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 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1081 NETIF_F_ALL_TSO) 1082 1083 static void bond_compute_features(struct bonding *bond) 1084 { 1085 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1086 IFF_XMIT_DST_RELEASE_PERM; 1087 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1088 netdev_features_t enc_features = BOND_ENC_FEATURES; 1089 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1090 struct net_device *bond_dev = bond->dev; 1091 struct list_head *iter; 1092 struct slave *slave; 1093 unsigned short max_hard_header_len = ETH_HLEN; 1094 unsigned int gso_max_size = GSO_MAX_SIZE; 1095 u16 gso_max_segs = GSO_MAX_SEGS; 1096 1097 if (!bond_has_slaves(bond)) 1098 goto done; 1099 vlan_features &= NETIF_F_ALL_FOR_ALL; 1100 mpls_features &= NETIF_F_ALL_FOR_ALL; 1101 1102 bond_for_each_slave(bond, slave, iter) { 1103 vlan_features = netdev_increment_features(vlan_features, 1104 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1105 1106 enc_features = netdev_increment_features(enc_features, 1107 slave->dev->hw_enc_features, 1108 BOND_ENC_FEATURES); 1109 1110 mpls_features = netdev_increment_features(mpls_features, 1111 slave->dev->mpls_features, 1112 BOND_MPLS_FEATURES); 1113 1114 dst_release_flag &= slave->dev->priv_flags; 1115 if (slave->dev->hard_header_len > max_hard_header_len) 1116 max_hard_header_len = slave->dev->hard_header_len; 1117 1118 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1119 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1120 } 1121 bond_dev->hard_header_len = max_hard_header_len; 1122 1123 done: 1124 bond_dev->vlan_features = vlan_features; 1125 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1126 NETIF_F_GSO_UDP_L4; 1127 bond_dev->mpls_features = mpls_features; 1128 bond_dev->gso_max_segs = gso_max_segs; 1129 netif_set_gso_max_size(bond_dev, gso_max_size); 1130 1131 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1132 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1133 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1134 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1135 1136 netdev_change_features(bond_dev); 1137 } 1138 1139 static void bond_setup_by_slave(struct net_device *bond_dev, 1140 struct net_device *slave_dev) 1141 { 1142 bond_dev->header_ops = slave_dev->header_ops; 1143 1144 bond_dev->type = slave_dev->type; 1145 bond_dev->hard_header_len = slave_dev->hard_header_len; 1146 bond_dev->addr_len = slave_dev->addr_len; 1147 1148 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1149 slave_dev->addr_len); 1150 } 1151 1152 /* On bonding slaves other than the currently active slave, suppress 1153 * duplicates except for alb non-mcast/bcast. 1154 */ 1155 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1156 struct slave *slave, 1157 struct bonding *bond) 1158 { 1159 if (bond_is_slave_inactive(slave)) { 1160 if (BOND_MODE(bond) == BOND_MODE_ALB && 1161 skb->pkt_type != PACKET_BROADCAST && 1162 skb->pkt_type != PACKET_MULTICAST) 1163 return false; 1164 return true; 1165 } 1166 return false; 1167 } 1168 1169 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1170 { 1171 struct sk_buff *skb = *pskb; 1172 struct slave *slave; 1173 struct bonding *bond; 1174 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1175 struct slave *); 1176 int ret = RX_HANDLER_ANOTHER; 1177 1178 skb = skb_share_check(skb, GFP_ATOMIC); 1179 if (unlikely(!skb)) 1180 return RX_HANDLER_CONSUMED; 1181 1182 *pskb = skb; 1183 1184 slave = bond_slave_get_rcu(skb->dev); 1185 bond = slave->bond; 1186 1187 recv_probe = READ_ONCE(bond->recv_probe); 1188 if (recv_probe) { 1189 ret = recv_probe(skb, bond, slave); 1190 if (ret == RX_HANDLER_CONSUMED) { 1191 consume_skb(skb); 1192 return ret; 1193 } 1194 } 1195 1196 /* 1197 * For packets determined by bond_should_deliver_exact_match() call to 1198 * be suppressed we want to make an exception for link-local packets. 1199 * This is necessary for e.g. LLDP daemons to be able to monitor 1200 * inactive slave links without being forced to bind to them 1201 * explicitly. 1202 * 1203 * At the same time, packets that are passed to the bonding master 1204 * (including link-local ones) can have their originating interface 1205 * determined via PACKET_ORIGDEV socket option. 1206 */ 1207 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1208 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1209 return RX_HANDLER_PASS; 1210 return RX_HANDLER_EXACT; 1211 } 1212 1213 skb->dev = bond->dev; 1214 1215 if (BOND_MODE(bond) == BOND_MODE_ALB && 1216 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1217 skb->pkt_type == PACKET_HOST) { 1218 1219 if (unlikely(skb_cow_head(skb, 1220 skb->data - skb_mac_header(skb)))) { 1221 kfree_skb(skb); 1222 return RX_HANDLER_CONSUMED; 1223 } 1224 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1225 bond->dev->addr_len); 1226 } 1227 1228 return ret; 1229 } 1230 1231 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1232 { 1233 switch (BOND_MODE(bond)) { 1234 case BOND_MODE_ROUNDROBIN: 1235 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1236 case BOND_MODE_ACTIVEBACKUP: 1237 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1238 case BOND_MODE_BROADCAST: 1239 return NETDEV_LAG_TX_TYPE_BROADCAST; 1240 case BOND_MODE_XOR: 1241 case BOND_MODE_8023AD: 1242 return NETDEV_LAG_TX_TYPE_HASH; 1243 default: 1244 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1245 } 1246 } 1247 1248 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1249 enum netdev_lag_tx_type type) 1250 { 1251 if (type != NETDEV_LAG_TX_TYPE_HASH) 1252 return NETDEV_LAG_HASH_NONE; 1253 1254 switch (bond->params.xmit_policy) { 1255 case BOND_XMIT_POLICY_LAYER2: 1256 return NETDEV_LAG_HASH_L2; 1257 case BOND_XMIT_POLICY_LAYER34: 1258 return NETDEV_LAG_HASH_L34; 1259 case BOND_XMIT_POLICY_LAYER23: 1260 return NETDEV_LAG_HASH_L23; 1261 case BOND_XMIT_POLICY_ENCAP23: 1262 return NETDEV_LAG_HASH_E23; 1263 case BOND_XMIT_POLICY_ENCAP34: 1264 return NETDEV_LAG_HASH_E34; 1265 default: 1266 return NETDEV_LAG_HASH_UNKNOWN; 1267 } 1268 } 1269 1270 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1271 struct netlink_ext_ack *extack) 1272 { 1273 struct netdev_lag_upper_info lag_upper_info; 1274 enum netdev_lag_tx_type type; 1275 1276 type = bond_lag_tx_type(bond); 1277 lag_upper_info.tx_type = type; 1278 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1279 1280 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1281 &lag_upper_info, extack); 1282 } 1283 1284 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1285 { 1286 netdev_upper_dev_unlink(slave->dev, bond->dev); 1287 slave->dev->flags &= ~IFF_SLAVE; 1288 } 1289 1290 static struct slave *bond_alloc_slave(struct bonding *bond) 1291 { 1292 struct slave *slave = NULL; 1293 1294 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1295 if (!slave) 1296 return NULL; 1297 1298 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1299 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1300 GFP_KERNEL); 1301 if (!SLAVE_AD_INFO(slave)) { 1302 kfree(slave); 1303 return NULL; 1304 } 1305 } 1306 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1307 1308 return slave; 1309 } 1310 1311 static void bond_free_slave(struct slave *slave) 1312 { 1313 struct bonding *bond = bond_get_bond_by_slave(slave); 1314 1315 cancel_delayed_work_sync(&slave->notify_work); 1316 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1317 kfree(SLAVE_AD_INFO(slave)); 1318 1319 kfree(slave); 1320 } 1321 1322 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1323 { 1324 info->bond_mode = BOND_MODE(bond); 1325 info->miimon = bond->params.miimon; 1326 info->num_slaves = bond->slave_cnt; 1327 } 1328 1329 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1330 { 1331 strcpy(info->slave_name, slave->dev->name); 1332 info->link = slave->link; 1333 info->state = bond_slave_state(slave); 1334 info->link_failure_count = slave->link_failure_count; 1335 } 1336 1337 static void bond_netdev_notify_work(struct work_struct *_work) 1338 { 1339 struct slave *slave = container_of(_work, struct slave, 1340 notify_work.work); 1341 1342 if (rtnl_trylock()) { 1343 struct netdev_bonding_info binfo; 1344 1345 bond_fill_ifslave(slave, &binfo.slave); 1346 bond_fill_ifbond(slave->bond, &binfo.master); 1347 netdev_bonding_info_change(slave->dev, &binfo); 1348 rtnl_unlock(); 1349 } else { 1350 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1351 } 1352 } 1353 1354 void bond_queue_slave_event(struct slave *slave) 1355 { 1356 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1357 } 1358 1359 void bond_lower_state_changed(struct slave *slave) 1360 { 1361 struct netdev_lag_lower_state_info info; 1362 1363 info.link_up = slave->link == BOND_LINK_UP || 1364 slave->link == BOND_LINK_FAIL; 1365 info.tx_enabled = bond_is_active_slave(slave); 1366 netdev_lower_state_changed(slave->dev, &info); 1367 } 1368 1369 /* enslave device <slave> to bond device <master> */ 1370 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1371 struct netlink_ext_ack *extack) 1372 { 1373 struct bonding *bond = netdev_priv(bond_dev); 1374 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1375 struct slave *new_slave = NULL, *prev_slave; 1376 struct sockaddr_storage ss; 1377 int link_reporting; 1378 int res = 0, i; 1379 1380 if (!bond->params.use_carrier && 1381 slave_dev->ethtool_ops->get_link == NULL && 1382 slave_ops->ndo_do_ioctl == NULL) { 1383 netdev_warn(bond_dev, "no link monitoring support for %s\n", 1384 slave_dev->name); 1385 } 1386 1387 /* already in-use? */ 1388 if (netdev_is_rx_handler_busy(slave_dev)) { 1389 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1390 netdev_err(bond_dev, 1391 "Error: Device is in use and cannot be enslaved\n"); 1392 return -EBUSY; 1393 } 1394 1395 if (bond_dev == slave_dev) { 1396 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1397 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1398 return -EPERM; 1399 } 1400 1401 /* vlan challenged mutual exclusion */ 1402 /* no need to lock since we're protected by rtnl_lock */ 1403 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1404 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n", 1405 slave_dev->name); 1406 if (vlan_uses_dev(bond_dev)) { 1407 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1408 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n", 1409 slave_dev->name, bond_dev->name); 1410 return -EPERM; 1411 } else { 1412 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n", 1413 slave_dev->name, slave_dev->name, 1414 bond_dev->name); 1415 } 1416 } else { 1417 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n", 1418 slave_dev->name); 1419 } 1420 1421 /* Old ifenslave binaries are no longer supported. These can 1422 * be identified with moderate accuracy by the state of the slave: 1423 * the current ifenslave will set the interface down prior to 1424 * enslaving it; the old ifenslave will not. 1425 */ 1426 if (slave_dev->flags & IFF_UP) { 1427 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1428 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n", 1429 slave_dev->name); 1430 return -EPERM; 1431 } 1432 1433 /* set bonding device ether type by slave - bonding netdevices are 1434 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1435 * there is a need to override some of the type dependent attribs/funcs. 1436 * 1437 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1438 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1439 */ 1440 if (!bond_has_slaves(bond)) { 1441 if (bond_dev->type != slave_dev->type) { 1442 netdev_dbg(bond_dev, "change device type from %d to %d\n", 1443 bond_dev->type, slave_dev->type); 1444 1445 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1446 bond_dev); 1447 res = notifier_to_errno(res); 1448 if (res) { 1449 netdev_err(bond_dev, "refused to change device type\n"); 1450 return -EBUSY; 1451 } 1452 1453 /* Flush unicast and multicast addresses */ 1454 dev_uc_flush(bond_dev); 1455 dev_mc_flush(bond_dev); 1456 1457 if (slave_dev->type != ARPHRD_ETHER) 1458 bond_setup_by_slave(bond_dev, slave_dev); 1459 else { 1460 ether_setup(bond_dev); 1461 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1462 } 1463 1464 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1465 bond_dev); 1466 } 1467 } else if (bond_dev->type != slave_dev->type) { 1468 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1469 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n", 1470 slave_dev->name, slave_dev->type, bond_dev->type); 1471 return -EINVAL; 1472 } 1473 1474 if (slave_dev->type == ARPHRD_INFINIBAND && 1475 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1476 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1477 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n", 1478 slave_dev->type); 1479 res = -EOPNOTSUPP; 1480 goto err_undo_flags; 1481 } 1482 1483 if (!slave_ops->ndo_set_mac_address || 1484 slave_dev->type == ARPHRD_INFINIBAND) { 1485 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n"); 1486 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1487 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1488 if (!bond_has_slaves(bond)) { 1489 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1490 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1491 } else { 1492 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1493 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"); 1494 res = -EOPNOTSUPP; 1495 goto err_undo_flags; 1496 } 1497 } 1498 } 1499 1500 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1501 1502 /* If this is the first slave, then we need to set the master's hardware 1503 * address to be the same as the slave's. 1504 */ 1505 if (!bond_has_slaves(bond) && 1506 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1507 res = bond_set_dev_addr(bond->dev, slave_dev); 1508 if (res) 1509 goto err_undo_flags; 1510 } 1511 1512 new_slave = bond_alloc_slave(bond); 1513 if (!new_slave) { 1514 res = -ENOMEM; 1515 goto err_undo_flags; 1516 } 1517 1518 new_slave->bond = bond; 1519 new_slave->dev = slave_dev; 1520 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1521 * is set via sysfs or module option if desired. 