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/icmp.h> 45 #include <linux/icmpv6.h> 46 #include <linux/tcp.h> 47 #include <linux/udp.h> 48 #include <linux/slab.h> 49 #include <linux/string.h> 50 #include <linux/init.h> 51 #include <linux/timer.h> 52 #include <linux/socket.h> 53 #include <linux/ctype.h> 54 #include <linux/inet.h> 55 #include <linux/bitops.h> 56 #include <linux/io.h> 57 #include <asm/dma.h> 58 #include <linux/uaccess.h> 59 #include <linux/errno.h> 60 #include <linux/netdevice.h> 61 #include <linux/inetdevice.h> 62 #include <linux/igmp.h> 63 #include <linux/etherdevice.h> 64 #include <linux/skbuff.h> 65 #include <net/sock.h> 66 #include <linux/rtnetlink.h> 67 #include <linux/smp.h> 68 #include <linux/if_ether.h> 69 #include <net/arp.h> 70 #include <linux/mii.h> 71 #include <linux/ethtool.h> 72 #include <linux/if_vlan.h> 73 #include <linux/if_bonding.h> 74 #include <linux/jiffies.h> 75 #include <linux/preempt.h> 76 #include <net/route.h> 77 #include <net/net_namespace.h> 78 #include <net/netns/generic.h> 79 #include <net/pkt_sched.h> 80 #include <linux/rculist.h> 81 #include <net/flow_dissector.h> 82 #include <net/bonding.h> 83 #include <net/bond_3ad.h> 84 #include <net/bond_alb.h> 85 86 #include "bonding_priv.h" 87 88 /*---------------------------- Module parameters ----------------------------*/ 89 90 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 91 92 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 93 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 94 static int num_peer_notif = 1; 95 static int miimon; 96 static int updelay; 97 static int downdelay; 98 static int use_carrier = 1; 99 static char *mode; 100 static char *primary; 101 static char *primary_reselect; 102 static char *lacp_rate; 103 static int min_links; 104 static char *ad_select; 105 static char *xmit_hash_policy; 106 static int arp_interval; 107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 108 static char *arp_validate; 109 static char *arp_all_targets; 110 static char *fail_over_mac; 111 static int all_slaves_active; 112 static struct bond_params bonding_defaults; 113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 114 static int packets_per_slave = 1; 115 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 116 117 module_param(max_bonds, int, 0); 118 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 119 module_param(tx_queues, int, 0); 120 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 121 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 122 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 123 "failover event (alias of num_unsol_na)"); 124 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 125 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 126 "failover event (alias of num_grat_arp)"); 127 module_param(miimon, int, 0); 128 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 129 module_param(updelay, int, 0); 130 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 131 module_param(downdelay, int, 0); 132 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 133 "in milliseconds"); 134 module_param(use_carrier, int, 0); 135 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 136 "0 for off, 1 for on (default)"); 137 module_param(mode, charp, 0); 138 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 139 "1 for active-backup, 2 for balance-xor, " 140 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 141 "6 for balance-alb"); 142 module_param(primary, charp, 0); 143 MODULE_PARM_DESC(primary, "Primary network device to use"); 144 module_param(primary_reselect, charp, 0); 145 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 146 "once it comes up; " 147 "0 for always (default), " 148 "1 for only if speed of primary is " 149 "better, " 150 "2 for only on active slave " 151 "failure"); 152 module_param(lacp_rate, charp, 0); 153 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 154 "0 for slow, 1 for fast"); 155 module_param(ad_select, charp, 0); 156 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 157 "0 for stable (default), 1 for bandwidth, " 158 "2 for count"); 159 module_param(min_links, int, 0); 160 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 161 162 module_param(xmit_hash_policy, charp, 0); 163 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 164 "0 for layer 2 (default), 1 for layer 3+4, " 165 "2 for layer 2+3, 3 for encap layer 2+3, " 166 "4 for encap layer 3+4"); 167 module_param(arp_interval, int, 0); 168 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 169 module_param_array(arp_ip_target, charp, NULL, 0); 170 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 171 module_param(arp_validate, charp, 0); 172 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 173 "0 for none (default), 1 for active, " 174 "2 for backup, 3 for all"); 175 module_param(arp_all_targets, charp, 0); 176 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 177 module_param(fail_over_mac, charp, 0); 178 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 179 "the same MAC; 0 for none (default), " 180 "1 for active, 2 for follow"); 181 module_param(all_slaves_active, int, 0); 182 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 183 "by setting active flag for all slaves; " 184 "0 for never (default), 1 for always."); 185 module_param(resend_igmp, int, 0); 186 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 187 "link failure"); 188 module_param(packets_per_slave, int, 0); 189 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 190 "mode; 0 for a random slave, 1 packet per " 191 "slave (default), >1 packets per slave."); 192 module_param(lp_interval, uint, 0); 193 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 194 "the bonding driver sends learning packets to " 195 "each slaves peer switch. The default is 1."); 196 197 /*----------------------------- Global variables ----------------------------*/ 198 199 #ifdef CONFIG_NET_POLL_CONTROLLER 200 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 201 #endif 202 203 unsigned int bond_net_id __read_mostly; 204 205 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 206 { 207 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 208 .offset = offsetof(struct flow_keys, control), 209 }, 210 { 211 .key_id = FLOW_DISSECTOR_KEY_BASIC, 212 .offset = offsetof(struct flow_keys, basic), 213 }, 214 { 215 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 216 .offset = offsetof(struct flow_keys, addrs.v4addrs), 217 }, 218 { 219 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 220 .offset = offsetof(struct flow_keys, addrs.v6addrs), 221 }, 222 { 223 .key_id = FLOW_DISSECTOR_KEY_TIPC, 224 .offset = offsetof(struct flow_keys, addrs.tipckey), 225 }, 226 { 227 .key_id = FLOW_DISSECTOR_KEY_PORTS, 228 .offset = offsetof(struct flow_keys, ports), 229 }, 230 { 231 .key_id = FLOW_DISSECTOR_KEY_ICMP, 232 .offset = offsetof(struct flow_keys, icmp), 233 }, 234 { 235 .key_id = FLOW_DISSECTOR_KEY_VLAN, 236 .offset = offsetof(struct flow_keys, vlan), 237 }, 238 { 239 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 240 .offset = offsetof(struct flow_keys, tags), 241 }, 242 { 243 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 244 .offset = offsetof(struct flow_keys, keyid), 245 }, 246 }; 247 248 static struct flow_dissector flow_keys_bonding __read_mostly; 249 250 /*-------------------------- Forward declarations ---------------------------*/ 251 252 static int bond_init(struct net_device *bond_dev); 253 static void bond_uninit(struct net_device *bond_dev); 254 static void bond_get_stats(struct net_device *bond_dev, 255 struct rtnl_link_stats64 *stats); 256 static void bond_slave_arr_handler(struct work_struct *work); 257 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 258 int mod); 259 static void bond_netdev_notify_work(struct work_struct *work); 260 261 /*---------------------------- General routines -----------------------------*/ 262 263 const char *bond_mode_name(int mode) 264 { 265 static const char *names[] = { 266 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 267 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 268 [BOND_MODE_XOR] = "load balancing (xor)", 269 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 270 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 271 [BOND_MODE_TLB] = "transmit load balancing", 272 [BOND_MODE_ALB] = "adaptive load balancing", 273 }; 274 275 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 276 return "unknown"; 277 278 return names[mode]; 279 } 280 281 /*---------------------------------- VLAN -----------------------------------*/ 282 283 /** 284 * bond_dev_queue_xmit - Prepare skb for xmit. 285 * 286 * @bond: bond device that got this skb for tx. 287 * @skb: hw accel VLAN tagged skb to transmit 288 * @slave_dev: slave that is supposed to xmit this skbuff 289 */ 290 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 291 struct net_device *slave_dev) 292 { 293 skb->dev = slave_dev; 294 295 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 296 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 297 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 298 299 if (unlikely(netpoll_tx_running(bond->dev))) 300 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 301 else 302 dev_queue_xmit(skb); 303 } 304 305 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 306 * We don't protect the slave list iteration with a lock because: 307 * a. This operation is performed in IOCTL context, 308 * b. The operation is protected by the RTNL semaphore in the 8021q code, 309 * c. Holding a lock with BH disabled while directly calling a base driver 310 * entry point is generally a BAD idea. 311 * 312 * The design of synchronization/protection for this operation in the 8021q 313 * module is good for one or more VLAN devices over a single physical device 314 * and cannot be extended for a teaming solution like bonding, so there is a 315 * potential race condition here where a net device from the vlan group might 316 * be referenced (either by a base driver or the 8021q code) while it is being 317 * removed from the system. However, it turns out we're not making matters 318 * worse, and if it works for regular VLAN usage it will work here too. 319 */ 320 321 /** 322 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 323 * @bond_dev: bonding net device that got called 324 * @vid: vlan id being added 325 */ 326 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 327 __be16 proto, u16 vid) 328 { 329 struct bonding *bond = netdev_priv(bond_dev); 330 struct slave *slave, *rollback_slave; 331 struct list_head *iter; 332 int res; 333 334 bond_for_each_slave(bond, slave, iter) { 335 res = vlan_vid_add(slave->dev, proto, vid); 336 if (res) 337 goto unwind; 338 } 339 340 return 0; 341 342 unwind: 343 /* unwind to the slave that failed */ 344 bond_for_each_slave(bond, rollback_slave, iter) { 345 if (rollback_slave == slave) 346 break; 347 348 vlan_vid_del(rollback_slave->dev, proto, vid); 349 } 350 351 return res; 352 } 353 354 /** 355 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 356 * @bond_dev: bonding net device that got called 357 * @vid: vlan id being removed 358 */ 359 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 360 __be16 proto, u16 vid) 361 { 362 struct bonding *bond = netdev_priv(bond_dev); 363 struct list_head *iter; 364 struct slave *slave; 365 366 bond_for_each_slave(bond, slave, iter) 367 vlan_vid_del(slave->dev, proto, vid); 368 369 if (bond_is_lb(bond)) 370 bond_alb_clear_vlan(bond, vid); 371 372 return 0; 373 } 374 375 /*------------------------------- Link status -------------------------------*/ 376 377 /* Set the carrier state for the master according to the state of its 378 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 379 * do special 802.3ad magic. 380 * 381 * Returns zero if carrier state does not change, nonzero if it does. 382 */ 383 int bond_set_carrier(struct bonding *bond) 384 { 385 struct list_head *iter; 386 struct slave *slave; 387 388 if (!bond_has_slaves(bond)) 389 goto down; 390 391 if (BOND_MODE(bond) == BOND_MODE_8023AD) 392 return bond_3ad_set_carrier(bond); 393 394 bond_for_each_slave(bond, slave, iter) { 395 if (slave->link == BOND_LINK_UP) { 396 if (!netif_carrier_ok(bond->dev)) { 397 netif_carrier_on(bond->dev); 398 return 1; 399 } 400 return 0; 401 } 402 } 403 404 down: 405 if (netif_carrier_ok(bond->dev)) { 406 netif_carrier_off(bond->dev); 407 return 1; 408 } 409 return 0; 410 } 411 412 /* Get link speed and duplex from the slave's base driver 413 * using ethtool. If for some reason the call fails or the 414 * values are invalid, set speed and duplex to -1, 415 * and return. Return 1 if speed or duplex settings are 416 * UNKNOWN; 0 otherwise. 417 */ 418 static int bond_update_speed_duplex(struct slave *slave) 419 { 420 struct net_device *slave_dev = slave->dev; 421 struct ethtool_link_ksettings ecmd; 422 int res; 423 424 slave->speed = SPEED_UNKNOWN; 425 slave->duplex = DUPLEX_UNKNOWN; 426 427 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 428 if (res < 0) 429 return 1; 430 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 431 return 1; 432 switch (ecmd.base.duplex) { 433 case DUPLEX_FULL: 434 case DUPLEX_HALF: 435 break; 436 default: 437 return 1; 438 } 439 440 slave->speed = ecmd.base.speed; 441 slave->duplex = ecmd.base.duplex; 442 443 return 0; 444 } 445 446 const char *bond_slave_link_status(s8 link) 447 { 448 switch (link) { 449 case BOND_LINK_UP: 450 return "up"; 451 case BOND_LINK_FAIL: 452 return "going down"; 453 case BOND_LINK_DOWN: 454 return "down"; 455 case BOND_LINK_BACK: 456 return "going back"; 457 default: 458 return "unknown"; 459 } 460 } 461 462 /* if <dev> supports MII link status reporting, check its link status. 463 * 464 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 465 * depending upon the setting of the use_carrier parameter. 466 * 467 * Return either BMSR_LSTATUS, meaning that the link is up (or we 468 * can't tell and just pretend it is), or 0, meaning that the link is 469 * down. 470 * 471 * If reporting is non-zero, instead of faking link up, return -1 if 472 * both ETHTOOL and MII ioctls fail (meaning the device does not 473 * support them). If use_carrier is set, return whatever it says. 474 * It'd be nice if there was a good way to tell if a driver supports 475 * netif_carrier, but there really isn't. 476 */ 477 static int bond_check_dev_link(struct bonding *bond, 478 struct net_device *slave_dev, int reporting) 479 { 480 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 481 int (*ioctl)(struct net_device *, struct ifreq *, int); 482 struct ifreq ifr; 483 struct mii_ioctl_data *mii; 484 485 if (!reporting && !netif_running(slave_dev)) 486 return 0; 487 488 if (bond->params.use_carrier) 489 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 490 491 /* Try to get link status using Ethtool first. */ 492 if (slave_dev->ethtool_ops->get_link) 493 return slave_dev->ethtool_ops->get_link(slave_dev) ? 494 BMSR_LSTATUS : 0; 495 496 /* Ethtool can't be used, fallback to MII ioctls. */ 497 ioctl = slave_ops->ndo_do_ioctl; 498 if (ioctl) { 499 /* TODO: set pointer to correct ioctl on a per team member 500 * bases to make this more efficient. that is, once 501 * we determine the correct ioctl, we will always 502 * call it and not the others for that team 503 * member. 504 */ 505 506 /* We cannot assume that SIOCGMIIPHY will also read a 507 * register; not all network drivers (e.g., e100) 508 * support that. 509 */ 510 511 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 512 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 513 mii = if_mii(&ifr); 514 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 515 mii->reg_num = MII_BMSR; 516 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 517 return mii->val_out & BMSR_LSTATUS; 518 } 519 } 520 521 /* If reporting, report that either there's no dev->do_ioctl, 522 * or both SIOCGMIIREG and get_link failed (meaning that we 523 * cannot report link status). If not reporting, pretend 524 * we're ok. 525 */ 526 return reporting ? -1 : BMSR_LSTATUS; 527 } 528 529 /*----------------------------- Multicast list ------------------------------*/ 530 531 /* Push the promiscuity flag down to appropriate slaves */ 532 static int bond_set_promiscuity(struct bonding *bond, int inc) 533 { 534 struct list_head *iter; 535 int err = 0; 536 537 if (bond_uses_primary(bond)) { 538 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 539 540 if (curr_active) 541 err = dev_set_promiscuity(curr_active->dev, inc); 542 } else { 543 struct slave *slave; 544 545 bond_for_each_slave(bond, slave, iter) { 546 err = dev_set_promiscuity(slave->dev, inc); 547 if (err) 548 return err; 549 } 550 } 551 return err; 552 } 553 554 /* Push the allmulti flag down to all slaves */ 555 static int bond_set_allmulti(struct bonding *bond, int inc) 556 { 557 struct list_head *iter; 558 int err = 0; 559 560 if (bond_uses_primary(bond)) { 561 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 562 563 if (curr_active) 564 err = dev_set_allmulti(curr_active->dev, inc); 565 } else { 566 struct slave *slave; 567 568 bond_for_each_slave(bond, slave, iter) { 569 err = dev_set_allmulti(slave->dev, inc); 570 if (err) 571 return err; 572 } 573 } 574 return err; 575 } 576 577 /* Retrieve the list of registered multicast addresses for the bonding 578 * device and retransmit an IGMP JOIN request to the current active 579 * slave. 580 */ 581 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 582 { 583 struct bonding *bond = container_of(work, struct bonding, 584 mcast_work.work); 585 586 if (!rtnl_trylock()) { 587 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 588 return; 589 } 590 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 591 592 if (bond->igmp_retrans > 1) { 593 bond->igmp_retrans--; 594 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 595 } 596 rtnl_unlock(); 597 } 598 599 /* Flush bond's hardware addresses from slave */ 600 static void bond_hw_addr_flush(struct net_device *bond_dev, 601 struct net_device *slave_dev) 602 { 603 struct bonding *bond = netdev_priv(bond_dev); 604 605 dev_uc_unsync(slave_dev, bond_dev); 606 dev_mc_unsync(slave_dev, bond_dev); 607 608 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 609 /* del lacpdu mc addr from mc list */ 610 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 611 612 dev_mc_del(slave_dev, lacpdu_multicast); 613 } 614 } 615 616 /*--------------------------- Active slave change ---------------------------*/ 617 618 /* Update the hardware address list and promisc/allmulti for the new and 619 * old active slaves (if any). Modes that are not using primary keep all 620 * slaves up date at all times; only the modes that use primary need to call 621 * this function to swap these settings during a failover. 