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/bonding.h> 81 #include <net/bond_3ad.h> 82 #include <net/bond_alb.h> 83 84 #include "bonding_priv.h" 85 86 /*---------------------------- Module parameters ----------------------------*/ 87 88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 89 90 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 91 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 92 static int num_peer_notif = 1; 93 static int miimon; 94 static int updelay; 95 static int downdelay; 96 static int use_carrier = 1; 97 static char *mode; 98 static char *primary; 99 static char *primary_reselect; 100 static char *lacp_rate; 101 static int min_links; 102 static char *ad_select; 103 static char *xmit_hash_policy; 104 static int arp_interval; 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 106 static char *arp_validate; 107 static char *arp_all_targets; 108 static char *fail_over_mac; 109 static int all_slaves_active; 110 static struct bond_params bonding_defaults; 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 112 static int packets_per_slave = 1; 113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 114 115 module_param(max_bonds, int, 0); 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 117 module_param(tx_queues, int, 0); 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 119 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 121 "failover event (alias of num_unsol_na)"); 122 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 124 "failover event (alias of num_grat_arp)"); 125 module_param(miimon, int, 0); 126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 127 module_param(updelay, int, 0); 128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 129 module_param(downdelay, int, 0); 130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 131 "in milliseconds"); 132 module_param(use_carrier, int, 0); 133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 134 "0 for off, 1 for on (default)"); 135 module_param(mode, charp, 0); 136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 137 "1 for active-backup, 2 for balance-xor, " 138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 139 "6 for balance-alb"); 140 module_param(primary, charp, 0); 141 MODULE_PARM_DESC(primary, "Primary network device to use"); 142 module_param(primary_reselect, charp, 0); 143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 144 "once it comes up; " 145 "0 for always (default), " 146 "1 for only if speed of primary is " 147 "better, " 148 "2 for only on active slave " 149 "failure"); 150 module_param(lacp_rate, charp, 0); 151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 152 "0 for slow, 1 for fast"); 153 module_param(ad_select, charp, 0); 154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 155 "0 for stable (default), 1 for bandwidth, " 156 "2 for count"); 157 module_param(min_links, int, 0); 158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 159 160 module_param(xmit_hash_policy, charp, 0); 161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 162 "0 for layer 2 (default), 1 for layer 3+4, " 163 "2 for layer 2+3, 3 for encap layer 2+3, " 164 "4 for encap layer 3+4"); 165 module_param(arp_interval, int, 0); 166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 167 module_param_array(arp_ip_target, charp, NULL, 0); 168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 169 module_param(arp_validate, charp, 0); 170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 171 "0 for none (default), 1 for active, " 172 "2 for backup, 3 for all"); 173 module_param(arp_all_targets, charp, 0); 174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 175 module_param(fail_over_mac, charp, 0); 176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 177 "the same MAC; 0 for none (default), " 178 "1 for active, 2 for follow"); 179 module_param(all_slaves_active, int, 0); 180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 181 "by setting active flag for all slaves; " 182 "0 for never (default), 1 for always."); 183 module_param(resend_igmp, int, 0); 184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 185 "link failure"); 186 module_param(packets_per_slave, int, 0); 187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 188 "mode; 0 for a random slave, 1 packet per " 189 "slave (default), >1 packets per slave."); 190 module_param(lp_interval, uint, 0); 191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 192 "the bonding driver sends learning packets to " 193 "each slaves peer switch. The default is 1."); 194 195 /*----------------------------- Global variables ----------------------------*/ 196 197 #ifdef CONFIG_NET_POLL_CONTROLLER 198 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 199 #endif 200 201 unsigned int bond_net_id __read_mostly; 202 203 /*-------------------------- Forward declarations ---------------------------*/ 204 205 static int bond_init(struct net_device *bond_dev); 206 static void bond_uninit(struct net_device *bond_dev); 207 static void bond_get_stats(struct net_device *bond_dev, 208 struct rtnl_link_stats64 *stats); 209 static void bond_slave_arr_handler(struct work_struct *work); 210 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 211 int mod); 212 static void bond_netdev_notify_work(struct work_struct *work); 213 214 /*---------------------------- General routines -----------------------------*/ 215 216 const char *bond_mode_name(int mode) 217 { 218 static const char *names[] = { 219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 221 [BOND_MODE_XOR] = "load balancing (xor)", 222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 224 [BOND_MODE_TLB] = "transmit load balancing", 225 [BOND_MODE_ALB] = "adaptive load balancing", 226 }; 227 228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 229 return "unknown"; 230 231 return names[mode]; 232 } 233 234 /*---------------------------------- VLAN -----------------------------------*/ 235 236 /** 237 * bond_dev_queue_xmit - Prepare skb for xmit. 238 * 239 * @bond: bond device that got this skb for tx. 240 * @skb: hw accel VLAN tagged skb to transmit 241 * @slave_dev: slave that is supposed to xmit this skbuff 242 */ 243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 244 struct net_device *slave_dev) 245 { 246 skb->dev = slave_dev; 247 248 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 250 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 251 252 if (unlikely(netpoll_tx_running(bond->dev))) 253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 254 else 255 dev_queue_xmit(skb); 256 } 257 258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 259 * We don't protect the slave list iteration with a lock because: 260 * a. This operation is performed in IOCTL context, 261 * b. The operation is protected by the RTNL semaphore in the 8021q code, 262 * c. Holding a lock with BH disabled while directly calling a base driver 263 * entry point is generally a BAD idea. 264 * 265 * The design of synchronization/protection for this operation in the 8021q 266 * module is good for one or more VLAN devices over a single physical device 267 * and cannot be extended for a teaming solution like bonding, so there is a 268 * potential race condition here where a net device from the vlan group might 269 * be referenced (either by a base driver or the 8021q code) while it is being 270 * removed from the system. However, it turns out we're not making matters 271 * worse, and if it works for regular VLAN usage it will work here too. 272 */ 273 274 /** 275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 276 * @bond_dev: bonding net device that got called 277 * @vid: vlan id being added 278 */ 279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 280 __be16 proto, u16 vid) 281 { 282 struct bonding *bond = netdev_priv(bond_dev); 283 struct slave *slave, *rollback_slave; 284 struct list_head *iter; 285 int res; 286 287 bond_for_each_slave(bond, slave, iter) { 288 res = vlan_vid_add(slave->dev, proto, vid); 289 if (res) 290 goto unwind; 291 } 292 293 return 0; 294 295 unwind: 296 /* unwind to the slave that failed */ 297 bond_for_each_slave(bond, rollback_slave, iter) { 298 if (rollback_slave == slave) 299 break; 300 301 vlan_vid_del(rollback_slave->dev, proto, vid); 302 } 303 304 return res; 305 } 306 307 /** 308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 309 * @bond_dev: bonding net device that got called 310 * @vid: vlan id being removed 311 */ 312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 313 __be16 proto, u16 vid) 314 { 315 struct bonding *bond = netdev_priv(bond_dev); 316 struct list_head *iter; 317 struct slave *slave; 318 319 bond_for_each_slave(bond, slave, iter) 320 vlan_vid_del(slave->dev, proto, vid); 321 322 if (bond_is_lb(bond)) 323 bond_alb_clear_vlan(bond, vid); 324 325 return 0; 326 } 327 328 /*------------------------------- Link status -------------------------------*/ 329 330 /* Set the carrier state for the master according to the state of its 331 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 332 * do special 802.3ad magic. 333 * 334 * Returns zero if carrier state does not change, nonzero if it does. 335 */ 336 int bond_set_carrier(struct bonding *bond) 337 { 338 struct list_head *iter; 339 struct slave *slave; 340 341 if (!bond_has_slaves(bond)) 342 goto down; 343 344 if (BOND_MODE(bond) == BOND_MODE_8023AD) 345 return bond_3ad_set_carrier(bond); 346 347 bond_for_each_slave(bond, slave, iter) { 348 if (slave->link == BOND_LINK_UP) { 349 if (!netif_carrier_ok(bond->dev)) { 350 netif_carrier_on(bond->dev); 351 return 1; 352 } 353 return 0; 354 } 355 } 356 357 down: 358 if (netif_carrier_ok(bond->dev)) { 359 netif_carrier_off(bond->dev); 360 return 1; 361 } 362 return 0; 363 } 364 365 /* Get link speed and duplex from the slave's base driver 366 * using ethtool. If for some reason the call fails or the 367 * values are invalid, set speed and duplex to -1, 368 * and return. Return 1 if speed or duplex settings are 369 * UNKNOWN; 0 otherwise. 370 */ 371 static int bond_update_speed_duplex(struct slave *slave) 372 { 373 struct net_device *slave_dev = slave->dev; 374 struct ethtool_link_ksettings ecmd; 375 int res; 376 377 slave->speed = SPEED_UNKNOWN; 378 slave->duplex = DUPLEX_UNKNOWN; 379 380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 381 if (res < 0) 382 return 1; 383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 384 return 1; 385 switch (ecmd.base.duplex) { 386 case DUPLEX_FULL: 387 case DUPLEX_HALF: 388 break; 389 default: 390 return 1; 391 } 392 393 slave->speed = ecmd.base.speed; 394 slave->duplex = ecmd.base.duplex; 395 396 return 0; 397 } 398 399 const char *bond_slave_link_status(s8 link) 400 { 401 switch (link) { 402 case BOND_LINK_UP: 403 return "up"; 404 case BOND_LINK_FAIL: 405 return "going down"; 406 case BOND_LINK_DOWN: 407 return "down"; 408 case BOND_LINK_BACK: 409 return "going back"; 410 default: 411 return "unknown"; 412 } 413 } 414 415 /* if <dev> supports MII link status reporting, check its link status. 416 * 417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 418 * depending upon the setting of the use_carrier parameter. 419 * 420 * Return either BMSR_LSTATUS, meaning that the link is up (or we 421 * can't tell and just pretend it is), or 0, meaning that the link is 422 * down. 423 * 424 * If reporting is non-zero, instead of faking link up, return -1 if 425 * both ETHTOOL and MII ioctls fail (meaning the device does not 426 * support them). If use_carrier is set, return whatever it says. 427 * It'd be nice if there was a good way to tell if a driver supports 428 * netif_carrier, but there really isn't. 429 */ 430 static int bond_check_dev_link(struct bonding *bond, 431 struct net_device *slave_dev, int reporting) 432 { 433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 434 int (*ioctl)(struct net_device *, struct ifreq *, int); 435 struct ifreq ifr; 436 struct mii_ioctl_data *mii; 437 438 if (!reporting && !netif_running(slave_dev)) 439 return 0; 440 441 if (bond->params.use_carrier) 442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 443 444 /* Try to get link status using Ethtool first. */ 445 if (slave_dev->ethtool_ops->get_link) 446 return slave_dev->ethtool_ops->get_link(slave_dev) ? 447 BMSR_LSTATUS : 0; 448 449 /* Ethtool can't be used, fallback to MII ioctls. */ 450 ioctl = slave_ops->ndo_do_ioctl; 451 if (ioctl) { 452 /* TODO: set pointer to correct ioctl on a per team member 453 * bases to make this more efficient. that is, once 454 * we determine the correct ioctl, we will always 455 * call it and not the others for that team 456 * member. 457 */ 458 459 /* We cannot assume that SIOCGMIIPHY will also read a 460 * register; not all network drivers (e.g., e100) 461 * support that. 462 */ 463 464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 466 mii = if_mii(&ifr); 467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 468 mii->reg_num = MII_BMSR; 469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 470 return mii->val_out & BMSR_LSTATUS; 471 } 472 } 473 474 /* If reporting, report that either there's no dev->do_ioctl, 475 * or both SIOCGMIIREG and get_link failed (meaning that we 476 * cannot report link status). If not reporting, pretend 477 * we're ok. 478 */ 479 return reporting ? -1 : BMSR_LSTATUS; 480 } 481 482 /*----------------------------- Multicast list ------------------------------*/ 483 484 /* Push the promiscuity flag down to appropriate slaves */ 485 static int bond_set_promiscuity(struct bonding *bond, int inc) 486 { 487 struct list_head *iter; 488 int err = 0; 489 490 if (bond_uses_primary(bond)) { 491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 492 493 if (curr_active) 494 err = dev_set_promiscuity(curr_active->dev, inc); 495 } else { 496 struct slave *slave; 497 498 bond_for_each_slave(bond, slave, iter) { 499 err = dev_set_promiscuity(slave->dev, inc); 500 if (err) 501 return err; 502 } 503 } 504 return err; 505 } 506 507 /* Push the allmulti flag down to all slaves */ 508 static int bond_set_allmulti(struct bonding *bond, int inc) 509 { 510 struct list_head *iter; 511 int err = 0; 512 513 if (bond_uses_primary(bond)) { 514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 515 516 if (curr_active) 517 err = dev_set_allmulti(curr_active->dev, inc); 518 } else { 519 struct slave *slave; 520 521 bond_for_each_slave(bond, slave, iter) { 522 err = dev_set_allmulti(slave->dev, inc); 523 if (err) 524 return err; 525 } 526 } 527 return err; 528 } 529 530 /* Retrieve the list of registered multicast addresses for the bonding 531 * device and retransmit an IGMP JOIN request to the current active 532 * slave. 533 */ 534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 535 { 536 struct bonding *bond = container_of(work, struct bonding, 537 mcast_work.work); 538 539 if (!rtnl_trylock()) { 540 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 541 return; 542 } 543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 544 545 if (bond->igmp_retrans > 1) { 546 bond->igmp_retrans--; 547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 548 } 549 rtnl_unlock(); 550 } 551 552 /* Flush bond's hardware addresses from slave */ 553 static void bond_hw_addr_flush(struct net_device *bond_dev, 554 struct net_device *slave_dev) 555 { 556 struct bonding *bond = netdev_priv(bond_dev); 557 558 dev_uc_unsync(slave_dev, bond_dev); 559 dev_mc_unsync(slave_dev, bond_dev); 560 561 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 562 /* del lacpdu mc addr from mc list */ 563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 564 565 dev_mc_del(slave_dev, lacpdu_multicast); 566 } 567 } 568 569 /*--------------------------- Active slave change ---------------------------*/ 570 571 /* Update the hardware address list and promisc/allmulti for the new and 572 * old active slaves (if any). Modes that are not using primary keep all 573 * slaves up date at all times; only the modes that use primary need to call 574 * this function to swap these settings during a failover. 575 */ 576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 577 struct slave *old_active) 578 { 579 if (old_active) { 580 if (bond->dev->flags & IFF_PROMISC) 581 dev_set_promiscuity(old_active->dev, -1); 582 583 if (bond->dev->flags & IFF_ALLMULTI) 584 dev_set_allmulti(old_active->dev, -1); 585 586 bond_hw_addr_flush(bond->dev, old_active->dev); 587 } 588 589 if (new_active) { 590 /* FIXME: Signal errors upstream. */ 591 if (bond->dev->flags & IFF_PROMISC) 592 dev_set_promiscuity(new_active->dev, 1); 593 594 if (bond->dev->flags & IFF_ALLMULTI) 595 dev_set_allmulti(new_active->dev, 1); 596 597 netif_addr_lock_bh(bond->dev); 598 dev_uc_sync(new_active->dev, bond->dev); 599 dev_mc_sync(new_active->dev, bond->dev); 600 netif_addr_unlock_bh(bond->dev); 601 } 602 } 603 604 /** 605 * bond_set_dev_addr - clone slave's address to bond 606 * @bond_dev: bond net device 607 * @slave_dev: slave net device 608 * 609 * Should be called with RTNL held. 610 */ 611 static int bond_set_dev_addr(struct net_device *bond_dev, 612 struct net_device *slave_dev) 613 { 614 int err; 615 616 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n", 617 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len); 618 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 619 if (err) 620 return err; 621 622 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 623 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 624 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 625 return 0; 626 } 627 628 static struct slave *bond_get_old_active(struct bonding *bond, 629 struct slave *new_active) 630 { 631 struct slave *slave; 632 struct list_head *iter; 633 634 bond_for_each_slave(bond, slave, iter) { 635 if (slave == new_active) 636 continue; 637 638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 639 return slave; 640 } 641 642 return NULL; 643 } 644 645 /* bond_do_fail_over_mac 646 * 647 * Perform special MAC address swapping for fail_over_mac settings 648 * 649 * Called with RTNL 650 */ 651 static void bond_do_fail_over_mac(struct bonding *bond, 652 struct slave *new_active, 653 struct slave *old_active) 654 { 655 u8 tmp_mac[MAX_ADDR_LEN]; 656 struct sockaddr_storage ss; 657 int rv; 658 659 switch (bond->params.fail_over_mac) { 660 case BOND_FOM_ACTIVE: 661 if (new_active) { 662 rv = bond_set_dev_addr(bond->dev, new_active->dev); 663 if (rv) 664 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 665 -rv, bond->dev->name); 666 } 667 break; 668 case BOND_FOM_FOLLOW: 669 /* if new_active && old_active, swap them 670 * if just old_active, do nothing (going to no active slave) 671 * if just new_active, set new_active to bond's MAC 672 */ 673 if (!new_active) 674 return; 675 676 if (!old_active) 677 old_active = bond_get_old_active(bond, new_active); 678 679 if (old_active) { 680 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 681 new_active->dev->addr_len); 682 bond_hw_addr_copy(ss.