV_FEAT_CHANGE: 3552 if (!bond->notifier_ctx) { 3553 bond->notifier_ctx = true; 3554 bond_compute_features(bond); 3555 bond->notifier_ctx = false; 3556 } 3557 break; 3558 case NETDEV_RESEND_IGMP: 3559 /* Propagate to master device */ 3560 call_netdevice_notifiers(event, slave->bond->dev); 3561 break; 3562 default: 3563 break; 3564 } 3565 3566 return NOTIFY_DONE; 3567} 3568 3569/* bond_netdev_event: handle netdev notifier chain events. 3570 * 3571 * This function receives events for the netdev chain. The caller (an 3572 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3573 * locks for us to safely manipulate the slave devices (RTNL lock, 3574 * dev_probe_lock). 3575 */ 3576static int bond_netdev_event(struct notifier_block *this, 3577 unsigned long event, void *ptr) 3578{ 3579 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3580 3581 netdev_dbg(event_dev, "%s received %s\n", 3582 __func__, netdev_cmd_to_name(event)); 3583 3584 if (!(event_dev->priv_flags & IFF_BONDING)) 3585 return NOTIFY_DONE; 3586 3587 if (event_dev->flags & IFF_MASTER) { 3588 int ret; 3589 3590 ret = bond_master_netdev_event(event, event_dev); 3591 if (ret != NOTIFY_DONE) 3592 return ret; 3593 } 3594 3595 if (event_dev->flags & IFF_SLAVE) 3596 return bond_slave_netdev_event(event, event_dev); 3597 3598 return NOTIFY_DONE; 3599} 3600 3601static struct notifier_block bond_netdev_notifier = { 3602 .notifier_call = bond_netdev_event, 3603}; 3604 3605/*---------------------------- Hashing Policies -----------------------------*/ 3606 3607/* L2 hash helper */ 3608static inline u32 bond_eth_hash(struct sk_buff *skb) 3609{ 3610 struct ethhdr *ep, hdr_tmp; 3611 3612 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3613 if (ep) 3614 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3615 return 0; 3616} 3617 3618static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, 3619 int *noff, int *proto, bool l34) 3620{ 3621 const struct ipv6hdr *iph6; 3622 const struct iphdr *iph; 3623 3624 if (skb->protocol == htons(ETH_P_IP)) { 3625 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph)))) 3626 return false; 3627 iph = (const struct iphdr *)(skb->data + *noff); 3628 iph_to_flow_copy_v4addrs(fk, iph); 3629 *noff += iph->ihl << 2; 3630 if (!ip_is_fragment(iph)) 3631 *proto = iph->protocol; 3632 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3633 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6)))) 3634 return false; 3635 iph6 = (const struct ipv6hdr *)(skb->data + *noff); 3636 iph_to_flow_copy_v6addrs(fk, iph6); 3637 *noff += sizeof(*iph6); 3638 *proto = iph6->nexthdr; 3639 } else { 3640 return false; 3641 } 3642 3643 if (l34 && *proto >= 0) 3644 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto); 3645 3646 return true; 3647} 3648 3649/* Extract the appropriate headers based on bond's xmit policy */ 3650static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3651 struct flow_keys *fk) 3652{ 3653 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 3654 int noff, proto = -1; 3655 3656 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) { 3657 memset(fk, 0, sizeof(*fk)); 3658 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 3659 fk, NULL, 0, 0, 0, 0); 3660 } 3661 3662 fk->ports.ports = 0; 3663 memset(&fk->icmp, 0, sizeof(fk->icmp)); 3664 noff = skb_network_offset(skb); 3665 if (!bond_flow_ip(skb, fk, &noff, &proto, l34)) 3666 return false; 3667 3668 /* ICMP error packets contains at least 8 bytes of the header 3669 * of the packet which generated the error. Use this information 3670 * to correlate ICMP error packets within the same flow which 3671 * generated the error. 3672 */ 3673 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) { 3674 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data, 3675 skb_transport_offset(skb), 3676 skb_headlen(skb)); 3677 if (proto == IPPROTO_ICMP) { 3678 if (!icmp_is_err(fk->icmp.type)) 3679 return true; 3680 3681 noff += sizeof(struct icmphdr); 3682 } else if (proto == IPPROTO_ICMPV6) { 3683 if (!icmpv6_is_err(fk->icmp.type)) 3684 return true; 3685 3686 noff += sizeof(struct icmp6hdr); 3687 } 3688 return bond_flow_ip(skb, fk, &noff, &proto, l34); 3689 } 3690 3691 return true; 3692} 3693 3694/** 3695 * bond_xmit_hash - generate a hash value based on the xmit policy 3696 * @bond: bonding device 3697 * @skb: buffer to use for headers 3698 * 3699 * This function will extract the necessary headers from the skb buffer and use 3700 * them to generate a hash based on the xmit_policy set in the bonding device 3701 */ 3702u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3703{ 3704 struct flow_keys flow; 3705 u32 hash; 3706 3707 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3708 skb->l4_hash) 3709 return skb->hash; 3710 3711 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3712 !bond_flow_dissect(bond, skb, &flow)) 3713 return bond_eth_hash(skb); 3714 3715 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3716 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 3717 hash = bond_eth_hash(skb); 3718 } else { 3719 if (flow.icmp.id) 3720 memcpy(&hash, &flow.icmp, sizeof(hash)); 3721 else 3722 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 3723 } 3724 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3725 (__force u32)flow_get_u32_src(&flow); 3726 hash ^= (hash >> 16); 3727 hash ^= (hash >> 8); 3728 3729 return hash >> 1; 3730} 3731 3732/*-------------------------- Device entry points ----------------------------*/ 3733 3734void bond_work_init_all(struct bonding *bond) 3735{ 3736 INIT_DELAYED_WORK(&bond->mcast_work, 3737 bond_resend_igmp_join_requests_delayed); 3738 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3739 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3740 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3741 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3742 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3743} 3744 3745static void bond_work_cancel_all(struct bonding *bond) 3746{ 3747 cancel_delayed_work_sync(&bond->mii_work); 3748 cancel_delayed_work_sync(&bond->arp_work); 3749 cancel_delayed_work_sync(&bond->alb_work); 3750 cancel_delayed_work_sync(&bond->ad_work); 3751 cancel_delayed_work_sync(&bond->mcast_work); 3752 cancel_delayed_work_sync(&bond->slave_arr_work); 3753} 3754 3755static int bond_open(struct net_device *bond_dev) 3756{ 3757 struct bonding *bond = netdev_priv(bond_dev); 3758 struct list_head *iter; 3759 struct slave *slave; 3760 3761 /* reset slave->backup and slave->inactive */ 3762 if (bond_has_slaves(bond)) { 3763 bond_for_each_slave(bond, slave, iter) { 3764 if (bond_uses_primary(bond) && 3765 slave != rcu_access_pointer(bond->curr_active_slave)) { 3766 bond_set_slave_inactive_flags(slave, 3767 BOND_SLAVE_NOTIFY_NOW); 3768 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3769 bond_set_slave_active_flags(slave, 3770 BOND_SLAVE_NOTIFY_NOW); 3771 } 3772 } 3773 } 3774 3775 if (bond_is_lb(bond)) { 3776 /* bond_alb_initialize must be called before the timer 3777 * is started. 