1 /* 2 * originally based on the dummy device. 3 * 4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 5 * Licensed under the GPL. Based on dummy.c, and eql.c devices. 6 * 7 * bonding.c: an Ethernet Bonding driver 8 * 9 * This is useful to talk to a Cisco EtherChannel compatible equipment: 10 * Cisco 5500 11 * Sun Trunking (Solaris) 12 * Alteon AceDirector Trunks 13 * Linux Bonding 14 * and probably many L2 switches ... 15 * 16 * How it works: 17 * ifconfig bond0 ipaddress netmask up 18 * will setup a network device, with an ip address. No mac address 19 * will be assigned at this time. The hw mac address will come from 20 * the first slave bonded to the channel. All slaves will then use 21 * this hw mac address. 22 * 23 * ifconfig bond0 down 24 * will release all slaves, marking them as down. 25 * 26 * ifenslave bond0 eth0 27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 28 * a: be used as initial mac address 29 * b: if a hw mac address already is there, eth0's hw mac address 30 * will then be set from bond0. 31 * 32 */ 33 34 #include <linux/kernel.h> 35 #include <linux/module.h> 36 #include <linux/types.h> 37 #include <linux/fcntl.h> 38 #include <linux/interrupt.h> 39 #include <linux/ptrace.h> 40 #include <linux/ioport.h> 41 #include <linux/in.h> 42 #include <net/ip.h> 43 #include <linux/ip.h> 44 #include <linux/tcp.h> 45 #include <linux/udp.h> 46 #include <linux/slab.h> 47 #include <linux/string.h> 48 #include <linux/init.h> 49 #include <linux/timer.h> 50 #include <linux/socket.h> 51 #include <linux/ctype.h> 52 #include <linux/inet.h> 53 #include <linux/bitops.h> 54 #include <linux/io.h> 55 #include <asm/dma.h> 56 #include <linux/uaccess.h> 57 #include <linux/errno.h> 58 #include <linux/netdevice.h> 59 #include <linux/inetdevice.h> 60 #include <linux/igmp.h> 61 #include <linux/etherdevice.h> 62 #include <linux/skbuff.h> 63 #include <net/sock.h> 64 #include <linux/rtnetlink.h> 65 #include <linux/smp.h> 66 #include <linux/if_ether.h> 67 #include <net/arp.h> 68 #include <linux/mii.h> 69 #include <linux/ethtool.h> 70 #include <linux/if_vlan.h> 71 #include <linux/if_bonding.h> 72 #include <linux/jiffies.h> 73 #include <linux/preempt.h> 74 #include <net/route.h> 75 #include <net/net_namespace.h> 76 #include <net/netns/generic.h> 77 #include <net/pkt_sched.h> 78 #include <linux/rculist.h> 79 #include <net/flow_dissector.h> 80 #include <net/switchdev.h> 81 #include <net/bonding.h> 82 #include <net/bond_3ad.h> 83 #include <net/bond_alb.h> 84 85 #include "bonding_priv.h" 86 87 /*---------------------------- Module parameters ----------------------------*/ 88 89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 90 91 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 92 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 93 static int num_peer_notif = 1; 94 static int miimon; 95 static int updelay; 96 static int downdelay; 97 static int use_carrier = 1; 98 static char *mode; 99 static char *primary; 100 static char *primary_reselect; 101 static char *lacp_rate; 102 static int min_links; 103 static char *ad_select; 104 static char *xmit_hash_policy; 105 static int arp_interval; 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 107 static char *arp_validate; 108 static char *arp_all_targets; 109 static char *fail_over_mac; 110 static int all_slaves_active; 111 static struct bond_params bonding_defaults; 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 113 static int packets_per_slave = 1; 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 115 116 module_param(max_bonds, int, 0); 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 118 module_param(tx_queues, int, 0); 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 122 "failover event (alias of num_unsol_na)"); 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 125 "failover event (alias of num_grat_arp)"); 126 module_param(miimon, int, 0); 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 128 module_param(updelay, int, 0); 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 130 module_param(downdelay, int, 0); 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 132 "in milliseconds"); 133 module_param(use_carrier, int, 0); 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 135 "0 for off, 1 for on (default)"); 136 module_param(mode, charp, 0); 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 138 "1 for active-backup, 2 for balance-xor, " 139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 140 "6 for balance-alb"); 141 module_param(primary, charp, 0); 142 MODULE_PARM_DESC(primary, "Primary network device to use"); 143 module_param(primary_reselect, charp, 0); 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 145 "once it comes up; " 146 "0 for always (default), " 147 "1 for only if speed of primary is " 148 "better, " 149 "2 for only on active slave " 150 "failure"); 151 module_param(lacp_rate, charp, 0); 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 153 "0 for slow, 1 for fast"); 154 module_param(ad_select, charp, 0); 155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 156 "0 for stable (default), 1 for bandwidth, " 157 "2 for count"); 158 module_param(min_links, int, 0); 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 160 161 module_param(xmit_hash_policy, charp, 0); 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 163 "0 for layer 2 (default), 1 for layer 3+4, " 164 "2 for layer 2+3, 3 for encap layer 2+3, " 165 "4 for encap layer 3+4"); 166 module_param(arp_interval, int, 0); 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 168 module_param_array(arp_ip_target, charp, NULL, 0); 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 170 module_param(arp_validate, charp, 0); 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 172 "0 for none (default), 1 for active, " 173 "2 for backup, 3 for all"); 174 module_param(arp_all_targets, charp, 0); 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 176 module_param(fail_over_mac, charp, 0); 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 178 "the same MAC; 0 for none (default), " 179 "1 for active, 2 for follow"); 180 module_param(all_slaves_active, int, 0); 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 182 "by setting active flag for all slaves; " 183 "0 for never (default), 1 for always."); 184 module_param(resend_igmp, int, 0); 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 186 "link failure"); 187 module_param(packets_per_slave, int, 0); 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 189 "mode; 0 for a random slave, 1 packet per " 190 "slave (default), >1 packets per slave."); 191 module_param(lp_interval, uint, 0); 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 193 "the bonding driver sends learning packets to " 194 "each slaves peer switch. The default is 1."); 195 196 /*----------------------------- Global variables ----------------------------*/ 197 198 #ifdef CONFIG_NET_POLL_CONTROLLER 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 200 #endif 201 202 unsigned int bond_net_id __read_mostly; 203 204 /*-------------------------- Forward declarations ---------------------------*/ 205 206 static int bond_init(struct net_device *bond_dev); 207 static void bond_uninit(struct net_device *bond_dev); 208 static void bond_get_stats(struct net_device *bond_dev, 209 struct rtnl_link_stats64 *stats); 210 static void bond_slave_arr_handler(struct work_struct *work); 211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 212 int mod); 213 static void bond_netdev_notify_work(struct work_struct *work); 214 215 /*---------------------------- General routines -----------------------------*/ 216 217 const char *bond_mode_name(int mode) 218 { 219 static const char *names[] = { 220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 222 [BOND_MODE_XOR] = "load balancing (xor)", 223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 225 [BOND_MODE_TLB] = "transmit load balancing", 226 [BOND_MODE_ALB] = "adaptive load balancing", 227 }; 228 229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 230 return "unknown"; 231 232 return names[mode]; 233 } 234 235 /*---------------------------------- VLAN -----------------------------------*/ 236 237 /** 238 * bond_dev_queue_xmit - Prepare skb for xmit. 239 * 240 * @bond: bond device that got this skb for tx. 241 * @skb: hw accel VLAN tagged skb to transmit 242 * @slave_dev: slave that is supposed to xmit this skbuff 243 */ 244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 245 struct net_device *slave_dev) 246 { 247 skb->dev = slave_dev; 248 249 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 251 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 252 253 if (unlikely(netpoll_tx_running(bond->dev))) 254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 255 else 256 dev_queue_xmit(skb); 257 } 258 259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 260 * We don't protect the slave list iteration with a lock because: 261 * a. This operation is performed in IOCTL context, 262 * b. The operation is protected by the RTNL semaphore in the 8021q code, 263 * c. Holding a lock with BH disabled while directly calling a base driver 264 * entry point is generally a BAD idea. 265 * 266 * The design of synchronization/protection for this operation in the 8021q 267 * module is good for one or more VLAN devices over a single physical device 268 * and cannot be extended for a teaming solution like bonding, so there is a 269 * potential race condition here where a net device from the vlan group might 270 * be referenced (either by a base driver or the 8021q code) while it is being 271 * removed from the system. However, it turns out we're not making matters 272 * worse, and if it works for regular VLAN usage it will work here too. 273 */ 274 275 /** 276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 277 * @bond_dev: bonding net device that got called 278 * @vid: vlan id being added 279 */ 280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 281 __be16 proto, u16 vid) 282 { 283 struct bonding *bond = netdev_priv(bond_dev); 284 struct slave *slave, *rollback_slave; 285 struct list_head *iter; 286 int res; 287 288 bond_for_each_slave(bond, slave, iter) { 289 res = vlan_vid_add(slave->dev, proto, vid); 290 if (res) 291 goto unwind; 292 } 293 294 return 0; 295 296 unwind: 297 /* unwind to the slave that failed */ 298 bond_for_each_slave(bond, rollback_slave, iter) { 299 if (rollback_slave == slave) 300 break; 301 302 vlan_vid_del(rollback_slave->dev, proto, vid); 303 } 304 305 return res; 306 } 307 308 /** 309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 310 * @bond_dev: bonding net device that got called 311 * @vid: vlan id being removed 312 */ 313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 314 __be16 proto, u16 vid) 315 { 316 struct bonding *bond = netdev_priv(bond_dev); 317 struct list_head *iter; 318 struct slave *slave; 319 320 bond_for_each_slave(bond, slave, iter) 321 vlan_vid_del(slave->dev, proto, vid); 322 323 if (bond_is_lb(bond)) 324 bond_alb_clear_vlan(bond, vid); 325 326 return 0; 327 } 328 329 /*------------------------------- Link status -------------------------------*/ 330 331 /* Set the carrier state for the master according to the state of its 332 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 333 * do special 802.3ad magic. 334 * 335 * Returns zero if carrier state does not change, nonzero if it does. 336 */ 337 int bond_set_carrier(struct bonding *bond) 338 { 339 struct list_head *iter; 340 struct slave *slave; 341 342 if (!bond_has_slaves(bond)) 343 goto down; 344 345 if (BOND_MODE(bond) == BOND_MODE_8023AD) 346 return bond_3ad_set_carrier(bond); 347 348 bond_for_each_slave(bond, slave, iter) { 349 if (slave->link == BOND_LINK_UP) { 350 if (!netif_carrier_ok(bond->dev)) { 351 netif_carrier_on(bond->dev); 352 return 1; 353 } 354 return 0; 355 } 356 } 357 358 down: 359 if (netif_carrier_ok(bond->dev)) { 360 netif_carrier_off(bond->dev); 361 return 1; 362 } 363 return 0; 364 } 365 366 /* Get link speed and duplex from the slave's base driver 367 * using ethtool. If for some reason the call fails or the 368 * values are invalid, set speed and duplex to -1, 369 * and return. Return 1 if speed or duplex settings are 370 * UNKNOWN; 0 otherwise. 371 */ 372 static int bond_update_speed_duplex(struct slave *slave) 373 { 374 struct net_device *slave_dev = slave->dev; 375 struct ethtool_link_ksettings ecmd; 376 int res; 377 378 slave->speed = SPEED_UNKNOWN; 379 slave->duplex = DUPLEX_UNKNOWN; 380 381 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 382 if (res < 0) 383 return 1; 384 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 385 return 1; 386 switch (ecmd.base.duplex) { 387 case DUPLEX_FULL: 388 case DUPLEX_HALF: 389 break; 390 default: 391 return 1; 392 } 393 394 slave->speed = ecmd.base.speed; 395 slave->duplex = ecmd.base.duplex; 396 397 return 0; 398 } 399 400 const char *bond_slave_link_status(s8 link) 401 { 402 switch (link) { 403 case BOND_LINK_UP: 404 return "up"; 405 case BOND_LINK_FAIL: 406 return "going down"; 407 case BOND_LINK_DOWN: 408 return "down"; 409 case BOND_LINK_BACK: 410 return "going back"; 411 default: 412 return "unknown"; 413 } 414 } 415 416 /* if <dev> supports MII link status reporting, check its link status. 417 * 418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 419 * depending upon the setting of the use_carrier parameter. 420 * 421 * Return either BMSR_LSTATUS, meaning that the link is up (or we 422 * can't tell and just pretend it is), or 0, meaning that the link is 423 * down. 424 * 425 * If reporting is non-zero, instead of faking link up, return -1 if 426 * both ETHTOOL and MII ioctls fail (meaning the device does not 427 * support them). If use_carrier is set, return whatever it says. 428 * It'd be nice if there was a good way to tell if a driver supports 429 * netif_carrier, but there really isn't. 430 */ 431 static int bond_check_dev_link(struct bonding *bond, 432 struct net_device *slave_dev, int reporting) 433 { 434 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 435 int (*ioctl)(struct net_device *, struct ifreq *, int); 436 struct ifreq ifr; 437 struct mii_ioctl_data *mii; 438 439 if (!reporting && !netif_running(slave_dev)) 440 return 0; 441 442 if (bond->params.use_carrier) 443 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 444 445 /* Try to get link status using Ethtool first. */ 446 if (slave_dev->ethtool_ops->get_link) 447 return slave_dev->ethtool_ops->get_link(slave_dev) ? 448 BMSR_LSTATUS : 0; 449 450 /* Ethtool can't be used, fallback to MII ioctls. */ 451 ioctl = slave_ops->ndo_do_ioctl; 452 if (ioctl) { 453 /* TODO: set pointer to correct ioctl on a per team member 454 * bases to make this more efficient. that is, once 455 * we determine the correct ioctl, we will always 456 * call it and not the others for that team 457 * member. 458 */ 459 460 /* We cannot assume that SIOCGMIIPHY will also read a 461 * register; not all network drivers (e.g., e100) 462 * support that. 463 */ 464 465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 466 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 467 mii = if_mii(&ifr); 468 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 469 mii->reg_num = MII_BMSR; 470 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 471 return mii->val_out & BMSR_LSTATUS; 472 } 473 } 474 475 /* If reporting, report that either there's no dev->do_ioctl, 476 * or both SIOCGMIIREG and get_link failed (meaning that we 477 * cannot report link status). If not reporting, pretend 478 * we're ok. 479 */ 480 return reporting ? -1 : BMSR_LSTATUS; 481 } 482 483 /*----------------------------- Multicast list ------------------------------*/ 484 485 /* Push the promiscuity flag down to appropriate slaves */ 486 static int bond_set_promiscuity(struct bonding *bond, int inc) 487 { 488 struct list_head *iter; 489 int err = 0; 490 491 if (bond_uses_primary(bond)) { 492 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 493 494 if (curr_active) 495 err = dev_set_promiscuity(curr_active->dev, inc); 496 } else { 497 struct slave *slave; 498 499 bond_for_each_slave(bond, slave, iter) { 500 err = dev_set_promiscuity(slave->dev, inc); 501 if (err) 502 return err; 503 } 504 } 505 return err; 506 } 507 508 /* Push the allmulti flag down to all slaves */ 509 static int bond_set_allmulti(struct bonding *bond, int inc) 510 { 511 struct list_head *iter; 512 int err = 0; 513 514 if (bond_uses_primary(bond)) { 515 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 516 517 if (curr_active) 518 err = dev_set_allmulti(curr_active->dev, inc); 519 } else { 520 struct slave *slave; 521 522 bond_for_each_slave(bond, slave, iter) { 523 err = dev_set_allmulti(slave->dev, inc); 524 if (err) 525 return err; 526 } 527 } 528 return err; 529 } 530 531 /* Retrieve the list of registered multicast addresses for the bonding 532 * device and retransmit an IGMP JOIN request to the current active 533 * slave. 534 */ 535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 536 { 537 struct bonding *bond = container_of(work, struct bonding, 538 mcast_work.work); 539 540 if (!rtnl_trylock()) { 541 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 542 return; 543 } 544 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 545 546 if (bond->igmp_retrans > 1) { 547 bond->igmp_retrans--; 548 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 549 } 550 rtnl_unlock(); 551 } 552 553 /* Flush bond's hardware addresses from slave */ 554 static void bond_hw_addr_flush(struct net_device *bond_dev, 555 struct net_device *slave_dev) 556 { 557 struct bonding *bond = netdev_priv(bond_dev); 558 559 dev_uc_unsync(slave_dev, bond_dev); 560 dev_mc_unsync(slave_dev, bond_dev); 561 562 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 563 /* del lacpdu mc addr from mc list */ 564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 565 566 dev_mc_del(slave_dev, lacpdu_multicast); 567 } 568 } 569 570 /*--------------------------- Active slave change ---------------------------*/ 571 572 /* Update the hardware address list and promisc/allmulti for the new and 573 * old active slaves (if any). Modes that are not using primary keep all 574 * slaves up date at all times; only the modes that use primary need to call 575 * this function to swap these settings during a failover. 