__data, 683 old_active->dev->dev_addr, 684 old_active->dev->addr_len); 685 ss.ss_family = new_active->dev->type; 686 } else { 687 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 688 bond->dev->addr_len); 689 ss.ss_family = bond->dev->type; 690 } 691 692 rv = dev_set_mac_address(new_active->dev, 693 (struct sockaddr *)&ss, NULL); 694 if (rv) { 695 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 696 -rv, new_active->dev->name); 697 goto out; 698 } 699 700 if (!old_active) 701 goto out; 702 703 bond_hw_addr_copy(ss.__data, tmp_mac, 704 new_active->dev->addr_len); 705 ss.ss_family = old_active->dev->type; 706 707 rv = dev_set_mac_address(old_active->dev, 708 (struct sockaddr *)&ss, NULL); 709 if (rv) 710 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n", 711 -rv, new_active->dev->name); 712 out: 713 break; 714 default: 715 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 716 bond->params.fail_over_mac); 717 break; 718 } 719 720 } 721 722 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 723 { 724 struct slave *prim = rtnl_dereference(bond->primary_slave); 725 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 726 727 if (!prim || prim->link != BOND_LINK_UP) { 728 if (!curr || curr->link != BOND_LINK_UP) 729 return NULL; 730 return curr; 731 } 732 733 if (bond->force_primary) { 734 bond->force_primary = false; 735 return prim; 736 } 737 738 if (!curr || curr->link != BOND_LINK_UP) 739 return prim; 740 741 /* At this point, prim and curr are both up */ 742 switch (bond->params.primary_reselect) { 743 case BOND_PRI_RESELECT_ALWAYS: 744 return prim; 745 case BOND_PRI_RESELECT_BETTER: 746 if (prim->speed < curr->speed) 747 return curr; 748 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 749 return curr; 750 return prim; 751 case BOND_PRI_RESELECT_FAILURE: 752 return curr; 753 default: 754 netdev_err(bond->dev, "impossible primary_reselect %d\n", 755 bond->params.primary_reselect); 756 return curr; 757 } 758 } 759 760 /** 761 * bond_find_best_slave - select the best available slave to be the active one 762 * @bond: our bonding struct 763 */ 764 static struct slave *bond_find_best_slave(struct bonding *bond) 765 { 766 struct slave *slave, *bestslave = NULL; 767 struct list_head *iter; 768 int mintime = bond->params.updelay; 769 770 slave = bond_choose_primary_or_current(bond); 771 if (slave) 772 return slave; 773 774 bond_for_each_slave(bond, slave, iter) { 775 if (slave->link == BOND_LINK_UP) 776 return slave; 777 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 778 slave->delay < mintime) { 779 mintime = slave->delay; 780 bestslave = slave; 781 } 782 } 783 784 return bestslave; 785 } 786 787 static bool bond_should_notify_peers(struct bonding *bond) 788 { 789 struct slave *slave; 790 791 rcu_read_lock(); 792 slave = rcu_dereference(bond->curr_active_slave); 793 rcu_read_unlock(); 794 795 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 796 slave ? slave->dev->name : "NULL"); 797 798 if (!slave || !bond->send_peer_notif || 799 !netif_carrier_ok(bond->dev) || 800 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 801 return false; 802 803 return true; 804 } 805 806 /** 807 * change_active_interface - change the active slave into the specified one 808 * @bond: our bonding struct 809 * @new: the new slave to make the active one 810 * 811 * Set the new slave to the bond's settings and unset them on the old 812 * curr_active_slave. 813 * Setting include flags, mc-list, promiscuity, allmulti, etc. 814 * 815 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 816 * because it is apparently the best available slave we have, even though its 817 * updelay hasn't timed out yet. 818 * 819 * Caller must hold RTNL. 820 */ 821 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 822 { 823 struct slave *old_active; 824 825 ASSERT_RTNL(); 826 827 old_active = rtnl_dereference(bond->curr_active_slave); 828 829 if (old_active == new_active) 830 return; 831 832 if (new_active) { 833 new_active->last_link_up = jiffies; 834 835 if (new_active->link == BOND_LINK_BACK) { 836 if (bond_uses_primary(bond)) { 837 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n", 838 new_active->dev->name, 839 (bond->params.updelay - new_active->delay) * bond->params.miimon); 840 } 841 842 new_active->delay = 0; 843 bond_set_slave_link_state(new_active, BOND_LINK_UP, 844 BOND_SLAVE_NOTIFY_NOW); 845 846 if (BOND_MODE(bond) == BOND_MODE_8023AD) 847 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 848 849 if (bond_is_lb(bond)) 850 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 851 } else { 852 if (bond_uses_primary(bond)) { 853 netdev_info(bond->dev, "making interface %s the new active one\n", 854 new_active->dev->name); 855 } 856 } 857 } 858 859 if (bond_uses_primary(bond)) 860 bond_hw_addr_swap(bond, new_active, old_active); 861 862 if (bond_is_lb(bond)) { 863 bond_alb_handle_active_change(bond, new_active); 864 if (old_active) 865 bond_set_slave_inactive_flags(old_active, 866 BOND_SLAVE_NOTIFY_NOW); 867 if (new_active) 868 bond_set_slave_active_flags(new_active, 869 BOND_SLAVE_NOTIFY_NOW); 870 } else { 871 rcu_assign_pointer(bond->curr_active_slave, new_active); 872 } 873 874 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 875 if (old_active) 876 bond_set_slave_inactive_flags(old_active, 877 BOND_SLAVE_NOTIFY_NOW); 878 879 if (new_active) { 880 bool should_notify_peers = false; 881 882 bond_set_slave_active_flags(new_active, 883 BOND_SLAVE_NOTIFY_NOW); 884 885 if (bond->params.fail_over_mac) 886 bond_do_fail_over_mac(bond, new_active, 887 old_active); 888 889 if (netif_running(bond->dev)) { 890 bond->send_peer_notif = 891 bond->params.num_peer_notif; 892 should_notify_peers = 893 bond_should_notify_peers(bond); 894 } 895 896 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 897 if (should_notify_peers) 898 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 899 bond->dev); 900 } 901 } 902 903 /* resend IGMP joins since active slave has changed or 904 * all were sent on curr_active_slave. 905 * resend only if bond is brought up with the affected 906 * bonding modes and the retransmission is enabled 907 */ 908 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 909 ((bond_uses_primary(bond) && new_active) || 910 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 911 bond->igmp_retrans = bond->params.resend_igmp; 912 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 913 } 914 } 915 916 /** 917 * bond_select_active_slave - select a new active slave, if needed 918 * @bond: our bonding struct 919 * 920 * This functions should be called when one of the following occurs: 921 * - The old curr_active_slave has been released or lost its link. 922 * - The primary_slave has got its link back. 923 * - A slave has got its link back and there's no old curr_active_slave. 924 * 925 * Caller must hold RTNL. 926 */ 927 void bond_select_active_slave(struct bonding *bond) 928 { 929 struct slave *best_slave; 930 int rv; 931 932 ASSERT_RTNL(); 933 934 best_slave = bond_find_best_slave(bond); 935 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 936 bond_change_active_slave(bond, best_slave); 937 rv = bond_set_carrier(bond); 938 if (!rv) 939 return; 940 941 if (netif_carrier_ok(bond->dev)) 942 netdev_info(bond->dev, "first active interface up!\n"); 943 else 944 netdev_info(bond->dev, "now running without any active interface!\n"); 945 } 946 } 947 948 #ifdef CONFIG_NET_POLL_CONTROLLER 949 static inline int slave_enable_netpoll(struct slave *slave) 950 { 951 struct netpoll *np; 952 int err = 0; 953 954 np = kzalloc(sizeof(*np), GFP_KERNEL); 955 err = -ENOMEM; 956 if (!np) 957 goto out; 958 959 err = __netpoll_setup(np, slave->dev); 960 if (err) { 961 kfree(np); 962 goto out; 963 } 964 slave->np = np; 965 out: 966 return err; 967 } 968 static inline void slave_disable_netpoll(struct slave *slave) 969 { 970 struct netpoll *np = slave->np; 971 972 if (!np) 973 return; 974 975 slave->np = NULL; 976 977 __netpoll_free(np); 978 } 979 980 static void bond_poll_controller(struct net_device *bond_dev) 981 { 982 struct bonding *bond = netdev_priv(bond_dev); 983 struct slave *slave = NULL; 984 struct list_head *iter; 985 struct ad_info ad_info; 986 987 if (BOND_MODE(bond) == BOND_MODE_8023AD) 988 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 989 return; 990 991 bond_for_each_slave_rcu(bond, slave, iter) { 992 if (!bond_slave_is_up(slave)) 993 continue; 994 995 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 996 struct aggregator *agg = 997 SLAVE_AD_INFO(slave)->port.aggregator; 998 999 if (agg && 1000 agg->aggregator_identifier != ad_info.aggregator_id) 1001 continue; 1002 } 1003 1004 netpoll_poll_dev(slave->dev); 1005 } 1006 } 1007 1008 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1009 { 1010 struct bonding *bond = netdev_priv(bond_dev); 1011 struct list_head *iter; 1012 struct slave *slave; 1013 1014 bond_for_each_slave(bond, slave, iter) 1015 if (bond_slave_is_up(slave)) 1016 slave_disable_netpoll(slave); 1017 } 1018 1019 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1020 { 1021 struct bonding *bond = netdev_priv(dev); 1022 struct list_head *iter; 1023 struct slave *slave; 1024 int err = 0; 1025 1026 bond_for_each_slave(bond, slave, iter) { 1027 err = slave_enable_netpoll(slave); 1028 if (err) { 1029 bond_netpoll_cleanup(dev); 1030 break; 1031 } 1032 } 1033 return err; 1034 } 1035 #else 1036 static inline int slave_enable_netpoll(struct slave *slave) 1037 { 1038 return 0; 1039 } 1040 static inline void slave_disable_netpoll(struct slave *slave) 1041 { 1042 } 1043 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1044 { 1045 } 1046 #endif 1047 1048 /*---------------------------------- IOCTL ----------------------------------*/ 1049 1050 static netdev_features_t bond_fix_features(struct net_device *dev, 1051 netdev_features_t features) 1052 { 1053 struct bonding *bond = netdev_priv(dev); 1054 struct list_head *iter; 1055 netdev_features_t mask; 1056 struct slave *slave; 1057 1058 mask = features; 1059 1060 features &= ~NETIF_F_ONE_FOR_ALL; 1061 features |= NETIF_F_ALL_FOR_ALL; 1062 1063 bond_for_each_slave(bond, slave, iter) { 1064 features = netdev_increment_features(features, 1065 slave->dev->features, 1066 mask); 1067 } 1068 features = netdev_add_tso_features(features, mask); 1069 1070 return features; 1071 } 1072 1073 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1074 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1075 NETIF_F_HIGHDMA | NETIF_F_LRO) 1076 1077 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1078 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1079 1080 static void bond_compute_features(struct bonding *bond) 1081 { 1082 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1083 IFF_XMIT_DST_RELEASE_PERM; 1084 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1085 netdev_features_t enc_features = BOND_ENC_FEATURES; 1086 struct net_device *bond_dev = bond->dev; 1087 struct list_head *iter; 1088 struct slave *slave; 1089 unsigned short max_hard_header_len = ETH_HLEN; 1090 unsigned int gso_max_size = GSO_MAX_SIZE; 1091 u16 gso_max_segs = GSO_MAX_SEGS; 1092 1093 if (!bond_has_slaves(bond)) 1094 goto done; 1095 vlan_features &= NETIF_F_ALL_FOR_ALL; 1096 1097 bond_for_each_slave(bond, slave, iter) { 1098 vlan_features = netdev_increment_features(vlan_features, 1099 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1100 1101 enc_features = netdev_increment_features(enc_features, 1102 slave->dev->hw_enc_features, 1103 BOND_ENC_FEATURES); 1104 dst_release_flag &= slave->dev->priv_flags; 1105 if (slave->dev->hard_header_len > max_hard_header_len) 1106 max_hard_header_len = slave->dev->hard_header_len; 1107 1108 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1109 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1110 } 1111 bond_dev->hard_header_len = max_hard_header_len; 1112 1113 done: 1114 bond_dev->vlan_features = vlan_features; 1115 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1116 NETIF_F_GSO_UDP_L4; 1117 bond_dev->gso_max_segs = gso_max_segs; 1118 netif_set_gso_max_size(bond_dev, gso_max_size); 1119 1120 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1121 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1122 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1123 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1124 1125 netdev_change_features(bond_dev); 1126 } 1127 1128 static void bond_setup_by_slave(struct net_device *bond_dev, 1129 struct net_device *slave_dev) 1130 { 1131 bond_dev->header_ops = slave_dev->header_ops; 1132 1133 bond_dev->type = slave_dev->type; 1134 bond_dev->hard_header_len = slave_dev->hard_header_len; 1135 bond_dev->addr_len = slave_dev->addr_len; 1136 1137 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1138 slave_dev->addr_len); 1139 } 1140 1141 /* On bonding slaves other than the currently active slave, suppress 1142 * duplicates except for alb non-mcast/bcast. 1143 */ 1144 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1145 struct slave *slave, 1146 struct bonding *bond) 1147 { 1148 if (bond_is_slave_inactive(slave)) { 1149 if (BOND_MODE(bond) == BOND_MODE_ALB && 1150 skb->pkt_type != PACKET_BROADCAST && 1151 skb->pkt_type != PACKET_MULTICAST) 1152 return false; 1153 return true; 1154 } 1155 return false; 1156 } 1157 1158 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1159 { 1160 struct sk_buff *skb = *pskb; 1161 struct slave *slave; 1162 struct bonding *bond; 1163 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1164 struct slave *); 1165 int ret = RX_HANDLER_ANOTHER; 1166 1167 skb = skb_share_check(skb, GFP_ATOMIC); 1168 if (unlikely(!skb)) 1169 return RX_HANDLER_CONSUMED; 1170 1171 *pskb = skb; 1172 1173 slave = bond_slave_get_rcu(skb->dev); 1174 bond = slave->bond; 1175 1176 recv_probe = READ_ONCE(bond->recv_probe); 1177 if (recv_probe) { 1178 ret = recv_probe(skb, bond, slave); 1179 if (ret == RX_HANDLER_CONSUMED) { 1180 consume_skb(skb); 1181 return ret; 1182 } 1183 } 1184 1185 /* 1186 * For packets determined by bond_should_deliver_exact_match() call to 1187 * be suppressed we want to make an exception for link-local packets. 1188 * This is necessary for e.g. LLDP daemons to be able to monitor 1189 * inactive slave links without being forced to bind to them 1190 * explicitly. 1191 * 1192 * At the same time, packets that are passed to the bonding master 1193 * (including link-local ones) can have their originating interface 1194 * determined via PACKET_ORIGDEV socket option. 1195 */ 1196 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1197 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1198 return RX_HANDLER_PASS; 1199 return RX_HANDLER_EXACT; 1200 } 1201 1202 skb->dev = bond->dev; 1203 1204 if (BOND_MODE(bond) == BOND_MODE_ALB && 1205 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1206 skb->pkt_type == PACKET_HOST) { 1207 1208 if (unlikely(skb_cow_head(skb, 1209 skb->data - skb_mac_header(skb)))) { 1210 kfree_skb(skb); 1211 return RX_HANDLER_CONSUMED; 1212 } 1213 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1214 bond->dev->addr_len); 1215 } 1216 1217 return ret; 1218 } 1219 1220 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1221 { 1222 switch (BOND_MODE(bond)) { 1223 case BOND_MODE_ROUNDROBIN: 1224 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1225 case BOND_MODE_ACTIVEBACKUP: 1226 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1227 case BOND_MODE_BROADCAST: 1228 return NETDEV_LAG_TX_TYPE_BROADCAST; 1229 case BOND_MODE_XOR: 1230 case BOND_MODE_8023AD: 1231 return NETDEV_LAG_TX_TYPE_HASH; 1232 default: 1233 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1234 } 1235 } 1236 1237 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1238 enum netdev_lag_tx_type type) 1239 { 1240 if (type != NETDEV_LAG_TX_TYPE_HASH) 1241 return NETDEV_LAG_HASH_NONE; 1242 1243 switch (bond->params.xmit_policy) { 1244 case BOND_XMIT_POLICY_LAYER2: 1245 return NETDEV_LAG_HASH_L2; 1246 case BOND_XMIT_POLICY_LAYER34: 1247 return NETDEV_LAG_HASH_L34; 1248 case BOND_XMIT_POLICY_LAYER23: 1249 return NETDEV_LAG_HASH_L23; 1250 case BOND_XMIT_POLICY_ENCAP23: 1251 return NETDEV_LAG_HASH_E23; 1252 case BOND_XMIT_POLICY_ENCAP34: 1253 return NETDEV_LAG_HASH_E34; 1254 default: 1255 return NETDEV_LAG_HASH_UNKNOWN; 1256 } 1257 } 1258 1259 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1260 struct netlink_ext_ack *extack) 1261 { 1262 struct netdev_lag_upper_info lag_upper_info; 1263 enum netdev_lag_tx_type type; 1264 1265 type = bond_lag_tx_type(bond); 1266 lag_upper_info.tx_type = type; 1267 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1268 1269 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1270 &lag_upper_info, extack); 1271 } 1272 1273 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1274 { 1275 netdev_upper_dev_unlink(slave->dev, bond->dev); 1276 slave->dev->flags &= ~IFF_SLAVE; 1277 } 1278 1279 static struct slave *bond_alloc_slave(struct bonding *bond) 1280 { 1281 struct slave *slave = NULL; 1282 1283 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1284 if (!slave) 1285 return NULL; 1286 1287 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1288 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1289 GFP_KERNEL); 1290 if (!SLAVE_AD_INFO(slave)) { 1291 kfree(slave); 1292 return NULL; 1293 } 1294 } 1295 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1296 1297 return slave; 1298 } 1299 1300 static void bond_free_slave(struct slave *slave) 1301 { 1302 struct bonding *bond = bond_get_bond_by_slave(slave); 1303 1304 cancel_delayed_work_sync(&slave->notify_work); 1305 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1306 kfree(SLAVE_AD_INFO(slave)); 1307 1308 kfree(slave); 1309 } 1310 1311 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1312 { 1313 info->bond_mode = BOND_MODE(bond); 1314 info->miimon = bond->params.miimon; 1315 info->num_slaves = bond->slave_cnt; 1316 } 1317 1318 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1319 { 1320 strcpy(info->slave_name, slave->dev->name); 1321 info->link = slave->link; 1322 info->state = bond_slave_state(slave); 1323 info->link_failure_count = slave->link_failure_count; 1324 } 1325 1326 static void bond_netdev_notify_work(struct work_struct *_work) 1327 { 1328 struct slave *slave = container_of(_work, struct slave, 1329 notify_work.work); 1330 1331 if (rtnl_trylock()) { 1332 struct netdev_bonding_info binfo; 1333 1334 bond_fill_ifslave(slave, &binfo.slave); 1335 bond_fill_ifbond(slave->bond, &binfo.master); 1336 netdev_bonding_info_change(slave->dev, &binfo); 1337 rtnl_unlock(); 1338 } else { 1339 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1340 } 1341 } 1342 1343 void bond_queue_slave_event(struct slave *slave) 1344 { 1345 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1346 } 1347 1348 void bond_lower_state_changed(struct slave *slave) 1349 { 1350 struct netdev_lag_lower_state_info info; 1351 1352 info.link_up = slave->link == BOND_LINK_UP || 1353 slave->link == BOND_LINK_FAIL; 1354 info.tx_enabled = bond_is_active_slave(slave); 1355 netdev_lower_state_changed(slave->dev, &info); 1356 } 1357 1358 /* enslave device <slave> to bond device <master> */ 1359 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1360 struct netlink_ext_ack *extack) 1361 { 1362 struct bonding *bond = netdev_priv(bond_dev); 1363 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1364 struct slave *new_slave = NULL, *prev_slave; 1365 struct sockaddr_storage ss; 1366 int link_reporting; 1367 int res = 0, i; 1368 1369 if (!bond->params.use_carrier && 1370 slave_dev->ethtool_ops->get_link == NULL && 1371 slave_ops->ndo_do_ioctl == NULL) { 1372 netdev_warn(bond_dev, "no link monitoring support for %s\n", 1373 slave_dev->name); 1374 } 1375 1376 /* already in-use? */ 1377 if (netdev_is_rx_handler_busy(slave_dev)) { 1378 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1379 netdev_err(bond_dev, 1380 "Error: Device is in use and cannot be enslaved\n"); 1381 return -EBUSY; 1382 } 1383 1384 if (bond_dev == slave_dev) { 1385 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1386 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1387 return -EPERM; 1388 } 1389 1390 /* vlan challenged mutual exclusion */ 1391 /* no need to lock since we're protected by rtnl_lock */ 1392 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1393 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n", 1394 slave_dev->name); 1395 if (vlan_uses_dev(bond_dev)) { 1396 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1397 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n", 1398 slave_dev->name, bond_dev->name); 1399 return -EPERM; 1400 } else { 1401 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n", 1402 slave_dev->name, slave_dev->name, 1403 bond_dev->name); 1404 } 1405 } else { 1406 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n", 1407 slave_dev->name); 1408 } 1409 1410 /* Old ifenslave binaries are no longer supported. These can 1411 * be identified with moderate accuracy by the state of the slave: 1412 * the current ifenslave will set the interface down prior to 1413 * enslaving it; the old ifenslave will not. 1414 */ 1415 if (slave_dev->flags & IFF_UP) { 1416 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1417 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n", 1418 slave_dev->name); 1419 return -EPERM; 1420 } 1421 1422 /* set bonding device ether type by slave - bonding netdevices are 1423 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1424 * there is a need to override some of the type dependent attribs/funcs. 1425 * 1426 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1427 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1428 */ 1429 if (!bond_has_slaves(bond)) { 1430 if (bond_dev->type != slave_dev->type) { 1431 netdev_dbg(bond_dev, "change device type from %d to %d\n", 1432 bond_dev->type, slave_dev->type); 1433 1434 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1435 bond_dev); 1436 res = notifier_to_errno(res); 1437 if (res) { 1438 netdev_err(bond_dev, "refused to change device type\n"); 1439 return -EBUSY; 1440 } 1441 1442 /* Flush unicast and multicast addresses */ 1443 dev_uc_flush(bond_dev); 1444 dev_mc_flush(bond_dev); 1445 1446 if (slave_dev->type != ARPHRD_ETHER) 1447 bond_setup_by_slave(bond_dev, slave_dev); 1448 else { 1449 ether_setup(bond_dev); 1450 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1451 } 1452 1453 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1454 bond_dev); 1455 } 1456 } else if (bond_dev->type != slave_dev->type) { 1457 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1458 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n", 1459 slave_dev->name, slave_dev->type, bond_dev->type); 1460 return -EINVAL; 1461 } 1462 1463 if (slave_dev->type == ARPHRD_INFINIBAND && 1464 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1465 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1466 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n", 1467 slave_dev->type); 1468 res = -EOPNOTSUPP; 1469 goto err_undo_flags; 1470 } 1471 1472 if (!slave_ops->ndo_set_mac_address || 1473 slave_dev->type == ARPHRD_INFINIBAND) { 1474 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n"); 1475 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1476 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1477 if (!bond_has_slaves(bond)) { 1478 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1479 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1480 } else { 1481 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1482 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"); 1483 res = -EOPNOTSUPP; 1484 goto err_undo_flags; 1485 } 1486 } 1487 } 1488 1489 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1490 1491 /* If this is the first slave, then we need to set the master's hardware 1492 * address to be the same as the slave's. 1493 */ 1494 if (!bond_has_slaves(bond) && 1495 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1496 res = bond_set_dev_addr(bond->dev, slave_dev); 1497 if (res) 1498 goto err_undo_flags; 1499 } 1500 1501 new_slave = bond_alloc_slave(bond); 1502 if (!new_slave) { 1503 res = -ENOMEM; 1504 goto err_undo_flags; 1505 } 1506 1507 new_slave->bond = bond; 1508 new_slave->dev = slave_dev; 1509 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1510 * is set via sysfs or module option if desired. 1511 */ 1512 new_slave->queue_id = 0; 1513 1514 /* Save slave's original mtu and then set it to match the bond */ 1515 new_slave->original_mtu = slave_dev->mtu; 1516 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1517 if (res) { 1518 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res); 1519 goto err_free; 1520 } 1521 1522 /* Save slave's original ("permanent") mac address for modes 1523 * that need it, and for restoring it upon release, and then 1524 * set it to the master's address 1525 */ 1526 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1527 slave_dev->addr_len); 1528 1529 if (!bond->params.fail_over_mac || 1530 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1531 /* Set slave to master's mac address. The application already 1532 * set the master's mac address to that of the first slave 1533 */ 1534 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1535 ss.ss_family = slave_dev->type; 1536 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1537 extack); 1538 if (res) { 1539 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res); 1540 goto err_restore_mtu; 1541 } 1542 } 1543 1544 /* set slave flag before open to prevent IPv6 addrconf */ 1545 slave_dev->flags |= IFF_SLAVE; 1546 1547 /* open the slave since the application closed it */ 1548 res = dev_open(slave_dev, extack); 1549 if (res) { 1550 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name); 1551 goto err_restore_mac; 1552 } 1553 1554 slave_dev->priv_flags |= IFF_BONDING; 1555 /* initialize slave stats */ 1556 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1557 1558 if (bond_is_lb(bond)) { 1559 /* bond_alb_init_slave() must be called before all other stages since 1560 * it might fail and we do not want to have to undo everything 1561 */ 1562 res = bond_alb_init_slave(bond, new_slave); 1563 if (res) 1564 goto err_close; 1565 } 1566 1567 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1568 if (res) { 1569 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n", 1570 slave_dev->name); 1571 goto err_close; 1572 } 1573 1574 prev_slave = bond_last_slave(bond); 1575 1576 new_slave->delay = 0; 1577 new_slave->link_failure_count = 0; 1578 1579 if (bond_update_speed_duplex(new_slave) && 1580 bond_needs_speed_duplex(bond)) 1581 new_slave->link = BOND_LINK_DOWN; 1582 1583 new_slave->last_rx = jiffies - 1584 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1585 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1586 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1587 1588 if (bond->params.miimon && !bond->params.use_carrier) { 1589 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1590 1591 if ((link_reporting == -1) && !bond->params.arp_interval) { 1592 /* miimon is set but a bonded network driver 1593 * does not support ETHTOOL/MII and 1594 * arp_interval is not set. Note: if 1595 * use_carrier is enabled, we will never go 1596 * here (because netif_carrier is always 1597 * supported); thus, we don't need to change 1598 * the messages for netif_carrier. 1599 */ 1600 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", 1601 slave_dev->name); 1602 } else if (link_reporting == -1) { 1603 /* unable get link status using mii/ethtool */ 1604 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", 1605 slave_dev->name); 1606 } 1607 } 1608 1609 /* check for initial state */ 1610 new_slave->link = BOND_LINK_NOCHANGE; 1611 if (bond->params.miimon) { 1612 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1613 if (bond->params.updelay) { 1614 bond_set_slave_link_state(new_slave, 1615 BOND_LINK_BACK, 1616 BOND_SLAVE_NOTIFY_NOW); 1617 new_slave->delay = bond->params.updelay; 1618 } else { 1619 bond_set_slave_link_state(new_slave, 1620 BOND_LINK_UP, 1621 BOND_SLAVE_NOTIFY_NOW); 1622 } 1623 } else { 1624 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1625 BOND_SLAVE_NOTIFY_NOW); 1626 } 1627 } else if (bond->params.arp_interval) { 1628 bond_set_slave_link_state(new_slave, 1629 (netif_carrier_ok(slave_dev) ? 1630 BOND_LINK_UP : BOND_LINK_DOWN), 1631 BOND_SLAVE_NOTIFY_NOW); 1632 } else { 1633 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1634 BOND_SLAVE_NOTIFY_NOW); 1635 } 1636 1637 if (new_slave->link != BOND_LINK_DOWN) 1638 new_slave->last_link_up = jiffies; 1639 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n", 1640 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1641 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1642 1643 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1644 /* if there is a primary slave, remember it */ 1645 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1646 rcu_assign_pointer(bond->primary_slave, new_slave); 1647 bond->force_primary = true; 1648 } 1649 } 1650 1651 switch (BOND_MODE(bond)) { 1652 case BOND_MODE_ACTIVEBACKUP: 1653 bond_set_slave_inactive_flags(new_slave, 1654 BOND_SLAVE_NOTIFY_NOW); 1655 break; 1656 case BOND_MODE_8023AD: 1657 /* in 802.3ad mode, the internal mechanism 1658 * will activate the slaves in the selected 1659 * aggregator 1660 */ 1661 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1662 /* if this is the first slave */ 1663 if (!prev_slave) { 1664 SLAVE_AD_INFO(new_slave)->id = 1; 1665 /* Initialize AD with the number of times that the AD timer is called in 1 second 1666 * can be called only after the mac address of the bond is set 1667 */ 1668 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1669 } else { 1670 SLAVE_AD_INFO(new_slave)->id = 1671 SLAVE_AD_INFO(prev_slave)->id + 1; 1672 } 1673 1674 bond_3ad_bind_slave(new_slave); 1675 break; 1676 case BOND_MODE_TLB: 1677 case BOND_MODE_ALB: 1678 bond_set_active_slave(new_slave); 1679 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1680 break; 1681 default: 1682 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n"); 1683 1684 /* always active in trunk mode */ 1685 bond_set_active_slave(new_slave); 1686 1687 /* In trunking mode there is little meaning to curr_active_slave 1688 * anyway (it holds no special properties of the bond device), 1689 * so we can change it without calling change_active_interface() 1690 */ 1691 if (!rcu_access_pointer(bond->curr_active_slave) && 1692 new_slave->link == BOND_LINK_UP) 1693 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1694 1695 break; 1696 } /* switch(bond_mode) */ 1697 1698 #ifdef CONFIG_NET_POLL_CONTROLLER 1699 if (bond->dev->npinfo) { 1700 if (slave_enable_netpoll(new_slave)) { 1701 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1702 res = -EBUSY; 1703 goto err_detach; 1704 } 1705 } 1706 #endif 1707 1708 if (!(bond_dev->features & NETIF_F_LRO)) 1709 dev_disable_lro(slave_dev); 1710 1711 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1712 new_slave); 1713 if (res) { 1714 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res); 1715 goto err_detach; 1716 } 1717 1718 res = bond_master_upper_dev_link(bond, new_slave, extack); 1719 if (res) { 1720 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1721 goto err_unregister; 1722 } 1723 1724 res = bond_sysfs_slave_add(new_slave); 1725 if (res) { 1726 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1727 goto err_upper_unlink; 1728 } 1729 1730 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1731 1732 /* If the mode uses primary, then the following is handled by 1733 * bond_change_active_slave(). 1734 */ 1735 if (!bond_uses_primary(bond)) { 1736 /* set promiscuity level to new slave */ 1737 if (bond_dev->flags & IFF_PROMISC) { 1738 res = dev_set_promiscuity(slave_dev, 1); 1739 if (res) 1740 goto err_sysfs_del; 1741 } 1742 1743 /* set allmulti level to new slave */ 1744 if (bond_dev->flags & IFF_ALLMULTI) { 1745 res = dev_set_allmulti(slave_dev, 1); 1746 if (res) { 1747 if (bond_dev->flags & IFF_PROMISC) 1748 dev_set_promiscuity(slave_dev, -1); 1749 goto err_sysfs_del; 1750 } 1751 } 1752 1753 netif_addr_lock_bh(bond_dev); 1754 dev_mc_sync_multiple(slave_dev, bond_dev); 1755 dev_uc_sync_multiple(slave_dev, bond_dev); 1756 netif_addr_unlock_bh(bond_dev); 1757 1758 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1759 /* add lacpdu mc addr to mc list */ 1760 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1761 1762 dev_mc_add(slave_dev, lacpdu_multicast); 1763 } 1764 } 1765 1766 bond->slave_cnt++; 1767 bond_compute_features(bond); 1768 bond_set_carrier(bond); 1769 1770 if (bond_uses_primary(bond)) { 1771 block_netpoll_tx(); 1772 bond_select_active_slave(bond); 1773 unblock_netpoll_tx(); 1774 } 1775 1776 if (bond_mode_can_use_xmit_hash(bond)) 1777 bond_update_slave_arr(bond, NULL); 1778 1779 1780 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n", 1781 slave_dev->name, 1782 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1783 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1784 1785 /* enslave is successful */ 1786 bond_queue_slave_event(new_slave); 1787 return 0; 1788 1789 /* Undo stages on error */ 1790 err_sysfs_del: 1791 bond_sysfs_slave_del(new_slave); 1792 1793 err_upper_unlink: 1794 bond_upper_dev_unlink(bond, new_slave); 1795 1796 err_unregister: 1797 netdev_rx_handler_unregister(slave_dev); 1798 1799 err_detach: 1800 vlan_vids_del_by_dev(slave_dev, bond_dev); 1801 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1802 RCU_INIT_POINTER(bond->primary_slave, NULL); 1803 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1804 block_netpoll_tx(); 1805 bond_change_active_slave(bond, NULL); 1806 bond_select_active_slave(bond); 1807 unblock_netpoll_tx(); 1808 } 1809 /* either primary_slave or curr_active_slave might've changed */ 1810 synchronize_rcu(); 1811 slave_disable_netpoll(new_slave); 1812 1813 err_close: 1814 slave_dev->priv_flags &= ~IFF_BONDING; 1815 dev_close(slave_dev); 1816 1817 err_restore_mac: 1818 slave_dev->flags &= ~IFF_SLAVE; 1819 if (!bond->params.fail_over_mac || 1820 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1821 /* XXX TODO - fom follow mode needs to change master's 1822 * MAC if this slave's MAC is in use by the bond, or at 1823 * least print a warning. 1824 */ 1825 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 1826 new_slave->dev->addr_len); 1827 ss.ss_family = slave_dev->type; 1828 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 1829 } 1830 1831 err_restore_mtu: 1832 dev_set_mtu(slave_dev, new_slave->original_mtu); 1833 1834 err_free: 1835 bond_free_slave(new_slave); 1836 1837 err_undo_flags: 1838 /* Enslave of first slave has failed and we need to fix master's mac */ 1839 if (!bond_has_slaves(bond)) { 1840 if (ether_addr_equal_64bits(bond_dev->dev_addr, 1841 slave_dev->dev_addr)) 1842 eth_hw_addr_random(bond_dev); 1843 if (bond_dev->type != ARPHRD_ETHER) { 1844 dev_close(bond_dev); 1845 ether_setup(bond_dev); 1846 bond_dev->flags |= IFF_MASTER; 1847 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1848 } 1849 } 1850 1851 return res; 1852 } 1853 1854 /* Try to release the slave device <slave> from the bond device <master> 1855 * It is legal to access curr_active_slave without a lock because all the function 1856 * is RTNL-locked. If "all" is true it means that the function is being called 1857 * while destroying a bond interface and all slaves are being released. 