3778 */ 3779 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3780 return -ENOMEM; 3781 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3782 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3783 } 3784 3785 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3786 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3787 3788 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3789 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3790 bond->recv_probe = bond_arp_rcv; 3791 } 3792 3793 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3794 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3795 /* register to receive LACPDUs */ 3796 bond->recv_probe = bond_3ad_lacpdu_recv; 3797 bond_3ad_initiate_agg_selection(bond, 1); 3798 3799 bond_for_each_slave(bond, slave, iter) 3800 dev_mc_add(slave->dev, lacpdu_mcast_addr); 3801 } 3802 3803 if (bond_mode_can_use_xmit_hash(bond)) 3804 bond_update_slave_arr(bond, NULL); 3805 3806 return 0; 3807} 3808 3809static int bond_close(struct net_device *bond_dev) 3810{ 3811 struct bonding *bond = netdev_priv(bond_dev); 3812 struct slave *slave; 3813 3814 bond_work_cancel_all(bond); 3815 bond->send_peer_notif = 0; 3816 if (bond_is_lb(bond)) 3817 bond_alb_deinitialize(bond); 3818 bond->recv_probe = NULL; 3819 3820 if (bond_uses_primary(bond)) { 3821 rcu_read_lock(); 3822 slave = rcu_dereference(bond->curr_active_slave); 3823 if (slave) 3824 bond_hw_addr_flush(bond_dev, slave->dev); 3825 rcu_read_unlock(); 3826 } else { 3827 struct list_head *iter; 3828 3829 bond_for_each_slave(bond, slave, iter) 3830 bond_hw_addr_flush(bond_dev, slave->dev); 3831 } 3832 3833 return 0; 3834} 3835 3836/* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3837 * that some drivers can provide 32bit values only. 3838 */ 3839static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3840 const struct rtnl_link_stats64 *_new, 3841 const struct rtnl_link_stats64 *_old) 3842{ 3843 const u64 *new = (const u64 *)_new; 3844 const u64 *old = (const u64 *)_old; 3845 u64 *res = (u64 *)_res; 3846 int i; 3847 3848 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3849 u64 nv = new[i]; 3850 u64 ov = old[i]; 3851 s64 delta = nv - ov; 3852 3853 /* detects if this particular field is 32bit only */ 3854 if (((nv | ov) >> 32) == 0) 3855 delta = (s64)(s32)((u32)nv - (u32)ov); 3856 3857 /* filter anomalies, some drivers reset their stats 3858 * at down/up events. 3859 */ 3860 if (delta > 0) 3861 res[i] += delta; 3862 } 3863} 3864 3865#ifdef CONFIG_LOCKDEP 3866static int bond_get_lowest_level_rcu(struct net_device *dev) 3867{ 3868 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 3869 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 3870 int cur = 0, max = 0; 3871 3872 now = dev; 3873 iter = &dev->adj_list.lower; 3874 3875 while (1) { 3876 next = NULL; 3877 while (1) { 3878 ldev = netdev_next_lower_dev_rcu(now, &iter); 3879 if (!ldev) 3880 break; 3881 3882 next = ldev; 3883 niter = &ldev->adj_list.lower; 3884 dev_stack[cur] = now; 3885 iter_stack[cur++] = iter; 3886 if (max <= cur) 3887 max = cur; 3888 break; 3889 } 3890 3891 if (!next) { 3892 if (!cur) 3893 return max; 3894 next = dev_stack[--cur]; 3895 niter = iter_stack[cur]; 3896 } 3897 3898 now = next; 3899 iter = niter; 3900 } 3901 3902 return max; 3903} 3904#endif 3905 3906static void bond_get_stats(struct net_device *bond_dev, 3907 struct rtnl_link_stats64 *stats) 3908{ 3909 struct bonding *bond = netdev_priv(bond_dev); 3910 struct rtnl_link_stats64 temp; 3911 struct list_head *iter; 3912 struct slave *slave; 3913 int nest_level = 0; 3914 3915 3916 rcu_read_lock(); 3917#ifdef CONFIG_LOCKDEP 3918 nest_level = bond_get_lowest_level_rcu(bond_dev); 3919#endif 3920 3921 spin_lock_nested(&bond->stats_lock, nest_level); 3922 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3923 3924 bond_for_each_slave_rcu(bond, slave, iter) { 3925 const struct rtnl_link_stats64 *new = 3926 dev_get_stats(slave->dev, &temp); 3927 3928 bond_fold_stats(stats, new, &slave->slave_stats); 3929 3930 /* save off the slave stats for the next run */ 3931 memcpy(&slave->slave_stats, new, sizeof(*new)); 3932 } 3933 3934 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3935 spin_unlock(&bond->stats_lock); 3936 rcu_read_unlock(); 3937} 3938 3939static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3940{ 3941 struct bonding *bond = netdev_priv(bond_dev); 3942 struct net_device *slave_dev = NULL; 3943 struct ifbond k_binfo; 3944 struct ifbond __user *u_binfo = NULL; 3945 struct ifslave k_sinfo; 3946 struct ifslave __user *u_sinfo = NULL; 3947 struct mii_ioctl_data *mii = NULL; 3948 struct bond_opt_value newval; 3949 struct net *net; 3950 int res = 0; 3951 3952 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3953 3954 switch (cmd) { 3955 case SIOCGMIIPHY: 3956 mii = if_mii(ifr); 3957 if (!mii) 3958 return -EINVAL; 3959 3960 mii->phy_id = 0; 3961 fallthrough; 3962 case SIOCGMIIREG: 3963 /* We do this again just in case we were called by SIOCGMIIREG 3964 * instead of SIOCGMIIPHY. 3965 */ 3966 mii = if_mii(ifr); 3967 if (!mii) 3968 return -EINVAL; 3969 3970 if (mii->reg_num == 1) { 3971 mii->val_out = 0; 3972 if (netif_carrier_ok(bond->dev)) 3973 mii->val_out = BMSR_LSTATUS; 3974 } 3975 3976 return 0; 3977 case BOND_INFO_QUERY_OLD: 3978 case SIOCBONDINFOQUERY: 3979 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3980 3981 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3982 return -EFAULT; 3983 3984 bond_info_query(bond_dev, &k_binfo); 3985 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3986 return -EFAULT; 3987 3988 return 0; 3989 case BOND_SLAVE_INFO_QUERY_OLD: 3990 case SIOCBONDSLAVEINFOQUERY: 3991 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3992 3993 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3994 return -EFAULT; 3995 3996 res = bond_slave_info_query(bond_dev, &k_sinfo); 3997 if (res == 0 && 3998 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3999 return -EFAULT; 4000 4001 return res; 4002 default: 4003 break; 4004 } 4005 4006 net = dev_net(bond_dev); 4007 4008 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4009 return -EPERM; 4010 4011 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4012 4013 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4014 4015 if (!slave_dev) 4016 return -ENODEV; 4017 4018 switch (cmd) { 4019 case BOND_ENSLAVE_OLD: 4020 case SIOCBONDENSLAVE: 4021 res = bond_enslave(bond_dev, slave_dev, NULL); 4022 break; 4023 case BOND_RELEASE_OLD: 4024 case SIOCBONDRELEASE: 4025 res = bond_release(bond_dev, slave_dev); 4026 break; 4027 case BOND_SETHWADDR_OLD: 4028 case SIOCBONDSETHWADDR: 4029 res = bond_set_dev_addr(bond_dev, slave_dev); 4030 break; 4031 case BOND_CHANGE_ACTIVE_OLD: 4032 case SIOCBONDCHANGEACTIVE: 4033 bond_opt_initstr(&newval, slave_dev->name); 4034 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4035 &newval); 4036 break; 4037 default: 4038 res = -EOPNOTSUPP; 4039 } 4040 4041 return res; 4042} 4043 4044static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4045{ 4046 struct bonding *bond = netdev_priv(bond_dev); 4047 4048 if (change & IFF_PROMISC) 4049 bond_set_promiscuity(bond, 4050 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4051 4052 if (change & IFF_ALLMULTI) 4053 bond_set_allmulti(bond, 4054 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4055} 4056 4057static void bond_set_rx_mode(struct net_device *bond_dev) 4058{ 4059 struct bonding *bond = netdev_priv(bond_dev); 4060 struct list_head *iter; 4061 struct slave *slave; 4062 4063 rcu_read_lock(); 4064 if (bond_uses_primary(bond)) { 4065 slave = rcu_dereference(bond->curr_active_slave); 4066 if (slave) { 4067 dev_uc_sync(slave->dev, bond_dev); 4068 dev_mc_sync(slave->dev, bond_dev); 4069 } 4070 } else { 4071 bond_for_each_slave_rcu(bond, slave, iter) { 4072 dev_uc_sync_multiple(slave->dev, bond_dev); 4073 dev_mc_sync_multiple(slave->dev, bond_dev); 4074 } 4075 } 4076 rcu_read_unlock(); 4077} 4078 4079static int bond_neigh_init(struct neighbour *n) 4080{ 4081 struct bonding *bond = netdev_priv(n->dev); 4082 const struct net_device_ops *slave_ops; 4083 struct neigh_parms parms; 4084 struct slave *slave; 4085 int ret = 0; 4086 4087 rcu_read_lock(); 4088 slave = bond_first_slave_rcu(bond); 4089 if (!slave) 4090 goto out; 4091 slave_ops = slave->dev->netdev_ops; 4092 if (!slave_ops->ndo_neigh_setup) 4093 goto out; 4094 4095 /* TODO: find another way [1] to implement this. 4096 * Passing a zeroed structure is fragile, 4097 * but at least we do not pass garbage. 