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/xfrm.h> 83 #include <net/bonding.h> 84 #include <net/bond_3ad.h> 85 #include <net/bond_alb.h> 86 87 #include "bonding_priv.h" 88 89 /*---------------------------- Module parameters ----------------------------*/ 90 91 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 92 93 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 94 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 95 static int num_peer_notif = 1; 96 static int miimon; 97 static int updelay; 98 static int downdelay; 99 static int use_carrier = 1; 100 static char *mode; 101 static char *primary; 102 static char *primary_reselect; 103 static char *lacp_rate; 104 static int min_links; 105 static char *ad_select; 106 static char *xmit_hash_policy; 107 static int arp_interval; 108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 109 static char *arp_validate; 110 static char *arp_all_targets; 111 static char *fail_over_mac; 112 static int all_slaves_active; 113 static struct bond_params bonding_defaults; 114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 115 static int packets_per_slave = 1; 116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 117 118 module_param(max_bonds, int, 0); 119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 120 module_param(tx_queues, int, 0); 121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 122 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 124 "failover event (alias of num_unsol_na)"); 125 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 127 "failover event (alias of num_grat_arp)"); 128 module_param(miimon, int, 0); 129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 130 module_param(updelay, int, 0); 131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 132 module_param(downdelay, int, 0); 133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 134 "in milliseconds"); 135 module_param(use_carrier, int, 0); 136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 137 "0 for off, 1 for on (default)"); 138 module_param(mode, charp, 0); 139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 140 "1 for active-backup, 2 for balance-xor, " 141 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 142 "6 for balance-alb"); 143 module_param(primary, charp, 0); 144 MODULE_PARM_DESC(primary, "Primary network device to use"); 145 module_param(primary_reselect, charp, 0); 146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 147 "once it comes up; " 148 "0 for always (default), " 149 "1 for only if speed of primary is " 150 "better, " 151 "2 for only on active slave " 152 "failure"); 153 module_param(lacp_rate, charp, 0); 154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 155 "0 for slow, 1 for fast"); 156 module_param(ad_select, charp, 0); 157 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 158 "0 for stable (default), 1 for bandwidth, " 159 "2 for count"); 160 module_param(min_links, int, 0); 161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 162 163 module_param(xmit_hash_policy, charp, 0); 164 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 165 "0 for layer 2 (default), 1 for layer 3+4, " 166 "2 for layer 2+3, 3 for encap layer 2+3, " 167 "4 for encap layer 3+4"); 168 module_param(arp_interval, int, 0); 169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 170 module_param_array(arp_ip_target, charp, NULL, 0); 171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 172 module_param(arp_validate, charp, 0); 173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 174 "0 for none (default), 1 for active, " 175 "2 for backup, 3 for all"); 176 module_param(arp_all_targets, charp, 0); 177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 178 module_param(fail_over_mac, charp, 0); 179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 180 "the same MAC; 0 for none (default), " 181 "1 for active, 2 for follow"); 182 module_param(all_slaves_active, int, 0); 183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 184 "by setting active flag for all slaves; " 185 "0 for never (default), 1 for always."); 186 module_param(resend_igmp, int, 0); 187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 188 "link failure"); 189 module_param(packets_per_slave, int, 0); 190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 191 "mode; 0 for a random slave, 1 packet per " 192 "slave (default), >1 packets per slave."); 193 module_param(lp_interval, uint, 0); 194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 195 "the bonding driver sends learning packets to " 196 "each slaves peer switch. The default is 1."); 197 198 /*----------------------------- Global variables ----------------------------*/ 199 200 #ifdef CONFIG_NET_POLL_CONTROLLER 201 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 202 #endif 203 204 unsigned int bond_net_id __read_mostly; 205 206 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 207 { 208 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 209 .offset = offsetof(struct flow_keys, control), 210 }, 211 { 212 .key_id = FLOW_DISSECTOR_KEY_BASIC, 213 .offset = offsetof(struct flow_keys, basic), 214 }, 215 { 216 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 217 .offset = offsetof(struct flow_keys, addrs.v4addrs), 218 }, 219 { 220 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 221 .offset = offsetof(struct flow_keys, addrs.v6addrs), 222 }, 223 { 224 .key_id = FLOW_DISSECTOR_KEY_TIPC, 225 .offset = offsetof(struct flow_keys, addrs.tipckey), 226 }, 227 { 228 .key_id = FLOW_DISSECTOR_KEY_PORTS, 229 .offset = offsetof(struct flow_keys, ports), 230 }, 231 { 232 .key_id = FLOW_DISSECTOR_KEY_ICMP, 233 .offset = offsetof(struct flow_keys, icmp), 234 }, 235 { 236 .key_id = FLOW_DISSECTOR_KEY_VLAN, 237 .offset = offsetof(struct flow_keys, vlan), 238 }, 239 { 240 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 241 .offset = offsetof(struct flow_keys, tags), 242 }, 243 { 244 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 245 .offset = offsetof(struct flow_keys, keyid), 246 }, 247 }; 248 249 static struct flow_dissector flow_keys_bonding __read_mostly; 250 251 /*-------------------------- Forward declarations ---------------------------*/ 252 253 static int bond_init(struct net_device *bond_dev); 254 static void bond_uninit(struct net_device *bond_dev); 255 static void bond_get_stats(struct net_device *bond_dev, 256 struct rtnl_link_stats64 *stats); 257 static void bond_slave_arr_handler(struct work_struct *work); 258 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 259 int mod); 260 static void bond_netdev_notify_work(struct work_struct *work); 261 262 /*---------------------------- General routines -----------------------------*/ 263 264 const char *bond_mode_name(int mode) 265 { 266 static const char *names[] = { 267 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 268 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 269 [BOND_MODE_XOR] = "load balancing (xor)", 270 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 271 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 272 [BOND_MODE_TLB] = "transmit load balancing", 273 [BOND_MODE_ALB] = "adaptive load balancing", 274 }; 275 276 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 277 return "unknown"; 278 279 return names[mode]; 280 } 281 282 /** 283 * bond_dev_queue_xmit - Prepare skb for xmit. 284 * 285 * @bond: bond device that got this skb for tx. 286 * @skb: hw accel VLAN tagged skb to transmit 287 * @slave_dev: slave that is supposed to xmit this skbuff 288 */ 289 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 290 struct net_device *slave_dev) 291 { 292 skb->dev = slave_dev; 293 294 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 295 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 296 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 297 298 if (unlikely(netpoll_tx_running(bond->dev))) 299 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 300 301 return dev_queue_xmit(skb); 302 } 303 304 /*---------------------------------- VLAN -----------------------------------*/ 305 306 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 307 * We don't protect the slave list iteration with a lock because: 308 * a. This operation is performed in IOCTL context, 309 * b. The operation is protected by the RTNL semaphore in the 8021q code, 310 * c. Holding a lock with BH disabled while directly calling a base driver 311 * entry point is generally a BAD idea. 312 * 313 * The design of synchronization/protection for this operation in the 8021q 314 * module is good for one or more VLAN devices over a single physical device 315 * and cannot be extended for a teaming solution like bonding, so there is a 316 * potential race condition here where a net device from the vlan group might 317 * be referenced (either by a base driver or the 8021q code) while it is being 318 * removed from the system. However, it turns out we're not making matters 319 * worse, and if it works for regular VLAN usage it will work here too. 320 */ 321 322 /** 323 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 324 * @bond_dev: bonding net device that got called 325 * @vid: vlan id being added 326 */ 327 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 328 __be16 proto, u16 vid) 329 { 330 struct bonding *bond = netdev_priv(bond_dev); 331 struct slave *slave, *rollback_slave; 332 struct list_head *iter; 333 int res; 334 335 bond_for_each_slave(bond, slave, iter) { 336 res = vlan_vid_add(slave->dev, proto, vid); 337 if (res) 338 goto unwind; 339 } 340 341 return 0; 342 343 unwind: 344 /* unwind to the slave that failed */ 345 bond_for_each_slave(bond, rollback_slave, iter) { 346 if (rollback_slave == slave) 347 break; 348 349 vlan_vid_del(rollback_slave->dev, proto, vid); 350 } 351 352 return res; 353 } 354 355 /** 356 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 357 * @bond_dev: bonding net device that got called 358 * @vid: vlan id being removed 359 */ 360 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 361 __be16 proto, u16 vid) 362 { 363 struct bonding *bond = netdev_priv(bond_dev); 364 struct list_head *iter; 365 struct slave *slave; 366 367 bond_for_each_slave(bond, slave, iter) 368 vlan_vid_del(slave->dev, proto, vid); 369 370 if (bond_is_lb(bond)) 371 bond_alb_clear_vlan(bond, vid); 372 373 return 0; 374 } 375 376 /*---------------------------------- XFRM -----------------------------------*/ 377 378 #ifdef CONFIG_XFRM_OFFLOAD 379 /** 380 * bond_ipsec_add_sa - program device with a security association 381 * @xs: pointer to transformer state struct 382 **/ 383 static int bond_ipsec_add_sa(struct xfrm_state *xs) 384 { 385 struct net_device *bond_dev = xs->xso.dev; 386 struct bonding *bond = netdev_priv(bond_dev); 387 struct slave *slave = rtnl_dereference(bond->curr_active_slave); 388 389 xs->xso.real_dev = slave->dev; 390 bond->xs = xs; 391 392 if (!(slave->dev->xfrmdev_ops 393 && slave->dev->xfrmdev_ops->xdo_dev_state_add)) { 394 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n"); 395 return -EINVAL; 396 } 397 398 return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs); 399 } 400 401 /** 402 * bond_ipsec_del_sa - clear out this specific SA 403 * @xs: pointer to transformer state struct 404 **/ 405 static void bond_ipsec_del_sa(struct xfrm_state *xs) 406 { 407 struct net_device *bond_dev = xs->xso.dev; 408 struct bonding *bond = netdev_priv(bond_dev); 409 struct slave *slave = rtnl_dereference(bond->curr_active_slave); 410 411 if (!slave) 412 return; 413 414 xs->xso.real_dev = slave->dev; 415 416 if (!(slave->dev->xfrmdev_ops 417 && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) { 418 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__); 419 return; 420 } 421 422 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs); 423 } 424 425 /** 426 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 427 * @skb: current data packet 428 * @xs: pointer to transformer state struct 429 **/ 430 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 431 { 432 struct net_device *bond_dev = xs->xso.dev; 433 struct bonding *bond = netdev_priv(bond_dev); 434 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 435 struct net_device *slave_dev = curr_active->dev; 436 437 if (!(slave_dev->xfrmdev_ops 438 && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) { 439 slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__); 440 return false; 441 } 442 443 xs->xso.real_dev = slave_dev; 444 return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 445 } 446 447 static const struct xfrmdev_ops bond_xfrmdev_ops = { 448 .xdo_dev_state_add = bond_ipsec_add_sa, 449 .xdo_dev_state_delete = bond_ipsec_del_sa, 450 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 451 }; 452 #endif /* CONFIG_XFRM_OFFLOAD */ 453 454 /*------------------------------- Link status -------------------------------*/ 455 456 /* Set the carrier state for the master according to the state of its 457 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 458 * do special 802.3ad magic. 459 * 460 * Returns zero if carrier state does not change, nonzero if it does. 461 */ 462 int bond_set_carrier(struct bonding *bond) 463 { 464 struct list_head *iter; 465 struct slave *slave; 466 467 if (!bond_has_slaves(bond)) 468 goto down; 469 470 if (BOND_MODE(bond) == BOND_MODE_8023AD) 471 return bond_3ad_set_carrier(bond); 472 473 bond_for_each_slave(bond, slave, iter) { 474 if (slave->link == BOND_LINK_UP) { 475 if (!netif_carrier_ok(bond->dev)) { 476 netif_carrier_on(bond->dev); 477 return 1; 478 } 479 return 0; 480 } 481 } 482 483 down: 484 if (netif_carrier_ok(bond->dev)) { 485 netif_carrier_off(bond->dev); 486 return 1; 487 } 488 return 0; 489 } 490 491 /* Get link speed and duplex from the slave's base driver 492 * using ethtool. If for some reason the call fails or the 493 * values are invalid, set speed and duplex to -1, 494 * and return. Return 1 if speed or duplex settings are 495 * UNKNOWN; 0 otherwise. 496 */ 497 static int bond_update_speed_duplex(struct slave *slave) 498 { 499 struct net_device *slave_dev = slave->dev; 500 struct ethtool_link_ksettings ecmd; 501 int res; 502 503 slave->speed = SPEED_UNKNOWN; 504 slave->duplex = DUPLEX_UNKNOWN; 505 506 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 507 if (res < 0) 508 return 1; 509 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 510 return 1; 511 switch (ecmd.base.duplex) { 512 case DUPLEX_FULL: 513 case DUPLEX_HALF: 514 break; 515 default: 516 return 1; 517 } 518 519 slave->speed = ecmd.base.speed; 520 slave->duplex = ecmd.base.duplex; 521 522 return 0; 523 } 524 525 const char *bond_slave_link_status(s8 link) 526 { 527 switch (link) { 528 case BOND_LINK_UP: 529 return "up"; 530 case BOND_LINK_FAIL: 531 return "going down"; 532 case BOND_LINK_DOWN: 533 return "down"; 534 case BOND_LINK_BACK: 535 return "going back"; 536 default: 537 return "unknown"; 538 } 539 } 540 541 /* if <dev> supports MII link status reporting, check its link status. 542 * 543 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 544 * depending upon the setting of the use_carrier parameter. 545 * 546 * Return either BMSR_LSTATUS, meaning that the link is up (or we 547 * can't tell and just pretend it is), or 0, meaning that the link is 548 * down. 549 * 550 * If reporting is non-zero, instead of faking link up, return -1 if 551 * both ETHTOOL and MII ioctls fail (meaning the device does not 552 * support them). If use_carrier is set, return whatever it says. 553 * It'd be nice if there was a good way to tell if a driver supports 554 * netif_carrier, but there really isn't. 555 */ 556 static int bond_check_dev_link(struct bonding *bond, 557 struct net_device *slave_dev, int reporting) 558 { 559 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 560 int (*ioctl)(struct net_device *, struct ifreq *, int); 561 struct ifreq ifr; 562 struct mii_ioctl_data *mii; 563 564 if (!reporting && !netif_running(slave_dev)) 565 return 0; 566 567 if (bond->params.use_carrier) 568 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 569 570 /* Try to get link status using Ethtool first. */ 571 if (slave_dev->ethtool_ops->get_link) 572 return slave_dev->ethtool_ops->get_link(slave_dev) ? 573 BMSR_LSTATUS : 0; 574 575 /* Ethtool can't be used, fallback to MII ioctls. */ 576 ioctl = slave_ops->ndo_do_ioctl; 577 if (ioctl) { 578 /* TODO: set pointer to correct ioctl on a per team member 579 * bases to make this more efficient. that is, once 580 * we determine the correct ioctl, we will always 581 * call it and not the others for that team 582 * member. 583 */ 584 585 /* We cannot assume that SIOCGMIIPHY will also read a 586 * register; not all network drivers (e.g., e100) 587 * support that. 588 */ 589 590 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 591 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 592 mii = if_mii(&ifr); 593 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 594 mii->reg_num = MII_BMSR; 595 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 596 return mii->val_out & BMSR_LSTATUS; 597 } 598 } 599 600 /* If reporting, report that either there's no dev->do_ioctl, 601 * or both SIOCGMIIREG and get_link failed (meaning that we 602 * cannot report link status). If not reporting, pretend 603 * we're ok. 604 */ 605 return reporting ? -1 : BMSR_LSTATUS; 606 } 607 608 /*----------------------------- Multicast list ------------------------------*/ 609 610 /* Push the promiscuity flag down to appropriate slaves */ 611 static int bond_set_promiscuity(struct bonding *bond, int inc) 612 { 613 struct list_head *iter; 614 int err = 0; 615 616 if (bond_uses_primary(bond)) { 617 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 618 619 if (curr_active) 620 err = dev_set_promiscuity(curr_active->dev, inc); 621 } else { 622 struct slave *slave; 623 624 bond_for_each_slave(bond, slave, iter) { 625 err = dev_set_promiscuity(slave->dev, inc); 626 if (err) 627 return err; 628 } 629 } 630 return err; 631 } 632 633 /* Push the allmulti flag down to all slaves */ 634 static int bond_set_allmulti(struct bonding *bond, int inc) 635 { 636 struct list_head *iter; 637 int err = 0; 638 639 if (bond_uses_primary(bond)) { 640 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 641 642 if (curr_active) 643 err = dev_set_allmulti(curr_active->dev, inc); 644 } else { 645 struct slave *slave; 646 647 bond_for_each_slave(bond, slave, iter) { 648 err = dev_set_allmulti(slave->dev, inc); 649 if (err) 650 return err; 651 } 652 } 653 return err; 654 } 655 656 /* Retrieve the list of registered multicast addresses for the bonding 657 * device and retransmit an IGMP JOIN request to the current active 658 * slave. 659 */ 660 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 661 { 662 struct bonding *bond = container_of(work, struct bonding, 663 mcast_work.work); 664 665 if (!rtnl_trylock()) { 666 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 667 return; 668 } 669 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 670 671 if (bond->igmp_retrans > 1) { 672 bond->igmp_retrans--; 673 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 674 } 675 rtnl_unlock(); 676 } 677 678 /* Flush bond's hardware addresses from slave */ 679 static void bond_hw_addr_flush(struct net_device *bond_dev, 680 struct net_device *slave_dev) 681 { 682 struct bonding *bond = netdev_priv(bond_dev); 683 684 dev_uc_unsync(slave_dev, bond_dev); 685 dev_mc_unsync(slave_dev, bond_dev); 686 687 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 688 /* del lacpdu mc addr from mc list */ 689 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 690 691 dev_mc_del(slave_dev, lacpdu_multicast); 692 } 693 } 694 695 /*--------------------------- Active slave change ---------------------------*/ 696 697 /* Update the hardware address list and promisc/allmulti for the new and 698 * old active slaves (if any). Modes that are not using primary keep all 699 * slaves up date at all times; only the modes that use primary need to call 700 * this function to swap these settings during a failover. 