576 */ 577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 578 struct slave *old_active) 579 { 580 if (old_active) { 581 if (bond->dev->flags & IFF_PROMISC) 582 dev_set_promiscuity(old_active->dev, -1); 583 584 if (bond->dev->flags & IFF_ALLMULTI) 585 dev_set_allmulti(old_active->dev, -1); 586 587 bond_hw_addr_flush(bond->dev, old_active->dev); 588 } 589 590 if (new_active) { 591 /* FIXME: Signal errors upstream. */ 592 if (bond->dev->flags & IFF_PROMISC) 593 dev_set_promiscuity(new_active->dev, 1); 594 595 if (bond->dev->flags & IFF_ALLMULTI) 596 dev_set_allmulti(new_active->dev, 1); 597 598 netif_addr_lock_bh(bond->dev); 599 dev_uc_sync(new_active->dev, bond->dev); 600 dev_mc_sync(new_active->dev, bond->dev); 601 netif_addr_unlock_bh(bond->dev); 602 } 603 } 604 605 /** 606 * bond_set_dev_addr - clone slave's address to bond 607 * @bond_dev: bond net device 608 * @slave_dev: slave net device 609 * 610 * Should be called with RTNL held. 611 */ 612 static int bond_set_dev_addr(struct net_device *bond_dev, 613 struct net_device *slave_dev) 614 { 615 int err; 616 617 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n", 618 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len); 619 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 620 if (err) 621 return err; 622 623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 624 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 626 return 0; 627 } 628 629 static struct slave *bond_get_old_active(struct bonding *bond, 630 struct slave *new_active) 631 { 632 struct slave *slave; 633 struct list_head *iter; 634 635 bond_for_each_slave(bond, slave, iter) { 636 if (slave == new_active) 637 continue; 638 639 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 640 return slave; 641 } 642 643 return NULL; 644 } 645 646 /* bond_do_fail_over_mac 647 * 648 * Perform special MAC address swapping for fail_over_mac settings 649 * 650 * Called with RTNL 651 */ 652 static void bond_do_fail_over_mac(struct bonding *bond, 653 struct slave *new_active, 654 struct slave *old_active) 655 { 656 u8 tmp_mac[MAX_ADDR_LEN]; 657 struct sockaddr_storage ss; 658 int rv; 659 660 switch (bond->params.fail_over_mac) { 661 case BOND_FOM_ACTIVE: 662 if (new_active) { 663 rv = bond_set_dev_addr(bond->dev, new_active->dev); 664 if (rv) 665 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 666 -rv, bond->dev->name); 667 } 668 break; 669 case BOND_FOM_FOLLOW: 670 /* if new_active && old_active, swap them 671 * if just old_active, do nothing (going to no active slave) 672 * if just new_active, set new_active to bond's MAC 673 */ 674 if (!new_active) 675 return; 676 677 if (!old_active) 678 old_active = bond_get_old_active(bond, new_active); 679 680 if (old_active) { 681 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 682 new_active->dev->addr_len); 683 bond_hw_addr_copy(ss.__data, 684 old_active->dev->dev_addr, 685 old_active->dev->addr_len); 686 ss.ss_family = new_active->dev->type; 687 } else { 688 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 689 bond->dev->addr_len); 690 ss.ss_family = bond->dev->type; 691 } 692 693 rv = dev_set_mac_address(new_active->dev, 694 (struct sockaddr *)&ss, NULL); 695 if (rv) { 696 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 697 -rv, new_active->dev->name); 698 goto out; 699 } 700 701 if (!old_active) 702 goto out; 703 704 bond_hw_addr_copy(ss.__data, tmp_mac, 705 new_active->dev->addr_len); 706 ss.ss_family = old_active->dev->type; 707 708 rv = dev_set_mac_address(old_active->dev, 709 (struct sockaddr *)&ss, NULL); 710 if (rv) 711 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 712 -rv, new_active->dev->name); 713 out: 714 break; 715 default: 716 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 717 bond->params.fail_over_mac); 718 break; 719 } 720 721 } 722 723 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 724 { 725 struct slave *prim = rtnl_dereference(bond->primary_slave); 726 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 727 728 if (!prim || prim->link != BOND_LINK_UP) { 729 if (!curr || curr->link != BOND_LINK_UP) 730 return NULL; 731 return curr; 732 } 733 734 if (bond->force_primary) { 735 bond->force_primary = false; 736 return prim; 737 } 738 739 if (!curr || curr->link != BOND_LINK_UP) 740 return prim; 741 742 /* At this point, prim and curr are both up */ 743 switch (bond->params.primary_reselect) { 744 case BOND_PRI_RESELECT_ALWAYS: 745 return prim; 746 case BOND_PRI_RESELECT_BETTER: 747 if (prim->speed < curr->speed) 748 return curr; 749 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 750 return curr; 751 return prim; 752 case BOND_PRI_RESELECT_FAILURE: 753 return curr; 754 default: 755 netdev_err(bond->dev, "impossible primary_reselect %d\n", 756 bond->params.primary_reselect); 757 return curr; 758 } 759 } 760 761 /** 762 * bond_find_best_slave - select the best available slave to be the active one 763 * @bond: our bonding struct 764 */ 765 static struct slave *bond_find_best_slave(struct bonding *bond) 766 { 767 struct slave *slave, *bestslave = NULL; 768 struct list_head *iter; 769 int mintime = bond->params.updelay; 770 771 slave = bond_choose_primary_or_current(bond); 772 if (slave) 773 return slave; 774 775 bond_for_each_slave(bond, slave, iter) { 776 if (slave->link == BOND_LINK_UP) 777 return slave; 778 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 779 slave->delay < mintime) { 780 mintime = slave->delay; 781 bestslave = slave; 782 } 783 } 784 785 return bestslave; 786 } 787 788 static bool bond_should_notify_peers(struct bonding *bond) 789 { 790 struct slave *slave; 791 792 rcu_read_lock(); 793 slave = rcu_dereference(bond->curr_active_slave); 794 rcu_read_unlock(); 795 796 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 797 slave ? slave->dev->name : "NULL"); 798 799 if (!slave || !bond->send_peer_notif || 800 !netif_carrier_ok(bond->dev) || 801 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 802 return false; 803 804 return true; 805 } 806 807 /** 808 * change_active_interface - change the active slave into the specified one 809 * @bond: our bonding struct 810 * @new: the new slave to make the active one 811 * 812 * Set the new slave to the bond's settings and unset them on the old 813 * curr_active_slave. 814 * Setting include flags, mc-list, promiscuity, allmulti, etc. 815 * 816 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 817 * because it is apparently the best available slave we have, even though its 818 * updelay hasn't timed out yet. 819 * 820 * Caller must hold RTNL. 821 */ 822 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 823 { 824 struct slave *old_active; 825 826 ASSERT_RTNL(); 827 828 old_active = rtnl_dereference(bond->curr_active_slave); 829 830 if (old_active == new_active) 831 return; 832 833 if (new_active) { 834 new_active->last_link_up = jiffies; 835 836 if (new_active->link == BOND_LINK_BACK) { 837 if (bond_uses_primary(bond)) { 838 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n", 839 new_active->dev->name, 840 (bond->params.updelay - new_active->delay) * bond->params.miimon); 841 } 842 843 new_active->delay = 0; 844 bond_set_slave_link_state(new_active, BOND_LINK_UP, 845 BOND_SLAVE_NOTIFY_NOW); 846 847 if (BOND_MODE(bond) == BOND_MODE_8023AD) 848 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 849 850 if (bond_is_lb(bond)) 851 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 852 } else { 853 if (bond_uses_primary(bond)) { 854 netdev_info(bond->dev, "making interface %s the new active one\n", 855 new_active->dev->name); 856 } 857 } 858 } 859 860 if (bond_uses_primary(bond)) 861 bond_hw_addr_swap(bond, new_active, old_active); 862 863 if (bond_is_lb(bond)) { 864 bond_alb_handle_active_change(bond, new_active); 865 if (old_active) 866 bond_set_slave_inactive_flags(old_active, 867 BOND_SLAVE_NOTIFY_NOW); 868 if (new_active) 869 bond_set_slave_active_flags(new_active, 870 BOND_SLAVE_NOTIFY_NOW); 871 } else { 872 rcu_assign_pointer(bond->curr_active_slave, new_active); 873 } 874 875 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 876 if (old_active) 877 bond_set_slave_inactive_flags(old_active, 878 BOND_SLAVE_NOTIFY_NOW); 879 880 if (new_active) { 881 bool should_notify_peers = false; 882 883 bond_set_slave_active_flags(new_active, 884 BOND_SLAVE_NOTIFY_NOW); 885 886 if (bond->params.fail_over_mac) 887 bond_do_fail_over_mac(bond, new_active, 888 old_active); 889 890 if (netif_running(bond->dev)) { 891 bond->send_peer_notif = 892 bond->params.num_peer_notif; 893 should_notify_peers = 894 bond_should_notify_peers(bond); 895 } 896 897 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 898 if (should_notify_peers) 899 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 900 bond->dev); 901 } 902 } 903 904 /* resend IGMP joins since active slave has changed or 905 * all were sent on curr_active_slave. 906 * resend only if bond is brought up with the affected 907 * bonding modes and the retransmission is enabled 908 */ 909 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 910 ((bond_uses_primary(bond) && new_active) || 911 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 912 bond->igmp_retrans = bond->params.resend_igmp; 913 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 914 } 915 } 916 917 /** 918 * bond_select_active_slave - select a new active slave, if needed 919 * @bond: our bonding struct 920 * 921 * This functions should be called when one of the following occurs: 922 * - The old curr_active_slave has been released or lost its link. 923 * - The primary_slave has got its link back. 924 * - A slave has got its link back and there's no old curr_active_slave. 925 * 926 * Caller must hold RTNL. 927 */ 928 void bond_select_active_slave(struct bonding *bond) 929 { 930 struct slave *best_slave; 931 int rv; 932 933 ASSERT_RTNL(); 934 935 best_slave = bond_find_best_slave(bond); 936 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 937 bond_change_active_slave(bond, best_slave); 938 rv = bond_set_carrier(bond); 939 if (!rv) 940 return; 941 942 if (netif_carrier_ok(bond->dev)) 943 netdev_info(bond->dev, "first active interface up!\n"); 944 else 945 netdev_info(bond->dev, "now running without any active interface!\n"); 946 } 947 } 948 949 #ifdef CONFIG_NET_POLL_CONTROLLER 950 static inline int slave_enable_netpoll(struct slave *slave) 951 { 952 struct netpoll *np; 953 int err = 0; 954 955 np = kzalloc(sizeof(*np), GFP_KERNEL); 956 err = -ENOMEM; 957 if (!np) 958 goto out; 959 960 err = __netpoll_setup(np, slave->dev); 961 if (err) { 962 kfree(np); 963 goto out; 964 } 965 slave->np = np; 966 out: 967 return err; 968 } 969 static inline void slave_disable_netpoll(struct slave *slave) 970 { 971 struct netpoll *np = slave->np; 972 973 if (!np) 974 return; 975 976 slave->np = NULL; 977 978 __netpoll_free(np); 979 } 980 981 static void bond_poll_controller(struct net_device *bond_dev) 982 { 983 struct bonding *bond = netdev_priv(bond_dev); 984 struct slave *slave = NULL; 985 struct list_head *iter; 986 struct ad_info ad_info; 987 988 if (BOND_MODE(bond) == BOND_MODE_8023AD) 989 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 990 return; 991 992 bond_for_each_slave_rcu(bond, slave, iter) { 993 if (!bond_slave_is_up(slave)) 994 continue; 995 996 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 997 struct aggregator *agg = 998 SLAVE_AD_INFO(slave)->port.aggregator; 999 1000 if (agg && 1001 agg->aggregator_identifier != ad_info.aggregator_id) 1002 continue; 1003 } 1004 1005 netpoll_poll_dev(slave->dev); 1006 } 1007 } 1008 1009 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1010 { 1011 struct bonding *bond = netdev_priv(bond_dev); 1012 struct list_head *iter; 1013 struct slave *slave; 1014 1015 bond_for_each_slave(bond, slave, iter) 1016 if (bond_slave_is_up(slave)) 1017 slave_disable_netpoll(slave); 1018 } 1019 1020 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1021 { 1022 struct bonding *bond = netdev_priv(dev); 1023 struct list_head *iter; 1024 struct slave *slave; 1025 int err = 0; 1026 1027 bond_for_each_slave(bond, slave, iter) { 1028 err = slave_enable_netpoll(slave); 1029 if (err) { 1030 bond_netpoll_cleanup(dev); 1031 break; 1032 } 1033 } 1034 return err; 1035 } 1036 #else 1037 static inline int slave_enable_netpoll(struct slave *slave) 1038 { 1039 return 0; 1040 } 1041 static inline void slave_disable_netpoll(struct slave *slave) 1042 { 1043 } 1044 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1045 { 1046 } 1047 #endif 1048 1049 /*---------------------------------- IOCTL ----------------------------------*/ 1050 1051 static netdev_features_t bond_fix_features(struct net_device *dev, 1052 netdev_features_t features) 1053 { 1054 struct bonding *bond = netdev_priv(dev); 1055 struct list_head *iter; 1056 netdev_features_t mask; 1057 struct slave *slave; 1058 1059 mask = features; 1060 1061 features &= ~NETIF_F_ONE_FOR_ALL; 1062 features |= NETIF_F_ALL_FOR_ALL; 1063 1064 bond_for_each_slave(bond, slave, iter) { 1065 features = netdev_increment_features(features, 1066 slave->dev->features, 1067 mask); 1068 } 1069 features = netdev_add_tso_features(features, mask); 1070 1071 return features; 1072 } 1073 1074 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1075 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1076 NETIF_F_HIGHDMA | NETIF_F_LRO) 1077 1078 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1079 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1080 1081 static void bond_compute_features(struct bonding *bond) 1082 { 1083 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1084 IFF_XMIT_DST_RELEASE_PERM; 1085 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1086 netdev_features_t enc_features = BOND_ENC_FEATURES; 1087 struct net_device *bond_dev = bond->dev; 1088 struct list_head *iter; 1089 struct slave *slave; 1090 unsigned short max_hard_header_len = ETH_HLEN; 1091 unsigned int gso_max_size = GSO_MAX_SIZE; 1092 u16 gso_max_segs = GSO_MAX_SEGS; 1093 1094 if (!bond_has_slaves(bond)) 1095 goto done; 1096 vlan_features &= NETIF_F_ALL_FOR_ALL; 1097 1098 bond_for_each_slave(bond, slave, iter) { 1099 vlan_features = netdev_increment_features(vlan_features, 1100 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1101 1102 enc_features = netdev_increment_features(enc_features, 1103 slave->dev->hw_enc_features, 1104 BOND_ENC_FEATURES); 1105 dst_release_flag &= slave->dev->priv_flags; 1106 if (slave->dev->hard_header_len > max_hard_header_len) 1107 max_hard_header_len = slave->dev->hard_header_len; 1108 1109 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1110 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1111 } 1112 bond_dev->hard_header_len = max_hard_header_len; 1113 1114 done: 1115 bond_dev->vlan_features = vlan_features; 1116 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1117 NETIF_F_GSO_UDP_L4; 1118 bond_dev->gso_max_segs = gso_max_segs; 1119 netif_set_gso_max_size(bond_dev, gso_max_size); 1120 1121 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1122 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1123 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1124 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1125 1126 netdev_change_features(bond_dev); 1127 } 1128 1129 static void bond_setup_by_slave(struct net_device *bond_dev, 1130 struct net_device *slave_dev) 1131 { 1132 bond_dev->header_ops = slave_dev->header_ops; 1133 1134 bond_dev->type = slave_dev->type; 1135 bond_dev->hard_header_len = slave_dev->hard_header_len; 1136 bond_dev->addr_len = slave_dev->addr_len; 1137 1138 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1139 slave_dev->addr_len); 1140 } 1141 1142 /* On bonding slaves other than the currently active slave, suppress 1143 * duplicates except for alb non-mcast/bcast. 1144 */ 1145 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1146 struct slave *slave, 1147 struct bonding *bond) 1148 { 1149 if (bond_is_slave_inactive(slave)) { 1150 if (BOND_MODE(bond) == BOND_MODE_ALB && 1151 skb->pkt_type != PACKET_BROADCAST && 1152 skb->pkt_type != PACKET_MULTICAST) 1153 return false; 1154 return true; 1155 } 1156 return false; 1157 } 1158 1159 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1160 { 1161 struct sk_buff *skb = *pskb; 1162 struct slave *slave; 1163 struct bonding *bond; 1164 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1165 struct slave *); 1166 int ret = RX_HANDLER_ANOTHER; 1167 1168 skb = skb_share_check(skb, GFP_ATOMIC); 1169 if (unlikely(!skb)) 1170 return RX_HANDLER_CONSUMED; 1171 1172 *pskb = skb; 1173 1174 slave = bond_slave_get_rcu(skb->dev); 1175 bond = slave->bond; 1176 1177 recv_probe = READ_ONCE(bond->recv_probe); 1178 if (recv_probe) { 1179 ret = recv_probe(skb, bond, slave); 1180 if (ret == RX_HANDLER_CONSUMED) { 1181 consume_skb(skb); 1182 return ret; 1183 } 1184 } 1185 1186 /* 1187 * For packets determined by bond_should_deliver_exact_match() call to 1188 * be suppressed we want to make an exception for link-local packets. 1189 * This is necessary for e.g. LLDP daemons to be able to monitor 1190 * inactive slave links without being forced to bind to them 1191 * explicitly. 1192 * 1193 * At the same time, packets that are passed to the bonding master 1194 * (including link-local ones) can have their originating interface 1195 * determined via PACKET_ORIGDEV socket option. 