4098 * 4099 * [1] One way would be that ndo_neigh_setup() never touch 4100 * struct neigh_parms, but propagate the new neigh_setup() 4101 * back to ___neigh_create() / neigh_parms_alloc() 4102 */ 4103 memset(&parms, 0, sizeof(parms)); 4104 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4105 4106 if (ret) 4107 goto out; 4108 4109 if (parms.neigh_setup) 4110 ret = parms.neigh_setup(n); 4111out: 4112 rcu_read_unlock(); 4113 return ret; 4114} 4115 4116/* The bonding ndo_neigh_setup is called at init time beofre any 4117 * slave exists. So we must declare proxy setup function which will 4118 * be used at run time to resolve the actual slave neigh param setup. 4119 * 4120 * It's also called by master devices (such as vlans) to setup their 4121 * underlying devices. In that case - do nothing, we're already set up from 4122 * our init. 4123 */ 4124static int bond_neigh_setup(struct net_device *dev, 4125 struct neigh_parms *parms) 4126{ 4127 /* modify only our neigh_parms */ 4128 if (parms->dev == dev) 4129 parms->neigh_setup = bond_neigh_init; 4130 4131 return 0; 4132} 4133 4134/* Change the MTU of all of a master's slaves to match the master */ 4135static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4136{ 4137 struct bonding *bond = netdev_priv(bond_dev); 4138 struct slave *slave, *rollback_slave; 4139 struct list_head *iter; 4140 int res = 0; 4141 4142 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4143 4144 bond_for_each_slave(bond, slave, iter) { 4145 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4146 slave, slave->dev->netdev_ops->ndo_change_mtu); 4147 4148 res = dev_set_mtu(slave->dev, new_mtu); 4149 4150 if (res) { 4151 /* If we failed to set the slave's mtu to the new value 4152 * we must abort the operation even in ACTIVE_BACKUP 4153 * mode, because if we allow the backup slaves to have 4154 * different mtu values than the active slave we'll 4155 * need to change their mtu when doing a failover. That 4156 * means changing their mtu from timer context, which 4157 * is probably not a good idea. 4158 */ 4159 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4160 res, new_mtu); 4161 goto unwind; 4162 } 4163 } 4164 4165 bond_dev->mtu = new_mtu; 4166 4167 return 0; 4168 4169unwind: 4170 /* unwind from head to the slave that failed */ 4171 bond_for_each_slave(bond, rollback_slave, iter) { 4172 int tmp_res; 4173 4174 if (rollback_slave == slave) 4175 break; 4176 4177 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4178 if (tmp_res) 4179 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4180 tmp_res); 4181 } 4182 4183 return res; 4184} 4185 4186/* Change HW address 4187 * 4188 * Note that many devices must be down to change the HW address, and 4189 * downing the master releases all slaves. We can make bonds full of 4190 * bonding devices to test this, however. 4191 */ 4192static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4193{ 4194 struct bonding *bond = netdev_priv(bond_dev); 4195 struct slave *slave, *rollback_slave; 4196 struct sockaddr_storage *ss = addr, tmp_ss; 4197 struct list_head *iter; 4198 int res = 0; 4199 4200 if (BOND_MODE(bond) == BOND_MODE_ALB) 4201 return bond_alb_set_mac_address(bond_dev, addr); 4202 4203 4204 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4205 4206 /* If fail_over_mac is enabled, do nothing and return success. 4207 * Returning an error causes ifenslave to fail. 4208 */ 4209 if (bond->params.fail_over_mac && 4210 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4211 return 0; 4212 4213 if (!is_valid_ether_addr(ss->__data)) 4214 return -EADDRNOTAVAIL; 4215 4216 bond_for_each_slave(bond, slave, iter) { 4217 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4218 __func__, slave); 4219 res = dev_set_mac_address(slave->dev, addr, NULL); 4220 if (res) { 4221 /* TODO: consider downing the slave 4222 * and retry ? 4223 * User should expect communications 4224 * breakage anyway until ARP finish 4225 * updating, so... 4226 */ 4227 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4228 __func__, res); 4229 goto unwind; 4230 } 4231 } 4232 4233 /* success */ 4234 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 4235 return 0; 4236 4237unwind: 4238 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4239 tmp_ss.ss_family = bond_dev->type; 4240 4241 /* unwind from head to the slave that failed */ 4242 bond_for_each_slave(bond, rollback_slave, iter) { 4243 int tmp_res; 4244 4245 if (rollback_slave == slave) 4246 break; 4247 4248 tmp_res = dev_set_mac_address(rollback_slave->dev, 4249 (struct sockaddr *)&tmp_ss, NULL); 4250 if (tmp_res) { 4251 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4252 __func__, tmp_res); 4253 } 4254 } 4255 4256 return res; 4257} 4258 4259/** 4260 * bond_get_slave_by_id - get xmit slave with slave_id 4261 * @bond: bonding device that is transmitting 4262 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4263 * 4264 * This function tries to get slave with slave_id but in case 4265 * it fails, it tries to find the first available slave for transmission. 4266 */ 4267static struct slave *bond_get_slave_by_id(struct bonding *bond, 4268 int slave_id) 4269{ 4270 struct list_head *iter; 4271 struct slave *slave; 4272 int i = slave_id; 4273 4274 /* Here we start from the slave with slave_id */ 4275 bond_for_each_slave_rcu(bond, slave, iter) { 4276 if (--i < 0) { 4277 if (bond_slave_can_tx(slave)) 4278 return slave; 4279 } 4280 } 4281 4282 /* Here we start from the first slave up to slave_id */ 4283 i = slave_id; 4284 bond_for_each_slave_rcu(bond, slave, iter) { 4285 if (--i < 0) 4286 break; 4287 if (bond_slave_can_tx(slave)) 4288 return slave; 4289 } 4290 /* no slave that can tx has been found */ 4291 return NULL; 4292} 4293 4294/** 4295 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4296 * @bond: bonding device to use 4297 * 4298 * Based on the value of the bonding device's packets_per_slave parameter 4299 * this function generates a slave id, which is usually used as the next 4300 * slave to transmit through. 