622 */ 623 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 624 struct slave *old_active) 625 { 626 if (old_active) { 627 if (bond->dev->flags & IFF_PROMISC) 628 dev_set_promiscuity(old_active->dev, -1); 629 630 if (bond->dev->flags & IFF_ALLMULTI) 631 dev_set_allmulti(old_active->dev, -1); 632 633 bond_hw_addr_flush(bond->dev, old_active->dev); 634 } 635 636 if (new_active) { 637 /* FIXME: Signal errors upstream. */ 638 if (bond->dev->flags & IFF_PROMISC) 639 dev_set_promiscuity(new_active->dev, 1); 640 641 if (bond->dev->flags & IFF_ALLMULTI) 642 dev_set_allmulti(new_active->dev, 1); 643 644 netif_addr_lock_bh(bond->dev); 645 dev_uc_sync(new_active->dev, bond->dev); 646 dev_mc_sync(new_active->dev, bond->dev); 647 netif_addr_unlock_bh(bond->dev); 648 } 649 } 650 651 /** 652 * bond_set_dev_addr - clone slave's address to bond 653 * @bond_dev: bond net device 654 * @slave_dev: slave net device 655 * 656 * Should be called with RTNL held. 657 */ 658 static int bond_set_dev_addr(struct net_device *bond_dev, 659 struct net_device *slave_dev) 660 { 661 int err; 662 663 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 664 bond_dev, slave_dev, slave_dev->addr_len); 665 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 666 if (err) 667 return err; 668 669 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 670 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 671 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 672 return 0; 673 } 674 675 static struct slave *bond_get_old_active(struct bonding *bond, 676 struct slave *new_active) 677 { 678 struct slave *slave; 679 struct list_head *iter; 680 681 bond_for_each_slave(bond, slave, iter) { 682 if (slave == new_active) 683 continue; 684 685 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 686 return slave; 687 } 688 689 return NULL; 690 } 691 692 /* bond_do_fail_over_mac 693 * 694 * Perform special MAC address swapping for fail_over_mac settings 695 * 696 * Called with RTNL 697 */ 698 static void bond_do_fail_over_mac(struct bonding *bond, 699 struct slave *new_active, 700 struct slave *old_active) 701 { 702 u8 tmp_mac[MAX_ADDR_LEN]; 703 struct sockaddr_storage ss; 704 int rv; 705 706 switch (bond->params.fail_over_mac) { 707 case BOND_FOM_ACTIVE: 708 if (new_active) { 709 rv = bond_set_dev_addr(bond->dev, new_active->dev); 710 if (rv) 711 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 712 -rv); 713 } 714 break; 715 case BOND_FOM_FOLLOW: 716 /* if new_active && old_active, swap them 717 * if just old_active, do nothing (going to no active slave) 718 * if just new_active, set new_active to bond's MAC 719 */ 720 if (!new_active) 721 return; 722 723 if (!old_active) 724 old_active = bond_get_old_active(bond, new_active); 725 726 if (old_active) { 727 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 728 new_active->dev->addr_len); 729 bond_hw_addr_copy(ss.__data, 730 old_active->dev->dev_addr, 731 old_active->dev->addr_len); 732 ss.ss_family = new_active->dev->type; 733 } else { 734 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 735 bond->dev->addr_len); 736 ss.ss_family = bond->dev->type; 737 } 738 739 rv = dev_set_mac_address(new_active->dev, 740 (struct sockaddr *)&ss, NULL); 741 if (rv) { 742 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 743 -rv); 744 goto out; 745 } 746 747 if (!old_active) 748 goto out; 749 750 bond_hw_addr_copy(ss.__data, tmp_mac, 751 new_active->dev->addr_len); 752 ss.ss_family = old_active->dev->type; 753 754 rv = dev_set_mac_address(old_active->dev, 755 (struct sockaddr *)&ss, NULL); 756 if (rv) 757 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 758 -rv); 759 out: 760 break; 761 default: 762 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 763 bond->params.fail_over_mac); 764 break; 765 } 766 767 } 768 769 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 770 { 771 struct slave *prim = rtnl_dereference(bond->primary_slave); 772 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 773 774 if (!prim || prim->link != BOND_LINK_UP) { 775 if (!curr || curr->link != BOND_LINK_UP) 776 return NULL; 777 return curr; 778 } 779 780 if (bond->force_primary) { 781 bond->force_primary = false; 782 return prim; 783 } 784 785 if (!curr || curr->link != BOND_LINK_UP) 786 return prim; 787 788 /* At this point, prim and curr are both up */ 789 switch (bond->params.primary_reselect) { 790 case BOND_PRI_RESELECT_ALWAYS: 791 return prim; 792 case BOND_PRI_RESELECT_BETTER: 793 if (prim->speed < curr->speed) 794 return curr; 795 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 796 return curr; 797 return prim; 798 case BOND_PRI_RESELECT_FAILURE: 799 return curr; 800 default: 801 netdev_err(bond->dev, "impossible primary_reselect %d\n", 802 bond->params.primary_reselect); 803 return curr; 804 } 805 } 806 807 /** 808 * bond_find_best_slave - select the best available slave to be the active one 809 * @bond: our bonding struct 810 */ 811 static struct slave *bond_find_best_slave(struct bonding *bond) 812 { 813 struct slave *slave, *bestslave = NULL; 814 struct list_head *iter; 815 int mintime = bond->params.updelay; 816 817 slave = bond_choose_primary_or_current(bond); 818 if (slave) 819 return slave; 820 821 bond_for_each_slave(bond, slave, iter) { 822 if (slave->link == BOND_LINK_UP) 823 return slave; 824 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 825 slave->delay < mintime) { 826 mintime = slave->delay; 827 bestslave = slave; 828 } 829 } 830 831 return bestslave; 832 } 833 834 static bool bond_should_notify_peers(struct bonding *bond) 835 { 836 struct slave *slave; 837 838 rcu_read_lock(); 839 slave = rcu_dereference(bond->curr_active_slave); 840 rcu_read_unlock(); 841 842 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 843 slave ? slave->dev->name : "NULL"); 844 845 if (!slave || !bond->send_peer_notif || 846 bond->send_peer_notif % 847 max(1, bond->params.peer_notif_delay) != 0 || 848 !netif_carrier_ok(bond->dev) || 849 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 850 return false; 851 852 return true; 853 } 854 855 /** 856 * change_active_interface - change the active slave into the specified one 857 * @bond: our bonding struct 858 * @new: the new slave to make the active one 859 * 860 * Set the new slave to the bond's settings and unset them on the old 861 * curr_active_slave. 862 * Setting include flags, mc-list, promiscuity, allmulti, etc. 863 * 864 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 865 * because it is apparently the best available slave we have, even though its 866 * updelay hasn't timed out yet. 867 * 868 * Caller must hold RTNL. 869 */ 870 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 871 { 872 struct slave *old_active; 873 874 ASSERT_RTNL(); 875 876 old_active = rtnl_dereference(bond->curr_active_slave); 877 878 if (old_active == new_active) 879 return; 880 881 if (new_active) { 882 new_active->last_link_up = jiffies; 883 884 if (new_active->link == BOND_LINK_BACK) { 885 if (bond_uses_primary(bond)) { 886 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 887 (bond->params.updelay - new_active->delay) * bond->params.miimon); 888 } 889 890 new_active->delay = 0; 891 bond_set_slave_link_state(new_active, BOND_LINK_UP, 892 BOND_SLAVE_NOTIFY_NOW); 893 894 if (BOND_MODE(bond) == BOND_MODE_8023AD) 895 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 896 897 if (bond_is_lb(bond)) 898 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 899 } else { 900 if (bond_uses_primary(bond)) { 901 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 902 } 903 } 904 } 905 906 if (bond_uses_primary(bond)) 907 bond_hw_addr_swap(bond, new_active, old_active); 908 909 if (bond_is_lb(bond)) { 910 bond_alb_handle_active_change(bond, new_active); 911 if (old_active) 912 bond_set_slave_inactive_flags(old_active, 913 BOND_SLAVE_NOTIFY_NOW); 914 if (new_active) 915 bond_set_slave_active_flags(new_active, 916 BOND_SLAVE_NOTIFY_NOW); 917 } else { 918 rcu_assign_pointer(bond->curr_active_slave, new_active); 919 } 920 921 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 922 if (old_active) 923 bond_set_slave_inactive_flags(old_active, 924 BOND_SLAVE_NOTIFY_NOW); 925 926 if (new_active) { 927 bool should_notify_peers = false; 928 929 bond_set_slave_active_flags(new_active, 930 BOND_SLAVE_NOTIFY_NOW); 931 932 if (bond->params.fail_over_mac) 933 bond_do_fail_over_mac(bond, new_active, 934 old_active); 935 936 if (netif_running(bond->dev)) { 937 bond->send_peer_notif = 938 bond->params.num_peer_notif * 939 max(1, bond->params.peer_notif_delay); 940 should_notify_peers = 941 bond_should_notify_peers(bond); 942 } 943 944 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 945 if (should_notify_peers) { 946 bond->send_peer_notif--; 947 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 948 bond->dev); 949 } 950 } 951 } 952 953 /* resend IGMP joins since active slave has changed or 954 * all were sent on curr_active_slave. 955 * resend only if bond is brought up with the affected 956 * bonding modes and the retransmission is enabled 957 */ 958 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 959 ((bond_uses_primary(bond) && new_active) || 960 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 961 bond->igmp_retrans = bond->params.resend_igmp; 962 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 963 } 964 } 965 966 /** 967 * bond_select_active_slave - select a new active slave, if needed 968 * @bond: our bonding struct 969 * 970 * This functions should be called when one of the following occurs: 971 * - The old curr_active_slave has been released or lost its link. 972 * - The primary_slave has got its link back. 973 * - A slave has got its link back and there's no old curr_active_slave. 974 * 975 * Caller must hold RTNL. 976 */ 977 void bond_select_active_slave(struct bonding *bond) 978 { 979 struct slave *best_slave; 980 int rv; 981 982 ASSERT_RTNL(); 983 984 best_slave = bond_find_best_slave(bond); 985 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 986 bond_change_active_slave(bond, best_slave); 987 rv = bond_set_carrier(bond); 988 if (!rv) 989 return; 990 991 if (netif_carrier_ok(bond->dev)) 992 netdev_info(bond->dev, "active interface up!\n"); 993 else 994 netdev_info(bond->dev, "now running without any active interface!\n"); 995 } 996 } 997 998 #ifdef CONFIG_NET_POLL_CONTROLLER 999 static inline int slave_enable_netpoll(struct slave *slave) 1000 { 1001 struct netpoll *np; 1002 int err = 0; 1003 1004 np = kzalloc(sizeof(*np), GFP_KERNEL); 1005 err = -ENOMEM; 1006 if (!np) 1007 goto out; 1008 1009 err = __netpoll_setup(np, slave->dev); 1010 if (err) { 1011 kfree(np); 1012 goto out; 1013 } 1014 slave->np = np; 1015 out: 1016 return err; 1017 } 1018 static inline void slave_disable_netpoll(struct slave *slave) 1019 { 1020 struct netpoll *np = slave->np; 1021 1022 if (!np) 1023 return; 1024 1025 slave->np = NULL; 1026 1027 __netpoll_free(np); 1028 } 1029 1030 static void bond_poll_controller(struct net_device *bond_dev) 1031 { 1032 struct bonding *bond = netdev_priv(bond_dev); 1033 struct slave *slave = NULL; 1034 struct list_head *iter; 1035 struct ad_info ad_info; 1036 1037 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1038 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1039 return; 1040 1041 bond_for_each_slave_rcu(bond, slave, iter) { 1042 if (!bond_slave_is_up(slave)) 1043 continue; 1044 1045 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1046 struct aggregator *agg = 1047 SLAVE_AD_INFO(slave)->port.aggregator; 1048 1049 if (agg && 1050 agg->aggregator_identifier != ad_info.aggregator_id) 1051 continue; 1052 } 1053 1054 netpoll_poll_dev(slave->dev); 1055 } 1056 } 1057 1058 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1059 { 1060 struct bonding *bond = netdev_priv(bond_dev); 1061 struct list_head *iter; 1062 struct slave *slave; 1063 1064 bond_for_each_slave(bond, slave, iter) 1065 if (bond_slave_is_up(slave)) 1066 slave_disable_netpoll(slave); 1067 } 1068 1069 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1070 { 1071 struct bonding *bond = netdev_priv(dev); 1072 struct list_head *iter; 1073 struct slave *slave; 1074 int err = 0; 1075 1076 bond_for_each_slave(bond, slave, iter) { 1077 err = slave_enable_netpoll(slave); 1078 if (err) { 1079 bond_netpoll_cleanup(dev); 1080 break; 1081 } 1082 } 1083 return err; 1084 } 1085 #else 1086 static inline int slave_enable_netpoll(struct slave *slave) 1087 { 1088 return 0; 1089 } 1090 static inline void slave_disable_netpoll(struct slave *slave) 1091 { 1092 } 1093 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1094 { 1095 } 1096 #endif 1097 1098 /*---------------------------------- IOCTL ----------------------------------*/ 1099 1100 static netdev_features_t bond_fix_features(struct net_device *dev, 1101 netdev_features_t features) 1102 { 1103 struct bonding *bond = netdev_priv(dev); 1104 struct list_head *iter; 1105 netdev_features_t mask; 1106 struct slave *slave; 1107 1108 mask = features; 1109 1110 features &= ~NETIF_F_ONE_FOR_ALL; 1111 features |= NETIF_F_ALL_FOR_ALL; 1112 1113 bond_for_each_slave(bond, slave, iter) { 1114 features = netdev_increment_features(features, 1115 slave->dev->features, 1116 mask); 1117 } 1118 features = netdev_add_tso_features(features, mask); 1119 1120 return features; 1121 } 1122 1123 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1124 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1125 NETIF_F_HIGHDMA | NETIF_F_LRO) 1126 1127 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1128 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1129 1130 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1131 NETIF_F_ALL_TSO) 1132 1133 static void bond_compute_features(struct bonding *bond) 1134 { 1135 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1136 IFF_XMIT_DST_RELEASE_PERM; 1137 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1138 netdev_features_t enc_features = BOND_ENC_FEATURES; 1139 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1140 struct net_device *bond_dev = bond->dev; 1141 struct list_head *iter; 1142 struct slave *slave; 1143 unsigned short max_hard_header_len = ETH_HLEN; 1144 unsigned int gso_max_size = GSO_MAX_SIZE; 1145 u16 gso_max_segs = GSO_MAX_SEGS; 1146 1147 if (!bond_has_slaves(bond)) 1148 goto done; 1149 vlan_features &= NETIF_F_ALL_FOR_ALL; 1150 mpls_features &= NETIF_F_ALL_FOR_ALL; 1151 1152 bond_for_each_slave(bond, slave, iter) { 1153 vlan_features = netdev_increment_features(vlan_features, 1154 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1155 1156 enc_features = netdev_increment_features(enc_features, 1157 slave->dev->hw_enc_features, 1158 BOND_ENC_FEATURES); 1159 1160 mpls_features = netdev_increment_features(mpls_features, 1161 slave->dev->mpls_features, 1162 BOND_MPLS_FEATURES); 1163 1164 dst_release_flag &= slave->dev->priv_flags; 1165 if (slave->dev->hard_header_len > max_hard_header_len) 1166 max_hard_header_len = slave->dev->hard_header_len; 1167 1168 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1169 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1170 } 1171 bond_dev->hard_header_len = max_hard_header_len; 1172 1173 done: 1174 bond_dev->vlan_features = vlan_features; 1175 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1176 NETIF_F_HW_VLAN_CTAG_TX | 1177 NETIF_F_HW_VLAN_STAG_TX | 1178 NETIF_F_GSO_UDP_L4; 1179 bond_dev->mpls_features = mpls_features; 1180 bond_dev->gso_max_segs = gso_max_segs; 1181 netif_set_gso_max_size(bond_dev, gso_max_size); 1182 1183 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1184 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1185 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1186 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1187 1188 netdev_change_features(bond_dev); 1189 } 1190 1191 static void bond_setup_by_slave(struct net_device *bond_dev, 1192 struct net_device *slave_dev) 1193 { 1194 bond_dev->header_ops = slave_dev->header_ops; 1195 1196 bond_dev->type = slave_dev->type; 1197 bond_dev->hard_header_len = slave_dev->hard_header_len; 1198 bond_dev->addr_len = slave_dev->addr_len; 1199 1200 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1201 slave_dev->addr_len); 1202 } 1203 1204 /* On bonding slaves other than the currently active slave, suppress 1205 * duplicates except for alb non-mcast/bcast. 1206 */ 1207 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1208 struct slave *slave, 1209 struct bonding *bond) 1210 { 1211 if (bond_is_slave_inactive(slave)) { 1212 if (BOND_MODE(bond) == BOND_MODE_ALB && 1213 skb->pkt_type != PACKET_BROADCAST && 1214 skb->pkt_type != PACKET_MULTICAST) 1215 return false; 1216 return true; 1217 } 1218 return false; 1219 } 1220 1221 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1222 { 1223 struct sk_buff *skb = *pskb; 1224 struct slave *slave; 1225 struct bonding *bond; 1226 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1227 struct slave *); 1228 int ret = RX_HANDLER_ANOTHER; 1229 1230 skb = skb_share_check(skb, GFP_ATOMIC); 1231 if (unlikely(!skb)) 1232 return RX_HANDLER_CONSUMED; 1233 1234 *pskb = skb; 1235 1236 slave = bond_slave_get_rcu(skb->dev); 1237 bond = slave->bond; 1238 1239 recv_probe = READ_ONCE(bond->recv_probe); 1240 if (recv_probe) { 1241 ret = recv_probe(skb, bond, slave); 1242 if (ret == RX_HANDLER_CONSUMED) { 1243 consume_skb(skb); 1244 return ret; 1245 } 1246 } 1247 1248 /* 1249 * For packets determined by bond_should_deliver_exact_match() call to 1250 * be suppressed we want to make an exception for link-local packets. 1251 * This is necessary for e.g. LLDP daemons to be able to monitor 1252 * inactive slave links without being forced to bind to them 1253 * explicitly. 1254 * 1255 * At the same time, packets that are passed to the bonding master 1256 * (including link-local ones) can have their originating interface 1257 * determined via PACKET_ORIGDEV socket option. 