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