4301 */ 4302static u32 bond_rr_gen_slave_id(struct bonding *bond) 4303{ 4304 u32 slave_id; 4305 struct reciprocal_value reciprocal_packets_per_slave; 4306 int packets_per_slave = bond->params.packets_per_slave; 4307 4308 switch (packets_per_slave) { 4309 case 0: 4310 slave_id = prandom_u32(); 4311 break; 4312 case 1: 4313 slave_id = bond->rr_tx_counter; 4314 break; 4315 default: 4316 reciprocal_packets_per_slave = 4317 bond->params.reciprocal_packets_per_slave; 4318 slave_id = reciprocal_divide(bond->rr_tx_counter, 4319 reciprocal_packets_per_slave); 4320 break; 4321 } 4322 bond->rr_tx_counter++; 4323 4324 return slave_id; 4325} 4326 4327static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4328 struct sk_buff *skb) 4329{ 4330 struct slave *slave; 4331 int slave_cnt; 4332 u32 slave_id; 4333 4334 /* Start with the curr_active_slave that joined the bond as the 4335 * default for sending IGMP traffic. For failover purposes one 4336 * needs to maintain some consistency for the interface that will 4337 * send the join/membership reports. The curr_active_slave found 4338 * will send all of this type of traffic. 4339 */ 4340 if (skb->protocol == htons(ETH_P_IP)) { 4341 int noff = skb_network_offset(skb); 4342 struct iphdr *iph; 4343 4344 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4345 goto non_igmp; 4346 4347 iph = ip_hdr(skb); 4348 if (iph->protocol == IPPROTO_IGMP) { 4349 slave = rcu_dereference(bond->curr_active_slave); 4350 if (slave) 4351 return slave; 4352 return bond_get_slave_by_id(bond, 0); 4353 } 4354 } 4355 4356non_igmp: 4357 slave_cnt = READ_ONCE(bond->slave_cnt); 4358 if (likely(slave_cnt)) { 4359 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4360 return bond_get_slave_by_id(bond, slave_id); 4361 } 4362 return NULL; 4363} 4364 4365static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4366 struct net_device *bond_dev) 4367{ 4368 struct bonding *bond = netdev_priv(bond_dev); 4369 struct slave *slave; 4370 4371 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4372 if (likely(slave)) 4373 return bond_dev_queue_xmit(bond, skb, slave->dev); 4374 4375 return bond_tx_drop(bond_dev, skb); 4376} 4377 4378static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond, 4379 struct sk_buff *skb) 4380{ 4381 return rcu_dereference(bond->curr_active_slave); 4382} 4383 4384/* In active-backup mode, we know that bond->curr_active_slave is always valid if 4385 * the bond has a usable interface. 4386 */ 4387static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4388 struct net_device *bond_dev) 4389{ 4390 struct bonding *bond = netdev_priv(bond_dev); 4391 struct slave *slave; 4392 4393 slave = bond_xmit_activebackup_slave_get(bond, skb); 4394 if (slave) 4395 return bond_dev_queue_xmit(bond, skb, slave->dev); 4396 4397 return bond_tx_drop(bond_dev, skb); 4398} 4399 4400/* Use this to update slave_array when (a) it's not appropriate to update 4401 * slave_array right away (note that update_slave_array() may sleep) 4402 * and / or (b) RTNL is not held. 4403 */ 4404void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4405{ 4406 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4407} 4408 4409/* Slave array work handler. Holds only RTNL */ 4410static void bond_slave_arr_handler(struct work_struct *work) 4411{ 4412 struct bonding *bond = container_of(work, struct bonding, 4413 slave_arr_work.work); 4414 int ret; 4415 4416 if (!rtnl_trylock()) 4417 goto err; 4418 4419 ret = bond_update_slave_arr(bond, NULL); 4420 rtnl_unlock(); 4421 if (ret) { 4422 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4423 goto err; 4424 } 4425 return; 4426 4427err: 4428 bond_slave_arr_work_rearm(bond, 1); 4429} 4430 4431static void bond_skip_slave(struct bond_up_slave *slaves, 4432 struct slave *skipslave) 4433{ 4434 int idx; 4435 4436 /* Rare situation where caller has asked to skip a specific 4437 * slave but allocation failed (most likely!). BTW this is 4438 * only possible when the call is initiated from 4439 * __bond_release_one(). In this situation; overwrite the 4440 * skipslave entry in the array with the last entry from the 4441 * array to avoid a situation where the xmit path may choose 4442 * this to-be-skipped slave to send a packet out. 4443 */ 4444 for (idx = 0; slaves && idx < slaves->count; idx++) { 4445 if (skipslave == slaves->arr[idx]) { 4446 slaves->arr[idx] = 4447 slaves->arr[slaves->count - 1]; 4448 slaves->count--; 4449 break; 4450 } 4451 } 4452} 4453 4454static void bond_set_slave_arr(struct bonding *bond, 4455 struct bond_up_slave *usable_slaves, 4456 struct bond_up_slave *all_slaves) 4457{ 4458 struct bond_up_slave *usable, *all; 4459 4460 usable = rtnl_dereference(bond->usable_slaves); 4461 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 4462 kfree_rcu(usable, rcu); 4463 4464 all = rtnl_dereference(bond->all_slaves); 4465 rcu_assign_pointer(bond->all_slaves, all_slaves); 4466 kfree_rcu(all, rcu); 4467} 4468 4469static void bond_reset_slave_arr(struct bonding *bond) 4470{ 4471 struct bond_up_slave *usable, *all; 4472 4473 usable = rtnl_dereference(bond->usable_slaves); 4474 if (usable) { 4475 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4476 kfree_rcu(usable, rcu); 4477 } 4478 4479 all = rtnl_dereference(bond->all_slaves); 4480 if (all) { 4481 RCU_INIT_POINTER(bond->all_slaves, NULL); 4482 kfree_rcu(all, rcu); 4483 } 4484} 4485 4486/* Build the usable slaves array in control path for modes that use xmit-hash 4487 * to determine the slave interface - 4488 * (a) BOND_MODE_8023AD 4489 * (b) BOND_MODE_XOR 4490 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 4491 * 4492 * The caller is expected to hold RTNL only and NO other lock! 4493 */ 4494int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 4495{ 4496 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 4497 struct slave *slave; 4498 struct list_head *iter; 4499 int agg_id = 0; 4500 int ret = 0; 4501 4502#ifdef CONFIG_LOCKDEP 4503 WARN_ON(lockdep_is_held(&bond->mode_lock)); 4504#endif 4505 4506 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 4507 bond->slave_cnt), GFP_KERNEL); 4508 all_slaves = kzalloc(struct_size(all_slaves, arr, 4509 bond->slave_cnt), GFP_KERNEL); 4510 if (!usable_slaves || !all_slaves) { 4511 ret = -ENOMEM; 4512 goto out; 4513 } 4514 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4515 struct ad_info ad_info; 4516 4517 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 4518 pr_debug("bond_3ad_get_active_agg_info failed\n"); 4519 /* No active aggragator means it's not safe to use 4520 * the previous array. 