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