701 */ 702 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 703 struct slave *old_active) 704 { 705 if (old_active) { 706 if (bond->dev->flags & IFF_PROMISC) 707 dev_set_promiscuity(old_active->dev, -1); 708 709 if (bond->dev->flags & IFF_ALLMULTI) 710 dev_set_allmulti(old_active->dev, -1); 711 712 bond_hw_addr_flush(bond->dev, old_active->dev); 713 } 714 715 if (new_active) { 716 /* FIXME: Signal errors upstream. */ 717 if (bond->dev->flags & IFF_PROMISC) 718 dev_set_promiscuity(new_active->dev, 1); 719 720 if (bond->dev->flags & IFF_ALLMULTI) 721 dev_set_allmulti(new_active->dev, 1); 722 723 netif_addr_lock_bh(bond->dev); 724 dev_uc_sync(new_active->dev, bond->dev); 725 dev_mc_sync(new_active->dev, bond->dev); 726 netif_addr_unlock_bh(bond->dev); 727 } 728 } 729 730 /** 731 * bond_set_dev_addr - clone slave's address to bond 732 * @bond_dev: bond net device 733 * @slave_dev: slave net device 734 * 735 * Should be called with RTNL held. 736 */ 737 static int bond_set_dev_addr(struct net_device *bond_dev, 738 struct net_device *slave_dev) 739 { 740 int err; 741 742 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 743 bond_dev, slave_dev, slave_dev->addr_len); 744 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 745 if (err) 746 return err; 747 748 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 749 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 750 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 751 return 0; 752 } 753 754 static struct slave *bond_get_old_active(struct bonding *bond, 755 struct slave *new_active) 756 { 757 struct slave *slave; 758 struct list_head *iter; 759 760 bond_for_each_slave(bond, slave, iter) { 761 if (slave == new_active) 762 continue; 763 764 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 765 return slave; 766 } 767 768 return NULL; 769 } 770 771 /* bond_do_fail_over_mac 772 * 773 * Perform special MAC address swapping for fail_over_mac settings 774 * 775 * Called with RTNL 776 */ 777 static void bond_do_fail_over_mac(struct bonding *bond, 778 struct slave *new_active, 779 struct slave *old_active) 780 { 781 u8 tmp_mac[MAX_ADDR_LEN]; 782 struct sockaddr_storage ss; 783 int rv; 784 785 switch (bond->params.fail_over_mac) { 786 case BOND_FOM_ACTIVE: 787 if (new_active) { 788 rv = bond_set_dev_addr(bond->dev, new_active->dev); 789 if (rv) 790 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 791 -rv); 792 } 793 break; 794 case BOND_FOM_FOLLOW: 795 /* if new_active && old_active, swap them 796 * if just old_active, do nothing (going to no active slave) 797 * if just new_active, set new_active to bond's MAC 798 */ 799 if (!new_active) 800 return; 801 802 if (!old_active) 803 old_active = bond_get_old_active(bond, new_active); 804 805 if (old_active) { 806 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 807 new_active->dev->addr_len); 808 bond_hw_addr_copy(ss.__data, 809 old_active->dev->dev_addr, 810 old_active->dev->addr_len); 811 ss.ss_family = new_active->dev->type; 812 } else { 813 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 814 bond->dev->addr_len); 815 ss.ss_family = bond->dev->type; 816 } 817 818 rv = dev_set_mac_address(new_active->dev, 819 (struct sockaddr *)&ss, NULL); 820 if (rv) { 821 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 822 -rv); 823 goto out; 824 } 825 826 if (!old_active) 827 goto out; 828 829 bond_hw_addr_copy(ss.__data, tmp_mac, 830 new_active->dev->addr_len); 831 ss.ss_family = old_active->dev->type; 832 833 rv = dev_set_mac_address(old_active->dev, 834 (struct sockaddr *)&ss, NULL); 835 if (rv) 836 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 837 -rv); 838 out: 839 break; 840 default: 841 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 842 bond->params.fail_over_mac); 843 break; 844 } 845 846 } 847 848 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 849 { 850 struct slave *prim = rtnl_dereference(bond->primary_slave); 851 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 852 853 if (!prim || prim->link != BOND_LINK_UP) { 854 if (!curr || curr->link != BOND_LINK_UP) 855 return NULL; 856 return curr; 857 } 858 859 if (bond->force_primary) { 860 bond->force_primary = false; 861 return prim; 862 } 863 864 if (!curr || curr->link != BOND_LINK_UP) 865 return prim; 866 867 /* At this point, prim and curr are both up */ 868 switch (bond->params.primary_reselect) { 869 case BOND_PRI_RESELECT_ALWAYS: 870 return prim; 871 case BOND_PRI_RESELECT_BETTER: 872 if (prim->speed < curr->speed) 873 return curr; 874 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 875 return curr; 876 return prim; 877 case BOND_PRI_RESELECT_FAILURE: 878 return curr; 879 default: 880 netdev_err(bond->dev, "impossible primary_reselect %d\n", 881 bond->params.primary_reselect); 882 return curr; 883 } 884 } 885 886 /** 887 * bond_find_best_slave - select the best available slave to be the active one 888 * @bond: our bonding struct 889 */ 890 static struct slave *bond_find_best_slave(struct bonding *bond) 891 { 892 struct slave *slave, *bestslave = NULL; 893 struct list_head *iter; 894 int mintime = bond->params.updelay; 895 896 slave = bond_choose_primary_or_current(bond); 897 if (slave) 898 return slave; 899 900 bond_for_each_slave(bond, slave, iter) { 901 if (slave->link == BOND_LINK_UP) 902 return slave; 903 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 904 slave->delay < mintime) { 905 mintime = slave->delay; 906 bestslave = slave; 907 } 908 } 909 910 return bestslave; 911 } 912 913 static bool bond_should_notify_peers(struct bonding *bond) 914 { 915 struct slave *slave; 916 917 rcu_read_lock(); 918 slave = rcu_dereference(bond->curr_active_slave); 919 rcu_read_unlock(); 920 921 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 922 slave ? slave->dev->name : "NULL"); 923 924 if (!slave || !bond->send_peer_notif || 925 bond->send_peer_notif % 926 max(1, bond->params.peer_notif_delay) != 0 || 927 !netif_carrier_ok(bond->dev) || 928 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 929 return false; 930 931 return true; 932 } 933 934 /** 935 * change_active_interface - change the active slave into the specified one 936 * @bond: our bonding struct 937 * @new: the new slave to make the active one 938 * 939 * Set the new slave to the bond's settings and unset them on the old 940 * curr_active_slave. 941 * Setting include flags, mc-list, promiscuity, allmulti, etc. 942 * 943 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 944 * because it is apparently the best available slave we have, even though its 945 * updelay hasn't timed out yet. 946 * 947 * Caller must hold RTNL. 948 */ 949 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 950 { 951 struct slave *old_active; 952 953 ASSERT_RTNL(); 954 955 old_active = rtnl_dereference(bond->curr_active_slave); 956 957 if (old_active == new_active) 958 return; 959 960 if (new_active) { 961 #ifdef CONFIG_XFRM_OFFLOAD 962 if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) && bond->xs) 963 bond_ipsec_del_sa(bond->xs); 964 #endif /* CONFIG_XFRM_OFFLOAD */ 965 966 new_active->last_link_up = jiffies; 967 968 if (new_active->link == BOND_LINK_BACK) { 969 if (bond_uses_primary(bond)) { 970 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 971 (bond->params.updelay - new_active->delay) * bond->params.miimon); 972 } 973 974 new_active->delay = 0; 975 bond_set_slave_link_state(new_active, BOND_LINK_UP, 976 BOND_SLAVE_NOTIFY_NOW); 977 978 if (BOND_MODE(bond) == BOND_MODE_8023AD) 979 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 980 981 if (bond_is_lb(bond)) 982 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 983 } else { 984 if (bond_uses_primary(bond)) { 985 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 986 } 987 } 988 } 989 990 if (bond_uses_primary(bond)) 991 bond_hw_addr_swap(bond, new_active, old_active); 992 993 if (bond_is_lb(bond)) { 994 bond_alb_handle_active_change(bond, new_active); 995 if (old_active) 996 bond_set_slave_inactive_flags(old_active, 997 BOND_SLAVE_NOTIFY_NOW); 998 if (new_active) 999 bond_set_slave_active_flags(new_active, 1000 BOND_SLAVE_NOTIFY_NOW); 1001 } else { 1002 rcu_assign_pointer(bond->curr_active_slave, new_active); 1003 } 1004 1005 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1006 if (old_active) 1007 bond_set_slave_inactive_flags(old_active, 1008 BOND_SLAVE_NOTIFY_NOW); 1009 1010 if (new_active) { 1011 bool should_notify_peers = false; 1012 1013 bond_set_slave_active_flags(new_active, 1014 BOND_SLAVE_NOTIFY_NOW); 1015 1016 if (bond->params.fail_over_mac) 1017 bond_do_fail_over_mac(bond, new_active, 1018 old_active); 1019 1020 if (netif_running(bond->dev)) { 1021 bond->send_peer_notif = 1022 bond->params.num_peer_notif * 1023 max(1, bond->params.peer_notif_delay); 1024 should_notify_peers = 1025 bond_should_notify_peers(bond); 1026 } 1027 1028 #ifdef CONFIG_XFRM_OFFLOAD 1029 if (old_active && bond->xs) { 1030 xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true); 1031 bond_ipsec_add_sa(bond->xs); 1032 } 1033 #endif /* CONFIG_XFRM_OFFLOAD */ 1034 1035 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1036 if (should_notify_peers) { 1037 bond->send_peer_notif--; 1038 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1039 bond->dev); 1040 } 1041 } 1042 } 1043 1044 /* resend IGMP joins since active slave has changed or 1045 * all were sent on curr_active_slave. 1046 * resend only if bond is brought up with the affected 1047 * bonding modes and the retransmission is enabled 1048 */ 1049 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1050 ((bond_uses_primary(bond) && new_active) || 1051 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1052 bond->igmp_retrans = bond->params.resend_igmp; 1053 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1054 } 1055 } 1056 1057 /** 1058 * bond_select_active_slave - select a new active slave, if needed 1059 * @bond: our bonding struct 1060 * 1061 * This functions should be called when one of the following occurs: 1062 * - The old curr_active_slave has been released or lost its link. 1063 * - The primary_slave has got its link back. 1064 * - A slave has got its link back and there's no old curr_active_slave. 1065 * 1066 * Caller must hold RTNL. 1067 */ 1068 void bond_select_active_slave(struct bonding *bond) 1069 { 1070 struct slave *best_slave; 1071 int rv; 1072 1073 ASSERT_RTNL(); 1074 1075 best_slave = bond_find_best_slave(bond); 1076 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1077 bond_change_active_slave(bond, best_slave); 1078 rv = bond_set_carrier(bond); 1079 if (!rv) 1080 return; 1081 1082 if (netif_carrier_ok(bond->dev)) 1083 netdev_info(bond->dev, "active interface up!\n"); 1084 else 1085 netdev_info(bond->dev, "now running without any active interface!\n"); 1086 } 1087 } 1088 1089 #ifdef CONFIG_NET_POLL_CONTROLLER 1090 static inline int slave_enable_netpoll(struct slave *slave) 1091 { 1092 struct netpoll *np; 1093 int err = 0; 1094 1095 np = kzalloc(sizeof(*np), GFP_KERNEL); 1096 err = -ENOMEM; 1097 if (!np) 1098 goto out; 1099 1100 err = __netpoll_setup(np, slave->dev); 1101 if (err) { 1102 kfree(np); 1103 goto out; 1104 } 1105 slave->np = np; 1106 out: 1107 return err; 1108 } 1109 static inline void slave_disable_netpoll(struct slave *slave) 1110 { 1111 struct netpoll *np = slave->np; 1112 1113 if (!np) 1114 return; 1115 1116 slave->np = NULL; 1117 1118 __netpoll_free(np); 1119 } 1120 1121 static void bond_poll_controller(struct net_device *bond_dev) 1122 { 1123 struct bonding *bond = netdev_priv(bond_dev); 1124 struct slave *slave = NULL; 1125 struct list_head *iter; 1126 struct ad_info ad_info; 1127 1128 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1129 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1130 return; 1131 1132 bond_for_each_slave_rcu(bond, slave, iter) { 1133 if (!bond_slave_is_up(slave)) 1134 continue; 1135 1136 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1137 struct aggregator *agg = 1138 SLAVE_AD_INFO(slave)->port.aggregator; 1139 1140 if (agg && 1141 agg->aggregator_identifier != ad_info.aggregator_id) 1142 continue; 1143 } 1144 1145 netpoll_poll_dev(slave->dev); 1146 } 1147 } 1148 1149 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1150 { 1151 struct bonding *bond = netdev_priv(bond_dev); 1152 struct list_head *iter; 1153 struct slave *slave; 1154 1155 bond_for_each_slave(bond, slave, iter) 1156 if (bond_slave_is_up(slave)) 1157 slave_disable_netpoll(slave); 1158 } 1159 1160 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1161 { 1162 struct bonding *bond = netdev_priv(dev); 1163 struct list_head *iter; 1164 struct slave *slave; 1165 int err = 0; 1166 1167 bond_for_each_slave(bond, slave, iter) { 1168 err = slave_enable_netpoll(slave); 1169 if (err) { 1170 bond_netpoll_cleanup(dev); 1171 break; 1172 } 1173 } 1174 return err; 1175 } 1176 #else 1177 static inline int slave_enable_netpoll(struct slave *slave) 1178 { 1179 return 0; 1180 } 1181 static inline void slave_disable_netpoll(struct slave *slave) 1182 { 1183 } 1184 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1185 { 1186 } 1187 #endif 1188 1189 /*---------------------------------- IOCTL ----------------------------------*/ 1190 1191 static netdev_features_t bond_fix_features(struct net_device *dev, 1192 netdev_features_t features) 1193 { 1194 struct bonding *bond = netdev_priv(dev); 1195 struct list_head *iter; 1196 netdev_features_t mask; 1197 struct slave *slave; 1198 1199 mask = features; 1200 1201 features &= ~NETIF_F_ONE_FOR_ALL; 1202 features |= NETIF_F_ALL_FOR_ALL; 1203 1204 bond_for_each_slave(bond, slave, iter) { 1205 features = netdev_increment_features(features, 1206 slave->dev->features, 1207 mask); 1208 } 1209 features = netdev_add_tso_features(features, mask); 1210 1211 return features; 1212 } 1213 1214 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1215 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \ 1216 NETIF_F_HIGHDMA | NETIF_F_LRO) 1217 1218 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1219 NETIF_F_RXCSUM | NETIF_F_ALL_TSO) 1220 1221 #ifdef CONFIG_XFRM_OFFLOAD 1222 #define BOND_XFRM_FEATURES (NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \ 1223 NETIF_F_GSO_ESP) 1224 #endif /* CONFIG_XFRM_OFFLOAD */ 1225 1226 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1227 NETIF_F_ALL_TSO) 1228 1229 1230 static void bond_compute_features(struct bonding *bond) 1231 { 1232 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1233 IFF_XMIT_DST_RELEASE_PERM; 1234 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1235 netdev_features_t enc_features = BOND_ENC_FEATURES; 1236 #ifdef CONFIG_XFRM_OFFLOAD 1237 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1238 #endif /* CONFIG_XFRM_OFFLOAD */ 1239 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1240 struct net_device *bond_dev = bond->dev; 1241 struct list_head *iter; 1242 struct slave *slave; 1243 unsigned short max_hard_header_len = ETH_HLEN; 1244 unsigned int gso_max_size = GSO_MAX_SIZE; 1245 u16 gso_max_segs = GSO_MAX_SEGS; 1246 1247 if (!bond_has_slaves(bond)) 1248 goto done; 1249 vlan_features &= NETIF_F_ALL_FOR_ALL; 1250 mpls_features &= NETIF_F_ALL_FOR_ALL; 1251 1252 bond_for_each_slave(bond, slave, iter) { 1253 vlan_features = netdev_increment_features(vlan_features, 1254 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1255 1256 enc_features = netdev_increment_features(enc_features, 1257 slave->dev->hw_enc_features, 1258 BOND_ENC_FEATURES); 1259 1260 #ifdef CONFIG_XFRM_OFFLOAD 1261 xfrm_features = netdev_increment_features(xfrm_features, 1262 slave->dev->hw_enc_features, 1263 BOND_XFRM_FEATURES); 1264 #endif /* CONFIG_XFRM_OFFLOAD */ 1265 1266 mpls_features = netdev_increment_features(mpls_features, 1267 slave->dev->mpls_features, 1268 BOND_MPLS_FEATURES); 1269 1270 dst_release_flag &= slave->dev->priv_flags; 1271 if (slave->dev->hard_header_len > max_hard_header_len) 1272 max_hard_header_len = slave->dev->hard_header_len; 1273 1274 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1275 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1276 } 1277 bond_dev->hard_header_len = max_hard_header_len; 1278 1279 done: 1280 bond_dev->vlan_features = vlan_features; 1281 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1282 NETIF_F_HW_VLAN_CTAG_TX | 1283 NETIF_F_HW_VLAN_STAG_TX | 1284 NETIF_F_GSO_UDP_L4; 1285 #ifdef CONFIG_XFRM_OFFLOAD 1286 bond_dev->hw_enc_features |= xfrm_features; 1287 #endif /* CONFIG_XFRM_OFFLOAD */ 1288 bond_dev->mpls_features = mpls_features; 1289 bond_dev->gso_max_segs = gso_max_segs; 1290 netif_set_gso_max_size(bond_dev, gso_max_size); 1291 1292 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1293 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1294 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1295 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1296 1297 netdev_change_features(bond_dev); 1298 } 1299 1300 static void bond_setup_by_slave(struct net_device *bond_dev, 1301 struct net_device *slave_dev) 1302 { 1303 bond_dev->header_ops = slave_dev->header_ops; 1304 1305 bond_dev->type = slave_dev->type; 1306 bond_dev->hard_header_len = slave_dev->hard_header_len; 1307 bond_dev->addr_len = slave_dev->addr_len; 1308 1309 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1310 slave_dev->addr_len); 1311 } 1312 1313 /* On bonding slaves other than the currently active slave, suppress 1314 * duplicates except for alb non-mcast/bcast. 1315 */ 1316 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1317 struct slave *slave, 1318 struct bonding *bond) 1319 { 1320 if (bond_is_slave_inactive(slave)) { 1321 if (BOND_MODE(bond) == BOND_MODE_ALB && 1322 skb->pkt_type != PACKET_BROADCAST && 1323 skb->pkt_type != PACKET_MULTICAST) 1324 return false; 1325 return true; 1326 } 1327 return false; 1328 } 1329 1330 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1331 { 1332 struct sk_buff *skb = *pskb; 1333 struct slave *slave; 1334 struct bonding *bond; 1335 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1336 struct slave *); 1337 int ret = RX_HANDLER_ANOTHER; 1338 1339 skb = skb_share_check(skb, GFP_ATOMIC); 1340 if (unlikely(!skb)) 1341 return RX_HANDLER_CONSUMED; 1342 1343 *pskb = skb; 1344 1345 slave = bond_slave_get_rcu(skb->dev); 1346 bond = slave->bond; 1347 1348 recv_probe = READ_ONCE(bond->recv_probe); 1349 if (recv_probe) { 1350 ret = recv_probe(skb, bond, slave); 1351 if (ret == RX_HANDLER_CONSUMED) { 1352 consume_skb(skb); 1353 return ret; 1354 } 1355 } 1356 1357 /* 1358 * For packets determined by bond_should_deliver_exact_match() call to 1359 * be suppressed we want to make an exception for link-local packets. 1360 * This is necessary for e.g. LLDP daemons to be able to monitor 1361 * inactive slave links without being forced to bind to them 1362 * explicitly. 1363 * 1364 * At the same time, packets that are passed to the bonding master 1365 * (including link-local ones) can have their originating interface 1366 * determined via PACKET_ORIGDEV socket option. 