1258 */ 1259 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1260 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1261 return RX_HANDLER_PASS; 1262 return RX_HANDLER_EXACT; 1263 } 1264 1265 skb->dev = bond->dev; 1266 1267 if (BOND_MODE(bond) == BOND_MODE_ALB && 1268 bond->dev->priv_flags & IFF_BRIDGE_PORT && 1269 skb->pkt_type == PACKET_HOST) { 1270 1271 if (unlikely(skb_cow_head(skb, 1272 skb->data - skb_mac_header(skb)))) { 1273 kfree_skb(skb); 1274 return RX_HANDLER_CONSUMED; 1275 } 1276 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1277 bond->dev->addr_len); 1278 } 1279 1280 return ret; 1281 } 1282 1283 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1284 { 1285 switch (BOND_MODE(bond)) { 1286 case BOND_MODE_ROUNDROBIN: 1287 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1288 case BOND_MODE_ACTIVEBACKUP: 1289 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1290 case BOND_MODE_BROADCAST: 1291 return NETDEV_LAG_TX_TYPE_BROADCAST; 1292 case BOND_MODE_XOR: 1293 case BOND_MODE_8023AD: 1294 return NETDEV_LAG_TX_TYPE_HASH; 1295 default: 1296 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1297 } 1298 } 1299 1300 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1301 enum netdev_lag_tx_type type) 1302 { 1303 if (type != NETDEV_LAG_TX_TYPE_HASH) 1304 return NETDEV_LAG_HASH_NONE; 1305 1306 switch (bond->params.xmit_policy) { 1307 case BOND_XMIT_POLICY_LAYER2: 1308 return NETDEV_LAG_HASH_L2; 1309 case BOND_XMIT_POLICY_LAYER34: 1310 return NETDEV_LAG_HASH_L34; 1311 case BOND_XMIT_POLICY_LAYER23: 1312 return NETDEV_LAG_HASH_L23; 1313 case BOND_XMIT_POLICY_ENCAP23: 1314 return NETDEV_LAG_HASH_E23; 1315 case BOND_XMIT_POLICY_ENCAP34: 1316 return NETDEV_LAG_HASH_E34; 1317 default: 1318 return NETDEV_LAG_HASH_UNKNOWN; 1319 } 1320 } 1321 1322 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1323 struct netlink_ext_ack *extack) 1324 { 1325 struct netdev_lag_upper_info lag_upper_info; 1326 enum netdev_lag_tx_type type; 1327 1328 type = bond_lag_tx_type(bond); 1329 lag_upper_info.tx_type = type; 1330 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1331 1332 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1333 &lag_upper_info, extack); 1334 } 1335 1336 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1337 { 1338 netdev_upper_dev_unlink(slave->dev, bond->dev); 1339 slave->dev->flags &= ~IFF_SLAVE; 1340 } 1341 1342 static struct slave *bond_alloc_slave(struct bonding *bond) 1343 { 1344 struct slave *slave = NULL; 1345 1346 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1347 if (!slave) 1348 return NULL; 1349 1350 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1351 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1352 GFP_KERNEL); 1353 if (!SLAVE_AD_INFO(slave)) { 1354 kfree(slave); 1355 return NULL; 1356 } 1357 } 1358 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1359 1360 return slave; 1361 } 1362 1363 static void bond_free_slave(struct slave *slave) 1364 { 1365 struct bonding *bond = bond_get_bond_by_slave(slave); 1366 1367 cancel_delayed_work_sync(&slave->notify_work); 1368 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1369 kfree(SLAVE_AD_INFO(slave)); 1370 1371 kfree(slave); 1372 } 1373 1374 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1375 { 1376 info->bond_mode = BOND_MODE(bond); 1377 info->miimon = bond->params.miimon; 1378 info->num_slaves = bond->slave_cnt; 1379 } 1380 1381 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1382 { 1383 strcpy(info->slave_name, slave->dev->name); 1384 info->link = slave->link; 1385 info->state = bond_slave_state(slave); 1386 info->link_failure_count = slave->link_failure_count; 1387 } 1388 1389 static void bond_netdev_notify_work(struct work_struct *_work) 1390 { 1391 struct slave *slave = container_of(_work, struct slave, 1392 notify_work.work); 1393 1394 if (rtnl_trylock()) { 1395 struct netdev_bonding_info binfo; 1396 1397 bond_fill_ifslave(slave, &binfo.slave); 1398 bond_fill_ifbond(slave->bond, &binfo.master); 1399 netdev_bonding_info_change(slave->dev, &binfo); 1400 rtnl_unlock(); 1401 } else { 1402 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1403 } 1404 } 1405 1406 void bond_queue_slave_event(struct slave *slave) 1407 { 1408 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1409 } 1410 1411 void bond_lower_state_changed(struct slave *slave) 1412 { 1413 struct netdev_lag_lower_state_info info; 1414 1415 info.link_up = slave->link == BOND_LINK_UP || 1416 slave->link == BOND_LINK_FAIL; 1417 info.tx_enabled = bond_is_active_slave(slave); 1418 netdev_lower_state_changed(slave->dev, &info); 1419 } 1420 1421 /* enslave device <slave> to bond device <master> */ 1422 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1423 struct netlink_ext_ack *extack) 1424 { 1425 struct bonding *bond = netdev_priv(bond_dev); 1426 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1427 struct slave *new_slave = NULL, *prev_slave; 1428 struct sockaddr_storage ss; 1429 int link_reporting; 1430 int res = 0, i; 1431 1432 if (!bond->params.use_carrier && 1433 slave_dev->ethtool_ops->get_link == NULL && 1434 slave_ops->ndo_do_ioctl == NULL) { 1435 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1436 } 1437 1438 /* already in-use? */ 1439 if (netdev_is_rx_handler_busy(slave_dev)) { 1440 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1441 slave_err(bond_dev, slave_dev, 1442 "Error: Device is in use and cannot be enslaved\n"); 1443 return -EBUSY; 1444 } 1445 1446 if (bond_dev == slave_dev) { 1447 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1448 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1449 return -EPERM; 1450 } 1451 1452 /* vlan challenged mutual exclusion */ 1453 /* no need to lock since we're protected by rtnl_lock */ 1454 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1455 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1456 if (vlan_uses_dev(bond_dev)) { 1457 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1458 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n"); 1459 return -EPERM; 1460 } else { 1461 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1462 } 1463 } else { 1464 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1465 } 1466 1467 /* Old ifenslave binaries are no longer supported. These can 1468 * be identified with moderate accuracy by the state of the slave: 1469 * the current ifenslave will set the interface down prior to 1470 * enslaving it; the old ifenslave will not. 1471 */ 1472 if (slave_dev->flags & IFF_UP) { 1473 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1474 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n"); 1475 return -EPERM; 1476 } 1477 1478 /* set bonding device ether type by slave - bonding netdevices are 1479 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1480 * there is a need to override some of the type dependent attribs/funcs. 1481 * 1482 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1483 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1484 */ 1485 if (!bond_has_slaves(bond)) { 1486 if (bond_dev->type != slave_dev->type) { 1487 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1488 bond_dev->type, slave_dev->type); 1489 1490 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1491 bond_dev); 1492 res = notifier_to_errno(res); 1493 if (res) { 1494 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1495 return -EBUSY; 1496 } 1497 1498 /* Flush unicast and multicast addresses */ 1499 dev_uc_flush(bond_dev); 1500 dev_mc_flush(bond_dev); 1501 1502 if (slave_dev->type != ARPHRD_ETHER) 1503 bond_setup_by_slave(bond_dev, slave_dev); 1504 else { 1505 ether_setup(bond_dev); 1506 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1507 } 1508 1509 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1510 bond_dev); 1511 } 1512 } else if (bond_dev->type != slave_dev->type) { 1513 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1514 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n", 1515 slave_dev->type, bond_dev->type); 1516 return -EINVAL; 1517 } 1518 1519 if (slave_dev->type == ARPHRD_INFINIBAND && 1520 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1521 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1522 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n", 1523 slave_dev->type); 1524 res = -EOPNOTSUPP; 1525 goto err_undo_flags; 1526 } 1527 1528 if (!slave_ops->ndo_set_mac_address || 1529 slave_dev->type == ARPHRD_INFINIBAND) { 1530 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1531 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1532 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1533 if (!bond_has_slaves(bond)) { 1534 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1535 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1536 } else { 1537 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1538 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n"); 1539 res = -EOPNOTSUPP; 1540 goto err_undo_flags; 1541 } 1542 } 1543 } 1544 1545 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1546 1547 /* If this is the first slave, then we need to set the master's hardware 1548 * address to be the same as the slave's. 1549 */ 1550 if (!bond_has_slaves(bond) && 1551 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1552 res = bond_set_dev_addr(bond->dev, slave_dev); 1553 if (res) 1554 goto err_undo_flags; 1555 } 1556 1557 new_slave = bond_alloc_slave(bond); 1558 if (!new_slave) { 1559 res = -ENOMEM; 1560 goto err_undo_flags; 1561 } 1562 1563 new_slave->bond = bond; 1564 new_slave->dev = slave_dev; 1565 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1566 * is set via sysfs or module option if desired. 1567 */ 1568 new_slave->queue_id = 0; 1569 1570 /* Save slave's original mtu and then set it to match the bond */ 1571 new_slave->original_mtu = slave_dev->mtu; 1572 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1573 if (res) { 1574 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1575 goto err_free; 1576 } 1577 1578 /* Save slave's original ("permanent") mac address for modes 1579 * that need it, and for restoring it upon release, and then 1580 * set it to the master's address 1581 */ 1582 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1583 slave_dev->addr_len); 1584 1585 if (!bond->params.fail_over_mac || 1586 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1587 /* Set slave to master's mac address. The application already 1588 * set the master's mac address to that of the first slave 1589 */ 1590 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1591 ss.ss_family = slave_dev->type; 1592 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1593 extack); 1594 if (res) { 1595 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1596 goto err_restore_mtu; 1597 } 1598 } 1599 1600 /* set slave flag before open to prevent IPv6 addrconf */ 1601 slave_dev->flags |= IFF_SLAVE; 1602 1603 /* open the slave since the application closed it */ 1604 res = dev_open(slave_dev, extack); 1605 if (res) { 1606 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1607 goto err_restore_mac; 1608 } 1609 1610 slave_dev->priv_flags |= IFF_BONDING; 1611 /* initialize slave stats */ 1612 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1613 1614 if (bond_is_lb(bond)) { 1615 /* bond_alb_init_slave() must be called before all other stages since 1616 * it might fail and we do not want to have to undo everything 1617 */ 1618 res = bond_alb_init_slave(bond, new_slave); 1619 if (res) 1620 goto err_close; 1621 } 1622 1623 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1624 if (res) { 1625 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1626 goto err_close; 1627 } 1628 1629 prev_slave = bond_last_slave(bond); 1630 1631 new_slave->delay = 0; 1632 new_slave->link_failure_count = 0; 1633 1634 if (bond_update_speed_duplex(new_slave) && 1635 bond_needs_speed_duplex(bond)) 1636 new_slave->link = BOND_LINK_DOWN; 1637 1638 new_slave->last_rx = jiffies - 1639 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1640 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1641 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1642 1643 if (bond->params.miimon && !bond->params.use_carrier) { 1644 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1645 1646 if ((link_reporting == -1) && !bond->params.arp_interval) { 1647 /* miimon is set but a bonded network driver 1648 * does not support ETHTOOL/MII and 1649 * arp_interval is not set. Note: if 1650 * use_carrier is enabled, we will never go 1651 * here (because netif_carrier is always 1652 * supported); thus, we don't need to change 1653 * the messages for netif_carrier. 1654 */ 1655 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n"); 1656 } else if (link_reporting == -1) { 1657 /* unable get link status using mii/ethtool */ 1658 slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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"); 1659 } 1660 } 1661 1662 /* check for initial state */ 1663 new_slave->link = BOND_LINK_NOCHANGE; 1664 if (bond->params.miimon) { 1665 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1666 if (bond->params.updelay) { 1667 bond_set_slave_link_state(new_slave, 1668 BOND_LINK_BACK, 1669 BOND_SLAVE_NOTIFY_NOW); 1670 new_slave->delay = bond->params.updelay; 1671 } else { 1672 bond_set_slave_link_state(new_slave, 1673 BOND_LINK_UP, 1674 BOND_SLAVE_NOTIFY_NOW); 1675 } 1676 } else { 1677 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1678 BOND_SLAVE_NOTIFY_NOW); 1679 } 1680 } else if (bond->params.arp_interval) { 1681 bond_set_slave_link_state(new_slave, 1682 (netif_carrier_ok(slave_dev) ? 1683 BOND_LINK_UP : BOND_LINK_DOWN), 1684 BOND_SLAVE_NOTIFY_NOW); 1685 } else { 1686 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1687 BOND_SLAVE_NOTIFY_NOW); 1688 } 1689 1690 if (new_slave->link != BOND_LINK_DOWN) 1691 new_slave->last_link_up = jiffies; 1692 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 1693 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1694 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1695 1696 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1697 /* if there is a primary slave, remember it */ 1698 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1699 rcu_assign_pointer(bond->primary_slave, new_slave); 1700 bond->force_primary = true; 1701 } 1702 } 1703 1704 switch (BOND_MODE(bond)) { 1705 case BOND_MODE_ACTIVEBACKUP: 1706 bond_set_slave_inactive_flags(new_slave, 1707 BOND_SLAVE_NOTIFY_NOW); 1708 break; 1709 case BOND_MODE_8023AD: 1710 /* in 802.3ad mode, the internal mechanism 1711 * will activate the slaves in the selected 1712 * aggregator 1713 */ 1714 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1715 /* if this is the first slave */ 1716 if (!prev_slave) { 1717 SLAVE_AD_INFO(new_slave)->id = 1; 1718 /* Initialize AD with the number of times that the AD timer is called in 1 second 1719 * can be called only after the mac address of the bond is set 1720 */ 1721 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1722 } else { 1723 SLAVE_AD_INFO(new_slave)->id = 1724 SLAVE_AD_INFO(prev_slave)->id + 1; 1725 } 1726 1727 bond_3ad_bind_slave(new_slave); 1728 break; 1729 case BOND_MODE_TLB: 1730 case BOND_MODE_ALB: 1731 bond_set_active_slave(new_slave); 1732 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1733 break; 1734 default: 1735 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 1736 1737 /* always active in trunk mode */ 1738 bond_set_active_slave(new_slave); 1739 1740 /* In trunking mode there is little meaning to curr_active_slave 1741 * anyway (it holds no special properties of the bond device), 1742 * so we can change it without calling change_active_interface() 1743 */ 1744 if (!rcu_access_pointer(bond->curr_active_slave) && 1745 new_slave->link == BOND_LINK_UP) 1746 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1747 1748 break; 1749 } /* switch(bond_mode) */ 1750 1751 #ifdef CONFIG_NET_POLL_CONTROLLER 1752 if (bond->dev->npinfo) { 1753 if (slave_enable_netpoll(new_slave)) { 1754 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1755 res = -EBUSY; 1756 goto err_detach; 1757 } 1758 } 1759 #endif 1760 1761 if (!(bond_dev->features & NETIF_F_LRO)) 1762 dev_disable_lro(slave_dev); 1763 1764 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1765 new_slave); 1766 if (res) { 1767 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 1768 goto err_detach; 1769 } 1770 1771 res = bond_master_upper_dev_link(bond, new_slave, extack); 1772 if (res) { 1773 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1774 goto err_unregister; 1775 } 1776 1777 res = bond_sysfs_slave_add(new_slave); 1778 if (res) { 1779 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1780 goto err_upper_unlink; 1781 } 1782 1783 /* If the mode uses primary, then the following is handled by 1784 * bond_change_active_slave(). 1785 */ 1786 if (!bond_uses_primary(bond)) { 1787 /* set promiscuity level to new slave */ 1788 if (bond_dev->flags & IFF_PROMISC) { 1789 res = dev_set_promiscuity(slave_dev, 1); 1790 if (res) 1791 goto err_sysfs_del; 1792 } 1793 1794 /* set allmulti level to new slave */ 1795 if (bond_dev->flags & IFF_ALLMULTI) { 1796 res = dev_set_allmulti(slave_dev, 1); 1797 if (res) { 1798 if (bond_dev->flags & IFF_PROMISC) 1799 dev_set_promiscuity(slave_dev, -1); 1800 goto err_sysfs_del; 1801 } 1802 } 1803 1804 netif_addr_lock_bh(bond_dev); 1805 dev_mc_sync_multiple(slave_dev, bond_dev); 1806 dev_uc_sync_multiple(slave_dev, bond_dev); 1807 netif_addr_unlock_bh(bond_dev); 1808 1809 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1810 /* add lacpdu mc addr to mc list */ 1811 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1812 1813 dev_mc_add(slave_dev, lacpdu_multicast); 1814 } 1815 } 1816 1817 bond->slave_cnt++; 1818 bond_compute_features(bond); 1819 bond_set_carrier(bond); 1820 1821 if (bond_uses_primary(bond)) { 1822 block_netpoll_tx(); 1823 bond_select_active_slave(bond); 1824 unblock_netpoll_tx(); 1825 } 1826 1827 if (bond_mode_can_use_xmit_hash(bond)) 1828 bond_update_slave_arr(bond, NULL); 1829 1830 1831 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 1832 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1833 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1834 1835 /* enslave is successful */ 1836 bond_queue_slave_event(new_slave); 1837 return 0; 1838 1839 /* Undo stages on error */ 1840 err_sysfs_del: 1841 bond_sysfs_slave_del(new_slave); 1842 1843 err_upper_unlink: 1844 bond_upper_dev_unlink(bond, new_slave); 1845 1846 err_unregister: 1847 netdev_rx_handler_unregister(slave_dev); 1848 1849 err_detach: 1850 vlan_vids_del_by_dev(slave_dev, bond_dev); 1851 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1852 RCU_INIT_POINTER(bond->primary_slave, NULL); 1853 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1854 block_netpoll_tx(); 1855 bond_change_active_slave(bond, NULL); 1856 bond_select_active_slave(bond); 1857 unblock_netpoll_tx(); 1858 } 1859 /* either primary_slave or curr_active_slave might've changed */ 1860 synchronize_rcu(); 1861 slave_disable_netpoll(new_slave); 1862 1863 err_close: 1864 if (!netif_is_bond_master(slave_dev)) 1865 slave_dev->priv_flags &= ~IFF_BONDING; 1866 dev_close(slave_dev); 1867 1868 err_restore_mac: 1869 slave_dev->flags &= ~IFF_SLAVE; 1870 if (!bond->params.fail_over_mac || 1871 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1872 /* XXX TODO - fom follow mode needs to change master's 1873 * MAC if this slave's MAC is in use by the bond, or at 1874 * least print a warning. 