1196 */ 1197 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1198 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1199 return RX_HANDLER_PASS; 1200 return RX_HANDLER_EXACT; 1201 } 1202 1203 skb->dev = bond->dev; 1204 1205 if (BOND_MODE(bond) == BOND_MODE_ALB && 1206 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1207 skb->pkt_type == PACKET_HOST) { 1208 1209 if (unlikely(skb_cow_head(skb, 1210 skb->data - skb_mac_header(skb)))) { 1211 kfree_skb(skb); 1212 return RX_HANDLER_CONSUMED; 1213 } 1214 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1215 bond->dev->addr_len); 1216 } 1217 1218 return ret; 1219 } 1220 1221 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1222 { 1223 switch (BOND_MODE(bond)) { 1224 case BOND_MODE_ROUNDROBIN: 1225 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1226 case BOND_MODE_ACTIVEBACKUP: 1227 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1228 case BOND_MODE_BROADCAST: 1229 return NETDEV_LAG_TX_TYPE_BROADCAST; 1230 case BOND_MODE_XOR: 1231 case BOND_MODE_8023AD: 1232 return NETDEV_LAG_TX_TYPE_HASH; 1233 default: 1234 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1235 } 1236 } 1237 1238 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1239 enum netdev_lag_tx_type type) 1240 { 1241 if (type != NETDEV_LAG_TX_TYPE_HASH) 1242 return NETDEV_LAG_HASH_NONE; 1243 1244 switch (bond->params.xmit_policy) { 1245 case BOND_XMIT_POLICY_LAYER2: 1246 return NETDEV_LAG_HASH_L2; 1247 case BOND_XMIT_POLICY_LAYER34: 1248 return NETDEV_LAG_HASH_L34; 1249 case BOND_XMIT_POLICY_LAYER23: 1250 return NETDEV_LAG_HASH_L23; 1251 case BOND_XMIT_POLICY_ENCAP23: 1252 return NETDEV_LAG_HASH_E23; 1253 case BOND_XMIT_POLICY_ENCAP34: 1254 return NETDEV_LAG_HASH_E34; 1255 default: 1256 return NETDEV_LAG_HASH_UNKNOWN; 1257 } 1258 } 1259 1260 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1261 struct netlink_ext_ack *extack) 1262 { 1263 struct netdev_lag_upper_info lag_upper_info; 1264 enum netdev_lag_tx_type type; 1265 1266 type = bond_lag_tx_type(bond); 1267 lag_upper_info.tx_type = type; 1268 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1269 1270 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1271 &lag_upper_info, extack); 1272 } 1273 1274 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1275 { 1276 netdev_upper_dev_unlink(slave->dev, bond->dev); 1277 slave->dev->flags &= ~IFF_SLAVE; 1278 } 1279 1280 static struct slave *bond_alloc_slave(struct bonding *bond) 1281 { 1282 struct slave *slave = NULL; 1283 1284 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1285 if (!slave) 1286 return NULL; 1287 1288 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1289 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1290 GFP_KERNEL); 1291 if (!SLAVE_AD_INFO(slave)) { 1292 kfree(slave); 1293 return NULL; 1294 } 1295 } 1296 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1297 1298 return slave; 1299 } 1300 1301 static void bond_free_slave(struct slave *slave) 1302 { 1303 struct bonding *bond = bond_get_bond_by_slave(slave); 1304 1305 cancel_delayed_work_sync(&slave->notify_work); 1306 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1307 kfree(SLAVE_AD_INFO(slave)); 1308 1309 kfree(slave); 1310 } 1311 1312 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1313 { 1314 info->bond_mode = BOND_MODE(bond); 1315 info->miimon = bond->params.miimon; 1316 info->num_slaves = bond->slave_cnt; 1317 } 1318 1319 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1320 { 1321 strcpy(info->slave_name, slave->dev->name); 1322 info->link = slave->link; 1323 info->state = bond_slave_state(slave); 1324 info->link_failure_count = slave->link_failure_count; 1325 } 1326 1327 static void bond_netdev_notify_work(struct work_struct *_work) 1328 { 1329 struct slave *slave = container_of(_work, struct slave, 1330 notify_work.work); 1331 1332 if (rtnl_trylock()) { 1333 struct netdev_bonding_info binfo; 1334 1335 bond_fill_ifslave(slave, &binfo.slave); 1336 bond_fill_ifbond(slave->bond, &binfo.master); 1337 netdev_bonding_info_change(slave->dev, &binfo); 1338 rtnl_unlock(); 1339 } else { 1340 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1341 } 1342 } 1343 1344 void bond_queue_slave_event(struct slave *slave) 1345 { 1346 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1347 } 1348 1349 void bond_lower_state_changed(struct slave *slave) 1350 { 1351 struct netdev_lag_lower_state_info info; 1352 1353 info.link_up = slave->link == BOND_LINK_UP || 1354 slave->link == BOND_LINK_FAIL; 1355 info.tx_enabled = bond_is_active_slave(slave); 1356 netdev_lower_state_changed(slave->dev, &info); 1357 } 1358 1359 /* enslave device <slave> to bond device <master> */ 1360 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1361 struct netlink_ext_ack *extack) 1362 { 1363 struct bonding *bond = netdev_priv(bond_dev); 1364 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1365 struct slave *new_slave = NULL, *prev_slave; 1366 struct sockaddr_storage ss; 1367 int link_reporting; 1368 int res = 0, i; 1369 1370 if (!bond->params.use_carrier && 1371 slave_dev->ethtool_ops->get_link == NULL && 1372 slave_ops->ndo_do_ioctl == NULL) { 1373 netdev_warn(bond_dev, "no link monitoring support for %s\n", 1374 slave_dev->name); 1375 } 1376 1377 /* already in-use? */ 1378 if (netdev_is_rx_handler_busy(slave_dev)) { 1379 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1380 netdev_err(bond_dev, 1381 "Error: Device is in use and cannot be enslaved\n"); 1382 return -EBUSY; 1383 } 1384 1385 if (bond_dev == slave_dev) { 1386 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1387 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1388 return -EPERM; 1389 } 1390 1391 /* vlan challenged mutual exclusion */ 1392 /* no need to lock since we're protected by rtnl_lock */ 1393 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1394 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n", 1395 slave_dev->name); 1396 if (vlan_uses_dev(bond_dev)) { 1397 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1398 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n", 1399 slave_dev->name, bond_dev->name); 1400 return -EPERM; 1401 } else { 1402 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n", 1403 slave_dev->name, slave_dev->name, 1404 bond_dev->name); 1405 } 1406 } else { 1407 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n", 1408 slave_dev->name); 1409 } 1410 1411 /* Old ifenslave binaries are no longer supported. These can 1412 * be identified with moderate accuracy by the state of the slave: 1413 * the current ifenslave will set the interface down prior to 1414 * enslaving it; the old ifenslave will not. 1415 */ 1416 if (slave_dev->flags & IFF_UP) { 1417 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1418 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n", 1419 slave_dev->name); 1420 return -EPERM; 1421 } 1422 1423 /* set bonding device ether type by slave - bonding netdevices are 1424 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1425 * there is a need to override some of the type dependent attribs/funcs. 1426 * 1427 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1428 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1429 */ 1430 if (!bond_has_slaves(bond)) { 1431 if (bond_dev->type != slave_dev->type) { 1432 netdev_dbg(bond_dev, "change device type from %d to %d\n", 1433 bond_dev->type, slave_dev->type); 1434 1435 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1436 bond_dev); 1437 res = notifier_to_errno(res); 1438 if (res) { 1439 netdev_err(bond_dev, "refused to change device type\n"); 1440 return -EBUSY; 1441 } 1442 1443 /* Flush unicast and multicast addresses */ 1444 dev_uc_flush(bond_dev); 1445 dev_mc_flush(bond_dev); 1446 1447 if (slave_dev->type != ARPHRD_ETHER) 1448 bond_setup_by_slave(bond_dev, slave_dev); 1449 else { 1450 ether_setup(bond_dev); 1451 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1452 } 1453 1454 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1455 bond_dev); 1456 } 1457 } else if (bond_dev->type != slave_dev->type) { 1458 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1459 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n", 1460 slave_dev->name, slave_dev->type, bond_dev->type); 1461 return -EINVAL; 1462 } 1463 1464 if (slave_dev->type == ARPHRD_INFINIBAND && 1465 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1466 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1467 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n", 1468 slave_dev->type); 1469 res = -EOPNOTSUPP; 1470 goto err_undo_flags; 1471 } 1472 1473 if (!slave_ops->ndo_set_mac_address || 1474 slave_dev->type == ARPHRD_INFINIBAND) { 1475 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n"); 1476 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1477 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1478 if (!bond_has_slaves(bond)) { 1479 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1480 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1481 } else { 1482 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1483 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n"); 1484 res = -EOPNOTSUPP; 1485 goto err_undo_flags; 1486 } 1487 } 1488 } 1489 1490 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1491 1492 /* If this is the first slave, then we need to set the master's hardware 1493 * address to be the same as the slave's. 1494 */ 1495 if (!bond_has_slaves(bond) && 1496 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1497 res = bond_set_dev_addr(bond->dev, slave_dev); 1498 if (res) 1499 goto err_undo_flags; 1500 } 1501 1502 new_slave = bond_alloc_slave(bond); 1503 if (!new_slave) { 1504 res = -ENOMEM; 1505 goto err_undo_flags; 1506 } 1507 1508 new_slave->bond = bond; 1509 new_slave->dev = slave_dev; 1510 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1511 * is set via sysfs or module option if desired. 1512 */ 1513 new_slave->queue_id = 0; 1514 1515 /* Save slave's original mtu and then set it to match the bond */ 1516 new_slave->original_mtu = slave_dev->mtu; 1517 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1518 if (res) { 1519 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res); 1520 goto err_free; 1521 } 1522 1523 /* Save slave's original ("permanent") mac address for modes 1524 * that need it, and for restoring it upon release, and then 1525 * set it to the master's address 1526 */ 1527 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1528 slave_dev->addr_len); 1529 1530 if (!bond->params.fail_over_mac || 1531 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1532 /* Set slave to master's mac address. The application already 1533 * set the master's mac address to that of the first slave 1534 */ 1535 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1536 ss.ss_family = slave_dev->type; 1537 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1538 extack); 1539 if (res) { 1540 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res); 1541 goto err_restore_mtu; 1542 } 1543 } 1544 1545 /* set slave flag before open to prevent IPv6 addrconf */ 1546 slave_dev->flags |= IFF_SLAVE; 1547 1548 /* open the slave since the application closed it */ 1549 res = dev_open(slave_dev, extack); 1550 if (res) { 1551 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name); 1552 goto err_restore_mac; 1553 } 1554 1555 slave_dev->priv_flags |= IFF_BONDING; 1556 /* initialize slave stats */ 1557 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1558 1559 if (bond_is_lb(bond)) { 1560 /* bond_alb_init_slave() must be called before all other stages since 1561 * it might fail and we do not want to have to undo everything 1562 */ 1563 res = bond_alb_init_slave(bond, new_slave); 1564 if (res) 1565 goto err_close; 1566 } 1567 1568 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1569 if (res) { 1570 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n", 1571 slave_dev->name); 1572 goto err_close; 1573 } 1574 1575 prev_slave = bond_last_slave(bond); 1576 1577 new_slave->delay = 0; 1578 new_slave->link_failure_count = 0; 1579 1580 if (bond_update_speed_duplex(new_slave) && 1581 bond_needs_speed_duplex(bond)) 1582 new_slave->link = BOND_LINK_DOWN; 1583 1584 new_slave->last_rx = jiffies - 1585 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1586 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1587 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1588 1589 if (bond->params.miimon && !bond->params.use_carrier) { 1590 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1591 1592 if ((link_reporting == -1) && !bond->params.arp_interval) { 1593 /* miimon is set but a bonded network driver 1594 * does not support ETHTOOL/MII and 1595 * arp_interval is not set. Note: if 1596 * use_carrier is enabled, we will never go 1597 * here (because netif_carrier is always 1598 * supported); thus, we don't need to change 1599 * the messages for netif_carrier. 1600 */ 1601 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n", 1602 slave_dev->name); 1603 } else if (link_reporting == -1) { 1604 /* unable get link status using mii/ethtool */ 1605 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n", 1606 slave_dev->name); 1607 } 1608 } 1609 1610 /* check for initial state */ 1611 new_slave->link = BOND_LINK_NOCHANGE; 1612 if (bond->params.miimon) { 1613 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1614 if (bond->params.updelay) { 1615 bond_set_slave_link_state(new_slave, 1616 BOND_LINK_BACK, 1617 BOND_SLAVE_NOTIFY_NOW); 1618 new_slave->delay = bond->params.updelay; 1619 } else { 1620 bond_set_slave_link_state(new_slave, 1621 BOND_LINK_UP, 1622 BOND_SLAVE_NOTIFY_NOW); 1623 } 1624 } else { 1625 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1626 BOND_SLAVE_NOTIFY_NOW); 1627 } 1628 } else if (bond->params.arp_interval) { 1629 bond_set_slave_link_state(new_slave, 1630 (netif_carrier_ok(slave_dev) ? 1631 BOND_LINK_UP : BOND_LINK_DOWN), 1632 BOND_SLAVE_NOTIFY_NOW); 1633 } else { 1634 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1635 BOND_SLAVE_NOTIFY_NOW); 1636 } 1637 1638 if (new_slave->link != BOND_LINK_DOWN) 1639 new_slave->last_link_up = jiffies; 1640 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n", 1641 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1642 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1643 1644 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1645 /* if there is a primary slave, remember it */ 1646 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1647 rcu_assign_pointer(bond->primary_slave, new_slave); 1648 bond->force_primary = true; 1649 } 1650 } 1651 1652 switch (BOND_MODE(bond)) { 1653 case BOND_MODE_ACTIVEBACKUP: 1654 bond_set_slave_inactive_flags(new_slave, 1655 BOND_SLAVE_NOTIFY_NOW); 1656 break; 1657 case BOND_MODE_8023AD: 1658 /* in 802.3ad mode, the internal mechanism 1659 * will activate the slaves in the selected 1660 * aggregator 1661 */ 1662 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1663 /* if this is the first slave */ 1664 if (!prev_slave) { 1665 SLAVE_AD_INFO(new_slave)->id = 1; 1666 /* Initialize AD with the number of times that the AD timer is called in 1 second 1667 * can be called only after the mac address of the bond is set 1668 */ 1669 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1670 } else { 1671 SLAVE_AD_INFO(new_slave)->id = 1672 SLAVE_AD_INFO(prev_slave)->id + 1; 1673 } 1674 1675 bond_3ad_bind_slave(new_slave); 1676 break; 1677 case BOND_MODE_TLB: 1678 case BOND_MODE_ALB: 1679 bond_set_active_slave(new_slave); 1680 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1681 break; 1682 default: 1683 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n"); 1684 1685 /* always active in trunk mode */ 1686 bond_set_active_slave(new_slave); 1687 1688 /* In trunking mode there is little meaning to curr_active_slave 1689 * anyway (it holds no special properties of the bond device), 1690 * so we can change it without calling change_active_interface() 1691 */ 1692 if (!rcu_access_pointer(bond->curr_active_slave) && 1693 new_slave->link == BOND_LINK_UP) 1694 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1695 1696 break; 1697 } /* switch(bond_mode) */ 1698 1699 #ifdef CONFIG_NET_POLL_CONTROLLER 1700 if (bond->dev->npinfo) { 1701 if (slave_enable_netpoll(new_slave)) { 1702 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1703 res = -EBUSY; 1704 goto err_detach; 1705 } 1706 } 1707 #endif 1708 1709 if (!(bond_dev->features & NETIF_F_LRO)) 1710 dev_disable_lro(slave_dev); 1711 1712 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1713 new_slave); 1714 if (res) { 1715 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res); 1716 goto err_detach; 1717 } 1718 1719 res = bond_master_upper_dev_link(bond, new_slave, extack); 1720 if (res) { 1721 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1722 goto err_unregister; 1723 } 1724 1725 res = bond_sysfs_slave_add(new_slave); 1726 if (res) { 1727 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1728 goto err_upper_unlink; 1729 } 1730 1731 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1732 1733 /* If the mode uses primary, then the following is handled by 1734 * bond_change_active_slave(). 1735 */ 1736 if (!bond_uses_primary(bond)) { 1737 /* set promiscuity level to new slave */ 1738 if (bond_dev->flags & IFF_PROMISC) { 1739 res = dev_set_promiscuity(slave_dev, 1); 1740 if (res) 1741 goto err_sysfs_del; 1742 } 1743 1744 /* set allmulti level to new slave */ 1745 if (bond_dev->flags & IFF_ALLMULTI) { 1746 res = dev_set_allmulti(slave_dev, 1); 1747 if (res) { 1748 if (bond_dev->flags & IFF_PROMISC) 1749 dev_set_promiscuity(slave_dev, -1); 1750 goto err_sysfs_del; 1751 } 1752 } 1753 1754 netif_addr_lock_bh(bond_dev); 1755 dev_mc_sync_multiple(slave_dev, bond_dev); 1756 dev_uc_sync_multiple(slave_dev, bond_dev); 1757 netif_addr_unlock_bh(bond_dev); 1758 1759 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1760 /* add lacpdu mc addr to mc list */ 1761 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1762 1763 dev_mc_add(slave_dev, lacpdu_multicast); 1764 } 1765 } 1766 1767 bond->slave_cnt++; 1768 bond_compute_features(bond); 1769 bond_set_carrier(bond); 1770 1771 if (bond_uses_primary(bond)) { 1772 block_netpoll_tx(); 1773 bond_select_active_slave(bond); 1774 unblock_netpoll_tx(); 1775 } 1776 1777 if (bond_mode_can_use_xmit_hash(bond)) 1778 bond_update_slave_arr(bond, NULL); 1779 1780 1781 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n", 1782 slave_dev->name, 1783 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1784 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1785 1786 /* enslave is successful */ 1787 bond_queue_slave_event(new_slave); 1788 return 0; 1789 1790 /* Undo stages on error */ 1791 err_sysfs_del: 1792 bond_sysfs_slave_del(new_slave); 1793 1794 err_upper_unlink: 1795 bond_upper_dev_unlink(bond, new_slave); 1796 1797 err_unregister: 1798 netdev_rx_handler_unregister(slave_dev); 1799 1800 err_detach: 1801 vlan_vids_del_by_dev(slave_dev, bond_dev); 1802 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1803 RCU_INIT_POINTER(bond->primary_slave, NULL); 1804 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1805 block_netpoll_tx(); 1806 bond_change_active_slave(bond, NULL); 1807 bond_select_active_slave(bond); 1808 unblock_netpoll_tx(); 1809 } 1810 /* either primary_slave or curr_active_slave might've changed */ 1811 synchronize_rcu(); 1812 slave_disable_netpoll(new_slave); 1813 1814 err_close: 1815 slave_dev->priv_flags &= ~IFF_BONDING; 1816 dev_close(slave_dev); 1817 1818 err_restore_mac: 1819 slave_dev->flags &= ~IFF_SLAVE; 1820 if (!bond->params.fail_over_mac || 1821 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1822 /* XXX TODO - fom follow mode needs to change master's 1823 * MAC if this slave's MAC is in use by the bond, or at 1824 * least print a warning. 