1522 */ 1523 new_slave->queue_id = 0; 1524 1525 /* Save slave's original mtu and then set it to match the bond */ 1526 new_slave->original_mtu = slave_dev->mtu; 1527 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1528 if (res) { 1529 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res); 1530 goto err_free; 1531 } 1532 1533 /* Save slave's original ("permanent") mac address for modes 1534 * that need it, and for restoring it upon release, and then 1535 * set it to the master's address 1536 */ 1537 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1538 slave_dev->addr_len); 1539 1540 if (!bond->params.fail_over_mac || 1541 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1542 /* Set slave to master's mac address. The application already 1543 * set the master's mac address to that of the first slave 1544 */ 1545 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1546 ss.ss_family = slave_dev->type; 1547 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1548 extack); 1549 if (res) { 1550 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res); 1551 goto err_restore_mtu; 1552 } 1553 } 1554 1555 /* set slave flag before open to prevent IPv6 addrconf */ 1556 slave_dev->flags |= IFF_SLAVE; 1557 1558 /* open the slave since the application closed it */ 1559 res = dev_open(slave_dev, extack); 1560 if (res) { 1561 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name); 1562 goto err_restore_mac; 1563 } 1564 1565 slave_dev->priv_flags |= IFF_BONDING; 1566 /* initialize slave stats */ 1567 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1568 1569 if (bond_is_lb(bond)) { 1570 /* bond_alb_init_slave() must be called before all other stages since 1571 * it might fail and we do not want to have to undo everything 1572 */ 1573 res = bond_alb_init_slave(bond, new_slave); 1574 if (res) 1575 goto err_close; 1576 } 1577 1578 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1579 if (res) { 1580 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n", 1581 slave_dev->name); 1582 goto err_close; 1583 } 1584 1585 prev_slave = bond_last_slave(bond); 1586 1587 new_slave->delay = 0; 1588 new_slave->link_failure_count = 0; 1589 1590 if (bond_update_speed_duplex(new_slave) && 1591 bond_needs_speed_duplex(bond)) 1592 new_slave->link = BOND_LINK_DOWN; 1593 1594 new_slave->last_rx = jiffies - 1595 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1596 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1597 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1598 1599 if (bond->params.miimon && !bond->params.use_carrier) { 1600 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1601 1602 if ((link_reporting == -1) && !bond->params.arp_interval) { 1603 /* miimon is set but a bonded network driver 1604 * does not support ETHTOOL/MII and 1605 * arp_interval is not set. Note: if 1606 * use_carrier is enabled, we will never go 1607 * here (because netif_carrier is always 1608 * supported); thus, we don't need to change 1609 * the messages for netif_carrier. 1610 */ 1611 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", 1612 slave_dev->name); 1613 } else if (link_reporting == -1) { 1614 /* unable get link status using mii/ethtool */ 1615 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", 1616 slave_dev->name); 1617 } 1618 } 1619 1620 /* check for initial state */ 1621 new_slave->link = BOND_LINK_NOCHANGE; 1622 if (bond->params.miimon) { 1623 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1624 if (bond->params.updelay) { 1625 bond_set_slave_link_state(new_slave, 1626 BOND_LINK_BACK, 1627 BOND_SLAVE_NOTIFY_NOW); 1628 new_slave->delay = bond->params.updelay; 1629 } else { 1630 bond_set_slave_link_state(new_slave, 1631 BOND_LINK_UP, 1632 BOND_SLAVE_NOTIFY_NOW); 1633 } 1634 } else { 1635 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1636 BOND_SLAVE_NOTIFY_NOW); 1637 } 1638 } else if (bond->params.arp_interval) { 1639 bond_set_slave_link_state(new_slave, 1640 (netif_carrier_ok(slave_dev) ? 1641 BOND_LINK_UP : BOND_LINK_DOWN), 1642 BOND_SLAVE_NOTIFY_NOW); 1643 } else { 1644 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1645 BOND_SLAVE_NOTIFY_NOW); 1646 } 1647 1648 if (new_slave->link != BOND_LINK_DOWN) 1649 new_slave->last_link_up = jiffies; 1650 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n", 1651 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1652 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1653 1654 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1655 /* if there is a primary slave, remember it */ 1656 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1657 rcu_assign_pointer(bond->primary_slave, new_slave); 1658 bond->force_primary = true; 1659 } 1660 } 1661 1662 switch (BOND_MODE(bond)) { 1663 case BOND_MODE_ACTIVEBACKUP: 1664 bond_set_slave_inactive_flags(new_slave, 1665 BOND_SLAVE_NOTIFY_NOW); 1666 break; 1667 case BOND_MODE_8023AD: 1668 /* in 802.3ad mode, the internal mechanism 1669 * will activate the slaves in the selected 1670 * aggregator 1671 */ 1672 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1673 /* if this is the first slave */ 1674 if (!prev_slave) { 1675 SLAVE_AD_INFO(new_slave)->id = 1; 1676 /* Initialize AD with the number of times that the AD timer is called in 1 second 1677 * can be called only after the mac address of the bond is set 1678 */ 1679 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1680 } else { 1681 SLAVE_AD_INFO(new_slave)->id = 1682 SLAVE_AD_INFO(prev_slave)->id + 1; 1683 } 1684 1685 bond_3ad_bind_slave(new_slave); 1686 break; 1687 case BOND_MODE_TLB: 1688 case BOND_MODE_ALB: 1689 bond_set_active_slave(new_slave); 1690 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1691 break; 1692 default: 1693 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n"); 1694 1695 /* always active in trunk mode */ 1696 bond_set_active_slave(new_slave); 1697 1698 /* In trunking mode there is little meaning to curr_active_slave 1699 * anyway (it holds no special properties of the bond device), 1700 * so we can change it without calling change_active_interface() 1701 */ 1702 if (!rcu_access_pointer(bond->curr_active_slave) && 1703 new_slave->link == BOND_LINK_UP) 1704 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1705 1706 break; 1707 } /* switch(bond_mode) */ 1708 1709 #ifdef CONFIG_NET_POLL_CONTROLLER 1710 if (bond->dev->npinfo) { 1711 if (slave_enable_netpoll(new_slave)) { 1712 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1713 res = -EBUSY; 1714 goto err_detach; 1715 } 1716 } 1717 #endif 1718 1719 if (!(bond_dev->features & NETIF_F_LRO)) 1720 dev_disable_lro(slave_dev); 1721 1722 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1723 new_slave); 1724 if (res) { 1725 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res); 1726 goto err_detach; 1727 } 1728 1729 res = bond_master_upper_dev_link(bond, new_slave, extack); 1730 if (res) { 1731 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1732 goto err_unregister; 1733 } 1734 1735 res = bond_sysfs_slave_add(new_slave); 1736 if (res) { 1737 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1738 goto err_upper_unlink; 1739 } 1740 1741 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1742 1743 /* If the mode uses primary, then the following is handled by 1744 * bond_change_active_slave(). 1745 */ 1746 if (!bond_uses_primary(bond)) { 1747 /* set promiscuity level to new slave */ 1748 if (bond_dev->flags & IFF_PROMISC) { 1749 res = dev_set_promiscuity(slave_dev, 1); 1750 if (res) 1751 goto err_sysfs_del; 1752 } 1753 1754 /* set allmulti level to new slave */ 1755 if (bond_dev->flags & IFF_ALLMULTI) { 1756 res = dev_set_allmulti(slave_dev, 1); 1757 if (res) { 1758 if (bond_dev->flags & IFF_PROMISC) 1759 dev_set_promiscuity(slave_dev, -1); 1760 goto err_sysfs_del; 1761 } 1762 } 1763 1764 netif_addr_lock_bh(bond_dev); 1765 dev_mc_sync_multiple(slave_dev, bond_dev); 1766 dev_uc_sync_multiple(slave_dev, bond_dev); 1767 netif_addr_unlock_bh(bond_dev); 1768 1769 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1770 /* add lacpdu mc addr to mc list */ 1771 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1772 1773 dev_mc_add(slave_dev, lacpdu_multicast); 1774 } 1775 } 1776 1777 bond->slave_cnt++; 1778 bond_compute_features(bond); 1779 bond_set_carrier(bond); 1780 1781 if (bond_uses_primary(bond)) { 1782 block_netpoll_tx(); 1783 bond_select_active_slave(bond); 1784 unblock_netpoll_tx(); 1785 } 1786 1787 if (bond_mode_can_use_xmit_hash(bond)) 1788 bond_update_slave_arr(bond, NULL); 1789 1790 1791 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n", 1792 slave_dev->name, 1793 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1794 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1795 1796 /* enslave is successful */ 1797 bond_queue_slave_event(new_slave); 1798 return 0; 1799 1800 /* Undo stages on error */ 1801 err_sysfs_del: 1802 bond_sysfs_slave_del(new_slave); 1803 1804 err_upper_unlink: 1805 bond_upper_dev_unlink(bond, new_slave); 1806 1807 err_unregister: 1808 netdev_rx_handler_unregister(slave_dev); 1809 1810 err_detach: 1811 vlan_vids_del_by_dev(slave_dev, bond_dev); 1812 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1813 RCU_INIT_POINTER(bond->primary_slave, NULL); 1814 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1815 block_netpoll_tx(); 1816 bond_change_active_slave(bond, NULL); 1817 bond_select_active_slave(bond); 1818 unblock_netpoll_tx(); 1819 } 1820 /* either primary_slave or curr_active_slave might've changed */ 1821 synchronize_rcu(); 1822 slave_disable_netpoll(new_slave); 1823 1824 err_close: 1825 slave_dev->priv_flags &= ~IFF_BONDING; 1826 dev_close(slave_dev); 1827 1828 err_restore_mac: 1829 slave_dev->flags &= ~IFF_SLAVE; 1830 if (!bond->params.fail_over_mac || 1831 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1832 /* XXX TODO - fom follow mode needs to change master's 1833 * MAC if this slave's MAC is in use by the bond, or at 1834 * least print a warning. 1835 */ 1836 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 1837 new_slave->dev->addr_len); 1838 ss.ss_family = slave_dev->type; 1839 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 1840 } 1841 1842 err_restore_mtu: 1843 dev_set_mtu(slave_dev, new_slave->original_mtu); 1844 1845 err_free: 1846 bond_free_slave(new_slave); 1847 1848 err_undo_flags: 1849 /* Enslave of first slave has failed and we need to fix master's mac */ 1850 if (!bond_has_slaves(bond)) { 1851 if (ether_addr_equal_64bits(bond_dev->dev_addr, 1852 slave_dev->dev_addr)) 1853 eth_hw_addr_random(bond_dev); 1854 if (bond_dev->type != ARPHRD_ETHER) { 1855 dev_close(bond_dev); 1856 ether_setup(bond_dev); 1857 bond_dev->flags |= IFF_MASTER; 1858 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1859 } 1860 } 1861 1862 return res; 1863 } 1864 1865 /* Try to release the slave device <slave> from the bond device <master> 1866 * It is legal to access curr_active_slave without a lock because all the function 1867 * is RTNL-locked. If "all" is true it means that the function is being called 1868 * while destroying a bond interface and all slaves are being released. 1869 * 1870 * The rules for slave state should be: 1871 * for Active/Backup: 1872 * Active stays on all backups go down 1873 * for Bonded connections: 1874 * The first up interface should be left on and all others downed. 1875 */ 1876 static int __bond_release_one(struct net_device *bond_dev, 1877 struct net_device *slave_dev, 1878 bool all, bool unregister) 1879 { 1880 struct bonding *bond = netdev_priv(bond_dev); 1881 struct slave *slave, *oldcurrent; 1882 struct sockaddr_storage ss; 1883 int old_flags = bond_dev->flags; 1884 netdev_features_t old_features = bond_dev->features; 1885 1886 /* slave is not a slave or master is not master of this slave */ 1887 if (!(slave_dev->flags & IFF_SLAVE) || 1888 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1889 netdev_dbg(bond_dev, "cannot release %s\n", 1890 slave_dev->name); 1891 return -EINVAL; 1892 } 1893 1894 block_netpoll_tx(); 1895 1896 slave = bond_get_slave_by_dev(bond, slave_dev); 1897 if (!slave) { 1898 /* not a slave of this bond */ 1899 netdev_info(bond_dev, "%s not enslaved\n", 1900 slave_dev->name); 1901 unblock_netpoll_tx(); 1902 return -EINVAL; 1903 } 1904 1905 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 1906 1907 bond_sysfs_slave_del(slave); 1908 1909 /* recompute stats just before removing the slave */ 1910 bond_get_stats(bond->dev, &bond->bond_stats); 1911 1912 bond_upper_dev_unlink(bond, slave); 1913 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1914 * for this slave anymore. 1915 */ 1916 netdev_rx_handler_unregister(slave_dev); 1917 1918 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1919 bond_3ad_unbind_slave(slave); 1920 1921 if (bond_mode_can_use_xmit_hash(bond)) 1922 bond_update_slave_arr(bond, slave); 1923 1924 netdev_info(bond_dev, "Releasing %s interface %s\n", 1925 bond_is_active_slave(slave) ? "active" : "backup", 1926 slave_dev->name); 1927 1928 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1929 1930 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1931 1932 if (!all && (!bond->params.fail_over_mac || 1933 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1934 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1935 bond_has_slaves(bond)) 1936 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", 1937 slave_dev->name, slave->perm_hwaddr, 1938 bond_dev->name, slave_dev->name); 1939 } 1940 1941 if (rtnl_dereference(bond->primary_slave) == slave) 1942 RCU_INIT_POINTER(bond->primary_slave, NULL); 1943 1944 if (oldcurrent == slave) 1945 bond_change_active_slave(bond, NULL); 1946 1947 if (bond_is_lb(bond)) { 1948 /* Must be called only after the slave has been 1949 * detached from the list and the curr_active_slave 1950 * has been cleared (if our_slave == old_current), 1951 * but before a new active slave is selected. 1952 */ 1953 bond_alb_deinit_slave(bond, slave); 1954 } 1955 1956 if (all) { 1957 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1958 } else if (oldcurrent == slave) { 1959 /* Note that we hold RTNL over this sequence, so there 1960 * is no concern that another slave add/remove event 1961 * will interfere. 