1875 */ 1876 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 1877 new_slave->dev->addr_len); 1878 ss.ss_family = slave_dev->type; 1879 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 1880 } 1881 1882 err_restore_mtu: 1883 dev_set_mtu(slave_dev, new_slave->original_mtu); 1884 1885 err_free: 1886 bond_free_slave(new_slave); 1887 1888 err_undo_flags: 1889 /* Enslave of first slave has failed and we need to fix master's mac */ 1890 if (!bond_has_slaves(bond)) { 1891 if (ether_addr_equal_64bits(bond_dev->dev_addr, 1892 slave_dev->dev_addr)) 1893 eth_hw_addr_random(bond_dev); 1894 if (bond_dev->type != ARPHRD_ETHER) { 1895 dev_close(bond_dev); 1896 ether_setup(bond_dev); 1897 bond_dev->flags |= IFF_MASTER; 1898 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1899 } 1900 } 1901 1902 return res; 1903 } 1904 1905 /* Try to release the slave device <slave> from the bond device <master> 1906 * It is legal to access curr_active_slave without a lock because all the function 1907 * is RTNL-locked. If "all" is true it means that the function is being called 1908 * while destroying a bond interface and all slaves are being released. 1909 * 1910 * The rules for slave state should be: 1911 * for Active/Backup: 1912 * Active stays on all backups go down 1913 * for Bonded connections: 1914 * The first up interface should be left on and all others downed. 1915 */ 1916 static int __bond_release_one(struct net_device *bond_dev, 1917 struct net_device *slave_dev, 1918 bool all, bool unregister) 1919 { 1920 struct bonding *bond = netdev_priv(bond_dev); 1921 struct slave *slave, *oldcurrent; 1922 struct sockaddr_storage ss; 1923 int old_flags = bond_dev->flags; 1924 netdev_features_t old_features = bond_dev->features; 1925 1926 /* slave is not a slave or master is not master of this slave */ 1927 if (!(slave_dev->flags & IFF_SLAVE) || 1928 !netdev_has_upper_dev(slave_dev, bond_dev)) { 1929 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 1930 return -EINVAL; 1931 } 1932 1933 block_netpoll_tx(); 1934 1935 slave = bond_get_slave_by_dev(bond, slave_dev); 1936 if (!slave) { 1937 /* not a slave of this bond */ 1938 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 1939 unblock_netpoll_tx(); 1940 return -EINVAL; 1941 } 1942 1943 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 1944 1945 bond_sysfs_slave_del(slave); 1946 1947 /* recompute stats just before removing the slave */ 1948 bond_get_stats(bond->dev, &bond->bond_stats); 1949 1950 bond_upper_dev_unlink(bond, slave); 1951 /* unregister rx_handler early so bond_handle_frame wouldn't be called 1952 * for this slave anymore. 1953 */ 1954 netdev_rx_handler_unregister(slave_dev); 1955 1956 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1957 bond_3ad_unbind_slave(slave); 1958 1959 if (bond_mode_can_use_xmit_hash(bond)) 1960 bond_update_slave_arr(bond, slave); 1961 1962 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 1963 bond_is_active_slave(slave) ? "active" : "backup"); 1964 1965 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 1966 1967 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 1968 1969 if (!all && (!bond->params.fail_over_mac || 1970 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 1971 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 1972 bond_has_slaves(bond)) 1973 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n", 1974 slave->perm_hwaddr); 1975 } 1976 1977 if (rtnl_dereference(bond->primary_slave) == slave) 1978 RCU_INIT_POINTER(bond->primary_slave, NULL); 1979 1980 if (oldcurrent == slave) 1981 bond_change_active_slave(bond, NULL); 1982 1983 if (bond_is_lb(bond)) { 1984 /* Must be called only after the slave has been 1985 * detached from the list and the curr_active_slave 1986 * has been cleared (if our_slave == old_current), 1987 * but before a new active slave is selected. 1988 */ 1989 bond_alb_deinit_slave(bond, slave); 1990 } 1991 1992 if (all) { 1993 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 1994 } else if (oldcurrent == slave) { 1995 /* Note that we hold RTNL over this sequence, so there 1996 * is no concern that another slave add/remove event 1997 * will interfere. 1998 */ 1999 bond_select_active_slave(bond); 2000 } 2001 2002 if (!bond_has_slaves(bond)) { 2003 bond_set_carrier(bond); 2004 eth_hw_addr_random(bond_dev); 2005 } 2006 2007 unblock_netpoll_tx(); 2008 synchronize_rcu(); 2009 bond->slave_cnt--; 2010 2011 if (!bond_has_slaves(bond)) { 2012 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2013 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2014 } 2015 2016 bond_compute_features(bond); 2017 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2018 (old_features & NETIF_F_VLAN_CHALLENGED)) 2019 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2020 2021 vlan_vids_del_by_dev(slave_dev, bond_dev); 2022 2023 /* If the mode uses primary, then this case was handled above by 2024 * bond_change_active_slave(..., NULL) 2025 */ 2026 if (!bond_uses_primary(bond)) { 2027 /* unset promiscuity level from slave 2028 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2029 * of the IFF_PROMISC flag in the bond_dev, but we need the 2030 * value of that flag before that change, as that was the value 2031 * when this slave was attached, so we cache at the start of the 2032 * function and use it here. Same goes for ALLMULTI below 2033 */ 2034 if (old_flags & IFF_PROMISC) 2035 dev_set_promiscuity(slave_dev, -1); 2036 2037 /* unset allmulti level from slave */ 2038 if (old_flags & IFF_ALLMULTI) 2039 dev_set_allmulti(slave_dev, -1); 2040 2041 bond_hw_addr_flush(bond_dev, slave_dev); 2042 } 2043 2044 slave_disable_netpoll(slave); 2045 2046 /* close slave before restoring its mac address */ 2047 dev_close(slave_dev); 2048 2049 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2050 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2051 /* restore original ("permanent") mac address */ 2052 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2053 slave->dev->addr_len); 2054 ss.ss_family = slave_dev->type; 2055 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2056 } 2057 2058 if (unregister) 2059 __dev_set_mtu(slave_dev, slave->original_mtu); 2060 else 2061 dev_set_mtu(slave_dev, slave->original_mtu); 2062 2063 if (!netif_is_bond_master(slave_dev)) 2064 slave_dev->priv_flags &= ~IFF_BONDING; 2065 2066 bond_free_slave(slave); 2067 2068 return 0; 2069 } 2070 2071 /* A wrapper used because of ndo_del_link */ 2072 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2073 { 2074 return __bond_release_one(bond_dev, slave_dev, false, false); 2075 } 2076 2077 /* First release a slave and then destroy the bond if no more slaves are left. 2078 * Must be under rtnl_lock when this function is called. 2079 */ 2080 static int bond_release_and_destroy(struct net_device *bond_dev, 2081 struct net_device *slave_dev) 2082 { 2083 struct bonding *bond = netdev_priv(bond_dev); 2084 int ret; 2085 2086 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2087 if (ret == 0 && !bond_has_slaves(bond)) { 2088 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2089 netdev_info(bond_dev, "Destroying bond\n"); 2090 bond_remove_proc_entry(bond); 2091 unregister_netdevice(bond_dev); 2092 } 2093 return ret; 2094 } 2095 2096 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2097 { 2098 struct bonding *bond = netdev_priv(bond_dev); 2099 bond_fill_ifbond(bond, info); 2100 } 2101 2102 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2103 { 2104 struct bonding *bond = netdev_priv(bond_dev); 2105 struct list_head *iter; 2106 int i = 0, res = -ENODEV; 2107 struct slave *slave; 2108 2109 bond_for_each_slave(bond, slave, iter) { 2110 if (i++ == (int)info->slave_id) { 2111 res = 0; 2112 bond_fill_ifslave(slave, info); 2113 break; 2114 } 2115 } 2116 2117 return res; 2118 } 2119 2120 /*-------------------------------- Monitoring -------------------------------*/ 2121 2122 /* called with rcu_read_lock() */ 2123 static int bond_miimon_inspect(struct bonding *bond) 2124 { 2125 int link_state, commit = 0; 2126 struct list_head *iter; 2127 struct slave *slave; 2128 bool ignore_updelay; 2129 2130 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2131 2132 bond_for_each_slave_rcu(bond, slave, iter) { 2133 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2134 2135 link_state = bond_check_dev_link(bond, slave->dev, 0); 2136 2137 switch (slave->link) { 2138 case BOND_LINK_UP: 2139 if (link_state) 2140 continue; 2141 2142 bond_propose_link_state(slave, BOND_LINK_FAIL); 2143 commit++; 2144 slave->delay = bond->params.downdelay; 2145 if (slave->delay) { 2146 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2147 (BOND_MODE(bond) == 2148 BOND_MODE_ACTIVEBACKUP) ? 2149 (bond_is_active_slave(slave) ? 2150 "active " : "backup ") : "", 2151 bond->params.downdelay * bond->params.miimon); 2152 } 2153 /*FALLTHRU*/ 2154 case BOND_LINK_FAIL: 2155 if (link_state) { 2156 /* recovered before downdelay expired */ 2157 bond_propose_link_state(slave, BOND_LINK_UP); 2158 slave->last_link_up = jiffies; 2159 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2160 (bond->params.downdelay - slave->delay) * 2161 bond->params.miimon); 2162 commit++; 2163 continue; 2164 } 2165 2166 if (slave->delay <= 0) { 2167 bond_propose_link_state(slave, BOND_LINK_DOWN); 2168 commit++; 2169 continue; 2170 } 2171 2172 slave->delay--; 2173 break; 2174 2175 case BOND_LINK_DOWN: 2176 if (!link_state) 2177 continue; 2178 2179 bond_propose_link_state(slave, BOND_LINK_BACK); 2180 commit++; 2181 slave->delay = bond->params.updelay; 2182 2183 if (slave->delay) { 2184 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2185 ignore_updelay ? 0 : 2186 bond->params.updelay * 2187 bond->params.miimon); 2188 } 2189 /*FALLTHRU*/ 2190 case BOND_LINK_BACK: 2191 if (!link_state) { 2192 bond_propose_link_state(slave, BOND_LINK_DOWN); 2193 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2194 (bond->params.updelay - slave->delay) * 2195 bond->params.miimon); 2196 commit++; 2197 continue; 2198 } 2199 2200 if (ignore_updelay) 2201 slave->delay = 0; 2202 2203 if (slave->delay <= 0) { 2204 bond_propose_link_state(slave, BOND_LINK_UP); 2205 commit++; 2206 ignore_updelay = false; 2207 continue; 2208 } 2209 2210 slave->delay--; 2211 break; 2212 } 2213 } 2214 2215 return commit; 2216 } 2217 2218 static void bond_miimon_link_change(struct bonding *bond, 2219 struct slave *slave, 2220 char link) 2221 { 2222 switch (BOND_MODE(bond)) { 2223 case BOND_MODE_8023AD: 2224 bond_3ad_handle_link_change(slave, link); 2225 break; 2226 case BOND_MODE_TLB: 2227 case BOND_MODE_ALB: 2228 bond_alb_handle_link_change(bond, slave, link); 2229 break; 2230 case BOND_MODE_XOR: 2231 bond_update_slave_arr(bond, NULL); 2232 break; 2233 } 2234 } 2235 2236 static void bond_miimon_commit(struct bonding *bond) 2237 { 2238 struct list_head *iter; 2239 struct slave *slave, *primary; 2240 2241 bond_for_each_slave(bond, slave, iter) { 2242 switch (slave->link_new_state) { 2243 case BOND_LINK_NOCHANGE: 2244 /* For 802.3ad mode, check current slave speed and 2245 * duplex again in case its port was disabled after 2246 * invalid speed/duplex reporting but recovered before 2247 * link monitoring could make a decision on the actual 2248 * link status 2249 */ 2250 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2251 slave->link == BOND_LINK_UP) 2252 bond_3ad_adapter_speed_duplex_changed(slave); 2253 continue; 2254 2255 case BOND_LINK_UP: 2256 if (bond_update_speed_duplex(slave) && 2257 bond_needs_speed_duplex(bond)) { 2258 slave->link = BOND_LINK_DOWN; 2259 if (net_ratelimit()) 2260 slave_warn(bond->dev, slave->dev, 2261 "failed to get link speed/duplex\n"); 2262 continue; 2263 } 2264 bond_set_slave_link_state(slave, BOND_LINK_UP, 2265 BOND_SLAVE_NOTIFY_NOW); 2266 slave->last_link_up = jiffies; 2267 2268 primary = rtnl_dereference(bond->primary_slave); 2269 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2270 /* prevent it from being the active one */ 2271 bond_set_backup_slave(slave); 2272 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2273 /* make it immediately active */ 2274 bond_set_active_slave(slave); 2275 } 2276 2277 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2278 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2279 slave->duplex ? "full" : "half"); 2280 2281 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2282 2283 if (!bond->curr_active_slave || slave == primary) 2284 goto do_failover; 2285 2286 continue; 2287 2288 case BOND_LINK_DOWN: 2289 if (slave->link_failure_count < UINT_MAX) 2290 slave->link_failure_count++; 2291 2292 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2293 BOND_SLAVE_NOTIFY_NOW); 2294 2295 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2296 BOND_MODE(bond) == BOND_MODE_8023AD) 2297 bond_set_slave_inactive_flags(slave, 2298 BOND_SLAVE_NOTIFY_NOW); 2299 2300 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2301 2302 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2303 2304 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2305 goto do_failover; 2306 2307 continue; 2308 2309 default: 2310 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2311 slave->link_new_state); 2312 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2313 2314 continue; 2315 } 2316 2317 do_failover: 2318 block_netpoll_tx(); 2319 bond_select_active_slave(bond); 2320 unblock_netpoll_tx(); 2321 } 2322 2323 bond_set_carrier(bond); 2324 } 2325 2326 /* bond_mii_monitor 2327 * 2328 * Really a wrapper that splits the mii monitor into two phases: an 2329 * inspection, then (if inspection indicates something needs to be done) 2330 * an acquisition of appropriate locks followed by a commit phase to 2331 * implement whatever link state changes are indicated. 2332 */ 2333 static void bond_mii_monitor(struct work_struct *work) 2334 { 2335 struct bonding *bond = container_of(work, struct bonding, 2336 mii_work.work); 2337 bool should_notify_peers = false; 2338 bool commit; 2339 unsigned long delay; 2340 struct slave *slave; 2341 struct list_head *iter; 2342 2343 delay = msecs_to_jiffies(bond->params.miimon); 2344 2345 if (!bond_has_slaves(bond)) 2346 goto re_arm; 2347 2348 rcu_read_lock(); 2349 should_notify_peers = bond_should_notify_peers(bond); 2350 commit = !!bond_miimon_inspect(bond); 2351 if (bond->send_peer_notif) { 2352 rcu_read_unlock(); 2353 if (rtnl_trylock()) { 2354 bond->send_peer_notif--; 2355 rtnl_unlock(); 2356 } 2357 } else { 2358 rcu_read_unlock(); 2359 } 2360 2361 if (commit) { 2362 /* Race avoidance with bond_close cancel of workqueue */ 2363 if (!rtnl_trylock()) { 2364 delay = 1; 2365 should_notify_peers = false; 2366 goto re_arm; 2367 } 2368 2369 bond_for_each_slave(bond, slave, iter) { 2370 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2371 } 2372 bond_miimon_commit(bond); 2373 2374 rtnl_unlock(); /* might sleep, hold no other locks */ 2375 } 2376 2377 re_arm: 2378 if (bond->params.miimon) 2379 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2380 2381 if (should_notify_peers) { 2382 if (!rtnl_trylock()) 2383 return; 2384 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2385 rtnl_unlock(); 2386 } 2387 } 2388 2389 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2390 { 2391 __be32 ip = *((__be32 *)data); 2392 2393 return ip == bond_confirm_addr(upper, 0, ip); 2394 } 2395 2396 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2397 { 2398 bool ret = false; 2399 2400 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2401 return true; 2402 2403 rcu_read_lock(); 2404 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2405 ret = true; 2406 rcu_read_unlock(); 2407 2408 return ret; 2409 } 2410 2411 /* We go to the (large) trouble of VLAN tagging ARP frames because 2412 * switches in VLAN mode (especially if ports are configured as 2413 * "native" to a VLAN) might not pass non-tagged frames. 2414 */ 2415 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2416 __be32 src_ip, struct bond_vlan_tag *tags) 2417 { 2418 struct sk_buff *skb; 2419 struct bond_vlan_tag *outer_tag = tags; 2420 struct net_device *slave_dev = slave->dev; 2421 struct net_device *bond_dev = slave->bond->dev; 2422 2423 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2424 arp_op, &dest_ip, &src_ip); 2425 2426 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2427 NULL, slave_dev->dev_addr, NULL); 2428 2429 if (!skb) { 2430 net_err_ratelimited("ARP packet allocation failed\n"); 2431 return; 2432 } 2433 2434 if (!tags || tags->vlan_proto == VLAN_N_VID) 2435 goto xmit; 2436 2437 tags++; 2438 2439 /* Go through all the tags backwards and add them to the packet */ 2440 while (tags->vlan_proto != VLAN_N_VID) { 2441 if (!tags->vlan_id) { 2442 tags++; 2443 continue; 2444 } 2445 2446 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2447 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2448 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2449 tags->vlan_id); 2450 if (!skb) { 2451 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2452 return; 2453 } 2454 2455 tags++; 2456 } 2457 /* Set the outer tag */ 2458 if (outer_tag->vlan_id) { 2459 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2460 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2461 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2462 outer_tag->vlan_id); 2463 } 2464 2465 xmit: 2466 arp_xmit(skb); 2467 } 2468 2469 /* Validate the device path between the @start_dev and the @end_dev. 2470 * The path is valid if the @end_dev is reachable through device 2471 * stacking. 