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 #if IS_ENABLED(CONFIG_TLS_DEVICE) 87 #include <net/tls.h> 88 #endif 89 90 #include "bonding_priv.h" 91 92 /*---------------------------- Module parameters ----------------------------*/ 93 94 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 95 96 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 97 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 98 static int num_peer_notif = 1; 99 static int miimon; 100 static int updelay; 101 static int downdelay; 102 static int use_carrier = 1; 103 static char *mode; 104 static char *primary; 105 static char *primary_reselect; 106 static char *lacp_rate; 107 static int min_links; 108 static char *ad_select; 109 static char *xmit_hash_policy; 110 static int arp_interval; 111 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 112 static char *arp_validate; 113 static char *arp_all_targets; 114 static char *fail_over_mac; 115 static int all_slaves_active; 116 static struct bond_params bonding_defaults; 117 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 118 static int packets_per_slave = 1; 119 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 120 121 module_param(max_bonds, int, 0); 122 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 123 module_param(tx_queues, int, 0); 124 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 125 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 126 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 127 "failover event (alias of num_unsol_na)"); 128 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 129 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 130 "failover event (alias of num_grat_arp)"); 131 module_param(miimon, int, 0); 132 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 133 module_param(updelay, int, 0); 134 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 135 module_param(downdelay, int, 0); 136 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 137 "in milliseconds"); 138 module_param(use_carrier, int, 0); 139 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 140 "0 for off, 1 for on (default)"); 141 module_param(mode, charp, 0); 142 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 143 "1 for active-backup, 2 for balance-xor, " 144 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 145 "6 for balance-alb"); 146 module_param(primary, charp, 0); 147 MODULE_PARM_DESC(primary, "Primary network device to use"); 148 module_param(primary_reselect, charp, 0); 149 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 150 "once it comes up; " 151 "0 for always (default), " 152 "1 for only if speed of primary is " 153 "better, " 154 "2 for only on active slave " 155 "failure"); 156 module_param(lacp_rate, charp, 0); 157 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 158 "0 for slow, 1 for fast"); 159 module_param(ad_select, charp, 0); 160 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 161 "0 for stable (default), 1 for bandwidth, " 162 "2 for count"); 163 module_param(min_links, int, 0); 164 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 165 166 module_param(xmit_hash_policy, charp, 0); 167 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 168 "0 for layer 2 (default), 1 for layer 3+4, " 169 "2 for layer 2+3, 3 for encap layer 2+3, " 170 "4 for encap layer 3+4, 5 for vlan+srcmac"); 171 module_param(arp_interval, int, 0); 172 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 173 module_param_array(arp_ip_target, charp, NULL, 0); 174 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 175 module_param(arp_validate, charp, 0); 176 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 177 "0 for none (default), 1 for active, " 178 "2 for backup, 3 for all"); 179 module_param(arp_all_targets, charp, 0); 180 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 181 module_param(fail_over_mac, charp, 0); 182 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 183 "the same MAC; 0 for none (default), " 184 "1 for active, 2 for follow"); 185 module_param(all_slaves_active, int, 0); 186 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 187 "by setting active flag for all slaves; " 188 "0 for never (default), 1 for always."); 189 module_param(resend_igmp, int, 0); 190 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 191 "link failure"); 192 module_param(packets_per_slave, int, 0); 193 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 194 "mode; 0 for a random slave, 1 packet per " 195 "slave (default), >1 packets per slave."); 196 module_param(lp_interval, uint, 0); 197 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 198 "the bonding driver sends learning packets to " 199 "each slaves peer switch. The default is 1."); 200 201 /*----------------------------- Global variables ----------------------------*/ 202 203 #ifdef CONFIG_NET_POLL_CONTROLLER 204 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 205 #endif 206 207 unsigned int bond_net_id __read_mostly; 208 209 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 210 { 211 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 212 .offset = offsetof(struct flow_keys, control), 213 }, 214 { 215 .key_id = FLOW_DISSECTOR_KEY_BASIC, 216 .offset = offsetof(struct flow_keys, basic), 217 }, 218 { 219 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 220 .offset = offsetof(struct flow_keys, addrs.v4addrs), 221 }, 222 { 223 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 224 .offset = offsetof(struct flow_keys, addrs.v6addrs), 225 }, 226 { 227 .key_id = FLOW_DISSECTOR_KEY_TIPC, 228 .offset = offsetof(struct flow_keys, addrs.tipckey), 229 }, 230 { 231 .key_id = FLOW_DISSECTOR_KEY_PORTS, 232 .offset = offsetof(struct flow_keys, ports), 233 }, 234 { 235 .key_id = FLOW_DISSECTOR_KEY_ICMP, 236 .offset = offsetof(struct flow_keys, icmp), 237 }, 238 { 239 .key_id = FLOW_DISSECTOR_KEY_VLAN, 240 .offset = offsetof(struct flow_keys, vlan), 241 }, 242 { 243 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 244 .offset = offsetof(struct flow_keys, tags), 245 }, 246 { 247 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 248 .offset = offsetof(struct flow_keys, keyid), 249 }, 250 }; 251 252 static struct flow_dissector flow_keys_bonding __read_mostly; 253 254 /*-------------------------- Forward declarations ---------------------------*/ 255 256 static int bond_init(struct net_device *bond_dev); 257 static void bond_uninit(struct net_device *bond_dev); 258 static void bond_get_stats(struct net_device *bond_dev, 259 struct rtnl_link_stats64 *stats); 260 static void bond_slave_arr_handler(struct work_struct *work); 261 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 262 int mod); 263 static void bond_netdev_notify_work(struct work_struct *work); 264 265 /*---------------------------- General routines -----------------------------*/ 266 267 const char *bond_mode_name(int mode) 268 { 269 static const char *names[] = { 270 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 271 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 272 [BOND_MODE_XOR] = "load balancing (xor)", 273 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 274 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 275 [BOND_MODE_TLB] = "transmit load balancing", 276 [BOND_MODE_ALB] = "adaptive load balancing", 277 }; 278 279 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 280 return "unknown"; 281 282 return names[mode]; 283 } 284 285 /** 286 * bond_dev_queue_xmit - Prepare skb for xmit. 287 * 288 * @bond: bond device that got this skb for tx. 289 * @skb: hw accel VLAN tagged skb to transmit 290 * @slave_dev: slave that is supposed to xmit this skbuff 291 */ 292 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 293 struct net_device *slave_dev) 294 { 295 skb->dev = slave_dev; 296 297 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 299 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 300 301 if (unlikely(netpoll_tx_running(bond->dev))) 302 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 303 304 return dev_queue_xmit(skb); 305 } 306 307 bool bond_sk_check(struct bonding *bond) 308 { 309 switch (BOND_MODE(bond)) { 310 case BOND_MODE_8023AD: 311 case BOND_MODE_XOR: 312 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 313 return true; 314 fallthrough; 315 default: 316 return false; 317 } 318 } 319 320 /*---------------------------------- VLAN -----------------------------------*/ 321 322 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 323 * We don't protect the slave list iteration with a lock because: 324 * a. This operation is performed in IOCTL context, 325 * b. The operation is protected by the RTNL semaphore in the 8021q code, 326 * c. Holding a lock with BH disabled while directly calling a base driver 327 * entry point is generally a BAD idea. 328 * 329 * The design of synchronization/protection for this operation in the 8021q 330 * module is good for one or more VLAN devices over a single physical device 331 * and cannot be extended for a teaming solution like bonding, so there is a 332 * potential race condition here where a net device from the vlan group might 333 * be referenced (either by a base driver or the 8021q code) while it is being 334 * removed from the system. However, it turns out we're not making matters 335 * worse, and if it works for regular VLAN usage it will work here too. 336 */ 337 338 /** 339 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 340 * @bond_dev: bonding net device that got called 341 * @proto: network protocol ID 342 * @vid: vlan id being added 343 */ 344 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 345 __be16 proto, u16 vid) 346 { 347 struct bonding *bond = netdev_priv(bond_dev); 348 struct slave *slave, *rollback_slave; 349 struct list_head *iter; 350 int res; 351 352 bond_for_each_slave(bond, slave, iter) { 353 res = vlan_vid_add(slave->dev, proto, vid); 354 if (res) 355 goto unwind; 356 } 357 358 return 0; 359 360 unwind: 361 /* unwind to the slave that failed */ 362 bond_for_each_slave(bond, rollback_slave, iter) { 363 if (rollback_slave == slave) 364 break; 365 366 vlan_vid_del(rollback_slave->dev, proto, vid); 367 } 368 369 return res; 370 } 371 372 /** 373 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 374 * @bond_dev: bonding net device that got called 375 * @proto: network protocol ID 376 * @vid: vlan id being removed 377 */ 378 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 379 __be16 proto, u16 vid) 380 { 381 struct bonding *bond = netdev_priv(bond_dev); 382 struct list_head *iter; 383 struct slave *slave; 384 385 bond_for_each_slave(bond, slave, iter) 386 vlan_vid_del(slave->dev, proto, vid); 387 388 if (bond_is_lb(bond)) 389 bond_alb_clear_vlan(bond, vid); 390 391 return 0; 392 } 393 394 /*---------------------------------- XFRM -----------------------------------*/ 395 396 #ifdef CONFIG_XFRM_OFFLOAD 397 /** 398 * bond_ipsec_add_sa - program device with a security association 399 * @xs: pointer to transformer state struct 400 **/ 401 static int bond_ipsec_add_sa(struct xfrm_state *xs) 402 { 403 struct net_device *bond_dev = xs->xso.dev; 404 struct bonding *bond; 405 struct slave *slave; 406 407 if (!bond_dev) 408 return -EINVAL; 409 410 bond = netdev_priv(bond_dev); 411 slave = rcu_dereference(bond->curr_active_slave); 412 xs->xso.real_dev = slave->dev; 413 bond->xs = xs; 414 415 if (!(slave->dev->xfrmdev_ops 416 && slave->dev->xfrmdev_ops->xdo_dev_state_add)) { 417 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n"); 418 return -EINVAL; 419 } 420 421 return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs); 422 } 423 424 /** 425 * bond_ipsec_del_sa - clear out this specific SA 426 * @xs: pointer to transformer state struct 427 **/ 428 static void bond_ipsec_del_sa(struct xfrm_state *xs) 429 { 430 struct net_device *bond_dev = xs->xso.dev; 431 struct bonding *bond; 432 struct slave *slave; 433 434 if (!bond_dev) 435 return; 436 437 bond = netdev_priv(bond_dev); 438 slave = rcu_dereference(bond->curr_active_slave); 439 440 if (!slave) 441 return; 442 443 xs->xso.real_dev = slave->dev; 444 445 if (!(slave->dev->xfrmdev_ops 446 && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) { 447 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__); 448 return; 449 } 450 451 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs); 452 } 453 454 /** 455 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 456 * @skb: current data packet 457 * @xs: pointer to transformer state struct 458 **/ 459 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 460 { 461 struct net_device *bond_dev = xs->xso.dev; 462 struct bonding *bond = netdev_priv(bond_dev); 463 struct slave *curr_active = rcu_dereference(bond->curr_active_slave); 464 struct net_device *slave_dev = curr_active->dev; 465 466 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 467 return true; 468 469 if (!(slave_dev->xfrmdev_ops 470 && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) { 471 slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__); 472 return false; 473 } 474 475 xs->xso.real_dev = slave_dev; 476 return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 477 } 478 479 static const struct xfrmdev_ops bond_xfrmdev_ops = { 480 .xdo_dev_state_add = bond_ipsec_add_sa, 481 .xdo_dev_state_delete = bond_ipsec_del_sa, 482 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 483 }; 484 #endif /* CONFIG_XFRM_OFFLOAD */ 485 486 /*------------------------------- Link status -------------------------------*/ 487 488 /* Set the carrier state for the master according to the state of its 489 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 490 * do special 802.3ad magic. 491 * 492 * Returns zero if carrier state does not change, nonzero if it does. 493 */ 494 int bond_set_carrier(struct bonding *bond) 495 { 496 struct list_head *iter; 497 struct slave *slave; 498 499 if (!bond_has_slaves(bond)) 500 goto down; 501 502 if (BOND_MODE(bond) == BOND_MODE_8023AD) 503 return bond_3ad_set_carrier(bond); 504 505 bond_for_each_slave(bond, slave, iter) { 506 if (slave->link == BOND_LINK_UP) { 507 if (!netif_carrier_ok(bond->dev)) { 508 netif_carrier_on(bond->dev); 509 return 1; 510 } 511 return 0; 512 } 513 } 514 515 down: 516 if (netif_carrier_ok(bond->dev)) { 517 netif_carrier_off(bond->dev); 518 return 1; 519 } 520 return 0; 521 } 522 523 /* Get link speed and duplex from the slave's base driver 524 * using ethtool. If for some reason the call fails or the 525 * values are invalid, set speed and duplex to -1, 526 * and return. Return 1 if speed or duplex settings are 527 * UNKNOWN; 0 otherwise. 