1962 */ 1963 bond_select_active_slave(bond); 1964 } 1965 1966 if (!bond_has_slaves(bond)) { 1967 bond_set_carrier(bond); 1968 eth_hw_addr_random(bond_dev); 1969 bond->nest_level = SINGLE_DEPTH_NESTING; 1970 } else { 1971 bond->nest_level = dev_get_nest_level(bond_dev) + 1; 1972 } 1973 1974 unblock_netpoll_tx(); 1975 synchronize_rcu(); 1976 bond->slave_cnt--; 1977 1978 if (!bond_has_slaves(bond)) { 1979 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 1980 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 1981 } 1982 1983 bond_compute_features(bond); 1984 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 1985 (old_features & NETIF_F_VLAN_CHALLENGED)) 1986 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n", 1987 slave_dev->name, bond_dev->name); 1988 1989 vlan_vids_del_by_dev(slave_dev, bond_dev); 1990 1991 /* If the mode uses primary, then this case was handled above by 1992 * bond_change_active_slave(..., NULL) 1993 */ 1994 if (!bond_uses_primary(bond)) { 1995 /* unset promiscuity level from slave 1996 * NOTE: The NETDEV_CHANGEADDR call above may change the value 1997 * of the IFF_PROMISC flag in the bond_dev, but we need the 1998 * value of that flag before that change, as that was the value 1999 * when this slave was attached, so we cache at the start of the 2000 * function and use it here. Same goes for ALLMULTI below 2001 */ 2002 if (old_flags & IFF_PROMISC) 2003 dev_set_promiscuity(slave_dev, -1); 2004 2005 /* unset allmulti level from slave */ 2006 if (old_flags & IFF_ALLMULTI) 2007 dev_set_allmulti(slave_dev, -1); 2008 2009 bond_hw_addr_flush(bond_dev, slave_dev); 2010 } 2011 2012 slave_disable_netpoll(slave); 2013 2014 /* close slave before restoring its mac address */ 2015 dev_close(slave_dev); 2016 2017 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2018 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2019 /* restore original ("permanent") mac address */ 2020 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2021 slave->dev->addr_len); 2022 ss.ss_family = slave_dev->type; 2023 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2024 } 2025 2026 if (unregister) 2027 __dev_set_mtu(slave_dev, slave->original_mtu); 2028 else 2029 dev_set_mtu(slave_dev, slave->original_mtu); 2030 2031 slave_dev->priv_flags &= ~IFF_BONDING; 2032 2033 bond_free_slave(slave); 2034 2035 return 0; 2036 } 2037 2038 /* A wrapper used because of ndo_del_link */ 2039 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2040 { 2041 return __bond_release_one(bond_dev, slave_dev, false, false); 2042 } 2043 2044 /* First release a slave and then destroy the bond if no more slaves are left. 2045 * Must be under rtnl_lock when this function is called. 2046 */ 2047 static int bond_release_and_destroy(struct net_device *bond_dev, 2048 struct net_device *slave_dev) 2049 { 2050 struct bonding *bond = netdev_priv(bond_dev); 2051 int ret; 2052 2053 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2054 if (ret == 0 && !bond_has_slaves(bond)) { 2055 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2056 netdev_info(bond_dev, "Destroying bond %s\n", 2057 bond_dev->name); 2058 bond_remove_proc_entry(bond); 2059 unregister_netdevice(bond_dev); 2060 } 2061 return ret; 2062 } 2063 2064 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2065 { 2066 struct bonding *bond = netdev_priv(bond_dev); 2067 bond_fill_ifbond(bond, info); 2068 } 2069 2070 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2071 { 2072 struct bonding *bond = netdev_priv(bond_dev); 2073 struct list_head *iter; 2074 int i = 0, res = -ENODEV; 2075 struct slave *slave; 2076 2077 bond_for_each_slave(bond, slave, iter) { 2078 if (i++ == (int)info->slave_id) { 2079 res = 0; 2080 bond_fill_ifslave(slave, info); 2081 break; 2082 } 2083 } 2084 2085 return res; 2086 } 2087 2088 /*-------------------------------- Monitoring -------------------------------*/ 2089 2090 /* called with rcu_read_lock() */ 2091 static int bond_miimon_inspect(struct bonding *bond) 2092 { 2093 int link_state, commit = 0; 2094 struct list_head *iter; 2095 struct slave *slave; 2096 bool ignore_updelay; 2097 2098 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2099 2100 bond_for_each_slave_rcu(bond, slave, iter) { 2101 slave->new_link = BOND_LINK_NOCHANGE; 2102 slave->link_new_state = slave->link; 2103 2104 link_state = bond_check_dev_link(bond, slave->dev, 0); 2105 2106 switch (slave->link) { 2107 case BOND_LINK_UP: 2108 if (link_state) 2109 continue; 2110 2111 bond_propose_link_state(slave, BOND_LINK_FAIL); 2112 commit++; 2113 slave->delay = bond->params.downdelay; 2114 if (slave->delay) { 2115 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n", 2116 (BOND_MODE(bond) == 2117 BOND_MODE_ACTIVEBACKUP) ? 2118 (bond_is_active_slave(slave) ? 2119 "active " : "backup ") : "", 2120 slave->dev->name, 2121 bond->params.downdelay * bond->params.miimon); 2122 } 2123 /*FALLTHRU*/ 2124 case BOND_LINK_FAIL: 2125 if (link_state) { 2126 /* recovered before downdelay expired */ 2127 bond_propose_link_state(slave, BOND_LINK_UP); 2128 slave->last_link_up = jiffies; 2129 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n", 2130 (bond->params.downdelay - slave->delay) * 2131 bond->params.miimon, 2132 slave->dev->name); 2133 commit++; 2134 continue; 2135 } 2136 2137 if (slave->delay <= 0) { 2138 slave->new_link = BOND_LINK_DOWN; 2139 commit++; 2140 continue; 2141 } 2142 2143 slave->delay--; 2144 break; 2145 2146 case BOND_LINK_DOWN: 2147 if (!link_state) 2148 continue; 2149 2150 bond_propose_link_state(slave, BOND_LINK_BACK); 2151 commit++; 2152 slave->delay = bond->params.updelay; 2153 2154 if (slave->delay) { 2155 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n", 2156 slave->dev->name, 2157 ignore_updelay ? 0 : 2158 bond->params.updelay * 2159 bond->params.miimon); 2160 } 2161 /*FALLTHRU*/ 2162 case BOND_LINK_BACK: 2163 if (!link_state) { 2164 bond_propose_link_state(slave, BOND_LINK_DOWN); 2165 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n", 2166 (bond->params.updelay - slave->delay) * 2167 bond->params.miimon, 2168 slave->dev->name); 2169 commit++; 2170 continue; 2171 } 2172 2173 if (ignore_updelay) 2174 slave->delay = 0; 2175 2176 if (slave->delay <= 0) { 2177 slave->new_link = BOND_LINK_UP; 2178 commit++; 2179 ignore_updelay = false; 2180 continue; 2181 } 2182 2183 slave->delay--; 2184 break; 2185 } 2186 } 2187 2188 return commit; 2189 } 2190 2191 static void bond_miimon_link_change(struct bonding *bond, 2192 struct slave *slave, 2193 char link) 2194 { 2195 switch (BOND_MODE(bond)) { 2196 case BOND_MODE_8023AD: 2197 bond_3ad_handle_link_change(slave, link); 2198 break; 2199 case BOND_MODE_TLB: 2200 case BOND_MODE_ALB: 2201 bond_alb_handle_link_change(bond, slave, link); 2202 break; 2203 case BOND_MODE_XOR: 2204 bond_update_slave_arr(bond, NULL); 2205 break; 2206 } 2207 } 2208 2209 static void bond_miimon_commit(struct bonding *bond) 2210 { 2211 struct list_head *iter; 2212 struct slave *slave, *primary; 2213 2214 bond_for_each_slave(bond, slave, iter) { 2215 switch (slave->new_link) { 2216 case BOND_LINK_NOCHANGE: 2217 continue; 2218 2219 case BOND_LINK_UP: 2220 if (bond_update_speed_duplex(slave) && 2221 bond_needs_speed_duplex(bond)) { 2222 slave->link = BOND_LINK_DOWN; 2223 if (net_ratelimit()) 2224 netdev_warn(bond->dev, 2225 "failed to get link speed/duplex for %s\n", 2226 slave->dev->name); 2227 continue; 2228 } 2229 bond_set_slave_link_state(slave, BOND_LINK_UP, 2230 BOND_SLAVE_NOTIFY_NOW); 2231 slave->last_link_up = jiffies; 2232 2233 primary = rtnl_dereference(bond->primary_slave); 2234 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2235 /* prevent it from being the active one */ 2236 bond_set_backup_slave(slave); 2237 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2238 /* make it immediately active */ 2239 bond_set_active_slave(slave); 2240 } else if (slave != primary) { 2241 /* prevent it from being the active one */ 2242 bond_set_backup_slave(slave); 2243 } 2244 2245 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n", 2246 slave->dev->name, 2247 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2248 slave->duplex ? "full" : "half"); 2249 2250 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2251 2252 if (!bond->curr_active_slave || slave == primary) 2253 goto do_failover; 2254 2255 continue; 2256 2257 case BOND_LINK_DOWN: 2258 if (slave->link_failure_count < UINT_MAX) 2259 slave->link_failure_count++; 2260 2261 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2262 BOND_SLAVE_NOTIFY_NOW); 2263 2264 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2265 BOND_MODE(bond) == BOND_MODE_8023AD) 2266 bond_set_slave_inactive_flags(slave, 2267 BOND_SLAVE_NOTIFY_NOW); 2268 2269 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2270 slave->dev->name); 2271 2272 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2273 2274 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2275 goto do_failover; 2276 2277 continue; 2278 2279 default: 2280 netdev_err(bond->dev, "invalid new link %d on slave %s\n", 2281 slave->new_link, slave->dev->name); 2282 slave->new_link = BOND_LINK_NOCHANGE; 2283 2284 continue; 2285 } 2286 2287 do_failover: 2288 block_netpoll_tx(); 2289 bond_select_active_slave(bond); 2290 unblock_netpoll_tx(); 2291 } 2292 2293 bond_set_carrier(bond); 2294 } 2295 2296 /* bond_mii_monitor 2297 * 2298 * Really a wrapper that splits the mii monitor into two phases: an 2299 * inspection, then (if inspection indicates something needs to be done) 2300 * an acquisition of appropriate locks followed by a commit phase to 2301 * implement whatever link state changes are indicated. 2302 */ 2303 static void bond_mii_monitor(struct work_struct *work) 2304 { 2305 struct bonding *bond = container_of(work, struct bonding, 2306 mii_work.work); 2307 bool should_notify_peers = false; 2308 unsigned long delay; 2309 struct slave *slave; 2310 struct list_head *iter; 2311 2312 delay = msecs_to_jiffies(bond->params.miimon); 2313 2314 if (!bond_has_slaves(bond)) 2315 goto re_arm; 2316 2317 rcu_read_lock(); 2318 2319 should_notify_peers = bond_should_notify_peers(bond); 2320 2321 if (bond_miimon_inspect(bond)) { 2322 rcu_read_unlock(); 2323 2324 /* Race avoidance with bond_close cancel of workqueue */ 2325 if (!rtnl_trylock()) { 2326 delay = 1; 2327 should_notify_peers = false; 2328 goto re_arm; 2329 } 2330 2331 bond_for_each_slave(bond, slave, iter) { 2332 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2333 } 2334 bond_miimon_commit(bond); 2335 2336 rtnl_unlock(); /* might sleep, hold no other locks */ 2337 } else 2338 rcu_read_unlock(); 2339 2340 re_arm: 2341 if (bond->params.miimon) 2342 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2343 2344 if (should_notify_peers) { 2345 if (!rtnl_trylock()) 2346 return; 2347 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2348 rtnl_unlock(); 2349 } 2350 } 2351 2352 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2353 { 2354 __be32 ip = *((__be32 *)data); 2355 2356 return ip == bond_confirm_addr(upper, 0, ip); 2357 } 2358 2359 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2360 { 2361 bool ret = false; 2362 2363 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2364 return true; 2365 2366 rcu_read_lock(); 2367 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2368 ret = true; 2369 rcu_read_unlock(); 2370 2371 return ret; 2372 } 2373 2374 /* We go to the (large) trouble of VLAN tagging ARP frames because 2375 * switches in VLAN mode (especially if ports are configured as 2376 * "native" to a VLAN) might not pass non-tagged frames. 2377 */ 2378 static void bond_arp_send(struct net_device *slave_dev, int arp_op, 2379 __be32 dest_ip, __be32 src_ip, 2380 struct bond_vlan_tag *tags) 2381 { 2382 struct sk_buff *skb; 2383 struct bond_vlan_tag *outer_tag = tags; 2384 2385 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n", 2386 arp_op, slave_dev->name, &dest_ip, &src_ip); 2387 2388 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2389 NULL, slave_dev->dev_addr, NULL); 2390 2391 if (!skb) { 2392 net_err_ratelimited("ARP packet allocation failed\n"); 2393 return; 2394 } 2395 2396 if (!tags || tags->vlan_proto == VLAN_N_VID) 2397 goto xmit; 2398 2399 tags++; 2400 2401 /* Go through all the tags backwards and add them to the packet */ 2402 while (tags->vlan_proto != VLAN_N_VID) { 2403 if (!tags->vlan_id) { 2404 tags++; 2405 continue; 2406 } 2407 2408 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n", 2409 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2410 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2411 tags->vlan_id); 2412 if (!skb) { 2413 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2414 return; 2415 } 2416 2417 tags++; 2418 } 2419 /* Set the outer tag */ 2420 if (outer_tag->vlan_id) { 2421 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n", 2422 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2423 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2424 outer_tag->vlan_id); 2425 } 2426 2427 xmit: 2428 arp_xmit(skb); 2429 } 2430 2431 /* Validate the device path between the @start_dev and the @end_dev. 2432 * The path is valid if the @end_dev is reachable through device 2433 * stacking. 