2472 * When the path is validated, collect any vlan information in the 2473 * path. 2474 */ 2475 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2476 struct net_device *end_dev, 2477 int level) 2478 { 2479 struct bond_vlan_tag *tags; 2480 struct net_device *upper; 2481 struct list_head *iter; 2482 2483 if (start_dev == end_dev) { 2484 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2485 if (!tags) 2486 return ERR_PTR(-ENOMEM); 2487 tags[level].vlan_proto = VLAN_N_VID; 2488 return tags; 2489 } 2490 2491 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2492 tags = bond_verify_device_path(upper, end_dev, level + 1); 2493 if (IS_ERR_OR_NULL(tags)) { 2494 if (IS_ERR(tags)) 2495 return tags; 2496 continue; 2497 } 2498 if (is_vlan_dev(upper)) { 2499 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2500 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2501 } 2502 2503 return tags; 2504 } 2505 2506 return NULL; 2507 } 2508 2509 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2510 { 2511 struct rtable *rt; 2512 struct bond_vlan_tag *tags; 2513 __be32 *targets = bond->params.arp_targets, addr; 2514 int i; 2515 2516 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2517 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2518 __func__, &targets[i]); 2519 tags = NULL; 2520 2521 /* Find out through which dev should the packet go */ 2522 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2523 RTO_ONLINK, 0); 2524 if (IS_ERR(rt)) { 2525 /* there's no route to target - try to send arp 2526 * probe to generate any traffic (arp_validate=0) 2527 */ 2528 if (bond->params.arp_validate) 2529 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2530 bond->dev->name, 2531 &targets[i]); 2532 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2533 0, tags); 2534 continue; 2535 } 2536 2537 /* bond device itself */ 2538 if (rt->dst.dev == bond->dev) 2539 goto found; 2540 2541 rcu_read_lock(); 2542 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2543 rcu_read_unlock(); 2544 2545 if (!IS_ERR_OR_NULL(tags)) 2546 goto found; 2547 2548 /* Not our device - skip */ 2549 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2550 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2551 2552 ip_rt_put(rt); 2553 continue; 2554 2555 found: 2556 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2557 ip_rt_put(rt); 2558 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2559 kfree(tags); 2560 } 2561 } 2562 2563 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2564 { 2565 int i; 2566 2567 if (!sip || !bond_has_this_ip(bond, tip)) { 2568 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2569 __func__, &sip, &tip); 2570 return; 2571 } 2572 2573 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2574 if (i == -1) { 2575 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2576 __func__, &sip); 2577 return; 2578 } 2579 slave->last_rx = jiffies; 2580 slave->target_last_arp_rx[i] = jiffies; 2581 } 2582 2583 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2584 struct slave *slave) 2585 { 2586 struct arphdr *arp = (struct arphdr *)skb->data; 2587 struct slave *curr_active_slave, *curr_arp_slave; 2588 unsigned char *arp_ptr; 2589 __be32 sip, tip; 2590 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2591 unsigned int alen; 2592 2593 if (!slave_do_arp_validate(bond, slave)) { 2594 if ((slave_do_arp_validate_only(bond) && is_arp) || 2595 !slave_do_arp_validate_only(bond)) 2596 slave->last_rx = jiffies; 2597 return RX_HANDLER_ANOTHER; 2598 } else if (!is_arp) { 2599 return RX_HANDLER_ANOTHER; 2600 } 2601 2602 alen = arp_hdr_len(bond->dev); 2603 2604 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 2605 __func__, skb->dev->name); 2606 2607 if (alen > skb_headlen(skb)) { 2608 arp = kmalloc(alen, GFP_ATOMIC); 2609 if (!arp) 2610 goto out_unlock; 2611 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2612 goto out_unlock; 2613 } 2614 2615 if (arp->ar_hln != bond->dev->addr_len || 2616 skb->pkt_type == PACKET_OTHERHOST || 2617 skb->pkt_type == PACKET_LOOPBACK || 2618 arp->ar_hrd != htons(ARPHRD_ETHER) || 2619 arp->ar_pro != htons(ETH_P_IP) || 2620 arp->ar_pln != 4) 2621 goto out_unlock; 2622 2623 arp_ptr = (unsigned char *)(arp + 1); 2624 arp_ptr += bond->dev->addr_len; 2625 memcpy(&sip, arp_ptr, 4); 2626 arp_ptr += 4 + bond->dev->addr_len; 2627 memcpy(&tip, arp_ptr, 4); 2628 2629 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2630 __func__, slave->dev->name, bond_slave_state(slave), 2631 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2632 &sip, &tip); 2633 2634 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2635 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2636 2637 /* We 'trust' the received ARP enough to validate it if: 2638 * 2639 * (a) the slave receiving the ARP is active (which includes the 2640 * current ARP slave, if any), or 2641 * 2642 * (b) the receiving slave isn't active, but there is a currently 2643 * active slave and it received valid arp reply(s) after it became 2644 * the currently active slave, or 2645 * 2646 * (c) there is an ARP slave that sent an ARP during the prior ARP 2647 * interval, and we receive an ARP reply on any slave. We accept 2648 * these because switch FDB update delays may deliver the ARP 2649 * reply to a slave other than the sender of the ARP request. 2650 * 2651 * Note: for (b), backup slaves are receiving the broadcast ARP 2652 * request, not a reply. This request passes from the sending 2653 * slave through the L2 switch(es) to the receiving slave. Since 2654 * this is checking the request, sip/tip are swapped for 2655 * validation. 2656 * 2657 * This is done to avoid endless looping when we can't reach the 2658 * arp_ip_target and fool ourselves with our own arp requests. 2659 */ 2660 if (bond_is_active_slave(slave)) 2661 bond_validate_arp(bond, slave, sip, tip); 2662 else if (curr_active_slave && 2663 time_after(slave_last_rx(bond, curr_active_slave), 2664 curr_active_slave->last_link_up)) 2665 bond_validate_arp(bond, slave, tip, sip); 2666 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2667 bond_time_in_interval(bond, 2668 dev_trans_start(curr_arp_slave->dev), 1)) 2669 bond_validate_arp(bond, slave, sip, tip); 2670 2671 out_unlock: 2672 if (arp != (struct arphdr *)skb->data) 2673 kfree(arp); 2674 return RX_HANDLER_ANOTHER; 2675 } 2676 2677 /* function to verify if we're in the arp_interval timeslice, returns true if 2678 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2679 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2680 */ 2681 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2682 int mod) 2683 { 2684 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2685 2686 return time_in_range(jiffies, 2687 last_act - delta_in_ticks, 2688 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2689 } 2690 2691 /* This function is called regularly to monitor each slave's link 2692 * ensuring that traffic is being sent and received when arp monitoring 2693 * is used in load-balancing mode. if the adapter has been dormant, then an 2694 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2695 * arp monitoring in active backup mode. 2696 */ 2697 static void bond_loadbalance_arp_mon(struct bonding *bond) 2698 { 2699 struct slave *slave, *oldcurrent; 2700 struct list_head *iter; 2701 int do_failover = 0, slave_state_changed = 0; 2702 2703 if (!bond_has_slaves(bond)) 2704 goto re_arm; 2705 2706 rcu_read_lock(); 2707 2708 oldcurrent = rcu_dereference(bond->curr_active_slave); 2709 /* see if any of the previous devices are up now (i.e. they have 2710 * xmt and rcv traffic). the curr_active_slave does not come into 2711 * the picture unless it is null. also, slave->last_link_up is not 2712 * needed here because we send an arp on each slave and give a slave 2713 * as long as it needs to get the tx/rx within the delta. 2714 * TODO: what about up/down delay in arp mode? it wasn't here before 2715 * so it can wait 2716 */ 2717 bond_for_each_slave_rcu(bond, slave, iter) { 2718 unsigned long trans_start = dev_trans_start(slave->dev); 2719 2720 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2721 2722 if (slave->link != BOND_LINK_UP) { 2723 if (bond_time_in_interval(bond, trans_start, 1) && 2724 bond_time_in_interval(bond, slave->last_rx, 1)) { 2725 2726 bond_propose_link_state(slave, BOND_LINK_UP); 2727 slave_state_changed = 1; 2728 2729 /* primary_slave has no meaning in round-robin 2730 * mode. the window of a slave being up and 2731 * curr_active_slave being null after enslaving 2732 * is closed. 2733 */ 2734 if (!oldcurrent) { 2735 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 2736 do_failover = 1; 2737 } else { 2738 slave_info(bond->dev, slave->dev, "interface is now up\n"); 2739 } 2740 } 2741 } else { 2742 /* slave->link == BOND_LINK_UP */ 2743 2744 /* not all switches will respond to an arp request 2745 * when the source ip is 0, so don't take the link down 2746 * if we don't know our ip yet 2747 */ 2748 if (!bond_time_in_interval(bond, trans_start, 2) || 2749 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2750 2751 bond_propose_link_state(slave, BOND_LINK_DOWN); 2752 slave_state_changed = 1; 2753 2754 if (slave->link_failure_count < UINT_MAX) 2755 slave->link_failure_count++; 2756 2757 slave_info(bond->dev, slave->dev, "interface is now down\n"); 2758 2759 if (slave == oldcurrent) 2760 do_failover = 1; 2761 } 2762 } 2763 2764 /* note: if switch is in round-robin mode, all links 2765 * must tx arp to ensure all links rx an arp - otherwise 2766 * links may oscillate or not come up at all; if switch is 2767 * in something like xor mode, there is nothing we can 2768 * do - all replies will be rx'ed on same link causing slaves 2769 * to be unstable during low/no traffic periods 2770 */ 2771 if (bond_slave_is_up(slave)) 2772 bond_arp_send_all(bond, slave); 2773 } 2774 2775 rcu_read_unlock(); 2776 2777 if (do_failover || slave_state_changed) { 2778 if (!rtnl_trylock()) 2779 goto re_arm; 2780 2781 bond_for_each_slave(bond, slave, iter) { 2782 if (slave->link_new_state != BOND_LINK_NOCHANGE) 2783 slave->link = slave->link_new_state; 2784 } 2785 2786 if (slave_state_changed) { 2787 bond_slave_state_change(bond); 2788 if (BOND_MODE(bond) == BOND_MODE_XOR) 2789 bond_update_slave_arr(bond, NULL); 2790 } 2791 if (do_failover) { 2792 block_netpoll_tx(); 2793 bond_select_active_slave(bond); 2794 unblock_netpoll_tx(); 2795 } 2796 rtnl_unlock(); 2797 } 2798 2799 re_arm: 2800 if (bond->params.arp_interval) 2801 queue_delayed_work(bond->wq, &bond->arp_work, 2802 msecs_to_jiffies(bond->params.arp_interval)); 2803 } 2804 2805 /* Called to inspect slaves for active-backup mode ARP monitor link state 2806 * changes. Sets proposed link state in slaves to specify what action 2807 * should take place for the slave. Returns 0 if no changes are found, >0 2808 * if changes to link states must be committed. 2809 * 2810 * Called with rcu_read_lock held. 2811 */ 2812 static int bond_ab_arp_inspect(struct bonding *bond) 2813 { 2814 unsigned long trans_start, last_rx; 2815 struct list_head *iter; 2816 struct slave *slave; 2817 int commit = 0; 2818 2819 bond_for_each_slave_rcu(bond, slave, iter) { 2820 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2821 last_rx = slave_last_rx(bond, slave); 2822 2823 if (slave->link != BOND_LINK_UP) { 2824 if (bond_time_in_interval(bond, last_rx, 1)) { 2825 bond_propose_link_state(slave, BOND_LINK_UP); 2826 commit++; 2827 } 2828 continue; 2829 } 2830 2831 /* Give slaves 2*delta after being enslaved or made 2832 * active. This avoids bouncing, as the last receive 2833 * times need a full ARP monitor cycle to be updated. 2834 */ 2835 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2836 continue; 2837 2838 /* Backup slave is down if: 2839 * - No current_arp_slave AND 2840 * - more than 3*delta since last receive AND 2841 * - the bond has an IP address 2842 * 2843 * Note: a non-null current_arp_slave indicates 2844 * the curr_active_slave went down and we are 2845 * searching for a new one; under this condition 2846 * we only take the curr_active_slave down - this 2847 * gives each slave a chance to tx/rx traffic 2848 * before being taken out 2849 */ 2850 if (!bond_is_active_slave(slave) && 2851 !rcu_access_pointer(bond->current_arp_slave) && 2852 !bond_time_in_interval(bond, last_rx, 3)) { 2853 bond_propose_link_state(slave, BOND_LINK_DOWN); 2854 commit++; 2855 } 2856 2857 /* Active slave is down if: 2858 * - more than 2*delta since transmitting OR 2859 * - (more than 2*delta since receive AND 2860 * the bond has an IP address) 2861 */ 2862 trans_start = dev_trans_start(slave->dev); 2863 if (bond_is_active_slave(slave) && 2864 (!bond_time_in_interval(bond, trans_start, 2) || 2865 !bond_time_in_interval(bond, last_rx, 2))) { 2866 bond_propose_link_state(slave, BOND_LINK_DOWN); 2867 commit++; 2868 } 2869 } 2870 2871 return commit; 2872 } 2873 2874 /* Called to commit link state changes noted by inspection step of 2875 * active-backup mode ARP monitor. 2876 * 2877 * Called with RTNL hold. 2878 */ 2879 static void bond_ab_arp_commit(struct bonding *bond) 2880 { 2881 unsigned long trans_start; 2882 struct list_head *iter; 2883 struct slave *slave; 2884 2885 bond_for_each_slave(bond, slave, iter) { 2886 switch (slave->link_new_state) { 2887 case BOND_LINK_NOCHANGE: 2888 continue; 2889 2890 case BOND_LINK_UP: 2891 trans_start = dev_trans_start(slave->dev); 2892 if (rtnl_dereference(bond->curr_active_slave) != slave || 2893 (!rtnl_dereference(bond->curr_active_slave) && 2894 bond_time_in_interval(bond, trans_start, 1))) { 2895 struct slave *current_arp_slave; 2896 2897 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 2898 bond_set_slave_link_state(slave, BOND_LINK_UP, 2899 BOND_SLAVE_NOTIFY_NOW); 2900 if (current_arp_slave) { 2901 bond_set_slave_inactive_flags( 2902 current_arp_slave, 2903 BOND_SLAVE_NOTIFY_NOW); 2904 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2905 } 2906 2907 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 2908 2909 if (!rtnl_dereference(bond->curr_active_slave) || 2910 slave == rtnl_dereference(bond->primary_slave)) 2911 goto do_failover; 2912 2913 } 2914 2915 continue; 2916 2917 case BOND_LINK_DOWN: 2918 if (slave->link_failure_count < UINT_MAX) 2919 slave->link_failure_count++; 2920 2921 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2922 BOND_SLAVE_NOTIFY_NOW); 2923 bond_set_slave_inactive_flags(slave, 2924 BOND_SLAVE_NOTIFY_NOW); 2925 2926 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2927 2928 if (slave == rtnl_dereference(bond->curr_active_slave)) { 2929 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2930 goto do_failover; 2931 } 2932 2933 continue; 2934 2935 default: 2936 slave_err(bond->dev, slave->dev, 2937 "impossible: link_new_state %d on slave\n", 2938 slave->link_new_state); 2939 continue; 2940 } 2941 2942 do_failover: 2943 block_netpoll_tx(); 2944 bond_select_active_slave(bond); 2945 unblock_netpoll_tx(); 2946 } 2947 2948 bond_set_carrier(bond); 2949 } 2950 2951 /* Send ARP probes for active-backup mode ARP monitor. 2952 * 2953 * Called with rcu_read_lock held. 2954 */ 2955 static bool bond_ab_arp_probe(struct bonding *bond) 2956 { 2957 struct slave *slave, *before = NULL, *new_slave = NULL, 2958 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 2959 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 2960 struct list_head *iter; 2961 bool found = false; 2962 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 2963 2964 if (curr_arp_slave && curr_active_slave) 2965 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 2966 curr_arp_slave->dev->name, 2967 curr_active_slave->dev->name); 2968 2969 if (curr_active_slave) { 2970 bond_arp_send_all(bond, curr_active_slave); 2971 return should_notify_rtnl; 2972 } 2973 2974 /* if we don't have a curr_active_slave, search for the next available 2975 * backup slave from the current_arp_slave and make it the candidate 2976 * for becoming the curr_active_slave 2977 */ 2978 2979 if (!curr_arp_slave) { 2980 curr_arp_slave = bond_first_slave_rcu(bond); 2981 if (!curr_arp_slave) 2982 return should_notify_rtnl; 2983 } 2984 2985 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 2986 2987 bond_for_each_slave_rcu(bond, slave, iter) { 2988 if (!found && !before && bond_slave_is_up(slave)) 2989 before = slave; 2990 2991 if (found && !new_slave && bond_slave_is_up(slave)) 2992 new_slave = slave; 2993 /* if the link state is up at this point, we 2994 * mark it down - this can happen if we have 2995 * simultaneous link failures and 2996 * reselect_active_interface doesn't make this 2997 * one the current slave so it is still marked 2998 * up when it is actually down 2999 */ 3000 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3001 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3002 BOND_SLAVE_NOTIFY_LATER); 3003 if (slave->link_failure_count < UINT_MAX) 3004 slave->link_failure_count++; 3005 3006 bond_set_slave_inactive_flags(slave, 3007 BOND_SLAVE_NOTIFY_LATER); 3008 3009 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3010 } 3011 if (slave == curr_arp_slave) 3012 found = true; 3013 } 3014 3015 if (!new_slave && before) 3016 new_slave = before; 3017 3018 if (!new_slave) 3019 goto check_state; 3020 3021 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3022 BOND_SLAVE_NOTIFY_LATER); 3023 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3024 bond_arp_send_all(bond, new_slave); 3025 new_slave->last_link_up = jiffies; 3026 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3027 3028 check_state: 3029 bond_for_each_slave_rcu(bond, slave, iter) { 3030 if (slave->should_notify || slave->should_notify_link) { 3031 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3032 break; 3033 } 3034 } 3035 return should_notify_rtnl; 3036 } 3037 3038 static void bond_activebackup_arp_mon(struct bonding *bond) 3039 { 3040 bool should_notify_peers = false; 3041 bool should_notify_rtnl = false; 3042 int delta_in_ticks; 3043 3044 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3045 3046 if (!bond_has_slaves(bond)) 3047 goto re_arm; 3048 3049 rcu_read_lock(); 3050 3051 should_notify_peers = bond_should_notify_peers(bond); 3052 3053 if (bond_ab_arp_inspect(bond)) { 3054 rcu_read_unlock(); 3055 3056 /* Race avoidance with bond_close flush of workqueue */ 3057 if (!rtnl_trylock()) { 3058 delta_in_ticks = 1; 3059 should_notify_peers = false; 3060 goto re_arm; 3061 } 3062 3063 bond_ab_arp_commit(bond); 3064 3065 rtnl_unlock(); 3066 rcu_read_lock(); 3067 } 3068 3069 should_notify_rtnl = bond_ab_arp_probe(bond); 3070 rcu_read_unlock(); 3071 3072 re_arm: 3073 if (bond->params.arp_interval) 3074 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3075 3076 if (should_notify_peers || should_notify_rtnl) { 3077 if (!rtnl_trylock()) 3078 return; 3079 3080 if (should_notify_peers) 3081 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3082 bond->dev); 3083 if (should_notify_rtnl) { 3084 bond_slave_state_notify(bond); 3085 bond_slave_link_notify(bond); 3086 } 3087 3088 rtnl_unlock(); 3089 } 3090 } 3091 3092 static void bond_arp_monitor(struct work_struct *work) 3093 { 3094 struct bonding *bond = container_of(work, struct bonding, 3095 arp_work.work); 3096 3097 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3098 bond_activebackup_arp_mon(bond); 3099 else 3100 bond_loadbalance_arp_mon(bond); 3101 } 3102 3103 /*-------------------------- netdev event handling --------------------------*/ 3104 3105 /* Change device name */ 3106 static int bond_event_changename(struct bonding *bond) 3107 { 3108 bond_remove_proc_entry(bond); 3109 bond_create_proc_entry(bond); 3110 3111 bond_debug_reregister(bond); 3112 3113 return NOTIFY_DONE; 3114 } 3115 3116 static int bond_master_netdev_event(unsigned long event, 3117 struct net_device *bond_dev) 3118 { 3119 struct bonding *event_bond = netdev_priv(bond_dev); 3120 3121 netdev_dbg(bond_dev, "%s called\n", __func__); 3122 3123 switch (event) { 3124 case NETDEV_CHANGENAME: 3125 return bond_event_changename(event_bond); 3126 case NETDEV_UNREGISTER: 3127 bond_remove_proc_entry(event_bond); 3128 break; 3129 case NETDEV_REGISTER: 3130 bond_create_proc_entry(event_bond); 3131 break; 3132 default: 3133 break; 3134 } 3135 3136 return NOTIFY_DONE; 3137 } 3138 3139 static int bond_slave_netdev_event(unsigned long event, 3140 struct net_device *slave_dev) 3141 { 3142 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3143 struct bonding *bond; 3144 struct net_device *bond_dev; 3145 3146 /* A netdev event can be generated while enslaving a device 3147 * before netdev_rx_handler_register is called in which case 3148 * slave will be NULL 3149 */ 3150 if (!slave) { 3151 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3152 return NOTIFY_DONE; 3153 } 3154 3155 bond_dev = slave->bond->dev; 3156 bond = slave->bond; 3157 primary = rtnl_dereference(bond->primary_slave); 3158 3159 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3160 3161 switch (event) { 3162 case NETDEV_UNREGISTER: 3163 if (bond_dev->type != ARPHRD_ETHER) 3164 bond_release_and_destroy(bond_dev, slave_dev); 3165 else 3166 __bond_release_one(bond_dev, slave_dev, false, true); 3167 break; 3168 case NETDEV_UP: 3169 case NETDEV_CHANGE: 3170 /* For 802.3ad mode only: 3171 * Getting invalid Speed/Duplex values here will put slave 3172 * in weird state. Mark it as link-fail if the link was 3173 * previously up or link-down if it hasn't yet come up, and 3174 * let link-monitoring (miimon) set it right when correct 3175 * speeds/duplex are available. 