528 */ 529 static int bond_update_speed_duplex(struct slave *slave) 530 { 531 struct net_device *slave_dev = slave->dev; 532 struct ethtool_link_ksettings ecmd; 533 int res; 534 535 slave->speed = SPEED_UNKNOWN; 536 slave->duplex = DUPLEX_UNKNOWN; 537 538 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 539 if (res < 0) 540 return 1; 541 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 542 return 1; 543 switch (ecmd.base.duplex) { 544 case DUPLEX_FULL: 545 case DUPLEX_HALF: 546 break; 547 default: 548 return 1; 549 } 550 551 slave->speed = ecmd.base.speed; 552 slave->duplex = ecmd.base.duplex; 553 554 return 0; 555 } 556 557 const char *bond_slave_link_status(s8 link) 558 { 559 switch (link) { 560 case BOND_LINK_UP: 561 return "up"; 562 case BOND_LINK_FAIL: 563 return "going down"; 564 case BOND_LINK_DOWN: 565 return "down"; 566 case BOND_LINK_BACK: 567 return "going back"; 568 default: 569 return "unknown"; 570 } 571 } 572 573 /* if <dev> supports MII link status reporting, check its link status. 574 * 575 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 576 * depending upon the setting of the use_carrier parameter. 577 * 578 * Return either BMSR_LSTATUS, meaning that the link is up (or we 579 * can't tell and just pretend it is), or 0, meaning that the link is 580 * down. 581 * 582 * If reporting is non-zero, instead of faking link up, return -1 if 583 * both ETHTOOL and MII ioctls fail (meaning the device does not 584 * support them). If use_carrier is set, return whatever it says. 585 * It'd be nice if there was a good way to tell if a driver supports 586 * netif_carrier, but there really isn't. 587 */ 588 static int bond_check_dev_link(struct bonding *bond, 589 struct net_device *slave_dev, int reporting) 590 { 591 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 592 int (*ioctl)(struct net_device *, struct ifreq *, int); 593 struct ifreq ifr; 594 struct mii_ioctl_data *mii; 595 596 if (!reporting && !netif_running(slave_dev)) 597 return 0; 598 599 if (bond->params.use_carrier) 600 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 601 602 /* Try to get link status using Ethtool first. */ 603 if (slave_dev->ethtool_ops->get_link) 604 return slave_dev->ethtool_ops->get_link(slave_dev) ? 605 BMSR_LSTATUS : 0; 606 607 /* Ethtool can't be used, fallback to MII ioctls. */ 608 ioctl = slave_ops->ndo_do_ioctl; 609 if (ioctl) { 610 /* TODO: set pointer to correct ioctl on a per team member 611 * bases to make this more efficient. that is, once 612 * we determine the correct ioctl, we will always 613 * call it and not the others for that team 614 * member. 615 */ 616 617 /* We cannot assume that SIOCGMIIPHY will also read a 618 * register; not all network drivers (e.g., e100) 619 * support that. 620 */ 621 622 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 623 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 624 mii = if_mii(&ifr); 625 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 626 mii->reg_num = MII_BMSR; 627 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 628 return mii->val_out & BMSR_LSTATUS; 629 } 630 } 631 632 /* If reporting, report that either there's no dev->do_ioctl, 633 * or both SIOCGMIIREG and get_link failed (meaning that we 634 * cannot report link status). If not reporting, pretend 635 * we're ok. 636 */ 637 return reporting ? -1 : BMSR_LSTATUS; 638 } 639 640 /*----------------------------- Multicast list ------------------------------*/ 641 642 /* Push the promiscuity flag down to appropriate slaves */ 643 static int bond_set_promiscuity(struct bonding *bond, int inc) 644 { 645 struct list_head *iter; 646 int err = 0; 647 648 if (bond_uses_primary(bond)) { 649 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 650 651 if (curr_active) 652 err = dev_set_promiscuity(curr_active->dev, inc); 653 } else { 654 struct slave *slave; 655 656 bond_for_each_slave(bond, slave, iter) { 657 err = dev_set_promiscuity(slave->dev, inc); 658 if (err) 659 return err; 660 } 661 } 662 return err; 663 } 664 665 /* Push the allmulti flag down to all slaves */ 666 static int bond_set_allmulti(struct bonding *bond, int inc) 667 { 668 struct list_head *iter; 669 int err = 0; 670 671 if (bond_uses_primary(bond)) { 672 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 673 674 if (curr_active) 675 err = dev_set_allmulti(curr_active->dev, inc); 676 } else { 677 struct slave *slave; 678 679 bond_for_each_slave(bond, slave, iter) { 680 err = dev_set_allmulti(slave->dev, inc); 681 if (err) 682 return err; 683 } 684 } 685 return err; 686 } 687 688 /* Retrieve the list of registered multicast addresses for the bonding 689 * device and retransmit an IGMP JOIN request to the current active 690 * slave. 691 */ 692 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 693 { 694 struct bonding *bond = container_of(work, struct bonding, 695 mcast_work.work); 696 697 if (!rtnl_trylock()) { 698 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 699 return; 700 } 701 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 702 703 if (bond->igmp_retrans > 1) { 704 bond->igmp_retrans--; 705 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 706 } 707 rtnl_unlock(); 708 } 709 710 /* Flush bond's hardware addresses from slave */ 711 static void bond_hw_addr_flush(struct net_device *bond_dev, 712 struct net_device *slave_dev) 713 { 714 struct bonding *bond = netdev_priv(bond_dev); 715 716 dev_uc_unsync(slave_dev, bond_dev); 717 dev_mc_unsync(slave_dev, bond_dev); 718 719 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 720 /* del lacpdu mc addr from mc list */ 721 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 722 723 dev_mc_del(slave_dev, lacpdu_multicast); 724 } 725 } 726 727 /*--------------------------- Active slave change ---------------------------*/ 728 729 /* Update the hardware address list and promisc/allmulti for the new and 730 * old active slaves (if any). Modes that are not using primary keep all 731 * slaves up date at all times; only the modes that use primary need to call 732 * this function to swap these settings during a failover. 733 */ 734 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 735 struct slave *old_active) 736 { 737 if (old_active) { 738 if (bond->dev->flags & IFF_PROMISC) 739 dev_set_promiscuity(old_active->dev, -1); 740 741 if (bond->dev->flags & IFF_ALLMULTI) 742 dev_set_allmulti(old_active->dev, -1); 743 744 bond_hw_addr_flush(bond->dev, old_active->dev); 745 } 746 747 if (new_active) { 748 /* FIXME: Signal errors upstream. */ 749 if (bond->dev->flags & IFF_PROMISC) 750 dev_set_promiscuity(new_active->dev, 1); 751 752 if (bond->dev->flags & IFF_ALLMULTI) 753 dev_set_allmulti(new_active->dev, 1); 754 755 netif_addr_lock_bh(bond->dev); 756 dev_uc_sync(new_active->dev, bond->dev); 757 dev_mc_sync(new_active->dev, bond->dev); 758 netif_addr_unlock_bh(bond->dev); 759 } 760 } 761 762 /** 763 * bond_set_dev_addr - clone slave's address to bond 764 * @bond_dev: bond net device 765 * @slave_dev: slave net device 766 * 767 * Should be called with RTNL held. 768 */ 769 static int bond_set_dev_addr(struct net_device *bond_dev, 770 struct net_device *slave_dev) 771 { 772 int err; 773 774 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 775 bond_dev, slave_dev, slave_dev->addr_len); 776 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 777 if (err) 778 return err; 779 780 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len); 781 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 782 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 783 return 0; 784 } 785 786 static struct slave *bond_get_old_active(struct bonding *bond, 787 struct slave *new_active) 788 { 789 struct slave *slave; 790 struct list_head *iter; 791 792 bond_for_each_slave(bond, slave, iter) { 793 if (slave == new_active) 794 continue; 795 796 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 797 return slave; 798 } 799 800 return NULL; 801 } 802 803 /* bond_do_fail_over_mac 804 * 805 * Perform special MAC address swapping for fail_over_mac settings 806 * 807 * Called with RTNL 808 */ 809 static void bond_do_fail_over_mac(struct bonding *bond, 810 struct slave *new_active, 811 struct slave *old_active) 812 { 813 u8 tmp_mac[MAX_ADDR_LEN]; 814 struct sockaddr_storage ss; 815 int rv; 816 817 switch (bond->params.fail_over_mac) { 818 case BOND_FOM_ACTIVE: 819 if (new_active) { 820 rv = bond_set_dev_addr(bond->dev, new_active->dev); 821 if (rv) 822 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 823 -rv); 824 } 825 break; 826 case BOND_FOM_FOLLOW: 827 /* if new_active && old_active, swap them 828 * if just old_active, do nothing (going to no active slave) 829 * if just new_active, set new_active to bond's MAC 830 */ 831 if (!new_active) 832 return; 833 834 if (!old_active) 835 old_active = bond_get_old_active(bond, new_active); 836 837 if (old_active) { 838 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 839 new_active->dev->addr_len); 840 bond_hw_addr_copy(ss.__data, 841 old_active->dev->dev_addr, 842 old_active->dev->addr_len); 843 ss.ss_family = new_active->dev->type; 844 } else { 845 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 846 bond->dev->addr_len); 847 ss.ss_family = bond->dev->type; 848 } 849 850 rv = dev_set_mac_address(new_active->dev, 851 (struct sockaddr *)&ss, NULL); 852 if (rv) { 853 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 854 -rv); 855 goto out; 856 } 857 858 if (!old_active) 859 goto out; 860 861 bond_hw_addr_copy(ss.__data, tmp_mac, 862 new_active->dev->addr_len); 863 ss.ss_family = old_active->dev->type; 864 865 rv = dev_set_mac_address(old_active->dev, 866 (struct sockaddr *)&ss, NULL); 867 if (rv) 868 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 869 -rv); 870 out: 871 break; 872 default: 873 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 874 bond->params.fail_over_mac); 875 break; 876 } 877 878 } 879 880 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 881 { 882 struct slave *prim = rtnl_dereference(bond->primary_slave); 883 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 884 885 if (!prim || prim->link != BOND_LINK_UP) { 886 if (!curr || curr->link != BOND_LINK_UP) 887 return NULL; 888 return curr; 889 } 890 891 if (bond->force_primary) { 892 bond->force_primary = false; 893 return prim; 894 } 895 896 if (!curr || curr->link != BOND_LINK_UP) 897 return prim; 898 899 /* At this point, prim and curr are both up */ 900 switch (bond->params.primary_reselect) { 901 case BOND_PRI_RESELECT_ALWAYS: 902 return prim; 903 case BOND_PRI_RESELECT_BETTER: 904 if (prim->speed < curr->speed) 905 return curr; 906 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 907 return curr; 908 return prim; 909 case BOND_PRI_RESELECT_FAILURE: 910 return curr; 911 default: 912 netdev_err(bond->dev, "impossible primary_reselect %d\n", 913 bond->params.primary_reselect); 914 return curr; 915 } 916 } 917 918 /** 919 * bond_find_best_slave - select the best available slave to be the active one 920 * @bond: our bonding struct 921 */ 922 static struct slave *bond_find_best_slave(struct bonding *bond) 923 { 924 struct slave *slave, *bestslave = NULL; 925 struct list_head *iter; 926 int mintime = bond->params.updelay; 927 928 slave = bond_choose_primary_or_current(bond); 929 if (slave) 930 return slave; 931 932 bond_for_each_slave(bond, slave, iter) { 933 if (slave->link == BOND_LINK_UP) 934 return slave; 935 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 936 slave->delay < mintime) { 937 mintime = slave->delay; 938 bestslave = slave; 939 } 940 } 941 942 return bestslave; 943 } 944 945 static bool bond_should_notify_peers(struct bonding *bond) 946 { 947 struct slave *slave; 948 949 rcu_read_lock(); 950 slave = rcu_dereference(bond->curr_active_slave); 951 rcu_read_unlock(); 952 953 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 954 slave ? slave->dev->name : "NULL"); 955 956 if (!slave || !bond->send_peer_notif || 957 bond->send_peer_notif % 958 max(1, bond->params.peer_notif_delay) != 0 || 959 !netif_carrier_ok(bond->dev) || 960 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 961 return false; 962 963 return true; 964 } 965 966 /** 967 * bond_change_active_slave - change the active slave into the specified one 968 * @bond: our bonding struct 969 * @new_active: the new slave to make the active one 970 * 971 * Set the new slave to the bond's settings and unset them on the old 972 * curr_active_slave. 973 * Setting include flags, mc-list, promiscuity, allmulti, etc. 974 * 975 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 976 * because it is apparently the best available slave we have, even though its 977 * updelay hasn't timed out yet. 978 * 979 * Caller must hold RTNL. 980 */ 981 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 982 { 983 struct slave *old_active; 984 985 ASSERT_RTNL(); 986 987 old_active = rtnl_dereference(bond->curr_active_slave); 988 989 if (old_active == new_active) 990 return; 991 992 #ifdef CONFIG_XFRM_OFFLOAD 993 if (old_active && bond->xs) 994 bond_ipsec_del_sa(bond->xs); 995 #endif /* CONFIG_XFRM_OFFLOAD */ 996 997 if (new_active) { 998 new_active->last_link_up = jiffies; 999 1000 if (new_active->link == BOND_LINK_BACK) { 1001 if (bond_uses_primary(bond)) { 1002 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1003 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1004 } 1005 1006 new_active->delay = 0; 1007 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1008 BOND_SLAVE_NOTIFY_NOW); 1009 1010 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1011 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1012 1013 if (bond_is_lb(bond)) 1014 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1015 } else { 1016 if (bond_uses_primary(bond)) 1017 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1018 } 1019 } 1020 1021 if (bond_uses_primary(bond)) 1022 bond_hw_addr_swap(bond, new_active, old_active); 1023 1024 if (bond_is_lb(bond)) { 1025 bond_alb_handle_active_change(bond, new_active); 1026 if (old_active) 1027 bond_set_slave_inactive_flags(old_active, 1028 BOND_SLAVE_NOTIFY_NOW); 1029 if (new_active) 1030 bond_set_slave_active_flags(new_active, 1031 BOND_SLAVE_NOTIFY_NOW); 1032 } else { 1033 rcu_assign_pointer(bond->curr_active_slave, new_active); 1034 } 1035 1036 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1037 if (old_active) 1038 bond_set_slave_inactive_flags(old_active, 1039 BOND_SLAVE_NOTIFY_NOW); 1040 1041 if (new_active) { 1042 bool should_notify_peers = false; 1043 1044 bond_set_slave_active_flags(new_active, 1045 BOND_SLAVE_NOTIFY_NOW); 1046 1047 if (bond->params.fail_over_mac) 1048 bond_do_fail_over_mac(bond, new_active, 1049 old_active); 1050 1051 if (netif_running(bond->dev)) { 1052 bond->send_peer_notif = 1053 bond->params.num_peer_notif * 1054 max(1, bond->params.peer_notif_delay); 1055 should_notify_peers = 1056 bond_should_notify_peers(bond); 1057 } 1058 1059 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1060 if (should_notify_peers) { 1061 bond->send_peer_notif--; 1062 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1063 bond->dev); 1064 } 1065 } 1066 } 1067 1068 #ifdef CONFIG_XFRM_OFFLOAD 1069 if (new_active && bond->xs) { 1070 xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true); 1071 bond_ipsec_add_sa(bond->xs); 1072 } 1073 #endif /* CONFIG_XFRM_OFFLOAD */ 1074 1075 /* resend IGMP joins since active slave has changed or 1076 * all were sent on curr_active_slave. 