1825 */ 1826 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 1827 new_slave->dev->addr_len); 1828 ss.ss_family = slave_dev->type; 1829 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 1830 } 1831 1832 err_restore_mtu: 1833 dev_set_mtu(slave_dev, new_slave->original_mtu); 1834 1835 err_free: 1836 bond_free_slave(new_slave); 1837 1838 err_undo_flags: 1839 /* Enslave of first slave has failed and we need to fix master's mac */ 1840 if (!bond_has_slaves(bond)) { 1841 if (ether_addr_equal_64bits(bond_dev->dev_addr, 1842 slave_dev->dev_addr)) 1843 eth_hw_addr_random(bond_dev); 1844 if (bond_dev->type != ARPHRD_ETHER) { 1845 dev_close(bond_dev); 1846 ether_setup(bond_dev); 1847 bond_dev->flags |= IFF_MASTER; 1848 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1849 } 1850 } 1851 1852 return res; 1853 } 1854 1855 /* Try to release the slave device <slave> from the bond device <master> 1856 * It is legal to access curr_active_slave without a lock because all the function 1857 * is RTNL-locked. If "all" is true it means that the function is being called 1858 * while destroying a bond interface and all slaves are being released. 1859 * 1860 * The rules for slave state should be: 1861 * for Active/Backup: 1862 * Active stays on all backups go down 1863 * for Bonded connections: 1864 * The first up interface should be left on and all others downed. 1865 */ 1866 static int __bond_release_one(struct net_device *bond_dev, 1867 struct net_device *slave_dev, 1868 bool all, bool unregister) 1869 { 1870 struct bonding *bond = netdev_priv(bond_dev); 1871 struct slave *slave, *oldcurrent; 1872 struct sockaddr_storage ss; 1873 int old_flags = bond_dev->flags; 1874 netdev_features_t old_features = bond_dev->features; 1875 1876 /* slave is not a slave or master is not master of this slave */ 1877 if (!(slave_dev->flags & IFF_SLAVE) || 1878 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1879 netdev_dbg(bond_dev, "cannot release %s\n", 1880 slave_dev->name); 1881 return -EINVAL; 1882 } 1883 1884 block_netpoll_tx(); 1885 1886 slave = bond_get_slave_by_dev(bond, slave_dev); 1887 if (!slave) { 1888 /* not a slave of this bond */ 1889 netdev_info(bond_dev, "%s not enslaved\n", 1890 slave_dev->name); 1891 unblock_netpoll_tx(); 1892 return -EINVAL; 1893 } 1894 1895 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 1896 1897 bond_sysfs_slave_del(slave); 1898 1899 /* recompute stats just before removing the slave */ 1900 bond_get_stats(bond->dev, &bond->bond_stats); 1901 1902 bond_upper_dev_unlink(bond, slave); 1903 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1904 * for this slave anymore. 1905 */ 1906 netdev_rx_handler_unregister(slave_dev); 1907 1908 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1909 bond_3ad_unbind_slave(slave); 1910 1911 if (bond_mode_can_use_xmit_hash(bond)) 1912 bond_update_slave_arr(bond, slave); 1913 1914 netdev_info(bond_dev, "Releasing %s interface %s\n", 1915 bond_is_active_slave(slave) ? "active" : "backup", 1916 slave_dev->name); 1917 1918 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1919 1920 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1921 1922 if (!all && (!bond->params.fail_over_mac || 1923 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1924 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1925 bond_has_slaves(bond)) 1926 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n", 1927 slave_dev->name, slave->perm_hwaddr, 1928 bond_dev->name, slave_dev->name); 1929 } 1930 1931 if (rtnl_dereference(bond->primary_slave) == slave) 1932 RCU_INIT_POINTER(bond->primary_slave, NULL); 1933 1934 if (oldcurrent == slave) 1935 bond_change_active_slave(bond, NULL); 1936 1937 if (bond_is_lb(bond)) { 1938 /* Must be called only after the slave has been 1939 * detached from the list and the curr_active_slave 1940 * has been cleared (if our_slave == old_current), 1941 * but before a new active slave is selected. 1942 */ 1943 bond_alb_deinit_slave(bond, slave); 1944 } 1945 1946 if (all) { 1947 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1948 } else if (oldcurrent == slave) { 1949 /* Note that we hold RTNL over this sequence, so there 1950 * is no concern that another slave add/remove event 1951 * will interfere. 1952 */ 1953 bond_select_active_slave(bond); 1954 } 1955 1956 if (!bond_has_slaves(bond)) { 1957 bond_set_carrier(bond); 1958 eth_hw_addr_random(bond_dev); 1959 bond->nest_level = SINGLE_DEPTH_NESTING; 1960 } else { 1961 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1962 } 1963 1964 unblock_netpoll_tx(); 1965 synchronize_rcu(); 1966 bond->slave_cnt--; 1967 1968 if (!bond_has_slaves(bond)) { 1969 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 1970 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 1971 } 1972 1973 bond_compute_features(bond); 1974 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 1975 (old_features & NETIF_F_VLAN_CHALLENGED)) 1976 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n", 1977 slave_dev->name, bond_dev->name); 1978 1979 vlan_vids_del_by_dev(slave_dev, bond_dev); 1980 1981 /* If the mode uses primary, then this case was handled above by 1982 * bond_change_active_slave(..., NULL) 1983 */ 1984 if (!bond_uses_primary(bond)) { 1985 /* unset promiscuity level from slave 1986 * NOTE: The NETDEV_CHANGEADDR call above may change the value 1987 * of the IFF_PROMISC flag in the bond_dev, but we need the 1988 * value of that flag before that change, as that was the value 1989 * when this slave was attached, so we cache at the start of the 1990 * function and use it here. Same goes for ALLMULTI below 1991 */ 1992 if (old_flags & IFF_PROMISC) 1993 dev_set_promiscuity(slave_dev, -1); 1994 1995 /* unset allmulti level from slave */ 1996 if (old_flags & IFF_ALLMULTI) 1997 dev_set_allmulti(slave_dev, -1); 1998 1999 bond_hw_addr_flush(bond_dev, slave_dev); 2000 } 2001 2002 slave_disable_netpoll(slave); 2003 2004 /* close slave before restoring its mac address */ 2005 dev_close(slave_dev); 2006 2007 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2008 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2009 /* restore original ("permanent") mac address */ 2010 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2011 slave->dev->addr_len); 2012 ss.ss_family = slave_dev->type; 2013 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2014 } 2015 2016 if (unregister) 2017 __dev_set_mtu(slave_dev, slave->original_mtu); 2018 else 2019 dev_set_mtu(slave_dev, slave->original_mtu); 2020 2021 slave_dev->priv_flags &= ~IFF_BONDING; 2022 2023 bond_free_slave(slave); 2024 2025 return 0; 2026 } 2027 2028 /* A wrapper used because of ndo_del_link */ 2029 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2030 { 2031 return __bond_release_one(bond_dev, slave_dev, false, false); 2032 } 2033 2034 /* First release a slave and then destroy the bond if no more slaves are left. 2035 * Must be under rtnl_lock when this function is called. 2036 */ 2037 static int bond_release_and_destroy(struct net_device *bond_dev, 2038 struct net_device *slave_dev) 2039 { 2040 struct bonding *bond = netdev_priv(bond_dev); 2041 int ret; 2042 2043 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2044 if (ret == 0 && !bond_has_slaves(bond)) { 2045 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2046 netdev_info(bond_dev, "Destroying bond %s\n", 2047 bond_dev->name); 2048 bond_remove_proc_entry(bond); 2049 unregister_netdevice(bond_dev); 2050 } 2051 return ret; 2052 } 2053 2054 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2055 { 2056 struct bonding *bond = netdev_priv(bond_dev); 2057 bond_fill_ifbond(bond, info); 2058 } 2059 2060 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2061 { 2062 struct bonding *bond = netdev_priv(bond_dev); 2063 struct list_head *iter; 2064 int i = 0, res = -ENODEV; 2065 struct slave *slave; 2066 2067 bond_for_each_slave(bond, slave, iter) { 2068 if (i++ == (int)info->slave_id) { 2069 res = 0; 2070 bond_fill_ifslave(slave, info); 2071 break; 2072 } 2073 } 2074 2075 return res; 2076 } 2077 2078 /*-------------------------------- Monitoring -------------------------------*/ 2079 2080 /* called with rcu_read_lock() */ 2081 static int bond_miimon_inspect(struct bonding *bond) 2082 { 2083 int link_state, commit = 0; 2084 struct list_head *iter; 2085 struct slave *slave; 2086 bool ignore_updelay; 2087 2088 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2089 2090 bond_for_each_slave_rcu(bond, slave, iter) { 2091 slave->new_link = BOND_LINK_NOCHANGE; 2092 slave->link_new_state = slave->link; 2093 2094 link_state = bond_check_dev_link(bond, slave->dev, 0); 2095 2096 switch (slave->link) { 2097 case BOND_LINK_UP: 2098 if (link_state) 2099 continue; 2100 2101 bond_propose_link_state(slave, BOND_LINK_FAIL); 2102 commit++; 2103 slave->delay = bond->params.downdelay; 2104 if (slave->delay) { 2105 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n", 2106 (BOND_MODE(bond) == 2107 BOND_MODE_ACTIVEBACKUP) ? 2108 (bond_is_active_slave(slave) ? 2109 "active " : "backup ") : "", 2110 slave->dev->name, 2111 bond->params.downdelay * bond->params.miimon); 2112 } 2113 /*FALLTHRU*/ 2114 case BOND_LINK_FAIL: 2115 if (link_state) { 2116 /* recovered before downdelay expired */ 2117 bond_propose_link_state(slave, BOND_LINK_UP); 2118 slave->last_link_up = jiffies; 2119 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n", 2120 (bond->params.downdelay - slave->delay) * 2121 bond->params.miimon, 2122 slave->dev->name); 2123 commit++; 2124 continue; 2125 } 2126 2127 if (slave->delay <= 0) { 2128 slave->new_link = BOND_LINK_DOWN; 2129 commit++; 2130 continue; 2131 } 2132 2133 slave->delay--; 2134 break; 2135 2136 case BOND_LINK_DOWN: 2137 if (!link_state) 2138 continue; 2139 2140 bond_propose_link_state(slave, BOND_LINK_BACK); 2141 commit++; 2142 slave->delay = bond->params.updelay; 2143 2144 if (slave->delay) { 2145 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n", 2146 slave->dev->name, 2147 ignore_updelay ? 0 : 2148 bond->params.updelay * 2149 bond->params.miimon); 2150 } 2151 /*FALLTHRU*/ 2152 case BOND_LINK_BACK: 2153 if (!link_state) { 2154 bond_propose_link_state(slave, BOND_LINK_DOWN); 2155 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n", 2156 (bond->params.updelay - slave->delay) * 2157 bond->params.miimon, 2158 slave->dev->name); 2159 commit++; 2160 continue; 2161 } 2162 2163 if (ignore_updelay) 2164 slave->delay = 0; 2165 2166 if (slave->delay <= 0) { 2167 slave->new_link = BOND_LINK_UP; 2168 commit++; 2169 ignore_updelay = false; 2170 continue; 2171 } 2172 2173 slave->delay--; 2174 break; 2175 } 2176 } 2177 2178 return commit; 2179 } 2180 2181 static void bond_miimon_link_change(struct bonding *bond, 2182 struct slave *slave, 2183 char link) 2184 { 2185 switch (BOND_MODE(bond)) { 2186 case BOND_MODE_8023AD: 2187 bond_3ad_handle_link_change(slave, link); 2188 break; 2189 case BOND_MODE_TLB: 2190 case BOND_MODE_ALB: 2191 bond_alb_handle_link_change(bond, slave, link); 2192 break; 2193 case BOND_MODE_XOR: 2194 bond_update_slave_arr(bond, NULL); 2195 break; 2196 } 2197 } 2198 2199 static void bond_miimon_commit(struct bonding *bond) 2200 { 2201 struct list_head *iter; 2202 struct slave *slave, *primary; 2203 2204 bond_for_each_slave(bond, slave, iter) { 2205 switch (slave->new_link) { 2206 case BOND_LINK_NOCHANGE: 2207 continue; 2208 2209 case BOND_LINK_UP: 2210 if (bond_update_speed_duplex(slave) && 2211 bond_needs_speed_duplex(bond)) { 2212 slave->link = BOND_LINK_DOWN; 2213 if (net_ratelimit()) 2214 netdev_warn(bond->dev, 2215 "failed to get link speed/duplex for %s\n", 2216 slave->dev->name); 2217 continue; 2218 } 2219 bond_set_slave_link_state(slave, BOND_LINK_UP, 2220 BOND_SLAVE_NOTIFY_NOW); 2221 slave->last_link_up = jiffies; 2222 2223 primary = rtnl_dereference(bond->primary_slave); 2224 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2225 /* prevent it from being the active one */ 2226 bond_set_backup_slave(slave); 2227 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2228 /* make it immediately active */ 2229 bond_set_active_slave(slave); 2230 } else if (slave != primary) { 2231 /* prevent it from being the active one */ 2232 bond_set_backup_slave(slave); 2233 } 2234 2235 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n", 2236 slave->dev->name, 2237 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2238 slave->duplex ? "full" : "half"); 2239 2240 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2241 2242 if (!bond->curr_active_slave || slave == primary) 2243 goto do_failover; 2244 2245 continue; 2246 2247 case BOND_LINK_DOWN: 2248 if (slave->link_failure_count < UINT_MAX) 2249 slave->link_failure_count++; 2250 2251 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2252 BOND_SLAVE_NOTIFY_NOW); 2253 2254 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2255 BOND_MODE(bond) == BOND_MODE_8023AD) 2256 bond_set_slave_inactive_flags(slave, 2257 BOND_SLAVE_NOTIFY_NOW); 2258 2259 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2260 slave->dev->name); 2261 2262 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2263 2264 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2265 goto do_failover; 2266 2267 continue; 2268 2269 default: 2270 netdev_err(bond->dev, "invalid new link %d on slave %s\n", 2271 slave->new_link, slave->dev->name); 2272 slave->new_link = BOND_LINK_NOCHANGE; 2273 2274 continue; 2275 } 2276 2277 do_failover: 2278 block_netpoll_tx(); 2279 bond_select_active_slave(bond); 2280 unblock_netpoll_tx(); 2281 } 2282 2283 bond_set_carrier(bond); 2284 } 2285 2286 /* bond_mii_monitor 2287 * 2288 * Really a wrapper that splits the mii monitor into two phases: an 2289 * inspection, then (if inspection indicates something needs to be done) 2290 * an acquisition of appropriate locks followed by a commit phase to 2291 * implement whatever link state changes are indicated. 2292 */ 2293 static void bond_mii_monitor(struct work_struct *work) 2294 { 2295 struct bonding *bond = container_of(work, struct bonding, 2296 mii_work.work); 2297 bool should_notify_peers = false; 2298 unsigned long delay; 2299 struct slave *slave; 2300 struct list_head *iter; 2301 2302 delay = msecs_to_jiffies(bond->params.miimon); 2303 2304 if (!bond_has_slaves(bond)) 2305 goto re_arm; 2306 2307 rcu_read_lock(); 2308 2309 should_notify_peers = bond_should_notify_peers(bond); 2310 2311 if (bond_miimon_inspect(bond)) { 2312 rcu_read_unlock(); 2313 2314 /* Race avoidance with bond_close cancel of workqueue */ 2315 if (!rtnl_trylock()) { 2316 delay = 1; 2317 should_notify_peers = false; 2318 goto re_arm; 2319 } 2320 2321 bond_for_each_slave(bond, slave, iter) { 2322 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2323 } 2324 bond_miimon_commit(bond); 2325 2326 rtnl_unlock(); /* might sleep, hold no other locks */ 2327 } else 2328 rcu_read_unlock(); 2329 2330 re_arm: 2331 if (bond->params.miimon) 2332 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2333 2334 if (should_notify_peers) { 2335 if (!rtnl_trylock()) 2336 return; 2337 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2338 rtnl_unlock(); 2339 } 2340 } 2341 2342 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2343 { 2344 __be32 ip = *((__be32 *)data); 2345 2346 return ip == bond_confirm_addr(upper, 0, ip); 2347 } 2348 2349 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2350 { 2351 bool ret = false; 2352 2353 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2354 return true; 2355 2356 rcu_read_lock(); 2357 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2358 ret = true; 2359 rcu_read_unlock(); 2360 2361 return ret; 2362 } 2363 2364 /* We go to the (large) trouble of VLAN tagging ARP frames because 2365 * switches in VLAN mode (especially if ports are configured as 2366 * "native" to a VLAN) might not pass non-tagged frames. 2367 */ 2368 static void bond_arp_send(struct net_device *slave_dev, int arp_op, 2369 __be32 dest_ip, __be32 src_ip, 2370 struct bond_vlan_tag *tags) 2371 { 2372 struct sk_buff *skb; 2373 struct bond_vlan_tag *outer_tag = tags; 2374 2375 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n", 2376 arp_op, slave_dev->name, &dest_ip, &src_ip); 2377 2378 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2379 NULL, slave_dev->dev_addr, NULL); 2380 2381 if (!skb) { 2382 net_err_ratelimited("ARP packet allocation failed\n"); 2383 return; 2384 } 2385 2386 if (!tags || tags->vlan_proto == VLAN_N_VID) 2387 goto xmit; 2388 2389 tags++; 2390 2391 /* Go through all the tags backwards and add them to the packet */ 2392 while (tags->vlan_proto != VLAN_N_VID) { 2393 if (!tags->vlan_id) { 2394 tags++; 2395 continue; 2396 } 2397 2398 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n", 2399 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2400 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2401 tags->vlan_id); 2402 if (!skb) { 2403 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2404 return; 2405 } 2406 2407 tags++; 2408 } 2409 /* Set the outer tag */ 2410 if (outer_tag->vlan_id) { 2411 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n", 2412 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2413 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2414 outer_tag->vlan_id); 2415 } 2416 2417 xmit: 2418 arp_xmit(skb); 2419 } 2420 2421 /* Validate the device path between the @start_dev and the @end_dev. 2422 * The path is valid if the @end_dev is reachable through device 2423 * stacking. 