1367 */ 1368 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1369 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1370 return RX_HANDLER_PASS; 1371 return RX_HANDLER_EXACT; 1372 } 1373 1374 skb->dev = bond->dev; 1375 1376 if (BOND_MODE(bond) == BOND_MODE_ALB && 1377 netif_is_bridge_port(bond->dev) && 1378 skb->pkt_type == PACKET_HOST) { 1379 1380 if (unlikely(skb_cow_head(skb, 1381 skb->data - skb_mac_header(skb)))) { 1382 kfree_skb(skb); 1383 return RX_HANDLER_CONSUMED; 1384 } 1385 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1386 bond->dev->addr_len); 1387 } 1388 1389 return ret; 1390 } 1391 1392 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1393 { 1394 switch (BOND_MODE(bond)) { 1395 case BOND_MODE_ROUNDROBIN: 1396 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1397 case BOND_MODE_ACTIVEBACKUP: 1398 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1399 case BOND_MODE_BROADCAST: 1400 return NETDEV_LAG_TX_TYPE_BROADCAST; 1401 case BOND_MODE_XOR: 1402 case BOND_MODE_8023AD: 1403 return NETDEV_LAG_TX_TYPE_HASH; 1404 default: 1405 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1406 } 1407 } 1408 1409 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1410 enum netdev_lag_tx_type type) 1411 { 1412 if (type != NETDEV_LAG_TX_TYPE_HASH) 1413 return NETDEV_LAG_HASH_NONE; 1414 1415 switch (bond->params.xmit_policy) { 1416 case BOND_XMIT_POLICY_LAYER2: 1417 return NETDEV_LAG_HASH_L2; 1418 case BOND_XMIT_POLICY_LAYER34: 1419 return NETDEV_LAG_HASH_L34; 1420 case BOND_XMIT_POLICY_LAYER23: 1421 return NETDEV_LAG_HASH_L23; 1422 case BOND_XMIT_POLICY_ENCAP23: 1423 return NETDEV_LAG_HASH_E23; 1424 case BOND_XMIT_POLICY_ENCAP34: 1425 return NETDEV_LAG_HASH_E34; 1426 default: 1427 return NETDEV_LAG_HASH_UNKNOWN; 1428 } 1429 } 1430 1431 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1432 struct netlink_ext_ack *extack) 1433 { 1434 struct netdev_lag_upper_info lag_upper_info; 1435 enum netdev_lag_tx_type type; 1436 1437 type = bond_lag_tx_type(bond); 1438 lag_upper_info.tx_type = type; 1439 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1440 1441 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1442 &lag_upper_info, extack); 1443 } 1444 1445 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1446 { 1447 netdev_upper_dev_unlink(slave->dev, bond->dev); 1448 slave->dev->flags &= ~IFF_SLAVE; 1449 } 1450 1451 static struct slave *bond_alloc_slave(struct bonding *bond) 1452 { 1453 struct slave *slave = NULL; 1454 1455 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1456 if (!slave) 1457 return NULL; 1458 1459 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1460 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1461 GFP_KERNEL); 1462 if (!SLAVE_AD_INFO(slave)) { 1463 kfree(slave); 1464 return NULL; 1465 } 1466 } 1467 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1468 1469 return slave; 1470 } 1471 1472 static void bond_free_slave(struct slave *slave) 1473 { 1474 struct bonding *bond = bond_get_bond_by_slave(slave); 1475 1476 cancel_delayed_work_sync(&slave->notify_work); 1477 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1478 kfree(SLAVE_AD_INFO(slave)); 1479 1480 kfree(slave); 1481 } 1482 1483 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1484 { 1485 info->bond_mode = BOND_MODE(bond); 1486 info->miimon = bond->params.miimon; 1487 info->num_slaves = bond->slave_cnt; 1488 } 1489 1490 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1491 { 1492 strcpy(info->slave_name, slave->dev->name); 1493 info->link = slave->link; 1494 info->state = bond_slave_state(slave); 1495 info->link_failure_count = slave->link_failure_count; 1496 } 1497 1498 static void bond_netdev_notify_work(struct work_struct *_work) 1499 { 1500 struct slave *slave = container_of(_work, struct slave, 1501 notify_work.work); 1502 1503 if (rtnl_trylock()) { 1504 struct netdev_bonding_info binfo; 1505 1506 bond_fill_ifslave(slave, &binfo.slave); 1507 bond_fill_ifbond(slave->bond, &binfo.master); 1508 netdev_bonding_info_change(slave->dev, &binfo); 1509 rtnl_unlock(); 1510 } else { 1511 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1512 } 1513 } 1514 1515 void bond_queue_slave_event(struct slave *slave) 1516 { 1517 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1518 } 1519 1520 void bond_lower_state_changed(struct slave *slave) 1521 { 1522 struct netdev_lag_lower_state_info info; 1523 1524 info.link_up = slave->link == BOND_LINK_UP || 1525 slave->link == BOND_LINK_FAIL; 1526 info.tx_enabled = bond_is_active_slave(slave); 1527 netdev_lower_state_changed(slave->dev, &info); 1528 } 1529 1530 /* enslave device <slave> to bond device <master> */ 1531 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1532 struct netlink_ext_ack *extack) 1533 { 1534 struct bonding *bond = netdev_priv(bond_dev); 1535 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1536 struct slave *new_slave = NULL, *prev_slave; 1537 struct sockaddr_storage ss; 1538 int link_reporting; 1539 int res = 0, i; 1540 1541 if (!bond->params.use_carrier && 1542 slave_dev->ethtool_ops->get_link == NULL && 1543 slave_ops->ndo_do_ioctl == NULL) { 1544 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1545 } 1546 1547 /* already in-use? */ 1548 if (netdev_is_rx_handler_busy(slave_dev)) { 1549 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1550 slave_err(bond_dev, slave_dev, 1551 "Error: Device is in use and cannot be enslaved\n"); 1552 return -EBUSY; 1553 } 1554 1555 if (bond_dev == slave_dev) { 1556 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1557 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1558 return -EPERM; 1559 } 1560 1561 /* vlan challenged mutual exclusion */ 1562 /* no need to lock since we're protected by rtnl_lock */ 1563 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1564 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1565 if (vlan_uses_dev(bond_dev)) { 1566 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1567 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n"); 1568 return -EPERM; 1569 } else { 1570 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1571 } 1572 } else { 1573 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1574 } 1575 1576 if (slave_dev->features & NETIF_F_HW_ESP) 1577 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1578 1579 /* Old ifenslave binaries are no longer supported. These can 1580 * be identified with moderate accuracy by the state of the slave: 1581 * the current ifenslave will set the interface down prior to 1582 * enslaving it; the old ifenslave will not. 1583 */ 1584 if (slave_dev->flags & IFF_UP) { 1585 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1586 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n"); 1587 return -EPERM; 1588 } 1589 1590 /* set bonding device ether type by slave - bonding netdevices are 1591 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1592 * there is a need to override some of the type dependent attribs/funcs. 1593 * 1594 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1595 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1596 */ 1597 if (!bond_has_slaves(bond)) { 1598 if (bond_dev->type != slave_dev->type) { 1599 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1600 bond_dev->type, slave_dev->type); 1601 1602 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1603 bond_dev); 1604 res = notifier_to_errno(res); 1605 if (res) { 1606 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1607 return -EBUSY; 1608 } 1609 1610 /* Flush unicast and multicast addresses */ 1611 dev_uc_flush(bond_dev); 1612 dev_mc_flush(bond_dev); 1613 1614 if (slave_dev->type != ARPHRD_ETHER) 1615 bond_setup_by_slave(bond_dev, slave_dev); 1616 else { 1617 ether_setup(bond_dev); 1618 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1619 } 1620 1621 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1622 bond_dev); 1623 } 1624 } else if (bond_dev->type != slave_dev->type) { 1625 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1626 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n", 1627 slave_dev->type, bond_dev->type); 1628 return -EINVAL; 1629 } 1630 1631 if (slave_dev->type == ARPHRD_INFINIBAND && 1632 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1633 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1634 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n", 1635 slave_dev->type); 1636 res = -EOPNOTSUPP; 1637 goto err_undo_flags; 1638 } 1639 1640 if (!slave_ops->ndo_set_mac_address || 1641 slave_dev->type == ARPHRD_INFINIBAND) { 1642 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1643 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1644 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1645 if (!bond_has_slaves(bond)) { 1646 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1647 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1648 } else { 1649 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1650 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"); 1651 res = -EOPNOTSUPP; 1652 goto err_undo_flags; 1653 } 1654 } 1655 } 1656 1657 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1658 1659 /* If this is the first slave, then we need to set the master's hardware 1660 * address to be the same as the slave's. 1661 */ 1662 if (!bond_has_slaves(bond) && 1663 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1664 res = bond_set_dev_addr(bond->dev, slave_dev); 1665 if (res) 1666 goto err_undo_flags; 1667 } 1668 1669 new_slave = bond_alloc_slave(bond); 1670 if (!new_slave) { 1671 res = -ENOMEM; 1672 goto err_undo_flags; 1673 } 1674 1675 new_slave->bond = bond; 1676 new_slave->dev = slave_dev; 1677 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1678 * is set via sysfs or module option if desired. 1679 */ 1680 new_slave->queue_id = 0; 1681 1682 /* Save slave's original mtu and then set it to match the bond */ 1683 new_slave->original_mtu = slave_dev->mtu; 1684 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1685 if (res) { 1686 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1687 goto err_free; 1688 } 1689 1690 /* Save slave's original ("permanent") mac address for modes 1691 * that need it, and for restoring it upon release, and then 1692 * set it to the master's address 1693 */ 1694 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1695 slave_dev->addr_len); 1696 1697 if (!bond->params.fail_over_mac || 1698 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1699 /* Set slave to master's mac address. The application already 1700 * set the master's mac address to that of the first slave 1701 */ 1702 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1703 ss.ss_family = slave_dev->type; 1704 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1705 extack); 1706 if (res) { 1707 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1708 goto err_restore_mtu; 1709 } 1710 } 1711 1712 /* set slave flag before open to prevent IPv6 addrconf */ 1713 slave_dev->flags |= IFF_SLAVE; 1714 1715 /* open the slave since the application closed it */ 1716 res = dev_open(slave_dev, extack); 1717 if (res) { 1718 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1719 goto err_restore_mac; 1720 } 1721 1722 slave_dev->priv_flags |= IFF_BONDING; 1723 /* initialize slave stats */ 1724 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1725 1726 if (bond_is_lb(bond)) { 1727 /* bond_alb_init_slave() must be called before all other stages since 1728 * it might fail and we do not want to have to undo everything 1729 */ 1730 res = bond_alb_init_slave(bond, new_slave); 1731 if (res) 1732 goto err_close; 1733 } 1734 1735 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1736 if (res) { 1737 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1738 goto err_close; 1739 } 1740 1741 prev_slave = bond_last_slave(bond); 1742 1743 new_slave->delay = 0; 1744 new_slave->link_failure_count = 0; 1745 1746 if (bond_update_speed_duplex(new_slave) && 1747 bond_needs_speed_duplex(bond)) 1748 new_slave->link = BOND_LINK_DOWN; 1749 1750 new_slave->last_rx = jiffies - 1751 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1752 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1753 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1754 1755 if (bond->params.miimon && !bond->params.use_carrier) { 1756 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1757 1758 if ((link_reporting == -1) && !bond->params.arp_interval) { 1759 /* miimon is set but a bonded network driver 1760 * does not support ETHTOOL/MII and 1761 * arp_interval is not set. Note: if 1762 * use_carrier is enabled, we will never go 1763 * here (because netif_carrier is always 1764 * supported); thus, we don't need to change 1765 * the messages for netif_carrier. 1766 */ 1767 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"); 1768 } else if (link_reporting == -1) { 1769 /* unable get link status using mii/ethtool */ 1770 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"); 1771 } 1772 } 1773 1774 /* check for initial state */ 1775 new_slave->link = BOND_LINK_NOCHANGE; 1776 if (bond->params.miimon) { 1777 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1778 if (bond->params.updelay) { 1779 bond_set_slave_link_state(new_slave, 1780 BOND_LINK_BACK, 1781 BOND_SLAVE_NOTIFY_NOW); 1782 new_slave->delay = bond->params.updelay; 1783 } else { 1784 bond_set_slave_link_state(new_slave, 1785 BOND_LINK_UP, 1786 BOND_SLAVE_NOTIFY_NOW); 1787 } 1788 } else { 1789 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1790 BOND_SLAVE_NOTIFY_NOW); 1791 } 1792 } else if (bond->params.arp_interval) { 1793 bond_set_slave_link_state(new_slave, 1794 (netif_carrier_ok(slave_dev) ? 1795 BOND_LINK_UP : BOND_LINK_DOWN), 1796 BOND_SLAVE_NOTIFY_NOW); 1797 } else { 1798 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1799 BOND_SLAVE_NOTIFY_NOW); 1800 } 1801 1802 if (new_slave->link != BOND_LINK_DOWN) 1803 new_slave->last_link_up = jiffies; 1804 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 1805 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1806 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1807 1808 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1809 /* if there is a primary slave, remember it */ 1810 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1811 rcu_assign_pointer(bond->primary_slave, new_slave); 1812 bond->force_primary = true; 1813 } 1814 } 1815 1816 switch (BOND_MODE(bond)) { 1817 case BOND_MODE_ACTIVEBACKUP: 1818 bond_set_slave_inactive_flags(new_slave, 1819 BOND_SLAVE_NOTIFY_NOW); 1820 break; 1821 case BOND_MODE_8023AD: 1822 /* in 802.3ad mode, the internal mechanism 1823 * will activate the slaves in the selected 1824 * aggregator 1825 */ 1826 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1827 /* if this is the first slave */ 1828 if (!prev_slave) { 1829 SLAVE_AD_INFO(new_slave)->id = 1; 1830 /* Initialize AD with the number of times that the AD timer is called in 1 second 1831 * can be called only after the mac address of the bond is set 1832 */ 1833 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1834 } else { 1835 SLAVE_AD_INFO(new_slave)->id = 1836 SLAVE_AD_INFO(prev_slave)->id + 1; 1837 } 1838 1839 bond_3ad_bind_slave(new_slave); 1840 break; 1841 case BOND_MODE_TLB: 1842 case BOND_MODE_ALB: 1843 bond_set_active_slave(new_slave); 1844 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1845 break; 1846 default: 1847 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 1848 1849 /* always active in trunk mode */ 1850 bond_set_active_slave(new_slave); 1851 1852 /* In trunking mode there is little meaning to curr_active_slave 1853 * anyway (it holds no special properties of the bond device), 1854 * so we can change it without calling change_active_interface() 1855 */ 1856 if (!rcu_access_pointer(bond->curr_active_slave) && 1857 new_slave->link == BOND_LINK_UP) 1858 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1859 1860 break; 1861 } /* switch(bond_mode) */ 1862 1863 #ifdef CONFIG_NET_POLL_CONTROLLER 1864 if (bond->dev->npinfo) { 1865 if (slave_enable_netpoll(new_slave)) { 1866 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1867 res = -EBUSY; 1868 goto err_detach; 1869 } 1870 } 1871 #endif 1872 1873 if (!(bond_dev->features & NETIF_F_LRO)) 1874 dev_disable_lro(slave_dev); 1875 1876 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1877 new_slave); 1878 if (res) { 1879 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 1880 goto err_detach; 1881 } 1882 1883 res = bond_master_upper_dev_link(bond, new_slave, extack); 1884 if (res) { 1885 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1886 goto err_unregister; 1887 } 1888 1889 res = bond_sysfs_slave_add(new_slave); 1890 if (res) { 1891 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1892 goto err_upper_unlink; 1893 } 1894 1895 /* If the mode uses primary, then the following is handled by 1896 * bond_change_active_slave(). 1897 */ 1898 if (!bond_uses_primary(bond)) { 1899 /* set promiscuity level to new slave */ 1900 if (bond_dev->flags & IFF_PROMISC) { 1901 res = dev_set_promiscuity(slave_dev, 1); 1902 if (res) 1903 goto err_sysfs_del; 1904 } 1905 1906 /* set allmulti level to new slave */ 1907 if (bond_dev->flags & IFF_ALLMULTI) { 1908 res = dev_set_allmulti(slave_dev, 1); 1909 if (res) { 1910 if (bond_dev->flags & IFF_PROMISC) 1911 dev_set_promiscuity(slave_dev, -1); 1912 goto err_sysfs_del; 1913 } 1914 } 1915 1916 netif_addr_lock_bh(bond_dev); 1917 dev_mc_sync_multiple(slave_dev, bond_dev); 1918 dev_uc_sync_multiple(slave_dev, bond_dev); 1919 netif_addr_unlock_bh(bond_dev); 1920 1921 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1922 /* add lacpdu mc addr to mc list */ 1923 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1924 1925 dev_mc_add(slave_dev, lacpdu_multicast); 1926 } 1927 } 1928 1929 bond->slave_cnt++; 1930 bond_compute_features(bond); 1931 bond_set_carrier(bond); 1932 1933 if (bond_uses_primary(bond)) { 1934 block_netpoll_tx(); 1935 bond_select_active_slave(bond); 1936 unblock_netpoll_tx(); 1937 } 1938 1939 if (bond_mode_can_use_xmit_hash(bond)) 1940 bond_update_slave_arr(bond, NULL); 1941 1942 1943 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 1944 bond_is_active_slave(new_slave) ? "an active" : "a backup", 1945 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 1946 1947 /* enslave is successful */ 1948 bond_queue_slave_event(new_slave); 1949 return 0; 1950 1951 /* Undo stages on error */ 1952 err_sysfs_del: 1953 bond_sysfs_slave_del(new_slave); 1954 1955 err_upper_unlink: 1956 bond_upper_dev_unlink(bond, new_slave); 1957 1958 err_unregister: 1959 netdev_rx_handler_unregister(slave_dev); 1960 1961 err_detach: 1962 vlan_vids_del_by_dev(slave_dev, bond_dev); 1963 if (rcu_access_pointer(bond->primary_slave) == new_slave) 1964 RCU_INIT_POINTER(bond->primary_slave, NULL); 1965 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 1966 block_netpoll_tx(); 1967 bond_change_active_slave(bond, NULL); 1968 bond_select_active_slave(bond); 1969 unblock_netpoll_tx(); 1970 } 1971 /* either primary_slave or curr_active_slave might've changed */ 1972 synchronize_rcu(); 1973 slave_disable_netpoll(new_slave); 1974 1975 err_close: 1976 if (!netif_is_bond_master(slave_dev)) 1977 slave_dev->priv_flags &= ~IFF_BONDING; 1978 dev_close(slave_dev); 1979 1980 err_restore_mac: 1981 slave_dev->flags &= ~IFF_SLAVE; 1982 if (!bond->params.fail_over_mac || 1983 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1984 /* XXX TODO - fom follow mode needs to change master's 1985 * MAC if this slave's MAC is in use by the bond, or at 1986 * least print a warning. 