4521 */ 4522 bond_reset_slave_arr(bond); 4523 goto out; 4524 } 4525 agg_id = ad_info.aggregator_id; 4526 } 4527 bond_for_each_slave(bond, slave, iter) { 4528 if (skipslave == slave) 4529 continue; 4530 4531 all_slaves->arr[all_slaves->count++] = slave; 4532 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4533 struct aggregator *agg; 4534 4535 agg = SLAVE_AD_INFO(slave)->port.aggregator; 4536 if (!agg || agg->aggregator_identifier != agg_id) 4537 continue; 4538 } 4539 if (!bond_slave_can_tx(slave)) 4540 continue; 4541 4542 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 4543 usable_slaves->count); 4544 4545 usable_slaves->arr[usable_slaves->count++] = slave; 4546 } 4547 4548 bond_set_slave_arr(bond, usable_slaves, all_slaves); 4549 return ret; 4550out: 4551 if (ret != 0 && skipslave) { 4552 bond_skip_slave(rtnl_dereference(bond->all_slaves), 4553 skipslave); 4554 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 4555 skipslave); 4556 } 4557 kfree_rcu(all_slaves, rcu); 4558 kfree_rcu(usable_slaves, rcu); 4559 4560 return ret; 4561} 4562 4563static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 4564 struct sk_buff *skb, 4565 struct bond_up_slave *slaves) 4566{ 4567 struct slave *slave; 4568 unsigned int count; 4569 u32 hash; 4570 4571 hash = bond_xmit_hash(bond, skb); 4572 count = slaves ? READ_ONCE(slaves->count) : 0; 4573 if (unlikely(!count)) 4574 return NULL; 4575 4576 slave = slaves->arr[hash % count]; 4577 return slave; 4578} 4579 4580/* Use this Xmit function for 3AD as well as XOR modes. The current 4581 * usable slave array is formed in the control path. The xmit function 4582 * just calculates hash and sends the packet out. 4583 */ 4584static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4585 struct net_device *dev) 4586{ 4587 struct bonding *bond = netdev_priv(dev); 4588 struct bond_up_slave *slaves; 4589 struct slave *slave; 4590 4591 slaves = rcu_dereference(bond->usable_slaves); 4592 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4593 if (likely(slave)) 4594 return bond_dev_queue_xmit(bond, skb, slave->dev); 4595 4596 return bond_tx_drop(dev, skb); 4597} 4598 4599/* in broadcast mode, we send everything to all usable interfaces. */ 4600static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4601 struct net_device *bond_dev) 4602{ 4603 struct bonding *bond = netdev_priv(bond_dev); 4604 struct slave *slave = NULL; 4605 struct list_head *iter; 4606 bool xmit_suc = false; 4607 bool skb_used = false; 4608 4609 bond_for_each_slave_rcu(bond, slave, iter) { 4610 struct sk_buff *skb2; 4611 4612 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 4613 continue; 4614 4615 if (bond_is_last_slave(bond, slave)) { 4616 skb2 = skb; 4617 skb_used = true; 4618 } else { 4619 skb2 = skb_clone(skb, GFP_ATOMIC); 4620 if (!skb2) { 4621 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4622 bond_dev->name, __func__); 4623 continue; 4624 } 4625 } 4626 4627 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 4628 xmit_suc = true; 4629 } 4630 4631 if (!skb_used) 4632 dev_kfree_skb_any(skb); 4633 4634 if (xmit_suc) 4635 return NETDEV_TX_OK; 4636 4637 atomic_long_inc(&bond_dev->tx_dropped); 4638 return NET_XMIT_DROP; 4639} 4640 4641/*------------------------- Device initialization ---------------------------*/ 4642 4643/* Lookup the slave that corresponds to a qid */ 4644static inline int bond_slave_override(struct bonding *bond, 4645 struct sk_buff *skb) 4646{ 4647 struct slave *slave = NULL; 4648 struct list_head *iter; 4649 4650 if (!skb_rx_queue_recorded(skb)) 4651 return 1; 4652 4653 /* Find out if any slaves have the same mapping as this skb. */ 4654 bond_for_each_slave_rcu(bond, slave, iter) { 4655 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4656 if (bond_slave_is_up(slave) && 4657 slave->link == BOND_LINK_UP) { 4658 bond_dev_queue_xmit(bond, skb, slave->dev); 4659 return 0; 4660 } 4661 /* If the slave isn't UP, use default transmit policy. */ 4662 break; 4663 } 4664 } 4665 4666 return 1; 4667} 4668 4669 4670static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4671 struct net_device *sb_dev) 4672{ 4673 /* This helper function exists to help dev_pick_tx get the correct 4674 * destination queue. Using a helper function skips a call to 4675 * skb_tx_hash and will put the skbs in the queue we expect on their 4676 * way down to the bonding driver. 4677 */ 4678 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4679 4680 /* Save the original txq to restore before passing to the driver */ 4681 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4682 4683 if (unlikely(txq >= dev->real_num_tx_queues)) { 4684 do { 4685 txq -= dev->real_num_tx_queues; 4686 } while (txq >= dev->real_num_tx_queues); 4687 } 4688 return txq; 4689} 4690 4691static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 4692 struct sk_buff *skb, 4693 bool all_slaves) 4694{ 4695 struct bonding *bond = netdev_priv(master_dev); 4696 struct bond_up_slave *slaves; 4697 struct slave *slave = NULL; 4698 4699 switch (BOND_MODE(bond)) { 4700 case BOND_MODE_ROUNDROBIN: 4701 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4702 break; 4703 case BOND_MODE_ACTIVEBACKUP: 4704 slave = bond_xmit_activebackup_slave_get(bond, skb); 4705 break; 4706 case BOND_MODE_8023AD: 4707 case BOND_MODE_XOR: 4708 if (all_slaves) 4709 slaves = rcu_dereference(bond->all_slaves); 4710 else 4711 slaves = rcu_dereference(bond->usable_slaves); 4712 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4713 break; 4714 case BOND_MODE_BROADCAST: 4715 break; 4716 case BOND_MODE_ALB: 4717 slave = bond_xmit_alb_slave_get(bond, skb); 4718 break; 4719 case BOND_MODE_TLB: 4720 slave = bond_xmit_tlb_slave_get(bond, skb); 4721 break; 4722 default: 4723 /* Should never happen, mode already checked */ 4724 WARN_ONCE(true, "Unknown bonding mode"); 4725 break; 4726 } 4727 4728 if (slave) 4729 return slave->dev; 4730 return NULL; 4731} 4732 4733static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4734{ 4735 struct bonding *bond = netdev_priv(dev); 4736 4737 if (bond_should_override_tx_queue(bond) && 4738 !bond_slave_override(bond, skb)) 4739 return NETDEV_TX_OK; 4740 4741 switch (BOND_MODE(bond)) { 4742 case BOND_MODE_ROUNDROBIN: 4743 return bond_xmit_roundrobin(skb, dev); 4744 case BOND_MODE_ACTIVEBACKUP: 4745 return bond_xmit_activebackup(skb, dev); 4746 case BOND_MODE_8023AD: 4747 case BOND_MODE_XOR: 4748 return bond_3ad_xor_xmit(skb, dev); 4749 case BOND_MODE_BROADCAST: 4750 return bond_xmit_broadcast(skb, dev); 4751 case BOND_MODE_ALB: 4752 return bond_alb_xmit(skb, dev); 4753 case BOND_MODE_TLB: 4754 return bond_tlb_xmit(skb, dev); 4755 default: 4756 /* Should never happen, mode already checked */ 4757 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4758 WARN_ON_ONCE(1); 4759 return bond_tx_drop(dev, skb); 4760 } 4761} 4762 4763static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4764{ 4765 struct bonding *bond = netdev_priv(dev); 4766 netdev_tx_t ret = NETDEV_TX_OK; 4767 4768 /* If we risk deadlock from transmitting this in the 4769 * netpoll path, tell netpoll to queue the frame for later tx 4770 */ 4771 if (unlikely(is_netpoll_tx_blocked(dev))) 4772 return NETDEV_TX_BUSY; 4773 4774 rcu_read_lock(); 4775 if (bond_has_slaves(bond)) 4776 ret = __bond_start_xmit(skb, dev); 4777 else 4778 ret = bond_tx_drop(dev, skb); 4779 rcu_read_unlock(); 4780 4781 return ret; 4782} 4783 4784static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 4785{ 4786 if (speed == 0 || speed == SPEED_UNKNOWN) 4787 speed = slave->speed; 4788 else 4789 speed = min(speed, slave->speed); 4790 4791 return speed; 4792} 4793 4794static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4795 struct ethtool_link_ksettings *cmd) 4796{ 4797 struct bonding *bond = netdev_priv(bond_dev); 4798 struct list_head *iter; 4799 struct slave *slave; 4800 u32 speed = 0; 4801 4802 cmd->base.duplex = DUPLEX_UNKNOWN; 4803 cmd->base.port = PORT_OTHER; 4804 4805 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4806 * do not need to check mode. Though link speed might not represent 4807 * the true receive or transmit bandwidth (not all modes are symmetric) 4808 * this is an accurate maximum. 