1077 * resend only if bond is brought up with the affected 1078 * bonding modes and the retransmission is enabled 1079 */ 1080 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1081 ((bond_uses_primary(bond) && new_active) || 1082 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1083 bond->igmp_retrans = bond->params.resend_igmp; 1084 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1085 } 1086 } 1087 1088 /** 1089 * bond_select_active_slave - select a new active slave, if needed 1090 * @bond: our bonding struct 1091 * 1092 * This functions should be called when one of the following occurs: 1093 * - The old curr_active_slave has been released or lost its link. 1094 * - The primary_slave has got its link back. 1095 * - A slave has got its link back and there's no old curr_active_slave. 1096 * 1097 * Caller must hold RTNL. 1098 */ 1099 void bond_select_active_slave(struct bonding *bond) 1100 { 1101 struct slave *best_slave; 1102 int rv; 1103 1104 ASSERT_RTNL(); 1105 1106 best_slave = bond_find_best_slave(bond); 1107 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1108 bond_change_active_slave(bond, best_slave); 1109 rv = bond_set_carrier(bond); 1110 if (!rv) 1111 return; 1112 1113 if (netif_carrier_ok(bond->dev)) 1114 netdev_info(bond->dev, "active interface up!\n"); 1115 else 1116 netdev_info(bond->dev, "now running without any active interface!\n"); 1117 } 1118 } 1119 1120 #ifdef CONFIG_NET_POLL_CONTROLLER 1121 static inline int slave_enable_netpoll(struct slave *slave) 1122 { 1123 struct netpoll *np; 1124 int err = 0; 1125 1126 np = kzalloc(sizeof(*np), GFP_KERNEL); 1127 err = -ENOMEM; 1128 if (!np) 1129 goto out; 1130 1131 err = __netpoll_setup(np, slave->dev); 1132 if (err) { 1133 kfree(np); 1134 goto out; 1135 } 1136 slave->np = np; 1137 out: 1138 return err; 1139 } 1140 static inline void slave_disable_netpoll(struct slave *slave) 1141 { 1142 struct netpoll *np = slave->np; 1143 1144 if (!np) 1145 return; 1146 1147 slave->np = NULL; 1148 1149 __netpoll_free(np); 1150 } 1151 1152 static void bond_poll_controller(struct net_device *bond_dev) 1153 { 1154 struct bonding *bond = netdev_priv(bond_dev); 1155 struct slave *slave = NULL; 1156 struct list_head *iter; 1157 struct ad_info ad_info; 1158 1159 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1160 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1161 return; 1162 1163 bond_for_each_slave_rcu(bond, slave, iter) { 1164 if (!bond_slave_is_up(slave)) 1165 continue; 1166 1167 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1168 struct aggregator *agg = 1169 SLAVE_AD_INFO(slave)->port.aggregator; 1170 1171 if (agg && 1172 agg->aggregator_identifier != ad_info.aggregator_id) 1173 continue; 1174 } 1175 1176 netpoll_poll_dev(slave->dev); 1177 } 1178 } 1179 1180 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1181 { 1182 struct bonding *bond = netdev_priv(bond_dev); 1183 struct list_head *iter; 1184 struct slave *slave; 1185 1186 bond_for_each_slave(bond, slave, iter) 1187 if (bond_slave_is_up(slave)) 1188 slave_disable_netpoll(slave); 1189 } 1190 1191 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1192 { 1193 struct bonding *bond = netdev_priv(dev); 1194 struct list_head *iter; 1195 struct slave *slave; 1196 int err = 0; 1197 1198 bond_for_each_slave(bond, slave, iter) { 1199 err = slave_enable_netpoll(slave); 1200 if (err) { 1201 bond_netpoll_cleanup(dev); 1202 break; 1203 } 1204 } 1205 return err; 1206 } 1207 #else 1208 static inline int slave_enable_netpoll(struct slave *slave) 1209 { 1210 return 0; 1211 } 1212 static inline void slave_disable_netpoll(struct slave *slave) 1213 { 1214 } 1215 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1216 { 1217 } 1218 #endif 1219 1220 /*---------------------------------- IOCTL ----------------------------------*/ 1221 1222 static netdev_features_t bond_fix_features(struct net_device *dev, 1223 netdev_features_t features) 1224 { 1225 struct bonding *bond = netdev_priv(dev); 1226 struct list_head *iter; 1227 netdev_features_t mask; 1228 struct slave *slave; 1229 1230 #if IS_ENABLED(CONFIG_TLS_DEVICE) 1231 if (bond_sk_check(bond)) 1232 features |= BOND_TLS_FEATURES; 1233 else 1234 features &= ~BOND_TLS_FEATURES; 1235 #endif 1236 1237 mask = features; 1238 1239 features &= ~NETIF_F_ONE_FOR_ALL; 1240 features |= NETIF_F_ALL_FOR_ALL; 1241 1242 bond_for_each_slave(bond, slave, iter) { 1243 features = netdev_increment_features(features, 1244 slave->dev->features, 1245 mask); 1246 } 1247 features = netdev_add_tso_features(features, mask); 1248 1249 return features; 1250 } 1251 1252 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1253 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1254 NETIF_F_HIGHDMA | NETIF_F_LRO) 1255 1256 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1257 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1258 1259 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1260 NETIF_F_GSO_SOFTWARE) 1261 1262 1263 static void bond_compute_features(struct bonding *bond) 1264 { 1265 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1266 IFF_XMIT_DST_RELEASE_PERM; 1267 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1268 netdev_features_t enc_features = BOND_ENC_FEATURES; 1269 #ifdef CONFIG_XFRM_OFFLOAD 1270 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1271 #endif /* CONFIG_XFRM_OFFLOAD */ 1272 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1273 struct net_device *bond_dev = bond->dev; 1274 struct list_head *iter; 1275 struct slave *slave; 1276 unsigned short max_hard_header_len = ETH_HLEN; 1277 unsigned int gso_max_size = GSO_MAX_SIZE; 1278 u16 gso_max_segs = GSO_MAX_SEGS; 1279 1280 if (!bond_has_slaves(bond)) 1281 goto done; 1282 vlan_features &= NETIF_F_ALL_FOR_ALL; 1283 mpls_features &= NETIF_F_ALL_FOR_ALL; 1284 1285 bond_for_each_slave(bond, slave, iter) { 1286 vlan_features = netdev_increment_features(vlan_features, 1287 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1288 1289 enc_features = netdev_increment_features(enc_features, 1290 slave->dev->hw_enc_features, 1291 BOND_ENC_FEATURES); 1292 1293 #ifdef CONFIG_XFRM_OFFLOAD 1294 xfrm_features = netdev_increment_features(xfrm_features, 1295 slave->dev->hw_enc_features, 1296 BOND_XFRM_FEATURES); 1297 #endif /* CONFIG_XFRM_OFFLOAD */ 1298 1299 mpls_features = netdev_increment_features(mpls_features, 1300 slave->dev->mpls_features, 1301 BOND_MPLS_FEATURES); 1302 1303 dst_release_flag &= slave->dev->priv_flags; 1304 if (slave->dev->hard_header_len > max_hard_header_len) 1305 max_hard_header_len = slave->dev->hard_header_len; 1306 1307 gso_max_size = min(gso_max_size, slave->dev->gso_max_size); 1308 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs); 1309 } 1310 bond_dev->hard_header_len = max_hard_header_len; 1311 1312 done: 1313 bond_dev->vlan_features = vlan_features; 1314 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1315 NETIF_F_HW_VLAN_CTAG_TX | 1316 NETIF_F_HW_VLAN_STAG_TX; 1317 #ifdef CONFIG_XFRM_OFFLOAD 1318 bond_dev->hw_enc_features |= xfrm_features; 1319 #endif /* CONFIG_XFRM_OFFLOAD */ 1320 bond_dev->mpls_features = mpls_features; 1321 bond_dev->gso_max_segs = gso_max_segs; 1322 netif_set_gso_max_size(bond_dev, gso_max_size); 1323 1324 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1325 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1326 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1327 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1328 1329 netdev_change_features(bond_dev); 1330 } 1331 1332 static void bond_setup_by_slave(struct net_device *bond_dev, 1333 struct net_device *slave_dev) 1334 { 1335 bond_dev->header_ops = slave_dev->header_ops; 1336 1337 bond_dev->type = slave_dev->type; 1338 bond_dev->hard_header_len = slave_dev->hard_header_len; 1339 bond_dev->needed_headroom = slave_dev->needed_headroom; 1340 bond_dev->addr_len = slave_dev->addr_len; 1341 1342 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1343 slave_dev->addr_len); 1344 } 1345 1346 /* On bonding slaves other than the currently active slave, suppress 1347 * duplicates except for alb non-mcast/bcast. 