2424 * When the path is validated, collect any vlan information in the 2425 * path. 2426 */ 2427 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2428 struct net_device *end_dev, 2429 int level) 2430 { 2431 struct bond_vlan_tag *tags; 2432 struct net_device *upper; 2433 struct list_head *iter; 2434 2435 if (start_dev == end_dev) { 2436 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2437 if (!tags) 2438 return ERR_PTR(-ENOMEM); 2439 tags[level].vlan_proto = VLAN_N_VID; 2440 return tags; 2441 } 2442 2443 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2444 tags = bond_verify_device_path(upper, end_dev, level + 1); 2445 if (IS_ERR_OR_NULL(tags)) { 2446 if (IS_ERR(tags)) 2447 return tags; 2448 continue; 2449 } 2450 if (is_vlan_dev(upper)) { 2451 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2452 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2453 } 2454 2455 return tags; 2456 } 2457 2458 return NULL; 2459 } 2460 2461 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2462 { 2463 struct rtable *rt; 2464 struct bond_vlan_tag *tags; 2465 __be32 *targets = bond->params.arp_targets, addr; 2466 int i; 2467 2468 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2469 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]); 2470 tags = NULL; 2471 2472 /* Find out through which dev should the packet go */ 2473 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2474 RTO_ONLINK, 0); 2475 if (IS_ERR(rt)) { 2476 /* there's no route to target - try to send arp 2477 * probe to generate any traffic (arp_validate=0) 2478 */ 2479 if (bond->params.arp_validate) 2480 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2481 bond->dev->name, 2482 &targets[i]); 2483 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2484 0, tags); 2485 continue; 2486 } 2487 2488 /* bond device itself */ 2489 if (rt->dst.dev == bond->dev) 2490 goto found; 2491 2492 rcu_read_lock(); 2493 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2494 rcu_read_unlock(); 2495 2496 if (!IS_ERR_OR_NULL(tags)) 2497 goto found; 2498 2499 /* Not our device - skip */ 2500 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2501 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2502 2503 ip_rt_put(rt); 2504 continue; 2505 2506 found: 2507 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2508 ip_rt_put(rt); 2509 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2510 addr, tags); 2511 kfree(tags); 2512 } 2513 } 2514 2515 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2516 { 2517 int i; 2518 2519 if (!sip || !bond_has_this_ip(bond, tip)) { 2520 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n", 2521 &sip, &tip); 2522 return; 2523 } 2524 2525 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2526 if (i == -1) { 2527 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n", 2528 &sip); 2529 return; 2530 } 2531 slave->last_rx = jiffies; 2532 slave->target_last_arp_rx[i] = jiffies; 2533 } 2534 2535 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2536 struct slave *slave) 2537 { 2538 struct arphdr *arp = (struct arphdr *)skb->data; 2539 struct slave *curr_active_slave, *curr_arp_slave; 2540 unsigned char *arp_ptr; 2541 __be32 sip, tip; 2542 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2543 unsigned int alen; 2544 2545 if (!slave_do_arp_validate(bond, slave)) { 2546 if ((slave_do_arp_validate_only(bond) && is_arp) || 2547 !slave_do_arp_validate_only(bond)) 2548 slave->last_rx = jiffies; 2549 return RX_HANDLER_ANOTHER; 2550 } else if (!is_arp) { 2551 return RX_HANDLER_ANOTHER; 2552 } 2553 2554 alen = arp_hdr_len(bond->dev); 2555 2556 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n", 2557 skb->dev->name); 2558 2559 if (alen > skb_headlen(skb)) { 2560 arp = kmalloc(alen, GFP_ATOMIC); 2561 if (!arp) 2562 goto out_unlock; 2563 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2564 goto out_unlock; 2565 } 2566 2567 if (arp->ar_hln != bond->dev->addr_len || 2568 skb->pkt_type == PACKET_OTHERHOST || 2569 skb->pkt_type == PACKET_LOOPBACK || 2570 arp->ar_hrd != htons(ARPHRD_ETHER) || 2571 arp->ar_pro != htons(ETH_P_IP) || 2572 arp->ar_pln != 4) 2573 goto out_unlock; 2574 2575 arp_ptr = (unsigned char *)(arp + 1); 2576 arp_ptr += bond->dev->addr_len; 2577 memcpy(&sip, arp_ptr, 4); 2578 arp_ptr += 4 + bond->dev->addr_len; 2579 memcpy(&tip, arp_ptr, 4); 2580 2581 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2582 slave->dev->name, bond_slave_state(slave), 2583 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2584 &sip, &tip); 2585 2586 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2587 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2588 2589 /* We 'trust' the received ARP enough to validate it if: 2590 * 2591 * (a) the slave receiving the ARP is active (which includes the 2592 * current ARP slave, if any), or 2593 * 2594 * (b) the receiving slave isn't active, but there is a currently 2595 * active slave and it received valid arp reply(s) after it became 2596 * the currently active slave, or 2597 * 2598 * (c) there is an ARP slave that sent an ARP during the prior ARP 2599 * interval, and we receive an ARP reply on any slave. We accept 2600 * these because switch FDB update delays may deliver the ARP 2601 * reply to a slave other than the sender of the ARP request. 2602 * 2603 * Note: for (b), backup slaves are receiving the broadcast ARP 2604 * request, not a reply. This request passes from the sending 2605 * slave through the L2 switch(es) to the receiving slave. Since 2606 * this is checking the request, sip/tip are swapped for 2607 * validation. 2608 * 2609 * This is done to avoid endless looping when we can't reach the 2610 * arp_ip_target and fool ourselves with our own arp requests. 2611 */ 2612 if (bond_is_active_slave(slave)) 2613 bond_validate_arp(bond, slave, sip, tip); 2614 else if (curr_active_slave && 2615 time_after(slave_last_rx(bond, curr_active_slave), 2616 curr_active_slave->last_link_up)) 2617 bond_validate_arp(bond, slave, tip, sip); 2618 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2619 bond_time_in_interval(bond, 2620 dev_trans_start(curr_arp_slave->dev), 1)) 2621 bond_validate_arp(bond, slave, sip, tip); 2622 2623 out_unlock: 2624 if (arp != (struct arphdr *)skb->data) 2625 kfree(arp); 2626 return RX_HANDLER_ANOTHER; 2627 } 2628 2629 /* function to verify if we're in the arp_interval timeslice, returns true if 2630 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2631 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2632 */ 2633 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2634 int mod) 2635 { 2636 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2637 2638 return time_in_range(jiffies, 2639 last_act - delta_in_ticks, 2640 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2641 } 2642 2643 /* This function is called regularly to monitor each slave's link 2644 * ensuring that traffic is being sent and received when arp monitoring 2645 * is used in load-balancing mode. if the adapter has been dormant, then an 2646 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2647 * arp monitoring in active backup mode. 2648 */ 2649 static void bond_loadbalance_arp_mon(struct bonding *bond) 2650 { 2651 struct slave *slave, *oldcurrent; 2652 struct list_head *iter; 2653 int do_failover = 0, slave_state_changed = 0; 2654 2655 if (!bond_has_slaves(bond)) 2656 goto re_arm; 2657 2658 rcu_read_lock(); 2659 2660 oldcurrent = rcu_dereference(bond->curr_active_slave); 2661 /* see if any of the previous devices are up now (i.e. they have 2662 * xmt and rcv traffic). the curr_active_slave does not come into 2663 * the picture unless it is null. also, slave->last_link_up is not 2664 * needed here because we send an arp on each slave and give a slave 2665 * as long as it needs to get the tx/rx within the delta. 2666 * TODO: what about up/down delay in arp mode? it wasn't here before 2667 * so it can wait 2668 */ 2669 bond_for_each_slave_rcu(bond, slave, iter) { 2670 unsigned long trans_start = dev_trans_start(slave->dev); 2671 2672 slave->new_link = BOND_LINK_NOCHANGE; 2673 2674 if (slave->link != BOND_LINK_UP) { 2675 if (bond_time_in_interval(bond, trans_start, 1) && 2676 bond_time_in_interval(bond, slave->last_rx, 1)) { 2677 2678 slave->new_link = BOND_LINK_UP; 2679 slave_state_changed = 1; 2680 2681 /* primary_slave has no meaning in round-robin 2682 * mode. the window of a slave being up and 2683 * curr_active_slave being null after enslaving 2684 * is closed. 2685 */ 2686 if (!oldcurrent) { 2687 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2688 slave->dev->name); 2689 do_failover = 1; 2690 } else { 2691 netdev_info(bond->dev, "interface %s is now up\n", 2692 slave->dev->name); 2693 } 2694 } 2695 } else { 2696 /* slave->link == BOND_LINK_UP */ 2697 2698 /* not all switches will respond to an arp request 2699 * when the source ip is 0, so don't take the link down 2700 * if we don't know our ip yet 2701 */ 2702 if (!bond_time_in_interval(bond, trans_start, 2) || 2703 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2704 2705 slave->new_link = BOND_LINK_DOWN; 2706 slave_state_changed = 1; 2707 2708 if (slave->link_failure_count < UINT_MAX) 2709 slave->link_failure_count++; 2710 2711 netdev_info(bond->dev, "interface %s is now down\n", 2712 slave->dev->name); 2713 2714 if (slave == oldcurrent) 2715 do_failover = 1; 2716 } 2717 } 2718 2719 /* note: if switch is in round-robin mode, all links 2720 * must tx arp to ensure all links rx an arp - otherwise 2721 * links may oscillate or not come up at all; if switch is 2722 * in something like xor mode, there is nothing we can 2723 * do - all replies will be rx'ed on same link causing slaves 2724 * to be unstable during low/no traffic periods 2725 */ 2726 if (bond_slave_is_up(slave)) 2727 bond_arp_send_all(bond, slave); 2728 } 2729 2730 rcu_read_unlock(); 2731 2732 if (do_failover || slave_state_changed) { 2733 if (!rtnl_trylock()) 2734 goto re_arm; 2735 2736 bond_for_each_slave(bond, slave, iter) { 2737 if (slave->new_link != BOND_LINK_NOCHANGE) 2738 slave->link = slave->new_link; 2739 } 2740 2741 if (slave_state_changed) { 2742 bond_slave_state_change(bond); 2743 if (BOND_MODE(bond) == BOND_MODE_XOR) 2744 bond_update_slave_arr(bond, NULL); 2745 } 2746 if (do_failover) { 2747 block_netpoll_tx(); 2748 bond_select_active_slave(bond); 2749 unblock_netpoll_tx(); 2750 } 2751 rtnl_unlock(); 2752 } 2753 2754 re_arm: 2755 if (bond->params.arp_interval) 2756 queue_delayed_work(bond->wq, &bond->arp_work, 2757 msecs_to_jiffies(bond->params.arp_interval)); 2758 } 2759 2760 /* Called to inspect slaves for active-backup mode ARP monitor link state 2761 * changes. Sets new_link in slaves to specify what action should take 2762 * place for the slave. Returns 0 if no changes are found, >0 if changes 2763 * to link states must be committed. 2764 * 2765 * Called with rcu_read_lock held. 2766 */ 2767 static int bond_ab_arp_inspect(struct bonding *bond) 2768 { 2769 unsigned long trans_start, last_rx; 2770 struct list_head *iter; 2771 struct slave *slave; 2772 int commit = 0; 2773 2774 bond_for_each_slave_rcu(bond, slave, iter) { 2775 slave->new_link = BOND_LINK_NOCHANGE; 2776 last_rx = slave_last_rx(bond, slave); 2777 2778 if (slave->link != BOND_LINK_UP) { 2779 if (bond_time_in_interval(bond, last_rx, 1)) { 2780 slave->new_link = BOND_LINK_UP; 2781 commit++; 2782 } 2783 continue; 2784 } 2785 2786 /* Give slaves 2*delta after being enslaved or made 2787 * active. This avoids bouncing, as the last receive 2788 * times need a full ARP monitor cycle to be updated. 2789 */ 2790 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2791 continue; 2792 2793 /* Backup slave is down if: 2794 * - No current_arp_slave AND 2795 * - more than 3*delta since last receive AND 2796 * - the bond has an IP address 2797 * 2798 * Note: a non-null current_arp_slave indicates 2799 * the curr_active_slave went down and we are 2800 * searching for a new one; under this condition 2801 * we only take the curr_active_slave down - this 2802 * gives each slave a chance to tx/rx traffic 2803 * before being taken out 2804 */ 2805 if (!bond_is_active_slave(slave) && 2806 !rcu_access_pointer(bond->current_arp_slave) && 2807 !bond_time_in_interval(bond, last_rx, 3)) { 2808 slave->new_link = BOND_LINK_DOWN; 2809 commit++; 2810 } 2811 2812 /* Active slave is down if: 2813 * - more than 2*delta since transmitting OR 2814 * - (more than 2*delta since receive AND 2815 * the bond has an IP address) 2816 */ 2817 trans_start = dev_trans_start(slave->dev); 2818 if (bond_is_active_slave(slave) && 2819 (!bond_time_in_interval(bond, trans_start, 2) || 2820 !bond_time_in_interval(bond, last_rx, 2))) { 2821 slave->new_link = BOND_LINK_DOWN; 2822 commit++; 2823 } 2824 } 2825 2826 return commit; 2827 } 2828 2829 /* Called to commit link state changes noted by inspection step of 2830 * active-backup mode ARP monitor. 2831 * 2832 * Called with RTNL hold. 2833 */ 2834 static void bond_ab_arp_commit(struct bonding *bond) 2835 { 2836 unsigned long trans_start; 2837 struct list_head *iter; 2838 struct slave *slave; 2839 2840 bond_for_each_slave(bond, slave, iter) { 2841 switch (slave->new_link) { 2842 case BOND_LINK_NOCHANGE: 2843 continue; 2844 2845 case BOND_LINK_UP: 2846 trans_start = dev_trans_start(slave->dev); 2847 if (rtnl_dereference(bond->curr_active_slave) != slave || 2848 (!rtnl_dereference(bond->curr_active_slave) && 2849 bond_time_in_interval(bond, trans_start, 1))) { 2850 struct slave *current_arp_slave; 2851 2852 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2853 bond_set_slave_link_state(slave, BOND_LINK_UP, 2854 BOND_SLAVE_NOTIFY_NOW); 2855 if (current_arp_slave) { 2856 bond_set_slave_inactive_flags( 2857 current_arp_slave, 2858 BOND_SLAVE_NOTIFY_NOW); 2859 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2860 } 2861 2862 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2863 slave->dev->name); 2864 2865 if (!rtnl_dereference(bond->curr_active_slave) || 2866 slave == rtnl_dereference(bond->primary_slave)) 2867 goto do_failover; 2868 2869 } 2870 2871 continue; 2872 2873 case BOND_LINK_DOWN: 2874 if (slave->link_failure_count < UINT_MAX) 2875 slave->link_failure_count++; 2876 2877 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2878 BOND_SLAVE_NOTIFY_NOW); 2879 bond_set_slave_inactive_flags(slave, 2880 BOND_SLAVE_NOTIFY_NOW); 2881 2882 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2883 slave->dev->name); 2884 2885 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2886 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2887 goto do_failover; 2888 } 2889 2890 continue; 2891 2892 default: 2893 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n", 2894 slave->new_link, slave->dev->name); 2895 continue; 2896 } 2897 2898 do_failover: 2899 block_netpoll_tx(); 2900 bond_select_active_slave(bond); 2901 unblock_netpoll_tx(); 2902 } 2903 2904 bond_set_carrier(bond); 2905 } 2906 2907 /* Send ARP probes for active-backup mode ARP monitor. 2908 * 2909 * Called with rcu_read_lock held. 2910 */ 2911 static bool bond_ab_arp_probe(struct bonding *bond) 2912 { 2913 struct slave *slave, *before = NULL, *new_slave = NULL, 2914 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2915 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2916 struct list_head *iter; 2917 bool found = false; 2918 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2919 2920 if (curr_arp_slave && curr_active_slave) 2921 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2922 curr_arp_slave->dev->name, 2923 curr_active_slave->dev->name); 2924 2925 if (curr_active_slave) { 2926 bond_arp_send_all(bond, curr_active_slave); 2927 return should_notify_rtnl; 2928 } 2929 2930 /* if we don't have a curr_active_slave, search for the next available 2931 * backup slave from the current_arp_slave and make it the candidate 2932 * for becoming the curr_active_slave 2933 */ 2934 2935 if (!curr_arp_slave) { 2936 curr_arp_slave = bond_first_slave_rcu(bond); 2937 if (!curr_arp_slave) 2938 return should_notify_rtnl; 2939 } 2940 2941 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2942 2943 bond_for_each_slave_rcu(bond, slave, iter) { 2944 if (!found && !before && bond_slave_is_up(slave)) 2945 before = slave; 2946 2947 if (found && !new_slave && bond_slave_is_up(slave)) 2948 new_slave = slave; 2949 /* if the link state is up at this point, we 2950 * mark it down - this can happen if we have 2951 * simultaneous link failures and 2952 * reselect_active_interface doesn't make this 2953 * one the current slave so it is still marked 2954 * up when it is actually down 2955 */ 2956 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 2957 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2958 BOND_SLAVE_NOTIFY_LATER); 2959 if (slave->link_failure_count < UINT_MAX) 2960 slave->link_failure_count++; 2961 2962 bond_set_slave_inactive_flags(slave, 2963 BOND_SLAVE_NOTIFY_LATER); 2964 2965 netdev_info(bond->dev, "backup interface %s is now down\n", 2966 slave->dev->name); 2967 } 2968 if (slave == curr_arp_slave) 2969 found = true; 2970 } 2971 2972 if (!new_slave && before) 2973 new_slave = before; 2974 2975 if (!new_slave) 2976 goto check_state; 2977 2978 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 2979 BOND_SLAVE_NOTIFY_LATER); 2980 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 2981 bond_arp_send_all(bond, new_slave); 2982 new_slave->last_link_up = jiffies; 2983 rcu_assign_pointer(bond->current_arp_slave, new_slave); 2984 2985 check_state: 2986 bond_for_each_slave_rcu(bond, slave, iter) { 2987 if (slave->should_notify || slave->should_notify_link) { 2988 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 2989 break; 2990 } 2991 } 2992 return should_notify_rtnl; 2993 } 2994 2995 static void bond_activebackup_arp_mon(struct bonding *bond) 2996 { 2997 bool should_notify_peers = false; 2998 bool should_notify_rtnl = false; 2999 int delta_in_ticks; 3000 3001 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3002 3003 if (!bond_has_slaves(bond)) 3004 goto re_arm; 3005 3006 rcu_read_lock(); 3007 3008 should_notify_peers = bond_should_notify_peers(bond); 3009 3010 if (bond_ab_arp_inspect(bond)) { 3011 rcu_read_unlock(); 3012 3013 /* Race avoidance with bond_close flush of workqueue */ 3014 if (!rtnl_trylock()) { 3015 delta_in_ticks = 1; 3016 should_notify_peers = false; 3017 goto re_arm; 3018 } 3019 3020 bond_ab_arp_commit(bond); 3021 3022 rtnl_unlock(); 3023 rcu_read_lock(); 3024 } 3025 3026 should_notify_rtnl = bond_ab_arp_probe(bond); 3027 rcu_read_unlock(); 3028 3029 re_arm: 3030 if (bond->params.arp_interval) 3031 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3032 3033 if (should_notify_peers || should_notify_rtnl) { 3034 if (!rtnl_trylock()) 3035 return; 3036 3037 if (should_notify_peers) 3038 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3039 bond->dev); 3040 if (should_notify_rtnl) { 3041 bond_slave_state_notify(bond); 3042 bond_slave_link_notify(bond); 3043 } 3044 3045 rtnl_unlock(); 3046 } 3047 } 3048 3049 static void bond_arp_monitor(struct work_struct *work) 3050 { 3051 struct bonding *bond = container_of(work, struct bonding, 3052 arp_work.work); 3053 3054 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3055 bond_activebackup_arp_mon(bond); 3056 else 3057 bond_loadbalance_arp_mon(bond); 3058 } 3059 3060 /*-------------------------- netdev event handling --------------------------*/ 3061 3062 /* Change device name */ 3063 static int bond_event_changename(struct bonding *bond) 3064 { 3065 bond_remove_proc_entry(bond); 3066 bond_create_proc_entry(bond); 3067 3068 bond_debug_reregister(bond); 3069 3070 return NOTIFY_DONE; 3071 } 3072 3073 static int bond_master_netdev_event(unsigned long event, 3074 struct net_device *bond_dev) 3075 { 3076 struct bonding *event_bond = netdev_priv(bond_dev); 3077 3078 switch (event) { 3079 case NETDEV_CHANGENAME: 3080 return bond_event_changename(event_bond); 3081 case NETDEV_UNREGISTER: 3082 bond_remove_proc_entry(event_bond); 3083 break; 3084 case NETDEV_REGISTER: 3085 bond_create_proc_entry(event_bond); 3086 break; 3087 case NETDEV_NOTIFY_PEERS: 3088 if (event_bond->send_peer_notif) 3089 event_bond->send_peer_notif--; 3090 break; 3091 default: 3092 break; 3093 } 3094 3095 return NOTIFY_DONE; 3096 } 3097 3098 static int bond_slave_netdev_event(unsigned long event, 3099 struct net_device *slave_dev) 3100 { 3101 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3102 struct bonding *bond; 3103 struct net_device *bond_dev; 3104 3105 /* A netdev event can be generated while enslaving a device 3106 * before netdev_rx_handler_register is called in which case 3107 * slave will be NULL 3108 */ 3109 if (!slave) 3110 return NOTIFY_DONE; 3111 bond_dev = slave->bond->dev; 3112 bond = slave->bond; 3113 primary = rtnl_dereference(bond->primary_slave); 3114 3115 switch (event) { 3116 case NETDEV_UNREGISTER: 3117 if (bond_dev->type != ARPHRD_ETHER) 3118 bond_release_and_destroy(bond_dev, slave_dev); 3119 else 3120 __bond_release_one(bond_dev, slave_dev, false, true); 3121 break; 3122 case NETDEV_UP: 3123 case NETDEV_CHANGE: 3124 /* For 802.3ad mode only: 3125 * Getting invalid Speed/Duplex values here will put slave 3126 * in weird state. So mark it as link-fail for the time 3127 * being and let link-monitoring (miimon) set it right when 3128 * correct speeds/duplex are available. 