2434 * When the path is validated, collect any vlan information in the 2435 * path. 2436 */ 2437 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2438 struct net_device *end_dev, 2439 int level) 2440 { 2441 struct bond_vlan_tag *tags; 2442 struct net_device *upper; 2443 struct list_head *iter; 2444 2445 if (start_dev == end_dev) { 2446 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2447 if (!tags) 2448 return ERR_PTR(-ENOMEM); 2449 tags[level].vlan_proto = VLAN_N_VID; 2450 return tags; 2451 } 2452 2453 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2454 tags = bond_verify_device_path(upper, end_dev, level + 1); 2455 if (IS_ERR_OR_NULL(tags)) { 2456 if (IS_ERR(tags)) 2457 return tags; 2458 continue; 2459 } 2460 if (is_vlan_dev(upper)) { 2461 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2462 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2463 } 2464 2465 return tags; 2466 } 2467 2468 return NULL; 2469 } 2470 2471 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2472 { 2473 struct rtable *rt; 2474 struct bond_vlan_tag *tags; 2475 __be32 *targets = bond->params.arp_targets, addr; 2476 int i; 2477 2478 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2479 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]); 2480 tags = NULL; 2481 2482 /* Find out through which dev should the packet go */ 2483 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2484 RTO_ONLINK, 0); 2485 if (IS_ERR(rt)) { 2486 /* there's no route to target - try to send arp 2487 * probe to generate any traffic (arp_validate=0) 2488 */ 2489 if (bond->params.arp_validate) 2490 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2491 bond->dev->name, 2492 &targets[i]); 2493 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2494 0, tags); 2495 continue; 2496 } 2497 2498 /* bond device itself */ 2499 if (rt->dst.dev == bond->dev) 2500 goto found; 2501 2502 rcu_read_lock(); 2503 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2504 rcu_read_unlock(); 2505 2506 if (!IS_ERR_OR_NULL(tags)) 2507 goto found; 2508 2509 /* Not our device - skip */ 2510 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2511 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2512 2513 ip_rt_put(rt); 2514 continue; 2515 2516 found: 2517 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2518 ip_rt_put(rt); 2519 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 2520 addr, tags); 2521 kfree(tags); 2522 } 2523 } 2524 2525 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2526 { 2527 int i; 2528 2529 if (!sip || !bond_has_this_ip(bond, tip)) { 2530 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n", 2531 &sip, &tip); 2532 return; 2533 } 2534 2535 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2536 if (i == -1) { 2537 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n", 2538 &sip); 2539 return; 2540 } 2541 slave->last_rx = jiffies; 2542 slave->target_last_arp_rx[i] = jiffies; 2543 } 2544 2545 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2546 struct slave *slave) 2547 { 2548 struct arphdr *arp = (struct arphdr *)skb->data; 2549 struct slave *curr_active_slave, *curr_arp_slave; 2550 unsigned char *arp_ptr; 2551 __be32 sip, tip; 2552 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2553 unsigned int alen; 2554 2555 if (!slave_do_arp_validate(bond, slave)) { 2556 if ((slave_do_arp_validate_only(bond) && is_arp) || 2557 !slave_do_arp_validate_only(bond)) 2558 slave->last_rx = jiffies; 2559 return RX_HANDLER_ANOTHER; 2560 } else if (!is_arp) { 2561 return RX_HANDLER_ANOTHER; 2562 } 2563 2564 alen = arp_hdr_len(bond->dev); 2565 2566 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n", 2567 skb->dev->name); 2568 2569 if (alen > skb_headlen(skb)) { 2570 arp = kmalloc(alen, GFP_ATOMIC); 2571 if (!arp) 2572 goto out_unlock; 2573 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2574 goto out_unlock; 2575 } 2576 2577 if (arp->ar_hln != bond->dev->addr_len || 2578 skb->pkt_type == PACKET_OTHERHOST || 2579 skb->pkt_type == PACKET_LOOPBACK || 2580 arp->ar_hrd != htons(ARPHRD_ETHER) || 2581 arp->ar_pro != htons(ETH_P_IP) || 2582 arp->ar_pln != 4) 2583 goto out_unlock; 2584 2585 arp_ptr = (unsigned char *)(arp + 1); 2586 arp_ptr += bond->dev->addr_len; 2587 memcpy(&sip, arp_ptr, 4); 2588 arp_ptr += 4 + bond->dev->addr_len; 2589 memcpy(&tip, arp_ptr, 4); 2590 2591 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2592 slave->dev->name, bond_slave_state(slave), 2593 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2594 &sip, &tip); 2595 2596 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2597 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2598 2599 /* We 'trust' the received ARP enough to validate it if: 2600 * 2601 * (a) the slave receiving the ARP is active (which includes the 2602 * current ARP slave, if any), or 2603 * 2604 * (b) the receiving slave isn't active, but there is a currently 2605 * active slave and it received valid arp reply(s) after it became 2606 * the currently active slave, or 2607 * 2608 * (c) there is an ARP slave that sent an ARP during the prior ARP 2609 * interval, and we receive an ARP reply on any slave. We accept 2610 * these because switch FDB update delays may deliver the ARP 2611 * reply to a slave other than the sender of the ARP request. 2612 * 2613 * Note: for (b), backup slaves are receiving the broadcast ARP 2614 * request, not a reply. This request passes from the sending 2615 * slave through the L2 switch(es) to the receiving slave. Since 2616 * this is checking the request, sip/tip are swapped for 2617 * validation. 2618 * 2619 * This is done to avoid endless looping when we can't reach the 2620 * arp_ip_target and fool ourselves with our own arp requests. 2621 */ 2622 if (bond_is_active_slave(slave)) 2623 bond_validate_arp(bond, slave, sip, tip); 2624 else if (curr_active_slave && 2625 time_after(slave_last_rx(bond, curr_active_slave), 2626 curr_active_slave->last_link_up)) 2627 bond_validate_arp(bond, slave, tip, sip); 2628 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2629 bond_time_in_interval(bond, 2630 dev_trans_start(curr_arp_slave->dev), 1)) 2631 bond_validate_arp(bond, slave, sip, tip); 2632 2633 out_unlock: 2634 if (arp != (struct arphdr *)skb->data) 2635 kfree(arp); 2636 return RX_HANDLER_ANOTHER; 2637 } 2638 2639 /* function to verify if we're in the arp_interval timeslice, returns true if 2640 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2641 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2642 */ 2643 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2644 int mod) 2645 { 2646 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2647 2648 return time_in_range(jiffies, 2649 last_act - delta_in_ticks, 2650 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2651 } 2652 2653 /* This function is called regularly to monitor each slave's link 2654 * ensuring that traffic is being sent and received when arp monitoring 2655 * is used in load-balancing mode. if the adapter has been dormant, then an 2656 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2657 * arp monitoring in active backup mode. 2658 */ 2659 static void bond_loadbalance_arp_mon(struct bonding *bond) 2660 { 2661 struct slave *slave, *oldcurrent; 2662 struct list_head *iter; 2663 int do_failover = 0, slave_state_changed = 0; 2664 2665 if (!bond_has_slaves(bond)) 2666 goto re_arm; 2667 2668 rcu_read_lock(); 2669 2670 oldcurrent = rcu_dereference(bond->curr_active_slave); 2671 /* see if any of the previous devices are up now (i.e. they have 2672 * xmt and rcv traffic). the curr_active_slave does not come into 2673 * the picture unless it is null. also, slave->last_link_up is not 2674 * needed here because we send an arp on each slave and give a slave 2675 * as long as it needs to get the tx/rx within the delta. 2676 * TODO: what about up/down delay in arp mode? it wasn't here before 2677 * so it can wait 2678 */ 2679 bond_for_each_slave_rcu(bond, slave, iter) { 2680 unsigned long trans_start = dev_trans_start(slave->dev); 2681 2682 slave->new_link = BOND_LINK_NOCHANGE; 2683 2684 if (slave->link != BOND_LINK_UP) { 2685 if (bond_time_in_interval(bond, trans_start, 1) && 2686 bond_time_in_interval(bond, slave->last_rx, 1)) { 2687 2688 slave->new_link = BOND_LINK_UP; 2689 slave_state_changed = 1; 2690 2691 /* primary_slave has no meaning in round-robin 2692 * mode. the window of a slave being up and 2693 * curr_active_slave being null after enslaving 2694 * is closed. 2695 */ 2696 if (!oldcurrent) { 2697 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2698 slave->dev->name); 2699 do_failover = 1; 2700 } else { 2701 netdev_info(bond->dev, "interface %s is now up\n", 2702 slave->dev->name); 2703 } 2704 } 2705 } else { 2706 /* slave->link == BOND_LINK_UP */ 2707 2708 /* not all switches will respond to an arp request 2709 * when the source ip is 0, so don't take the link down 2710 * if we don't know our ip yet 2711 */ 2712 if (!bond_time_in_interval(bond, trans_start, 2) || 2713 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2714 2715 slave->new_link = BOND_LINK_DOWN; 2716 slave_state_changed = 1; 2717 2718 if (slave->link_failure_count < UINT_MAX) 2719 slave->link_failure_count++; 2720 2721 netdev_info(bond->dev, "interface %s is now down\n", 2722 slave->dev->name); 2723 2724 if (slave == oldcurrent) 2725 do_failover = 1; 2726 } 2727 } 2728 2729 /* note: if switch is in round-robin mode, all links 2730 * must tx arp to ensure all links rx an arp - otherwise 2731 * links may oscillate or not come up at all; if switch is 2732 * in something like xor mode, there is nothing we can 2733 * do - all replies will be rx'ed on same link causing slaves 2734 * to be unstable during low/no traffic periods 2735 */ 2736 if (bond_slave_is_up(slave)) 2737 bond_arp_send_all(bond, slave); 2738 } 2739 2740 rcu_read_unlock(); 2741 2742 if (do_failover || slave_state_changed) { 2743 if (!rtnl_trylock()) 2744 goto re_arm; 2745 2746 bond_for_each_slave(bond, slave, iter) { 2747 if (slave->new_link != BOND_LINK_NOCHANGE) 2748 slave->link = slave->new_link; 2749 } 2750 2751 if (slave_state_changed) { 2752 bond_slave_state_change(bond); 2753 if (BOND_MODE(bond) == BOND_MODE_XOR) 2754 bond_update_slave_arr(bond, NULL); 2755 } 2756 if (do_failover) { 2757 block_netpoll_tx(); 2758 bond_select_active_slave(bond); 2759 unblock_netpoll_tx(); 2760 } 2761 rtnl_unlock(); 2762 } 2763 2764 re_arm: 2765 if (bond->params.arp_interval) 2766 queue_delayed_work(bond->wq, &bond->arp_work, 2767 msecs_to_jiffies(bond->params.arp_interval)); 2768 } 2769 2770 /* Called to inspect slaves for active-backup mode ARP monitor link state 2771 * changes. Sets new_link in slaves to specify what action should take 2772 * place for the slave. Returns 0 if no changes are found, >0 if changes 2773 * to link states must be committed. 2774 * 2775 * Called with rcu_read_lock held. 2776 */ 2777 static int bond_ab_arp_inspect(struct bonding *bond) 2778 { 2779 unsigned long trans_start, last_rx; 2780 struct list_head *iter; 2781 struct slave *slave; 2782 int commit = 0; 2783 2784 bond_for_each_slave_rcu(bond, slave, iter) { 2785 slave->new_link = BOND_LINK_NOCHANGE; 2786 last_rx = slave_last_rx(bond, slave); 2787 2788 if (slave->link != BOND_LINK_UP) { 2789 if (bond_time_in_interval(bond, last_rx, 1)) { 2790 slave->new_link = BOND_LINK_UP; 2791 commit++; 2792 } 2793 continue; 2794 } 2795 2796 /* Give slaves 2*delta after being enslaved or made 2797 * active. This avoids bouncing, as the last receive 2798 * times need a full ARP monitor cycle to be updated. 2799 */ 2800 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2801 continue; 2802 2803 /* Backup slave is down if: 2804 * - No current_arp_slave AND 2805 * - more than 3*delta since last receive AND 2806 * - the bond has an IP address 2807 * 2808 * Note: a non-null current_arp_slave indicates 2809 * the curr_active_slave went down and we are 2810 * searching for a new one; under this condition 2811 * we only take the curr_active_slave down - this 2812 * gives each slave a chance to tx/rx traffic 2813 * before being taken out 2814 */ 2815 if (!bond_is_active_slave(slave) && 2816 !rcu_access_pointer(bond->current_arp_slave) && 2817 !bond_time_in_interval(bond, last_rx, 3)) { 2818 slave->new_link = BOND_LINK_DOWN; 2819 commit++; 2820 } 2821 2822 /* Active slave is down if: 2823 * - more than 2*delta since transmitting OR 2824 * - (more than 2*delta since receive AND 2825 * the bond has an IP address) 2826 */ 2827 trans_start = dev_trans_start(slave->dev); 2828 if (bond_is_active_slave(slave) && 2829 (!bond_time_in_interval(bond, trans_start, 2) || 2830 !bond_time_in_interval(bond, last_rx, 2))) { 2831 slave->new_link = BOND_LINK_DOWN; 2832 commit++; 2833 } 2834 } 2835 2836 return commit; 2837 } 2838 2839 /* Called to commit link state changes noted by inspection step of 2840 * active-backup mode ARP monitor. 2841 * 2842 * Called with RTNL hold. 2843 */ 2844 static void bond_ab_arp_commit(struct bonding *bond) 2845 { 2846 unsigned long trans_start; 2847 struct list_head *iter; 2848 struct slave *slave; 2849 2850 bond_for_each_slave(bond, slave, iter) { 2851 switch (slave->new_link) { 2852 case BOND_LINK_NOCHANGE: 2853 continue; 2854 2855 case BOND_LINK_UP: 2856 trans_start = dev_trans_start(slave->dev); 2857 if (rtnl_dereference(bond->curr_active_slave) != slave || 2858 (!rtnl_dereference(bond->curr_active_slave) && 2859 bond_time_in_interval(bond, trans_start, 1))) { 2860 struct slave *current_arp_slave; 2861 2862 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2863 bond_set_slave_link_state(slave, BOND_LINK_UP, 2864 BOND_SLAVE_NOTIFY_NOW); 2865 if (current_arp_slave) { 2866 bond_set_slave_inactive_flags( 2867 current_arp_slave, 2868 BOND_SLAVE_NOTIFY_NOW); 2869 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2870 } 2871 2872 netdev_info(bond->dev, "link status definitely up for interface %s\n", 2873 slave->dev->name); 2874 2875 if (!rtnl_dereference(bond->curr_active_slave) || 2876 slave == rtnl_dereference(bond->primary_slave)) 2877 goto do_failover; 2878 2879 } 2880 2881 continue; 2882 2883 case BOND_LINK_DOWN: 2884 if (slave->link_failure_count < UINT_MAX) 2885 slave->link_failure_count++; 2886 2887 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2888 BOND_SLAVE_NOTIFY_NOW); 2889 bond_set_slave_inactive_flags(slave, 2890 BOND_SLAVE_NOTIFY_NOW); 2891 2892 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n", 2893 slave->dev->name); 2894 2895 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2896 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2897 goto do_failover; 2898 } 2899 2900 continue; 2901 2902 default: 2903 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n", 2904 slave->new_link, slave->dev->name); 2905 continue; 2906 } 2907 2908 do_failover: 2909 block_netpoll_tx(); 2910 bond_select_active_slave(bond); 2911 unblock_netpoll_tx(); 2912 } 2913 2914 bond_set_carrier(bond); 2915 } 2916 2917 /* Send ARP probes for active-backup mode ARP monitor. 