4809 */ 4810 bond_for_each_slave(bond, slave, iter) { 4811 if (bond_slave_can_tx(slave)) { 4812 if (slave->speed != SPEED_UNKNOWN) { 4813 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 4814 speed = bond_mode_bcast_speed(slave, 4815 speed); 4816 else 4817 speed += slave->speed; 4818 } 4819 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4820 slave->duplex != DUPLEX_UNKNOWN) 4821 cmd->base.duplex = slave->duplex; 4822 } 4823 } 4824 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4825 4826 return 0; 4827} 4828 4829static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4830 struct ethtool_drvinfo *drvinfo) 4831{ 4832 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4833 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4834 BOND_ABI_VERSION); 4835} 4836 4837static const struct ethtool_ops bond_ethtool_ops = { 4838 .get_drvinfo = bond_ethtool_get_drvinfo, 4839 .get_link = ethtool_op_get_link, 4840 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4841}; 4842 4843static const struct net_device_ops bond_netdev_ops = { 4844 .ndo_init = bond_init, 4845 .ndo_uninit = bond_uninit, 4846 .ndo_open = bond_open, 4847 .ndo_stop = bond_close, 4848 .ndo_start_xmit = bond_start_xmit, 4849 .ndo_select_queue = bond_select_queue, 4850 .ndo_get_stats64 = bond_get_stats, 4851 .ndo_do_ioctl = bond_do_ioctl, 4852 .ndo_change_rx_flags = bond_change_rx_flags, 4853 .ndo_set_rx_mode = bond_set_rx_mode, 4854 .ndo_change_mtu = bond_change_mtu, 4855 .ndo_set_mac_address = bond_set_mac_address, 4856 .ndo_neigh_setup = bond_neigh_setup, 4857 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4858 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4859#ifdef CONFIG_NET_POLL_CONTROLLER 4860 .ndo_netpoll_setup = bond_netpoll_setup, 4861 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4862 .ndo_poll_controller = bond_poll_controller, 4863#endif 4864 .ndo_add_slave = bond_enslave, 4865 .ndo_del_slave = bond_release, 4866 .ndo_fix_features = bond_fix_features, 4867 .ndo_features_check = passthru_features_check, 4868 .ndo_get_xmit_slave = bond_xmit_get_slave, 4869}; 4870 4871static const struct device_type bond_type = { 4872 .name = "bond", 4873}; 4874 4875static void bond_destructor(struct net_device *bond_dev) 4876{ 4877 struct bonding *bond = netdev_priv(bond_dev); 4878 if (bond->wq) 4879 destroy_workqueue(bond->wq); 4880} 4881 4882void bond_setup(struct net_device *bond_dev) 4883{ 4884 struct bonding *bond = netdev_priv(bond_dev); 4885 4886 spin_lock_init(&bond->mode_lock); 4887 bond->params = bonding_defaults; 4888 4889 /* Initialize pointers */ 4890 bond->dev = bond_dev; 4891 4892 /* Initialize the device entry points */ 4893 ether_setup(bond_dev); 4894 bond_dev->max_mtu = ETH_MAX_MTU; 4895 bond_dev->netdev_ops = &bond_netdev_ops; 4896 bond_dev->ethtool_ops = &bond_ethtool_ops; 4897 4898 bond_dev->needs_free_netdev = true; 4899 bond_dev->priv_destructor = bond_destructor; 4900 4901 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4902 4903 /* Initialize the device options */ 4904 bond_dev->flags |= IFF_MASTER; 4905 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4906 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4907 4908#ifdef CONFIG_XFRM_OFFLOAD 4909 /* set up xfrm device ops (only supported in active-backup right now) */ 4910 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 4911 INIT_LIST_HEAD(&bond->ipsec_list); 4912 spin_lock_init(&bond->ipsec_lock); 4913#endif /* CONFIG_XFRM_OFFLOAD */ 4914 4915 /* don't acquire bond device's netif_tx_lock when transmitting */ 4916 bond_dev->features |= NETIF_F_LLTX; 4917 4918 /* By default, we declare the bond to be fully 4919 * VLAN hardware accelerated capable. Special 4920 * care is taken in the various xmit functions 4921 * when there are slaves that are not hw accel 4922 * capable 4923 */ 4924 4925 /* Don't allow bond devices to change network namespaces. */ 4926 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4927 4928 bond_dev->hw_features = BOND_VLAN_FEATURES | 4929 NETIF_F_HW_VLAN_CTAG_RX | 4930 NETIF_F_HW_VLAN_CTAG_FILTER | 4931 NETIF_F_HW_VLAN_STAG_RX | 4932 NETIF_F_HW_VLAN_STAG_FILTER; 4933 4934 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4935#ifdef CONFIG_XFRM_OFFLOAD 4936 bond_dev->hw_features |= BOND_XFRM_FEATURES; 4937#endif /* CONFIG_XFRM_OFFLOAD */ 4938 bond_dev->features |= bond_dev->hw_features; 4939 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 4940#ifdef CONFIG_XFRM_OFFLOAD 4941 /* Disable XFRM features if this isn't an active-backup config */ 4942 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 4943 bond_dev->features &= ~BOND_XFRM_FEATURES; 4944#endif /* CONFIG_XFRM_OFFLOAD */ 4945} 4946 4947/* Destroy a bonding device. 4948 * Must be under rtnl_lock when this function is called. 4949 */ 4950static void bond_uninit(struct net_device *bond_dev) 4951{ 4952 struct bonding *bond = netdev_priv(bond_dev); 4953 struct bond_up_slave *usable, *all; 4954 struct list_head *iter; 4955 struct slave *slave; 4956 4957 bond_netpoll_cleanup(bond_dev); 4958 4959 /* Release the bonded slaves */ 4960 bond_for_each_slave(bond, slave, iter) 4961 __bond_release_one(bond_dev, slave->dev, true, true); 4962 netdev_info(bond_dev, "Released all slaves\n"); 4963 4964 usable = rtnl_dereference(bond->usable_slaves); 4965 if (usable) { 4966 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4967 kfree_rcu(usable, rcu); 4968 } 4969 4970 all = rtnl_dereference(bond->all_slaves); 4971 if (all) { 4972 RCU_INIT_POINTER(bond->all_slaves, NULL); 4973 kfree_rcu(all, rcu); 4974 } 4975 4976 list_del(&bond->bond_list); 4977 4978 bond_debug_unregister(bond); 4979} 4980 4981/*------------------------- Module initialization ---------------------------*/ 4982 4983static int bond_check_params(struct bond_params *params) 4984{ 4985 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4986 struct bond_opt_value newval; 4987 const struct bond_opt_value *valptr; 4988 int arp_all_targets_value = 0; 4989 u16 ad_actor_sys_prio = 0; 4990 u16 ad_user_port_key = 0; 4991 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4992 int arp_ip_count; 4993 int bond_mode = BOND_MODE_ROUNDROBIN; 4994 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4995 int lacp_fast = 0; 4996 int tlb_dynamic_lb; 4997 4998 /* Convert string parameters. */ 4999 if (mode) { 5000 bond_opt_initstr(&newval, mode); 5001 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5002 if (!valptr) { 5003 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5004 return -EINVAL; 5005 } 5006 bond_mode = valptr->value; 5007 } 5008 5009 if (xmit_hash_policy) { 5010 if (bond_mode == BOND_MODE_ROUNDROBIN || 5011 bond_mode == BOND_MODE_ACTIVEBACKUP || 5012 bond_mode == BOND_MODE_BROADCAST) { 5013 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 5014 bond_mode_name(bond_mode)); 5015 } else { 5016 bond_opt_initstr(&newval, xmit_hash_policy); 5017 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5018 &newval); 5019 if (!valptr) { 5020 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5021 xmit_hash_policy); 5022 return -EINVAL; 5023 } 5024 xmit_hashtype = valptr->value; 5025 } 5026 } 5027 5028 if (lacp_rate) { 5029 if (bond_mode != BOND_MODE_8023AD) { 5030 pr_info("lacp_rate param is irrelevant in mode %s\n", 5031 bond_mode_name(bond_mode)); 5032 } else { 5033 bond_opt_initstr(&newval, lacp_rate); 5034 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5035 &newval); 5036 if (!valptr) { 5037 