1348 */ 1349 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1350 struct slave *slave, 1351 struct bonding *bond) 1352 { 1353 if (bond_is_slave_inactive(slave)) { 1354 if (BOND_MODE(bond) == BOND_MODE_ALB && 1355 skb->pkt_type != PACKET_BROADCAST && 1356 skb->pkt_type != PACKET_MULTICAST) 1357 return false; 1358 return true; 1359 } 1360 return false; 1361 } 1362 1363 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1364 { 1365 struct sk_buff *skb = *pskb; 1366 struct slave *slave; 1367 struct bonding *bond; 1368 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1369 struct slave *); 1370 int ret = RX_HANDLER_ANOTHER; 1371 1372 skb = skb_share_check(skb, GFP_ATOMIC); 1373 if (unlikely(!skb)) 1374 return RX_HANDLER_CONSUMED; 1375 1376 *pskb = skb; 1377 1378 slave = bond_slave_get_rcu(skb->dev); 1379 bond = slave->bond; 1380 1381 recv_probe = READ_ONCE(bond->recv_probe); 1382 if (recv_probe) { 1383 ret = recv_probe(skb, bond, slave); 1384 if (ret == RX_HANDLER_CONSUMED) { 1385 consume_skb(skb); 1386 return ret; 1387 } 1388 } 1389 1390 /* 1391 * For packets determined by bond_should_deliver_exact_match() call to 1392 * be suppressed we want to make an exception for link-local packets. 1393 * This is necessary for e.g. LLDP daemons to be able to monitor 1394 * inactive slave links without being forced to bind to them 1395 * explicitly. 1396 * 1397 * At the same time, packets that are passed to the bonding master 1398 * (including link-local ones) can have their originating interface 1399 * determined via PACKET_ORIGDEV socket option. 1400 */ 1401 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1402 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1403 return RX_HANDLER_PASS; 1404 return RX_HANDLER_EXACT; 1405 } 1406 1407 skb->dev = bond->dev; 1408 1409 if (BOND_MODE(bond) == BOND_MODE_ALB && 1410 netif_is_bridge_port(bond->dev) && 1411 skb->pkt_type == PACKET_HOST) { 1412 1413 if (unlikely(skb_cow_head(skb, 1414 skb->data - skb_mac_header(skb)))) { 1415 kfree_skb(skb); 1416 return RX_HANDLER_CONSUMED; 1417 } 1418 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1419 bond->dev->addr_len); 1420 } 1421 1422 return ret; 1423 } 1424 1425 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1426 { 1427 switch (BOND_MODE(bond)) { 1428 case BOND_MODE_ROUNDROBIN: 1429 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1430 case BOND_MODE_ACTIVEBACKUP: 1431 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1432 case BOND_MODE_BROADCAST: 1433 return NETDEV_LAG_TX_TYPE_BROADCAST; 1434 case BOND_MODE_XOR: 1435 case BOND_MODE_8023AD: 1436 return NETDEV_LAG_TX_TYPE_HASH; 1437 default: 1438 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1439 } 1440 } 1441 1442 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1443 enum netdev_lag_tx_type type) 1444 { 1445 if (type != NETDEV_LAG_TX_TYPE_HASH) 1446 return NETDEV_LAG_HASH_NONE; 1447 1448 switch (bond->params.xmit_policy) { 1449 case BOND_XMIT_POLICY_LAYER2: 1450 return NETDEV_LAG_HASH_L2; 1451 case BOND_XMIT_POLICY_LAYER34: 1452 return NETDEV_LAG_HASH_L34; 1453 case BOND_XMIT_POLICY_LAYER23: 1454 return NETDEV_LAG_HASH_L23; 1455 case BOND_XMIT_POLICY_ENCAP23: 1456 return NETDEV_LAG_HASH_E23; 1457 case BOND_XMIT_POLICY_ENCAP34: 1458 return NETDEV_LAG_HASH_E34; 1459 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1460 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1461 default: 1462 return NETDEV_LAG_HASH_UNKNOWN; 1463 } 1464 } 1465 1466 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1467 struct netlink_ext_ack *extack) 1468 { 1469 struct netdev_lag_upper_info lag_upper_info; 1470 enum netdev_lag_tx_type type; 1471 1472 type = bond_lag_tx_type(bond); 1473 lag_upper_info.tx_type = type; 1474 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1475 1476 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1477 &lag_upper_info, extack); 1478 } 1479 1480 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1481 { 1482 netdev_upper_dev_unlink(slave->dev, bond->dev); 1483 slave->dev->flags &= ~IFF_SLAVE; 1484 } 1485 1486 static void slave_kobj_release(struct kobject *kobj) 1487 { 1488 struct slave *slave = to_slave(kobj); 1489 struct bonding *bond = bond_get_bond_by_slave(slave); 1490 1491 cancel_delayed_work_sync(&slave->notify_work); 1492 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1493 kfree(SLAVE_AD_INFO(slave)); 1494 1495 kfree(slave); 1496 } 1497 1498 static struct kobj_type slave_ktype = { 1499 .release = slave_kobj_release, 1500 #ifdef CONFIG_SYSFS 1501 .sysfs_ops = &slave_sysfs_ops, 1502 #endif 1503 }; 1504 1505 static int bond_kobj_init(struct slave *slave) 1506 { 1507 int err; 1508 1509 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1510 &(slave->dev->dev.kobj), "bonding_slave"); 1511 if (err) 1512 kobject_put(&slave->kobj); 1513 1514 return err; 1515 } 1516 1517 static struct slave *bond_alloc_slave(struct bonding *bond, 1518 struct net_device *slave_dev) 1519 { 1520 struct slave *slave = NULL; 1521 1522 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1523 if (!slave) 1524 return NULL; 1525 1526 slave->bond = bond; 1527 slave->dev = slave_dev; 1528 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1529 1530 if (bond_kobj_init(slave)) 1531 return NULL; 1532 1533 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1534 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1535 GFP_KERNEL); 1536 if (!SLAVE_AD_INFO(slave)) { 1537 kobject_put(&slave->kobj); 1538 return NULL; 1539 } 1540 } 1541 1542 return slave; 1543 } 1544 1545 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1546 { 1547 info->bond_mode = BOND_MODE(bond); 1548 info->miimon = bond->params.miimon; 1549 info->num_slaves = bond->slave_cnt; 1550 } 1551 1552 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1553 { 1554 strcpy(info->slave_name, slave->dev->name); 1555 info->link = slave->link; 1556 info->state = bond_slave_state(slave); 1557 info->link_failure_count = slave->link_failure_count; 1558 } 1559 1560 static void bond_netdev_notify_work(struct work_struct *_work) 1561 { 1562 struct slave *slave = container_of(_work, struct slave, 1563 notify_work.work); 1564 1565 if (rtnl_trylock()) { 1566 struct netdev_bonding_info binfo; 1567 1568 bond_fill_ifslave(slave, &binfo.slave); 1569 bond_fill_ifbond(slave->bond, &binfo.master); 1570 netdev_bonding_info_change(slave->dev, &binfo); 1571 rtnl_unlock(); 1572 } else { 1573 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1574 } 1575 } 1576 1577 void bond_queue_slave_event(struct slave *slave) 1578 { 1579 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1580 } 1581 1582 void bond_lower_state_changed(struct