3176 */ 3177 if (bond_update_speed_duplex(slave) && 3178 BOND_MODE(bond) == BOND_MODE_8023AD) { 3179 if (slave->last_link_up) 3180 slave->link = BOND_LINK_FAIL; 3181 else 3182 slave->link = BOND_LINK_DOWN; 3183 } 3184 3185 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3186 bond_3ad_adapter_speed_duplex_changed(slave); 3187 /* Fallthrough */ 3188 case NETDEV_DOWN: 3189 /* Refresh slave-array if applicable! 3190 * If the setup does not use miimon or arpmon (mode-specific!), 3191 * then these events will not cause the slave-array to be 3192 * refreshed. This will cause xmit to use a slave that is not 3193 * usable. Avoid such situation by refeshing the array at these 3194 * events. If these (miimon/arpmon) parameters are configured 3195 * then array gets refreshed twice and that should be fine! 3196 */ 3197 if (bond_mode_can_use_xmit_hash(bond)) 3198 bond_update_slave_arr(bond, NULL); 3199 break; 3200 case NETDEV_CHANGEMTU: 3201 /* TODO: Should slaves be allowed to 3202 * independently alter their MTU? For 3203 * an active-backup bond, slaves need 3204 * not be the same type of device, so 3205 * MTUs may vary. For other modes, 3206 * slaves arguably should have the 3207 * same MTUs. To do this, we'd need to 3208 * take over the slave's change_mtu 3209 * function for the duration of their 3210 * servitude. 3211 */ 3212 break; 3213 case NETDEV_CHANGENAME: 3214 /* we don't care if we don't have primary set */ 3215 if (!bond_uses_primary(bond) || 3216 !bond->params.primary[0]) 3217 break; 3218 3219 if (slave == primary) { 3220 /* slave's name changed - he's no longer primary */ 3221 RCU_INIT_POINTER(bond->primary_slave, NULL); 3222 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3223 /* we have a new primary slave */ 3224 rcu_assign_pointer(bond->primary_slave, slave); 3225 } else { /* we didn't change primary - exit */ 3226 break; 3227 } 3228 3229 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3230 primary ? slave_dev->name : "none"); 3231 3232 block_netpoll_tx(); 3233 bond_select_active_slave(bond); 3234 unblock_netpoll_tx(); 3235 break; 3236 case NETDEV_FEAT_CHANGE: 3237 bond_compute_features(bond); 3238 break; 3239 case NETDEV_RESEND_IGMP: 3240 /* Propagate to master device */ 3241 call_netdevice_notifiers(event, slave->bond->dev); 3242 break; 3243 default: 3244 break; 3245 } 3246 3247 return NOTIFY_DONE; 3248 } 3249 3250 /* bond_netdev_event: handle netdev notifier chain events. 3251 * 3252 * This function receives events for the netdev chain. The caller (an 3253 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3254 * locks for us to safely manipulate the slave devices (RTNL lock, 3255 * dev_probe_lock). 3256 */ 3257 static int bond_netdev_event(struct notifier_block *this, 3258 unsigned long event, void *ptr) 3259 { 3260 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3261 3262 netdev_dbg(event_dev, "%s received %s\n", 3263 __func__, netdev_cmd_to_name(event)); 3264 3265 if (!(event_dev->priv_flags & IFF_BONDING)) 3266 return NOTIFY_DONE; 3267 3268 if (event_dev->flags & IFF_MASTER) { 3269 int ret; 3270 3271 ret = bond_master_netdev_event(event, event_dev); 3272 if (ret != NOTIFY_DONE) 3273 return ret; 3274 } 3275 3276 if (event_dev->flags & IFF_SLAVE) 3277 return bond_slave_netdev_event(event, event_dev); 3278 3279 return NOTIFY_DONE; 3280 } 3281 3282 static struct notifier_block bond_netdev_notifier = { 3283 .notifier_call = bond_netdev_event, 3284 }; 3285 3286 /*---------------------------- Hashing Policies -----------------------------*/ 3287 3288 /* L2 hash helper */ 3289 static inline u32 bond_eth_hash(struct sk_buff *skb) 3290 { 3291 struct ethhdr *ep, hdr_tmp; 3292 3293 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3294 if (ep) 3295 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3296 return 0; 3297 } 3298 3299 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, 3300 int *noff, int *proto, bool l34) 3301 { 3302 const struct ipv6hdr *iph6; 3303 const struct iphdr *iph; 3304 3305 if (skb->protocol == htons(ETH_P_IP)) { 3306 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph)))) 3307 return false; 3308 iph = (const struct iphdr *)(skb->data + *noff); 3309 iph_to_flow_copy_v4addrs(fk, iph); 3310 *noff += iph->ihl << 2; 3311 if (!ip_is_fragment(iph)) 3312 *proto = iph->protocol; 3313 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3314 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6)))) 3315 return false; 3316 iph6 = (const struct ipv6hdr *)(skb->data + *noff); 3317 iph_to_flow_copy_v6addrs(fk, iph6); 3318 *noff += sizeof(*iph6); 3319 *proto = iph6->nexthdr; 3320 } else { 3321 return false; 3322 } 3323 3324 if (l34 && *proto >= 0) 3325 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto); 3326 3327 return true; 3328 } 3329 3330 /* Extract the appropriate headers based on bond's xmit policy */ 3331 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3332 struct flow_keys *fk) 3333 { 3334 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 3335 int noff, proto = -1; 3336 3337 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) { 3338 memset(fk, 0, sizeof(*fk)); 3339 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 3340 fk, NULL, 0, 0, 0, 0); 3341 } 3342 3343 fk->ports.ports = 0; 3344 memset(&fk->icmp, 0, sizeof(fk->icmp)); 3345 noff = skb_network_offset(skb); 3346 if (!bond_flow_ip(skb, fk, &noff, &proto, l34)) 3347 return false; 3348 3349 /* ICMP error packets contains at least 8 bytes of the header 3350 * of the packet which generated the error. Use this information 3351 * to correlate ICMP error packets within the same flow which 3352 * generated the error. 3353 */ 3354 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) { 3355 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data, 3356 skb_transport_offset(skb), 3357 skb_headlen(skb)); 3358 if (proto == IPPROTO_ICMP) { 3359 if (!icmp_is_err(fk->icmp.type)) 3360 return true; 3361 3362 noff += sizeof(struct icmphdr); 3363 } else if (proto == IPPROTO_ICMPV6) { 3364 if (!icmpv6_is_err(fk->icmp.type)) 3365 return true; 3366 3367 noff += sizeof(struct icmp6hdr); 3368 } 3369 return bond_flow_ip(skb, fk, &noff, &proto, l34); 3370 } 3371 3372 return true; 3373 } 3374 3375 /** 3376 * bond_xmit_hash - generate a hash value based on the xmit policy 3377 * @bond: bonding device 3378 * @skb: buffer to use for headers 3379 * 3380 * This function will extract the necessary headers from the skb buffer and use 3381 * them to generate a hash based on the xmit_policy set in the bonding device 3382 */ 3383 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3384 { 3385 struct flow_keys flow; 3386 u32 hash; 3387 3388 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3389 skb->l4_hash) 3390 return skb->hash; 3391 3392 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3393 !bond_flow_dissect(bond, skb, &flow)) 3394 return bond_eth_hash(skb); 3395 3396 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3397 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 3398 hash = bond_eth_hash(skb); 3399 } else { 3400 if (flow.icmp.id) 3401 memcpy(&hash, &flow.icmp, sizeof(hash)); 3402 else 3403 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 3404 } 3405 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3406 (__force u32)flow_get_u32_src(&flow); 3407 hash ^= (hash >> 16); 3408 hash ^= (hash >> 8); 3409 3410 return hash >> 1; 3411 } 3412 3413 /*-------------------------- Device entry points ----------------------------*/ 3414 3415 void bond_work_init_all(struct bonding *bond) 3416 { 3417 INIT_DELAYED_WORK(&bond->mcast_work, 3418 bond_resend_igmp_join_requests_delayed); 3419 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3420 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3421 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3422 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3423 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3424 } 3425 3426 static void bond_work_cancel_all(struct bonding *bond) 3427 { 3428 cancel_delayed_work_sync(&bond->mii_work); 3429 cancel_delayed_work_sync(&bond->arp_work); 3430 cancel_delayed_work_sync(&bond->alb_work); 3431 cancel_delayed_work_sync(&bond->ad_work); 3432 cancel_delayed_work_sync(&bond->mcast_work); 3433 cancel_delayed_work_sync(&bond->slave_arr_work); 3434 } 3435 3436 static int bond_open(struct net_device *bond_dev) 3437 { 3438 struct bonding *bond = netdev_priv(bond_dev); 3439 struct list_head *iter; 3440 struct slave *slave; 3441 3442 /* reset slave->backup and slave->inactive */ 3443 if (bond_has_slaves(bond)) { 3444 bond_for_each_slave(bond, slave, iter) { 3445 if (bond_uses_primary(bond) && 3446 slave != rcu_access_pointer(bond->curr_active_slave)) { 3447 bond_set_slave_inactive_flags(slave, 3448 BOND_SLAVE_NOTIFY_NOW); 3449 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3450 bond_set_slave_active_flags(slave, 3451 BOND_SLAVE_NOTIFY_NOW); 3452 } 3453 } 3454 } 3455 3456 if (bond_is_lb(bond)) { 3457 /* bond_alb_initialize must be called before the timer 3458 * is started. 3459 */ 3460 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3461 return -ENOMEM; 3462 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3463 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3464 } 3465 3466 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3467 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3468 3469 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3470 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3471 bond->recv_probe = bond_arp_rcv; 3472 } 3473 3474 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3475 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3476 /* register to receive LACPDUs */ 3477 bond->recv_probe = bond_3ad_lacpdu_recv; 3478 bond_3ad_initiate_agg_selection(bond, 1); 3479 } 3480 3481 if (bond_mode_can_use_xmit_hash(bond)) 3482 bond_update_slave_arr(bond, NULL); 3483 3484 return 0; 3485 } 3486 3487 static int bond_close(struct net_device *bond_dev) 3488 { 3489 struct bonding *bond = netdev_priv(bond_dev); 3490 3491 bond_work_cancel_all(bond); 3492 bond->send_peer_notif = 0; 3493 if (bond_is_lb(bond)) 3494 bond_alb_deinitialize(bond); 3495 bond->recv_probe = NULL; 3496 3497 return 0; 3498 } 3499 3500 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3501 * that some drivers can provide 32bit values only. 3502 */ 3503 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3504 const struct rtnl_link_stats64 *_new, 3505 const struct rtnl_link_stats64 *_old) 3506 { 3507 const u64 *new = (const u64 *)_new; 3508 const u64 *old = (const u64 *)_old; 3509 u64 *res = (u64 *)_res; 3510 int i; 3511 3512 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3513 u64 nv = new[i]; 3514 u64 ov = old[i]; 3515 s64 delta = nv - ov; 3516 3517 /* detects if this particular field is 32bit only */ 3518 if (((nv | ov) >> 32) == 0) 3519 delta = (s64)(s32)((u32)nv - (u32)ov); 3520 3521 /* filter anomalies, some drivers reset their stats 3522 * at down/up events. 3523 */ 3524 if (delta > 0) 3525 res[i] += delta; 3526 } 3527 } 3528 3529 static void bond_get_stats(struct net_device *bond_dev, 3530 struct rtnl_link_stats64 *stats) 3531 { 3532 struct bonding *bond = netdev_priv(bond_dev); 3533 struct rtnl_link_stats64 temp; 3534 struct list_head *iter; 3535 struct slave *slave; 3536 3537 spin_lock(&bond->stats_lock); 3538 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3539 3540 rcu_read_lock(); 3541 bond_for_each_slave_rcu(bond, slave, iter) { 3542 const struct rtnl_link_stats64 *new = 3543 dev_get_stats(slave->dev, &temp); 3544 3545 bond_fold_stats(stats, new, &slave->slave_stats); 3546 3547 /* save off the slave stats for the next run */ 3548 memcpy(&slave->slave_stats, new, sizeof(*new)); 3549 } 3550 rcu_read_unlock(); 3551 3552 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3553 spin_unlock(&bond->stats_lock); 3554 } 3555 3556 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3557 { 3558 struct bonding *bond = netdev_priv(bond_dev); 3559 struct net_device *slave_dev = NULL; 3560 struct ifbond k_binfo; 3561 struct ifbond __user *u_binfo = NULL; 3562 struct ifslave k_sinfo; 3563 struct ifslave __user *u_sinfo = NULL; 3564 struct mii_ioctl_data *mii = NULL; 3565 struct bond_opt_value newval; 3566 struct net *net; 3567 int res = 0; 3568 3569 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3570 3571 switch (cmd) { 3572 case SIOCGMIIPHY: 3573 mii = if_mii(ifr); 3574 if (!mii) 3575 return -EINVAL; 3576 3577 mii->phy_id = 0; 3578 /* Fall Through */ 3579 case SIOCGMIIREG: 3580 /* We do this again just in case we were called by SIOCGMIIREG 3581 * instead of SIOCGMIIPHY. 3582 */ 3583 mii = if_mii(ifr); 3584 if (!mii) 3585 return -EINVAL; 3586 3587 if (mii->reg_num == 1) { 3588 mii->val_out = 0; 3589 if (netif_carrier_ok(bond->dev)) 3590 mii->val_out = BMSR_LSTATUS; 3591 } 3592 3593 return 0; 3594 case BOND_INFO_QUERY_OLD: 3595 case SIOCBONDINFOQUERY: 3596 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3597 3598 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3599 return -EFAULT; 3600 3601 bond_info_query(bond_dev, &k_binfo); 3602 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3603 return -EFAULT; 3604 3605 return 0; 3606 case BOND_SLAVE_INFO_QUERY_OLD: 3607 case SIOCBONDSLAVEINFOQUERY: 3608 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3609 3610 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3611 return -EFAULT; 3612 3613 res = bond_slave_info_query(bond_dev, &k_sinfo); 3614 if (res == 0 && 3615 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3616 return -EFAULT; 3617 3618 return res; 3619 default: 3620 break; 3621 } 3622 3623 net = dev_net(bond_dev); 3624 3625 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3626 return -EPERM; 3627 3628 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3629 3630 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 3631 3632 if (!slave_dev) 3633 return -ENODEV; 3634 3635 switch (cmd) { 3636 case BOND_ENSLAVE_OLD: 3637 case SIOCBONDENSLAVE: 3638 res = bond_enslave(bond_dev, slave_dev, NULL); 3639 break; 3640 case BOND_RELEASE_OLD: 3641 case SIOCBONDRELEASE: 3642 res = bond_release(bond_dev, slave_dev); 3643 break; 3644 case BOND_SETHWADDR_OLD: 3645 case SIOCBONDSETHWADDR: 3646 res = bond_set_dev_addr(bond_dev, slave_dev); 3647 break; 3648 case BOND_CHANGE_ACTIVE_OLD: 3649 case SIOCBONDCHANGEACTIVE: 3650 bond_opt_initstr(&newval, slave_dev->name); 3651 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3652 &newval); 3653 break; 3654 default: 3655 res = -EOPNOTSUPP; 3656 } 3657 3658 return res; 3659 } 3660 3661 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3662 { 3663 struct bonding *bond = netdev_priv(bond_dev); 3664 3665 if (change & IFF_PROMISC) 3666 bond_set_promiscuity(bond, 3667 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3668 3669 if (change & IFF_ALLMULTI) 3670 bond_set_allmulti(bond, 3671 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3672 } 3673 3674 static void bond_set_rx_mode(struct net_device *bond_dev) 3675 { 3676 struct bonding *bond = netdev_priv(bond_dev); 3677 struct list_head *iter; 3678 struct slave *slave; 3679 3680 rcu_read_lock(); 3681 if (bond_uses_primary(bond)) { 3682 slave = rcu_dereference(bond->curr_active_slave); 3683 if (slave) { 3684 dev_uc_sync(slave->dev, bond_dev); 3685 dev_mc_sync(slave->dev, bond_dev); 3686 } 3687 } else { 3688 bond_for_each_slave_rcu(bond, slave, iter) { 3689 dev_uc_sync_multiple(slave->dev, bond_dev); 3690 dev_mc_sync_multiple(slave->dev, bond_dev); 3691 } 3692 } 3693 rcu_read_unlock(); 3694 } 3695 3696 static int bond_neigh_init(struct neighbour *n) 3697 { 3698 struct bonding *bond = netdev_priv(n->dev); 3699 const struct net_device_ops *slave_ops; 3700 struct neigh_parms parms; 3701 struct slave *slave; 3702 int ret = 0; 3703 3704 rcu_read_lock(); 3705 slave = bond_first_slave_rcu(bond); 3706 if (!slave) 3707 goto out; 3708 slave_ops = slave->dev->netdev_ops; 3709 if (!slave_ops->ndo_neigh_setup) 3710 goto out; 3711 3712 /* TODO: find another way [1] to implement this. 3713 * Passing a zeroed structure is fragile, 3714 * but at least we do not pass garbage. 3715 * 3716 * [1] One way would be that ndo_neigh_setup() never touch 3717 * struct neigh_parms, but propagate the new neigh_setup() 3718 * back to ___neigh_create() / neigh_parms_alloc() 3719 */ 3720 memset(&parms, 0, sizeof(parms)); 3721 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3722 3723 if (ret) 3724 goto out; 3725 3726 if (parms.neigh_setup) 3727 ret = parms.neigh_setup(n); 3728 out: 3729 rcu_read_unlock(); 3730 return ret; 3731 } 3732 3733 /* The bonding ndo_neigh_setup is called at init time beofre any 3734 * slave exists. So we must declare proxy setup function which will 3735 * be used at run time to resolve the actual slave neigh param setup. 