1987 */ 1988 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 1989 new_slave->dev->addr_len); 1990 ss.ss_family = slave_dev->type; 1991 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 1992 } 1993 1994 err_restore_mtu: 1995 dev_set_mtu(slave_dev, new_slave->original_mtu); 1996 1997 err_free: 1998 bond_free_slave(new_slave); 1999 2000 err_undo_flags: 2001 /* Enslave of first slave has failed and we need to fix master's mac */ 2002 if (!bond_has_slaves(bond)) { 2003 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2004 slave_dev->dev_addr)) 2005 eth_hw_addr_random(bond_dev); 2006 if (bond_dev->type != ARPHRD_ETHER) { 2007 dev_close(bond_dev); 2008 ether_setup(bond_dev); 2009 bond_dev->flags |= IFF_MASTER; 2010 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2011 } 2012 } 2013 2014 return res; 2015 } 2016 2017 /* Try to release the slave device <slave> from the bond device <master> 2018 * It is legal to access curr_active_slave without a lock because all the function 2019 * is RTNL-locked. If "all" is true it means that the function is being called 2020 * while destroying a bond interface and all slaves are being released. 2021 * 2022 * The rules for slave state should be: 2023 * for Active/Backup: 2024 * Active stays on all backups go down 2025 * for Bonded connections: 2026 * The first up interface should be left on and all others downed. 2027 */ 2028 static int __bond_release_one(struct net_device *bond_dev, 2029 struct net_device *slave_dev, 2030 bool all, bool unregister) 2031 { 2032 struct bonding *bond = netdev_priv(bond_dev); 2033 struct slave *slave, *oldcurrent; 2034 struct sockaddr_storage ss; 2035 int old_flags = bond_dev->flags; 2036 netdev_features_t old_features = bond_dev->features; 2037 2038 /* slave is not a slave or master is not master of this slave */ 2039 if (!(slave_dev->flags & IFF_SLAVE) || 2040 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2041 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2042 return -EINVAL; 2043 } 2044 2045 block_netpoll_tx(); 2046 2047 slave = bond_get_slave_by_dev(bond, slave_dev); 2048 if (!slave) { 2049 /* not a slave of this bond */ 2050 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2051 unblock_netpoll_tx(); 2052 return -EINVAL; 2053 } 2054 2055 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2056 2057 bond_sysfs_slave_del(slave); 2058 2059 /* recompute stats just before removing the slave */ 2060 bond_get_stats(bond->dev, &bond->bond_stats); 2061 2062 bond_upper_dev_unlink(bond, slave); 2063 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2064 * for this slave anymore. 2065 */ 2066 netdev_rx_handler_unregister(slave_dev); 2067 2068 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2069 bond_3ad_unbind_slave(slave); 2070 2071 if (bond_mode_can_use_xmit_hash(bond)) 2072 bond_update_slave_arr(bond, slave); 2073 2074 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2075 bond_is_active_slave(slave) ? "active" : "backup"); 2076 2077 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2078 2079 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2080 2081 if (!all && (!bond->params.fail_over_mac || 2082 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2083 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2084 bond_has_slaves(bond)) 2085 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", 2086 slave->perm_hwaddr); 2087 } 2088 2089 if (rtnl_dereference(bond->primary_slave) == slave) 2090 RCU_INIT_POINTER(bond->primary_slave, NULL); 2091 2092 if (oldcurrent == slave) 2093 bond_change_active_slave(bond, NULL); 2094 2095 if (bond_is_lb(bond)) { 2096 /* Must be called only after the slave has been 2097 * detached from the list and the curr_active_slave 2098 * has been cleared (if our_slave == old_current), 2099 * but before a new active slave is selected. 2100 */ 2101 bond_alb_deinit_slave(bond, slave); 2102 } 2103 2104 if (all) { 2105 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2106 } else if (oldcurrent == slave) { 2107 /* Note that we hold RTNL over this sequence, so there 2108 * is no concern that another slave add/remove event 2109 * will interfere. 2110 */ 2111 bond_select_active_slave(bond); 2112 } 2113 2114 if (!bond_has_slaves(bond)) { 2115 bond_set_carrier(bond); 2116 eth_hw_addr_random(bond_dev); 2117 } 2118 2119 unblock_netpoll_tx(); 2120 synchronize_rcu(); 2121 bond->slave_cnt--; 2122 2123 if (!bond_has_slaves(bond)) { 2124 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2125 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2126 } 2127 2128 bond_compute_features(bond); 2129 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2130 (old_features & NETIF_F_VLAN_CHALLENGED)) 2131 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2132 2133 vlan_vids_del_by_dev(slave_dev, bond_dev); 2134 2135 /* If the mode uses primary, then this case was handled above by 2136 * bond_change_active_slave(..., NULL) 2137 */ 2138 if (!bond_uses_primary(bond)) { 2139 /* unset promiscuity level from slave 2140 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2141 * of the IFF_PROMISC flag in the bond_dev, but we need the 2142 * value of that flag before that change, as that was the value 2143 * when this slave was attached, so we cache at the start of the 2144 * function and use it here. Same goes for ALLMULTI below 2145 */ 2146 if (old_flags & IFF_PROMISC) 2147 dev_set_promiscuity(slave_dev, -1); 2148 2149 /* unset allmulti level from slave */ 2150 if (old_flags & IFF_ALLMULTI) 2151 dev_set_allmulti(slave_dev, -1); 2152 2153 bond_hw_addr_flush(bond_dev, slave_dev); 2154 } 2155 2156 slave_disable_netpoll(slave); 2157 2158 /* close slave before restoring its mac address */ 2159 dev_close(slave_dev); 2160 2161 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2162 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2163 /* restore original ("permanent") mac address */ 2164 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2165 slave->dev->addr_len); 2166 ss.ss_family = slave_dev->type; 2167 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2168 } 2169 2170 if (unregister) 2171 __dev_set_mtu(slave_dev, slave->original_mtu); 2172 else 2173 dev_set_mtu(slave_dev, slave->original_mtu); 2174 2175 if (!netif_is_bond_master(slave_dev)) 2176 slave_dev->priv_flags &= ~IFF_BONDING; 2177 2178 bond_free_slave(slave); 2179 2180 return 0; 2181 } 2182 2183 /* A wrapper used because of ndo_del_link */ 2184 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2185 { 2186 return __bond_release_one(bond_dev, slave_dev, false, false); 2187 } 2188 2189 /* First release a slave and then destroy the bond if no more slaves are left. 2190 * Must be under rtnl_lock when this function is called. 2191 */ 2192 static int bond_release_and_destroy(struct net_device *bond_dev, 2193 struct net_device *slave_dev) 2194 { 2195 struct bonding *bond = netdev_priv(bond_dev); 2196 int ret; 2197 2198 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2199 if (ret == 0 && !bond_has_slaves(bond)) { 2200 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2201 netdev_info(bond_dev, "Destroying bond\n"); 2202 bond_remove_proc_entry(bond); 2203 unregister_netdevice(bond_dev); 2204 } 2205 return ret; 2206 } 2207 2208 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2209 { 2210 struct bonding *bond = netdev_priv(bond_dev); 2211 bond_fill_ifbond(bond, info); 2212 } 2213 2214 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2215 { 2216 struct bonding *bond = netdev_priv(bond_dev); 2217 struct list_head *iter; 2218 int i = 0, res = -ENODEV; 2219 struct slave *slave; 2220 2221 bond_for_each_slave(bond, slave, iter) { 2222 if (i++ == (int)info->slave_id) { 2223 res = 0; 2224 bond_fill_ifslave(slave, info); 2225 break; 2226 } 2227 } 2228 2229 return res; 2230 } 2231 2232 /*-------------------------------- Monitoring -------------------------------*/ 2233 2234 /* called with rcu_read_lock() */ 2235 static int bond_miimon_inspect(struct bonding *bond) 2236 { 2237 int link_state, commit = 0; 2238 struct list_head *iter; 2239 struct slave *slave; 2240 bool ignore_updelay; 2241 2242 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2243 2244 bond_for_each_slave_rcu(bond, slave, iter) { 2245 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2246 2247 link_state = bond_check_dev_link(bond, slave->dev, 0); 2248 2249 switch (slave->link) { 2250 case BOND_LINK_UP: 2251 if (link_state) 2252 continue; 2253 2254 bond_propose_link_state(slave, BOND_LINK_FAIL); 2255 commit++; 2256 slave->delay = bond->params.downdelay; 2257 if (slave->delay) { 2258 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2259 (BOND_MODE(bond) == 2260 BOND_MODE_ACTIVEBACKUP) ? 2261 (bond_is_active_slave(slave) ? 2262 "active " : "backup ") : "", 2263 bond->params.downdelay * bond->params.miimon); 2264 } 2265 /*FALLTHRU*/ 2266 case BOND_LINK_FAIL: 2267 if (link_state) { 2268 /* recovered before downdelay expired */ 2269 bond_propose_link_state(slave, BOND_LINK_UP); 2270 slave->last_link_up = jiffies; 2271 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2272 (bond->params.downdelay - slave->delay) * 2273 bond->params.miimon); 2274 commit++; 2275 continue; 2276 } 2277 2278 if (slave->delay <= 0) { 2279 bond_propose_link_state(slave, BOND_LINK_DOWN); 2280 commit++; 2281 continue; 2282 } 2283 2284 slave->delay--; 2285 break; 2286 2287 case BOND_LINK_DOWN: 2288 if (!link_state) 2289 continue; 2290 2291 bond_propose_link_state(slave, BOND_LINK_BACK); 2292 commit++; 2293 slave->delay = bond->params.updelay; 2294 2295 if (slave->delay) { 2296 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2297 ignore_updelay ? 0 : 2298 bond->params.updelay * 2299 bond->params.miimon); 2300 } 2301 /*FALLTHRU*/ 2302 case BOND_LINK_BACK: 2303 if (!link_state) { 2304 bond_propose_link_state(slave, BOND_LINK_DOWN); 2305 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2306 (bond->params.updelay - slave->delay) * 2307 bond->params.miimon); 2308 commit++; 2309 continue; 2310 } 2311 2312 if (ignore_updelay) 2313 slave->delay = 0; 2314 2315 if (slave->delay <= 0) { 2316 bond_propose_link_state(slave, BOND_LINK_UP); 2317 commit++; 2318 ignore_updelay = false; 2319 continue; 2320 } 2321 2322 slave->delay--; 2323 break; 2324 } 2325 } 2326 2327 return commit; 2328 } 2329 2330 static void bond_miimon_link_change(struct bonding *bond, 2331 struct slave *slave, 2332 char link) 2333 { 2334 switch (BOND_MODE(bond)) { 2335 case BOND_MODE_8023AD: 2336 bond_3ad_handle_link_change(slave, link); 2337 break; 2338 case BOND_MODE_TLB: 2339 case BOND_MODE_ALB: 2340 bond_alb_handle_link_change(bond, slave, link); 2341 break; 2342 case BOND_MODE_XOR: 2343 bond_update_slave_arr(bond, NULL); 2344 break; 2345 } 2346 } 2347 2348 static void bond_miimon_commit(struct bonding *bond) 2349 { 2350 struct list_head *iter; 2351 struct slave *slave, *primary; 2352 2353 bond_for_each_slave(bond, slave, iter) { 2354 switch (slave->link_new_state) { 2355 case BOND_LINK_NOCHANGE: 2356 /* For 802.3ad mode, check current slave speed and 2357 * duplex again in case its port was disabled after 2358 * invalid speed/duplex reporting but recovered before 2359 * link monitoring could make a decision on the actual 2360 * link status 2361 */ 2362 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2363 slave->link == BOND_LINK_UP) 2364 bond_3ad_adapter_speed_duplex_changed(slave); 2365 continue; 2366 2367 case BOND_LINK_UP: 2368 if (bond_update_speed_duplex(slave) && 2369 bond_needs_speed_duplex(bond)) { 2370 slave->link = BOND_LINK_DOWN; 2371 if (net_ratelimit()) 2372 slave_warn(bond->dev, slave->dev, 2373 "failed to get link speed/duplex\n"); 2374 continue; 2375 } 2376 bond_set_slave_link_state(slave, BOND_LINK_UP, 2377 BOND_SLAVE_NOTIFY_NOW); 2378 slave->last_link_up = jiffies; 2379 2380 primary = rtnl_dereference(bond->primary_slave); 2381 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2382 /* prevent it from being the active one */ 2383 bond_set_backup_slave(slave); 2384 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2385 /* make it immediately active */ 2386 bond_set_active_slave(slave); 2387 } 2388 2389 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2390 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2391 slave->duplex ? "full" : "half"); 2392 2393 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2394 2395 if (!bond->curr_active_slave || slave == primary) 2396 goto do_failover; 2397 2398 continue; 2399 2400 case BOND_LINK_DOWN: 2401 if (slave->link_failure_count < UINT_MAX) 2402 slave->link_failure_count++; 2403 2404 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2405 BOND_SLAVE_NOTIFY_NOW); 2406 2407 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2408 BOND_MODE(bond) == BOND_MODE_8023AD) 2409 bond_set_slave_inactive_flags(slave, 2410 BOND_SLAVE_NOTIFY_NOW); 2411 2412 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2413 2414 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2415 2416 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2417 goto do_failover; 2418 2419 continue; 2420 2421 default: 2422 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2423 slave->link_new_state); 2424 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2425 2426 continue; 2427 } 2428 2429 do_failover: 2430 block_netpoll_tx(); 2431 bond_select_active_slave(bond); 2432 unblock_netpoll_tx(); 2433 } 2434 2435 bond_set_carrier(bond); 2436 } 2437 2438 /* bond_mii_monitor 2439 * 2440 * Really a wrapper that splits the mii monitor into two phases: an 2441 * inspection, then (if inspection indicates something needs to be done) 2442 * an acquisition of appropriate locks followed by a commit phase to 2443 * implement whatever link state changes are indicated. 2444 */ 2445 static void bond_mii_monitor(struct work_struct *work) 2446 { 2447 struct bonding *bond = container_of(work, struct bonding, 2448 mii_work.work); 2449 bool should_notify_peers = false; 2450 bool commit; 2451 unsigned long delay; 2452 struct slave *slave; 2453 struct list_head *iter; 2454 2455 delay = msecs_to_jiffies(bond->params.miimon); 2456 2457 if (!bond_has_slaves(bond)) 2458 goto re_arm; 2459 2460 rcu_read_lock(); 2461 should_notify_peers = bond_should_notify_peers(bond); 2462 commit = !!bond_miimon_inspect(bond); 2463 if (bond->send_peer_notif) { 2464 rcu_read_unlock(); 2465 if (rtnl_trylock()) { 2466 bond->send_peer_notif--; 2467 rtnl_unlock(); 2468 } 2469 } else { 2470 rcu_read_unlock(); 2471 } 2472 2473 if (commit) { 2474 /* Race avoidance with bond_close cancel of workqueue */ 2475 if (!rtnl_trylock()) { 2476 delay = 1; 2477 should_notify_peers = false; 2478 goto re_arm; 2479 } 2480 2481 bond_for_each_slave(bond, slave, iter) { 2482 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2483 } 2484 bond_miimon_commit(bond); 2485 2486 rtnl_unlock(); /* might sleep, hold no other locks */ 2487 } 2488 2489 re_arm: 2490 if (bond->params.miimon) 2491 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2492 2493 if (should_notify_peers) { 2494 if (!rtnl_trylock()) 2495 return; 2496 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2497 rtnl_unlock(); 2498 } 2499 } 2500 2501 static int bond_upper_dev_walk(struct net_device *upper, void *data) 2502 { 2503 __be32 ip = *((__be32 *)data); 2504 2505 return ip == bond_confirm_addr(upper, 0, ip); 2506 } 2507 2508 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2509 { 2510 bool ret = false; 2511 2512 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2513 return true; 2514 2515 rcu_read_lock(); 2516 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip)) 2517 ret = true; 2518 rcu_read_unlock(); 2519 2520 return ret; 2521 } 2522 2523 /* We go to the (large) trouble of VLAN tagging ARP frames because 2524 * switches in VLAN mode (especially if ports are configured as 2525 * "native" to a VLAN) might not pass non-tagged frames. 2526 */ 2527 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2528 __be32 src_ip, struct bond_vlan_tag *tags) 2529 { 2530 struct sk_buff *skb; 2531 struct bond_vlan_tag *outer_tag = tags; 2532 struct net_device *slave_dev = slave->dev; 2533 struct net_device *bond_dev = slave->bond->dev; 2534 2535 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2536 arp_op, &dest_ip, &src_ip); 2537 2538 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2539 NULL, slave_dev->dev_addr, NULL); 2540 2541 if (!skb) { 2542 net_err_ratelimited("ARP packet allocation failed\n"); 2543 return; 2544 } 2545 2546 if (!tags || tags->vlan_proto == VLAN_N_VID) 2547 goto xmit; 2548 2549 tags++; 2550 2551 /* Go through all the tags backwards and add them to the packet */ 2552 while (tags->vlan_proto != VLAN_N_VID) { 2553 if (!tags->vlan_id) { 2554 tags++; 2555 continue; 2556 } 2557 2558 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2559 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2560 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2561 tags->vlan_id); 2562 if (!skb) { 2563 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2564 return; 2565 } 2566 2567 tags++; 2568 } 2569 /* Set the outer tag */ 2570 if (outer_tag->vlan_id) { 2571 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2572 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2573 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2574 outer_tag->vlan_id); 2575 } 2576 2577 xmit: 2578 arp_xmit(skb); 2579 } 2580 2581 /* Validate the device path between the @start_dev and the @end_dev. 2582 * The path is valid if the @end_dev is reachable through device 2583 * stacking. 