slave *slave) 1583 { 1584 struct netdev_lag_lower_state_info info; 1585 1586 info.link_up = slave->link == BOND_LINK_UP || 1587 slave->link == BOND_LINK_FAIL; 1588 info.tx_enabled = bond_is_active_slave(slave); 1589 netdev_lower_state_changed(slave->dev, &info); 1590 } 1591 1592 /* enslave device <slave> to bond device <master> */ 1593 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1594 struct netlink_ext_ack *extack) 1595 { 1596 struct bonding *bond = netdev_priv(bond_dev); 1597 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1598 struct slave *new_slave = NULL, *prev_slave; 1599 struct sockaddr_storage ss; 1600 int link_reporting; 1601 int res = 0, i; 1602 1603 if (slave_dev->flags & IFF_MASTER && 1604 !netif_is_bond_master(slave_dev)) { 1605 NL_SET_ERR_MSG(extack, "Device with IFF_MASTER cannot be enslaved"); 1606 netdev_err(bond_dev, 1607 "Error: Device with IFF_MASTER cannot be enslaved\n"); 1608 return -EPERM; 1609 } 1610 1611 if (!bond->params.use_carrier && 1612 slave_dev->ethtool_ops->get_link == NULL && 1613 slave_ops->ndo_do_ioctl == NULL) { 1614 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1615 } 1616 1617 /* already in-use? */ 1618 if (netdev_is_rx_handler_busy(slave_dev)) { 1619 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved"); 1620 slave_err(bond_dev, slave_dev, 1621 "Error: Device is in use and cannot be enslaved\n"); 1622 return -EBUSY; 1623 } 1624 1625 if (bond_dev == slave_dev) { 1626 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself."); 1627 netdev_err(bond_dev, "cannot enslave bond to itself.\n"); 1628 return -EPERM; 1629 } 1630 1631 /* vlan challenged mutual exclusion */ 1632 /* no need to lock since we're protected by rtnl_lock */ 1633 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1634 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1635 if (vlan_uses_dev(bond_dev)) { 1636 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond"); 1637 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n"); 1638 return -EPERM; 1639 } else { 1640 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1641 } 1642 } else { 1643 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1644 } 1645 1646 if (slave_dev->features & NETIF_F_HW_ESP) 1647 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1648 1649 /* Old ifenslave binaries are no longer supported. These can 1650 * be identified with moderate accuracy by the state of the slave: 1651 * the current ifenslave will set the interface down prior to 1652 * enslaving it; the old ifenslave will not. 1653 */ 1654 if (slave_dev->flags & IFF_UP) { 1655 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up"); 1656 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n"); 1657 return -EPERM; 1658 } 1659 1660 /* set bonding device ether type by slave - bonding netdevices are 1661 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1662 * there is a need to override some of the type dependent attribs/funcs. 1663 * 1664 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1665 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1666 */ 1667 if (!bond_has_slaves(bond)) { 1668 if (bond_dev->type != slave_dev->type) { 1669 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1670 bond_dev->type, slave_dev->type); 1671 1672 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1673 bond_dev); 1674 res = notifier_to_errno(res); 1675 if (res) { 1676 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1677 return -EBUSY; 1678 } 1679 1680 /* Flush unicast and multicast addresses */ 1681 dev_uc_flush(bond_dev); 1682 dev_mc_flush(bond_dev); 1683 1684 if (slave_dev->type != ARPHRD_ETHER) 1685 bond_setup_by_slave(bond_dev, slave_dev); 1686 else { 1687 ether_setup(bond_dev); 1688 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1689 } 1690 1691 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1692 bond_dev); 1693 } 1694 } else if (bond_dev->type != slave_dev->type) { 1695 NL_SET_ERR_MSG(extack, "Device type is different from other slaves"); 1696 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n", 1697 slave_dev->type, bond_dev->type); 1698 return -EINVAL; 1699 } 1700 1701 if (slave_dev->type == ARPHRD_INFINIBAND && 1702 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1703 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves"); 1704 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n", 1705 slave_dev->type); 1706 res = -EOPNOTSUPP; 1707 goto err_undo_flags; 1708 } 1709 1710 if (!slave_ops->ndo_set_mac_address || 1711 slave_dev->type == ARPHRD_INFINIBAND) { 1712 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1713 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1714 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1715 if (!bond_has_slaves(bond)) { 1716 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1717 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1718 } else { 1719 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1720 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"); 1721 res = -EOPNOTSUPP; 1722 goto err_undo_flags; 1723 } 1724 } 1725 } 1726 1727 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1728 1729 /* If this is the first slave, then we need to set the master's hardware 1730 * address to be the same as the slave's. 1731 */ 1732 if (!bond_has_slaves(bond) && 1733 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1734 res = bond_set_dev_addr(bond->dev, slave_dev); 1735 if (res) 1736 goto err_undo_flags; 1737 } 1738 1739 new_slave = bond_alloc_slave(bond, slave_dev); 1740 if (!new_slave) { 1741 res = -ENOMEM; 1742 goto err_undo_flags; 1743 } 1744 1745 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1746 * is set via sysfs or module option if desired. 1747 */ 1748 new_slave->queue_id = 0; 1749 1750 /* Save slave's original mtu and then set it to match the bond */ 1751 new_slave->original_mtu = slave_dev->mtu; 1752 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1753 if (res) { 1754 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1755 goto err_free; 1756 } 1757 1758 /* Save slave's original ("permanent") mac address for modes 1759 * that need it, and for restoring it upon release, and then 1760 * set it to the master's address 1761 */ 1762 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1763 slave_dev->addr_len); 1764 1765 if (!bond->params.fail_over_mac || 1766 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1767 /* Set slave to master's mac address. The application already 1768 * set the master's mac address to that of the first slave 1769 */ 1770 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1771 ss.ss_family = slave_dev->type; 1772 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1773 extack); 1774 if (res) { 1775 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1776 goto err_restore_mtu; 1777 } 1778 } 1779 1780 /* set slave flag before open to prevent IPv6 addrconf */ 1781 slave_dev->flags |= IFF_SLAVE; 1782 1783 /* open the slave since the application closed it */ 1784 res = dev_open(slave_dev, extack); 1785 if (res) { 1786 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1787 goto err_restore_mac; 1788 } 1789 1790 slave_dev->priv_flags |= IFF_BONDING; 1791 /* initialize slave stats */ 1792 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1793 1794 if (bond_is_lb(bond)) { 1795 /* bond_alb_init_slave() must be called before all other stages since 1796 * it might fail and we do not want to have to undo everything 1797 */ 1798 