1858 * 1859 * The rules for slave state should be: 1860 * for Active/Backup: 1861 * Active stays on all backups go down 1862 * for Bonded connections: 1863 * The first up interface should be left on and all others downed. 1864 */ 1865 static int __bond_release_one(struct net_device *bond_dev, 1866 struct net_device *slave_dev, 1867 bool all, bool unregister) 1868 { 1869 struct bonding *bond = netdev_priv(bond_dev); 1870 struct slave *slave, *oldcurrent; 1871 struct sockaddr_storage ss; 1872 int old_flags = bond_dev->flags; 1873 netdev_features_t old_features = bond_dev->features; 1874 1875 /* slave is not a slave or master is not master of this slave */ 1876 if (!(slave_dev->flags & IFF_SLAVE) || 1877 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1878 netdev_dbg(bond_dev, "cannot release %s\n", 1879 slave_dev->name); 1880 return -EINVAL; 1881 } 1882 1883 block_netpoll_tx(); 1884 1885 slave = bond_get_slave_by_dev(bond, slave_dev); 1886 if (!slave) { 1887 /* not a slave of this bond */ 1888 netdev_info(bond_dev, "%s not enslaved\n", 1889 slave_dev->name); 1890 unblock_netpoll_tx(); 1891 return -EINVAL; 1892 } 1893 1894 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 1895 1896 bond_sysfs_slave_del(slave); 1897 1898 /* recompute stats just before removing the slave */ 1899 bond_get_stats(bond->dev, &bond->bond_stats); 1900 1901 bond_upper_dev_unlink(bond, slave); 1902 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1903 * for this slave anymore. 1904 */ 1905 netdev_rx_handler_unregister(slave_dev); 1906 1907 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1908 bond_3ad_unbind_slave(slave); 1909 1910 if (bond_mode_can_use_xmit_hash(bond)) 1911 bond_update_slave_arr(bond, slave); 1912 1913 netdev_info(bond_dev, "Releasing %s interface %s\n", 1914 bond_is_active_slave(slave) ? "active" : "backup", 1915 slave_dev->name); 1916 1917 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1918 1919 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1920 1921 if (!all && (!bond->params.fail_over_mac || 1922 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1923 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1924 bond_has_slaves(bond)) 1925 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", 1926 slave_dev->name, slave->perm_hwaddr, 1927 bond_dev->name, slave_dev->name); 1928 } 1929 1930 if (rtnl_dereference(bond->primary_slave) == slave) 1931 RCU_INIT_POINTER(bond->primary_slave, NULL); 1932 1933 if (oldcurrent == slave) 1934 bond_change_active_slave(bond, NULL); 1935 1936 if (bond_is_lb(bond)) { 1937 /* Must be called only after the slave has been 1938 * detached from the list and the curr_active_slave 1939 * has been cleared (if our_slave == old_current), 1940 * but before a new active slave is selected. 1941 */ 1942 bond_alb_deinit_slave(bond, slave); 1943 } 1944 1945 if (all) { 1946 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1947 } else if (oldcurrent == slave) { 1948 /* Note that we hold RTNL over this sequence, so there 1949 * is no concern that another slave add/remove event 1950 * will interfere. 1951 */ 1952 bond_select_active_slave(bond); 1953 } 1954 1955 if (!bond_has_slaves(bond)) { 1956 bond_set_carrier(bond); 1957 eth_hw_addr_random(bond_dev); 1958 bond->nest_level = SINGLE_DEPTH_NESTING; 1959 } else { 1960 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1961 } 1962 1963 unblock_netpoll_tx(); 1964 synchronize_rcu(); 1965 bond->slave_cnt--; 1966 1967 if (!bond_has_slaves(bond)) { 1968 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 1969 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 1970 } 1971 1972 bond_compute_features(bond); 1973 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 1974 (old_features & NETIF_F_VLAN_CHALLENGED)) 1975 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n", 1976 slave_dev->name, bond_dev->name); 1977 1978 vlan_vids_del_by_dev(slave_dev, bond_dev); 1979 1980 /* If the mode uses primary, then this case was handled above by 1981 * bond_change_active_slave(..., NULL) 1982 */ 1983 if (!bond_uses_primary(bond)) { 1984 /* unset promiscuity level from slave 1985 * NOTE: The NETDEV_CHANGEADDR call above may change the value 1986 * of the IFF_PROMISC flag in the bond_dev, but we need the 1987 * value of that flag before that change, as that was the value 1988 * when this slave was attached, so we cache at the start of the 1989 * function and use it here. Same goes for ALLMULTI below 1990 */ 1991 if (old_flags & IFF_PROMISC) 1992 dev_set_promiscuity(slave_dev, -1); 1993 1994 /* unset allmulti level from slave */ 1995 if (old_flags & IFF_ALLMULTI) 1996 dev_set_allmulti(slave_dev, -1); 1997 1998 bond_hw_addr_flush(bond_dev, slave_dev); 1999 } 2000 2001 slave_disable_netpoll(slave); 2002 2003 /* close slave before restoring its mac address */ 2004 dev_close(slave_dev); 2005 2006 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2007 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2008 /* restore original ("permanent") mac address */ 2009 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2010 slave->dev->addr_len); 2011 ss.ss_family = slave_dev->type; 2012 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2013 } 2014 2015 if (unregister) 2016 __dev_set_mtu(slave_dev, slave->original_mtu); 2017 else 2018 dev_set_mtu(slave_dev, slave->original_mtu); 2019 2020 slave_dev->priv_flags &= ~IFF_BONDING; 2021 2022 bond_free_slave(slave); 2023 2024 return 0; 2025 } 2026 2027 /* A wrapper used because of ndo_del_link */ 2028 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2029 { 2030 return __bond_release_one(bond_dev, slave_dev, false, false); 2031 } 2032 2033 /* First release a slave and then destroy the bond if no more slaves are left. 2034 * Must be under rtnl_lock when this function is called. 2035 */ 2036 static int bond_release_and_destroy(struct net_device *bond_dev, 2037 struct net_device *slave_dev) 2038 { 2039 struct bonding *bond = netdev_priv(bond_dev); 2040 int ret; 2041 2042 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2043 if (ret == 0 && !bond_has_slaves(bond)) { 2044 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2045 netdev_info(bond_dev, "Destroying bond %s\n", 2046 bond_dev->name); 2047 bond_remove_proc_entry(bond); 2048 unregister_netdevice(bond_dev); 2049 } 2050 return ret; 2051 } 2052 2053 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2054 { 2055 struct bonding *bond = netdev_priv(bond_dev); 2056 bond_fill_ifbond(bond, info); 2057 } 2058 2059 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2060 { 2061 struct bonding *bond = netdev_priv(bond_dev); 2062 struct list_head *iter; 2063 int i = 0, res = -ENODEV; 2064 struct slave *slave; 2065 2066 bond_for_each_slave(bond, slave, iter) { 2067 if (i++ == (int)info->slave_id) { 2068 res = 0; 2069 bond_fill_ifslave(slave, info); 2070 break; 2071 } 2072 } 2073 2074 return res; 2075 } 2076 2077 /*-------------------------------- Monitoring -------------------------------*/ 2078 2079 /* called with rcu_read_lock() */ 2080 static int bond_miimon_inspect(struct bonding *bond) 2081 { 2082 int link_state, commit = 0; 2083 struct list_head *iter; 2084 struct slave *slave; 2085 bool ignore_updelay; 2086 2087 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2088 2089 bond_for_each_slave_rcu(bond, slave, iter) { 2090 slave->new_link = BOND_LINK_NOCHANGE; 2091 slave->link_new_state = slave->link; 2092 2093 link_state = bond_check_dev_link(bond, slave->dev, 0); 2094 2095 switch (slave->link) { 2096 case BOND_LINK_UP: 2097 if (link_state) 2098 continue; 2099 2100 bond_propose_link_state(slave, BOND_LINK_FAIL); 2101 commit++; 2102 slave->delay = bond->params.downdelay; 2103 if (slave->delay) { 2104 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n", 2105 (BOND_MODE(bond) == 2106 BOND_MODE_ACTIVEBACKUP) ? 2107 (bond_is_active_slave(slave) ? 2108 "active " : "backup ") : "", 2109 slave->dev->name, 2110 bond->params.downdelay * bond->params.miimon); 2111 } 2112 /*FALLTHRU*/ 2113 case BOND_LINK_FAIL: 2114 if (link_state) { 2115 /* recovered before downdelay expired */ 2116 bond_propose_link_state(slave, BOND_LINK_UP); 2117 slave->last_link_up = jiffies; 2118 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n", 2119 (bond->params.downdelay - slave->delay) * 2120 bond->params.miimon, 2121 slave->dev->name); 2122 commit++; 2123 continue; 2124 } 2125 2126 if (slave->delay <= 0) { 2127 slave->new_link = BOND_LINK_DOWN; 2128 commit++; 2129 continue; 2130 } 2131 2132 slave->delay--; 2133 break; 2134 2135 case BOND_LINK_DOWN: 2136 if (!link_state) 2137 continue; 2138 2139 bond_propose_link_state(slave, BOND_LINK_BACK); 2140 commit++; 2141 slave->delay = bond->params.updelay; 2142 2143 if (slave->delay) { 2144 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n", 2145 slave->dev->name, 2146 ignore_updelay ? 0 : 2147 bond->params.updelay * 2148 bond->params.miimon); 2149 } 2150 /*FALLTHRU*/ 2151 case BOND_LINK_BACK: 2152 if (!link_state) { 2153 bond_propose_link_state(slave, BOND_LINK_DOWN); 2154 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n", 2155 (bond->params.updelay - slave->delay) * 2156 bond->params.miimon, 2157 slave->dev->name); 2158 commit++; 2159 continue; 2160 } 2161 2162 if (ignore_updelay) 2163 slave->delay = 0; 2164 2165 if (slave->delay <= 0) { 2166 slave->new_link = BOND_LINK_UP; 2167 commit++; 2168 ignore_updelay = false; 2169 continue; 2170 } 2171 2172 slave->delay--; 2173 break; 2174 } 2175 } 2176 2177 return commit; 2178 } 2179 2180 static void bond_miimon_link_change(struct bonding *bond, 2181 struct slave *slave, 2182 char link) 2183 { 2184 switch (BOND_MODE(bond)) { 2185 case BOND_MODE_8023AD: 2186 bond_3ad_handle_link_change(slave, link); 2187 break; 2188 case BOND_MODE_TLB: 2189 case BOND_MODE_ALB: 2190 bond_alb_handle_link_change(bond, slave, link); 2191 break; 2192 case BOND_MODE_XOR: 2193 bond_update_slave_arr(bond, NULL); 2194 break; 2195 } 2196 } 2197 2198 static void bond_miimon_commit(struct bonding *bond) 2199 { 2200 struct list_head *iter; 2201 struct slave *slave, *primary; 2202 2203 bond_for_each_slave(bond, slave, iter) { 2204 switch (slave->new_link) { 2205 case BOND_LINK_NOCHANGE: 2206 continue; 2207 2208 case BOND_LINK_UP: 2209 if (bond_update_speed_duplex(slave) && 2210 bond_needs_speed_duplex(bond)) { 2211 slave->link = BOND_LINK_DOWN; 2212 if (net_ratelimit()) 2213 netdev_warn(bond->dev, 2214 "failed to get link speed/duplex for %s\n", 2215 slave->dev->name); 2216 continue; 2217 } 2218 bond_set_slave_link_state(slave, BOND_LINK_UP, 2219 BOND_SLAVE_NOTIFY_NOW); 2220 slave->last_link_up = jiffies; 2221 2222 primary = rtnl_dereference(bond->primary_slave); 2223 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2224 /* prevent it from being the active one */ 2225 bond_set_backup_slave(slave); 2226 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2227 /* make it immediately active */ 2228 bond_set_active_slave(slave); 2229 } else if (slave != primary) { 2230 /* prevent it from being the active one */ 2231 bond_set_backup_slave(slave); 2232 } 2233 2234 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n", 2235 slave->dev->name, 2236 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2237 slave->duplex ? "full" : "half"); 2238 2239 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2240 2241 if (!bond->curr_active_slave || slave == primary) 2242 goto do_failover; 2243 2244 continue; 2245 2246 case BOND_LINK_DOWN: 2247 if (slave->link_failure_count < UINT_MAX) 2248 slave->link_failure_count++; 2249 2250 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2251 BOND_SLAVE_NOTIFY_NOW); 2252 2253 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2254 BOND_MODE(bond) == BOND_MODE_8023AD) 2255 bond_set_slave_inactive_flags(slave, 2256 BOND_SLAVE_NOTIFY_NOW); 2257 2258 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2259 slave->dev->name); 2260 2261 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2262 2263 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2264 goto do_failover; 2265 2266 continue; 2267 2268 default: 2269 netdev_err(bond->dev, "invalid new link %d on slave %s\n", 2270 slave->new_link, slave->dev->name); 2271 slave->new_link = BOND_LINK_NOCHANGE; 2272 2273 continue; 2274 } 2275 2276 do_failover: 2277 block_netpoll_tx(); 2278 bond_select_active_slave(bond); 2279 unblock_netpoll_tx(); 2280 } 2281 2282 bond_set_carrier(bond); 2283 } 2284 2285 /* bond_mii_monitor 2286 * 2287 * Really a wrapper that splits the mii monitor into two phases: an 2288 * inspection, then (if inspection indicates something needs to be done) 2289 * an acquisition of appropriate locks followed by a commit phase to 2290 * implement whatever link state changes are indicated. 2291 */ 2292 static void bond_mii_monitor(struct work_struct *work) 2293 { 2294 struct bonding *bond = container_of(work, struct bonding, 2295 mii_work.work); 2296 bool should_notify_peers = false; 2297 unsigned long delay; 2298 struct slave *slave; 2299 struct list_head *iter; 2300 2301 delay = msecs_to_jiffies(bond->params.miimon); 2302 2303 if (!bond_has_slaves(bond)) 2304 goto re_arm; 2305 2306 rcu_read_lock(); 2307 2308 should_notify_peers = bond_should_notify_peers(bond); 2309 2310 if (bond_miimon_inspect(bond)) { 2311 rcu_read_unlock(); 2312 2313 /* Race avoidance with bond_close cancel of workqueue */ 2314 if (!rtnl_trylock()) { 2315 delay = 1; 2316 should_notify_peers = false; 2317 goto re_arm; 2318 } 2319 2320 bond_for_each_slave(bond, slave, iter) { 2321 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2322 } 2323 bond_miimon_commit(bond); 2324 2325 rtnl_unlock(); /* might sleep, hold no other locks */ 2326 } else 2327 rcu_read_unlock(); 2328 2329 re_arm: 2330 if (bond->params.miimon) 2331 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2332 2333 if (should_notify_peers) { 2334 if (!rtnl_trylock()) 2335 return; 2336 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2337 rtnl_unlock(); 2338 } 2339 } 2340 2341 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2342 { 2343 __be32 ip = *((__be32 *)data); 2344 2345 return ip == bond_confirm_addr(upper, 0, ip); 2346 } 2347 2348 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2349 { 2350 bool ret = false; 2351 2352 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2353 return true; 2354 2355 rcu_read_lock(); 2356 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2357 ret = true; 2358 rcu_read_unlock(); 2359 2360 return ret; 2361 } 2362 2363 /* We go to the (large) trouble of VLAN tagging ARP frames because 2364 * switches in VLAN mode (especially if ports are configured as 2365 * "native" to a VLAN) might not pass non-tagged frames. 2366 */ 2367 static void bond_arp_send(struct net_device *slave_dev, int arp_op, 2368 __be32 dest_ip, __be32 src_ip, 2369 struct bond_vlan_tag *tags) 2370 { 2371 struct sk_buff *skb; 2372 struct bond_vlan_tag *outer_tag = tags; 2373 2374 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n", 2375 arp_op, slave_dev->name, &dest_ip, &src_ip); 2376 2377 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2378 NULL, slave_dev->dev_addr, NULL); 2379 2380 if (!skb) { 2381 net_err_ratelimited("ARP packet allocation failed\n"); 2382 return; 2383 } 2384 2385 if (!tags || tags->vlan_proto == VLAN_N_VID) 2386 goto xmit; 2387 2388 tags++; 2389 2390 /* Go through all the tags backwards and add them to the packet */ 2391 while (tags->vlan_proto != VLAN_N_VID) { 2392 if (!tags->vlan_id) { 2393 tags++; 2394 continue; 2395 } 2396 2397 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n", 2398 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2399 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2400 tags->vlan_id); 2401 if (!skb) { 2402 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2403 return; 2404 } 2405 2406 tags++; 2407 } 2408 /* Set the outer tag */ 2409 if (outer_tag->vlan_id) { 2410 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n", 2411 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2412 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2413 outer_tag->vlan_id); 2414 } 2415 2416 xmit: 2417 arp_xmit(skb); 2418 } 2419 2420 /* Validate the device path between the @start_dev and the @end_dev. 2421 * The path is valid if the @end_dev is reachable through device 2422 * stacking. 