2584 * When the path is validated, collect any vlan information in the 2585 * path. 2586 */ 2587 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2588 struct net_device *end_dev, 2589 int level) 2590 { 2591 struct bond_vlan_tag *tags; 2592 struct net_device *upper; 2593 struct list_head *iter; 2594 2595 if (start_dev == end_dev) { 2596 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2597 if (!tags) 2598 return ERR_PTR(-ENOMEM); 2599 tags[level].vlan_proto = VLAN_N_VID; 2600 return tags; 2601 } 2602 2603 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2604 tags = bond_verify_device_path(upper, end_dev, level + 1); 2605 if (IS_ERR_OR_NULL(tags)) { 2606 if (IS_ERR(tags)) 2607 return tags; 2608 continue; 2609 } 2610 if (is_vlan_dev(upper)) { 2611 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2612 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2613 } 2614 2615 return tags; 2616 } 2617 2618 return NULL; 2619 } 2620 2621 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2622 { 2623 struct rtable *rt; 2624 struct bond_vlan_tag *tags; 2625 __be32 *targets = bond->params.arp_targets, addr; 2626 int i; 2627 2628 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2629 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2630 __func__, &targets[i]); 2631 tags = NULL; 2632 2633 /* Find out through which dev should the packet go */ 2634 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2635 RTO_ONLINK, 0); 2636 if (IS_ERR(rt)) { 2637 /* there's no route to target - try to send arp 2638 * probe to generate any traffic (arp_validate=0) 2639 */ 2640 if (bond->params.arp_validate) 2641 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2642 bond->dev->name, 2643 &targets[i]); 2644 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2645 0, tags); 2646 continue; 2647 } 2648 2649 /* bond device itself */ 2650 if (rt->dst.dev == bond->dev) 2651 goto found; 2652 2653 rcu_read_lock(); 2654 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2655 rcu_read_unlock(); 2656 2657 if (!IS_ERR_OR_NULL(tags)) 2658 goto found; 2659 2660 /* Not our device - skip */ 2661 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2662 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2663 2664 ip_rt_put(rt); 2665 continue; 2666 2667 found: 2668 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2669 ip_rt_put(rt); 2670 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2671 kfree(tags); 2672 } 2673 } 2674 2675 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2676 { 2677 int i; 2678 2679 if (!sip || !bond_has_this_ip(bond, tip)) { 2680 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2681 __func__, &sip, &tip); 2682 return; 2683 } 2684 2685 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2686 if (i == -1) { 2687 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2688 __func__, &sip); 2689 return; 2690 } 2691 slave->last_rx = jiffies; 2692 slave->target_last_arp_rx[i] = jiffies; 2693 } 2694 2695 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2696 struct slave *slave) 2697 { 2698 struct arphdr *arp = (struct arphdr *)skb->data; 2699 struct slave *curr_active_slave, *curr_arp_slave; 2700 unsigned char *arp_ptr; 2701 __be32 sip, tip; 2702 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2703 unsigned int alen; 2704 2705 if (!slave_do_arp_validate(bond, slave)) { 2706 if ((slave_do_arp_validate_only(bond) && is_arp) || 2707 !slave_do_arp_validate_only(bond)) 2708 slave->last_rx = jiffies; 2709 return RX_HANDLER_ANOTHER; 2710 } else if (!is_arp) { 2711 return RX_HANDLER_ANOTHER; 2712 } 2713 2714 alen = arp_hdr_len(bond->dev); 2715 2716 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 2717 __func__, skb->dev->name); 2718 2719 if (alen > skb_headlen(skb)) { 2720 arp = kmalloc(alen, GFP_ATOMIC); 2721 if (!arp) 2722 goto out_unlock; 2723 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2724 goto out_unlock; 2725 } 2726 2727 if (arp->ar_hln != bond->dev->addr_len || 2728 skb->pkt_type == PACKET_OTHERHOST || 2729 skb->pkt_type == PACKET_LOOPBACK || 2730 arp->ar_hrd != htons(ARPHRD_ETHER) || 2731 arp->ar_pro != htons(ETH_P_IP) || 2732 arp->ar_pln != 4) 2733 goto out_unlock; 2734 2735 arp_ptr = (unsigned char *)(arp + 1); 2736 arp_ptr += bond->dev->addr_len; 2737 memcpy(&sip, arp_ptr, 4); 2738 arp_ptr += 4 + bond->dev->addr_len; 2739 memcpy(&tip, arp_ptr, 4); 2740 2741 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2742 __func__, slave->dev->name, bond_slave_state(slave), 2743 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2744 &sip, &tip); 2745 2746 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2747 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2748 2749 /* We 'trust' the received ARP enough to validate it if: 2750 * 2751 * (a) the slave receiving the ARP is active (which includes the 2752 * current ARP slave, if any), or 2753 * 2754 * (b) the receiving slave isn't active, but there is a currently 2755 * active slave and it received valid arp reply(s) after it became 2756 * the currently active slave, or 2757 * 2758 * (c) there is an ARP slave that sent an ARP during the prior ARP 2759 * interval, and we receive an ARP reply on any slave. We accept 2760 * these because switch FDB update delays may deliver the ARP 2761 * reply to a slave other than the sender of the ARP request. 2762 * 2763 * Note: for (b), backup slaves are receiving the broadcast ARP 2764 * request, not a reply. This request passes from the sending 2765 * slave through the L2 switch(es) to the receiving slave. Since 2766 * this is checking the request, sip/tip are swapped for 2767 * validation. 2768 * 2769 * This is done to avoid endless looping when we can't reach the 2770 * arp_ip_target and fool ourselves with our own arp requests. 2771 */ 2772 if (bond_is_active_slave(slave)) 2773 bond_validate_arp(bond, slave, sip, tip); 2774 else if (curr_active_slave && 2775 time_after(slave_last_rx(bond, curr_active_slave), 2776 curr_active_slave->last_link_up)) 2777 bond_validate_arp(bond, slave, tip, sip); 2778 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2779 bond_time_in_interval(bond, 2780 dev_trans_start(curr_arp_slave->dev), 1)) 2781 bond_validate_arp(bond, slave, sip, tip); 2782 2783 out_unlock: 2784 if (arp != (struct arphdr *)skb->data) 2785 kfree(arp); 2786 return RX_HANDLER_ANOTHER; 2787 } 2788 2789 /* function to verify if we're in the arp_interval timeslice, returns true if 2790 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2791 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2792 */ 2793 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2794 int mod) 2795 { 2796 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2797 2798 return time_in_range(jiffies, 2799 last_act - delta_in_ticks, 2800 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2801 } 2802 2803 /* This function is called regularly to monitor each slave's link 2804 * ensuring that traffic is being sent and received when arp monitoring 2805 * is used in load-balancing mode. if the adapter has been dormant, then an 2806 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2807 * arp monitoring in active backup mode. 2808 */ 2809 static void bond_loadbalance_arp_mon(struct bonding *bond) 2810 { 2811 struct slave *slave, *oldcurrent; 2812 struct list_head *iter; 2813 int do_failover = 0, slave_state_changed = 0; 2814 2815 if (!bond_has_slaves(bond)) 2816 goto re_arm; 2817 2818 rcu_read_lock(); 2819 2820 oldcurrent = rcu_dereference(bond->curr_active_slave); 2821 /* see if any of the previous devices are up now (i.e. they have 2822 * xmt and rcv traffic). the curr_active_slave does not come into 2823 * the picture unless it is null. also, slave->last_link_up is not 2824 * needed here because we send an arp on each slave and give a slave 2825 * as long as it needs to get the tx/rx within the delta. 2826 * TODO: what about up/down delay in arp mode? it wasn't here before 2827 * so it can wait 2828 */ 2829 bond_for_each_slave_rcu(bond, slave, iter) { 2830 unsigned long trans_start = dev_trans_start(slave->dev); 2831 2832 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2833 2834 if (slave->link != BOND_LINK_UP) { 2835 if (bond_time_in_interval(bond, trans_start, 1) && 2836 bond_time_in_interval(bond, slave->last_rx, 1)) { 2837 2838 bond_propose_link_state(slave, BOND_LINK_UP); 2839 slave_state_changed = 1; 2840 2841 /* primary_slave has no meaning in round-robin 2842 * mode. the window of a slave being up and 2843 * curr_active_slave being null after enslaving 2844 * is closed. 2845 */ 2846 if (!oldcurrent) { 2847 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 2848 do_failover = 1; 2849 } else { 2850 slave_info(bond->dev, slave->dev, "interface is now up\n"); 2851 } 2852 } 2853 } else { 2854 /* slave->link == BOND_LINK_UP */ 2855 2856 /* not all switches will respond to an arp request 2857 * when the source ip is 0, so don't take the link down 2858 * if we don't know our ip yet 2859 */ 2860 if (!bond_time_in_interval(bond, trans_start, 2) || 2861 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2862 2863 bond_propose_link_state(slave, BOND_LINK_DOWN); 2864 slave_state_changed = 1; 2865 2866 if (slave->link_failure_count < UINT_MAX) 2867 slave->link_failure_count++; 2868 2869 slave_info(bond->dev, slave->dev, "interface is now down\n"); 2870 2871 if (slave == oldcurrent) 2872 do_failover = 1; 2873 } 2874 } 2875 2876 /* note: if switch is in round-robin mode, all links 2877 * must tx arp to ensure all links rx an arp - otherwise 2878 * links may oscillate or not come up at all; if switch is 2879 * in something like xor mode, there is nothing we can 2880 * do - all replies will be rx'ed on same link causing slaves 2881 * to be unstable during low/no traffic periods 2882 */ 2883 if (bond_slave_is_up(slave)) 2884 bond_arp_send_all(bond, slave); 2885 } 2886 2887 rcu_read_unlock(); 2888 2889 if (do_failover || slave_state_changed) { 2890 if (!rtnl_trylock()) 2891 goto re_arm; 2892 2893 bond_for_each_slave(bond, slave, iter) { 2894 if (slave->link_new_state != BOND_LINK_NOCHANGE) 2895 slave->link = slave->link_new_state; 2896 } 2897 2898 if (slave_state_changed) { 2899 bond_slave_state_change(bond); 2900 if (BOND_MODE(bond) == BOND_MODE_XOR) 2901 bond_update_slave_arr(bond, NULL); 2902 } 2903 if (do_failover) { 2904 block_netpoll_tx(); 2905 bond_select_active_slave(bond); 2906 unblock_netpoll_tx(); 2907 } 2908 rtnl_unlock(); 2909 } 2910 2911 re_arm: 2912 if (bond->params.arp_interval) 2913 queue_delayed_work(bond->wq, &bond->arp_work, 2914 msecs_to_jiffies(bond->params.arp_interval)); 2915 } 2916 2917 /* Called to inspect slaves for active-backup mode ARP monitor link state 2918 * changes. Sets proposed link state in slaves to specify what action 2919 * should take place for the slave. Returns 0 if no changes are found, >0 2920 * if changes to link states must be committed. 2921 * 2922 * Called with rcu_read_lock held. 2923 */ 2924 static int bond_ab_arp_inspect(struct bonding *bond) 2925 { 2926 unsigned long trans_start, last_rx; 2927 struct list_head *iter; 2928 struct slave *slave; 2929 int commit = 0; 2930 2931 bond_for_each_slave_rcu(bond, slave, iter) { 2932 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2933 last_rx = slave_last_rx(bond, slave); 2934 2935 if (slave->link != BOND_LINK_UP) { 2936 if (bond_time_in_interval(bond, last_rx, 1)) { 2937 bond_propose_link_state(slave, BOND_LINK_UP); 2938 commit++; 2939 } 2940 continue; 2941 } 2942 2943 /* Give slaves 2*delta after being enslaved or made 2944 * active. This avoids bouncing, as the last receive 2945 * times need a full ARP monitor cycle to be updated. 2946 */ 2947 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 2948 continue; 2949 2950 /* Backup slave is down if: 2951 * - No current_arp_slave AND 2952 * - more than 3*delta since last receive AND 2953 * - the bond has an IP address 2954 * 2955 * Note: a non-null current_arp_slave indicates 2956 * the curr_active_slave went down and we are 2957 * searching for a new one; under this condition 2958 * we only take the curr_active_slave down - this 2959 * gives each slave a chance to tx/rx traffic 2960 * before being taken out 2961 */ 2962 if (!bond_is_active_slave(slave) && 2963 !rcu_access_pointer(bond->current_arp_slave) && 2964 !bond_time_in_interval(bond, last_rx, 3)) { 2965 bond_propose_link_state(slave, BOND_LINK_DOWN); 2966 commit++; 2967 } 2968 2969 /* Active slave is down if: 2970 * - more than 2*delta since transmitting OR 2971 * - (more than 2*delta since receive AND 2972 * the bond has an IP address) 2973 */ 2974 trans_start = dev_trans_start(slave->dev); 2975 if (bond_is_active_slave(slave) && 2976 (!bond_time_in_interval(bond, trans_start, 2) || 2977 !bond_time_in_interval(bond, last_rx, 2))) { 2978 bond_propose_link_state(slave, BOND_LINK_DOWN); 2979 commit++; 2980 } 2981 } 2982 2983 return commit; 2984 } 2985 2986 /* Called to commit link state changes noted by inspection step of 2987 * active-backup mode ARP monitor. 2988 * 2989 * Called with RTNL hold. 2990 */ 2991 static void bond_ab_arp_commit(struct bonding *bond) 2992 { 2993 unsigned long trans_start; 2994 struct list_head *iter; 2995 struct slave *slave; 2996 2997 bond_for_each_slave(bond, slave, iter) { 2998 switch (slave->link_new_state) { 2999 case BOND_LINK_NOCHANGE: 3000 continue; 3001 3002 case BOND_LINK_UP: 3003 trans_start = dev_trans_start(slave->dev); 3004 if (rtnl_dereference(bond->curr_active_slave) != slave || 3005 (!rtnl_dereference(bond->curr_active_slave) && 3006 bond_time_in_interval(bond, trans_start, 1))) { 3007 struct slave *current_arp_slave; 3008 3009 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3010 bond_set_slave_link_state(slave, BOND_LINK_UP, 3011 BOND_SLAVE_NOTIFY_NOW); 3012 if (current_arp_slave) { 3013 bond_set_slave_inactive_flags( 3014 current_arp_slave, 3015 BOND_SLAVE_NOTIFY_NOW); 3016 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3017 } 3018 3019 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3020 3021 if (!rtnl_dereference(bond->curr_active_slave) || 3022 slave == rtnl_dereference(bond->primary_slave)) 3023 goto do_failover; 3024 3025 } 3026 3027 continue; 3028 3029 case BOND_LINK_DOWN: 3030 if (slave->link_failure_count < UINT_MAX) 3031 slave->link_failure_count++; 3032 3033 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3034 BOND_SLAVE_NOTIFY_NOW); 3035 bond_set_slave_inactive_flags(slave, 3036 BOND_SLAVE_NOTIFY_NOW); 3037 3038 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3039 3040 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3041 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3042 goto do_failover; 3043 } 3044 3045 continue; 3046 3047 default: 3048 slave_err(bond->dev, slave->dev, 3049 "impossible: link_new_state %d on slave\n", 3050 slave->link_new_state); 3051 continue; 3052 } 3053 3054 do_failover: 3055 block_netpoll_tx(); 3056 bond_select_active_slave(bond); 3057 unblock_netpoll_tx(); 3058 } 3059 3060 bond_set_carrier(bond); 3061 } 3062 3063 /* Send ARP probes for active-backup mode ARP monitor. 3064 * 3065 * Called with rcu_read_lock held. 3066 */ 3067 static bool bond_ab_arp_probe(struct bonding *bond) 3068 { 3069 struct slave *slave, *before = NULL, *new_slave = NULL, 3070 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3071 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3072 struct list_head *iter; 3073 bool found = false; 3074 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3075 3076 if (curr_arp_slave && curr_active_slave) 3077 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3078 curr_arp_slave->dev->name, 3079 curr_active_slave->dev->name); 3080 3081 if (curr_active_slave) { 3082 bond_arp_send_all(bond, curr_active_slave); 3083 return should_notify_rtnl; 3084 } 3085 3086 /* if we don't have a curr_active_slave, search for the next available 3087 * backup slave from the current_arp_slave and make it the candidate 3088 * for becoming the curr_active_slave 3089 */ 3090 3091 if (!curr_arp_slave) { 3092 curr_arp_slave = bond_first_slave_rcu(bond); 3093 if (!curr_arp_slave) 3094 return should_notify_rtnl; 3095 } 3096 3097 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER); 3098 3099 bond_for_each_slave_rcu(bond, slave, iter) { 3100 if (!found && !before && bond_slave_is_up(slave)) 3101 before = slave; 3102 3103 if (found && !new_slave && bond_slave_is_up(slave)) 3104 new_slave = slave; 3105 /* if the link state is up at this point, we 3106 * mark it down - this can happen if we have 3107 * simultaneous link failures and 3108 * reselect_active_interface doesn't make this 3109 * one the current slave so it is still marked 3110 * up when it is actually down 3111 */ 3112 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3113 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3114 BOND_SLAVE_NOTIFY_LATER); 3115 if (slave->link_failure_count < UINT_MAX) 3116 slave->link_failure_count++; 3117 3118 bond_set_slave_inactive_flags(slave, 3119 BOND_SLAVE_NOTIFY_LATER); 3120 3121 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3122 } 3123 if (slave == curr_arp_slave) 3124 found = true; 3125 } 3126 3127 if (!new_slave && before) 3128 new_slave = before; 3129 3130 if (!new_slave) 3131 goto check_state; 3132 3133 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3134 BOND_SLAVE_NOTIFY_LATER); 3135 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3136 bond_arp_send_all(bond, new_slave); 3137 new_slave->last_link_up = jiffies; 3138 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3139 3140 check_state: 3141 bond_for_each_slave_rcu(bond, slave, iter) { 3142 if (slave->should_notify || slave->should_notify_link) { 3143 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3144 break; 3145 } 3146 } 3147 return should_notify_rtnl; 3148 } 3149 3150 static void bond_activebackup_arp_mon(struct bonding *bond) 3151 { 3152 bool should_notify_peers = false; 3153 bool should_notify_rtnl = false; 3154 int delta_in_ticks; 3155 3156 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3157 3158 if (!bond_has_slaves(bond)) 3159 goto re_arm; 3160 3161 rcu_read_lock(); 3162 3163 should_notify_peers = bond_should_notify_peers(bond); 3164 3165 if (bond_ab_arp_inspect(bond)) { 3166 rcu_read_unlock(); 3167 3168 /* Race avoidance with bond_close flush of workqueue */ 3169 if (!rtnl_trylock()) { 3170 delta_in_ticks = 1; 3171 should_notify_peers = false; 3172 goto re_arm; 3173 } 3174 3175 bond_ab_arp_commit(bond); 3176 3177 rtnl_unlock(); 3178 rcu_read_lock(); 3179 } 3180 3181 should_notify_rtnl = bond_ab_arp_probe(bond); 3182 rcu_read_unlock(); 3183 3184 re_arm: 3185 if (bond->params.arp_interval) 3186 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3187 3188 if (should_notify_peers || should_notify_rtnl) { 3189 if (!rtnl_trylock()) 3190 return; 3191 3192 if (should_notify_peers) 3193 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3194 bond->dev); 3195 if (should_notify_rtnl) { 3196 bond_slave_state_notify(bond); 3197 bond_slave_link_notify(bond); 3198 } 3199 3200 rtnl_unlock(); 3201 } 3202 } 3203 3204 static void bond_arp_monitor(struct work_struct *work) 3205 { 3206 struct bonding *bond = container_of(work, struct bonding, 3207 arp_work.work); 3208 3209 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3210 bond_activebackup_arp_mon(bond); 3211 else 3212 bond_loadbalance_arp_mon(bond); 3213 } 3214 3215 /*-------------------------- netdev event handling --------------------------*/ 3216 3217 /* Change device name */ 3218 static int bond_event_changename(struct bonding *bond) 3219 { 3220 bond_remove_proc_entry(bond); 3221 bond_create_proc_entry(bond); 3222 3223 bond_debug_reregister(bond); 3224 3225 return NOTIFY_DONE; 3226 } 3227 3228 static int bond_master_netdev_event(unsigned long event, 3229 struct net_device *bond_dev) 3230 { 3231 struct bonding *event_bond = netdev_priv(bond_dev); 3232 3233 netdev_dbg(bond_dev, "%s called\n", __func__); 3234 3235 switch (event) { 3236 case NETDEV_CHANGENAME: 3237 return bond_event_changename(event_bond); 3238 case NETDEV_UNREGISTER: 3239 bond_remove_proc_entry(event_bond); 3240 break; 3241 case NETDEV_REGISTER: 3242 bond_create_proc_entry(event_bond); 3243 break; 3244 default: 3245 break; 3246 } 3247 3248 return NOTIFY_DONE; 3249 } 3250 3251 static int bond_slave_netdev_event(unsigned long event, 3252 struct net_device *slave_dev) 3253 { 3254 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3255 struct bonding *bond; 3256 struct net_device *bond_dev; 3257 3258 /* A netdev event can be generated while enslaving a device 3259 * before netdev_rx_handler_register is called in which case 3260 * slave will be NULL 3261 */ 3262 if (!slave) { 3263 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3264 return NOTIFY_DONE; 3265 } 3266 3267 bond_dev = slave->bond->dev; 3268 bond = slave->bond; 3269 primary = rtnl_dereference(bond->primary_slave); 3270 3271 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3272 3273 switch (event) { 3274 case NETDEV_UNREGISTER: 3275 if (bond_dev->type != ARPHRD_ETHER) 3276 bond_release_and_destroy(bond_dev, slave_dev); 3277 else 3278 __bond_release_one(bond_dev, slave_dev, false, true); 3279 break; 3280 case NETDEV_UP: 3281 case NETDEV_CHANGE: 3282 /* For 802.3ad mode only: 3283 * Getting invalid Speed/Duplex values here will put slave 3284 * in weird state. Mark it as link-fail if the link was 3285 * previously up or link-down if it hasn't yet come up, and 3286 * let link-monitoring (miimon) set it right when correct 3287 * speeds/duplex are available. 3288 */ 3289 if (bond_update_speed_duplex(slave) && 3290 BOND_MODE(bond) == BOND_MODE_8023AD) { 3291 if (slave->last_link_up) 3292 slave->link = BOND_LINK_FAIL; 3293 else 3294 slave->link = BOND_LINK_DOWN; 3295 } 3296 3297 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3298 bond_3ad_adapter_speed_duplex_changed(slave); 3299 /* Fallthrough */ 3300 case NETDEV_DOWN: 3301 /* Refresh slave-array if applicable! 