2918 * 2919 * Called with rcu_read_lock held. 2920 */ 2921 static bool bond_ab_arp_probe(struct bonding *bond) 2922 { 2923 struct slave *slave, *before = NULL, *new_slave = NULL, 2924 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2925 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2926 struct list_head *iter; 2927 bool found = false; 2928 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2929 2930 if (curr_arp_slave && curr_active_slave) 2931 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2932 curr_arp_slave->dev->name, 2933 curr_active_slave->dev->name); 2934 2935 if (curr_active_slave) { 2936 bond_arp_send_all(bond, curr_active_slave); 2937 return should_notify_rtnl; 2938 } 2939 2940 /* if we don't have a curr_active_slave, search for the next available 2941 * backup slave from the current_arp_slave and make it the candidate 2942 * for becoming the curr_active_slave 2943 */ 2944 2945 if (!curr_arp_slave) { 2946 curr_arp_slave = bond_first_slave_rcu(bond); 2947 if (!curr_arp_slave) 2948 return should_notify_rtnl; 2949 } 2950 2951 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2952 2953 bond_for_each_slave_rcu(bond, slave, iter) { 2954 if (!found && !before && bond_slave_is_up(slave)) 2955 before = slave; 2956 2957 if (found && !new_slave && bond_slave_is_up(slave)) 2958 new_slave = slave; 2959 /* if the link state is up at this point, we 2960 * mark it down - this can happen if we have 2961 * simultaneous link failures and 2962 * reselect_active_interface doesn't make this 2963 * one the current slave so it is still marked 2964 * up when it is actually down 2965 */ 2966 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 2967 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2968 BOND_SLAVE_NOTIFY_LATER); 2969 if (slave->link_failure_count < UINT_MAX) 2970 slave->link_failure_count++; 2971 2972 bond_set_slave_inactive_flags(slave, 2973 BOND_SLAVE_NOTIFY_LATER); 2974 2975 netdev_info(bond->dev, "backup interface %s is now down\n", 2976 slave->dev->name); 2977 } 2978 if (slave == curr_arp_slave) 2979 found = true; 2980 } 2981 2982 if (!new_slave && before) 2983 new_slave = before; 2984 2985 if (!new_slave) 2986 goto check_state; 2987 2988 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 2989 BOND_SLAVE_NOTIFY_LATER); 2990 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 2991 bond_arp_send_all(bond, new_slave); 2992 new_slave->last_link_up = jiffies; 2993 rcu_assign_pointer(bond->current_arp_slave, new_slave); 2994 2995 check_state: 2996 bond_for_each_slave_rcu(bond, slave, iter) { 2997 if (slave->should_notify || slave->should_notify_link) { 2998 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 2999 break; 3000 } 3001 } 3002 return should_notify_rtnl; 3003 } 3004 3005 static void bond_activebackup_arp_mon(struct bonding *bond) 3006 { 3007 bool should_notify_peers = false; 3008 bool should_notify_rtnl = false; 3009 int delta_in_ticks; 3010 3011 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3012 3013 if (!bond_has_slaves(bond)) 3014 goto re_arm; 3015 3016 rcu_read_lock(); 3017 3018 should_notify_peers = bond_should_notify_peers(bond); 3019 3020 if (bond_ab_arp_inspect(bond)) { 3021 rcu_read_unlock(); 3022 3023 /* Race avoidance with bond_close flush of workqueue */ 3024 if (!rtnl_trylock()) { 3025 delta_in_ticks = 1; 3026 should_notify_peers = false; 3027 goto re_arm; 3028 } 3029 3030 bond_ab_arp_commit(bond); 3031 3032 rtnl_unlock(); 3033 rcu_read_lock(); 3034 } 3035 3036 should_notify_rtnl = bond_ab_arp_probe(bond); 3037 rcu_read_unlock(); 3038 3039 re_arm: 3040 if (bond->params.arp_interval) 3041 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3042 3043 if (should_notify_peers || should_notify_rtnl) { 3044 if (!rtnl_trylock()) 3045 return; 3046 3047 if (should_notify_peers) 3048 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3049 bond->dev); 3050 if (should_notify_rtnl) { 3051 bond_slave_state_notify(bond); 3052 bond_slave_link_notify(bond); 3053 } 3054 3055 rtnl_unlock(); 3056 } 3057 } 3058 3059 static void bond_arp_monitor(struct work_struct *work) 3060 { 3061 struct bonding *bond = container_of(work, struct bonding, 3062 arp_work.work); 3063 3064 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3065 bond_activebackup_arp_mon(bond); 3066 else 3067 bond_loadbalance_arp_mon(bond); 3068 } 3069 3070 /*-------------------------- netdev event handling --------------------------*/ 3071 3072 /* Change device name */ 3073 static int bond_event_changename(struct bonding *bond) 3074 { 3075 bond_remove_proc_entry(bond); 3076 bond_create_proc_entry(bond); 3077 3078 bond_debug_reregister(bond); 3079 3080 return NOTIFY_DONE; 3081 } 3082 3083 static int bond_master_netdev_event(unsigned long event, 3084 struct net_device *bond_dev) 3085 { 3086 struct bonding *event_bond = netdev_priv(bond_dev); 3087 3088 switch (event) { 3089 case NETDEV_CHANGENAME: 3090 return bond_event_changename(event_bond); 3091 case NETDEV_UNREGISTER: 3092 bond_remove_proc_entry(event_bond); 3093 break; 3094 case NETDEV_REGISTER: 3095 bond_create_proc_entry(event_bond); 3096 break; 3097 case NETDEV_NOTIFY_PEERS: 3098 if (event_bond->send_peer_notif) 3099 event_bond->send_peer_notif--; 3100 break; 3101 default: 3102 break; 3103 } 3104 3105 return NOTIFY_DONE; 3106 } 3107 3108 static int bond_slave_netdev_event(unsigned long event, 3109 struct net_device *slave_dev) 3110 { 3111 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3112 struct bonding *bond; 3113 struct net_device *bond_dev; 3114 3115 /* A netdev event can be generated while enslaving a device 3116 * before netdev_rx_handler_register is called in which case 3117 * slave will be NULL 3118 */ 3119 if (!slave) 3120 return NOTIFY_DONE; 3121 bond_dev = slave->bond->dev; 3122 bond = slave->bond; 3123 primary = rtnl_dereference(bond->primary_slave); 3124 3125 switch (event) { 3126 case NETDEV_UNREGISTER: 3127 if (bond_dev->type != ARPHRD_ETHER) 3128 bond_release_and_destroy(bond_dev, slave_dev); 3129 else 3130 __bond_release_one(bond_dev, slave_dev, false, true); 3131 break; 3132 case NETDEV_UP: 3133 case NETDEV_CHANGE: 3134 /* For 802.3ad mode only: 3135 * Getting invalid Speed/Duplex values here will put slave 3136 * in weird state. Mark it as link-fail if the link was 3137 * previously up or link-down if it hasn't yet come up, and 3138 * let link-monitoring (miimon) set it right when correct 3139 * speeds/duplex are available. 3140 */ 3141 if (bond_update_speed_duplex(slave) && 3142 BOND_MODE(bond) == BOND_MODE_8023AD) { 3143 if (slave->last_link_up) 3144 slave->link = BOND_LINK_FAIL; 3145 else 3146 slave->link = BOND_LINK_DOWN; 3147 } 3148 3149 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3150 bond_3ad_adapter_speed_duplex_changed(slave); 3151 /* Fallthrough */ 3152 case NETDEV_DOWN: 3153 /* Refresh slave-array if applicable! 3154 * If the setup does not use miimon or arpmon (mode-specific!), 3155 * then these events will not cause the slave-array to be 3156 * refreshed. This will cause xmit to use a slave that is not 3157 * usable. Avoid such situation by refeshing the array at these 3158 * events. If these (miimon/arpmon) parameters are configured 3159 * then array gets refreshed twice and that should be fine! 3160 */ 3161 if (bond_mode_can_use_xmit_hash(bond)) 3162 bond_update_slave_arr(bond, NULL); 3163 break; 3164 case NETDEV_CHANGEMTU: 3165 /* TODO: Should slaves be allowed to 3166 * independently alter their MTU? For 3167 * an active-backup bond, slaves need 3168 * not be the same type of device, so 3169 * MTUs may vary. For other modes, 3170 * slaves arguably should have the 3171 * same MTUs. To do this, we'd need to 3172 * take over the slave's change_mtu 3173 * function for the duration of their 3174 * servitude. 3175 */ 3176 break; 3177 case NETDEV_CHANGENAME: 3178 /* we don't care if we don't have primary set */ 3179 if (!bond_uses_primary(bond) || 3180 !bond->params.primary[0]) 3181 break; 3182 3183 if (slave == primary) { 3184 /* slave's name changed - he's no longer primary */ 3185 RCU_INIT_POINTER(bond->primary_slave, NULL); 3186 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3187 /* we have a new primary slave */ 3188 rcu_assign_pointer(bond->primary_slave, slave); 3189 } else { /* we didn't change primary - exit */ 3190 break; 3191 } 3192 3193 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3194 primary ? slave_dev->name : "none"); 3195 3196 block_netpoll_tx(); 3197 bond_select_active_slave(bond); 3198 unblock_netpoll_tx(); 3199 break; 3200 case NETDEV_FEAT_CHANGE: 3201 bond_compute_features(bond); 3202 break; 3203 case NETDEV_RESEND_IGMP: 3204 /* Propagate to master device */ 3205 call_netdevice_notifiers(event, slave->bond->dev); 3206 break; 3207 default: 3208 break; 3209 } 3210 3211 return NOTIFY_DONE; 3212 } 3213 3214 /* bond_netdev_event: handle netdev notifier chain events. 3215 * 3216 * This function receives events for the netdev chain. The caller (an 3217 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3218 * locks for us to safely manipulate the slave devices (RTNL lock, 3219 * dev_probe_lock). 3220 */ 3221 static int bond_netdev_event(struct notifier_block *this, 3222 unsigned long event, void *ptr) 3223 { 3224 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3225 3226 netdev_dbg(event_dev, "event: %lx\n", event); 3227 3228 if (!(event_dev->priv_flags & IFF_BONDING)) 3229 return NOTIFY_DONE; 3230 3231 if (event_dev->flags & IFF_MASTER) { 3232 int ret; 3233 3234 netdev_dbg(event_dev, "IFF_MASTER\n"); 3235 ret = bond_master_netdev_event(event, event_dev); 3236 if (ret != NOTIFY_DONE) 3237 return ret; 3238 } 3239 3240 if (event_dev->flags & IFF_SLAVE) { 3241 netdev_dbg(event_dev, "IFF_SLAVE\n"); 3242 return bond_slave_netdev_event(event, event_dev); 3243 } 3244 3245 return NOTIFY_DONE; 3246 } 3247 3248 static struct notifier_block bond_netdev_notifier = { 3249 .notifier_call = bond_netdev_event, 3250 }; 3251 3252 /*---------------------------- Hashing Policies -----------------------------*/ 3253 3254 /* L2 hash helper */ 3255 static inline u32 bond_eth_hash(struct sk_buff *skb) 3256 { 3257 struct ethhdr *ep, hdr_tmp; 3258 3259 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3260 if (ep) 3261 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3262 return 0; 3263 } 3264 3265 /* Extract the appropriate headers based on bond's xmit policy */ 3266 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3267 struct flow_keys *fk) 3268 { 3269 const struct ipv6hdr *iph6; 3270 const struct iphdr *iph; 3271 int noff, proto = -1; 3272 3273 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) 3274 return skb_flow_dissect_flow_keys(skb, fk, 0); 3275 3276 fk->ports.ports = 0; 3277 noff = skb_network_offset(skb); 3278 if (skb->protocol == htons(ETH_P_IP)) { 3279 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 3280 return false; 3281 iph = ip_hdr(skb); 3282 iph_to_flow_copy_v4addrs(fk, iph); 3283 noff += iph->ihl << 2; 3284 if (!ip_is_fragment(iph)) 3285 proto = iph->protocol; 3286 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3287 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6)))) 3288 return false; 3289 iph6 = ipv6_hdr(skb); 3290 iph_to_flow_copy_v6addrs(fk, iph6); 3291 noff += sizeof(*iph6); 3292 proto = iph6->nexthdr; 3293 } else { 3294 return false; 3295 } 3296 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0) 3297 fk->ports.ports = skb_flow_get_ports(skb, noff, proto); 3298 3299 return true; 3300 } 3301 3302 /** 3303 * bond_xmit_hash - generate a hash value based on the xmit policy 3304 * @bond: bonding device 3305 * @skb: buffer to use for headers 3306 * 3307 * This function will extract the necessary headers from the skb buffer and use 3308 * them to generate a hash based on the xmit_policy set in the bonding device 3309 */ 3310 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3311 { 3312 struct flow_keys flow; 3313 u32 hash; 3314 3315 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3316 skb->l4_hash) 3317 return skb->hash; 3318 3319 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3320 !bond_flow_dissect(bond, skb, &flow)) 3321 return bond_eth_hash(skb); 3322 3323 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3324 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) 3325 hash = bond_eth_hash(skb); 3326 else 3327 hash = (__force u32)flow.ports.ports; 3328 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3329 (__force u32)flow_get_u32_src(&flow); 3330 hash ^= (hash >> 16); 3331 hash ^= (hash >> 8); 3332 3333 return hash >> 1; 3334 } 3335 3336 /*-------------------------- Device entry points ----------------------------*/ 3337 3338 void bond_work_init_all(struct bonding *bond) 3339 { 3340 INIT_DELAYED_WORK(&bond->mcast_work, 3341 bond_resend_igmp_join_requests_delayed); 3342 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3343 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3344 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3345 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3346 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3347 } 3348 3349 static void bond_work_cancel_all(struct bonding *bond) 3350 { 3351 cancel_delayed_work_sync(&bond->mii_work); 3352 cancel_delayed_work_sync(&bond->arp_work); 3353 cancel_delayed_work_sync(&bond->alb_work); 3354 cancel_delayed_work_sync(&bond->ad_work); 3355 cancel_delayed_work_sync(&bond->mcast_work); 3356 cancel_delayed_work_sync(&bond->slave_arr_work); 3357 } 3358 3359 static int bond_open(struct net_device *bond_dev) 3360 { 3361 struct bonding *bond = netdev_priv(bond_dev); 3362 struct list_head *iter; 3363 struct slave *slave; 3364 3365 /* reset slave->backup and slave->inactive */ 3366 if (bond_has_slaves(bond)) { 3367 bond_for_each_slave(bond, slave, iter) { 3368 if (bond_uses_primary(bond) && 3369 slave != rcu_access_pointer(bond->curr_active_slave)) { 3370 bond_set_slave_inactive_flags(slave, 3371 BOND_SLAVE_NOTIFY_NOW); 3372 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3373 bond_set_slave_active_flags(slave, 3374 BOND_SLAVE_NOTIFY_NOW); 3375 } 3376 } 3377 } 3378 3379 if (bond_is_lb(bond)) { 3380 /* bond_alb_initialize must be called before the timer 3381 * is started. 3382 */ 3383 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3384 return -ENOMEM; 3385 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3386 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3387 } 3388 3389 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3390 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3391 3392 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3393 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3394 bond->recv_probe = bond_arp_rcv; 3395 } 3396 3397 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3398 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3399 /* register to receive LACPDUs */ 3400 bond->recv_probe = bond_3ad_lacpdu_recv; 3401 bond_3ad_initiate_agg_selection(bond, 1); 3402 } 3403 3404 if (bond_mode_can_use_xmit_hash(bond)) 3405 bond_update_slave_arr(bond, NULL); 3406 3407 return 0; 3408 } 3409 3410 static int bond_close(struct net_device *bond_dev) 3411 { 3412 struct bonding *bond = netdev_priv(bond_dev); 3413 3414 bond_work_cancel_all(bond); 3415 bond->send_peer_notif = 0; 3416 if (bond_is_lb(bond)) 3417 bond_alb_deinitialize(bond); 3418 bond->recv_probe = NULL; 3419 3420 return 0; 3421 } 3422 3423 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3424 * that some drivers can provide 32bit values only. 3425 */ 3426 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3427 const struct rtnl_link_stats64 *_new, 3428 const struct rtnl_link_stats64 *_old) 3429 { 3430 const u64 *new = (const u64 *)_new; 3431 const u64 *old = (const u64 *)_old; 3432 u64 *res = (u64 *)_res; 3433 int i; 3434 3435 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3436 u64 nv = new[i]; 3437 u64 ov = old[i]; 3438 s64 delta = nv - ov; 3439 3440 /* detects if this particular field is 32bit only */ 3441 if (((nv | ov) >> 32) == 0) 3442 delta = (s64)(s32)((u32)nv - (u32)ov); 3443 3444 /* filter anomalies, some drivers reset their stats 3445 * at down/up events. 