2423 * When the path is validated, collect any vlan information in the 2424 * path. 2425 */ 2426 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2427 struct net_device *end_dev, 2428 int level) 2429 { 2430 struct bond_vlan_tag *tags; 2431 struct net_device *upper; 2432 struct list_head *iter; 2433 2434 if (start_dev == end_dev) { 2435 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2436 if (!tags) 2437 return ERR_PTR(-ENOMEM); 2438 tags[level].vlan_proto = VLAN_N_VID; 2439 return tags; 2440 } 2441 2442 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2443 tags = bond_verify_device_path(upper, end_dev, level + 1); 2444 if (IS_ERR_OR_NULL(tags)) { 2445 if (IS_ERR(tags)) 2446 return tags; 2447 continue; 2448 } 2449 if (is_vlan_dev(upper)) { 2450 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2451 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2452 } 2453 2454 return tags; 2455 } 2456 2457 return NULL; 2458 } 2459 2460 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2461 { 2462 struct rtable *rt; 2463 struct bond_vlan_tag *tags; 2464 __be32 *targets = bond->params.arp_targets, addr; 2465 int i; 2466 2467 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2468 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]); 2469 tags = NULL; 2470 2471 /* Find out through which dev should the packet go */ 2472 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2473 RTO_ONLINK, 0); 2474 if (IS_ERR(rt)) { 2475 /* there's no route to target - try to send arp 2476 * probe to generate any traffic (arp_validate=0) 2477 */ 2478 if (bond->params.arp_validate) 2479 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2480 bond->dev->name, 2481 &targets[i]); 2482 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2483 0, tags); 2484 continue; 2485 } 2486 2487 /* bond device itself */ 2488 if (rt->dst.dev == bond->dev) 2489 goto found; 2490 2491 rcu_read_lock(); 2492 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2493 rcu_read_unlock(); 2494 2495 if (!IS_ERR_OR_NULL(tags)) 2496 goto found; 2497 2498 /* Not our device - skip */ 2499 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2500 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2501 2502 ip_rt_put(rt); 2503 continue; 2504 2505 found: 2506 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2507 ip_rt_put(rt); 2508 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2509 addr, tags); 2510 kfree(tags); 2511 } 2512 } 2513 2514 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2515 { 2516 int i; 2517 2518 if (!sip || !bond_has_this_ip(bond, tip)) { 2519 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n", 2520 &sip, &tip); 2521 return; 2522 } 2523 2524 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2525 if (i == -1) { 2526 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n", 2527 &sip); 2528 return; 2529 } 2530 slave->last_rx = jiffies; 2531 slave->target_last_arp_rx[i] = jiffies; 2532 } 2533 2534 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2535 struct slave *slave) 2536 { 2537 struct arphdr *arp = (struct arphdr *)skb->data; 2538 struct slave *curr_active_slave, *curr_arp_slave; 2539 unsigned char *arp_ptr; 2540 __be32 sip, tip; 2541 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2542 unsigned int alen; 2543 2544 if (!slave_do_arp_validate(bond, slave)) { 2545 if ((slave_do_arp_validate_only(bond) && is_arp) || 2546 !slave_do_arp_validate_only(bond)) 2547 slave->last_rx = jiffies; 2548 return RX_HANDLER_ANOTHER; 2549 } else if (!is_arp) { 2550 return RX_HANDLER_ANOTHER; 2551 } 2552 2553 alen = arp_hdr_len(bond->dev); 2554 2555 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n", 2556 skb->dev->name); 2557 2558 if (alen > skb_headlen(skb)) { 2559 arp = kmalloc(alen, GFP_ATOMIC); 2560 if (!arp) 2561 goto out_unlock; 2562 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2563 goto out_unlock; 2564 } 2565 2566 if (arp->ar_hln != bond->dev->addr_len || 2567 skb->pkt_type == PACKET_OTHERHOST || 2568 skb->pkt_type == PACKET_LOOPBACK || 2569 arp->ar_hrd != htons(ARPHRD_ETHER) || 2570 arp->ar_pro != htons(ETH_P_IP) || 2571 arp->ar_pln != 4) 2572 goto out_unlock; 2573 2574 arp_ptr = (unsigned char *)(arp + 1); 2575 arp_ptr += bond->dev->addr_len; 2576 memcpy(&sip, arp_ptr, 4); 2577 arp_ptr += 4 + bond->dev->addr_len; 2578 memcpy(&tip, arp_ptr, 4); 2579 2580 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2581 slave->dev->name, bond_slave_state(slave), 2582 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2583 &sip, &tip); 2584 2585 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2586 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2587 2588 /* We 'trust' the received ARP enough to validate it if: 2589 * 2590 * (a) the slave receiving the ARP is active (which includes the 2591 * current ARP slave, if any), or 2592 * 2593 * (b) the receiving slave isn't active, but there is a currently 2594 * active slave and it received valid arp reply(s) after it became 2595 * the currently active slave, or 2596 * 2597 * (c) there is an ARP slave that sent an ARP during the prior ARP 2598 * interval, and we receive an ARP reply on any slave. We accept 2599 * these because switch FDB update delays may deliver the ARP 2600 * reply to a slave other than the sender of the ARP request. 2601 * 2602 * Note: for (b), backup slaves are receiving the broadcast ARP 2603 * request, not a reply. This request passes from the sending 2604 * slave through the L2 switch(es) to the receiving slave. Since 2605 * this is checking the request, sip/tip are swapped for 2606 * validation. 2607 * 2608 * This is done to avoid endless looping when we can't reach the 2609 * arp_ip_target and fool ourselves with our own arp requests. 2610 */ 2611 if (bond_is_active_slave(slave)) 2612 bond_validate_arp(bond, slave, sip, tip); 2613 else if (curr_active_slave && 2614 time_after(slave_last_rx(bond, curr_active_slave), 2615 curr_active_slave->last_link_up)) 2616 bond_validate_arp(bond, slave, tip, sip); 2617 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2618 bond_time_in_interval(bond, 2619 dev_trans_start(curr_arp_slave->dev), 1)) 2620 bond_validate_arp(bond, slave, sip, tip); 2621 2622 out_unlock: 2623 if (arp != (struct arphdr *)skb->data) 2624 kfree(arp); 2625 return RX_HANDLER_ANOTHER; 2626 } 2627 2628 /* function to verify if we're in the arp_interval timeslice, returns true if 2629 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2630 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2631 */ 2632 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2633 int mod) 2634 { 2635 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2636 2637 return time_in_range(jiffies, 2638 last_act - delta_in_ticks, 2639 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2640 } 2641 2642 /* This function is called regularly to monitor each slave's link 2643 * ensuring that traffic is being sent and received when arp monitoring 2644 * is used in load-balancing mode. if the adapter has been dormant, then an 2645 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2646 * arp monitoring in active backup mode. 2647 */ 2648 static void bond_loadbalance_arp_mon(struct bonding *bond) 2649 { 2650 struct slave *slave, *oldcurrent; 2651 struct list_head *iter; 2652 int do_failover = 0, slave_state_changed = 0; 2653 2654 if (!bond_has_slaves(bond)) 2655 goto re_arm; 2656 2657 rcu_read_lock(); 2658 2659 oldcurrent = rcu_dereference(bond->curr_active_slave); 2660 /* see if any of the previous devices are up now (i.e. they have 2661 * xmt and rcv traffic). the curr_active_slave does not come into 2662 * the picture unless it is null. also, slave->last_link_up is not 2663 * needed here because we send an arp on each slave and give a slave 2664 * as long as it needs to get the tx/rx within the delta. 2665 * TODO: what about up/down delay in arp mode? it wasn't here before 2666 * so it can wait 2667 */ 2668 bond_for_each_slave_rcu(bond, slave, iter) { 2669 unsigned long trans_start = dev_trans_start(slave->dev); 2670 2671 slave->new_link = BOND_LINK_NOCHANGE; 2672 2673 if (slave->link != BOND_LINK_UP) { 2674 if (bond_time_in_interval(bond, trans_start, 1) && 2675 bond_time_in_interval(bond, slave->last_rx, 1)) { 2676 2677 slave->new_link = BOND_LINK_UP; 2678 slave_state_changed = 1; 2679 2680 /* primary_slave has no meaning in round-robin 2681 * mode. the window of a slave being up and 2682 * curr_active_slave being null after enslaving 2683 * is closed. 2684 */ 2685 if (!oldcurrent) { 2686 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2687 slave->dev->name); 2688 do_failover = 1; 2689 } else { 2690 netdev_info(bond->dev, "interface %s is now up\n", 2691 slave->dev->name); 2692 } 2693 } 2694 } else { 2695 /* slave->link == BOND_LINK_UP */ 2696 2697 /* not all switches will respond to an arp request 2698 * when the source ip is 0, so don't take the link down 2699 * if we don't know our ip yet 2700 */ 2701 if (!bond_time_in_interval(bond, trans_start, 2) || 2702 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2703 2704 slave->new_link = BOND_LINK_DOWN; 2705 slave_state_changed = 1; 2706 2707 if (slave->link_failure_count < UINT_MAX) 2708 slave->link_failure_count++; 2709 2710 netdev_info(bond->dev, "interface %s is now down\n", 2711 slave->dev->name); 2712 2713 if (slave == oldcurrent) 2714 do_failover = 1; 2715 } 2716 } 2717 2718 /* note: if switch is in round-robin mode, all links 2719 * must tx arp to ensure all links rx an arp - otherwise 2720 * links may oscillate or not come up at all; if switch is 2721 * in something like xor mode, there is nothing we can 2722 * do - all replies will be rx'ed on same link causing slaves 2723 * to be unstable during low/no traffic periods 2724 */ 2725 if (bond_slave_is_up(slave)) 2726 bond_arp_send_all(bond, slave); 2727 } 2728 2729 rcu_read_unlock(); 2730 2731 if (do_failover || slave_state_changed) { 2732 if (!rtnl_trylock()) 2733 goto re_arm; 2734 2735 bond_for_each_slave(bond, slave, iter) { 2736 if (slave->new_link != BOND_LINK_NOCHANGE) 2737 slave->link = slave->new_link; 2738 } 2739 2740 if (slave_state_changed) { 2741 bond_slave_state_change(bond); 2742 if (BOND_MODE(bond) == BOND_MODE_XOR) 2743 bond_update_slave_arr(bond, NULL); 2744 } 2745 if (do_failover) { 2746 block_netpoll_tx(); 2747 bond_select_active_slave(bond); 2748 unblock_netpoll_tx(); 2749 } 2750 rtnl_unlock(); 2751 } 2752 2753 re_arm: 2754 if (bond->params.arp_interval) 2755 queue_delayed_work(bond->wq, &bond->arp_work, 2756 msecs_to_jiffies(bond->params.arp_interval)); 2757 } 2758 2759 /* Called to inspect slaves for active-backup mode ARP monitor link state 2760 * changes. Sets new_link in slaves to specify what action should take 2761 * place for the slave. Returns 0 if no changes are found, >0 if changes 2762 * to link states must be committed. 2763 * 2764 * Called with rcu_read_lock held. 2765 */ 2766 static int bond_ab_arp_inspect(struct bonding *bond) 2767 { 2768 unsigned long trans_start, last_rx; 2769 struct list_head *iter; 2770 struct slave *slave; 2771 int commit = 0; 2772 2773 bond_for_each_slave_rcu(bond, slave, iter) { 2774 slave->new_link = BOND_LINK_NOCHANGE; 2775 last_rx = slave_last_rx(bond, slave); 2776 2777 if (slave->link != BOND_LINK_UP) { 2778 if (bond_time_in_interval(bond, last_rx, 1)) { 2779 slave->new_link = BOND_LINK_UP; 2780 commit++; 2781 } 2782 continue; 2783 } 2784 2785 /* Give slaves 2*delta after being enslaved or made 2786 * active. This avoids bouncing, as the last receive 2787 * times need a full ARP monitor cycle to be updated. 2788 */ 2789 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2790 continue; 2791 2792 /* Backup slave is down if: 2793 * - No current_arp_slave AND 2794 * - more than 3*delta since last receive AND 2795 * - the bond has an IP address 2796 * 2797 * Note: a non-null current_arp_slave indicates 2798 * the curr_active_slave went down and we are 2799 * searching for a new one; under this condition 2800 * we only take the curr_active_slave down - this 2801 * gives each slave a chance to tx/rx traffic 2802 * before being taken out 2803 */ 2804 if (!bond_is_active_slave(slave) && 2805 !rcu_access_pointer(bond->current_arp_slave) && 2806 !bond_time_in_interval(bond, last_rx, 3)) { 2807 slave->new_link = BOND_LINK_DOWN; 2808 commit++; 2809 } 2810 2811 /* Active slave is down if: 2812 * - more than 2*delta since transmitting OR 2813 * - (more than 2*delta since receive AND 2814 * the bond has an IP address) 2815 */ 2816 trans_start = dev_trans_start(slave->dev); 2817 if (bond_is_active_slave(slave) && 2818 (!bond_time_in_interval(bond, trans_start, 2) || 2819 !bond_time_in_interval(bond, last_rx, 2))) { 2820 slave->new_link = BOND_LINK_DOWN; 2821 commit++; 2822 } 2823 } 2824 2825 return commit; 2826 } 2827 2828 /* Called to commit link state changes noted by inspection step of 2829 * active-backup mode ARP monitor. 2830 * 2831 * Called with RTNL hold. 2832 */ 2833 static void bond_ab_arp_commit(struct bonding *bond) 2834 { 2835 unsigned long trans_start; 2836 struct list_head *iter; 2837 struct slave *slave; 2838 2839 bond_for_each_slave(bond, slave, iter) { 2840 switch (slave->new_link) { 2841 case BOND_LINK_NOCHANGE: 2842 continue; 2843 2844 case BOND_LINK_UP: 2845 trans_start = dev_trans_start(slave->dev); 2846 if (rtnl_dereference(bond->curr_active_slave) != slave || 2847 (!rtnl_dereference(bond->curr_active_slave) && 2848 bond_time_in_interval(bond, trans_start, 1))) { 2849 struct slave *current_arp_slave; 2850 2851 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2852 bond_set_slave_link_state(slave, BOND_LINK_UP, 2853 BOND_SLAVE_NOTIFY_NOW); 2854 if (current_arp_slave) { 2855 bond_set_slave_inactive_flags( 2856 current_arp_slave, 2857 BOND_SLAVE_NOTIFY_NOW); 2858 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2859 } 2860 2861 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2862 slave->dev->name); 2863 2864 if (!rtnl_dereference(bond->curr_active_slave) || 2865 slave == rtnl_dereference(bond->primary_slave)) 2866 goto do_failover; 2867 2868 } 2869 2870 continue; 2871 2872 case BOND_LINK_DOWN: 2873 if (slave->link_failure_count < UINT_MAX) 2874 slave->link_failure_count++; 2875 2876 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2877 BOND_SLAVE_NOTIFY_NOW); 2878 bond_set_slave_inactive_flags(slave, 2879 BOND_SLAVE_NOTIFY_NOW); 2880 2881 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2882 slave->dev->name); 2883 2884 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2885 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2886 goto do_failover; 2887 } 2888 2889 continue; 2890 2891 default: 2892 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n", 2893 slave->new_link, slave->dev->name); 2894 continue; 2895 } 2896 2897 do_failover: 2898 block_netpoll_tx(); 2899 bond_select_active_slave(bond); 2900 unblock_netpoll_tx(); 2901 } 2902 2903 bond_set_carrier(bond); 2904 } 2905 2906 /* Send ARP probes for active-backup mode ARP monitor. 2907 * 2908 * Called with rcu_read_lock held. 2909 */ 2910 static bool bond_ab_arp_probe(struct bonding *bond) 2911 { 2912 struct slave *slave, *before = NULL, *new_slave = NULL, 2913 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2914 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2915 struct list_head *iter; 2916 bool found = false; 2917 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2918 2919 if (curr_arp_slave && curr_active_slave) 2920 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2921 curr_arp_slave->dev->name, 2922 curr_active_slave->dev->name); 2923 2924 if (curr_active_slave) { 2925 bond_arp_send_all(bond, curr_active_slave); 2926 return should_notify_rtnl; 2927 } 2928 2929 /* if we don't have a curr_active_slave, search for the next available 2930 * backup slave from the current_arp_slave and make it the candidate 2931 * for becoming the curr_active_slave 2932 */ 2933 2934 if (!curr_arp_slave) { 2935 curr_arp_slave = bond_first_slave_rcu(bond); 2936 if (!curr_arp_slave) 2937 return should_notify_rtnl; 2938 } 2939 2940 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2941 2942 bond_for_each_slave_rcu(bond, slave, iter) { 2943 if (!found && !before && bond_slave_is_up(slave)) 2944 before = slave; 2945 2946 if (found && !new_slave && bond_slave_is_up(slave)) 2947 new_slave = slave; 2948 /* if the link state is up at this point, we 2949 * mark it down - this can happen if we have 2950 * simultaneous link failures and 2951 * reselect_active_interface doesn't make this 2952 * one the current slave so it is still marked 2953 * up when it is actually down 2954 */ 2955 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 2956 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2957 BOND_SLAVE_NOTIFY_LATER); 2958 if (slave->link_failure_count < UINT_MAX) 2959 slave->link_failure_count++; 2960 2961 bond_set_slave_inactive_flags(slave, 2962 BOND_SLAVE_NOTIFY_LATER); 2963 2964 netdev_info(bond->dev, "backup interface %s is now down\n", 2965 slave->dev->name); 2966 } 2967 if (slave == curr_arp_slave) 2968 found = true; 2969 } 2970 2971 if (!new_slave && before) 2972 new_slave = before; 2973 2974 if (!new_slave) 2975 goto check_state; 2976 2977 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 2978 BOND_SLAVE_NOTIFY_LATER); 2979 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 2980 bond_arp_send_all(bond, new_slave); 2981 new_slave->last_link_up = jiffies; 2982 rcu_assign_pointer(bond->current_arp_slave, new_slave); 2983 2984 check_state: 2985 bond_for_each_slave_rcu(bond, slave, iter) { 2986 if (slave->should_notify || slave->should_notify_link) { 2987 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 2988 break; 2989 } 2990 } 2991 return should_notify_rtnl; 2992 } 2993 2994 static void bond_activebackup_arp_mon(struct bonding *bond) 2995 { 2996 bool should_notify_peers = false; 2997 bool should_notify_rtnl = false; 2998 int delta_in_ticks; 2999 3000 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3001 3002 if (!bond_has_slaves(bond)) 3003 goto re_arm; 3004 3005 rcu_read_lock(); 3006 3007 should_notify_peers = bond_should_notify_peers(bond); 3008 3009 if (bond_ab_arp_inspect(bond)) { 3010 rcu_read_unlock(); 3011 3012 /* Race avoidance with bond_close flush of workqueue */ 3013 if (!rtnl_trylock()) { 3014 delta_in_ticks = 1; 3015 should_notify_peers = false; 3016 goto re_arm; 3017 } 3018 3019 bond_ab_arp_commit(bond); 3020 3021 rtnl_unlock(); 3022 rcu_read_lock(); 3023 } 3024 3025 should_notify_rtnl = bond_ab_arp_probe(bond); 3026 rcu_read_unlock(); 3027 3028 re_arm: 3029 if (bond->params.arp_interval) 3030 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3031 3032 if (should_notify_peers || should_notify_rtnl) { 3033 if (!rtnl_trylock()) 3034 return; 3035 3036 if (should_notify_peers) 3037 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3038 bond->dev); 3039 if (should_notify_rtnl) { 3040 bond_slave_state_notify(bond); 3041 bond_slave_link_notify(bond); 3042 } 3043 3044 rtnl_unlock(); 3045 } 3046 } 3047 3048 static void bond_arp_monitor(struct work_struct *work) 3049 { 3050 struct bonding *bond = container_of(work, struct bonding, 3051 arp_work.work); 3052 3053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3054 bond_activebackup_arp_mon(bond); 3055 else 3056 bond_loadbalance_arp_mon(bond); 3057 } 3058 3059 /*-------------------------- netdev event handling --------------------------*/ 3060 3061 /* Change device name */ 3062 static int bond_event_changename(struct bonding *bond) 3063 { 3064 bond_remove_proc_entry(bond); 3065 bond_create_proc_entry(bond); 3066 3067 bond_debug_reregister(bond); 3068 3069 return NOTIFY_DONE; 3070 } 3071 3072 static int bond_master_netdev_event(unsigned long event, 3073 struct net_device *bond_dev) 3074 { 3075 struct bonding *event_bond = netdev_priv(bond_dev); 3076 3077 switch (event) { 3078 case NETDEV_CHANGENAME: 3079 return bond_event_changename(event_bond); 3080 case NETDEV_UNREGISTER: 3081 bond_remove_proc_entry(event_bond); 3082 break; 3083 case NETDEV_REGISTER: 3084 bond_create_proc_entry(event_bond); 3085 break; 3086 case NETDEV_NOTIFY_PEERS: 3087 if (event_bond->send_peer_notif) 3088 event_bond->send_peer_notif--; 3089 break; 3090 default: 3091 break; 3092 } 3093 3094 return NOTIFY_DONE; 3095 } 3096 3097 static int bond_slave_netdev_event(unsigned long event, 3098 struct net_device *slave_dev) 3099 { 3100 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3101 struct bonding *bond; 3102 struct net_device *bond_dev; 3103 3104 /* A netdev event can be generated while enslaving a device 3105 * before netdev_rx_handler_register is called in which case 3106 * slave will be NULL 3107 */ 3108 if (!slave) 3109 return NOTIFY_DONE; 3110 bond_dev = slave->bond->dev; 3111 bond = slave->bond; 3112 primary = rtnl_dereference(bond->primary_slave); 3113 3114 switch (event) { 3115 case NETDEV_UNREGISTER: 3116 if (bond_dev->type != ARPHRD_ETHER) 3117 bond_release_and_destroy(bond_dev, slave_dev); 3118 else 3119 __bond_release_one(bond_dev, slave_dev, false, true); 3120 break; 3121 case NETDEV_UP: 3122 case NETDEV_CHANGE: 3123 /* For 802.3ad mode only: 3124 * Getting invalid Speed/Duplex values here will put slave 3125 * in weird state. So mark it as link-fail for the time 3126 * being and let link-monitoring (miimon) set it right when 3127 * correct speeds/duplex are available. 