3302 * If the setup does not use miimon or arpmon (mode-specific!), 3303 * then these events will not cause the slave-array to be 3304 * refreshed. This will cause xmit to use a slave that is not 3305 * usable. Avoid such situation by refeshing the array at these 3306 * events. If these (miimon/arpmon) parameters are configured 3307 * then array gets refreshed twice and that should be fine! 3308 */ 3309 if (bond_mode_can_use_xmit_hash(bond)) 3310 bond_update_slave_arr(bond, NULL); 3311 break; 3312 case NETDEV_CHANGEMTU: 3313 /* TODO: Should slaves be allowed to 3314 * independently alter their MTU? For 3315 * an active-backup bond, slaves need 3316 * not be the same type of device, so 3317 * MTUs may vary. For other modes, 3318 * slaves arguably should have the 3319 * same MTUs. To do this, we'd need to 3320 * take over the slave's change_mtu 3321 * function for the duration of their 3322 * servitude. 3323 */ 3324 break; 3325 case NETDEV_CHANGENAME: 3326 /* we don't care if we don't have primary set */ 3327 if (!bond_uses_primary(bond) || 3328 !bond->params.primary[0]) 3329 break; 3330 3331 if (slave == primary) { 3332 /* slave's name changed - he's no longer primary */ 3333 RCU_INIT_POINTER(bond->primary_slave, NULL); 3334 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3335 /* we have a new primary slave */ 3336 rcu_assign_pointer(bond->primary_slave, slave); 3337 } else { /* we didn't change primary - exit */ 3338 break; 3339 } 3340 3341 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3342 primary ? slave_dev->name : "none"); 3343 3344 block_netpoll_tx(); 3345 bond_select_active_slave(bond); 3346 unblock_netpoll_tx(); 3347 break; 3348 case NETDEV_FEAT_CHANGE: 3349 bond_compute_features(bond); 3350 break; 3351 case NETDEV_RESEND_IGMP: 3352 /* Propagate to master device */ 3353 call_netdevice_notifiers(event, slave->bond->dev); 3354 break; 3355 default: 3356 break; 3357 } 3358 3359 return NOTIFY_DONE; 3360 } 3361 3362 /* bond_netdev_event: handle netdev notifier chain events. 3363 * 3364 * This function receives events for the netdev chain. The caller (an 3365 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3366 * locks for us to safely manipulate the slave devices (RTNL lock, 3367 * dev_probe_lock). 3368 */ 3369 static int bond_netdev_event(struct notifier_block *this, 3370 unsigned long event, void *ptr) 3371 { 3372 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3373 3374 netdev_dbg(event_dev, "%s received %s\n", 3375 __func__, netdev_cmd_to_name(event)); 3376 3377 if (!(event_dev->priv_flags & IFF_BONDING)) 3378 return NOTIFY_DONE; 3379 3380 if (event_dev->flags & IFF_MASTER) { 3381 int ret; 3382 3383 ret = bond_master_netdev_event(event, event_dev); 3384 if (ret != NOTIFY_DONE) 3385 return ret; 3386 } 3387 3388 if (event_dev->flags & IFF_SLAVE) 3389 return bond_slave_netdev_event(event, event_dev); 3390 3391 return NOTIFY_DONE; 3392 } 3393 3394 static struct notifier_block bond_netdev_notifier = { 3395 .notifier_call = bond_netdev_event, 3396 }; 3397 3398 /*---------------------------- Hashing Policies -----------------------------*/ 3399 3400 /* L2 hash helper */ 3401 static inline u32 bond_eth_hash(struct sk_buff *skb) 3402 { 3403 struct ethhdr *ep, hdr_tmp; 3404 3405 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3406 if (ep) 3407 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3408 return 0; 3409 } 3410 3411 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, 3412 int *noff, int *proto, bool l34) 3413 { 3414 const struct ipv6hdr *iph6; 3415 const struct iphdr *iph; 3416 3417 if (skb->protocol == htons(ETH_P_IP)) { 3418 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph)))) 3419 return false; 3420 iph = (const struct iphdr *)(skb->data + *noff); 3421 iph_to_flow_copy_v4addrs(fk, iph); 3422 *noff += iph->ihl << 2; 3423 if (!ip_is_fragment(iph)) 3424 *proto = iph->protocol; 3425 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3426 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6)))) 3427 return false; 3428 iph6 = (const struct ipv6hdr *)(skb->data + *noff); 3429 iph_to_flow_copy_v6addrs(fk, iph6); 3430 *noff += sizeof(*iph6); 3431 *proto = iph6->nexthdr; 3432 } else { 3433 return false; 3434 } 3435 3436 if (l34 && *proto >= 0) 3437 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto); 3438 3439 return true; 3440 } 3441 3442 /* Extract the appropriate headers based on bond's xmit policy */ 3443 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3444 struct flow_keys *fk) 3445 { 3446 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 3447 int noff, proto = -1; 3448 3449 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) { 3450 memset(fk, 0, sizeof(*fk)); 3451 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 3452 fk, NULL, 0, 0, 0, 0); 3453 } 3454 3455 fk->ports.ports = 0; 3456 memset(&fk->icmp, 0, sizeof(fk->icmp)); 3457 noff = skb_network_offset(skb); 3458 if (!bond_flow_ip(skb, fk, &noff, &proto, l34)) 3459 return false; 3460 3461 /* ICMP error packets contains at least 8 bytes of the header 3462 * of the packet which generated the error. Use this information 3463 * to correlate ICMP error packets within the same flow which 3464 * generated the error. 3465 */ 3466 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) { 3467 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data, 3468 skb_transport_offset(skb), 3469 skb_headlen(skb)); 3470 if (proto == IPPROTO_ICMP) { 3471 if (!icmp_is_err(fk->icmp.type)) 3472 return true; 3473 3474 noff += sizeof(struct icmphdr); 3475 } else if (proto == IPPROTO_ICMPV6) { 3476 if (!icmpv6_is_err(fk->icmp.type)) 3477 return true; 3478 3479 noff += sizeof(struct icmp6hdr); 3480 } 3481 return bond_flow_ip(skb, fk, &noff, &proto, l34); 3482 } 3483 3484 return true; 3485 } 3486 3487 /** 3488 * bond_xmit_hash - generate a hash value based on the xmit policy 3489 * @bond: bonding device 3490 * @skb: buffer to use for headers 3491 * 3492 * This function will extract the necessary headers from the skb buffer and use 3493 * them to generate a hash based on the xmit_policy set in the bonding device 3494 */ 3495 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3496 { 3497 struct flow_keys flow; 3498 u32 hash; 3499 3500 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3501 skb->l4_hash) 3502 return skb->hash; 3503 3504 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3505 !bond_flow_dissect(bond, skb, &flow)) 3506 return bond_eth_hash(skb); 3507 3508 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3509 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 3510 hash = bond_eth_hash(skb); 3511 } else { 3512 if (flow.icmp.id) 3513 memcpy(&hash, &flow.icmp, sizeof(hash)); 3514 else 3515 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 3516 } 3517 hash ^= (__force u32)flow_get_u32_dst(&flow) ^ 3518 (__force u32)flow_get_u32_src(&flow); 3519 hash ^= (hash >> 16); 3520 hash ^= (hash >> 8); 3521 3522 return hash >> 1; 3523 } 3524 3525 /*-------------------------- Device entry points ----------------------------*/ 3526 3527 void bond_work_init_all(struct bonding *bond) 3528 { 3529 INIT_DELAYED_WORK(&bond->mcast_work, 3530 bond_resend_igmp_join_requests_delayed); 3531 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3532 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3533 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3534 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3535 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3536 } 3537 3538 static void bond_work_cancel_all(struct bonding *bond) 3539 { 3540 cancel_delayed_work_sync(&bond->mii_work); 3541 cancel_delayed_work_sync(&bond->arp_work); 3542 cancel_delayed_work_sync(&bond->alb_work); 3543 cancel_delayed_work_sync(&bond->ad_work); 3544 cancel_delayed_work_sync(&bond->mcast_work); 3545 cancel_delayed_work_sync(&bond->slave_arr_work); 3546 } 3547 3548 static int bond_open(struct net_device *bond_dev) 3549 { 3550 struct bonding *bond = netdev_priv(bond_dev); 3551 struct list_head *iter; 3552 struct slave *slave; 3553 3554 /* reset slave->backup and slave->inactive */ 3555 if (bond_has_slaves(bond)) { 3556 bond_for_each_slave(bond, slave, iter) { 3557 if (bond_uses_primary(bond) && 3558 slave != rcu_access_pointer(bond->curr_active_slave)) { 3559 bond_set_slave_inactive_flags(slave, 3560 BOND_SLAVE_NOTIFY_NOW); 3561 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3562 bond_set_slave_active_flags(slave, 3563 BOND_SLAVE_NOTIFY_NOW); 3564 } 3565 } 3566 } 3567 3568 if (bond_is_lb(bond)) { 3569 /* bond_alb_initialize must be called before the timer 3570 * is started. 3571 */ 3572 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3573 return -ENOMEM; 3574 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3575 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3576 } 3577 3578 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3579 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3580 3581 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3582 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3583 bond->recv_probe = bond_arp_rcv; 3584 } 3585 3586 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3587 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3588 /* register to receive LACPDUs */ 3589 bond->recv_probe = bond_3ad_lacpdu_recv; 3590 bond_3ad_initiate_agg_selection(bond, 1); 3591 } 3592 3593 if (bond_mode_can_use_xmit_hash(bond)) 3594 bond_update_slave_arr(bond, NULL); 3595 3596 return 0; 3597 } 3598 3599 static int bond_close(struct net_device *bond_dev) 3600 { 3601 struct bonding *bond = netdev_priv(bond_dev); 3602 3603 bond_work_cancel_all(bond); 3604 bond->send_peer_notif = 0; 3605 if (bond_is_lb(bond)) 3606 bond_alb_deinitialize(bond); 3607 bond->recv_probe = NULL; 3608 3609 return 0; 3610 } 3611 3612 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3613 * that some drivers can provide 32bit values only. 3614 */ 3615 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3616 const struct rtnl_link_stats64 *_new, 3617 const struct rtnl_link_stats64 *_old) 3618 { 3619 const u64 *new = (const u64 *)_new; 3620 const u64 *old = (const u64 *)_old; 3621 u64 *res = (u64 *)_res; 3622 int i; 3623 3624 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3625 u64 nv = new[i]; 3626 u64 ov = old[i]; 3627 s64 delta = nv - ov; 3628 3629 /* detects if this particular field is 32bit only */ 3630 if (((nv | ov) >> 32) == 0) 3631 delta = (s64)(s32)((u32)nv - (u32)ov); 3632 3633 /* filter anomalies, some drivers reset their stats 3634 * at down/up events. 3635 */ 3636 if (delta > 0) 3637 res[i] += delta; 3638 } 3639 } 3640 3641 #ifdef CONFIG_LOCKDEP 3642 static int bond_get_lowest_level_rcu(struct net_device *dev) 3643 { 3644 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 3645 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 3646 int cur = 0, max = 0; 3647 3648 now = dev; 3649 iter = &dev->adj_list.lower; 3650 3651 while (1) { 3652 next = NULL; 3653 while (1) { 3654 ldev = netdev_next_lower_dev_rcu(now, &iter); 3655 if (!ldev) 3656 break; 3657 3658 next = ldev; 3659 niter = &ldev->adj_list.lower; 3660 dev_stack[cur] = now; 3661 iter_stack[cur++] = iter; 3662 if (max <= cur) 3663 max = cur; 3664 break; 3665 } 3666 3667 if (!next) { 3668 if (!cur) 3669 return max; 3670 next = dev_stack[--cur]; 3671 niter = iter_stack[cur]; 3672 } 3673 3674 now = next; 3675 iter = niter; 3676 } 3677 3678 return max; 3679 } 3680 #endif 3681 3682 static void bond_get_stats(struct net_device *bond_dev, 3683 struct rtnl_link_stats64 *stats) 3684 { 3685 struct bonding *bond = netdev_priv(bond_dev); 3686 struct rtnl_link_stats64 temp; 3687 struct list_head *iter; 3688 struct slave *slave; 3689 int nest_level = 0; 3690 3691 3692 rcu_read_lock(); 3693 #ifdef CONFIG_LOCKDEP 3694 nest_level = bond_get_lowest_level_rcu(bond_dev); 3695 #endif 3696 3697 spin_lock_nested(&bond->stats_lock, nest_level); 3698 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3699 3700 bond_for_each_slave_rcu(bond, slave, iter) { 3701 const struct rtnl_link_stats64 *new = 3702 dev_get_stats(slave->dev, &temp); 3703 3704 bond_fold_stats(stats, new, &slave->slave_stats); 3705 3706 /* save off the slave stats for the next run */ 3707 memcpy(&slave->slave_stats, new, sizeof(*new)); 3708 } 3709 3710 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3711 spin_unlock(&bond->stats_lock); 3712 rcu_read_unlock(); 3713 } 3714 3715 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3716 { 3717 struct bonding *bond = netdev_priv(bond_dev); 3718 struct net_device *slave_dev = NULL; 3719 struct ifbond k_binfo; 3720 struct ifbond __user *u_binfo = NULL; 3721 struct ifslave k_sinfo; 3722 struct ifslave __user *u_sinfo = NULL; 3723 struct mii_ioctl_data *mii = NULL; 3724 struct bond_opt_value newval; 3725 struct net *net; 3726 int res = 0; 3727 3728 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3729 3730 switch (cmd) { 3731 case SIOCGMIIPHY: 3732 mii = if_mii(ifr); 3733 if (!mii) 3734 return -EINVAL; 3735 3736 mii->phy_id = 0; 3737 /* Fall Through */ 3738 case SIOCGMIIREG: 3739 /* We do this again just in case we were called by SIOCGMIIREG 3740 * instead of SIOCGMIIPHY. 3741 */ 3742 mii = if_mii(ifr); 3743 if (!mii) 3744 return -EINVAL; 3745 3746 if (mii->reg_num == 1) { 3747 mii->val_out = 0; 3748 if (netif_carrier_ok(bond->dev)) 3749 mii->val_out = BMSR_LSTATUS; 3750 } 3751 3752 return 0; 3753 case BOND_INFO_QUERY_OLD: 3754 case SIOCBONDINFOQUERY: 3755 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3756 3757 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3758 return -EFAULT; 3759 3760 bond_info_query(bond_dev, &k_binfo); 3761 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3762 return -EFAULT; 3763 3764 return 0; 3765 case BOND_SLAVE_INFO_QUERY_OLD: 3766 case SIOCBONDSLAVEINFOQUERY: 3767 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3768 3769 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3770 return -EFAULT; 3771 3772 res = bond_slave_info_query(bond_dev, &k_sinfo); 3773 if (res == 0 && 3774 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3775 return -EFAULT; 3776 3777 return res; 3778 default: 3779 break; 3780 } 3781 3782 net = dev_net(bond_dev); 3783 3784 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3785 return -EPERM; 3786 3787 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3788 3789 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 3790 3791 if (!slave_dev) 3792 return -ENODEV; 3793 3794 switch (cmd) { 3795 case BOND_ENSLAVE_OLD: 3796 case SIOCBONDENSLAVE: 3797 res = bond_enslave(bond_dev, slave_dev, NULL); 3798 break; 3799 case BOND_RELEASE_OLD: 3800 case SIOCBONDRELEASE: 3801 res = bond_release(bond_dev, slave_dev); 3802 break; 3803 case BOND_SETHWADDR_OLD: 3804 case SIOCBONDSETHWADDR: 3805 res = bond_set_dev_addr(bond_dev, slave_dev); 3806 break; 3807 case BOND_CHANGE_ACTIVE_OLD: 3808 case SIOCBONDCHANGEACTIVE: 3809 bond_opt_initstr(&newval, slave_dev->name); 3810 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3811 &newval); 3812 break; 3813 default: 3814 res = -EOPNOTSUPP; 3815 } 3816 3817 return res; 3818 } 3819 3820 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3821 { 3822 struct bonding *bond = netdev_priv(bond_dev); 3823 3824 if (change & IFF_PROMISC) 3825 bond_set_promiscuity(bond, 3826 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3827 3828 if (change & IFF_ALLMULTI) 3829 bond_set_allmulti(bond, 3830 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3831 } 3832 3833 static void bond_set_rx_mode(struct net_device *bond_dev) 3834 { 3835 struct bonding *bond = netdev_priv(bond_dev); 3836 struct list_head *iter; 3837 struct slave *slave; 3838 3839 rcu_read_lock(); 3840 if (bond_uses_primary(bond)) { 3841 slave = rcu_dereference(bond->curr_active_slave); 3842 if (slave) { 3843 dev_uc_sync(slave->dev, bond_dev); 3844 dev_mc_sync(slave->dev, bond_dev); 3845 } 3846 } else { 3847 bond_for_each_slave_rcu(bond, slave, iter) { 3848 dev_uc_sync_multiple(slave->dev, bond_dev); 3849 dev_mc_sync_multiple(slave->dev, bond_dev); 3850 } 3851 } 3852 rcu_read_unlock(); 3853 } 3854 3855 static int bond_neigh_init(struct neighbour *n) 3856 { 3857 struct bonding *bond = netdev_priv(n->dev); 3858 const struct net_device_ops *slave_ops; 3859 struct neigh_parms parms; 3860 struct slave *slave; 3861 int ret = 0; 3862 3863 rcu_read_lock(); 3864 slave = bond_first_slave_rcu(bond); 3865 if (!slave) 3866 goto out; 3867 slave_ops = slave->dev->netdev_ops; 3868 if (!slave_ops->ndo_neigh_setup) 3869 goto out; 3870 3871 /* TODO: find another way [1] to implement this. 3872 * Passing a zeroed structure is fragile, 3873 * but at least we do not pass garbage. 3874 * 3875 * [1] One way would be that ndo_neigh_setup() never touch 3876 * struct neigh_parms, but propagate the new neigh_setup() 3877 * back to ___neigh_create() / neigh_parms_alloc() 3878 */ 3879 memset(&parms, 0, sizeof(parms)); 3880 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 3881 3882 if (ret) 3883 goto out; 3884 3885 if (parms.neigh_setup) 3886 ret = parms.neigh_setup(n); 3887 out: 3888 rcu_read_unlock(); 3889 return ret; 3890 } 3891 3892 /* The bonding ndo_neigh_setup is called at init time beofre any 3893 * slave exists. So we must declare proxy setup function which will 3894 * be used at run time to resolve the actual slave neigh param setup. 3895 * 3896 * It's also called by master devices (such as vlans) to setup their 3897 * underlying devices. In that case - do nothing, we're already set up from 3898 * our init. 3899 */ 3900 static int bond_neigh_setup(struct net_device *dev, 3901 struct neigh_parms *parms) 3902 { 3903 /* modify only our neigh_parms */ 3904 if (parms->dev == dev) 3905 parms->neigh_setup = bond_neigh_init; 3906 3907 return 0; 3908 } 3909 3910 /* Change the MTU of all of a master's slaves to match the master */ 3911 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 3912 { 3913 struct bonding *bond = netdev_priv(bond_dev); 3914 struct slave *slave, *rollback_slave; 3915 struct list_head *iter; 3916 int res = 0; 3917 3918 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 3919 3920 bond_for_each_slave(bond, slave, iter) { 3921 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 3922 slave, slave->dev->netdev_ops->ndo_change_mtu); 3923 3924 res = dev_set_mtu(slave->dev, new_mtu); 3925 3926 if (res) { 3927 /* If we failed to set the slave's mtu to the new value 3928 * we must abort the operation even in ACTIVE_BACKUP 3929 * mode, because if we allow the backup slaves to have 3930 * different mtu values than the active slave we'll 3931 * need to change their mtu when doing a failover. That 3932 * means changing their mtu from timer context, which 3933 * is probably not a good idea. 3934 */ 3935 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 3936 res, new_mtu); 3937 goto unwind; 3938 } 3939 } 3940 3941 bond_dev->mtu = new_mtu; 3942 3943 return 0; 3944 3945 unwind: 3946 /* unwind from head to the slave that failed */ 3947 bond_for_each_slave(bond, rollback_slave, iter) { 3948 int tmp_res; 3949 3950 if (rollback_slave == slave) 3951 break; 3952 3953 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 3954 if (tmp_res) 3955 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 3956 tmp_res); 3957 } 3958 3959 return res; 3960 } 3961 3962 /* Change HW address 3963 * 3964 * Note that many devices must be down to change the HW address, and 3965 * downing the master releases all slaves. We can make bonds full of 3966 * bonding devices to test this, however. 