3129 */ 3130 if (bond_update_speed_duplex(slave) && 3131 BOND_MODE(bond) == BOND_MODE_8023AD) 3132 slave->link = BOND_LINK_FAIL; 3133 3134 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3135 bond_3ad_adapter_speed_duplex_changed(slave); 3136 /* Fallthrough */ 3137 case NETDEV_DOWN: 3138 /* Refresh slave-array if applicable! 3139 * If the setup does not use miimon or arpmon (mode-specific!), 3140 * then these events will not cause the slave-array to be 3141 * refreshed. This will cause xmit to use a slave that is not 3142 * usable. Avoid such situation by refeshing the array at these 3143 * events. If these (miimon/arpmon) parameters are configured 3144 * then array gets refreshed twice and that should be fine! 3145 */ 3146 if (bond_mode_can_use_xmit_hash(bond)) 3147 bond_update_slave_arr(bond, NULL); 3148 break; 3149 case NETDEV_CHANGEMTU: 3150 /* TODO: Should slaves be allowed to 3151 * independently alter their MTU? For 3152 * an active-backup bond, slaves need 3153 * not be the same type of device, so 3154 * MTUs may vary. For other modes, 3155 * slaves arguably should have the 3156 * same MTUs. To do this, we'd need to 3157 * take over the slave's change_mtu 3158 * function for the duration of their 3159 * servitude. 3160 */ 3161 break; 3162 case NETDEV_CHANGENAME: 3163 /* we don't care if we don't have primary set */ 3164 if (!bond_uses_primary(bond) || 3165 !bond->params.primary[0]) 3166 break; 3167 3168 if (slave == primary) { 3169 /* slave's name changed - he's no longer primary */ 3170 RCU_INIT_POINTER(bond->primary_slave, NULL); 3171 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3172 /* we have a new primary slave */ 3173 rcu_assign_pointer(bond->primary_slave, slave); 3174 } else { /* we didn't change primary - exit */ 3175 break; 3176 } 3177 3178 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3179 primary ? slave_dev->name : "none"); 3180 3181 block_netpoll_tx(); 3182 bond_select_active_slave(bond); 3183 unblock_netpoll_tx(); 3184 break; 3185 case NETDEV_FEAT_CHANGE: 3186 bond_compute_features(bond); 3187 break; 3188 case NETDEV_RESEND_IGMP: 3189 /* Propagate to master device */ 3190 call_netdevice_notifiers(event, slave->bond->dev); 3191 break; 3192 default: 3193 break; 3194 } 3195 3196 return NOTIFY_DONE; 3197 } 3198 3199 /* bond_netdev_event: handle netdev notifier chain events. 3200 * 3201 * This function receives events for the netdev chain. The caller (an 3202 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3203 * locks for us to safely manipulate the slave devices (RTNL lock, 3204 * dev_probe_lock). 3205 */ 3206 static int bond_netdev_event(struct notifier_block *this, 3207 unsigned long event, void *ptr) 3208 { 3209 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3210 3211 netdev_dbg(event_dev, "event: %lx\n", event); 3212 3213 if (!(event_dev->priv_flags & IFF_BONDING)) 3214 return NOTIFY_DONE; 3215 3216 if (event_dev->flags & IFF_MASTER) { 3217 netdev_dbg(event_dev, "IFF_MASTER\n"); 3218 return bond_master_netdev_event(event, event_dev); 3219 } 3220 3221 if (event_dev->flags & IFF_SLAVE) { 3222 netdev_dbg(event_dev, "IFF_SLAVE\n"); 3223 return bond_slave_netdev_event(event, event_dev); 3224 } 3225 3226 return NOTIFY_DONE; 3227 } 3228 3229 static struct notifier_block bond_netdev_notifier = { 3230 .notifier_call = bond_netdev_event, 3231 }; 3232 3233 /*---------------------------- Hashing Policies -----------------------------*/ 3234 3235 /* L2 hash helper */ 3236 static inline u32 bond_eth_hash(struct sk_buff *skb) 3237 { 3238 struct ethhdr *ep, hdr_tmp; 3239 3240 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3241 if (ep) 3242 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3243 return 0; 3244 } 3245 3246 /* Extract the appropriate headers based on bond's xmit policy */ 3247 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3248 struct flow_keys *fk) 3249 { 3250 const struct ipv6hdr *iph6; 3251 const struct iphdr *iph; 3252 int noff, proto = -1; 3253 3254 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) 3255 return skb_flow_dissect_flow_keys(skb, fk, 0); 3256 3257 fk->ports.ports = 0; 3258 noff = skb_network_offset(skb); 3259 if (skb->protocol == htons(ETH_P_IP)) { 3260 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 3261 return false; 3262 iph = ip_hdr(skb); 3263 iph_to_flow_copy_v4addrs(fk, iph); 3264 noff += iph->ihl << 2; 3265 if (!ip_is_fragment(iph)) 3266 proto = iph->protocol; 3267 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3268 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6)))) 3269 return false; 3270 iph6 = ipv6_hdr(skb); 3271 iph_to_flow_copy_v6addrs(fk, iph6); 3272 noff += sizeof(*iph6); 3273 proto = iph6->nexthdr; 3274 } else { 3275 return false; 3276 } 3277 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) 3278 fk->ports.ports = skb_flow_get_ports(skb, noff, proto); 3279 3280 return true; 3281 } 3282 3283 /** 3284 * bond_xmit_hash - generate a hash value based on the xmit policy 3285 * @bond: bonding device 3286 * @skb: buffer to use for headers 3287 * 3288 * This function will extract the necessary headers from the skb buffer and use 3289 * them to generate a hash based on the xmit_policy set in the bonding device 3290 */ 3291 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3292 { 3293 struct flow_keys flow; 3294 u32 hash; 3295 3296 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3297 skb->l4_hash) 3298 return skb->hash; 3299 3300 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3301 !bond_flow_dissect(bond, skb, &flow)) 3302 return bond_eth_hash(skb); 3303 3304 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3305 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) 3306 hash = bond_eth_hash(skb); 3307 else 3308 hash = (__force u32)flow.ports.ports; 3309 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3310 (__force u32)flow_get_u32_src(&flow); 3311 hash ^= (hash >> 16); 3312 hash ^= (hash >> 8); 3313 3314 return hash >> 1; 3315 } 3316 3317 /*-------------------------- Device entry points ----------------------------*/ 3318 3319 void bond_work_init_all(struct bonding *bond) 3320 { 3321 INIT_DELAYED_WORK(&bond->mcast_work, 3322 bond_resend_igmp_join_requests_delayed); 3323 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3324 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3325 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3326 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3327 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3328 } 3329 3330 static void bond_work_cancel_all(struct bonding *bond) 3331 { 3332 cancel_delayed_work_sync(&bond->mii_work); 3333 cancel_delayed_work_sync(&bond->arp_work); 3334 cancel_delayed_work_sync(&bond->alb_work); 3335 cancel_delayed_work_sync(&bond->ad_work); 3336 cancel_delayed_work_sync(&bond->mcast_work); 3337 cancel_delayed_work_sync(&bond->slave_arr_work); 3338 } 3339 3340 static int bond_open(struct net_device *bond_dev) 3341 { 3342 struct bonding *bond = netdev_priv(bond_dev); 3343 struct list_head *iter; 3344 struct slave *slave; 3345 3346 /* reset slave->backup and slave->inactive */ 3347 if (bond_has_slaves(bond)) { 3348 bond_for_each_slave(bond, slave, iter) { 3349 if (bond_uses_primary(bond) && 3350 slave != rcu_access_pointer(bond->curr_active_slave)) { 3351 bond_set_slave_inactive_flags(slave, 3352 BOND_SLAVE_NOTIFY_NOW); 3353 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3354 bond_set_slave_active_flags(slave, 3355 BOND_SLAVE_NOTIFY_NOW); 3356 } 3357 } 3358 } 3359 3360 if (bond_is_lb(bond)) { 3361 /* bond_alb_initialize must be called before the timer 3362 * is started. 3363 */ 3364 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3365 return -ENOMEM; 3366 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3367 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3368 } 3369 3370 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3371 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3372 3373 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3374 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3375 bond->recv_probe = bond_arp_rcv; 3376 } 3377 3378 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3379 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3380 /* register to receive LACPDUs */ 3381 bond->recv_probe = bond_3ad_lacpdu_recv; 3382 bond_3ad_initiate_agg_selection(bond, 1); 3383 } 3384 3385 if (bond_mode_can_use_xmit_hash(bond)) 3386 bond_update_slave_arr(bond, NULL); 3387 3388 return 0; 3389 } 3390 3391 static int bond_close(struct net_device *bond_dev) 3392 { 3393 struct bonding *bond = netdev_priv(bond_dev); 3394 3395 bond_work_cancel_all(bond); 3396 bond->send_peer_notif = 0; 3397 if (bond_is_lb(bond)) 3398 bond_alb_deinitialize(bond); 3399 bond->recv_probe = NULL; 3400 3401 return 0; 3402 } 3403 3404 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3405 * that some drivers can provide 32bit values only. 3406 */ 3407 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3408 const struct rtnl_link_stats64 *_new, 3409 const struct rtnl_link_stats64 *_old) 3410 { 3411 const u64 *new = (const u64 *)_new; 3412 const u64 *old = (const u64 *)_old; 3413 u64 *res = (u64 *)_res; 3414 int i; 3415 3416 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3417 u64 nv = new[i]; 3418 u64 ov = old[i]; 3419 s64 delta = nv - ov; 3420 3421 /* detects if this particular field is 32bit only */ 3422 if (((nv | ov) >> 32) == 0) 3423 delta = (s64)(s32)((u32)nv - (u32)ov); 3424 3425 /* filter anomalies, some drivers reset their stats 3426 * at down/up events. 3427 */ 3428 if (delta > 0) 3429 res[i] += delta; 3430 } 3431 } 3432 3433 static int bond_get_nest_level(struct net_device *bond_dev) 3434 { 3435 struct bonding *bond = netdev_priv(bond_dev); 3436 3437 return bond->nest_level; 3438 } 3439 3440 static void bond_get_stats(struct net_device *bond_dev, 3441 struct rtnl_link_stats64 *stats) 3442 { 3443 struct bonding *bond = netdev_priv(bond_dev); 3444 struct rtnl_link_stats64 temp; 3445 struct list_head *iter; 3446 struct slave *slave; 3447 3448 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev)); 3449 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3450 3451 rcu_read_lock(); 3452 bond_for_each_slave_rcu(bond, slave, iter) { 3453 const struct rtnl_link_stats64 *new = 3454 dev_get_stats(slave->dev, &temp); 3455 3456 bond_fold_stats(stats, new, &slave->slave_stats); 3457 3458 /* save off the slave stats for the next run */ 3459 memcpy(&slave->slave_stats, new, sizeof(*new)); 3460 } 3461 rcu_read_unlock(); 3462 3463 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3464 spin_unlock(&bond->stats_lock); 3465 } 3466 3467 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3468 { 3469 struct bonding *bond = netdev_priv(bond_dev); 3470 struct net_device *slave_dev = NULL; 3471 struct ifbond k_binfo; 3472 struct ifbond __user *u_binfo = NULL; 3473 struct ifslave k_sinfo; 3474 struct ifslave __user *u_sinfo = NULL; 3475 struct mii_ioctl_data *mii = NULL; 3476 struct bond_opt_value newval; 3477 struct net *net; 3478 int res = 0; 3479 3480 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3481 3482 switch (cmd) { 3483 case SIOCGMIIPHY: 3484 mii = if_mii(ifr); 3485 if (!mii) 3486 return -EINVAL; 3487 3488 mii->phy_id = 0; 3489 /* Fall Through */ 3490 case SIOCGMIIREG: 3491 /* We do this again just in case we were called by SIOCGMIIREG 3492 * instead of SIOCGMIIPHY. 3493 */ 3494 mii = if_mii(ifr); 3495 if (!mii) 3496 return -EINVAL; 3497 3498 if (mii->reg_num == 1) { 3499 mii->val_out = 0; 3500 if (netif_carrier_ok(bond->dev)) 3501 mii->val_out = BMSR_LSTATUS; 3502 } 3503 3504 return 0; 3505 case BOND_INFO_QUERY_OLD: 3506 case SIOCBONDINFOQUERY: 3507 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3508 3509 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3510 return -EFAULT; 3511 3512 bond_info_query(bond_dev, &k_binfo); 3513 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3514 return -EFAULT; 3515 3516 return 0; 3517 case BOND_SLAVE_INFO_QUERY_OLD: 3518 case SIOCBONDSLAVEINFOQUERY: 3519 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3520 3521 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3522 return -EFAULT; 3523 3524 res = bond_slave_info_query(bond_dev, &k_sinfo); 3525 if (res == 0 && 3526 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3527 return -EFAULT; 3528 3529 return res; 3530 default: 3531 break; 3532 } 3533 3534 net = dev_net(bond_dev); 3535 3536 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3537 return -EPERM; 3538 3539 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3540 3541 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev); 3542 3543 if (!slave_dev) 3544 return -ENODEV; 3545 3546 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name); 3547 switch (cmd) { 3548 case BOND_ENSLAVE_OLD: 3549 case SIOCBONDENSLAVE: 3550 res = bond_enslave(bond_dev, slave_dev, NULL); 3551 break; 3552 case BOND_RELEASE_OLD: 3553 case SIOCBONDRELEASE: 3554 res = bond_release(bond_dev, slave_dev); 3555 break; 3556 case BOND_SETHWADDR_OLD: 3557 case SIOCBONDSETHWADDR: 3558 res = bond_set_dev_addr(bond_dev, slave_dev); 3559 break; 3560 case BOND_CHANGE_ACTIVE_OLD: 3561 case SIOCBONDCHANGEACTIVE: 3562 bond_opt_initstr(&newval, slave_dev->name); 3563 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3564 &newval); 3565 break; 3566 default: 3567 res = -EOPNOTSUPP; 3568 } 3569 3570 return res; 3571 } 3572 3573 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3574 { 3575 struct bonding *bond = netdev_priv(bond_dev); 3576 3577 if (change & IFF_PROMISC) 3578 bond_set_promiscuity(bond, 3579 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3580 3581 if (change & IFF_ALLMULTI) 3582 bond_set_allmulti(bond, 3583 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3584 } 3585 3586 static void bond_set_rx_mode(struct net_device *bond_dev) 3587 { 3588 struct bonding *bond = netdev_priv(bond_dev); 3589 struct list_head *iter; 3590 struct slave *slave; 3591 3592 rcu_read_lock(); 3593 if (bond_uses_primary(bond)) { 3594 slave = rcu_dereference(bond->curr_active_slave); 3595 if (slave) { 3596 dev_uc_sync(slave->dev, bond_dev); 3597 dev_mc_sync(slave->dev, bond_dev); 3598 } 3599 } else { 3600 bond_for_each_slave_rcu(bond, slave, iter) { 3601 dev_uc_sync_multiple(slave->dev, bond_dev); 3602 dev_mc_sync_multiple(slave->dev, bond_dev); 3603 } 3604 } 3605 rcu_read_unlock(); 3606 } 3607 3608 static int bond_neigh_init(struct neighbour *n) 3609 { 3610 struct bonding *bond = netdev_priv(n->dev); 3611 const struct net_device_ops *slave_ops; 3612 struct neigh_parms parms; 3613 struct slave *slave; 3614 int ret; 3615 3616 slave = bond_first_slave(bond); 3617 if (!slave) 3618 return 0; 3619 slave_ops = slave->dev->netdev_ops; 3620 if (!slave_ops->ndo_neigh_setup) 3621 return 0; 3622 3623 parms.neigh_setup = NULL; 3624 parms.neigh_cleanup = NULL; 3625 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3626 if (ret) 3627 return ret; 3628 3629 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called 3630 * after the last slave has been detached. Assumes that all slaves 3631 * utilize the same neigh_cleanup (true at this writing as only user 3632 * is ipoib). 3633 */ 3634 n->parms->neigh_cleanup = parms.neigh_cleanup; 3635 3636 if (!parms.neigh_setup) 3637 return 0; 3638 3639 return parms.neigh_setup(n); 3640 } 3641 3642 /* The bonding ndo_neigh_setup is called at init time beofre any 3643 * slave exists. So we must declare proxy setup function which will 3644 * be used at run time to resolve the actual slave neigh param setup. 