3446 */ 3447 if (delta > 0) 3448 res[i] += delta; 3449 } 3450 } 3451 3452 static int bond_get_nest_level(struct net_device *bond_dev) 3453 { 3454 struct bonding *bond = netdev_priv(bond_dev); 3455 3456 return bond->nest_level; 3457 } 3458 3459 static void bond_get_stats(struct net_device *bond_dev, 3460 struct rtnl_link_stats64 *stats) 3461 { 3462 struct bonding *bond = netdev_priv(bond_dev); 3463 struct rtnl_link_stats64 temp; 3464 struct list_head *iter; 3465 struct slave *slave; 3466 3467 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev)); 3468 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3469 3470 rcu_read_lock(); 3471 bond_for_each_slave_rcu(bond, slave, iter) { 3472 const struct rtnl_link_stats64 *new = 3473 dev_get_stats(slave->dev, &temp); 3474 3475 bond_fold_stats(stats, new, &slave->slave_stats); 3476 3477 /* save off the slave stats for the next run */ 3478 memcpy(&slave->slave_stats, new, sizeof(*new)); 3479 } 3480 rcu_read_unlock(); 3481 3482 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3483 spin_unlock(&bond->stats_lock); 3484 } 3485 3486 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3487 { 3488 struct bonding *bond = netdev_priv(bond_dev); 3489 struct net_device *slave_dev = NULL; 3490 struct ifbond k_binfo; 3491 struct ifbond __user *u_binfo = NULL; 3492 struct ifslave k_sinfo; 3493 struct ifslave __user *u_sinfo = NULL; 3494 struct mii_ioctl_data *mii = NULL; 3495 struct bond_opt_value newval; 3496 struct net *net; 3497 int res = 0; 3498 3499 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3500 3501 switch (cmd) { 3502 case SIOCGMIIPHY: 3503 mii = if_mii(ifr); 3504 if (!mii) 3505 return -EINVAL; 3506 3507 mii->phy_id = 0; 3508 /* Fall Through */ 3509 case SIOCGMIIREG: 3510 /* We do this again just in case we were called by SIOCGMIIREG 3511 * instead of SIOCGMIIPHY. 3512 */ 3513 mii = if_mii(ifr); 3514 if (!mii) 3515 return -EINVAL; 3516 3517 if (mii->reg_num == 1) { 3518 mii->val_out = 0; 3519 if (netif_carrier_ok(bond->dev)) 3520 mii->val_out = BMSR_LSTATUS; 3521 } 3522 3523 return 0; 3524 case BOND_INFO_QUERY_OLD: 3525 case SIOCBONDINFOQUERY: 3526 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3527 3528 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3529 return -EFAULT; 3530 3531 bond_info_query(bond_dev, &k_binfo); 3532 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3533 return -EFAULT; 3534 3535 return 0; 3536 case BOND_SLAVE_INFO_QUERY_OLD: 3537 case SIOCBONDSLAVEINFOQUERY: 3538 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3539 3540 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3541 return -EFAULT; 3542 3543 res = bond_slave_info_query(bond_dev, &k_sinfo); 3544 if (res == 0 && 3545 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3546 return -EFAULT; 3547 3548 return res; 3549 default: 3550 break; 3551 } 3552 3553 net = dev_net(bond_dev); 3554 3555 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3556 return -EPERM; 3557 3558 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3559 3560 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev); 3561 3562 if (!slave_dev) 3563 return -ENODEV; 3564 3565 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name); 3566 switch (cmd) { 3567 case BOND_ENSLAVE_OLD: 3568 case SIOCBONDENSLAVE: 3569 res = bond_enslave(bond_dev, slave_dev, NULL); 3570 break; 3571 case BOND_RELEASE_OLD: 3572 case SIOCBONDRELEASE: 3573 res = bond_release(bond_dev, slave_dev); 3574 break; 3575 case BOND_SETHWADDR_OLD: 3576 case SIOCBONDSETHWADDR: 3577 res = bond_set_dev_addr(bond_dev, slave_dev); 3578 break; 3579 case BOND_CHANGE_ACTIVE_OLD: 3580 case SIOCBONDCHANGEACTIVE: 3581 bond_opt_initstr(&newval, slave_dev->name); 3582 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3583 &newval); 3584 break; 3585 default: 3586 res = -EOPNOTSUPP; 3587 } 3588 3589 return res; 3590 } 3591 3592 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3593 { 3594 struct bonding *bond = netdev_priv(bond_dev); 3595 3596 if (change & IFF_PROMISC) 3597 bond_set_promiscuity(bond, 3598 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3599 3600 if (change & IFF_ALLMULTI) 3601 bond_set_allmulti(bond, 3602 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3603 } 3604 3605 static void bond_set_rx_mode(struct net_device *bond_dev) 3606 { 3607 struct bonding *bond = netdev_priv(bond_dev); 3608 struct list_head *iter; 3609 struct slave *slave; 3610 3611 rcu_read_lock(); 3612 if (bond_uses_primary(bond)) { 3613 slave = rcu_dereference(bond->curr_active_slave); 3614 if (slave) { 3615 dev_uc_sync(slave->dev, bond_dev); 3616 dev_mc_sync(slave->dev, bond_dev); 3617 } 3618 } else { 3619 bond_for_each_slave_rcu(bond, slave, iter) { 3620 dev_uc_sync_multiple(slave->dev, bond_dev); 3621 dev_mc_sync_multiple(slave->dev, bond_dev); 3622 } 3623 } 3624 rcu_read_unlock(); 3625 } 3626 3627 static int bond_neigh_init(struct neighbour *n) 3628 { 3629 struct bonding *bond = netdev_priv(n->dev); 3630 const struct net_device_ops *slave_ops; 3631 struct neigh_parms parms; 3632 struct slave *slave; 3633 int ret; 3634 3635 slave = bond_first_slave(bond); 3636 if (!slave) 3637 return 0; 3638 slave_ops = slave->dev->netdev_ops; 3639 if (!slave_ops->ndo_neigh_setup) 3640 return 0; 3641 3642 parms.neigh_setup = NULL; 3643 parms.neigh_cleanup = NULL; 3644 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3645 if (ret) 3646 return ret; 3647 3648 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called 3649 * after the last slave has been detached. Assumes that all slaves 3650 * utilize the same neigh_cleanup (true at this writing as only user 3651 * is ipoib). 3652 */ 3653 n->parms->neigh_cleanup = parms.neigh_cleanup; 3654 3655 if (!parms.neigh_setup) 3656 return 0; 3657 3658 return parms.neigh_setup(n); 3659 } 3660 3661 /* The bonding ndo_neigh_setup is called at init time beofre any 3662 * slave exists. So we must declare proxy setup function which will 3663 * be used at run time to resolve the actual slave neigh param setup. 3664 * 3665 * It's also called by master devices (such as vlans) to setup their 3666 * underlying devices. In that case - do nothing, we're already set up from 3667 * our init. 3668 */ 3669 static int bond_neigh_setup(struct net_device *dev, 3670 struct neigh_parms *parms) 3671 { 3672 /* modify only our neigh_parms */ 3673 if (parms->dev == dev) 3674 parms->neigh_setup = bond_neigh_init; 3675 3676 return 0; 3677 } 3678 3679 /* Change the MTU of all of a master's slaves to match the master */ 3680 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3681 { 3682 struct bonding *bond = netdev_priv(bond_dev); 3683 struct slave *slave, *rollback_slave; 3684 struct list_head *iter; 3685 int res = 0; 3686 3687 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3688 3689 bond_for_each_slave(bond, slave, iter) { 3690 netdev_dbg(bond_dev, "s %p c_m %p\n", 3691 slave, slave->dev->netdev_ops->ndo_change_mtu); 3692 3693 res = dev_set_mtu(slave->dev, new_mtu); 3694 3695 if (res) { 3696 /* If we failed to set the slave's mtu to the new value 3697 * we must abort the operation even in ACTIVE_BACKUP 3698 * mode, because if we allow the backup slaves to have 3699 * different mtu values than the active slave we'll 3700 * need to change their mtu when doing a failover. That 3701 * means changing their mtu from timer context, which 3702 * is probably not a good idea. 3703 */ 3704 netdev_dbg(bond_dev, "err %d %s\n", res, 3705 slave->dev->name); 3706 goto unwind; 3707 } 3708 } 3709 3710 bond_dev->mtu = new_mtu; 3711 3712 return 0; 3713 3714 unwind: 3715 /* unwind from head to the slave that failed */ 3716 bond_for_each_slave(bond, rollback_slave, iter) { 3717 int tmp_res; 3718 3719 if (rollback_slave == slave) 3720 break; 3721 3722 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3723 if (tmp_res) { 3724 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3725 tmp_res, rollback_slave->dev->name); 3726 } 3727 } 3728 3729 return res; 3730 } 3731 3732 /* Change HW address 3733 * 3734 * Note that many devices must be down to change the HW address, and 3735 * downing the master releases all slaves. We can make bonds full of 3736 * bonding devices to test this, however. 3737 */ 3738 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3739 { 3740 struct bonding *bond = netdev_priv(bond_dev); 3741 struct slave *slave, *rollback_slave; 3742 struct sockaddr_storage *ss = addr, tmp_ss; 3743 struct list_head *iter; 3744 int res = 0; 3745 3746 if (BOND_MODE(bond) == BOND_MODE_ALB) 3747 return bond_alb_set_mac_address(bond_dev, addr); 3748 3749 3750 netdev_dbg(bond_dev, "bond=%p\n", bond); 3751 3752 /* If fail_over_mac is enabled, do nothing and return success. 3753 * Returning an error causes ifenslave to fail. 3754 */ 3755 if (bond->params.fail_over_mac && 3756 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3757 return 0; 3758 3759 if (!is_valid_ether_addr(ss->__data)) 3760 return -EADDRNOTAVAIL; 3761 3762 bond_for_each_slave(bond, slave, iter) { 3763 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name); 3764 res = dev_set_mac_address(slave->dev, addr, NULL); 3765 if (res) { 3766 /* TODO: consider downing the slave 3767 * and retry ? 3768 * User should expect communications 3769 * breakage anyway until ARP finish 3770 * updating, so... 3771 */ 3772 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name); 3773 goto unwind; 3774 } 3775 } 3776 3777 /* success */ 3778 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 3779 return 0; 3780 3781 unwind: 3782 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 3783 tmp_ss.ss_family = bond_dev->type; 3784 3785 /* unwind from head to the slave that failed */ 3786 bond_for_each_slave(bond, rollback_slave, iter) { 3787 int tmp_res; 3788 3789 if (rollback_slave == slave) 3790 break; 3791 3792 tmp_res = dev_set_mac_address(rollback_slave->dev, 3793 (struct sockaddr *)&tmp_ss, NULL); 3794 if (tmp_res) { 3795 netdev_dbg(bond_dev, "unwind err %d dev %s\n", 3796 tmp_res, rollback_slave->dev->name); 3797 } 3798 } 3799 3800 return res; 3801 } 3802 3803 /** 3804 * bond_xmit_slave_id - transmit skb through slave with slave_id 3805 * @bond: bonding device that is transmitting 3806 * @skb: buffer to transmit 3807 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3808 * 3809 * This function tries to transmit through slave with slave_id but in case 3810 * it fails, it tries to find the first available slave for transmission. 3811 * The skb is consumed in all cases, thus the function is void. 3812 */ 3813 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3814 { 3815 struct list_head *iter; 3816 struct slave *slave; 3817 int i = slave_id; 3818 3819 /* Here we start from the slave with slave_id */ 3820 bond_for_each_slave_rcu(bond, slave, iter) { 3821 if (--i < 0) { 3822 if (bond_slave_can_tx(slave)) { 3823 bond_dev_queue_xmit(bond, skb, slave->dev); 3824 return; 3825 } 3826 } 3827 } 3828 3829 /* Here we start from the first slave up to slave_id */ 3830 i = slave_id; 3831 bond_for_each_slave_rcu(bond, slave, iter) { 3832 if (--i < 0) 3833 break; 3834 if (bond_slave_can_tx(slave)) { 3835 bond_dev_queue_xmit(bond, skb, slave->dev); 3836 return; 3837 } 3838 } 3839 /* no slave that can tx has been found */ 3840 bond_tx_drop(bond->dev, skb); 3841 } 3842 3843 /** 3844 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3845 * @bond: bonding device to use 3846 * 3847 * Based on the value of the bonding device's packets_per_slave parameter 3848 * this function generates a slave id, which is usually used as the next 3849 * slave to transmit through. 3850 */ 3851 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3852 { 3853 u32 slave_id; 3854 struct reciprocal_value reciprocal_packets_per_slave; 3855 int packets_per_slave = bond->params.packets_per_slave; 3856 3857 switch (packets_per_slave) { 3858 case 0: 3859 slave_id = prandom_u32(); 3860 break; 3861 case 1: 3862 slave_id = bond->rr_tx_counter; 3863 break; 3864 default: 3865 reciprocal_packets_per_slave = 3866 bond->params.reciprocal_packets_per_slave; 3867 slave_id = reciprocal_divide(bond->rr_tx_counter, 3868 reciprocal_packets_per_slave); 3869 break; 3870 } 3871 bond->rr_tx_counter++; 3872 3873 return slave_id; 3874 } 3875 3876 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 3877 struct net_device *bond_dev) 3878 { 3879 struct bonding *bond = netdev_priv(bond_dev); 3880 struct iphdr *iph = ip_hdr(skb); 3881 struct slave *slave; 3882 u32 slave_id; 3883 3884 /* Start with the curr_active_slave that joined the bond as the 3885 * default for sending IGMP traffic. For failover purposes one 3886 * needs to maintain some consistency for the interface that will 3887 * send the join/membership reports. The curr_active_slave found 3888 * will send all of this type of traffic. 3889 */ 3890 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) { 3891 slave = rcu_dereference(bond->curr_active_slave); 3892 if (slave) 3893 bond_dev_queue_xmit(bond, skb, slave->dev); 3894 else 3895 bond_xmit_slave_id(bond, skb, 0); 3896 } else { 3897 int slave_cnt = READ_ONCE(bond->slave_cnt); 3898 3899 if (likely(slave_cnt)) { 3900 slave_id = bond_rr_gen_slave_id(bond); 3901 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3902 } else { 3903 bond_tx_drop(bond_dev, skb); 3904 } 3905 } 3906 3907 return NETDEV_TX_OK; 3908 } 3909 3910 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3911 * the bond has a usable interface. 3912 */ 3913 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 3914 struct net_device *bond_dev) 3915 { 3916 struct bonding *bond = netdev_priv(bond_dev); 3917 struct slave *slave; 3918 3919 slave = rcu_dereference(bond->curr_active_slave); 3920 if (slave) 3921 bond_dev_queue_xmit(bond, skb, slave->dev); 3922 else 3923 bond_tx_drop(bond_dev, skb); 3924 3925 return NETDEV_TX_OK; 3926 } 3927 3928 /* Use this to update slave_array when (a) it's not appropriate to update 3929 * slave_array right away (note that update_slave_array() may sleep) 3930 * and / or (b) RTNL is not held. 3931 */ 3932 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 3933 { 3934 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 3935 } 3936 3937 /* Slave array work handler. Holds only RTNL */ 3938 static void bond_slave_arr_handler(struct work_struct *work) 3939 { 3940 struct bonding *bond = container_of(work, struct bonding, 3941 slave_arr_work.work); 3942 int ret; 3943 3944 if (!rtnl_trylock()) 3945 goto err; 3946 3947 ret = bond_update_slave_arr(bond, NULL); 3948 rtnl_unlock(); 3949 if (ret) { 3950 pr_warn_ratelimited("Failed to update slave array from WT\n"); 3951 goto err; 3952 } 3953 return; 3954 3955 err: 3956 bond_slave_arr_work_rearm(bond, 1); 3957 } 3958 3959 /* Build the usable slaves array in control path for modes that use xmit-hash 3960 * to determine the slave interface - 3961 * (a) BOND_MODE_8023AD 3962 * (b) BOND_MODE_XOR 3963 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 3964 * 3965 * The caller is expected to hold RTNL only and NO other lock! 3966 */ 3967 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 3968 { 3969 struct slave *slave; 3970 struct list_head *iter; 3971 struct bond_up_slave *new_arr, *old_arr; 3972 int agg_id = 0; 3973 int ret = 0; 3974 3975 #ifdef CONFIG_LOCKDEP 3976 WARN_ON(lockdep_is_held(&bond->mode_lock)); 3977 #endif 3978 3979 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 3980 GFP_KERNEL); 3981 if (!new_arr) { 3982 ret = -ENOMEM; 3983 pr_err("Failed to build slave-array.