3128 */ 3129 if (bond_update_speed_duplex(slave) && 3130 BOND_MODE(bond) == BOND_MODE_8023AD) 3131 slave->link = BOND_LINK_FAIL; 3132 3133 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3134 bond_3ad_adapter_speed_duplex_changed(slave); 3135 /* Fallthrough */ 3136 case NETDEV_DOWN: 3137 /* Refresh slave-array if applicable! 3138 * If the setup does not use miimon or arpmon (mode-specific!), 3139 * then these events will not cause the slave-array to be 3140 * refreshed. This will cause xmit to use a slave that is not 3141 * usable. Avoid such situation by refeshing the array at these 3142 * events. If these (miimon/arpmon) parameters are configured 3143 * then array gets refreshed twice and that should be fine! 3144 */ 3145 if (bond_mode_can_use_xmit_hash(bond)) 3146 bond_update_slave_arr(bond, NULL); 3147 break; 3148 case NETDEV_CHANGEMTU: 3149 /* TODO: Should slaves be allowed to 3150 * independently alter their MTU? For 3151 * an active-backup bond, slaves need 3152 * not be the same type of device, so 3153 * MTUs may vary. For other modes, 3154 * slaves arguably should have the 3155 * same MTUs. To do this, we'd need to 3156 * take over the slave's change_mtu 3157 * function for the duration of their 3158 * servitude. 3159 */ 3160 break; 3161 case NETDEV_CHANGENAME: 3162 /* we don't care if we don't have primary set */ 3163 if (!bond_uses_primary(bond) || 3164 !bond->params.primary[0]) 3165 break; 3166 3167 if (slave == primary) { 3168 /* slave's name changed - he's no longer primary */ 3169 RCU_INIT_POINTER(bond->primary_slave, NULL); 3170 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3171 /* we have a new primary slave */ 3172 rcu_assign_pointer(bond->primary_slave, slave); 3173 } else { /* we didn't change primary - exit */ 3174 break; 3175 } 3176 3177 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3178 primary ? slave_dev->name : "none"); 3179 3180 block_netpoll_tx(); 3181 bond_select_active_slave(bond); 3182 unblock_netpoll_tx(); 3183 break; 3184 case NETDEV_FEAT_CHANGE: 3185 bond_compute_features(bond); 3186 break; 3187 case NETDEV_RESEND_IGMP: 3188 /* Propagate to master device */ 3189 call_netdevice_notifiers(event, slave->bond->dev); 3190 break; 3191 default: 3192 break; 3193 } 3194 3195 return NOTIFY_DONE; 3196 } 3197 3198 /* bond_netdev_event: handle netdev notifier chain events. 3199 * 3200 * This function receives events for the netdev chain. The caller (an 3201 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3202 * locks for us to safely manipulate the slave devices (RTNL lock, 3203 * dev_probe_lock). 3204 */ 3205 static int bond_netdev_event(struct notifier_block *this, 3206 unsigned long event, void *ptr) 3207 { 3208 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3209 3210 netdev_dbg(event_dev, "event: %lx\n", event); 3211 3212 if (!(event_dev->priv_flags & IFF_BONDING)) 3213 return NOTIFY_DONE; 3214 3215 if (event_dev->flags & IFF_MASTER) { 3216 int ret; 3217 3218 netdev_dbg(event_dev, "IFF_MASTER\n"); 3219 ret = bond_master_netdev_event(event, event_dev); 3220 if (ret != NOTIFY_DONE) 3221 return ret; 3222 } 3223 3224 if (event_dev->flags & IFF_SLAVE) { 3225 netdev_dbg(event_dev, "IFF_SLAVE\n"); 3226 return bond_slave_netdev_event(event, event_dev); 3227 } 3228 3229 return NOTIFY_DONE; 3230 } 3231 3232 static struct notifier_block bond_netdev_notifier = { 3233 .notifier_call = bond_netdev_event, 3234 }; 3235 3236 /*---------------------------- Hashing Policies -----------------------------*/ 3237 3238 /* L2 hash helper */ 3239 static inline u32 bond_eth_hash(struct sk_buff *skb) 3240 { 3241 struct ethhdr *ep, hdr_tmp; 3242 3243 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3244 if (ep) 3245 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3246 return 0; 3247 } 3248 3249 /* Extract the appropriate headers based on bond's xmit policy */ 3250 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3251 struct flow_keys *fk) 3252 { 3253 const struct ipv6hdr *iph6; 3254 const struct iphdr *iph; 3255 int noff, proto = -1; 3256 3257 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) 3258 return skb_flow_dissect_flow_keys(skb, fk, 0); 3259 3260 fk->ports.ports = 0; 3261 noff = skb_network_offset(skb); 3262 if (skb->protocol == htons(ETH_P_IP)) { 3263 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 3264 return false; 3265 iph = ip_hdr(skb); 3266 iph_to_flow_copy_v4addrs(fk, iph); 3267 noff += iph->ihl << 2; 3268 if (!ip_is_fragment(iph)) 3269 proto = iph->protocol; 3270 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3271 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6)))) 3272 return false; 3273 iph6 = ipv6_hdr(skb); 3274 iph_to_flow_copy_v6addrs(fk, iph6); 3275 noff += sizeof(*iph6); 3276 proto = iph6->nexthdr; 3277 } else { 3278 return false; 3279 } 3280 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) 3281 fk->ports.ports = skb_flow_get_ports(skb, noff, proto); 3282 3283 return true; 3284 } 3285 3286 /** 3287 * bond_xmit_hash - generate a hash value based on the xmit policy 3288 * @bond: bonding device 3289 * @skb: buffer to use for headers 3290 * 3291 * This function will extract the necessary headers from the skb buffer and use 3292 * them to generate a hash based on the xmit_policy set in the bonding device 3293 */ 3294 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3295 { 3296 struct flow_keys flow; 3297 u32 hash; 3298 3299 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3300 skb->l4_hash) 3301 return skb->hash; 3302 3303 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3304 !bond_flow_dissect(bond, skb, &flow)) 3305 return bond_eth_hash(skb); 3306 3307 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3308 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) 3309 hash = bond_eth_hash(skb); 3310 else 3311 hash = (__force u32)flow.ports.ports; 3312 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3313 (__force u32)flow_get_u32_src(&flow); 3314 hash ^= (hash >> 16); 3315 hash ^= (hash >> 8); 3316 3317 return hash >> 1; 3318 } 3319 3320 /*-------------------------- Device entry points ----------------------------*/ 3321 3322 void bond_work_init_all(struct bonding *bond) 3323 { 3324 INIT_DELAYED_WORK(&bond->mcast_work, 3325 bond_resend_igmp_join_requests_delayed); 3326 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3327 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3328 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3329 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3330 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3331 } 3332 3333 static void bond_work_cancel_all(struct bonding *bond) 3334 { 3335 cancel_delayed_work_sync(&bond->mii_work); 3336 cancel_delayed_work_sync(&bond->arp_work); 3337 cancel_delayed_work_sync(&bond->alb_work); 3338 cancel_delayed_work_sync(&bond->ad_work); 3339 cancel_delayed_work_sync(&bond->mcast_work); 3340 cancel_delayed_work_sync(&bond->slave_arr_work); 3341 } 3342 3343 static int bond_open(struct net_device *bond_dev) 3344 { 3345 struct bonding *bond = netdev_priv(bond_dev); 3346 struct list_head *iter; 3347 struct slave *slave; 3348 3349 /* reset slave->backup and slave->inactive */ 3350 if (bond_has_slaves(bond)) { 3351 bond_for_each_slave(bond, slave, iter) { 3352 if (bond_uses_primary(bond) && 3353 slave != rcu_access_pointer(bond->curr_active_slave)) { 3354 bond_set_slave_inactive_flags(slave, 3355 BOND_SLAVE_NOTIFY_NOW); 3356 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3357 bond_set_slave_active_flags(slave, 3358 BOND_SLAVE_NOTIFY_NOW); 3359 } 3360 } 3361 } 3362 3363 if (bond_is_lb(bond)) { 3364 /* bond_alb_initialize must be called before the timer 3365 * is started. 3366 */ 3367 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3368 return -ENOMEM; 3369 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3370 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3371 } 3372 3373 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3374 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3375 3376 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3377 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3378 bond->recv_probe = bond_arp_rcv; 3379 } 3380 3381 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3382 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3383 /* register to receive LACPDUs */ 3384 bond->recv_probe = bond_3ad_lacpdu_recv; 3385 bond_3ad_initiate_agg_selection(bond, 1); 3386 } 3387 3388 if (bond_mode_can_use_xmit_hash(bond)) 3389 bond_update_slave_arr(bond, NULL); 3390 3391 return 0; 3392 } 3393 3394 static int bond_close(struct net_device *bond_dev) 3395 { 3396 struct bonding *bond = netdev_priv(bond_dev); 3397 3398 bond_work_cancel_all(bond); 3399 bond->send_peer_notif = 0; 3400 if (bond_is_lb(bond)) 3401 bond_alb_deinitialize(bond); 3402 bond->recv_probe = NULL; 3403 3404 return 0; 3405 } 3406 3407 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3408 * that some drivers can provide 32bit values only. 3409 */ 3410 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3411 const struct rtnl_link_stats64 *_new, 3412 const struct rtnl_link_stats64 *_old) 3413 { 3414 const u64 *new = (const u64 *)_new; 3415 const u64 *old = (const u64 *)_old; 3416 u64 *res = (u64 *)_res; 3417 int i; 3418 3419 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3420 u64 nv = new[i]; 3421 u64 ov = old[i]; 3422 s64 delta = nv - ov; 3423 3424 /* detects if this particular field is 32bit only */ 3425 if (((nv | ov) >> 32) == 0) 3426 delta = (s64)(s32)((u32)nv - (u32)ov); 3427 3428 /* filter anomalies, some drivers reset their stats 3429 * at down/up events. 3430 */ 3431 if (delta > 0) 3432 res[i] += delta; 3433 } 3434 } 3435 3436 static int bond_get_nest_level(struct net_device *bond_dev) 3437 { 3438 struct bonding *bond = netdev_priv(bond_dev); 3439 3440 return bond->nest_level; 3441 } 3442 3443 static void bond_get_stats(struct net_device *bond_dev, 3444 struct rtnl_link_stats64 *stats) 3445 { 3446 struct bonding *bond = netdev_priv(bond_dev); 3447 struct rtnl_link_stats64 temp; 3448 struct list_head *iter; 3449 struct slave *slave; 3450 3451 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev)); 3452 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3453 3454 rcu_read_lock(); 3455 bond_for_each_slave_rcu(bond, slave, iter) { 3456 const struct rtnl_link_stats64 *new = 3457 dev_get_stats(slave->dev, &temp); 3458 3459 bond_fold_stats(stats, new, &slave->slave_stats); 3460 3461 /* save off the slave stats for the next run */ 3462 memcpy(&slave->slave_stats, new, sizeof(*new)); 3463 } 3464 rcu_read_unlock(); 3465 3466 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3467 spin_unlock(&bond->stats_lock); 3468 } 3469 3470 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3471 { 3472 struct bonding *bond = netdev_priv(bond_dev); 3473 struct net_device *slave_dev = NULL; 3474 struct ifbond k_binfo; 3475 struct ifbond __user *u_binfo = NULL; 3476 struct ifslave k_sinfo; 3477 struct ifslave __user *u_sinfo = NULL; 3478 struct mii_ioctl_data *mii = NULL; 3479 struct bond_opt_value newval; 3480 struct net *net; 3481 int res = 0; 3482 3483 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3484 3485 switch (cmd) { 3486 case SIOCGMIIPHY: 3487 mii = if_mii(ifr); 3488 if (!mii) 3489 return -EINVAL; 3490 3491 mii->phy_id = 0; 3492 /* Fall Through */ 3493 case SIOCGMIIREG: 3494 /* We do this again just in case we were called by SIOCGMIIREG 3495 * instead of SIOCGMIIPHY. 3496 */ 3497 mii = if_mii(ifr); 3498 if (!mii) 3499 return -EINVAL; 3500 3501 if (mii->reg_num == 1) { 3502 mii->val_out = 0; 3503 if (netif_carrier_ok(bond->dev)) 3504 mii->val_out = BMSR_LSTATUS; 3505 } 3506 3507 return 0; 3508 case BOND_INFO_QUERY_OLD: 3509 case SIOCBONDINFOQUERY: 3510 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3511 3512 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3513 return -EFAULT; 3514 3515 bond_info_query(bond_dev, &k_binfo); 3516 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3517 return -EFAULT; 3518 3519 return 0; 3520 case BOND_SLAVE_INFO_QUERY_OLD: 3521 case SIOCBONDSLAVEINFOQUERY: 3522 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3523 3524 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3525 return -EFAULT; 3526 3527 res = bond_slave_info_query(bond_dev, &k_sinfo); 3528 if (res == 0 && 3529 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3530 return -EFAULT; 3531 3532 return res; 3533 default: 3534 break; 3535 } 3536 3537 net = dev_net(bond_dev); 3538 3539 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3540 return -EPERM; 3541 3542 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3543 3544 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev); 3545 3546 if (!slave_dev) 3547 return -ENODEV; 3548 3549 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name); 3550 switch (cmd) { 3551 case BOND_ENSLAVE_OLD: 3552 case SIOCBONDENSLAVE: 3553 res = bond_enslave(bond_dev, slave_dev, NULL); 3554 break; 3555 case BOND_RELEASE_OLD: 3556 case SIOCBONDRELEASE: 3557 res = bond_release(bond_dev, slave_dev); 3558 break; 3559 case BOND_SETHWADDR_OLD: 3560 case SIOCBONDSETHWADDR: 3561 res = bond_set_dev_addr(bond_dev, slave_dev); 3562 break; 3563 case BOND_CHANGE_ACTIVE_OLD: 3564 case SIOCBONDCHANGEACTIVE: 3565 bond_opt_initstr(&newval, slave_dev->name); 3566 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3567 &newval); 3568 break; 3569 default: 3570 res = -EOPNOTSUPP; 3571 } 3572 3573 return res; 3574 } 3575 3576 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3577 { 3578 struct bonding *bond = netdev_priv(bond_dev); 3579 3580 if (change & IFF_PROMISC) 3581 bond_set_promiscuity(bond, 3582 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3583 3584 if (change & IFF_ALLMULTI) 3585 bond_set_allmulti(bond, 3586 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3587 } 3588 3589 static void bond_set_rx_mode(struct net_device *bond_dev) 3590 { 3591 struct bonding *bond = netdev_priv(bond_dev); 3592 struct list_head *iter; 3593 struct slave *slave; 3594 3595 rcu_read_lock(); 3596 if (bond_uses_primary(bond)) { 3597 slave = rcu_dereference(bond->curr_active_slave); 3598 if (slave) { 3599 dev_uc_sync(slave->dev, bond_dev); 3600 dev_mc_sync(slave->dev, bond_dev); 3601 } 3602 } else { 3603 bond_for_each_slave_rcu(bond, slave, iter) { 3604 dev_uc_sync_multiple(slave->dev, bond_dev); 3605 dev_mc_sync_multiple(slave->dev, bond_dev); 3606 } 3607 } 3608 rcu_read_unlock(); 3609 } 3610 3611 static int bond_neigh_init(struct neighbour *n) 3612 { 3613 struct bonding *bond = netdev_priv(n->dev); 3614 const struct net_device_ops *slave_ops; 3615 struct neigh_parms parms; 3616 struct slave *slave; 3617 int ret; 3618 3619 slave = bond_first_slave(bond); 3620 if (!slave) 3621 return 0; 3622 slave_ops = slave->dev->netdev_ops; 3623 if (!slave_ops->ndo_neigh_setup) 3624 return 0; 3625 3626 parms.neigh_setup = NULL; 3627 parms.neigh_cleanup = NULL; 3628 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3629 if (ret) 3630 return ret; 3631 3632 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called 3633 * after the last slave has been detached. Assumes that all slaves 3634 * utilize the same neigh_cleanup (true at this writing as only user 3635 * is ipoib). 3636 */ 3637 n->parms->neigh_cleanup = parms.neigh_cleanup; 3638 3639 if (!parms.neigh_setup) 3640 return 0; 3641 3642 return parms.neigh_setup(n); 3643 } 3644 3645 /* The bonding ndo_neigh_setup is called at init time beofre any 3646 * slave exists. So we must declare proxy setup function which will 3647 * be used at run time to resolve the actual slave neigh param setup. 3648 * 3649 * It's also called by master devices (such as vlans) to setup their 3650 * underlying devices. In that case - do nothing, we're already set up from 3651 * our init. 3652 */ 3653 static int bond_neigh_setup(struct net_device *dev, 3654 struct neigh_parms *parms) 3655 { 3656 /* modify only our neigh_parms */ 3657 if (parms->dev == dev) 3658 parms->neigh_setup = bond_neigh_init; 3659 3660 return 0; 3661 } 3662 3663 /* Change the MTU of all of a master's slaves to match the master */ 3664 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3665 { 3666 struct bonding *bond = netdev_priv(bond_dev); 3667 struct slave *slave, *rollback_slave; 3668 struct list_head *iter; 3669 int res = 0; 3670 3671 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3672 3673 bond_for_each_slave(bond, slave, iter) { 3674 netdev_dbg(bond_dev, "s %p c_m %p\n", 3675 slave, slave->dev->netdev_ops->ndo_change_mtu); 3676 3677 res = dev_set_mtu(slave->dev, new_mtu); 3678 3679 if (res) { 3680 /* If we failed to set the slave's mtu to the new value 3681 * we must abort the operation even in ACTIVE_BACKUP 3682 * mode, because if we allow the backup slaves to have 3683 * different mtu values than the active slave we'll 3684 * need to change their mtu when doing a failover. That 3685 * means changing their mtu from timer context, which 3686 * is probably not a good idea. 3687 */ 3688 netdev_dbg(bond_dev, "err %d %s\n", res, 3689 slave->dev->name); 3690 goto unwind; 3691 } 3692 } 3693 3694 bond_dev->mtu = new_mtu; 3695 3696 return 0; 3697 3698 unwind: 3699 /* unwind from head to the slave that failed */ 3700 bond_for_each_slave(bond, rollback_slave, iter) { 3701 int tmp_res; 3702 3703 if (rollback_slave == slave) 3704 break; 3705 3706 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3707 if (tmp_res) { 3708 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3709 tmp_res, rollback_slave->dev->name); 3710 } 3711 } 3712 3713 return res; 3714 } 3715 3716 /* Change HW address 3717 * 3718 * Note that many devices must be down to change the HW address, and 3719 * downing the master releases all slaves. We can make bonds full of 3720 * bonding devices to test this, however. 