3736 * 3737 * It's also called by master devices (such as vlans) to setup their 3738 * underlying devices. In that case - do nothing, we're already set up from 3739 * our init. 3740 */ 3741 static int bond_neigh_setup(struct net_device *dev, 3742 struct neigh_parms *parms) 3743 { 3744 /* modify only our neigh_parms */ 3745 if (parms->dev == dev) 3746 parms->neigh_setup = bond_neigh_init; 3747 3748 return 0; 3749 } 3750 3751 /* Change the MTU of all of a master's slaves to match the master */ 3752 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3753 { 3754 struct bonding *bond = netdev_priv(bond_dev); 3755 struct slave *slave, *rollback_slave; 3756 struct list_head *iter; 3757 int res = 0; 3758 3759 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3760 3761 bond_for_each_slave(bond, slave, iter) { 3762 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 3763 slave, slave->dev->netdev_ops->ndo_change_mtu); 3764 3765 res = dev_set_mtu(slave->dev, new_mtu); 3766 3767 if (res) { 3768 /* If we failed to set the slave's mtu to the new value 3769 * we must abort the operation even in ACTIVE_BACKUP 3770 * mode, because if we allow the backup slaves to have 3771 * different mtu values than the active slave we'll 3772 * need to change their mtu when doing a failover. That 3773 * means changing their mtu from timer context, which 3774 * is probably not a good idea. 3775 */ 3776 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 3777 res, new_mtu); 3778 goto unwind; 3779 } 3780 } 3781 3782 bond_dev->mtu = new_mtu; 3783 3784 return 0; 3785 3786 unwind: 3787 /* unwind from head to the slave that failed */ 3788 bond_for_each_slave(bond, rollback_slave, iter) { 3789 int tmp_res; 3790 3791 if (rollback_slave == slave) 3792 break; 3793 3794 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3795 if (tmp_res) 3796 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 3797 tmp_res); 3798 } 3799 3800 return res; 3801 } 3802 3803 /* Change HW address 3804 * 3805 * Note that many devices must be down to change the HW address, and 3806 * downing the master releases all slaves. We can make bonds full of 3807 * bonding devices to test this, however. 3808 */ 3809 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3810 { 3811 struct bonding *bond = netdev_priv(bond_dev); 3812 struct slave *slave, *rollback_slave; 3813 struct sockaddr_storage *ss = addr, tmp_ss; 3814 struct list_head *iter; 3815 int res = 0; 3816 3817 if (BOND_MODE(bond) == BOND_MODE_ALB) 3818 return bond_alb_set_mac_address(bond_dev, addr); 3819 3820 3821 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 3822 3823 /* If fail_over_mac is enabled, do nothing and return success. 3824 * Returning an error causes ifenslave to fail. 3825 */ 3826 if (bond->params.fail_over_mac && 3827 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3828 return 0; 3829 3830 if (!is_valid_ether_addr(ss->__data)) 3831 return -EADDRNOTAVAIL; 3832 3833 bond_for_each_slave(bond, slave, iter) { 3834 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 3835 __func__, slave); 3836 res = dev_set_mac_address(slave->dev, addr, NULL); 3837 if (res) { 3838 /* TODO: consider downing the slave 3839 * and retry ? 3840 * User should expect communications 3841 * breakage anyway until ARP finish 3842 * updating, so... 3843 */ 3844 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 3845 __func__, res); 3846 goto unwind; 3847 } 3848 } 3849 3850 /* success */ 3851 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 3852 return 0; 3853 3854 unwind: 3855 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 3856 tmp_ss.ss_family = bond_dev->type; 3857 3858 /* unwind from head to the slave that failed */ 3859 bond_for_each_slave(bond, rollback_slave, iter) { 3860 int tmp_res; 3861 3862 if (rollback_slave == slave) 3863 break; 3864 3865 tmp_res = dev_set_mac_address(rollback_slave->dev, 3866 (struct sockaddr *)&tmp_ss, NULL); 3867 if (tmp_res) { 3868 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 3869 __func__, tmp_res); 3870 } 3871 } 3872 3873 return res; 3874 } 3875 3876 /** 3877 * bond_xmit_slave_id - transmit skb through slave with slave_id 3878 * @bond: bonding device that is transmitting 3879 * @skb: buffer to transmit 3880 * @slave_id: slave id up to slave_cnt-1 through which to transmit 3881 * 3882 * This function tries to transmit through slave with slave_id but in case 3883 * it fails, it tries to find the first available slave for transmission. 3884 * The skb is consumed in all cases, thus the function is void. 3885 */ 3886 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id) 3887 { 3888 struct list_head *iter; 3889 struct slave *slave; 3890 int i = slave_id; 3891 3892 /* Here we start from the slave with slave_id */ 3893 bond_for_each_slave_rcu(bond, slave, iter) { 3894 if (--i < 0) { 3895 if (bond_slave_can_tx(slave)) { 3896 bond_dev_queue_xmit(bond, skb, slave->dev); 3897 return; 3898 } 3899 } 3900 } 3901 3902 /* Here we start from the first slave up to slave_id */ 3903 i = slave_id; 3904 bond_for_each_slave_rcu(bond, slave, iter) { 3905 if (--i < 0) 3906 break; 3907 if (bond_slave_can_tx(slave)) { 3908 bond_dev_queue_xmit(bond, skb, slave->dev); 3909 return; 3910 } 3911 } 3912 /* no slave that can tx has been found */ 3913 bond_tx_drop(bond->dev, skb); 3914 } 3915 3916 /** 3917 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 3918 * @bond: bonding device to use 3919 * 3920 * Based on the value of the bonding device's packets_per_slave parameter 3921 * this function generates a slave id, which is usually used as the next 3922 * slave to transmit through. 3923 */ 3924 static u32 bond_rr_gen_slave_id(struct bonding *bond) 3925 { 3926 u32 slave_id; 3927 struct reciprocal_value reciprocal_packets_per_slave; 3928 int packets_per_slave = bond->params.packets_per_slave; 3929 3930 switch (packets_per_slave) { 3931 case 0: 3932 slave_id = prandom_u32(); 3933 break; 3934 case 1: 3935 slave_id = bond->rr_tx_counter; 3936 break; 3937 default: 3938 reciprocal_packets_per_slave = 3939 bond->params.reciprocal_packets_per_slave; 3940 slave_id = reciprocal_divide(bond->rr_tx_counter, 3941 reciprocal_packets_per_slave); 3942 break; 3943 } 3944 bond->rr_tx_counter++; 3945 3946 return slave_id; 3947 } 3948 3949 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 3950 struct net_device *bond_dev) 3951 { 3952 struct bonding *bond = netdev_priv(bond_dev); 3953 struct slave *slave; 3954 int slave_cnt; 3955 u32 slave_id; 3956 3957 /* Start with the curr_active_slave that joined the bond as the 3958 * default for sending IGMP traffic. For failover purposes one 3959 * needs to maintain some consistency for the interface that will 3960 * send the join/membership reports. The curr_active_slave found 3961 * will send all of this type of traffic. 3962 */ 3963 if (skb->protocol == htons(ETH_P_IP)) { 3964 int noff = skb_network_offset(skb); 3965 struct iphdr *iph; 3966 3967 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 3968 goto non_igmp; 3969 3970 iph = ip_hdr(skb); 3971 if (iph->protocol == IPPROTO_IGMP) { 3972 slave = rcu_dereference(bond->curr_active_slave); 3973 if (slave) 3974 bond_dev_queue_xmit(bond, skb, slave->dev); 3975 else 3976 bond_xmit_slave_id(bond, skb, 0); 3977 return NETDEV_TX_OK; 3978 } 3979 } 3980 3981 non_igmp: 3982 slave_cnt = READ_ONCE(bond->slave_cnt); 3983 if (likely(slave_cnt)) { 3984 slave_id = bond_rr_gen_slave_id(bond); 3985 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt); 3986 } else { 3987 bond_tx_drop(bond_dev, skb); 3988 } 3989 return NETDEV_TX_OK; 3990 } 3991 3992 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 3993 * the bond has a usable interface. 3994 */ 3995 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 3996 struct net_device *bond_dev) 3997 { 3998 struct bonding *bond = netdev_priv(bond_dev); 3999 struct slave *slave; 4000 4001 slave = rcu_dereference(bond->curr_active_slave); 4002 if (slave) 4003 bond_dev_queue_xmit(bond, skb, slave->dev); 4004 else 4005 bond_tx_drop(bond_dev, skb); 4006 4007 return NETDEV_TX_OK; 4008 } 4009 4010 /* Use this to update slave_array when (a) it's not appropriate to update 4011 * slave_array right away (note that update_slave_array() may sleep) 4012 * and / or (b) RTNL is not held. 4013 */ 4014 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4015 { 4016 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4017 } 4018 4019 /* Slave array work handler. Holds only RTNL */ 4020 static void bond_slave_arr_handler(struct work_struct *work) 4021 { 4022 struct bonding *bond = container_of(work, struct bonding, 4023 slave_arr_work.work); 4024 int ret; 4025 4026 if (!rtnl_trylock()) 4027 goto err; 4028 4029 ret = bond_update_slave_arr(bond, NULL); 4030 rtnl_unlock(); 4031 if (ret) { 4032 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4033 goto err; 4034 } 4035 return; 4036 4037 err: 4038 bond_slave_arr_work_rearm(bond, 1); 4039 } 4040 4041 /* Build the usable slaves array in control path for modes that use xmit-hash 4042 * to determine the slave interface - 4043 * (a) BOND_MODE_8023AD 4044 * (b) BOND_MODE_XOR 4045 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 4046 * 4047 * The caller is expected to hold RTNL only and NO other lock! 4048 */ 4049 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 4050 { 4051 struct slave *slave; 4052 struct list_head *iter; 4053 struct bond_up_slave *new_arr, *old_arr; 4054 int agg_id = 0; 4055 int ret = 0; 4056 4057 #ifdef CONFIG_LOCKDEP 4058 WARN_ON(lockdep_is_held(&bond->mode_lock)); 4059 #endif 4060 4061 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]), 4062 GFP_KERNEL); 4063 if (!new_arr) { 4064 ret = -ENOMEM; 4065 pr_err("Failed to build slave-array.\n"); 4066 goto out; 4067 } 4068 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4069 struct ad_info ad_info; 4070 4071 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 4072 pr_debug("bond_3ad_get_active_agg_info failed\n"); 4073 kfree_rcu(new_arr, rcu); 4074 /* No active aggragator means it's not safe to use 4075 * the previous array. 4076 */ 4077 old_arr = rtnl_dereference(bond->slave_arr); 4078 if (old_arr) { 4079 RCU_INIT_POINTER(bond->slave_arr, NULL); 4080 kfree_rcu(old_arr, rcu); 4081 } 4082 goto out; 4083 } 4084 agg_id = ad_info.aggregator_id; 4085 } 4086 bond_for_each_slave(bond, slave, iter) { 4087 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4088 struct aggregator *agg; 4089 4090 agg = SLAVE_AD_INFO(slave)->port.aggregator; 4091 if (!agg || agg->aggregator_identifier != agg_id) 4092 continue; 4093 } 4094 if (!bond_slave_can_tx(slave)) 4095 continue; 4096 if (skipslave == slave) 4097 continue; 4098 4099 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 4100 new_arr->count); 4101 4102 new_arr->arr[new_arr->count++] = slave; 4103 } 4104 4105 old_arr = rtnl_dereference(bond->slave_arr); 4106 rcu_assign_pointer(bond->slave_arr, new_arr); 4107 if (old_arr) 4108 kfree_rcu(old_arr, rcu); 4109 out: 4110 if (ret != 0 && skipslave) { 4111 int idx; 4112 4113 /* Rare situation where caller has asked to skip a specific 4114 * slave but allocation failed (most likely!). BTW this is 4115 * only possible when the call is initiated from 4116 * __bond_release_one(). In this situation; overwrite the 4117 * skipslave entry in the array with the last entry from the 4118 * array to avoid a situation where the xmit path may choose 4119 * this to-be-skipped slave to send a packet out. 4120 */ 4121 old_arr = rtnl_dereference(bond->slave_arr); 4122 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) { 4123 if (skipslave == old_arr->arr[idx]) { 4124 old_arr->arr[idx] = 4125 old_arr->arr[old_arr->count-1]; 4126 old_arr->count--; 4127 break; 4128 } 4129 } 4130 } 4131 return ret; 4132 } 4133 4134 /* Use this Xmit function for 3AD as well as XOR modes. The current 4135 * usable slave array is formed in the control path. The xmit function 4136 * just calculates hash and sends the packet out. 4137 */ 4138 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4139 struct net_device *dev) 4140 { 4141 struct bonding *bond = netdev_priv(dev); 4142 struct slave *slave; 4143 struct bond_up_slave *slaves; 4144 unsigned int count; 4145 4146 slaves = rcu_dereference(bond->slave_arr); 4147 count = slaves ? READ_ONCE(slaves->count) : 0; 4148 if (likely(count)) { 4149 slave = slaves->arr[bond_xmit_hash(bond, skb) % count]; 4150 bond_dev_queue_xmit(bond, skb, slave->dev); 4151 } else { 4152 bond_tx_drop(dev, skb); 4153 } 4154 4155 return NETDEV_TX_OK; 4156 } 4157 4158 /* in broadcast mode, we send everything to all usable interfaces. */ 4159 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4160 struct net_device *bond_dev) 4161 { 4162 struct bonding *bond = netdev_priv(bond_dev); 4163 struct slave *slave = NULL; 4164 struct list_head *iter; 4165 4166 bond_for_each_slave_rcu(bond, slave, iter) { 4167 if (bond_is_last_slave(bond, slave)) 4168 break; 4169 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4170 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4171 4172 if (!skb2) { 4173 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4174 bond_dev->name, __func__); 4175 continue; 4176 } 4177 bond_dev_queue_xmit(bond, skb2, slave->dev); 4178 } 4179 } 4180 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4181 bond_dev_queue_xmit(bond, skb, slave->dev); 4182 else 4183 bond_tx_drop(bond_dev, skb); 4184 4185 return NETDEV_TX_OK; 4186 } 4187 4188 /*------------------------- Device initialization ---------------------------*/ 4189 4190 /* Lookup the slave that corresponds to a qid */ 4191 static inline int bond_slave_override(struct bonding *bond, 4192 struct sk_buff *skb) 4193 { 4194 struct slave *slave = NULL; 4195 struct list_head *iter; 4196 4197 if (!skb_rx_queue_recorded(skb)) 4198 return 1; 4199 4200 /* Find out if any slaves have the same mapping as this skb. */ 4201 bond_for_each_slave_rcu(bond, slave, iter) { 4202 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4203 if (bond_slave_is_up(slave) && 4204 slave->link == BOND_LINK_UP) { 4205 bond_dev_queue_xmit(bond, skb, slave->dev); 4206 return 0; 4207 } 4208 /* If the slave isn't UP, use default transmit policy. */ 4209 break; 4210 } 4211 } 4212 4213 return 1; 4214 } 4215 4216 4217 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4218 struct net_device *sb_dev) 4219 { 4220 /* This helper function exists to help dev_pick_tx get the correct 4221 * destination queue. Using a helper function skips a call to 4222 * skb_tx_hash and will put the skbs in the queue we expect on their 4223 * way down to the bonding driver. 4224 */ 4225 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4226 4227 /* Save the original txq to restore before passing to the driver */ 4228 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4229 4230 if (unlikely(txq >= dev->real_num_tx_queues)) { 4231 do { 4232 txq -= dev->real_num_tx_queues; 4233 } while (txq >= dev->real_num_tx_queues); 4234 } 4235 return txq; 4236 } 4237 4238 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4239 { 4240 struct bonding *bond = netdev_priv(dev); 4241 4242 if (bond_should_override_tx_queue(bond) && 4243 !bond_slave_override(bond, skb)) 4244 return NETDEV_TX_OK; 4245 4246 switch (BOND_MODE(bond)) { 4247 case BOND_MODE_ROUNDROBIN: 4248 return bond_xmit_roundrobin(skb, dev); 4249 case BOND_MODE_ACTIVEBACKUP: 4250 return bond_xmit_activebackup(skb, dev); 4251 case BOND_MODE_8023AD: 4252 case BOND_MODE_XOR: 4253 return bond_3ad_xor_xmit(skb, dev); 4254 case BOND_MODE_BROADCAST: 4255 return bond_xmit_broadcast(skb, dev); 4256 case BOND_MODE_ALB: 4257 return bond_alb_xmit(skb, dev); 4258 case BOND_MODE_TLB: 4259 return bond_tlb_xmit(skb, dev); 4260 default: 4261 /* Should never happen, mode already checked */ 4262 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4263 WARN_ON_ONCE(1); 4264 bond_tx_drop(dev, skb); 4265 return NETDEV_TX_OK; 4266 } 4267 } 4268 4269 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4270 { 4271 struct bonding *bond = netdev_priv(dev); 4272 netdev_tx_t ret = NETDEV_TX_OK; 4273 4274 /* If we risk deadlock from transmitting this in the 4275 * netpoll path, tell netpoll to queue the frame for later tx 4276 */ 4277 if (unlikely(is_netpoll_tx_blocked(dev))) 4278 return NETDEV_TX_BUSY; 4279 4280 rcu_read_lock(); 4281 if (bond_has_slaves(bond)) 4282 ret = __bond_start_xmit(skb, dev); 4283 else 4284 bond_tx_drop(dev, skb); 4285 rcu_read_unlock(); 4286 4287 return ret; 4288 } 4289 4290 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4291 struct ethtool_link_ksettings *cmd) 4292 { 4293 struct bonding *bond = netdev_priv(bond_dev); 4294 unsigned long speed = 0; 4295 struct list_head *iter; 4296 struct slave *slave; 4297 4298 cmd->base.duplex = DUPLEX_UNKNOWN; 4299 cmd->base.port = PORT_OTHER; 4300 4301 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4302 * do not need to check mode. Though link speed might not represent 4303 * the true receive or transmit bandwidth (not all modes are symmetric) 4304 * this is an accurate maximum. 4305 */ 4306 bond_for_each_slave(bond, slave, iter) { 4307 if (bond_slave_can_tx(slave)) { 4308 if (slave->speed != SPEED_UNKNOWN) 4309 speed += slave->speed; 4310 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4311 slave->duplex != DUPLEX_UNKNOWN) 4312 cmd->base.duplex = slave->duplex; 4313 } 4314 } 4315 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4316 4317 return 0; 4318 } 4319 4320 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4321 struct ethtool_drvinfo *drvinfo) 4322 { 4323 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4324 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version)); 4325 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4326 BOND_ABI_VERSION); 4327 } 4328 4329 static const struct ethtool_ops bond_ethtool_ops = { 4330 .get_drvinfo = bond_ethtool_get_drvinfo, 4331 .get_link = ethtool_op_get_link, 4332 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4333 }; 4334 4335 static const struct net_device_ops bond_netdev_ops = { 4336 .ndo_init = bond_init, 4337 .ndo_uninit = bond_uninit, 4338 .ndo_open = bond_open, 4339 .ndo_stop = bond_close, 4340 .ndo_start_xmit = bond_start_xmit, 4341 .ndo_select_queue = bond_select_queue, 4342 .ndo_get_stats64 = bond_get_stats, 4343 .ndo_do_ioctl = bond_do_ioctl, 4344 .ndo_change_rx_flags = bond_change_rx_flags, 4345 .ndo_set_rx_mode = bond_set_rx_mode, 4346 .ndo_change_mtu = bond_change_mtu, 4347 .ndo_set_mac_address = bond_set_mac_address, 4348 .ndo_neigh_setup = bond_neigh_setup, 4349 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4350 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4351 #ifdef CONFIG_NET_POLL_CONTROLLER 4352 .ndo_netpoll_setup = bond_netpoll_setup, 4353 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4354 .ndo_poll_controller = bond_poll_controller, 4355 #endif 4356 .ndo_add_slave = bond_enslave, 4357 .ndo_del_slave = bond_release, 4358 .ndo_fix_features = bond_fix_features, 4359 .ndo_features_check = passthru_features_check, 4360 }; 4361 4362 static const struct device_type bond_type = { 4363 .name = "bond", 4364 }; 4365 4366 static void bond_destructor(struct net_device *bond_dev) 4367 { 4368 struct bonding *bond = netdev_priv(bond_dev); 4369 if (bond->wq) 4370 destroy_workqueue(bond->wq); 4371 } 4372 4373 void bond_setup(struct net_device *bond_dev) 4374 { 4375 struct bonding *bond = netdev_priv(bond_dev); 4376 4377 spin_lock_init(&bond->mode_lock); 4378 bond->params = bonding_defaults; 4379 4380 /* Initialize pointers */ 4381 bond->dev = bond_dev; 4382 4383 /* Initialize the device entry points */ 4384 ether_setup(bond_dev); 4385 bond_dev->max_mtu = ETH_MAX_MTU; 4386 bond_dev->netdev_ops = &bond_netdev_ops; 4387 bond_dev->ethtool_ops = &bond_ethtool_ops; 4388 4389 bond_dev->needs_free_netdev = true; 4390 bond_dev->priv_destructor = bond_destructor; 4391 4392 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4393 4394 /* Initialize the device options */ 4395 bond_dev->flags |= IFF_MASTER; 4396 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4397 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4398 4399 /* don't acquire bond device's netif_tx_lock when transmitting */ 4400 bond_dev->features |= NETIF_F_LLTX; 4401 4402 /* By default, we declare the bond to be fully 4403 * VLAN hardware accelerated capable. Special 4404 * care is taken in the various xmit functions 4405 * when there are slaves that are not hw accel 4406 * capable 4407 */ 4408 4409 /* Don't allow bond devices to change network namespaces. */ 4410 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4411 4412 bond_dev->hw_features = BOND_VLAN_FEATURES | 4413 NETIF_F_HW_VLAN_CTAG_RX | 4414 NETIF_F_HW_VLAN_CTAG_FILTER; 4415 4416 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4417 bond_dev->features |= bond_dev->hw_features; 4418 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 4419 } 4420 4421 /* Destroy a bonding device. 