pr_err("Error: Invalid lacp rate \"%s\"\n", 5038 lacp_rate); 5039 return -EINVAL; 5040 } 5041 lacp_fast = valptr->value; 5042 } 5043 } 5044 5045 if (ad_select) { 5046 bond_opt_initstr(&newval, ad_select); 5047 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5048 &newval); 5049 if (!valptr) { 5050 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5051 return -EINVAL; 5052 } 5053 params->ad_select = valptr->value; 5054 if (bond_mode != BOND_MODE_8023AD) 5055 pr_warn("ad_select param only affects 802.3ad mode\n"); 5056 } else { 5057 params->ad_select = BOND_AD_STABLE; 5058 } 5059 5060 if (max_bonds < 0) { 5061 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5062 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5063 max_bonds = BOND_DEFAULT_MAX_BONDS; 5064 } 5065 5066 if (miimon < 0) { 5067 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5068 miimon, INT_MAX); 5069 miimon = 0; 5070 } 5071 5072 if (updelay < 0) { 5073 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5074 updelay, INT_MAX); 5075 updelay = 0; 5076 } 5077 5078 if (downdelay < 0) { 5079 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5080 downdelay, INT_MAX); 5081 downdelay = 0; 5082 } 5083 5084 if ((use_carrier != 0) && (use_carrier != 1)) { 5085 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5086 use_carrier); 5087 use_carrier = 1; 5088 } 5089 5090 if (num_peer_notif < 0 || num_peer_notif > 255) { 5091 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5092 num_peer_notif); 5093 num_peer_notif = 1; 5094 } 5095 5096 /* reset values for 802.3ad/TLB/ALB */ 5097 if (!bond_mode_uses_arp(bond_mode)) { 5098 if (!miimon) { 5099 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"); 5100 pr_warn("Forcing miimon to 100msec\n"); 5101 miimon = BOND_DEFAULT_MIIMON; 5102 } 5103 } 5104 5105 if (tx_queues < 1 || tx_queues > 255) { 5106 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5107 tx_queues, BOND_DEFAULT_TX_QUEUES); 5108 tx_queues = BOND_DEFAULT_TX_QUEUES; 5109 } 5110 5111 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5112 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5113 all_slaves_active); 5114 all_slaves_active = 0; 5115 } 5116 5117 if (resend_igmp < 0 || resend_igmp > 255) { 5118 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 5119 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 5120 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 5121 } 5122 5123 bond_opt_initval(&newval, packets_per_slave); 5124 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 5125 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 5126 packets_per_slave, USHRT_MAX); 5127 packets_per_slave = 1; 5128 } 5129 5130 if (bond_mode == BOND_MODE_ALB) { 5131 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", 5132 updelay); 5133 } 5134 5135 if (!miimon) { 5136 if (updelay || downdelay) { 5137 /* just warn the user the up/down delay will have 5138 * no effect since miimon is zero... 5139 */ 5140 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", 5141 updelay, downdelay); 5142 } 5143 } else { 5144 /* don't allow arp monitoring */ 5145 if (arp_interval) { 5146 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 5147 miimon, arp_interval); 5148 arp_interval = 0; 5149 } 5150 5151 if ((updelay % miimon) != 0) { 5152 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 5153 updelay, miimon, (updelay / miimon) * miimon); 5154 } 5155 5156 updelay /= miimon; 5157 5158 if ((downdelay % miimon) != 0) { 5159 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 5160 downdelay, miimon, 5161 (downdelay / miimon) * miimon); 5162 } 5163 5164 downdelay /= miimon; 5165 } 5166 5167 if (arp_interval < 0) { 5168 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5169 arp_interval, INT_MAX); 5170 arp_interval = 0; 5171 } 5172 5173 for (arp_ip_count = 0, i = 0; 5174 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 5175 __be32 ip; 5176 5177 /* not a complete check, but good enough to catch mistakes */ 5178 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 5179 !bond_is_ip_target_ok(ip)) { 5180 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 5181 arp_ip_target[i]); 5182 arp_interval = 0; 5183 } else { 5184 if (bond_get_targets_ip(arp_target, ip) == -1) 5185 arp_target[arp_ip_count++] = ip; 5186 else 5187 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 5188 &ip); 5189 } 5190 } 5191 5192 if (arp_interval && !arp_ip_count) { 5193 /* don't allow arping if no arp_ip_target given... */ 5194 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 5195 arp_interval); 5196 arp_interval = 0; 5197 } 5198 5199 if (arp_validate) { 5200 if (!arp_interval) { 5201 pr_err("arp_validate requires arp_interval\n"); 5202 return -EINVAL; 5203 } 5204 5205 bond_opt_initstr(&newval, arp_validate); 5206 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 5207 &newval); 5208 if (!valptr) { 5209 pr_err("Error: invalid arp_validate \"%s\"\n", 5210 arp_validate); 5211 return -EINVAL; 5212 } 5213 arp_validate_value = valptr->value; 5214 } else { 5215 arp_validate_value = 0; 5216 } 5217 5218 if (arp_all_targets) { 5219 bond_opt_initstr(&newval, arp_all_targets); 5220 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 5221 &newval); 5222 if (!valptr) { 5223 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 5224 arp_all_targets); 5225 arp_all_targets_value = 0; 5226 } else { 5227 arp_all_targets_value = valptr->value; 5228 } 5229 } 5230 5231 if (miimon) { 5232 pr_info("MII link monitoring set to %d ms\n", miimon); 5233 } else if (arp_interval) { 5234 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 5235 arp_validate_value); 5236 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 5237 arp_interval, valptr->string, arp_ip_count); 5238 5239 for (i = 0; i < arp_ip_count; i++) 5240 pr_cont(" %s", arp_ip_target[i]); 5241 5242 pr_cont("\n"); 5243 5244 } else if (max_bonds) { 5245 /* miimon and arp_interval not set, we need one so things 5246 * work as expected, see bonding.txt for details 5247 */ 5248 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"); 5249 } 5250 5251 if (primary && !bond_mode_uses_primary(bond_mode)) { 5252 /* currently, using a primary only makes sense 5253 * in active backup, TLB or ALB modes 5254 */ 5255 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 5256 primary, bond_mode_name(bond_mode)); 5257 primary = NULL; 5258 } 5259 5260 if (primary && primary_reselect) { 5261 bond_opt_initstr(&newval, primary_reselect); 5262 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 5263 &newval); 5264 if (!valptr) { 5265 pr_err("Error: Invalid primary_reselect \"%s\"\n", 5266 primary_reselect); 5267 return -EINVAL; 5268 } 5269 primary_reselect_value = valptr->value; 5270 } else { 5271 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 5272 } 5273 5274 if (fail_over_mac) { 5275 bond_opt_initstr(&newval, fail_over_mac); 5276 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 5277 &newval); 5278 if (!valptr) { 5279 pr_err("Error: invalid fail_over_mac \"%s\"\n", 5280 fail_over_mac); 5281 return -EINVAL; 5282 } 5283 fail_over_mac_value = valptr->value; 5284 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 5285 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 5286 } else { 5287 fail_over_mac_value = BOND_FOM_NONE; 5288 } 5289 5290 