3721 */ 3722 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3723 { 3724 struct bonding *bond = netdev_priv(bond_dev); 3725 struct slave *slave, *rollback_slave; 3726 struct sockaddr_storage *ss = addr, tmp_ss; 3727 struct list_head *iter; 3728 int res = 0; 3729 3730 if (BOND_MODE(bond) == BOND_MODE_ALB) 3731 return bond_alb_set_mac_address(bond_dev, addr); 3732 3733 3734 netdev_dbg(bond_dev, "bond=%p\n", bond); 3735 3736 /* If fail_over_mac is enabled, do nothing and return success. 3737 * Returning an error causes ifenslave to fail. 3738 */ 3739 if (bond->params.fail_over_mac && 3740 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3741 return 0; 3742 3743 if (!is_valid_ether_addr(ss->__data)) 3744 return -EADDRNOTAVAIL; 3745 3746 bond_for_each_slave(bond, slave, iter) { 3747 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name); 3748 res = dev_set_mac_address(slave->dev, addr, NULL); 3749 if (res) { 3750 /* TODO: consider downing the slave 3751 * and retry ? 3752 * User should expect communications 3753 * breakage anyway until ARP finish 3754 * updating, so... 3755 */ 3756 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name); 3757 goto unwind; 3758 } 3759 } 3760 3761 /* success */ 3762 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 3763 return 0; 3764 3765 unwind: 3766 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 3767 tmp_ss.ss_family = bond_dev->type; 3768 3769 /* unwind from head to the slave that failed */ 3770 bond_for_each_slave(bond, rollback_slave, iter) { 3771 int tmp_res; 3772 3773 if (rollback_slave == slave) 3774 break; 3775 3776 tmp_res = dev_set_mac_address(rollback_slave->dev, 3777 (struct sockaddr *)&tmp_ss, NULL); 3778 if (tmp_res) { 3779 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3780 tmp_res, rollback_slave->dev->name); 3781 } 3782 } 3783 3784 return res; 3785 } 3786 3787 /** 3788 * bond_xmit_slave_id - transmit skb through slave with slave_id 3789 * @bond: bonding device that is transmitting 3790 * @skb: buffer to transmit 3791 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3792 * 3793 * This function tries to transmit through slave with slave_id but in case 3794 * it fails, it tries to find the first available slave for transmission. 3795 * The skb is consumed in all cases, thus the function is void. 3796 */ 3797 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3798 { 3799 struct list_head *iter; 3800 struct slave *slave; 3801 int i = slave_id; 3802 3803 /* Here we start from the slave with slave_id */ 3804 bond_for_each_slave_rcu(bond, slave, iter) { 3805 if (--i < 0) { 3806 if (bond_slave_can_tx(slave)) { 3807 bond_dev_queue_xmit(bond, skb, slave->dev); 3808 return; 3809 } 3810 } 3811 } 3812 3813 /* Here we start from the first slave up to slave_id */ 3814 i = slave_id; 3815 bond_for_each_slave_rcu(bond, slave, iter) { 3816 if (--i < 0) 3817 break; 3818 if (bond_slave_can_tx(slave)) { 3819 bond_dev_queue_xmit(bond, skb, slave->dev); 3820 return; 3821 } 3822 } 3823 /* no slave that can tx has been found */ 3824 bond_tx_drop(bond->dev, skb); 3825 } 3826 3827 /** 3828 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3829 * @bond: bonding device to use 3830 * 3831 * Based on the value of the bonding device's packets_per_slave parameter 3832 * this function generates a slave id, which is usually used as the next 3833 * slave to transmit through. 3834 */ 3835 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3836 { 3837 u32 slave_id; 3838 struct reciprocal_value reciprocal_packets_per_slave; 3839 int packets_per_slave = bond->params.packets_per_slave; 3840 3841 switch (packets_per_slave) { 3842 case 0: 3843 slave_id = prandom_u32(); 3844 break; 3845 case 1: 3846 slave_id = bond->rr_tx_counter; 3847 break; 3848 default: 3849 reciprocal_packets_per_slave = 3850 bond->params.reciprocal_packets_per_slave; 3851 slave_id = reciprocal_divide(bond->rr_tx_counter, 3852 reciprocal_packets_per_slave); 3853 break; 3854 } 3855 bond->rr_tx_counter++; 3856 3857 return slave_id; 3858 } 3859 3860 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 3861 struct net_device *bond_dev) 3862 { 3863 struct bonding *bond = netdev_priv(bond_dev); 3864 struct iphdr *iph = ip_hdr(skb); 3865 struct slave *slave; 3866 u32 slave_id; 3867 3868 /* Start with the curr_active_slave that joined the bond as the 3869 * default for sending IGMP traffic. For failover purposes one 3870 * needs to maintain some consistency for the interface that will 3871 * send the join/membership reports. The curr_active_slave found 3872 * will send all of this type of traffic. 3873 */ 3874 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) { 3875 slave = rcu_dereference(bond->curr_active_slave); 3876 if (slave) 3877 bond_dev_queue_xmit(bond, skb, slave->dev); 3878 else 3879 bond_xmit_slave_id(bond, skb, 0); 3880 } else { 3881 int slave_cnt = READ_ONCE(bond->slave_cnt); 3882 3883 if (likely(slave_cnt)) { 3884 slave_id = bond_rr_gen_slave_id(bond); 3885 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3886 } else { 3887 bond_tx_drop(bond_dev, skb); 3888 } 3889 } 3890 3891 return NETDEV_TX_OK; 3892 } 3893 3894 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3895 * the bond has a usable interface. 3896 */ 3897 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 3898 struct net_device *bond_dev) 3899 { 3900 struct bonding *bond = netdev_priv(bond_dev); 3901 struct slave *slave; 3902 3903 slave = rcu_dereference(bond->curr_active_slave); 3904 if (slave) 3905 bond_dev_queue_xmit(bond, skb, slave->dev); 3906 else 3907 bond_tx_drop(bond_dev, skb); 3908 3909 return NETDEV_TX_OK; 3910 } 3911 3912 /* Use this to update slave_array when (a) it's not appropriate to update 3913 * slave_array right away (note that update_slave_array() may sleep) 3914 * and / or (b) RTNL is not held. 3915 */ 3916 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 3917 { 3918 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 3919 } 3920 3921 /* Slave array work handler. Holds only RTNL */ 3922 static void bond_slave_arr_handler(struct work_struct *work) 3923 { 3924 struct bonding *bond = container_of(work, struct bonding, 3925 slave_arr_work.work); 3926 int ret; 3927 3928 if (!rtnl_trylock()) 3929 goto err; 3930 3931 ret = bond_update_slave_arr(bond, NULL); 3932 rtnl_unlock(); 3933 if (ret) { 3934 pr_warn_ratelimited("Failed to update slave array from WT\n"); 3935 goto err; 3936 } 3937 return; 3938 3939 err: 3940 bond_slave_arr_work_rearm(bond, 1); 3941 } 3942 3943 /* Build the usable slaves array in control path for modes that use xmit-hash 3944 * to determine the slave interface - 3945 * (a) BOND_MODE_8023AD 3946 * (b) BOND_MODE_XOR 3947 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 3948 * 3949 * The caller is expected to hold RTNL only and NO other lock! 3950 */ 3951 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 3952 { 3953 struct slave *slave; 3954 struct list_head *iter; 3955 struct bond_up_slave *new_arr, *old_arr; 3956 int agg_id = 0; 3957 int ret = 0; 3958 3959 #ifdef CONFIG_LOCKDEP 3960 WARN_ON(lockdep_is_held(&bond->mode_lock)); 3961 #endif 3962 3963 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 3964 GFP_KERNEL); 3965 if (!new_arr) { 3966 ret = -ENOMEM; 3967 pr_err("Failed to build slave-array.\n"); 3968 goto out; 3969 } 3970 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3971 struct ad_info ad_info; 3972 3973 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 3974 pr_debug("bond_3ad_get_active_agg_info failed\n"); 3975 kfree_rcu(new_arr, rcu); 3976 /* No active aggragator means it's not safe to use 3977 * the previous array. 3978 */ 3979 old_arr = rtnl_dereference(bond->slave_arr); 3980 if (old_arr) { 3981 RCU_INIT_POINTER(bond->slave_arr, NULL); 3982 kfree_rcu(old_arr, rcu); 3983 } 3984 goto out; 3985 } 3986 agg_id = ad_info.aggregator_id; 3987 } 3988 bond_for_each_slave(bond, slave, iter) { 3989 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3990 struct aggregator *agg; 3991 3992 agg = SLAVE_AD_INFO(slave)->port.aggregator; 3993 if (!agg || agg->aggregator_identifier != agg_id) 3994 continue; 3995 } 3996 if (!bond_slave_can_tx(slave)) 3997 continue; 3998 if (skipslave == slave) 3999 continue; 4000 4001 netdev_dbg(bond->dev, 4002 "Adding slave dev %s to tx hash array[%d]\n", 4003 slave->dev->name, new_arr->count); 4004 4005 new_arr->arr[new_arr->count++] = slave; 4006 } 4007 4008 old_arr = rtnl_dereference(bond->slave_arr); 4009 rcu_assign_pointer(bond->slave_arr, new_arr); 4010 if (old_arr) 4011 kfree_rcu(old_arr, rcu); 4012 out: 4013 if (ret != 0 && skipslave) { 4014 int idx; 4015 4016 /* Rare situation where caller has asked to skip a specific 4017 * slave but allocation failed (most likely!). BTW this is 4018 * only possible when the call is initiated from 4019 * __bond_release_one(). In this situation; overwrite the 4020 * skipslave entry in the array with the last entry from the 4021 * array to avoid a situation where the xmit path may choose 4022 * this to-be-skipped slave to send a packet out. 4023 */ 4024 old_arr = rtnl_dereference(bond->slave_arr); 4025 for (idx = 0; idx < old_arr->count; idx++) { 4026 if (skipslave == old_arr->arr[idx]) { 4027 old_arr->arr[idx] = 4028 old_arr->arr[old_arr->count-1]; 4029 old_arr->count--; 4030 break; 4031 } 4032 } 4033 } 4034 return ret; 4035 } 4036 4037 /* Use this Xmit function for 3AD as well as XOR modes. The current 4038 * usable slave array is formed in the control path. The xmit function 4039 * just calculates hash and sends the packet out. 4040 */ 4041 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4042 struct net_device *dev) 4043 { 4044 struct bonding *bond = netdev_priv(dev); 4045 struct slave *slave; 4046 struct bond_up_slave *slaves; 4047 unsigned int count; 4048 4049 slaves = rcu_dereference(bond->slave_arr); 4050 count = slaves ? READ_ONCE(slaves->count) : 0; 4051 if (likely(count)) { 4052 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 4053 bond_dev_queue_xmit(bond, skb, slave->dev); 4054 } else { 4055 bond_tx_drop(dev, skb); 4056 } 4057 4058 return NETDEV_TX_OK; 4059 } 4060 4061 /* in broadcast mode, we send everything to all usable interfaces. */ 4062 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4063 struct net_device *bond_dev) 4064 { 4065 struct bonding *bond = netdev_priv(bond_dev); 4066 struct slave *slave = NULL; 4067 struct list_head *iter; 4068 4069 bond_for_each_slave_rcu(bond, slave, iter) { 4070 if (bond_is_last_slave(bond, slave)) 4071 break; 4072 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4073 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4074 4075 if (!skb2) { 4076 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4077 bond_dev->name, __func__); 4078 continue; 4079 } 4080 bond_dev_queue_xmit(bond, skb2, slave->dev); 4081 } 4082 } 4083 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4084 bond_dev_queue_xmit(bond, skb, slave->dev); 4085 else 4086 bond_tx_drop(bond_dev, skb); 4087 4088 return NETDEV_TX_OK; 4089 } 4090 4091 /*------------------------- Device initialization ---------------------------*/ 4092 4093 /* Lookup the slave that corresponds to a qid */ 4094 static inline int bond_slave_override(struct bonding *bond, 4095 struct sk_buff *skb) 4096 { 4097 struct slave *slave = NULL; 4098 struct list_head *iter; 4099 4100 if (!skb_rx_queue_recorded(skb)) 4101 return 1; 4102 4103 /* Find out if any slaves have the same mapping as this skb. */ 4104 bond_for_each_slave_rcu(bond, slave, iter) { 4105 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4106 if (bond_slave_is_up(slave) && 4107 slave->link == BOND_LINK_UP) { 4108 bond_dev_queue_xmit(bond, skb, slave->dev); 4109 return 0; 4110 } 4111 /* If the slave isn't UP, use default transmit policy. */ 4112 break; 4113 } 4114 } 4115 4116 return 1; 4117 } 4118 4119 4120 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4121 struct net_device *sb_dev, 4122 select_queue_fallback_t fallback) 4123 { 4124 /* This helper function exists to help dev_pick_tx get the correct 4125 * destination queue. Using a helper function skips a call to 4126 * skb_tx_hash and will put the skbs in the queue we expect on their 4127 * way down to the bonding driver. 4128 */ 4129 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4130 4131 /* Save the original txq to restore before passing to the driver */ 4132 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4133 4134 if (unlikely(txq >= dev->real_num_tx_queues)) { 4135 do { 4136 txq -= dev->real_num_tx_queues; 4137 } while (txq >= dev->real_num_tx_queues); 4138 } 4139 return txq; 4140 } 4141 4142 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4143 { 4144 struct bonding *bond = netdev_priv(dev); 4145 4146 if (bond_should_override_tx_queue(bond) && 4147 !bond_slave_override(bond, skb)) 4148 return NETDEV_TX_OK; 4149 4150 switch (BOND_MODE(bond)) { 4151 case BOND_MODE_ROUNDROBIN: 4152 return bond_xmit_roundrobin(skb, dev); 4153 case BOND_MODE_ACTIVEBACKUP: 4154 return bond_xmit_activebackup(skb, dev); 4155 case BOND_MODE_8023AD: 4156 case BOND_MODE_XOR: 4157 return bond_3ad_xor_xmit(skb, dev); 4158 case BOND_MODE_BROADCAST: 4159 return bond_xmit_broadcast(skb, dev); 4160 case BOND_MODE_ALB: 4161 return bond_alb_xmit(skb, dev); 4162 case BOND_MODE_TLB: 4163 return bond_tlb_xmit(skb, dev); 4164 default: 4165 /* Should never happen, mode already checked */ 4166 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4167 WARN_ON_ONCE(1); 4168 bond_tx_drop(dev, skb); 4169 return NETDEV_TX_OK; 4170 } 4171 } 4172 4173 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4174 { 4175 struct bonding *bond = netdev_priv(dev); 4176 netdev_tx_t ret = NETDEV_TX_OK; 4177 4178 /* If we risk deadlock from transmitting this in the 4179 * netpoll path, tell netpoll to queue the frame for later tx 4180 */ 4181 if (unlikely(is_netpoll_tx_blocked(dev))) 4182 return NETDEV_TX_BUSY; 4183 4184 rcu_read_lock(); 4185 if (bond_has_slaves(bond)) 4186 ret = __bond_start_xmit(skb, dev); 4187 else 4188 bond_tx_drop(dev, skb); 4189 rcu_read_unlock(); 4190 4191 return ret; 4192 } 4193 4194 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4195 struct ethtool_link_ksettings *cmd) 4196 { 4197 struct bonding *bond = netdev_priv(bond_dev); 4198 unsigned long speed = 0; 4199 struct list_head *iter; 4200 struct slave *slave; 4201 4202 cmd->base.duplex = DUPLEX_UNKNOWN; 4203 cmd->base.port = PORT_OTHER; 4204 4205 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4206 * do not need to check mode. Though link speed might not represent 4207 * the true receive or transmit bandwidth (not all modes are symmetric) 4208 * this is an accurate maximum. 4209 */ 4210 bond_for_each_slave(bond, slave, iter) { 4211 if (bond_slave_can_tx(slave)) { 4212 if (slave->speed != SPEED_UNKNOWN) 4213 speed += slave->speed; 4214 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4215 slave->duplex != DUPLEX_UNKNOWN) 4216 cmd->base.duplex = slave->duplex; 4217 } 4218 } 4219 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4220 4221 return 0; 4222 } 4223 4224 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4225 struct ethtool_drvinfo *drvinfo) 4226 { 4227 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4228 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 4229 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4230 BOND_ABI_VERSION); 4231 } 4232 4233 static const struct ethtool_ops bond_ethtool_ops = { 4234 .get_drvinfo = bond_ethtool_get_drvinfo, 4235 .get_link = ethtool_op_get_link, 4236 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4237 }; 4238 4239 static const struct net_device_ops bond_netdev_ops = { 4240 .ndo_init = bond_init, 4241 .ndo_uninit = bond_uninit, 4242 .ndo_open = bond_open, 4243 .ndo_stop = bond_close, 4244 .ndo_start_xmit = bond_start_xmit, 4245 .ndo_select_queue = bond_select_queue, 4246 .ndo_get_stats64 = bond_get_stats, 4247 .ndo_do_ioctl = bond_do_ioctl, 4248 .ndo_change_rx_flags = bond_change_rx_flags, 4249 .ndo_set_rx_mode = bond_set_rx_mode, 4250 .ndo_change_mtu = bond_change_mtu, 4251 .ndo_set_mac_address = bond_set_mac_address, 4252 .ndo_neigh_setup = bond_neigh_setup, 4253 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4254 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4255 .ndo_get_lock_subclass = bond_get_nest_level, 4256 #ifdef CONFIG_NET_POLL_CONTROLLER 4257 .ndo_netpoll_setup = bond_netpoll_setup, 4258 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4259 .ndo_poll_controller = bond_poll_controller, 4260 #endif 4261 .ndo_add_slave = bond_enslave, 4262 .ndo_del_slave = bond_release, 4263 .ndo_fix_features = bond_fix_features, 4264 .ndo_features_check = passthru_features_check, 4265 }; 4266 4267 static const struct device_type bond_type = { 4268 .name = "bond", 4269 }; 4270 4271 static void bond_destructor(struct net_device *bond_dev) 4272 { 4273 struct bonding *bond = netdev_priv(bond_dev); 4274 if (bond->wq) 4275 destroy_workqueue(bond->wq); 4276 } 4277 4278 void bond_setup(struct net_device *bond_dev) 4279 { 4280 struct bonding *bond = netdev_priv(bond_dev); 4281 4282 spin_lock_init(&bond->mode_lock); 4283 spin_lock_init(&bond->stats_lock); 4284 bond->params = bonding_defaults; 4285 4286 /* Initialize pointers */ 4287 bond->dev = bond_dev; 4288 4289 /* Initialize the device entry points */ 4290 ether_setup(bond_dev); 4291 bond_dev->max_mtu = ETH_MAX_MTU; 4292 bond_dev->netdev_ops = &bond_netdev_ops; 4293 bond_dev->ethtool_ops = &bond_ethtool_ops; 4294 4295 bond_dev->needs_free_netdev = true; 4296 bond_dev->priv_destructor = bond_destructor; 4297 4298 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4299 4300 /* Initialize the device options */ 4301 bond_dev->flags |= IFF_MASTER; 4302 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4303 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4304 4305 /* don't acquire bond device's netif_tx_lock when transmitting */ 4306 bond_dev->features |= NETIF_F_LLTX; 4307 4308 /* By default, we declare the bond to be fully 4309 * VLAN hardware accelerated capable. Special 4310 * care is taken in the various xmit functions 4311 * when there are slaves that are not hw accel 4312 * capable 4313 */ 4314 4315 /* Don't allow bond devices to change network namespaces. */ 4316 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4317 4318 bond_dev->hw_features = BOND_VLAN_FEATURES | 4319 NETIF_F_HW_VLAN_CTAG_TX | 4320 NETIF_F_HW_VLAN_CTAG_RX | 4321 NETIF_F_HW_VLAN_CTAG_FILTER; 4322 4323 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4324 bond_dev->features |= bond_dev->hw_features; 4325 } 4326 4327 /* Destroy a bonding device. 