3645 * 3646 * It's also called by master devices (such as vlans) to setup their 3647 * underlying devices. In that case - do nothing, we're already set up from 3648 * our init. 3649 */ 3650 static int bond_neigh_setup(struct net_device *dev, 3651 struct neigh_parms *parms) 3652 { 3653 /* modify only our neigh_parms */ 3654 if (parms->dev == dev) 3655 parms->neigh_setup = bond_neigh_init; 3656 3657 return 0; 3658 } 3659 3660 /* Change the MTU of all of a master's slaves to match the master */ 3661 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3662 { 3663 struct bonding *bond = netdev_priv(bond_dev); 3664 struct slave *slave, *rollback_slave; 3665 struct list_head *iter; 3666 int res = 0; 3667 3668 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3669 3670 bond_for_each_slave(bond, slave, iter) { 3671 netdev_dbg(bond_dev, "s %p c_m %p\n", 3672 slave, slave->dev->netdev_ops->ndo_change_mtu); 3673 3674 res = dev_set_mtu(slave->dev, new_mtu); 3675 3676 if (res) { 3677 /* If we failed to set the slave's mtu to the new value 3678 * we must abort the operation even in ACTIVE_BACKUP 3679 * mode, because if we allow the backup slaves to have 3680 * different mtu values than the active slave we'll 3681 * need to change their mtu when doing a failover. That 3682 * means changing their mtu from timer context, which 3683 * is probably not a good idea. 3684 */ 3685 netdev_dbg(bond_dev, "err %d %s\n", res, 3686 slave->dev->name); 3687 goto unwind; 3688 } 3689 } 3690 3691 bond_dev->mtu = new_mtu; 3692 3693 return 0; 3694 3695 unwind: 3696 /* unwind from head to the slave that failed */ 3697 bond_for_each_slave(bond, rollback_slave, iter) { 3698 int tmp_res; 3699 3700 if (rollback_slave == slave) 3701 break; 3702 3703 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3704 if (tmp_res) { 3705 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3706 tmp_res, rollback_slave->dev->name); 3707 } 3708 } 3709 3710 return res; 3711 } 3712 3713 /* Change HW address 3714 * 3715 * Note that many devices must be down to change the HW address, and 3716 * downing the master releases all slaves. We can make bonds full of 3717 * bonding devices to test this, however. 3718 */ 3719 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3720 { 3721 struct bonding *bond = netdev_priv(bond_dev); 3722 struct slave *slave, *rollback_slave; 3723 struct sockaddr_storage *ss = addr, tmp_ss; 3724 struct list_head *iter; 3725 int res = 0; 3726 3727 if (BOND_MODE(bond) == BOND_MODE_ALB) 3728 return bond_alb_set_mac_address(bond_dev, addr); 3729 3730 3731 netdev_dbg(bond_dev, "bond=%p\n", bond); 3732 3733 /* If fail_over_mac is enabled, do nothing and return success. 3734 * Returning an error causes ifenslave to fail. 3735 */ 3736 if (bond->params.fail_over_mac && 3737 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3738 return 0; 3739 3740 if (!is_valid_ether_addr(ss->__data)) 3741 return -EADDRNOTAVAIL; 3742 3743 bond_for_each_slave(bond, slave, iter) { 3744 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name); 3745 res = dev_set_mac_address(slave->dev, addr, NULL); 3746 if (res) { 3747 /* TODO: consider downing the slave 3748 * and retry ? 3749 * User should expect communications 3750 * breakage anyway until ARP finish 3751 * updating, so... 3752 */ 3753 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name); 3754 goto unwind; 3755 } 3756 } 3757 3758 /* success */ 3759 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 3760 return 0; 3761 3762 unwind: 3763 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 3764 tmp_ss.ss_family = bond_dev->type; 3765 3766 /* unwind from head to the slave that failed */ 3767 bond_for_each_slave(bond, rollback_slave, iter) { 3768 int tmp_res; 3769 3770 if (rollback_slave == slave) 3771 break; 3772 3773 tmp_res = dev_set_mac_address(rollback_slave->dev, 3774 (struct sockaddr *)&tmp_ss, NULL); 3775 if (tmp_res) { 3776 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3777 tmp_res, rollback_slave->dev->name); 3778 } 3779 } 3780 3781 return res; 3782 } 3783 3784 /** 3785 * bond_xmit_slave_id - transmit skb through slave with slave_id 3786 * @bond: bonding device that is transmitting 3787 * @skb: buffer to transmit 3788 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3789 * 3790 * This function tries to transmit through slave with slave_id but in case 3791 * it fails, it tries to find the first available slave for transmission. 3792 * The skb is consumed in all cases, thus the function is void. 3793 */ 3794 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3795 { 3796 struct list_head *iter; 3797 struct slave *slave; 3798 int i = slave_id; 3799 3800 /* Here we start from the slave with slave_id */ 3801 bond_for_each_slave_rcu(bond, slave, iter) { 3802 if (--i < 0) { 3803 if (bond_slave_can_tx(slave)) { 3804 bond_dev_queue_xmit(bond, skb, slave->dev); 3805 return; 3806 } 3807 } 3808 } 3809 3810 /* Here we start from the first slave up to slave_id */ 3811 i = slave_id; 3812 bond_for_each_slave_rcu(bond, slave, iter) { 3813 if (--i < 0) 3814 break; 3815 if (bond_slave_can_tx(slave)) { 3816 bond_dev_queue_xmit(bond, skb, slave->dev); 3817 return; 3818 } 3819 } 3820 /* no slave that can tx has been found */ 3821 bond_tx_drop(bond->dev, skb); 3822 } 3823 3824 /** 3825 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3826 * @bond: bonding device to use 3827 * 3828 * Based on the value of the bonding device's packets_per_slave parameter 3829 * this function generates a slave id, which is usually used as the next 3830 * slave to transmit through. 3831 */ 3832 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3833 { 3834 u32 slave_id; 3835 struct reciprocal_value reciprocal_packets_per_slave; 3836 int packets_per_slave = bond->params.packets_per_slave; 3837 3838 switch (packets_per_slave) { 3839 case 0: 3840 slave_id = prandom_u32(); 3841 break; 3842 case 1: 3843 slave_id = bond->rr_tx_counter; 3844 break; 3845 default: 3846 reciprocal_packets_per_slave = 3847 bond->params.reciprocal_packets_per_slave; 3848 slave_id = reciprocal_divide(bond->rr_tx_counter, 3849 reciprocal_packets_per_slave); 3850 break; 3851 } 3852 bond->rr_tx_counter++; 3853 3854 return slave_id; 3855 } 3856 3857 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 3858 struct net_device *bond_dev) 3859 { 3860 struct bonding *bond = netdev_priv(bond_dev); 3861 struct iphdr *iph = ip_hdr(skb); 3862 struct slave *slave; 3863 u32 slave_id; 3864 3865 /* Start with the curr_active_slave that joined the bond as the 3866 * default for sending IGMP traffic. For failover purposes one 3867 * needs to maintain some consistency for the interface that will 3868 * send the join/membership reports. The curr_active_slave found 3869 * will send all of this type of traffic. 3870 */ 3871 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) { 3872 slave = rcu_dereference(bond->curr_active_slave); 3873 if (slave) 3874 bond_dev_queue_xmit(bond, skb, slave->dev); 3875 else 3876 bond_xmit_slave_id(bond, skb, 0); 3877 } else { 3878 int slave_cnt = READ_ONCE(bond->slave_cnt); 3879 3880 if (likely(slave_cnt)) { 3881 slave_id = bond_rr_gen_slave_id(bond); 3882 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3883 } else { 3884 bond_tx_drop(bond_dev, skb); 3885 } 3886 } 3887 3888 return NETDEV_TX_OK; 3889 } 3890 3891 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3892 * the bond has a usable interface. 3893 */ 3894 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 3895 struct net_device *bond_dev) 3896 { 3897 struct bonding *bond = netdev_priv(bond_dev); 3898 struct slave *slave; 3899 3900 slave = rcu_dereference(bond->curr_active_slave); 3901 if (slave) 3902 bond_dev_queue_xmit(bond, skb, slave->dev); 3903 else 3904 bond_tx_drop(bond_dev, skb); 3905 3906 return NETDEV_TX_OK; 3907 } 3908 3909 /* Use this to update slave_array when (a) it's not appropriate to update 3910 * slave_array right away (note that update_slave_array() may sleep) 3911 * and / or (b) RTNL is not held. 3912 */ 3913 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 3914 { 3915 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 3916 } 3917 3918 /* Slave array work handler. Holds only RTNL */ 3919 static void bond_slave_arr_handler(struct work_struct *work) 3920 { 3921 struct bonding *bond = container_of(work, struct bonding, 3922 slave_arr_work.work); 3923 int ret; 3924 3925 if (!rtnl_trylock()) 3926 goto err; 3927 3928 ret = bond_update_slave_arr(bond, NULL); 3929 rtnl_unlock(); 3930 if (ret) { 3931 pr_warn_ratelimited("Failed to update slave array from WT\n"); 3932 goto err; 3933 } 3934 return; 3935 3936 err: 3937 bond_slave_arr_work_rearm(bond, 1); 3938 } 3939 3940 /* Build the usable slaves array in control path for modes that use xmit-hash 3941 * to determine the slave interface - 3942 * (a) BOND_MODE_8023AD 3943 * (b) BOND_MODE_XOR 3944 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 3945 * 3946 * The caller is expected to hold RTNL only and NO other lock! 3947 */ 3948 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 3949 { 3950 struct slave *slave; 3951 struct list_head *iter; 3952 struct bond_up_slave *new_arr, *old_arr; 3953 int agg_id = 0; 3954 int ret = 0; 3955 3956 #ifdef CONFIG_LOCKDEP 3957 WARN_ON(lockdep_is_held(&bond->mode_lock)); 3958 #endif 3959 3960 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 3961 GFP_KERNEL); 3962 if (!new_arr) { 3963 ret = -ENOMEM; 3964 pr_err("Failed to build slave-array.\n"); 3965 goto out; 3966 } 3967 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3968 struct ad_info ad_info; 3969 3970 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 3971 pr_debug("bond_3ad_get_active_agg_info failed\n"); 3972 kfree_rcu(new_arr, rcu); 3973 /* No active aggragator means it's not safe to use 3974 * the previous array. 3975 */ 3976 old_arr = rtnl_dereference(bond->slave_arr); 3977 if (old_arr) { 3978 RCU_INIT_POINTER(bond->slave_arr, NULL); 3979 kfree_rcu(old_arr, rcu); 3980 } 3981 goto out; 3982 } 3983 agg_id = ad_info.aggregator_id; 3984 } 3985 bond_for_each_slave(bond, slave, iter) { 3986 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3987 struct aggregator *agg; 3988 3989 agg = SLAVE_AD_INFO(slave)->port.aggregator; 3990 if (!agg || agg->aggregator_identifier != agg_id) 3991 continue; 3992 } 3993 if (!bond_slave_can_tx(slave)) 3994 continue; 3995 if (skipslave == slave) 3996 continue; 3997 3998 netdev_dbg(bond->dev, 3999 "Adding slave dev %s to tx hash array[%d]\n", 4000 slave->dev->name, new_arr->count); 4001 4002 new_arr->arr[new_arr->count++] = slave; 4003 } 4004 4005 old_arr = rtnl_dereference(bond->slave_arr); 4006 rcu_assign_pointer(bond->slave_arr, new_arr); 4007 if (old_arr) 4008 kfree_rcu(old_arr, rcu); 4009 out: 4010 if (ret != 0 && skipslave) { 4011 int idx; 4012 4013 /* Rare situation where caller has asked to skip a specific 4014 * slave but allocation failed (most likely!). BTW this is 4015 * only possible when the call is initiated from 4016 * __bond_release_one(). In this situation; overwrite the 4017 * skipslave entry in the array with the last entry from the 4018 * array to avoid a situation where the xmit path may choose 4019 * this to-be-skipped slave to send a packet out. 4020 */ 4021 old_arr = rtnl_dereference(bond->slave_arr); 4022 for (idx = 0; idx < old_arr->count; idx++) { 4023 if (skipslave == old_arr->arr[idx]) { 4024 old_arr->arr[idx] = 4025 old_arr->arr[old_arr->count-1]; 4026 old_arr->count--; 4027 break; 4028 } 4029 } 4030 } 4031 return ret; 4032 } 4033 4034 /* Use this Xmit function for 3AD as well as XOR modes. The current 4035 * usable slave array is formed in the control path. The xmit function 4036 * just calculates hash and sends the packet out. 4037 */ 4038 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4039 struct net_device *dev) 4040 { 4041 struct bonding *bond = netdev_priv(dev); 4042 struct slave *slave; 4043 struct bond_up_slave *slaves; 4044 unsigned int count; 4045 4046 slaves = rcu_dereference(bond->slave_arr); 4047 count = slaves ? READ_ONCE(slaves->count) : 0; 4048 if (likely(count)) { 4049 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 4050 bond_dev_queue_xmit(bond, skb, slave->dev); 4051 } else { 4052 bond_tx_drop(dev, skb); 4053 } 4054 4055 return NETDEV_TX_OK; 4056 } 4057 4058 /* in broadcast mode, we send everything to all usable interfaces. */ 4059 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4060 struct net_device *bond_dev) 4061 { 4062 struct bonding *bond = netdev_priv(bond_dev); 4063 struct slave *slave = NULL; 4064 struct list_head *iter; 4065 4066 bond_for_each_slave_rcu(bond, slave, iter) { 4067 if (bond_is_last_slave(bond, slave)) 4068 break; 4069 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4070 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4071 4072 if (!skb2) { 4073 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4074 bond_dev->name, __func__); 4075 continue; 4076 } 4077 bond_dev_queue_xmit(bond, skb2, slave->dev); 4078 } 4079 } 4080 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4081 bond_dev_queue_xmit(bond, skb, slave->dev); 4082 else 4083 bond_tx_drop(bond_dev, skb); 4084 4085 return NETDEV_TX_OK; 4086 } 4087 4088 /*------------------------- Device initialization ---------------------------*/ 4089 4090 /* Lookup the slave that corresponds to a qid */ 4091 static inline int bond_slave_override(struct bonding *bond, 4092 struct sk_buff *skb) 4093 { 4094 struct slave *slave = NULL; 4095 struct list_head *iter; 4096 4097 if (!skb_rx_queue_recorded(skb)) 4098 return 1; 4099 4100 /* Find out if any slaves have the same mapping as this skb. */ 4101 bond_for_each_slave_rcu(bond, slave, iter) { 4102 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4103 if (bond_slave_is_up(slave) && 4104 slave->link == BOND_LINK_UP) { 4105 bond_dev_queue_xmit(bond, skb, slave->dev); 4106 return 0; 4107 } 4108 /* If the slave isn't UP, use default transmit policy. */ 4109 break; 4110 } 4111 } 4112 4113 return 1; 4114 } 4115 4116 4117 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4118 struct net_device *sb_dev, 4119 select_queue_fallback_t fallback) 4120 { 4121 /* This helper function exists to help dev_pick_tx get the correct 4122 * destination queue. Using a helper function skips a call to 4123 * skb_tx_hash and will put the skbs in the queue we expect on their 4124 * way down to the bonding driver. 4125 */ 4126 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4127 4128 /* Save the original txq to restore before passing to the driver */ 4129 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4130 4131 if (unlikely(txq >= dev->real_num_tx_queues)) { 4132 do { 4133 txq -= dev->real_num_tx_queues; 4134 } while (txq >= dev->real_num_tx_queues); 4135 } 4136 return txq; 4137 } 4138 4139 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4140 { 4141 struct bonding *bond = netdev_priv(dev); 4142 4143 if (bond_should_override_tx_queue(bond) && 4144 !bond_slave_override(bond, skb)) 4145 return NETDEV_TX_OK; 4146 4147 switch (BOND_MODE(bond)) { 4148 case BOND_MODE_ROUNDROBIN: 4149 return bond_xmit_roundrobin(skb, dev); 4150 case BOND_MODE_ACTIVEBACKUP: 4151 return bond_xmit_activebackup(skb, dev); 4152 case BOND_MODE_8023AD: 4153 case BOND_MODE_XOR: 4154 return bond_3ad_xor_xmit(skb, dev); 4155 case BOND_MODE_BROADCAST: 4156 return bond_xmit_broadcast(skb, dev); 4157 case BOND_MODE_ALB: 4158 return bond_alb_xmit(skb, dev); 4159 case BOND_MODE_TLB: 4160 return bond_tlb_xmit(skb, dev); 4161 default: 4162 /* Should never happen, mode already checked */ 4163 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4164 WARN_ON_ONCE(1); 4165 bond_tx_drop(dev, skb); 4166 return NETDEV_TX_OK; 4167 } 4168 } 4169 4170 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4171 { 4172 struct bonding *bond = netdev_priv(dev); 4173 netdev_tx_t ret = NETDEV_TX_OK; 4174 4175 /* If we risk deadlock from transmitting this in the 4176 * netpoll path, tell netpoll to queue the frame for later tx 4177 */ 4178 if (unlikely(is_netpoll_tx_blocked(dev))) 4179 return NETDEV_TX_BUSY; 4180 4181 rcu_read_lock(); 4182 if (bond_has_slaves(bond)) 4183 ret = __bond_start_xmit(skb, dev); 4184 else 4185 bond_tx_drop(dev, skb); 4186 rcu_read_unlock(); 4187 4188 return ret; 4189 } 4190 4191 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4192 struct ethtool_link_ksettings *cmd) 4193 { 4194 struct bonding *bond = netdev_priv(bond_dev); 4195 unsigned long speed = 0; 4196 struct list_head *iter; 4197 struct slave *slave; 4198 4199 cmd->base.duplex = DUPLEX_UNKNOWN; 4200 cmd->base.port = PORT_OTHER; 4201 4202 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4203 * do not need to check mode. Though link speed might not represent 4204 * the true receive or transmit bandwidth (not all modes are symmetric) 4205 * this is an accurate maximum. 4206 */ 4207 bond_for_each_slave(bond, slave, iter) { 4208 if (bond_slave_can_tx(slave)) { 4209 if (slave->speed != SPEED_UNKNOWN) 4210 speed += slave->speed; 4211 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4212 slave->duplex != DUPLEX_UNKNOWN) 4213 cmd->base.duplex = slave->duplex; 4214 } 4215 } 4216 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4217 4218 return 0; 4219 } 4220 4221 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4222 struct ethtool_drvinfo *drvinfo) 4223 { 4224 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4225 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 4226 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4227 BOND_ABI_VERSION); 4228 } 4229 4230 static const struct ethtool_ops bond_ethtool_ops = { 4231 .get_drvinfo = bond_ethtool_get_drvinfo, 4232 .get_link = ethtool_op_get_link, 4233 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4234 }; 4235 4236 static const struct net_device_ops bond_netdev_ops = { 4237 .ndo_init = bond_init, 4238 .ndo_uninit = bond_uninit, 4239 .ndo_open = bond_open, 4240 .ndo_stop = bond_close, 4241 .ndo_start_xmit = bond_start_xmit, 4242 .ndo_select_queue = bond_select_queue, 4243 .ndo_get_stats64 = bond_get_stats, 4244 .ndo_do_ioctl = bond_do_ioctl, 4245 .ndo_change_rx_flags = bond_change_rx_flags, 4246 .ndo_set_rx_mode = bond_set_rx_mode, 4247 .ndo_change_mtu = bond_change_mtu, 4248 .ndo_set_mac_address = bond_set_mac_address, 4249 .ndo_neigh_setup = bond_neigh_setup, 4250 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4251 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4252 .ndo_get_lock_subclass = bond_get_nest_level, 4253 #ifdef CONFIG_NET_POLL_CONTROLLER 4254 .ndo_netpoll_setup = bond_netpoll_setup, 4255 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4256 .ndo_poll_controller = bond_poll_controller, 4257 #endif 4258 .ndo_add_slave = bond_enslave, 4259 .ndo_del_slave = bond_release, 4260 .ndo_fix_features = bond_fix_features, 4261 .ndo_features_check = passthru_features_check, 4262 }; 4263 4264 static const struct device_type bond_type = { 4265 .name = "bond", 4266 }; 4267 4268 static void bond_destructor(struct net_device *bond_dev) 4269 { 4270 struct bonding *bond = netdev_priv(bond_dev); 4271 if (bond->wq) 4272 destroy_workqueue(bond->wq); 4273 } 4274 4275 void bond_setup(struct net_device *bond_dev) 4276 { 4277 struct bonding *bond = netdev_priv(bond_dev); 4278 4279 spin_lock_init(&bond->mode_lock); 4280 spin_lock_init(&bond->stats_lock); 4281 bond->params = bonding_defaults; 4282 4283 /* Initialize pointers */ 4284 bond->dev = bond_dev; 4285 4286 /* Initialize the device entry points */ 4287 ether_setup(bond_dev); 4288 bond_dev->max_mtu = ETH_MAX_MTU; 4289 bond_dev->netdev_ops = &bond_netdev_ops; 4290 bond_dev->ethtool_ops = &bond_ethtool_ops; 4291 4292 bond_dev->needs_free_netdev = true; 4293 bond_dev->priv_destructor = bond_destructor; 4294 4295 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4296 4297 /* Initialize the device options */ 4298 bond_dev->flags |= IFF_MASTER; 4299 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4300 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4301 4302 /* don't acquire bond device's netif_tx_lock when transmitting */ 4303 bond_dev->features |= NETIF_F_LLTX; 4304 4305 /* By default, we declare the bond to be fully 4306 * VLAN hardware accelerated capable. Special 4307 * care is taken in the various xmit functions 4308 * when there are slaves that are not hw accel 4309 * capable 4310 */ 4311 4312 /* Don't allow bond devices to change network namespaces. */ 4313 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4314 4315 bond_dev->hw_features = BOND_VLAN_FEATURES | 4316 NETIF_F_HW_VLAN_CTAG_TX | 4317 NETIF_F_HW_VLAN_CTAG_RX | 4318 NETIF_F_HW_VLAN_CTAG_FILTER; 4319 4320 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4321 bond_dev->features |= bond_dev->hw_features; 4322 } 4323 4324 /* Destroy a bonding device. 