\n"); 3984 goto out; 3985 } 3986 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3987 struct ad_info ad_info; 3988 3989 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 3990 pr_debug("bond_3ad_get_active_agg_info failed\n"); 3991 kfree_rcu(new_arr, rcu); 3992 /* No active aggragator means it's not safe to use 3993 * the previous array. 3994 */ 3995 old_arr = rtnl_dereference(bond->slave_arr); 3996 if (old_arr) { 3997 RCU_INIT_POINTER(bond->slave_arr, NULL); 3998 kfree_rcu(old_arr, rcu); 3999 } 4000 goto out; 4001 } 4002 agg_id = ad_info.aggregator_id; 4003 } 4004 bond_for_each_slave(bond, slave, iter) { 4005 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4006 struct aggregator *agg; 4007 4008 agg = SLAVE_AD_INFO(slave)->port.aggregator; 4009 if (!agg || agg->aggregator_identifier != agg_id) 4010 continue; 4011 } 4012 if (!bond_slave_can_tx(slave)) 4013 continue; 4014 if (skipslave == slave) 4015 continue; 4016 4017 netdev_dbg(bond->dev, 4018 "Adding slave dev %s to tx hash array[%d]\n", 4019 slave->dev->name, new_arr->count); 4020 4021 new_arr->arr[new_arr->count++] = slave; 4022 } 4023 4024 old_arr = rtnl_dereference(bond->slave_arr); 4025 rcu_assign_pointer(bond->slave_arr, new_arr); 4026 if (old_arr) 4027 kfree_rcu(old_arr, rcu); 4028 out: 4029 if (ret != 0 && skipslave) { 4030 int idx; 4031 4032 /* Rare situation where caller has asked to skip a specific 4033 * slave but allocation failed (most likely!). BTW this is 4034 * only possible when the call is initiated from 4035 * __bond_release_one(). In this situation; overwrite the 4036 * skipslave entry in the array with the last entry from the 4037 * array to avoid a situation where the xmit path may choose 4038 * this to-be-skipped slave to send a packet out. 4039 */ 4040 old_arr = rtnl_dereference(bond->slave_arr); 4041 for (idx = 0; idx < old_arr->count; idx++) { 4042 if (skipslave == old_arr->arr[idx]) { 4043 old_arr->arr[idx] = 4044 old_arr->arr[old_arr->count-1]; 4045 old_arr->count--; 4046 break; 4047 } 4048 } 4049 } 4050 return ret; 4051 } 4052 4053 /* Use this Xmit function for 3AD as well as XOR modes. The current 4054 * usable slave array is formed in the control path. The xmit function 4055 * just calculates hash and sends the packet out. 4056 */ 4057 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4058 struct net_device *dev) 4059 { 4060 struct bonding *bond = netdev_priv(dev); 4061 struct slave *slave; 4062 struct bond_up_slave *slaves; 4063 unsigned int count; 4064 4065 slaves = rcu_dereference(bond->slave_arr); 4066 count = slaves ? READ_ONCE(slaves->count) : 0; 4067 if (likely(count)) { 4068 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 4069 bond_dev_queue_xmit(bond, skb, slave->dev); 4070 } else { 4071 bond_tx_drop(dev, skb); 4072 } 4073 4074 return NETDEV_TX_OK; 4075 } 4076 4077 /* in broadcast mode, we send everything to all usable interfaces. */ 4078 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4079 struct net_device *bond_dev) 4080 { 4081 struct bonding *bond = netdev_priv(bond_dev); 4082 struct slave *slave = NULL; 4083 struct list_head *iter; 4084 4085 bond_for_each_slave_rcu(bond, slave, iter) { 4086 if (bond_is_last_slave(bond, slave)) 4087 break; 4088 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4089 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4090 4091 if (!skb2) { 4092 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4093 bond_dev->name, __func__); 4094 continue; 4095 } 4096 bond_dev_queue_xmit(bond, skb2, slave->dev); 4097 } 4098 } 4099 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4100 bond_dev_queue_xmit(bond, skb, slave->dev); 4101 else 4102 bond_tx_drop(bond_dev, skb); 4103 4104 return NETDEV_TX_OK; 4105 } 4106 4107 /*------------------------- Device initialization ---------------------------*/ 4108 4109 /* Lookup the slave that corresponds to a qid */ 4110 static inline int bond_slave_override(struct bonding *bond, 4111 struct sk_buff *skb) 4112 { 4113 struct slave *slave = NULL; 4114 struct list_head *iter; 4115 4116 if (!skb_rx_queue_recorded(skb)) 4117 return 1; 4118 4119 /* Find out if any slaves have the same mapping as this skb. */ 4120 bond_for_each_slave_rcu(bond, slave, iter) { 4121 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4122 if (bond_slave_is_up(slave) && 4123 slave->link == BOND_LINK_UP) { 4124 bond_dev_queue_xmit(bond, skb, slave->dev); 4125 return 0; 4126 } 4127 /* If the slave isn't UP, use default transmit policy. */ 4128 break; 4129 } 4130 } 4131 4132 return 1; 4133 } 4134 4135 4136 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4137 struct net_device *sb_dev) 4138 { 4139 /* This helper function exists to help dev_pick_tx get the correct 4140 * destination queue. Using a helper function skips a call to 4141 * skb_tx_hash and will put the skbs in the queue we expect on their 4142 * way down to the bonding driver. 4143 */ 4144 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4145 4146 /* Save the original txq to restore before passing to the driver */ 4147 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4148 4149 if (unlikely(txq >= dev->real_num_tx_queues)) { 4150 do { 4151 txq -= dev->real_num_tx_queues; 4152 } while (txq >= dev->real_num_tx_queues); 4153 } 4154 return txq; 4155 } 4156 4157 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4158 { 4159 struct bonding *bond = netdev_priv(dev); 4160 4161 if (bond_should_override_tx_queue(bond) && 4162 !bond_slave_override(bond, skb)) 4163 return NETDEV_TX_OK; 4164 4165 switch (BOND_MODE(bond)) { 4166 case BOND_MODE_ROUNDROBIN: 4167 return bond_xmit_roundrobin(skb, dev); 4168 case BOND_MODE_ACTIVEBACKUP: 4169 return bond_xmit_activebackup(skb, dev); 4170 case BOND_MODE_8023AD: 4171 case BOND_MODE_XOR: 4172 return bond_3ad_xor_xmit(skb, dev); 4173 case BOND_MODE_BROADCAST: 4174 return bond_xmit_broadcast(skb, dev); 4175 case BOND_MODE_ALB: 4176 return bond_alb_xmit(skb, dev); 4177 case BOND_MODE_TLB: 4178 return bond_tlb_xmit(skb, dev); 4179 default: 4180 /* Should never happen, mode already checked */ 4181 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4182 WARN_ON_ONCE(1); 4183 bond_tx_drop(dev, skb); 4184 return NETDEV_TX_OK; 4185 } 4186 } 4187 4188 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4189 { 4190 struct bonding *bond = netdev_priv(dev); 4191 netdev_tx_t ret = NETDEV_TX_OK; 4192 4193 /* If we risk deadlock from transmitting this in the 4194 * netpoll path, tell netpoll to queue the frame for later tx 4195 */ 4196 if (unlikely(is_netpoll_tx_blocked(dev))) 4197 return NETDEV_TX_BUSY; 4198 4199 rcu_read_lock(); 4200 if (bond_has_slaves(bond)) 4201 ret = __bond_start_xmit(skb, dev); 4202 else 4203 bond_tx_drop(dev, skb); 4204 rcu_read_unlock(); 4205 4206 return ret; 4207 } 4208 4209 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4210 struct ethtool_link_ksettings *cmd) 4211 { 4212 struct bonding *bond = netdev_priv(bond_dev); 4213 unsigned long speed = 0; 4214 struct list_head *iter; 4215 struct slave *slave; 4216 4217 cmd->base.duplex = DUPLEX_UNKNOWN; 4218 cmd->base.port = PORT_OTHER; 4219 4220 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4221 * do not need to check mode. Though link speed might not represent 4222 * the true receive or transmit bandwidth (not all modes are symmetric) 4223 * this is an accurate maximum. 4224 */ 4225 bond_for_each_slave(bond, slave, iter) { 4226 if (bond_slave_can_tx(slave)) { 4227 if (slave->speed != SPEED_UNKNOWN) 4228 speed += slave->speed; 4229 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4230 slave->duplex != DUPLEX_UNKNOWN) 4231 cmd->base.duplex = slave->duplex; 4232 } 4233 } 4234 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4235 4236 return 0; 4237 } 4238 4239 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4240 struct ethtool_drvinfo *drvinfo) 4241 { 4242 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4243 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 4244 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4245 BOND_ABI_VERSION); 4246 } 4247 4248 static const struct ethtool_ops bond_ethtool_ops = { 4249 .get_drvinfo = bond_ethtool_get_drvinfo, 4250 .get_link = ethtool_op_get_link, 4251 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4252 }; 4253 4254 static const struct net_device_ops bond_netdev_ops = { 4255 .ndo_init = bond_init, 4256 .ndo_uninit = bond_uninit, 4257 .ndo_open = bond_open, 4258 .ndo_stop = bond_close, 4259 .ndo_start_xmit = bond_start_xmit, 4260 .ndo_select_queue = bond_select_queue, 4261 .ndo_get_stats64 = bond_get_stats, 4262 .ndo_do_ioctl = bond_do_ioctl, 4263 .ndo_change_rx_flags = bond_change_rx_flags, 4264 .ndo_set_rx_mode = bond_set_rx_mode, 4265 .ndo_change_mtu = bond_change_mtu, 4266 .ndo_set_mac_address = bond_set_mac_address, 4267 .ndo_neigh_setup = bond_neigh_setup, 4268 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4269 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4270 .ndo_get_lock_subclass = bond_get_nest_level, 4271 #ifdef CONFIG_NET_POLL_CONTROLLER 4272 .ndo_netpoll_setup = bond_netpoll_setup, 4273 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4274 .ndo_poll_controller = bond_poll_controller, 4275 #endif 4276 .ndo_add_slave = bond_enslave, 4277 .ndo_del_slave = bond_release, 4278 .ndo_fix_features = bond_fix_features, 4279 .ndo_features_check = passthru_features_check, 4280 }; 4281 4282 static const struct device_type bond_type = { 4283 .name = "bond", 4284 }; 4285 4286 static void bond_destructor(struct net_device *bond_dev) 4287 { 4288 struct bonding *bond = netdev_priv(bond_dev); 4289 if (bond->wq) 4290 destroy_workqueue(bond->wq); 4291 } 4292 4293 void bond_setup(struct net_device *bond_dev) 4294 { 4295 struct bonding *bond = netdev_priv(bond_dev); 4296 4297 spin_lock_init(&bond->mode_lock); 4298 spin_lock_init(&bond->stats_lock); 4299 bond->params = bonding_defaults; 4300 4301 /* Initialize pointers */ 4302 bond->dev = bond_dev; 4303 4304 /* Initialize the device entry points */ 4305 ether_setup(bond_dev); 4306 bond_dev->max_mtu = ETH_MAX_MTU; 4307 bond_dev->netdev_ops = &bond_netdev_ops; 4308 bond_dev->ethtool_ops = &bond_ethtool_ops; 4309 4310 bond_dev->needs_free_netdev = true; 4311 bond_dev->priv_destructor = bond_destructor; 4312 4313 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4314 4315 /* Initialize the device options */ 4316 bond_dev->flags |= IFF_MASTER; 4317 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4318 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4319 4320 /* don't acquire bond device's netif_tx_lock when transmitting */ 4321 bond_dev->features |= NETIF_F_LLTX; 4322 4323 /* By default, we declare the bond to be fully 4324 * VLAN hardware accelerated capable. Special 4325 * care is taken in the various xmit functions 4326 * when there are slaves that are not hw accel 4327 * capable 4328 */ 4329 4330 /* Don't allow bond devices to change network namespaces. */ 4331 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4332 4333 bond_dev->hw_features = BOND_VLAN_FEATURES | 4334 NETIF_F_HW_VLAN_CTAG_TX | 4335 NETIF_F_HW_VLAN_CTAG_RX | 4336 NETIF_F_HW_VLAN_CTAG_FILTER; 4337 4338 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4339 bond_dev->features |= bond_dev->hw_features; 4340 } 4341 4342 /* Destroy a bonding device. 4343 * Must be under rtnl_lock when this function is called. 4344 */ 4345 static void bond_uninit(struct net_device *bond_dev) 4346 { 4347 struct bonding *bond = netdev_priv(bond_dev); 4348 struct list_head *iter; 4349 struct slave *slave; 4350 struct bond_up_slave *arr; 4351 4352 bond_netpoll_cleanup(bond_dev); 4353 4354 /* Release the bonded slaves */ 4355 bond_for_each_slave(bond, slave, iter) 4356 __bond_release_one(bond_dev, slave->dev, true, true); 4357 netdev_info(bond_dev, "Released all slaves\n"); 4358 4359 arr = rtnl_dereference(bond->slave_arr); 4360 if (arr) { 4361 RCU_INIT_POINTER(bond->slave_arr, NULL); 4362 kfree_rcu(arr, rcu); 4363 } 4364 4365 list_del(&bond->bond_list); 4366 4367 bond_debug_unregister(bond); 4368 } 4369 4370 /*------------------------- Module initialization ---------------------------*/ 4371 4372 static int bond_check_params(struct bond_params *params) 4373 { 4374 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4375 struct bond_opt_value newval; 4376 const struct bond_opt_value *valptr; 4377 int arp_all_targets_value = 0; 4378 u16 ad_actor_sys_prio = 0; 4379 u16 ad_user_port_key = 0; 4380 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4381 int arp_ip_count; 4382 int bond_mode = BOND_MODE_ROUNDROBIN; 4383 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4384 int lacp_fast = 0; 4385 int tlb_dynamic_lb; 4386 4387 /* Convert string parameters. */ 4388 if (mode) { 4389 bond_opt_initstr(&newval, mode); 4390 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4391 if (!valptr) { 4392 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4393 return -EINVAL; 4394 } 4395 bond_mode = valptr->value; 4396 } 4397 4398 if (xmit_hash_policy) { 4399 if (bond_mode == BOND_MODE_ROUNDROBIN || 4400 bond_mode == BOND_MODE_ACTIVEBACKUP || 4401 bond_mode == BOND_MODE_BROADCAST) { 4402 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4403 bond_mode_name(bond_mode)); 4404 } else { 4405 bond_opt_initstr(&newval, xmit_hash_policy); 4406 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4407 &newval); 4408 if (!valptr) { 4409 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4410 xmit_hash_policy); 4411 return -EINVAL; 4412 } 4413 xmit_hashtype = valptr->value; 4414 } 4415 } 4416 4417 if (lacp_rate) { 4418 if (bond_mode != BOND_MODE_8023AD) { 4419 pr_info("lacp_rate param is irrelevant in mode %s\n", 4420 bond_mode_name(bond_mode)); 4421 } else { 4422 bond_opt_initstr(&newval, lacp_rate); 