3967 */ 3968 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 3969 { 3970 struct bonding *bond = netdev_priv(bond_dev); 3971 struct slave *slave, *rollback_slave; 3972 struct sockaddr_storage *ss = addr, tmp_ss; 3973 struct list_head *iter; 3974 int res = 0; 3975 3976 if (BOND_MODE(bond) == BOND_MODE_ALB) 3977 return bond_alb_set_mac_address(bond_dev, addr); 3978 3979 3980 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 3981 3982 /* If fail_over_mac is enabled, do nothing and return success. 3983 * Returning an error causes ifenslave to fail. 3984 */ 3985 if (bond->params.fail_over_mac && 3986 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3987 return 0; 3988 3989 if (!is_valid_ether_addr(ss->__data)) 3990 return -EADDRNOTAVAIL; 3991 3992 bond_for_each_slave(bond, slave, iter) { 3993 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 3994 __func__, slave); 3995 res = dev_set_mac_address(slave->dev, addr, NULL); 3996 if (res) { 3997 /* TODO: consider downing the slave 3998 * and retry ? 3999 * User should expect communications 4000 * breakage anyway until ARP finish 4001 * updating, so... 4002 */ 4003 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4004 __func__, res); 4005 goto unwind; 4006 } 4007 } 4008 4009 /* success */ 4010 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 4011 return 0; 4012 4013 unwind: 4014 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4015 tmp_ss.ss_family = bond_dev->type; 4016 4017 /* unwind from head to the slave that failed */ 4018 bond_for_each_slave(bond, rollback_slave, iter) { 4019 int tmp_res; 4020 4021 if (rollback_slave == slave) 4022 break; 4023 4024 tmp_res = dev_set_mac_address(rollback_slave->dev, 4025 (struct sockaddr *)&tmp_ss, NULL); 4026 if (tmp_res) { 4027 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4028 __func__, tmp_res); 4029 } 4030 } 4031 4032 return res; 4033 } 4034 4035 /** 4036 * bond_get_slave_by_id - get xmit slave with slave_id 4037 * @bond: bonding device that is transmitting 4038 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4039 * 4040 * This function tries to get slave with slave_id but in case 4041 * it fails, it tries to find the first available slave for transmission. 4042 */ 4043 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4044 int slave_id) 4045 { 4046 struct list_head *iter; 4047 struct slave *slave; 4048 int i = slave_id; 4049 4050 /* Here we start from the slave with slave_id */ 4051 bond_for_each_slave_rcu(bond, slave, iter) { 4052 if (--i < 0) { 4053 if (bond_slave_can_tx(slave)) 4054 return slave; 4055 } 4056 } 4057 4058 /* Here we start from the first slave up to slave_id */ 4059 i = slave_id; 4060 bond_for_each_slave_rcu(bond, slave, iter) { 4061 if (--i < 0) 4062 break; 4063 if (bond_slave_can_tx(slave)) 4064 return slave; 4065 } 4066 /* no slave that can tx has been found */ 4067 return NULL; 4068 } 4069 4070 /** 4071 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4072 * @bond: bonding device to use 4073 * 4074 * Based on the value of the bonding device's packets_per_slave parameter 4075 * this function generates a slave id, which is usually used as the next 4076 * slave to transmit through. 4077 */ 4078 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4079 { 4080 u32 slave_id; 4081 struct reciprocal_value reciprocal_packets_per_slave; 4082 int packets_per_slave = bond->params.packets_per_slave; 4083 4084 switch (packets_per_slave) { 4085 case 0: 4086 slave_id = prandom_u32(); 4087 break; 4088 case 1: 4089 slave_id = bond->rr_tx_counter; 4090 break; 4091 default: 4092 reciprocal_packets_per_slave = 4093 bond->params.reciprocal_packets_per_slave; 4094 slave_id = reciprocal_divide(bond->rr_tx_counter, 4095 reciprocal_packets_per_slave); 4096 break; 4097 } 4098 bond->rr_tx_counter++; 4099 4100 return slave_id; 4101 } 4102 4103 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4104 struct sk_buff *skb) 4105 { 4106 struct slave *slave; 4107 int slave_cnt; 4108 u32 slave_id; 4109 4110 /* Start with the curr_active_slave that joined the bond as the 4111 * default for sending IGMP traffic. For failover purposes one 4112 * needs to maintain some consistency for the interface that will 4113 * send the join/membership reports. The curr_active_slave found 4114 * will send all of this type of traffic. 4115 */ 4116 if (skb->protocol == htons(ETH_P_IP)) { 4117 int noff = skb_network_offset(skb); 4118 struct iphdr *iph; 4119 4120 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4121 goto non_igmp; 4122 4123 iph = ip_hdr(skb); 4124 if (iph->protocol == IPPROTO_IGMP) { 4125 slave = rcu_dereference(bond->curr_active_slave); 4126 if (slave) 4127 return slave; 4128 return bond_get_slave_by_id(bond, 0); 4129 } 4130 } 4131 4132 non_igmp: 4133 slave_cnt = READ_ONCE(bond->slave_cnt); 4134 if (likely(slave_cnt)) { 4135 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4136 return bond_get_slave_by_id(bond, slave_id); 4137 } 4138 return NULL; 4139 } 4140 4141 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4142 struct net_device *bond_dev) 4143 { 4144 struct bonding *bond = netdev_priv(bond_dev); 4145 struct slave *slave; 4146 4147 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4148 if (likely(slave)) 4149 return bond_dev_queue_xmit(bond, skb, slave->dev); 4150 4151 return bond_tx_drop(bond_dev, skb); 4152 } 4153 4154 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond, 4155 struct sk_buff *skb) 4156 { 4157 return rcu_dereference(bond->curr_active_slave); 4158 } 4159 4160 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4161 * the bond has a usable interface. 4162 */ 4163 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4164 struct net_device *bond_dev) 4165 { 4166 struct bonding *bond = netdev_priv(bond_dev); 4167 struct slave *slave; 4168 4169 slave = bond_xmit_activebackup_slave_get(bond, skb); 4170 if (slave) 4171 return bond_dev_queue_xmit(bond, skb, slave->dev); 4172 4173 return bond_tx_drop(bond_dev, skb); 4174 } 4175 4176 /* Use this to update slave_array when (a) it's not appropriate to update 4177 * slave_array right away (note that update_slave_array() may sleep) 4178 * and / or (b) RTNL is not held. 4179 */ 4180 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4181 { 4182 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4183 } 4184 4185 /* Slave array work handler. Holds only RTNL */ 4186 static void bond_slave_arr_handler(struct work_struct *work) 4187 { 4188 struct bonding *bond = container_of(work, struct bonding, 4189 slave_arr_work.work); 4190 int ret; 4191 4192 if (!rtnl_trylock()) 4193 goto err; 4194 4195 ret = bond_update_slave_arr(bond, NULL); 4196 rtnl_unlock(); 4197 if (ret) { 4198 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4199 goto err; 4200 } 4201 return; 4202 4203 err: 4204 bond_slave_arr_work_rearm(bond, 1); 4205 } 4206 4207 static void bond_skip_slave(struct bond_up_slave *slaves, 4208 struct slave *skipslave) 4209 { 4210 int idx; 4211 4212 /* Rare situation where caller has asked to skip a specific 4213 * slave but allocation failed (most likely!). BTW this is 4214 * only possible when the call is initiated from 4215 * __bond_release_one(). In this situation; overwrite the 4216 * skipslave entry in the array with the last entry from the 4217 * array to avoid a situation where the xmit path may choose 4218 * this to-be-skipped slave to send a packet out. 4219 */ 4220 for (idx = 0; slaves && idx < slaves->count; idx++) { 4221 if (skipslave == slaves->arr[idx]) { 4222 slaves->arr[idx] = 4223 slaves->arr[slaves->count - 1]; 4224 slaves->count--; 4225 break; 4226 } 4227 } 4228 } 4229 4230 static void bond_set_slave_arr(struct bonding *bond, 4231 struct bond_up_slave *usable_slaves, 4232 struct bond_up_slave *all_slaves) 4233 { 4234 struct bond_up_slave *usable, *all; 4235 4236 usable = rtnl_dereference(bond->usable_slaves); 4237 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 4238 kfree_rcu(usable, rcu); 4239 4240 all = rtnl_dereference(bond->all_slaves); 4241 rcu_assign_pointer(bond->all_slaves, all_slaves); 4242 kfree_rcu(all, rcu); 4243 } 4244 4245 static void bond_reset_slave_arr(struct bonding *bond) 4246 { 4247 struct bond_up_slave *usable, *all; 4248 4249 usable = rtnl_dereference(bond->usable_slaves); 4250 if (usable) { 4251 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4252 kfree_rcu(usable, rcu); 4253 } 4254 4255 all = rtnl_dereference(bond->all_slaves); 4256 if (all) { 4257 RCU_INIT_POINTER(bond->all_slaves, NULL); 4258 kfree_rcu(all, rcu); 4259 } 4260 } 4261 4262 /* Build the usable slaves array in control path for modes that use xmit-hash 4263 * to determine the slave interface - 4264 * (a) BOND_MODE_8023AD 4265 * (b) BOND_MODE_XOR 4266 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 4267 * 4268 * The caller is expected to hold RTNL only and NO other lock! 4269 */ 4270 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 4271 { 4272 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 4273 struct slave *slave; 4274 struct list_head *iter; 4275 int agg_id = 0; 4276 int ret = 0; 4277 4278 #ifdef CONFIG_LOCKDEP 4279 WARN_ON(lockdep_is_held(&bond->mode_lock)); 4280 #endif 4281 4282 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 4283 bond->slave_cnt), GFP_KERNEL); 4284 all_slaves = kzalloc(struct_size(all_slaves, arr, 4285 bond->slave_cnt), GFP_KERNEL); 4286 if (!usable_slaves || !all_slaves) { 4287 ret = -ENOMEM; 4288 goto out; 4289 } 4290 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4291 struct ad_info ad_info; 4292 4293 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 4294 pr_debug("bond_3ad_get_active_agg_info failed\n"); 4295 /* No active aggragator means it's not safe to use 4296 * the previous array. 4297 */ 4298 bond_reset_slave_arr(bond); 4299 goto out; 4300 } 4301 agg_id = ad_info.aggregator_id; 4302 } 4303 bond_for_each_slave(bond, slave, iter) { 4304 if (skipslave == slave) 4305 continue; 4306 4307 all_slaves->arr[all_slaves->count++] = slave; 4308 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4309 struct aggregator *agg; 4310 4311 agg = SLAVE_AD_INFO(slave)->port.aggregator; 4312 if (!agg || agg->aggregator_identifier != agg_id) 4313 continue; 4314 } 4315 if (!bond_slave_can_tx(slave)) 4316 continue; 4317 4318 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 4319 usable_slaves->count); 4320 4321 usable_slaves->arr[usable_slaves->count++] = slave; 4322 } 4323 4324 bond_set_slave_arr(bond, usable_slaves, all_slaves); 4325 return ret; 4326 out: 4327 if (ret != 0 && skipslave) { 4328 bond_skip_slave(rtnl_dereference(bond->all_slaves), 4329 skipslave); 4330 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 4331 skipslave); 4332 } 4333 kfree_rcu(all_slaves, rcu); 4334 kfree_rcu(usable_slaves, rcu); 4335 4336 return ret; 4337 } 4338 4339 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 4340 struct sk_buff *skb, 4341 struct bond_up_slave *slaves) 4342 { 4343 struct slave *slave; 4344 unsigned int count; 4345 u32 hash; 4346 4347 hash = bond_xmit_hash(bond, skb); 4348 count = slaves ? READ_ONCE(slaves->count) : 0; 4349 if (unlikely(!count)) 4350 return NULL; 4351 4352 slave = slaves->arr[hash % count]; 4353 return slave; 4354 } 4355 4356 /* Use this Xmit function for 3AD as well as XOR modes. The current 4357 * usable slave array is formed in the control path. The xmit function 4358 * just calculates hash and sends the packet out. 4359 */ 4360 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4361 struct net_device *dev) 4362 { 4363 struct bonding *bond = netdev_priv(dev); 4364 struct bond_up_slave *slaves; 4365 struct slave *slave; 4366 4367 slaves = rcu_dereference(bond->usable_slaves); 4368 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4369 if (likely(slave)) 4370 return bond_dev_queue_xmit(bond, skb, slave->dev); 4371 4372 return bond_tx_drop(dev, skb); 4373 } 4374 4375 /* in broadcast mode, we send everything to all usable interfaces. */ 4376 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4377 struct net_device *bond_dev) 4378 { 4379 struct bonding *bond = netdev_priv(bond_dev); 4380 struct slave *slave = NULL; 4381 struct list_head *iter; 4382 4383 bond_for_each_slave_rcu(bond, slave, iter) { 4384 if (bond_is_last_slave(bond, slave)) 4385 break; 4386 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4387 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4388 4389 if (!skb2) { 4390 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4391 bond_dev->name, __func__); 4392 continue; 4393 } 4394 bond_dev_queue_xmit(bond, skb2, slave->dev); 4395 } 4396 } 4397 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4398 return bond_dev_queue_xmit(bond, skb, slave->dev); 4399 4400 return bond_tx_drop(bond_dev, skb); 4401 } 4402 4403 /*------------------------- Device initialization ---------------------------*/ 4404 4405 /* Lookup the slave that corresponds to a qid */ 4406 static inline int bond_slave_override(struct bonding *bond, 4407 struct sk_buff *skb) 4408 { 4409 struct slave *slave = NULL; 4410 struct list_head *iter; 4411 4412 if (!skb_rx_queue_recorded(skb)) 4413 return 1; 4414 4415 /* Find out if any slaves have the same mapping as this skb. */ 4416 bond_for_each_slave_rcu(bond, slave, iter) { 4417 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4418 if (bond_slave_is_up(slave) && 4419 slave->link == BOND_LINK_UP) { 4420 bond_dev_queue_xmit(bond, skb, slave->dev); 4421 return 0; 4422 } 4423 /* If the slave isn't UP, use default transmit policy. */ 4424 break; 4425 } 4426 } 4427 4428 return 1; 4429 } 4430 4431 4432 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4433 struct net_device *sb_dev) 4434 { 4435 /* This helper function exists to help dev_pick_tx get the correct 4436 * destination queue. Using a helper function skips a call to 4437 * skb_tx_hash and will put the skbs in the queue we expect on their 4438 * way down to the bonding driver. 4439 */ 4440 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4441 4442 /* Save the original txq to restore before passing to the driver */ 4443 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4444 4445 if (unlikely(txq >= dev->real_num_tx_queues)) { 4446 do { 4447 txq -= dev->real_num_tx_queues; 4448 } while (txq >= dev->real_num_tx_queues); 4449 } 4450 return txq; 4451 } 4452 4453 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 4454 struct sk_buff *skb, 4455 bool all_slaves) 4456 { 4457 struct bonding *bond = netdev_priv(master_dev); 4458 struct bond_up_slave *slaves; 4459 struct slave *slave = NULL; 4460 4461 switch (BOND_MODE(bond)) { 4462 case BOND_MODE_ROUNDROBIN: 4463 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4464 break; 4465 case BOND_MODE_ACTIVEBACKUP: 4466 slave = bond_xmit_activebackup_slave_get(bond, skb); 4467 break; 4468 case BOND_MODE_8023AD: 4469 case BOND_MODE_XOR: 4470 if (all_slaves) 4471 slaves = rcu_dereference(bond->all_slaves); 4472 else 4473 slaves = rcu_dereference(bond->usable_slaves); 4474 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4475 break; 4476 case BOND_MODE_BROADCAST: 4477 break; 4478 case BOND_MODE_ALB: 4479 slave = bond_xmit_alb_slave_get(bond, skb); 4480 break; 4481 case BOND_MODE_TLB: 4482 slave = bond_xmit_tlb_slave_get(bond, skb); 4483 break; 4484 default: 4485 /* Should never happen, mode already checked */ 4486 WARN_ONCE(true, "Unknown bonding mode"); 4487 break; 4488 } 4489 4490 if (slave) 4491 return slave->dev; 4492 return NULL; 4493 } 4494 4495 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4496 { 4497 struct bonding *bond = netdev_priv(dev); 4498 4499 if (bond_should_override_tx_queue(bond) && 4500 !bond_slave_override(bond, skb)) 4501 return NETDEV_TX_OK; 4502 4503 switch (BOND_MODE(bond)) { 4504 case BOND_MODE_ROUNDROBIN: 4505 return bond_xmit_roundrobin(skb, dev); 4506 case BOND_MODE_ACTIVEBACKUP: 4507 return bond_xmit_activebackup(skb, dev); 4508 case BOND_MODE_8023AD: 4509 case BOND_MODE_XOR: 4510 return bond_3ad_xor_xmit(skb, dev); 4511 case BOND_MODE_BROADCAST: 4512 return bond_xmit_broadcast(skb, dev); 4513 case BOND_MODE_ALB: 4514 return bond_alb_xmit(skb, dev); 4515 case BOND_MODE_TLB: 4516 return bond_tlb_xmit(skb, dev); 4517 default: 4518 /* Should never happen, mode already checked */ 4519 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4520 WARN_ON_ONCE(1); 4521 return bond_tx_drop(dev, skb); 4522 } 4523 } 4524 4525 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4526 { 4527 struct bonding *bond = netdev_priv(dev); 4528 netdev_tx_t ret = NETDEV_TX_OK; 4529 4530 /* If we risk deadlock from transmitting this in the 4531 * netpoll path, tell netpoll to queue the frame for later tx 4532 */ 4533 if (unlikely(is_netpoll_tx_blocked(dev))) 4534 return NETDEV_TX_BUSY; 4535 4536 rcu_read_lock(); 4537 if (bond_has_slaves(bond)) 4538 ret = __bond_start_xmit(skb, dev); 4539 else 4540 ret = bond_tx_drop(dev, skb); 4541 rcu_read_unlock(); 4542 4543 return ret; 4544 } 4545 4546 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4547 struct ethtool_link_ksettings *cmd) 4548 { 4549 struct bonding *bond = netdev_priv(bond_dev); 4550 unsigned long speed = 0; 4551 struct list_head *iter; 4552 struct slave *slave; 4553 4554 cmd->base.duplex = DUPLEX_UNKNOWN; 4555 cmd->base.port = PORT_OTHER; 4556 4557 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4558 * do not need to check mode. Though link speed might not represent 4559 * the true receive or transmit bandwidth (not all modes are symmetric) 4560 * this is an accurate maximum. 