bond_opt_initstr(&newval, "default"); 5291 valptr = bond_opt_parse( 5292 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 5293 &newval); 5294 if (!valptr) { 5295 pr_err("Error: No ad_actor_sys_prio default value"); 5296 return -EINVAL; 5297 } 5298 ad_actor_sys_prio = valptr->value; 5299 5300 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 5301 &newval); 5302 if (!valptr) { 5303 pr_err("Error: No ad_user_port_key default value"); 5304 return -EINVAL; 5305 } 5306 ad_user_port_key = valptr->value; 5307 5308 bond_opt_initstr(&newval, "default"); 5309 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 5310 if (!valptr) { 5311 pr_err("Error: No tlb_dynamic_lb default value"); 5312 return -EINVAL; 5313 } 5314 tlb_dynamic_lb = valptr->value; 5315 5316 if (lp_interval == 0) { 5317 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 5318 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 5319 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 5320 } 5321 5322 /* fill params struct with the proper values */ 5323 params->mode = bond_mode; 5324 params->xmit_policy = xmit_hashtype; 5325 params->miimon = miimon; 5326 params->num_peer_notif = num_peer_notif; 5327 params->arp_interval = arp_interval; 5328 params->arp_validate = arp_validate_value; 5329 params->arp_all_targets = arp_all_targets_value; 5330 params->updelay = updelay; 5331 params->downdelay = downdelay; 5332 params->peer_notif_delay = 0; 5333 params->use_carrier = use_carrier; 5334 params->lacp_fast = lacp_fast; 5335 params->primary[0] = 0; 5336 params->primary_reselect = primary_reselect_value; 5337 params->fail_over_mac = fail_over_mac_value; 5338 params->tx_queues = tx_queues; 5339 params->all_slaves_active = all_slaves_active; 5340 params->resend_igmp = resend_igmp; 5341 params->min_links = min_links; 5342 params->lp_interval = lp_interval; 5343 params->packets_per_slave = packets_per_slave; 5344 params->tlb_dynamic_lb = tlb_dynamic_lb; 5345 params->ad_actor_sys_prio = ad_actor_sys_prio; 5346 eth_zero_addr(params->ad_actor_system); 5347 params->ad_user_port_key = ad_user_port_key; 5348 if (packets_per_slave > 0) { 5349 params->reciprocal_packets_per_slave = 5350 reciprocal_value(packets_per_slave); 5351 } else { 5352 /* reciprocal_packets_per_slave is unused if 5353 * packets_per_slave is 0 or 1, just initialize it 5354 */ 5355 params->reciprocal_packets_per_slave = 5356 (struct reciprocal_value) { 0 }; 5357 } 5358 5359 if (primary) { 5360 strncpy(params->primary, primary, IFNAMSIZ); 5361 params->primary[IFNAMSIZ - 1] = 0; 5362 } 5363 5364 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 5365 5366 return 0; 5367} 5368 5369/* Called from registration process */ 5370static int bond_init(struct net_device *bond_dev) 5371{ 5372 struct bonding *bond = netdev_priv(bond_dev); 5373 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 5374 5375 netdev_dbg(bond_dev, "Begin bond_init\n"); 5376 5377 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 5378 if (!bond->wq) 5379 return -ENOMEM; 5380 5381 bond->notifier_ctx = false; 5382 5383 spin_lock_init(&bond->stats_lock); 5384 netdev_lockdep_set_classes(bond_dev); 5385 5386 list_add_tail(&bond->bond_list, &bn->dev_list); 5387 5388 bond_prepare_sysfs_group(bond); 5389 5390 bond_debug_register(bond); 5391 5392 /* Ensure valid dev_addr */ 5393 if (is_zero_ether_addr(bond_dev->dev_addr) && 5394 bond_dev->addr_assign_type == NET_ADDR_PERM) 5395 eth_hw_addr_random(bond_dev); 5396 5397 return 0; 5398} 5399 5400unsigned int bond_get_num_tx_queues(void) 5401{ 5402 return tx_queues; 5403} 5404 5405/* Create a new bond based on the specified name and bonding parameters. 5406 * If name is NULL, obtain a suitable "bond%d" name for us. 5407 * Caller must NOT hold rtnl_lock; we need to release it here before we 5408 * set up our sysfs entries. 5409 */ 5410int bond_create(struct net *net, const char *name) 5411{ 5412 struct net_device *bond_dev; 5413 struct bonding *bond; 5414 struct alb_bond_info *bond_info; 5415 int res; 5416 5417 rtnl_lock(); 5418 5419 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 5420 name ? name : "bond%d", NET_NAME_UNKNOWN, 5421 bond_setup, tx_queues); 5422 if (!bond_dev) { 5423 pr_err("%s: eek! can't alloc netdev!\n", name); 5424 rtnl_unlock(); 5425 return -ENOMEM; 5426 } 5427 5428 /* 5429 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 5430 * It is set to 0 by default which is wrong. 5431 */ 5432 bond = netdev_priv(bond_dev); 5433 bond_info = &(BOND_ALB_INFO(bond)); 5434 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 5435 5436 dev_net_set(bond_dev, net); 5437 bond_dev->rtnl_link_ops = &bond_link_ops; 5438 5439 res = register_netdevice(bond_dev); 5440 if (res < 0) { 5441 free_netdev(bond_dev); 5442 rtnl_unlock(); 5443 5444 return res; 5445 } 5446 5447 netif_carrier_off(bond_dev); 5448 5449 bond_work_init_all(bond); 5450 5451 rtnl_unlock(); 5452 return 0; 5453} 5454 5455static int __net_init bond_net_init(struct net *net) 5456{ 5457 struct bond_net *bn = net_generic(net, bond_net_id); 5458 5459 bn->net = net; 5460 INIT_LIST_HEAD(&bn->dev_list); 5461 5462 bond_create_proc_dir(bn); 5463 bond_create_sysfs(bn); 5464 5465 return 0; 5466} 5467 5468static void __net_exit bond_net_exit(struct net *net) 5469{ 5470 struct bond_net *bn = net_generic(net, bond_net_id); 5471 struct bonding *bond, *tmp_bond; 5472 LIST_HEAD(list); 5473 5474 bond_destroy_sysfs(bn); 5475 5476 /* Kill off any bonds created after unregistering bond rtnl ops */ 5477 rtnl_lock(); 5478 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 5479 unregister_netdevice_queue(bond->dev, &list); 5480 unregister_netdevice_many(&list); 5481 rtnl_unlock(); 5482 5483 bond_destroy_proc_dir(bn); 5484} 5485 5486static struct pernet_operations bond_net_ops = { 5487 .init = bond_net_init, 5488 .exit = bond_net_exit, 5489 .id = &bond_net_id, 5490 .size = sizeof(struct bond_net), 5491}; 5492 5493static int __init bonding_init(void) 5494{ 5495 int i; 5496 int res; 5497 5498 res = bond_check_params(&bonding_defaults); 5499 if (res) 5500 goto out; 5501 5502 res = register_pernet_subsys(&bond_net_ops); 5503 if (res) 5504 goto out; 5505 5506 res = bond_netlink_init(); 5507 if (res) 5508 goto err_link; 5509 5510 bond_create_debugfs(); 5511 5512 for (i = 0; i < max_bonds; i++) { 5513 res = bond_create(&init_net, NULL); 5514 if (res) 5515 goto err; 5516 } 5517 5518 skb_flow_dissector_init(&flow_keys_bonding, 5519 flow_keys_bonding_keys, 5520 ARRAY_SIZE(flow_keys_bonding_keys)); 5521 5522 register_netdevice_notifier(&bond_netdev_notifier); 5523out: 5524 return res; 5525err: 5526 bond_destroy_debugfs(); 5527 bond_netlink_fini(); 5528err_link: 5529 unregister_pernet_subsys(&bond_net_ops); 5530 goto out; 5531 5532} 5533 5534static void __exit bonding_exit(void) 5535{ 5536 unregister_netdevice_notifier(&bond_netdev_notifier); 5537 5538 bond_destroy_debugfs(); 5539 5540 bond_netlink_fini(); 5541 unregister_pernet_subsys(&bond_net_ops); 5542 5543#ifdef CONFIG_NET_POLL_CONTROLLER 5544 /* Make sure we don't have an imbalance on our netpoll blocking */ 5545 WARN_ON(atomic_read(&netpoll_block_tx)); 5546#endif 5547} 5548 5549module_init(bonding_init); 5550module_exit(bonding_exit); 5551MODULE_LICENSE("GPL"); 5552MODULE_DESCRIPTION(DRV_DESCRIPTION); 5553MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 5554