4423 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4424 &newval); 4425 if (!valptr) { 4426 pr_err("Error: Invalid lacp rate \"%s\"\n", 4427 lacp_rate); 4428 return -EINVAL; 4429 } 4430 lacp_fast = valptr->value; 4431 } 4432 } 4433 4434 if (ad_select) { 4435 bond_opt_initstr(&newval, ad_select); 4436 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4437 &newval); 4438 if (!valptr) { 4439 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4440 return -EINVAL; 4441 } 4442 params->ad_select = valptr->value; 4443 if (bond_mode != BOND_MODE_8023AD) 4444 pr_warn("ad_select param only affects 802.3ad mode\n"); 4445 } else { 4446 params->ad_select = BOND_AD_STABLE; 4447 } 4448 4449 if (max_bonds < 0) { 4450 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4451 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4452 max_bonds = BOND_DEFAULT_MAX_BONDS; 4453 } 4454 4455 if (miimon < 0) { 4456 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4457 miimon, INT_MAX); 4458 miimon = 0; 4459 } 4460 4461 if (updelay < 0) { 4462 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4463 updelay, INT_MAX); 4464 updelay = 0; 4465 } 4466 4467 if (downdelay < 0) { 4468 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4469 downdelay, INT_MAX); 4470 downdelay = 0; 4471 } 4472 4473 if ((use_carrier != 0) && (use_carrier != 1)) { 4474 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4475 use_carrier); 4476 use_carrier = 1; 4477 } 4478 4479 if (num_peer_notif < 0 || num_peer_notif > 255) { 4480 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4481 num_peer_notif); 4482 num_peer_notif = 1; 4483 } 4484 4485 /* reset values for 802.3ad/TLB/ALB */ 4486 if (!bond_mode_uses_arp(bond_mode)) { 4487 if (!miimon) { 4488 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"); 4489 pr_warn("Forcing miimon to 100msec\n"); 4490 miimon = BOND_DEFAULT_MIIMON; 4491 } 4492 } 4493 4494 if (tx_queues < 1 || tx_queues > 255) { 4495 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4496 tx_queues, BOND_DEFAULT_TX_QUEUES); 4497 tx_queues = BOND_DEFAULT_TX_QUEUES; 4498 } 4499 4500 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4501 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4502 all_slaves_active); 4503 all_slaves_active = 0; 4504 } 4505 4506 if (resend_igmp < 0 || resend_igmp > 255) { 4507 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4508 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4509 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4510 } 4511 4512 bond_opt_initval(&newval, packets_per_slave); 4513 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4514 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4515 packets_per_slave, USHRT_MAX); 4516 packets_per_slave = 1; 4517 } 4518 4519 if (bond_mode == BOND_MODE_ALB) { 4520 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", 4521 updelay); 4522 } 4523 4524 if (!miimon) { 4525 if (updelay || downdelay) { 4526 /* just warn the user the up/down delay will have 4527 * no effect since miimon is zero... 4528 */ 4529 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", 4530 updelay, downdelay); 4531 } 4532 } else { 4533 /* don't allow arp monitoring */ 4534 if (arp_interval) { 4535 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4536 miimon, arp_interval); 4537 arp_interval = 0; 4538 } 4539 4540 if ((updelay % miimon) != 0) { 4541 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4542 updelay, miimon, (updelay / miimon) * miimon); 4543 } 4544 4545 updelay /= miimon; 4546 4547 if ((downdelay % miimon) != 0) { 4548 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4549 downdelay, miimon, 4550 (downdelay / miimon) * miimon); 4551 } 4552 4553 downdelay /= miimon; 4554 } 4555 4556 if (arp_interval < 0) { 4557 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4558 arp_interval, INT_MAX); 4559 arp_interval = 0; 4560 } 4561 4562 for (arp_ip_count = 0, i = 0; 4563 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4564 __be32 ip; 4565 4566 /* not a complete check, but good enough to catch mistakes */ 4567 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4568 !bond_is_ip_target_ok(ip)) { 4569 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4570 arp_ip_target[i]); 4571 arp_interval = 0; 4572 } else { 4573 if (bond_get_targets_ip(arp_target, ip) == -1) 4574 arp_target[arp_ip_count++] = ip; 4575 else 4576 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4577 &ip); 4578 } 4579 } 4580 4581 if (arp_interval && !arp_ip_count) { 4582 /* don't allow arping if no arp_ip_target given... */ 4583 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4584 arp_interval); 4585 arp_interval = 0; 4586 } 4587 4588 if (arp_validate) { 4589 if (!arp_interval) { 4590 pr_err("arp_validate requires arp_interval\n"); 4591 return -EINVAL; 4592 } 4593 4594 bond_opt_initstr(&newval, arp_validate); 4595 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4596 &newval); 4597 if (!valptr) { 4598 pr_err("Error: invalid arp_validate \"%s\"\n", 4599 arp_validate); 4600 return -EINVAL; 4601 } 4602 arp_validate_value = valptr->value; 4603 } else { 4604 arp_validate_value = 0; 4605 } 4606 4607 if (arp_all_targets) { 4608 bond_opt_initstr(&newval, arp_all_targets); 4609 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4610 &newval); 4611 if (!valptr) { 4612 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4613 arp_all_targets); 4614 arp_all_targets_value = 0; 4615 } else { 4616 arp_all_targets_value = valptr->value; 4617 } 4618 } 4619 4620 if (miimon) { 4621 pr_info("MII link monitoring set to %d ms\n", miimon); 4622 } else if (arp_interval) { 4623 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4624 arp_validate_value); 4625 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4626 arp_interval, valptr->string, arp_ip_count); 4627 4628 for (i = 0; i < arp_ip_count; i++) 4629 pr_cont(" %s", arp_ip_target[i]); 4630 4631 pr_cont("\n"); 4632 4633 } else if (max_bonds) { 4634 /* miimon and arp_interval not set, we need one so things 4635 * work as expected, see bonding.txt for details 4636 */ 4637 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"); 4638 } 4639 4640 if (primary && !bond_mode_uses_primary(bond_mode)) { 4641 /* currently, using a primary only makes sense 4642 * in active backup, TLB or ALB modes 4643 */ 4644 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4645 primary, bond_mode_name(bond_mode)); 4646 primary = NULL; 4647 } 4648 4649 if (primary && primary_reselect) { 4650 bond_opt_initstr(&newval, primary_reselect); 4651 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4652 &newval); 4653 if (!valptr) { 4654 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4655 primary_reselect); 4656 return -EINVAL; 4657 } 4658 primary_reselect_value = valptr->value; 4659 } else { 4660 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4661 } 4662 4663 if (fail_over_mac) { 4664 bond_opt_initstr(&newval, fail_over_mac); 4665 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4666 &newval); 4667 if (!valptr) { 4668 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4669 fail_over_mac); 4670 return -EINVAL; 4671 } 4672 fail_over_mac_value = valptr->value; 4673 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4674 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4675 } else { 4676 fail_over_mac_value = BOND_FOM_NONE; 4677 } 4678 4679 bond_opt_initstr(&newval, "default"); 4680 valptr = bond_opt_parse( 4681 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 4682 &newval); 4683 if (!valptr) { 4684 pr_err("Error: No ad_actor_sys_prio default value"); 4685 return -EINVAL; 4686 } 4687 ad_actor_sys_prio = valptr->value; 4688 4689 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 4690 &newval); 4691 if (!valptr) { 4692 pr_err("Error: No ad_user_port_key default value"); 4693 return -EINVAL; 4694 } 4695 ad_user_port_key = valptr->value; 4696 4697 bond_opt_initstr(&newval, "default"); 4698 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 4699 if (!valptr) { 4700 pr_err("Error: No tlb_dynamic_lb default value"); 4701 return -EINVAL; 4702 } 4703 tlb_dynamic_lb = valptr->value; 4704 4705 if (lp_interval == 0) { 4706 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4707 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4708 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4709 } 4710 4711 /* fill params struct with the proper values */ 4712 params->mode = bond_mode; 4713 params->xmit_policy = xmit_hashtype; 4714 params->miimon = miimon; 4715 params->num_peer_notif = num_peer_notif; 4716 params->arp_interval = arp_interval; 4717 params->arp_validate = arp_validate_value; 4718 params->arp_all_targets = arp_all_targets_value; 4719 params->updelay = updelay; 4720 params->downdelay = downdelay; 4721 params->use_carrier = use_carrier; 4722 params->lacp_fast = lacp_fast; 4723 params->primary[0] = 0; 4724 params->primary_reselect = primary_reselect_value; 4725 params->fail_over_mac = fail_over_mac_value; 4726 params->tx_queues = tx_queues; 4727 params->all_slaves_active = all_slaves_active; 4728 params->resend_igmp = resend_igmp; 4729 params->min_links = min_links; 4730 params->lp_interval = lp_interval; 4731 params->packets_per_slave = packets_per_slave; 4732 params->tlb_dynamic_lb = tlb_dynamic_lb; 4733 params->ad_actor_sys_prio = ad_actor_sys_prio; 4734 eth_zero_addr(params->ad_actor_system); 4735 params->ad_user_port_key = ad_user_port_key; 4736 if (packets_per_slave > 0) { 4737 params->reciprocal_packets_per_slave = 4738 reciprocal_value(packets_per_slave); 4739 } else { 4740 /* reciprocal_packets_per_slave is unused if 4741 * packets_per_slave is 0 or 1, just initialize it 4742 */ 4743 params->reciprocal_packets_per_slave = 4744 (struct reciprocal_value) { 0 }; 4745 } 4746 4747 if (primary) { 4748 strncpy(params->primary, primary, IFNAMSIZ); 4749 params->primary[IFNAMSIZ - 1] = 0; 4750 } 4751 4752 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4753 4754 return 0; 4755 } 4756 4757 /* Called from registration process */ 4758 static int bond_init(struct net_device *bond_dev) 4759 { 4760 struct bonding *bond = netdev_priv(bond_dev); 4761 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4762 4763 netdev_dbg(bond_dev, "Begin bond_init\n"); 4764 4765 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 4766 if (!bond->wq) 4767 return -ENOMEM; 4768 4769 bond->nest_level = SINGLE_DEPTH_NESTING; 4770 netdev_lockdep_set_classes(bond_dev); 4771 4772 list_add_tail(&bond->bond_list, &bn->dev_list); 4773 4774 bond_prepare_sysfs_group(bond); 4775 4776 bond_debug_register(bond); 4777 4778 /* Ensure valid dev_addr */ 4779 if (is_zero_ether_addr(bond_dev->dev_addr) && 4780 bond_dev->addr_assign_type == NET_ADDR_PERM) 4781 eth_hw_addr_random(bond_dev); 4782 4783 return 0; 4784 } 4785 4786 unsigned int bond_get_num_tx_queues(void) 4787 { 4788 return tx_queues; 4789 } 4790 4791 /* Create a new bond based on the specified name and bonding parameters. 4792 * If name is NULL, obtain a suitable "bond%d" name for us. 4793 * Caller must NOT hold rtnl_lock; we need to release it here before we 4794 * set up our sysfs entries. 4795 */ 4796 int bond_create(struct net *net, const char *name) 4797 { 4798 struct net_device *bond_dev; 4799 struct bonding *bond; 4800 struct alb_bond_info *bond_info; 4801 int res; 4802 4803 rtnl_lock(); 4804 4805 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4806 name ? name : "bond%d", NET_NAME_UNKNOWN, 4807 bond_setup, tx_queues); 4808 if (!bond_dev) { 4809 pr_err("%s: eek! can't alloc netdev!\n", name); 4810 rtnl_unlock(); 4811 return -ENOMEM; 4812 } 4813 4814 /* 4815 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 4816 * It is set to 0 by default which is wrong. 4817 */ 4818 bond = netdev_priv(bond_dev); 4819 bond_info = &(BOND_ALB_INFO(bond)); 4820 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 4821 4822 dev_net_set(bond_dev, net); 4823 bond_dev->rtnl_link_ops = &bond_link_ops; 4824 4825 res = register_netdevice(bond_dev); 4826 4827 netif_carrier_off(bond_dev); 4828 4829 bond_work_init_all(bond); 4830 4831 rtnl_unlock(); 4832 if (res < 0) 4833 free_netdev(bond_dev); 4834 return res; 4835 } 4836 4837 static int __net_init bond_net_init(struct net *net) 4838 { 4839 struct bond_net *bn = net_generic(net, bond_net_id); 4840 4841 bn->net = net; 4842 INIT_LIST_HEAD(&bn->dev_list); 4843 4844 bond_create_proc_dir(bn); 4845 bond_create_sysfs(bn); 4846 4847 return 0; 4848 } 4849 4850 static void __net_exit bond_net_exit(struct net *net) 4851 { 4852 struct bond_net *bn = net_generic(net, bond_net_id); 4853 struct bonding *bond, *tmp_bond; 4854 LIST_HEAD(list); 4855 4856 bond_destroy_sysfs(bn); 4857 4858 /* Kill off any bonds created after unregistering bond rtnl ops */ 4859 rtnl_lock(); 4860 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4861 unregister_netdevice_queue(bond->dev, &list); 4862 unregister_netdevice_many(&list); 4863 rtnl_unlock(); 4864 4865 bond_destroy_proc_dir(bn); 4866 } 4867 4868 static struct pernet_operations bond_net_ops = { 4869 .init = bond_net_init, 4870 .exit = bond_net_exit, 4871 .id = &bond_net_id, 4872 .size = sizeof(struct bond_net), 4873 }; 4874 4875 static int __init bonding_init(void) 4876 { 4877 int i; 4878 int res; 4879 4880 pr_info("%s", bond_version); 4881 4882 res = bond_check_params(&bonding_defaults); 4883 if (res) 4884 goto out; 4885 4886 res = register_pernet_subsys(&bond_net_ops); 4887 if (res) 4888 goto out; 4889 4890 res = bond_netlink_init(); 4891 if (res) 4892 goto err_link; 4893 4894 bond_create_debugfs(); 4895 4896 for (i = 0; i < max_bonds; i++) { 4897 res = bond_create(&init_net, NULL); 4898 if (res) 4899 goto err; 4900 } 4901 4902 register_netdevice_notifier(&bond_netdev_notifier); 4903 out: 4904 return res; 4905 err: 4906 bond_destroy_debugfs(); 4907 bond_netlink_fini(); 4908 err_link: 4909 unregister_pernet_subsys(&bond_net_ops); 4910 goto out; 4911 4912 } 4913 4914 static void __exit bonding_exit(void) 4915 { 4916 unregister_netdevice_notifier(&bond_netdev_notifier); 4917 4918 bond_destroy_debugfs(); 4919 4920 bond_netlink_fini(); 4921 unregister_pernet_subsys(&bond_net_ops); 4922 4923 #ifdef CONFIG_NET_POLL_CONTROLLER 4924 /* Make sure we don't have an imbalance on our netpoll blocking */ 4925 WARN_ON(atomic_read(&netpoll_block_tx)); 4926 #endif 4927 } 4928 4929 module_init(bonding_init); 4930 module_exit(bonding_exit); 4931 MODULE_LICENSE("GPL"); 4932 MODULE_VERSION(DRV_VERSION); 4933 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 4934 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 4935