4328 * Must be under rtnl_lock when this function is called. 4329 */ 4330 static void bond_uninit(struct net_device *bond_dev) 4331 { 4332 struct bonding *bond = netdev_priv(bond_dev); 4333 struct list_head *iter; 4334 struct slave *slave; 4335 struct bond_up_slave *arr; 4336 4337 bond_netpoll_cleanup(bond_dev); 4338 4339 /* Release the bonded slaves */ 4340 bond_for_each_slave(bond, slave, iter) 4341 __bond_release_one(bond_dev, slave->dev, true, true); 4342 netdev_info(bond_dev, "Released all slaves\n"); 4343 4344 arr = rtnl_dereference(bond->slave_arr); 4345 if (arr) { 4346 RCU_INIT_POINTER(bond->slave_arr, NULL); 4347 kfree_rcu(arr, rcu); 4348 } 4349 4350 list_del(&bond->bond_list); 4351 4352 bond_debug_unregister(bond); 4353 } 4354 4355 /*------------------------- Module initialization ---------------------------*/ 4356 4357 static int bond_check_params(struct bond_params *params) 4358 { 4359 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4360 struct bond_opt_value newval; 4361 const struct bond_opt_value *valptr; 4362 int arp_all_targets_value = 0; 4363 u16 ad_actor_sys_prio = 0; 4364 u16 ad_user_port_key = 0; 4365 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4366 int arp_ip_count; 4367 int bond_mode = BOND_MODE_ROUNDROBIN; 4368 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4369 int lacp_fast = 0; 4370 int tlb_dynamic_lb; 4371 4372 /* Convert string parameters. */ 4373 if (mode) { 4374 bond_opt_initstr(&newval, mode); 4375 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4376 if (!valptr) { 4377 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4378 return -EINVAL; 4379 } 4380 bond_mode = valptr->value; 4381 } 4382 4383 if (xmit_hash_policy) { 4384 if (bond_mode == BOND_MODE_ROUNDROBIN || 4385 bond_mode == BOND_MODE_ACTIVEBACKUP || 4386 bond_mode == BOND_MODE_BROADCAST) { 4387 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4388 bond_mode_name(bond_mode)); 4389 } else { 4390 bond_opt_initstr(&newval, xmit_hash_policy); 4391 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4392 &newval); 4393 if (!valptr) { 4394 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4395 xmit_hash_policy); 4396 return -EINVAL; 4397 } 4398 xmit_hashtype = valptr->value; 4399 } 4400 } 4401 4402 if (lacp_rate) { 4403 if (bond_mode != BOND_MODE_8023AD) { 4404 pr_info("lacp_rate param is irrelevant in mode %s\n", 4405 bond_mode_name(bond_mode)); 4406 } else { 4407 bond_opt_initstr(&newval, lacp_rate); 4408 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4409 &newval); 4410 if (!valptr) { 4411 pr_err("Error: Invalid lacp rate \"%s\"\n", 4412 lacp_rate); 4413 return -EINVAL; 4414 } 4415 lacp_fast = valptr->value; 4416 } 4417 } 4418 4419 if (ad_select) { 4420 bond_opt_initstr(&newval, ad_select); 4421 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4422 &newval); 4423 if (!valptr) { 4424 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4425 return -EINVAL; 4426 } 4427 params->ad_select = valptr->value; 4428 if (bond_mode != BOND_MODE_8023AD) 4429 pr_warn("ad_select param only affects 802.3ad mode\n"); 4430 } else { 4431 params->ad_select = BOND_AD_STABLE; 4432 } 4433 4434 if (max_bonds < 0) { 4435 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4436 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4437 max_bonds = BOND_DEFAULT_MAX_BONDS; 4438 } 4439 4440 if (miimon < 0) { 4441 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4442 miimon, INT_MAX); 4443 miimon = 0; 4444 } 4445 4446 if (updelay < 0) { 4447 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4448 updelay, INT_MAX); 4449 updelay = 0; 4450 } 4451 4452 if (downdelay < 0) { 4453 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4454 downdelay, INT_MAX); 4455 downdelay = 0; 4456 } 4457 4458 if ((use_carrier != 0) && (use_carrier != 1)) { 4459 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4460 use_carrier); 4461 use_carrier = 1; 4462 } 4463 4464 if (num_peer_notif < 0 || num_peer_notif > 255) { 4465 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4466 num_peer_notif); 4467 num_peer_notif = 1; 4468 } 4469 4470 /* reset values for 802.3ad/TLB/ALB */ 4471 if (!bond_mode_uses_arp(bond_mode)) { 4472 if (!miimon) { 4473 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"); 4474 pr_warn("Forcing miimon to 100msec\n"); 4475 miimon = BOND_DEFAULT_MIIMON; 4476 } 4477 } 4478 4479 if (tx_queues < 1 || tx_queues > 255) { 4480 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4481 tx_queues, BOND_DEFAULT_TX_QUEUES); 4482 tx_queues = BOND_DEFAULT_TX_QUEUES; 4483 } 4484 4485 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4486 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4487 all_slaves_active); 4488 all_slaves_active = 0; 4489 } 4490 4491 if (resend_igmp < 0 || resend_igmp > 255) { 4492 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4493 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4494 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4495 } 4496 4497 bond_opt_initval(&newval, packets_per_slave); 4498 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4499 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4500 packets_per_slave, USHRT_MAX); 4501 packets_per_slave = 1; 4502 } 4503 4504 if (bond_mode == BOND_MODE_ALB) { 4505 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", 4506 updelay); 4507 } 4508 4509 if (!miimon) { 4510 if (updelay || downdelay) { 4511 /* just warn the user the up/down delay will have 4512 * no effect since miimon is zero... 4513 */ 4514 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", 4515 updelay, downdelay); 4516 } 4517 } else { 4518 /* don't allow arp monitoring */ 4519 if (arp_interval) { 4520 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4521 miimon, arp_interval); 4522 arp_interval = 0; 4523 } 4524 4525 if ((updelay % miimon) != 0) { 4526 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4527 updelay, miimon, (updelay / miimon) * miimon); 4528 } 4529 4530 updelay /= miimon; 4531 4532 if ((downdelay % miimon) != 0) { 4533 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4534 downdelay, miimon, 4535 (downdelay / miimon) * miimon); 4536 } 4537 4538 downdelay /= miimon; 4539 } 4540 4541 if (arp_interval < 0) { 4542 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4543 arp_interval, INT_MAX); 4544 arp_interval = 0; 4545 } 4546 4547 for (arp_ip_count = 0, i = 0; 4548 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4549 __be32 ip; 4550 4551 /* not a complete check, but good enough to catch mistakes */ 4552 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4553 !bond_is_ip_target_ok(ip)) { 4554 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4555 arp_ip_target[i]); 4556 arp_interval = 0; 4557 } else { 4558 if (bond_get_targets_ip(arp_target, ip) == -1) 4559 arp_target[arp_ip_count++] = ip; 4560 else 4561 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4562 &ip); 4563 } 4564 } 4565 4566 if (arp_interval && !arp_ip_count) { 4567 /* don't allow arping if no arp_ip_target given... */ 4568 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4569 arp_interval); 4570 arp_interval = 0; 4571 } 4572 4573 if (arp_validate) { 4574 if (!arp_interval) { 4575 pr_err("arp_validate requires arp_interval\n"); 4576 return -EINVAL; 4577 } 4578 4579 bond_opt_initstr(&newval, arp_validate); 4580 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4581 &newval); 4582 if (!valptr) { 4583 pr_err("Error: invalid arp_validate \"%s\"\n", 4584 arp_validate); 4585 return -EINVAL; 4586 } 4587 arp_validate_value = valptr->value; 4588 } else { 4589 arp_validate_value = 0; 4590 } 4591 4592 if (arp_all_targets) { 4593 bond_opt_initstr(&newval, arp_all_targets); 4594 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4595 &newval); 4596 if (!valptr) { 4597 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4598 arp_all_targets); 4599 arp_all_targets_value = 0; 4600 } else { 4601 arp_all_targets_value = valptr->value; 4602 } 4603 } 4604 4605 if (miimon) { 4606 pr_info("MII link monitoring set to %d ms\n", miimon); 4607 } else if (arp_interval) { 4608 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4609 arp_validate_value); 4610 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4611 arp_interval, valptr->string, arp_ip_count); 4612 4613 for (i = 0; i < arp_ip_count; i++) 4614 pr_cont(" %s", arp_ip_target[i]); 4615 4616 pr_cont("\n"); 4617 4618 } else if (max_bonds) { 4619 /* miimon and arp_interval not set, we need one so things 4620 * work as expected, see bonding.txt for details 4621 */ 4622 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"); 4623 } 4624 4625 if (primary && !bond_mode_uses_primary(bond_mode)) { 4626 /* currently, using a primary only makes sense 4627 * in active backup, TLB or ALB modes 4628 */ 4629 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4630 primary, bond_mode_name(bond_mode)); 4631 primary = NULL; 4632 } 4633 4634 if (primary && primary_reselect) { 4635 bond_opt_initstr(&newval, primary_reselect); 4636 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4637 &newval); 4638 if (!valptr) { 4639 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4640 primary_reselect); 4641 return -EINVAL; 4642 } 4643 primary_reselect_value = valptr->value; 4644 } else { 4645 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4646 } 4647 4648 if (fail_over_mac) { 4649 bond_opt_initstr(&newval, fail_over_mac); 4650 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4651 &newval); 4652 if (!valptr) { 4653 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4654 fail_over_mac); 4655 return -EINVAL; 4656 } 4657 fail_over_mac_value = valptr->value; 4658 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4659 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4660 } else { 4661 fail_over_mac_value = BOND_FOM_NONE; 4662 } 4663 4664 bond_opt_initstr(&newval, "default"); 4665 valptr = bond_opt_parse( 4666 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 4667 &newval); 4668 if (!valptr) { 4669 pr_err("Error: No ad_actor_sys_prio default value"); 4670 return -EINVAL; 4671 } 4672 ad_actor_sys_prio = valptr->value; 4673 4674 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 4675 &newval); 4676 if (!valptr) { 4677 pr_err("Error: No ad_user_port_key default value"); 4678 return -EINVAL; 4679 } 4680 ad_user_port_key = valptr->value; 4681 4682 bond_opt_initstr(&newval, "default"); 4683 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 4684 if (!valptr) { 4685 pr_err("Error: No tlb_dynamic_lb default value"); 4686 return -EINVAL; 4687 } 4688 tlb_dynamic_lb = valptr->value; 4689 4690 if (lp_interval == 0) { 4691 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4692 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4693 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4694 } 4695 4696 /* fill params struct with the proper values */ 4697 params->mode = bond_mode; 4698 params->xmit_policy = xmit_hashtype; 4699 params->miimon = miimon; 4700 params->num_peer_notif = num_peer_notif; 4701 params->arp_interval = arp_interval; 4702 params->arp_validate = arp_validate_value; 4703 params->arp_all_targets = arp_all_targets_value; 4704 params->updelay = updelay; 4705 params->downdelay = downdelay; 4706 params->use_carrier = use_carrier; 4707 params->lacp_fast = lacp_fast; 4708 params->primary[0] = 0; 4709 params->primary_reselect = primary_reselect_value; 4710 params->fail_over_mac = fail_over_mac_value; 4711 params->tx_queues = tx_queues; 4712 params->all_slaves_active = all_slaves_active; 4713 params->resend_igmp = resend_igmp; 4714 params->min_links = min_links; 4715 params->lp_interval = lp_interval; 4716 params->packets_per_slave = packets_per_slave; 4717 params->tlb_dynamic_lb = tlb_dynamic_lb; 4718 params->ad_actor_sys_prio = ad_actor_sys_prio; 4719 eth_zero_addr(params->ad_actor_system); 4720 params->ad_user_port_key = ad_user_port_key; 4721 if (packets_per_slave > 0) { 4722 params->reciprocal_packets_per_slave = 4723 reciprocal_value(packets_per_slave); 4724 } else { 4725 /* reciprocal_packets_per_slave is unused if 4726 * packets_per_slave is 0 or 1, just initialize it 4727 */ 4728 params->reciprocal_packets_per_slave = 4729 (struct reciprocal_value) { 0 }; 4730 } 4731 4732 if (primary) { 4733 strncpy(params->primary, primary, IFNAMSIZ); 4734 params->primary[IFNAMSIZ - 1] = 0; 4735 } 4736 4737 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4738 4739 return 0; 4740 } 4741 4742 /* Called from registration process */ 4743 static int bond_init(struct net_device *bond_dev) 4744 { 4745 struct bonding *bond = netdev_priv(bond_dev); 4746 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4747 4748 netdev_dbg(bond_dev, "Begin bond_init\n"); 4749 4750 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 4751 if (!bond->wq) 4752 return -ENOMEM; 4753 4754 bond->nest_level = SINGLE_DEPTH_NESTING; 4755 netdev_lockdep_set_classes(bond_dev); 4756 4757 list_add_tail(&bond->bond_list, &bn->dev_list); 4758 4759 bond_prepare_sysfs_group(bond); 4760 4761 bond_debug_register(bond); 4762 4763 /* Ensure valid dev_addr */ 4764 if (is_zero_ether_addr(bond_dev->dev_addr) && 4765 bond_dev->addr_assign_type == NET_ADDR_PERM) 4766 eth_hw_addr_random(bond_dev); 4767 4768 return 0; 4769 } 4770 4771 unsigned int bond_get_num_tx_queues(void) 4772 { 4773 return tx_queues; 4774 } 4775 4776 /* Create a new bond based on the specified name and bonding parameters. 4777 * If name is NULL, obtain a suitable "bond%d" name for us. 4778 * Caller must NOT hold rtnl_lock; we need to release it here before we 4779 * set up our sysfs entries. 4780 */ 4781 int bond_create(struct net *net, const char *name) 4782 { 4783 struct net_device *bond_dev; 4784 struct bonding *bond; 4785 struct alb_bond_info *bond_info; 4786 int res; 4787 4788 rtnl_lock(); 4789 4790 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4791 name ? name : "bond%d", NET_NAME_UNKNOWN, 4792 bond_setup, tx_queues); 4793 if (!bond_dev) { 4794 pr_err("%s: eek! can't alloc netdev!\n", name); 4795 rtnl_unlock(); 4796 return -ENOMEM; 4797 } 4798 4799 /* 4800 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 4801 * It is set to 0 by default which is wrong. 4802 */ 4803 bond = netdev_priv(bond_dev); 4804 bond_info = &(BOND_ALB_INFO(bond)); 4805 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 4806 4807 dev_net_set(bond_dev, net); 4808 bond_dev->rtnl_link_ops = &bond_link_ops; 4809 4810 res = register_netdevice(bond_dev); 4811 4812 netif_carrier_off(bond_dev); 4813 4814 bond_work_init_all(bond); 4815 4816 rtnl_unlock(); 4817 if (res < 0) 4818 free_netdev(bond_dev); 4819 return res; 4820 } 4821 4822 static int __net_init bond_net_init(struct net *net) 4823 { 4824 struct bond_net *bn = net_generic(net, bond_net_id); 4825 4826 bn->net = net; 4827 INIT_LIST_HEAD(&bn->dev_list); 4828 4829 bond_create_proc_dir(bn); 4830 bond_create_sysfs(bn); 4831 4832 return 0; 4833 } 4834 4835 static void __net_exit bond_net_exit(struct net *net) 4836 { 4837 struct bond_net *bn = net_generic(net, bond_net_id); 4838 struct bonding *bond, *tmp_bond; 4839 LIST_HEAD(list); 4840 4841 bond_destroy_sysfs(bn); 4842 4843 /* Kill off any bonds created after unregistering bond rtnl ops */ 4844 rtnl_lock(); 4845 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4846 unregister_netdevice_queue(bond->dev, &list); 4847 unregister_netdevice_many(&list); 4848 rtnl_unlock(); 4849 4850 bond_destroy_proc_dir(bn); 4851 } 4852 4853 static struct pernet_operations bond_net_ops = { 4854 .init = bond_net_init, 4855 .exit = bond_net_exit, 4856 .id = &bond_net_id, 4857 .size = sizeof(struct bond_net), 4858 }; 4859 4860 static int __init bonding_init(void) 4861 { 4862 int i; 4863 int res; 4864 4865 pr_info("%s", bond_version); 4866 4867 res = bond_check_params(&bonding_defaults); 4868 if (res) 4869 goto out; 4870 4871 res = register_pernet_subsys(&bond_net_ops); 4872 if (res) 4873 goto out; 4874 4875 res = bond_netlink_init(); 4876 if (res) 4877 goto err_link; 4878 4879 bond_create_debugfs(); 4880 4881 for (i = 0; i < max_bonds; i++) { 4882 res = bond_create(&init_net, NULL); 4883 if (res) 4884 goto err; 4885 } 4886 4887 register_netdevice_notifier(&bond_netdev_notifier); 4888 out: 4889 return res; 4890 err: 4891 bond_destroy_debugfs(); 4892 bond_netlink_fini(); 4893 err_link: 4894 unregister_pernet_subsys(&bond_net_ops); 4895 goto out; 4896 4897 } 4898 4899 static void __exit bonding_exit(void) 4900 { 4901 unregister_netdevice_notifier(&bond_netdev_notifier); 4902 4903 bond_destroy_debugfs(); 4904 4905 bond_netlink_fini(); 4906 unregister_pernet_subsys(&bond_net_ops); 4907 4908 #ifdef CONFIG_NET_POLL_CONTROLLER 4909 /* Make sure we don't have an imbalance on our netpoll blocking */ 4910 WARN_ON(atomic_read(&netpoll_block_tx)); 4911 #endif 4912 } 4913 4914 module_init(bonding_init); 4915 module_exit(bonding_exit); 4916 MODULE_LICENSE("GPL"); 4917 MODULE_VERSION(DRV_VERSION); 4918 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 4919 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 4920