4561 */ 4562 bond_for_each_slave(bond, slave, iter) { 4563 if (bond_slave_can_tx(slave)) { 4564 if (slave->speed != SPEED_UNKNOWN) 4565 speed += slave->speed; 4566 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4567 slave->duplex != DUPLEX_UNKNOWN) 4568 cmd->base.duplex = slave->duplex; 4569 } 4570 } 4571 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4572 4573 return 0; 4574 } 4575 4576 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4577 struct ethtool_drvinfo *drvinfo) 4578 { 4579 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4580 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4581 BOND_ABI_VERSION); 4582 } 4583 4584 static const struct ethtool_ops bond_ethtool_ops = { 4585 .get_drvinfo = bond_ethtool_get_drvinfo, 4586 .get_link = ethtool_op_get_link, 4587 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4588 }; 4589 4590 static const struct net_device_ops bond_netdev_ops = { 4591 .ndo_init = bond_init, 4592 .ndo_uninit = bond_uninit, 4593 .ndo_open = bond_open, 4594 .ndo_stop = bond_close, 4595 .ndo_start_xmit = bond_start_xmit, 4596 .ndo_select_queue = bond_select_queue, 4597 .ndo_get_stats64 = bond_get_stats, 4598 .ndo_do_ioctl = bond_do_ioctl, 4599 .ndo_change_rx_flags = bond_change_rx_flags, 4600 .ndo_set_rx_mode = bond_set_rx_mode, 4601 .ndo_change_mtu = bond_change_mtu, 4602 .ndo_set_mac_address = bond_set_mac_address, 4603 .ndo_neigh_setup = bond_neigh_setup, 4604 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4605 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4606 #ifdef CONFIG_NET_POLL_CONTROLLER 4607 .ndo_netpoll_setup = bond_netpoll_setup, 4608 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4609 .ndo_poll_controller = bond_poll_controller, 4610 #endif 4611 .ndo_add_slave = bond_enslave, 4612 .ndo_del_slave = bond_release, 4613 .ndo_fix_features = bond_fix_features, 4614 .ndo_features_check = passthru_features_check, 4615 .ndo_get_xmit_slave = bond_xmit_get_slave, 4616 }; 4617 4618 static const struct device_type bond_type = { 4619 .name = "bond", 4620 }; 4621 4622 static void bond_destructor(struct net_device *bond_dev) 4623 { 4624 struct bonding *bond = netdev_priv(bond_dev); 4625 if (bond->wq) 4626 destroy_workqueue(bond->wq); 4627 } 4628 4629 void bond_setup(struct net_device *bond_dev) 4630 { 4631 struct bonding *bond = netdev_priv(bond_dev); 4632 4633 spin_lock_init(&bond->mode_lock); 4634 bond->params = bonding_defaults; 4635 4636 /* Initialize pointers */ 4637 bond->dev = bond_dev; 4638 4639 /* Initialize the device entry points */ 4640 ether_setup(bond_dev); 4641 bond_dev->max_mtu = ETH_MAX_MTU; 4642 bond_dev->netdev_ops = &bond_netdev_ops; 4643 bond_dev->ethtool_ops = &bond_ethtool_ops; 4644 4645 bond_dev->needs_free_netdev = true; 4646 bond_dev->priv_destructor = bond_destructor; 4647 4648 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4649 4650 /* Initialize the device options */ 4651 bond_dev->flags |= IFF_MASTER; 4652 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4653 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4654 4655 #ifdef CONFIG_XFRM_OFFLOAD 4656 /* set up xfrm device ops (only supported in active-backup right now) */ 4657 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4658 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 4659 bond->xs = NULL; 4660 #endif /* CONFIG_XFRM_OFFLOAD */ 4661 4662 /* don't acquire bond device's netif_tx_lock when transmitting */ 4663 bond_dev->features |= NETIF_F_LLTX; 4664 4665 /* By default, we declare the bond to be fully 4666 * VLAN hardware accelerated capable. Special 4667 * care is taken in the various xmit functions 4668 * when there are slaves that are not hw accel 4669 * capable 4670 */ 4671 4672 /* Don't allow bond devices to change network namespaces. */ 4673 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4674 4675 bond_dev->hw_features = BOND_VLAN_FEATURES | 4676 NETIF_F_HW_VLAN_CTAG_RX | 4677 NETIF_F_HW_VLAN_CTAG_FILTER; 4678 4679 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4; 4680 #ifdef CONFIG_XFRM_OFFLOAD 4681 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4682 bond_dev->hw_features |= BOND_XFRM_FEATURES; 4683 #endif /* CONFIG_XFRM_OFFLOAD */ 4684 bond_dev->features |= bond_dev->hw_features; 4685 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 4686 } 4687 4688 /* Destroy a bonding device. 4689 * Must be under rtnl_lock when this function is called. 4690 */ 4691 static void bond_uninit(struct net_device *bond_dev) 4692 { 4693 struct bonding *bond = netdev_priv(bond_dev); 4694 struct bond_up_slave *usable, *all; 4695 struct list_head *iter; 4696 struct slave *slave; 4697 4698 bond_netpoll_cleanup(bond_dev); 4699 4700 /* Release the bonded slaves */ 4701 bond_for_each_slave(bond, slave, iter) 4702 __bond_release_one(bond_dev, slave->dev, true, true); 4703 netdev_info(bond_dev, "Released all slaves\n"); 4704 4705 usable = rtnl_dereference(bond->usable_slaves); 4706 if (usable) { 4707 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4708 kfree_rcu(usable, rcu); 4709 } 4710 4711 all = rtnl_dereference(bond->all_slaves); 4712 if (all) { 4713 RCU_INIT_POINTER(bond->all_slaves, NULL); 4714 kfree_rcu(all, rcu); 4715 } 4716 4717 list_del(&bond->bond_list); 4718 4719 bond_debug_unregister(bond); 4720 } 4721 4722 /*------------------------- Module initialization ---------------------------*/ 4723 4724 static int bond_check_params(struct bond_params *params) 4725 { 4726 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4727 struct bond_opt_value newval; 4728 const struct bond_opt_value *valptr; 4729 int arp_all_targets_value = 0; 4730 u16 ad_actor_sys_prio = 0; 4731 u16 ad_user_port_key = 0; 4732 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4733 int arp_ip_count; 4734 int bond_mode = BOND_MODE_ROUNDROBIN; 4735 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4736 int lacp_fast = 0; 4737 int tlb_dynamic_lb; 4738 4739 /* Convert string parameters. */ 4740 if (mode) { 4741 bond_opt_initstr(&newval, mode); 4742 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4743 if (!valptr) { 4744 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4745 return -EINVAL; 4746 } 4747 bond_mode = valptr->value; 4748 } 4749 4750 if (xmit_hash_policy) { 4751 if (bond_mode == BOND_MODE_ROUNDROBIN || 4752 bond_mode == BOND_MODE_ACTIVEBACKUP || 4753 bond_mode == BOND_MODE_BROADCAST) { 4754 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4755 bond_mode_name(bond_mode)); 4756 } else { 4757 bond_opt_initstr(&newval, xmit_hash_policy); 4758 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 4759 &newval); 4760 if (!valptr) { 4761 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 4762 xmit_hash_policy); 4763 return -EINVAL; 4764 } 4765 xmit_hashtype = valptr->value; 4766 } 4767 } 4768 4769 if (lacp_rate) { 4770 if (bond_mode != BOND_MODE_8023AD) { 4771 pr_info("lacp_rate param is irrelevant in mode %s\n", 4772 bond_mode_name(bond_mode)); 4773 } else { 4774 bond_opt_initstr(&newval, lacp_rate); 4775 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 4776 &newval); 4777 if (!valptr) { 4778 pr_err("Error: Invalid lacp rate \"%s\"\n", 4779 lacp_rate); 4780 return -EINVAL; 4781 } 4782 lacp_fast = valptr->value; 4783 } 4784 } 4785 4786 if (ad_select) { 4787 bond_opt_initstr(&newval, ad_select); 4788 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 4789 &newval); 4790 if (!valptr) { 4791 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 4792 return -EINVAL; 4793 } 4794 params->ad_select = valptr->value; 4795 if (bond_mode != BOND_MODE_8023AD) 4796 pr_warn("ad_select param only affects 802.3ad mode\n"); 4797 } else { 4798 params->ad_select = BOND_AD_STABLE; 4799 } 4800 4801 if (max_bonds < 0) { 4802 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 4803 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 4804 max_bonds = BOND_DEFAULT_MAX_BONDS; 4805 } 4806 4807 if (miimon < 0) { 4808 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4809 miimon, INT_MAX); 4810 miimon = 0; 4811 } 4812 4813 if (updelay < 0) { 4814 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4815 updelay, INT_MAX); 4816 updelay = 0; 4817 } 4818 4819 if (downdelay < 0) { 4820 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4821 downdelay, INT_MAX); 4822 downdelay = 0; 4823 } 4824 4825 if ((use_carrier != 0) && (use_carrier != 1)) { 4826 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 4827 use_carrier); 4828 use_carrier = 1; 4829 } 4830 4831 if (num_peer_notif < 0 || num_peer_notif > 255) { 4832 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 4833 num_peer_notif); 4834 num_peer_notif = 1; 4835 } 4836 4837 /* reset values for 802.3ad/TLB/ALB */ 4838 if (!bond_mode_uses_arp(bond_mode)) { 4839 if (!miimon) { 4840 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"); 4841 pr_warn("Forcing miimon to 100msec\n"); 4842 miimon = BOND_DEFAULT_MIIMON; 4843 } 4844 } 4845 4846 if (tx_queues < 1 || tx_queues > 255) { 4847 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 4848 tx_queues, BOND_DEFAULT_TX_QUEUES); 4849 tx_queues = BOND_DEFAULT_TX_QUEUES; 4850 } 4851 4852 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 4853 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 4854 all_slaves_active); 4855 all_slaves_active = 0; 4856 } 4857 4858 if (resend_igmp < 0 || resend_igmp > 255) { 4859 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 4860 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 4861 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 4862 } 4863 4864 bond_opt_initval(&newval, packets_per_slave); 4865 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 4866 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 4867 packets_per_slave, USHRT_MAX); 4868 packets_per_slave = 1; 4869 } 4870 4871 if (bond_mode == BOND_MODE_ALB) { 4872 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", 4873 updelay); 4874 } 4875 4876 if (!miimon) { 4877 if (updelay || downdelay) { 4878 /* just warn the user the up/down delay will have 4879 * no effect since miimon is zero... 4880 */ 4881 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", 4882 updelay, downdelay); 4883 } 4884 } else { 4885 /* don't allow arp monitoring */ 4886 if (arp_interval) { 4887 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 4888 miimon, arp_interval); 4889 arp_interval = 0; 4890 } 4891 4892 if ((updelay % miimon) != 0) { 4893 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 4894 updelay, miimon, (updelay / miimon) * miimon); 4895 } 4896 4897 updelay /= miimon; 4898 4899 if ((downdelay % miimon) != 0) { 4900 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 4901 downdelay, miimon, 4902 (downdelay / miimon) * miimon); 4903 } 4904 4905 downdelay /= miimon; 4906 } 4907 4908 if (arp_interval < 0) { 4909 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 4910 arp_interval, INT_MAX); 4911 arp_interval = 0; 4912 } 4913 4914 for (arp_ip_count = 0, i = 0; 4915 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 4916 __be32 ip; 4917 4918 /* not a complete check, but good enough to catch mistakes */ 4919 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 4920 !bond_is_ip_target_ok(ip)) { 4921 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 4922 arp_ip_target[i]); 4923 arp_interval = 0; 4924 } else { 4925 if (bond_get_targets_ip(arp_target, ip) == -1) 4926 arp_target[arp_ip_count++] = ip; 4927 else 4928 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 4929 &ip); 4930 } 4931 } 4932 4933 if (arp_interval && !arp_ip_count) { 4934 /* don't allow arping if no arp_ip_target given... */ 4935 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 4936 arp_interval); 4937 arp_interval = 0; 4938 } 4939 4940 if (arp_validate) { 4941 if (!arp_interval) { 4942 pr_err("arp_validate requires arp_interval\n"); 4943 return -EINVAL; 4944 } 4945 4946 bond_opt_initstr(&newval, arp_validate); 4947 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 4948 &newval); 4949 if (!valptr) { 4950 pr_err("Error: invalid arp_validate \"%s\"\n", 4951 arp_validate); 4952 return -EINVAL; 4953 } 4954 arp_validate_value = valptr->value; 4955 } else { 4956 arp_validate_value = 0; 4957 } 4958 4959 if (arp_all_targets) { 4960 bond_opt_initstr(&newval, arp_all_targets); 4961 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 4962 &newval); 4963 if (!valptr) { 4964 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 4965 arp_all_targets); 4966 arp_all_targets_value = 0; 4967 } else { 4968 arp_all_targets_value = valptr->value; 4969 } 4970 } 4971 4972 if (miimon) { 4973 pr_info("MII link monitoring set to %d ms\n", miimon); 4974 } else if (arp_interval) { 4975 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 4976 arp_validate_value); 4977 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 4978 arp_interval, valptr->string, arp_ip_count); 4979 4980 for (i = 0; i < arp_ip_count; i++) 4981 pr_cont(" %s", arp_ip_target[i]); 4982 4983 pr_cont("\n"); 4984 4985 } else if (max_bonds) { 4986 /* miimon and arp_interval not set, we need one so things 4987 * work as expected, see bonding.txt for details 4988 */ 4989 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"); 4990 } 4991 4992 if (primary && !bond_mode_uses_primary(bond_mode)) { 4993 /* currently, using a primary only makes sense 4994 * in active backup, TLB or ALB modes 4995 */ 4996 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 4997 primary, bond_mode_name(bond_mode)); 4998 primary = NULL; 4999 } 5000 5001 if (primary && primary_reselect) { 5002 bond_opt_initstr(&newval, primary_reselect); 5003 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 5004 &newval); 5005 if (!valptr) { 5006 pr_err("Error: Invalid primary_reselect \"%s\"\n", 5007 primary_reselect); 5008 return -EINVAL; 5009 } 5010 primary_reselect_value = valptr->value; 5011 } else { 5012 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 5013 } 5014 5015 if (fail_over_mac) { 5016 bond_opt_initstr(&newval, fail_over_mac); 5017 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 5018 &newval); 5019 if (!valptr) { 5020 pr_err("Error: invalid fail_over_mac \"%s\"\n", 5021 fail_over_mac); 5022 return -EINVAL; 5023 } 5024 fail_over_mac_value = valptr->value; 5025 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 5026 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 5027 } else { 5028 fail_over_mac_value = BOND_FOM_NONE; 5029 } 5030 5031 bond_opt_initstr(&newval, "default"); 5032 valptr = bond_opt_parse( 5033 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 5034 &newval); 5035 if (!valptr) { 5036 pr_err("Error: No ad_actor_sys_prio default value"); 5037 return -EINVAL; 5038 } 5039 ad_actor_sys_prio = valptr->value; 5040 5041 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 5042 &newval); 5043 if (!valptr) { 5044 pr_err("Error: No ad_user_port_key default value"); 5045 return -EINVAL; 5046 } 5047 ad_user_port_key = valptr->value; 5048 5049 bond_opt_initstr(&newval, "default"); 5050 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 5051 if (!valptr) { 5052 pr_err("Error: No tlb_dynamic_lb default value"); 5053 return -EINVAL; 5054 } 5055 tlb_dynamic_lb = valptr->value; 5056 5057 if (lp_interval == 0) { 5058 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 5059 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 5060 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 5061 } 5062 5063 /* fill params struct with the proper values */ 5064 params->mode = bond_mode; 5065 params->xmit_policy = xmit_hashtype; 5066 params->miimon = miimon; 5067 params->num_peer_notif = num_peer_notif; 5068 params->arp_interval = arp_interval; 5069 params->arp_validate = arp_validate_value; 5070 params->arp_all_targets = arp_all_targets_value; 5071 params->updelay = updelay; 5072 params->downdelay = downdelay; 5073 params->peer_notif_delay = 0; 5074 params->use_carrier = use_carrier; 5075 params->lacp_fast = lacp_fast; 5076 params->primary[0] = 0; 5077 params->primary_reselect = primary_reselect_value; 5078 params->fail_over_mac = fail_over_mac_value; 5079 params->tx_queues = tx_queues; 5080 params->all_slaves_active = all_slaves_active; 5081 params->resend_igmp = resend_igmp; 5082 params->min_links = min_links; 5083 params->lp_interval = lp_interval; 5084 params->packets_per_slave = packets_per_slave; 5085 params->tlb_dynamic_lb = tlb_dynamic_lb; 5086 params->ad_actor_sys_prio = ad_actor_sys_prio; 5087 eth_zero_addr(params->ad_actor_system); 5088 params->ad_user_port_key = ad_user_port_key; 5089 if (packets_per_slave > 0) { 5090 params->reciprocal_packets_per_slave = 5091 reciprocal_value(packets_per_slave); 5092 } else { 5093 /* reciprocal_packets_per_slave is unused if 5094 * packets_per_slave is 0 or 1, just initialize it 5095 */ 5096 params->reciprocal_packets_per_slave = 5097 (struct reciprocal_value) { 0 }; 5098 } 5099 5100 if (primary) { 5101 strncpy(params->primary, primary, IFNAMSIZ); 5102 params->primary[IFNAMSIZ - 1] = 0; 5103 } 5104 5105 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 5106 5107 return 0; 5108 } 5109 5110 /* Called from registration process */ 5111 static int bond_init(struct net_device *bond_dev) 5112 { 5113 struct bonding *bond = netdev_priv(bond_dev); 5114 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 5115 5116 netdev_dbg(bond_dev, "Begin bond_init\n"); 5117 5118 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 5119 if (!bond->wq) 5120 return -ENOMEM; 5121 5122 spin_lock_init(&bond->stats_lock); 5123 netdev_lockdep_set_classes(bond_dev); 5124 5125 list_add_tail(&bond->bond_list, &bn->dev_list); 5126 5127 bond_prepare_sysfs_group(bond); 5128 5129 bond_debug_register(bond); 5130 5131 /* Ensure valid dev_addr */ 5132 if (is_zero_ether_addr(bond_dev->dev_addr) && 5133 bond_dev->addr_assign_type == NET_ADDR_PERM) 5134 eth_hw_addr_random(bond_dev); 5135 5136 return 0; 5137 } 5138 5139 unsigned int bond_get_num_tx_queues(void) 5140 { 5141 return tx_queues; 5142 } 5143 5144 /* Create a new bond based on the specified name and bonding parameters. 5145 * If name is NULL, obtain a suitable "bond%d" name for us. 5146 * Caller must NOT hold rtnl_lock; we need to release it here before we 5147 * set up our sysfs entries. 5148 */ 5149 int bond_create(struct net *net, const char *name) 5150 { 5151 struct net_device *bond_dev; 5152 struct bonding *bond; 5153 struct alb_bond_info *bond_info; 5154 int res; 5155 5156 rtnl_lock(); 5157 5158 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 5159 name ? name : "bond%d", NET_NAME_UNKNOWN, 5160 bond_setup, tx_queues); 5161 if (!bond_dev) { 5162 pr_err("%s: eek! can't alloc netdev!\n", name); 5163 rtnl_unlock(); 5164 return -ENOMEM; 5165 } 5166 5167 /* 5168 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 5169 * It is set to 0 by default which is wrong. 5170 */ 5171 bond = netdev_priv(bond_dev); 5172 bond_info = &(BOND_ALB_INFO(bond)); 5173 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 5174 5175 dev_net_set(bond_dev, net); 5176 bond_dev->rtnl_link_ops = &bond_link_ops; 5177 5178 res = register_netdevice(bond_dev); 5179 5180 netif_carrier_off(bond_dev); 5181 5182 bond_work_init_all(bond); 5183 5184 rtnl_unlock(); 5185 if (res < 0) 5186 free_netdev(bond_dev); 5187 return res; 5188 } 5189 5190 static int __net_init bond_net_init(struct net *net) 5191 { 5192 struct bond_net *bn = net_generic(net, bond_net_id); 5193 5194 bn->net = net; 5195 INIT_LIST_HEAD(&bn->dev_list); 5196 5197 bond_create_proc_dir(bn); 5198 bond_create_sysfs(bn); 5199 5200 return 0; 5201 } 5202 5203 static void __net_exit bond_net_exit(struct net *net) 5204 { 5205 struct bond_net *bn = net_generic(net, bond_net_id); 5206 struct bonding *bond, *tmp_bond; 5207 LIST_HEAD(list); 5208 5209 bond_destroy_sysfs(bn); 5210 5211 /* Kill off any bonds created after unregistering bond rtnl ops */ 5212 rtnl_lock(); 5213 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 5214 unregister_netdevice_queue(bond->dev, &list); 5215 unregister_netdevice_many(&list); 5216 rtnl_unlock(); 5217 5218 bond_destroy_proc_dir(bn); 5219 } 5220 5221 static struct pernet_operations bond_net_ops = { 5222 .init = bond_net_init, 5223 .exit = bond_net_exit, 5224 .id = &bond_net_id, 5225 .size = sizeof(struct bond_net), 5226 }; 5227 5228 static int __init bonding_init(void) 5229 { 5230 int i; 5231 int res; 5232 5233 res = bond_check_params(&bonding_defaults); 5234 if (res) 5235 goto out; 5236 5237 res = register_pernet_subsys(&bond_net_ops); 5238 if (res) 5239 goto out; 5240 5241 res = bond_netlink_init(); 5242 if (res) 5243 goto err_link; 5244 5245 bond_create_debugfs(); 5246 5247 for (i = 0; i < max_bonds; i++) { 5248 res = bond_create(&init_net, NULL); 5249 if (res) 5250 goto err; 5251 } 5252 5253 skb_flow_dissector_init(&flow_keys_bonding, 5254 flow_keys_bonding_keys, 5255 ARRAY_SIZE(flow_keys_bonding_keys)); 5256 5257 register_netdevice_notifier(&bond_netdev_notifier); 5258 out: 5259 return res; 5260 err: 5261 bond_destroy_debugfs(); 5262 bond_netlink_fini(); 5263 err_link: 5264 unregister_pernet_subsys(&bond_net_ops); 5265 goto out; 5266 5267 } 5268 5269 static void __exit bonding_exit(void) 5270 { 5271 unregister_netdevice_notifier(&bond_netdev_notifier); 5272 5273 bond_destroy_debugfs(); 5274 5275 bond_netlink_fini(); 5276 unregister_pernet_subsys(&bond_net_ops); 5277 5278 #ifdef CONFIG_NET_POLL_CONTROLLER 5279 /* Make sure we don't have an imbalance on our netpoll blocking */ 5280 WARN_ON(atomic_read(&netpoll_block_tx)); 5281 #endif 5282 } 5283 5284 module_init(bonding_init); 5285 module_exit(bonding_exit); 5286 MODULE_LICENSE("GPL"); 5287 MODULE_DESCRIPTION(DRV_DESCRIPTION); 5288 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 5289