4422 * Must be under rtnl_lock when this function is called. 4423 */ 4424 static void bond_uninit(struct net_device *bond_dev) 4425 { 4426 struct bonding *bond = netdev_priv(bond_dev); 4427 struct list_head *iter; 4428 struct slave *slave; 4429 struct bond_up_slave *arr; 4430 4431 bond_netpoll_cleanup(bond_dev); 4432 4433 /* Release the bonded slaves */ 4434 bond_for_each_slave(bond, slave, iter) 4435 __bond_release_one(bond_dev, slave->dev, true, true); 4436 netdev_info(bond_dev, "Released all slaves\n"); 4437 4438 arr = rtnl_dereference(bond->slave_arr); 4439 if (arr) { 4440 RCU_INIT_POINTER(bond->slave_arr, NULL); 4441 kfree_rcu(arr, rcu); 4442 } 4443 4444 list_del(&bond->bond_list); 4445 4446 lockdep_unregister_key(&bond->stats_lock_key); 4447 bond_debug_unregister(bond); 4448 } 4449 4450 /*------------------------- Module initialization ---------------------------*/ 4451 4452 static int bond_check_params(struct bond_params *params) 4453 { 4454 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4455 struct bond_opt_value newval; 4456 const struct bond_opt_value *valptr; 4457 int arp_all_targets_value = 0; 4458 u16 ad_actor_sys_prio = 0; 4459 u16 ad_user_port_key = 0; 4460 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4461 int arp_ip_count; 4462 int bond_mode = BOND_MODE_ROUNDROBIN; 4463 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4464 int lacp_fast = 0; 4465 int tlb_dynamic_lb; 4466 4467 /* Convert string parameters. */ 4468 if (mode) { 4469 bond_opt_initstr(&newval, mode); 4470 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4471 if (!valptr) { 4472 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4473 return -EINVAL; 4474 } 4475 bond_mode = valptr->value; 4476 } 4477 4478 if (xmit_hash_policy) { 4479 if (bond_mode == BOND_MODE_ROUNDROBIN || 4480 bond_mode == BOND_MODE_ACTIVEBACKUP || 4481 bond_mode == BOND_MODE_BROADCAST) { 4482 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4483 bond_mode_name(bond_mode)); 4484 } else { 4485 bond_opt_initstr(&newval, xmit_hash_policy); 4486 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4487 &newval); 4488 if (!valptr) { 4489 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4490 xmit_hash_policy); 4491 return -EINVAL; 4492 } 4493 xmit_hashtype = valptr->value; 4494 } 4495 } 4496 4497 if (lacp_rate) { 4498 if (bond_mode != BOND_MODE_8023AD) { 4499 pr_info("lacp_rate param is irrelevant in mode %s\n", 4500 bond_mode_name(bond_mode)); 4501 } else { 4502 bond_opt_initstr(&newval, lacp_rate); 4503 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4504 &newval); 4505 if (!valptr) { 4506 pr_err("Error: Invalid lacp rate \"%s\"\n", 4507 lacp_rate); 4508 return -EINVAL; 4509 } 4510 lacp_fast = valptr->value; 4511 } 4512 } 4513 4514 if (ad_select) { 4515 bond_opt_initstr(&newval, ad_select); 4516 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4517 &newval); 4518 if (!valptr) { 4519 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4520 return -EINVAL; 4521 } 4522 params->ad_select = valptr->value; 4523 if (bond_mode != BOND_MODE_8023AD) 4524 pr_warn("ad_select param only affects 802.3ad mode\n"); 4525 } else { 4526 params->ad_select = BOND_AD_STABLE; 4527 } 4528 4529 if (max_bonds < 0) { 4530 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4531 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4532 max_bonds = BOND_DEFAULT_MAX_BONDS; 4533 } 4534 4535 if (miimon < 0) { 4536 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4537 miimon, INT_MAX); 4538 miimon = 0; 4539 } 4540 4541 if (updelay < 0) { 4542 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4543 updelay, INT_MAX); 4544 updelay = 0; 4545 } 4546 4547 if (downdelay < 0) { 4548 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4549 downdelay, INT_MAX); 4550 downdelay = 0; 4551 } 4552 4553 if ((use_carrier != 0) && (use_carrier != 1)) { 4554 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4555 use_carrier); 4556 use_carrier = 1; 4557 } 4558 4559 if (num_peer_notif < 0 || num_peer_notif > 255) { 4560 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4561 num_peer_notif); 4562 num_peer_notif = 1; 4563 } 4564 4565 /* reset values for 802.3ad/TLB/ALB */ 4566 if (!bond_mode_uses_arp(bond_mode)) { 4567 if (!miimon) { 4568 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"); 4569 pr_warn("Forcing miimon to 100msec\n"); 4570 miimon = BOND_DEFAULT_MIIMON; 4571 } 4572 } 4573 4574 if (tx_queues < 1 || tx_queues > 255) { 4575 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4576 tx_queues, BOND_DEFAULT_TX_QUEUES); 4577 tx_queues = BOND_DEFAULT_TX_QUEUES; 4578 } 4579 4580 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4581 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4582 all_slaves_active); 4583 all_slaves_active = 0; 4584 } 4585 4586 if (resend_igmp < 0 || resend_igmp > 255) { 4587 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4588 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4589 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4590 } 4591 4592 bond_opt_initval(&newval, packets_per_slave); 4593 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4594 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4595 packets_per_slave, USHRT_MAX); 4596 packets_per_slave = 1; 4597 } 4598 4599 if (bond_mode == BOND_MODE_ALB) { 4600 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", 4601 updelay); 4602 } 4603 4604 if (!miimon) { 4605 if (updelay || downdelay) { 4606 /* just warn the user the up/down delay will have 4607 * no effect since miimon is zero... 4608 */ 4609 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", 4610 updelay, downdelay); 4611 } 4612 } else { 4613 /* don't allow arp monitoring */ 4614 if (arp_interval) { 4615 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4616 miimon, arp_interval); 4617 arp_interval = 0; 4618 } 4619 4620 if ((updelay % miimon) != 0) { 4621 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4622 updelay, miimon, (updelay / miimon) * miimon); 4623 } 4624 4625 updelay /= miimon; 4626 4627 if ((downdelay % miimon) != 0) { 4628 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4629 downdelay, miimon, 4630 (downdelay / miimon) * miimon); 4631 } 4632 4633 downdelay /= miimon; 4634 } 4635 4636 if (arp_interval < 0) { 4637 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4638 arp_interval, INT_MAX); 4639 arp_interval = 0; 4640 } 4641 4642 for (arp_ip_count = 0, i = 0; 4643 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4644 __be32 ip; 4645 4646 /* not a complete check, but good enough to catch mistakes */ 4647 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4648 !bond_is_ip_target_ok(ip)) { 4649 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4650 arp_ip_target[i]); 4651 arp_interval = 0; 4652 } else { 4653 if (bond_get_targets_ip(arp_target, ip) == -1) 4654 arp_target[arp_ip_count++] = ip; 4655 else 4656 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4657 &ip); 4658 } 4659 } 4660 4661 if (arp_interval && !arp_ip_count) { 4662 /* don't allow arping if no arp_ip_target given... */ 4663 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4664 arp_interval); 4665 arp_interval = 0; 4666 } 4667 4668 if (arp_validate) { 4669 if (!arp_interval) { 4670 pr_err("arp_validate requires arp_interval\n"); 4671 return -EINVAL; 4672 } 4673 4674 bond_opt_initstr(&newval, arp_validate); 4675 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4676 &newval); 4677 if (!valptr) { 4678 pr_err("Error: invalid arp_validate \"%s\"\n", 4679 arp_validate); 4680 return -EINVAL; 4681 } 4682 arp_validate_value = valptr->value; 4683 } else { 4684 arp_validate_value = 0; 4685 } 4686 4687 if (arp_all_targets) { 4688 bond_opt_initstr(&newval, arp_all_targets); 4689 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4690 &newval); 4691 if (!valptr) { 4692 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4693 arp_all_targets); 4694 arp_all_targets_value = 0; 4695 } else { 4696 arp_all_targets_value = valptr->value; 4697 } 4698 } 4699 4700 if (miimon) { 4701 pr_info("MII link monitoring set to %d ms\n", miimon); 4702 } else if (arp_interval) { 4703 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4704 arp_validate_value); 4705 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4706 arp_interval, valptr->string, arp_ip_count); 4707 4708 for (i = 0; i < arp_ip_count; i++) 4709 pr_cont(" %s", arp_ip_target[i]); 4710 4711 pr_cont("\n"); 4712 4713 } else if (max_bonds) { 4714 /* miimon and arp_interval not set, we need one so things 4715 * work as expected, see bonding.txt for details 4716 */ 4717 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"); 4718 } 4719 4720 if (primary && !bond_mode_uses_primary(bond_mode)) { 4721 /* currently, using a primary only makes sense 4722 * in active backup, TLB or ALB modes 4723 */ 4724 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4725 primary, bond_mode_name(bond_mode)); 4726 primary = NULL; 4727 } 4728 4729 if (primary && primary_reselect) { 4730 bond_opt_initstr(&newval, primary_reselect); 4731 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 4732 &newval); 4733 if (!valptr) { 4734 pr_err("Error: Invalid primary_reselect \"%s\"\n", 4735 primary_reselect); 4736 return -EINVAL; 4737 } 4738 primary_reselect_value = valptr->value; 4739 } else { 4740 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 4741 } 4742 4743 if (fail_over_mac) { 4744 bond_opt_initstr(&newval, fail_over_mac); 4745 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 4746 &newval); 4747 if (!valptr) { 4748 pr_err("Error: invalid fail_over_mac \"%s\"\n", 4749 fail_over_mac); 4750 return -EINVAL; 4751 } 4752 fail_over_mac_value = valptr->value; 4753 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 4754 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 4755 } else { 4756 fail_over_mac_value = BOND_FOM_NONE; 4757 } 4758 4759 bond_opt_initstr(&newval, "default"); 4760 valptr = bond_opt_parse( 4761 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 4762 &newval); 4763 if (!valptr) { 4764 pr_err("Error: No ad_actor_sys_prio default value"); 4765 return -EINVAL; 4766 } 4767 ad_actor_sys_prio = valptr->value; 4768 4769 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 4770 &newval); 4771 if (!valptr) { 4772 pr_err("Error: No ad_user_port_key default value"); 4773 return -EINVAL; 4774 } 4775 ad_user_port_key = valptr->value; 4776 4777 bond_opt_initstr(&newval, "default"); 4778 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 4779 if (!valptr) { 4780 pr_err("Error: No tlb_dynamic_lb default value"); 4781 return -EINVAL; 4782 } 4783 tlb_dynamic_lb = valptr->value; 4784 4785 if (lp_interval == 0) { 4786 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 4787 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 4788 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 4789 } 4790 4791 /* fill params struct with the proper values */ 4792 params->mode = bond_mode; 4793 params->xmit_policy = xmit_hashtype; 4794 params->miimon = miimon; 4795 params->num_peer_notif = num_peer_notif; 4796 params->arp_interval = arp_interval; 4797 params->arp_validate = arp_validate_value; 4798 params->arp_all_targets = arp_all_targets_value; 4799 params->updelay = updelay; 4800 params->downdelay = downdelay; 4801 params->peer_notif_delay = 0; 4802 params->use_carrier = use_carrier; 4803 params->lacp_fast = lacp_fast; 4804 params->primary[0] = 0; 4805 params->primary_reselect = primary_reselect_value; 4806 params->fail_over_mac = fail_over_mac_value; 4807 params->tx_queues = tx_queues; 4808 params->all_slaves_active = all_slaves_active; 4809 params->resend_igmp = resend_igmp; 4810 params->min_links = min_links; 4811 params->lp_interval = lp_interval; 4812 params->packets_per_slave = packets_per_slave; 4813 params->tlb_dynamic_lb = tlb_dynamic_lb; 4814 params->ad_actor_sys_prio = ad_actor_sys_prio; 4815 eth_zero_addr(params->ad_actor_system); 4816 params->ad_user_port_key = ad_user_port_key; 4817 if (packets_per_slave > 0) { 4818 params->reciprocal_packets_per_slave = 4819 reciprocal_value(packets_per_slave); 4820 } else { 4821 /* reciprocal_packets_per_slave is unused if 4822 * packets_per_slave is 0 or 1, just initialize it 4823 */ 4824 params->reciprocal_packets_per_slave = 4825 (struct reciprocal_value) { 0 }; 4826 } 4827 4828 if (primary) { 4829 strncpy(params->primary, primary, IFNAMSIZ); 4830 params->primary[IFNAMSIZ - 1] = 0; 4831 } 4832 4833 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 4834 4835 return 0; 4836 } 4837 4838 /* Called from registration process */ 4839 static int bond_init(struct net_device *bond_dev) 4840 { 4841 struct bonding *bond = netdev_priv(bond_dev); 4842 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 4843 4844 netdev_dbg(bond_dev, "Begin bond_init\n"); 4845 4846 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 4847 if (!bond->wq) 4848 return -ENOMEM; 4849 4850 spin_lock_init(&bond->stats_lock); 4851 lockdep_register_key(&bond->stats_lock_key); 4852 lockdep_set_class(&bond->stats_lock, &bond->stats_lock_key); 4853 4854 list_add_tail(&bond->bond_list, &bn->dev_list); 4855 4856 bond_prepare_sysfs_group(bond); 4857 4858 bond_debug_register(bond); 4859 4860 /* Ensure valid dev_addr */ 4861 if (is_zero_ether_addr(bond_dev->dev_addr) && 4862 bond_dev->addr_assign_type == NET_ADDR_PERM) 4863 eth_hw_addr_random(bond_dev); 4864 4865 return 0; 4866 } 4867 4868 unsigned int bond_get_num_tx_queues(void) 4869 { 4870 return tx_queues; 4871 } 4872 4873 /* Create a new bond based on the specified name and bonding parameters. 4874 * If name is NULL, obtain a suitable "bond%d" name for us. 4875 * Caller must NOT hold rtnl_lock; we need to release it here before we 4876 * set up our sysfs entries. 4877 */ 4878 int bond_create(struct net *net, const char *name) 4879 { 4880 struct net_device *bond_dev; 4881 struct bonding *bond; 4882 struct alb_bond_info *bond_info; 4883 int res; 4884 4885 rtnl_lock(); 4886 4887 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 4888 name ? name : "bond%d", NET_NAME_UNKNOWN, 4889 bond_setup, tx_queues); 4890 if (!bond_dev) { 4891 pr_err("%s: eek! can't alloc netdev!\n", name); 4892 rtnl_unlock(); 4893 return -ENOMEM; 4894 } 4895 4896 /* 4897 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 4898 * It is set to 0 by default which is wrong. 4899 */ 4900 bond = netdev_priv(bond_dev); 4901 bond_info = &(BOND_ALB_INFO(bond)); 4902 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 4903 4904 dev_net_set(bond_dev, net); 4905 bond_dev->rtnl_link_ops = &bond_link_ops; 4906 4907 res = register_netdevice(bond_dev); 4908 4909 netif_carrier_off(bond_dev); 4910 4911 bond_work_init_all(bond); 4912 4913 rtnl_unlock(); 4914 if (res < 0) 4915 free_netdev(bond_dev); 4916 return res; 4917 } 4918 4919 static int __net_init bond_net_init(struct net *net) 4920 { 4921 struct bond_net *bn = net_generic(net, bond_net_id); 4922 4923 bn->net = net; 4924 INIT_LIST_HEAD(&bn->dev_list); 4925 4926 bond_create_proc_dir(bn); 4927 bond_create_sysfs(bn); 4928 4929 return 0; 4930 } 4931 4932 static void __net_exit bond_net_exit(struct net *net) 4933 { 4934 struct bond_net *bn = net_generic(net, bond_net_id); 4935 struct bonding *bond, *tmp_bond; 4936 LIST_HEAD(list); 4937 4938 bond_destroy_sysfs(bn); 4939 4940 /* Kill off any bonds created after unregistering bond rtnl ops */ 4941 rtnl_lock(); 4942 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 4943 unregister_netdevice_queue(bond->dev, &list); 4944 unregister_netdevice_many(&list); 4945 rtnl_unlock(); 4946 4947 bond_destroy_proc_dir(bn); 4948 } 4949 4950 static struct pernet_operations bond_net_ops = { 4951 .init = bond_net_init, 4952 .exit = bond_net_exit, 4953 .id = &bond_net_id, 4954 .size = sizeof(struct bond_net), 4955 }; 4956 4957 static int __init bonding_init(void) 4958 { 4959 int i; 4960 int res; 4961 4962 pr_info("%s", bond_version); 4963 4964 res = bond_check_params(&bonding_defaults); 4965 if (res) 4966 goto out; 4967 4968 res = register_pernet_subsys(&bond_net_ops); 4969 if (res) 4970 goto out; 4971 4972 res = bond_netlink_init(); 4973 if (res) 4974 goto err_link; 4975 4976 bond_create_debugfs(); 4977 4978 for (i = 0; i < max_bonds; i++) { 4979 res = bond_create(&init_net, NULL); 4980 if (res) 4981 goto err; 4982 } 4983 4984 skb_flow_dissector_init(&flow_keys_bonding, 4985 flow_keys_bonding_keys, 4986 ARRAY_SIZE(flow_keys_bonding_keys)); 4987 4988 register_netdevice_notifier(&bond_netdev_notifier); 4989 out: 4990 return res; 4991 err: 4992 bond_destroy_debugfs(); 4993 bond_netlink_fini(); 4994 err_link: 4995 unregister_pernet_subsys(&bond_net_ops); 4996 goto out; 4997 4998 } 4999 5000 static void __exit bonding_exit(void) 5001 { 5002 unregister_netdevice_notifier(&bond_netdev_notifier); 5003 5004 bond_destroy_debugfs(); 5005 5006 bond_netlink_fini(); 5007 unregister_pernet_subsys(&bond_net_ops); 5008 5009 #ifdef CONFIG_NET_POLL_CONTROLLER 5010 /* Make sure we don't have an imbalance on our netpoll blocking */ 5011 WARN_ON(atomic_read(&netpoll_block_tx)); 5012 #endif 5013 } 5014 5015 module_init(bonding_init); 5016 module_exit(bonding_exit); 5017 MODULE_LICENSE("GPL"); 5018 MODULE_VERSION(DRV_VERSION); 5019 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION); 5020 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 5021