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