4325 * Must be under rtnl_lock when this function is called. 4326 */ 4327 static void bond_uninit(struct net_device *bond_dev) 4328 { 4329 struct bonding *bond = netdev_priv(bond_dev); 4330 struct list_head *iter; 4331 struct slave *slave; 4332 struct bond_up_slave *arr; 4333 4334 bond_netpoll_cleanup(bond_dev); 4335 4336 /* Release the bonded slaves */ 4337 bond_for_each_slave(bond, slave, iter) 4338 __bond_release_one(bond_dev, slave->dev, true, true); 4339 netdev_info(bond_dev, "Released all slaves\n"); 4340 4341 arr = rtnl_dereference(bond->slave_arr); 4342 if (arr) { 4343 RCU_INIT_POINTER(bond->slave_arr, NULL); 4344 kfree_rcu(arr, rcu); 4345 } 4346 4347 list_del(&bond->bond_list); 4348 4349 bond_debug_unregister(bond); 4350 } 4351 4352 /*------------------------- Module initialization ---------------------------*/ 4353 4354 static int bond_check_params(struct bond_params *params) 4355 { 4356 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4357 struct bond_opt_value newval; 4358 const struct bond_opt_value *valptr; 4359 int arp_all_targets_value = 0; 4360 u16 ad_actor_sys_prio = 0; 4361 u16 ad_user_port_key = 0; 4362 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4363 int arp_ip_count; 4364 int bond_mode = BOND_MODE_ROUNDROBIN; 4365 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4366 int lacp_fast = 0; 4367 int tlb_dynamic_lb; 4368 4369 /* Convert string parameters. */ 4370 if (mode) { 4371 bond_opt_initstr(&newval, mode); 4372 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4373 if (!valptr) { 4374 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4375 return -EINVAL; 4376 } 4377 bond_mode = valptr->value; 4378 } 4379 4380 if (xmit_hash_policy) { 4381 if (bond_mode == BOND_MODE_ROUNDROBIN || 4382 bond_mode == BOND_MODE_ACTIVEBACKUP || 4383 bond_mode == BOND_MODE_BROADCAST) { 4384 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4385 bond_mode_name(bond_mode)); 4386 } else { 4387 bond_opt_initstr(&newval, xmit_hash_policy); 4388 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4389 &newval); 4390 if (!valptr) { 4391 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4392 xmit_hash_policy); 4393 return -EINVAL; 4394 } 4395 xmit_hashtype = valptr->value; 4396 } 4397 } 4398 4399 if (lacp_rate) { 4400 if (bond_mode != BOND_MODE_8023AD) { 4401 pr_info("lacp_rate param is irrelevant in mode %s\n", 4402 bond_mode_name(bond_mode)); 4403 } else { 4404 bond_opt_initstr(&newval, lacp_rate); 4405 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4406 &newval); 4407 if (!valptr) { 4408 pr_err("Error: Invalid lacp rate \"%s\"\n", 4409 lacp_rate); 4410 return -EINVAL; 4411 } 4412 lacp_fast = valptr->value; 4413 } 4414 } 4415 4416 if (ad_select) { 4417 bond_opt_initstr(&newval, ad_select); 4418 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4419 &newval); 4420 if (!valptr) { 4421 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4422 return -EINVAL; 4423 } 4424 params->ad_select = valptr->value; 4425 if (bond_mode != BOND_MODE_8023AD) 4426 pr_warn("ad_select param only affects 802.3ad mode\n"); 4427 } else { 4428 params->ad_select = BOND_AD_STABLE; 4429 } 4430 4431 if (max_bonds < 0) { 4432 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4433 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4434 max_bonds = BOND_DEFAULT_MAX_BONDS; 4435 } 4436 4437 if (miimon < 0) { 4438 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4439 miimon, INT_MAX); 4440 miimon = 0; 4441 } 4442 4443 if (updelay < 0) { 4444 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4445 updelay, INT_MAX); 4446 updelay = 0; 4447 } 4448 4449 if (downdelay < 0) { 4450 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4451 downdelay, INT_MAX); 4452 downdelay = 0; 4453 } 4454 4455 if ((use_carrier != 0) && (use_carrier != 1)) { 4456 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4457 use_carrier); 4458 use_carrier = 1; 4459 } 4460 4461 if (num_peer_notif < 0 || num_peer_notif > 255) { 4462 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4463 num_peer_notif); 4464 num_peer_notif = 1; 4465 } 4466 4467 /* reset values for 802.3ad/TLB/ALB */ 4468 if (!bond_mode_uses_arp(bond_mode)) { 4469 if (!miimon) { 4470 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"); 4471 pr_warn("Forcing miimon to 100msec\n"); 4472 miimon = BOND_DEFAULT_MIIMON; 4473 } 4474 } 4475 4476 if (tx_queues < 1 || tx_queues > 255) { 4477 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4478 tx_queues, BOND_DEFAULT_TX_QUEUES); 4479 tx_queues = BOND_DEFAULT_TX_QUEUES; 4480 } 4481 4482 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4483 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4484 all_slaves_active); 4485 all_slaves_active = 0; 4486 } 4487 4488 if (resend_igmp < 0 || resend_igmp > 255) { 4489 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4490 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4491 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4492 } 4493 4494 bond_opt_initval(&newval, packets_per_slave); 4495 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4496 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4497 packets_per_slave, USHRT_MAX); 4498 packets_per_slave = 1; 4499 } 4500 4501 if (bond_mode == BOND_MODE_ALB) { 4502 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", 4503 updelay); 4504 } 4505 4506 if (!miimon) { 4507 if (updelay || downdelay) { 4508 /* just warn the user the up/down delay will have 4509 * no effect since miimon is zero... 4510 */ 4511 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", 4512 updelay, downdelay); 4513 } 4514 } else { 4515 /* don't allow arp monitoring */ 4516 if (arp_interval) { 4517 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4518 miimon, arp_interval); 4519 arp_interval = 0; 4520 } 4521 4522 if ((updelay % miimon) != 0) { 4523 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4524 updelay, miimon, (updelay / miimon) * miimon); 4525 } 4526 4527 updelay /= miimon; 4528 4529 if ((downdelay % miimon) != 0) { 4530 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4531 downdelay, miimon, 4532 (downdelay / miimon) * miimon); 4533 } 4534 4535 downdelay /= miimon; 4536 } 4537 4538 if (arp_interval < 0) { 4539 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4540 arp_interval, INT_MAX); 4541 arp_interval = 0; 4542 } 4543 4544 for (arp_ip_count = 0, i = 0; 4545 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4546 __be32 ip; 4547 4548 /* not a complete check, but good enough to catch mistakes */ 4549 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4550 !bond_is_ip_target_ok(ip)) { 4551 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4552 arp_ip_target[i]); 4553 arp_interval = 0; 4554 } else { 4555 if (bond_get_targets_ip(arp_target, ip) == -1) 4556 arp_target[arp_ip_count++] = ip; 4557 else 4558 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4559 &ip); 4560 } 4561 } 4562 4563 if (arp_interval && !arp_ip_count) { 4564 /* don't allow arping if no arp_ip_target given... */ 4565 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4566 arp_interval); 4567 arp_interval = 0; 4568 } 4569 4570 if (arp_validate) { 4571 if (!arp_interval) { 4572 pr_err("arp_validate requires arp_interval\n"); 4573 return -EINVAL; 4574 } 4575 4576 bond_opt_initstr(&newval, arp_validate); 4577 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4578 &newval); 4579 if (!valptr) { 4580 pr_err("Error: invalid arp_validate \"%s\"\n", 4581 arp_validate); 4582 return -EINVAL; 4583 } 4584 arp_validate_value = valptr->value; 4585 } else { 4586 arp_validate_value = 0; 4587 } 4588 4589 if (arp_all_targets) { 4590 bond_opt_initstr(&newval, arp_all_targets); 4591 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4592 &newval); 4593 if (!valptr) { 4594 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4595 arp_all_targets); 4596 arp_all_targets_value = 0; 4597 } else { 4598 arp_all_targets_value = valptr->value; 4599 } 4600 } 4601 4602 if (miimon) { 4603 pr_info("MII link monitoring set to %d ms\n", miimon); 4604 } else if (arp_interval) { 4605 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4606 arp_validate_value); 4607 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4608 arp_interval, valptr->string, arp_ip_count); 4609 4610 for (i = 0; i < arp_ip_count; i++) 4611 pr_cont(" %s", arp_ip_target[i]); 4612 4613 pr_cont("\n"); 4614 4615 } else if (max_bonds) { 4616 /* miimon and arp_interval not set, we need one so things 4617 * work as expected, see bonding.txt for details 4618 */ 4619 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"); 4620 } 4621 4622 if (primary && !bond_mode_uses_primary(bond_mode)) { 4623 /* currently, using a primary only makes sense 4624 * in active backup, TLB or ALB modes 4625 */ 4626 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4627 primary, bond_mode_name(bond_mode)); 4628 primary = NULL; 4629 } 4630 4631 if (primary && primary_reselect) { 4632 bond_opt_initstr(&newval, primary_reselect); 4633 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4634 &newval); 4635 if (!valptr) { 4636 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4637 primary_reselect); 4638 return -EINVAL; 4639 } 4640 primary_reselect_value = valptr->value; 4641 } else { 4642 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4643 } 4644 4645 if (fail_over_mac) { 4646 bond_opt_initstr(&newval, fail_over_mac); 4647 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4648 &newval); 4649 if (!valptr) { 4650 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4651 fail_over_mac); 4652 return -EINVAL; 4653 } 4654 fail_over_mac_value = valptr->value; 4655 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4656 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4657 } else { 4658 fail_over_mac_value = BOND_FOM_NONE; 4659 } 4660 4661 bond_opt_initstr(&newval, "default"); 4662 valptr = bond_opt_parse( 4663 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 4664 &newval); 4665 if (!valptr) { 4666 pr_err("Error: No ad_actor_sys_prio default value"); 4667 return -EINVAL; 4668 } 4669 ad_actor_sys_prio = valptr->value; 4670 4671 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 4672 &newval); 4673 if (!valptr) { 4674 pr_err("Error: No ad_user_port_key default value"); 4675 return -EINVAL; 4676 } 4677 ad_user_port_key = valptr->value; 4678 4679 bond_opt_initstr(&newval, "default"); 4680 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 4681 if (!valptr) { 4682 pr_err("Error: No tlb_dynamic_lb default value"); 4683 return -EINVAL; 4684 } 4685 tlb_dynamic_lb = valptr->value; 4686 4687 if (lp_interval == 0) { 4688 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4689 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4690 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4691 } 4692 4693 /* fill params struct with the proper values */ 4694 params->mode = bond_mode; 4695 params->xmit_policy = xmit_hashtype; 4696 params->miimon = miimon; 4697 params->num_peer_notif = num_peer_notif; 4698 params->arp_interval = arp_interval; 4699 params->arp_validate = arp_validate_value; 4700 params->arp_all_targets = arp_all_targets_value; 4701 params->updelay = updelay; 4702 params->downdelay = downdelay; 4703 params->use_carrier = use_carrier; 4704 params->lacp_fast = lacp_fast; 4705 params->primary[0] = 0; 4706 params->primary_reselect = primary_reselect_value; 4707 params->fail_over_mac = fail_over_mac_value; 4708 params->tx_queues = tx_queues; 4709 params->all_slaves_active = all_slaves_active; 4710 params->resend_igmp = resend_igmp; 4711 params->min_links = min_links; 4712 params->lp_interval = lp_interval; 4713 params->packets_per_slave = packets_per_slave; 4714 params->tlb_dynamic_lb = tlb_dynamic_lb; 4715 params->ad_actor_sys_prio = ad_actor_sys_prio; 4716 eth_zero_addr(params->ad_actor_system); 4717 params->ad_user_port_key = ad_user_port_key; 4718 if (packets_per_slave > 0) { 4719 params->reciprocal_packets_per_slave = 4720 reciprocal_value(packets_per_slave); 4721 } else { 4722 /* reciprocal_packets_per_slave is unused if 4723 * packets_per_slave is 0 or 1, just initialize it 4724 */ 4725 params->reciprocal_packets_per_slave = 4726 (struct reciprocal_value) { 0 }; 4727 } 4728 4729 if (primary) { 4730 strncpy(params->primary, primary, IFNAMSIZ); 4731 params->primary[IFNAMSIZ - 1] = 0; 4732 } 4733 4734 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4735 4736 return 0; 4737 } 4738 4739 /* Called from registration process */ 4740 static int bond_init(struct net_device *bond_dev) 4741 { 4742 struct bonding *bond = netdev_priv(bond_dev); 4743 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4744 4745 netdev_dbg(bond_dev, "Begin bond_init\n"); 4746 4747 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 4748 if (!bond->wq) 4749 return -ENOMEM; 4750 4751 bond->nest_level = SINGLE_DEPTH_NESTING; 4752 netdev_lockdep_set_classes(bond_dev); 4753 4754 list_add_tail(&bond->bond_list, &bn->dev_list); 4755 4756 bond_prepare_sysfs_group(bond); 4757 4758 bond_debug_register(bond); 4759 4760 /* Ensure valid dev_addr */ 4761 if (is_zero_ether_addr(bond_dev->dev_addr) && 4762 bond_dev->addr_assign_type == NET_ADDR_PERM) 4763 eth_hw_addr_random(bond_dev); 4764 4765 return 0; 4766 } 4767 4768 unsigned int bond_get_num_tx_queues(void) 4769 { 4770 return tx_queues; 4771 } 4772 4773 /* Create a new bond based on the specified name and bonding parameters. 4774 * If name is NULL, obtain a suitable "bond%d" name for us. 4775 * Caller must NOT hold rtnl_lock; we need to release it here before we 4776 * set up our sysfs entries. 4777 */ 4778 int bond_create(struct net *net, const char *name) 4779 { 4780 struct net_device *bond_dev; 4781 struct bonding *bond; 4782 struct alb_bond_info *bond_info; 4783 int res; 4784 4785 rtnl_lock(); 4786 4787 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4788 name ? name : "bond%d", NET_NAME_UNKNOWN, 4789 bond_setup, tx_queues); 4790 if (!bond_dev) { 4791 pr_err("%s: eek! can't alloc netdev!\n", name); 4792 rtnl_unlock(); 4793 return -ENOMEM; 4794 } 4795 4796 /* 4797 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 4798 * It is set to 0 by default which is wrong. 4799 */ 4800 bond = netdev_priv(bond_dev); 4801 bond_info = &(BOND_ALB_INFO(bond)); 4802 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 4803 4804 dev_net_set(bond_dev, net); 4805 bond_dev->rtnl_link_ops = &bond_link_ops; 4806 4807 res = register_netdevice(bond_dev); 4808 4809 netif_carrier_off(bond_dev); 4810 4811 bond_work_init_all(bond); 4812 4813 rtnl_unlock(); 4814 if (res < 0) 4815 free_netdev(bond_dev); 4816 return res; 4817 } 4818 4819 static int __net_init bond_net_init(struct net *net) 4820 { 4821 struct bond_net *bn = net_generic(net, bond_net_id); 4822 4823 bn->net = net; 4824 INIT_LIST_HEAD(&bn->dev_list); 4825 4826 bond_create_proc_dir(bn); 4827 bond_create_sysfs(bn); 4828 4829 return 0; 4830 } 4831 4832 static void __net_exit bond_net_exit(struct net *net) 4833 { 4834 struct bond_net *bn = net_generic(net, bond_net_id); 4835 struct bonding *bond, *tmp_bond; 4836 LIST_HEAD(list); 4837 4838 bond_destroy_sysfs(bn); 4839 4840 /* Kill off any bonds created after unregistering bond rtnl ops */ 4841 rtnl_lock(); 4842 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4843 unregister_netdevice_queue(bond->dev, &list); 4844 unregister_netdevice_many(&list); 4845 rtnl_unlock(); 4846 4847 bond_destroy_proc_dir(bn); 4848 } 4849 4850 static struct pernet_operations bond_net_ops = { 4851 .init = bond_net_init, 4852 .exit = bond_net_exit, 4853 .id = &bond_net_id, 4854 .size = sizeof(struct bond_net), 4855 }; 4856 4857 static int __init bonding_init(void) 4858 { 4859 int i; 4860 int res; 4861 4862 pr_info("%s", bond_version); 4863 4864 res = bond_check_params(&bonding_defaults); 4865 if (res) 4866 goto out; 4867 4868 res = register_pernet_subsys(&bond_net_ops); 4869 if (res) 4870 goto out; 4871 4872 res = bond_netlink_init(); 4873 if (res) 4874 goto err_link; 4875 4876 bond_create_debugfs(); 4877 4878 for (i = 0; i < max_bonds; i++) { 4879 res = bond_create(&init_net, NULL); 4880 if (res) 4881 goto err; 4882 } 4883 4884 register_netdevice_notifier(&bond_netdev_notifier); 4885 out: 4886 return res; 4887 err: 4888 bond_destroy_debugfs(); 4889 bond_netlink_fini(); 4890 err_link: 4891 unregister_pernet_subsys(&bond_net_ops); 4892 goto out; 4893 4894 } 4895 4896 static void __exit bonding_exit(void) 4897 { 4898 unregister_netdevice_notifier(&bond_netdev_notifier); 4899 4900 bond_destroy_debugfs(); 4901 4902 bond_netlink_fini(); 4903 unregister_pernet_subsys(&bond_net_ops); 4904 4905 #ifdef CONFIG_NET_POLL_CONTROLLER 4906 /* Make sure we don't have an imbalance on our netpoll blocking */ 4907 WARN_ON(atomic_read(&netpoll_block_tx)); 4908 #endif 4909 } 4910 4911 module_init(bonding_init); 4912 module_exit(bonding_exit); 4913 MODULE_LICENSE("GPL"); 4914 MODULE_VERSION(DRV_VERSION); 4915 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 4916 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 4917