res = bond_alb_init_slave(bond, new_slave); 1799 if (res) 1800 goto err_close; 1801 } 1802 1803 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1804 if (res) { 1805 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1806 goto err_close; 1807 } 1808 1809 prev_slave = bond_last_slave(bond); 1810 1811 new_slave->delay = 0; 1812 new_slave->link_failure_count = 0; 1813 1814 if (bond_update_speed_duplex(new_slave) && 1815 bond_needs_speed_duplex(bond)) 1816 new_slave->link = BOND_LINK_DOWN; 1817 1818 new_slave->last_rx = jiffies - 1819 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1820 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1821 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1822 1823 if (bond->params.miimon && !bond->params.use_carrier) { 1824 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 1825 1826 if ((link_reporting == -1) && !bond->params.arp_interval) { 1827 /* miimon is set but a bonded network driver 1828 * does not support ETHTOOL/MII and 1829 * arp_interval is not set. Note: if 1830 * use_carrier is enabled, we will never go 1831 * here (because netif_carrier is always 1832 * supported); thus, we don't need to change 1833 * the messages for netif_carrier. 1834 */ 1835 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"); 1836 } else if (link_reporting == -1) { 1837 /* unable get link status using mii/ethtool */ 1838 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"); 1839 } 1840 } 1841 1842 /* check for initial state */ 1843 new_slave->link = BOND_LINK_NOCHANGE; 1844 if (bond->params.miimon) { 1845 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 1846 if (bond->params.updelay) { 1847 bond_set_slave_link_state(new_slave, 1848 BOND_LINK_BACK, 1849 BOND_SLAVE_NOTIFY_NOW); 1850 new_slave->delay = bond->params.updelay; 1851 } else { 1852 bond_set_slave_link_state(new_slave, 1853 BOND_LINK_UP, 1854 BOND_SLAVE_NOTIFY_NOW); 1855 } 1856 } else { 1857 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 1858 BOND_SLAVE_NOTIFY_NOW); 1859 } 1860 } else if (bond->params.arp_interval) { 1861 bond_set_slave_link_state(new_slave, 1862 (netif_carrier_ok(slave_dev) ? 1863 BOND_LINK_UP : BOND_LINK_DOWN), 1864 BOND_SLAVE_NOTIFY_NOW); 1865 } else { 1866 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 1867 BOND_SLAVE_NOTIFY_NOW); 1868 } 1869 1870 if (new_slave->link != BOND_LINK_DOWN) 1871 new_slave->last_link_up = jiffies; 1872 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 1873 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 1874 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 1875 1876 if (bond_uses_primary(bond) && bond->params.primary[0]) { 1877 /* if there is a primary slave, remember it */ 1878 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 1879 rcu_assign_pointer(bond->primary_slave, new_slave); 1880 bond->force_primary = true; 1881 } 1882 } 1883 1884 switch (BOND_MODE(bond)) { 1885 case BOND_MODE_ACTIVEBACKUP: 1886 bond_set_slave_inactive_flags(new_slave, 1887 BOND_SLAVE_NOTIFY_NOW); 1888 break; 1889 case BOND_MODE_8023AD: 1890 /* in 802.3ad mode, the internal mechanism 1891 * will activate the slaves in the selected 1892 * aggregator 1893 */ 1894 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1895 /* if this is the first slave */ 1896 if (!prev_slave) { 1897 SLAVE_AD_INFO(new_slave)->id = 1; 1898 /* Initialize AD with the number of times that the AD timer is called in 1 second 1899 * can be called only after the mac address of the bond is set 1900 */ 1901 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 1902 } else { 1903 SLAVE_AD_INFO(new_slave)->id = 1904 SLAVE_AD_INFO(prev_slave)->id + 1; 1905 } 1906 1907 bond_3ad_bind_slave(new_slave); 1908 break; 1909 case BOND_MODE_TLB: 1910 case BOND_MODE_ALB: 1911 bond_set_active_slave(new_slave); 1912 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 1913 break; 1914 default: 1915 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 1916 1917 /* always active in trunk mode */ 1918 bond_set_active_slave(new_slave); 1919 1920 /* In trunking mode there is little meaning to curr_active_slave 1921 * anyway (it holds no special properties of the bond device), 1922 * so we can change it without calling change_active_interface() 1923 */ 1924 if (!rcu_access_pointer(bond->curr_active_slave) && 1925 new_slave->link == BOND_LINK_UP) 1926 rcu_assign_pointer(bond->curr_active_slave, new_slave); 1927 1928 break; 1929 } /* switch(bond_mode) */ 1930 1931 #ifdef CONFIG_NET_POLL_CONTROLLER 1932 if (bond->dev->npinfo) { 1933 if (slave_enable_netpoll(new_slave)) { 1934 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 1935 res = -EBUSY; 1936 goto err_detach; 1937 } 1938 } 1939 #endif 1940 1941 if (!(bond_dev->features & NETIF_F_LRO)) 1942 dev_disable_lro(slave_dev); 1943 1944 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 1945 new_slave); 1946 if (res) { 1947 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 1948 goto err_detach; 1949 } 1950 1951 res = bond_master_upper_dev_link(bond, new_slave, extack); 1952 if (res) { 1953 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 1954 goto err_unregister; 1955 } 1956 1957 bond_lower_state_changed(new_slave); 1958 1959 res = bond_sysfs_slave_add(new_slave); 1960 if (res) { 1961 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 1962 goto err_upper_unlink; 1963 } 1964 1965 /* If the mode uses primary, then the following is handled by 1966 * bond_change_active_slave(). 1967 */ 1968 if (!bond_uses_primary(bond)) { 1969 /* set promiscuity level to new slave */ 1970 if (bond_dev->flags & IFF_PROMISC) { 1971 res = dev_set_promiscuity(slave_dev, 1); 1972 if (res) 1973 goto err_sysfs_del; 1974 } 1975 1976 /* set allmulti level to new slave */ 1977 if (bond_dev->flags & IFF_ALLMULTI) { 1978 res = dev_set_allmulti(slave_dev, 1); 1979 if (res) { 1980 if (bond_dev->flags & IFF_PROMISC) 1981 dev_set_promiscuity(slave_dev, -1); 1982 goto err_sysfs_del; 1983 } 1984 } 1985 1986 netif_addr_lock_bh(bond_dev); 1987 dev_mc_sync_multiple(slave_dev, bond_dev); 1988 dev_uc_sync_multiple(slave_dev, bond_dev); 1989 netif_addr_unlock_bh(bond_dev); 1990 1991 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1992 /* add lacpdu mc addr to mc list */ 1993 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 1994 1995 dev_mc_add(slave_dev, lacpdu_multicast); 1996 } 1997 } 1998 1999 bond->slave_cnt++; 2000 bond_compute_features(bond); 2001 bond_set_carrier(bond); 2002 2003 if (bond_uses_primary(bond)) { 2004 block_netpoll_tx(); 2005 bond_select_active_slave(bond); 2006 unblock_netpoll_tx(); 2007 } 2008 2009 if (bond_mode_can_use_xmit_hash(bond)) 2010 bond_update_slave_arr(bond, NULL); 2011 2012 2013 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2014 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2015 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2016 2017 /* enslave is successful */ 2018 bond_queue_slave_event(new_slave); 2019 return 0; 2020 2021 /* Undo stages on error */ 2022 err_sysfs_del: 2023 bond_sysfs_slave_del(new_slave); 2024 2025 err_upper_unlink: 2026 bond_upper_dev_unlink(bond, new_slave); 2027 2028 err_unregister: 2029 netdev_rx_handler_unregister(slave_dev); 2030 2031 err_detach: 2032 vlan_vids_del_by_dev(slave_dev, bond_dev); 2033 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2034 RCU_INIT_POINTER(bond->primary_slave, NULL); 2035 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2036 block_netpoll_tx(); 2037 bond_change_active_slave(bond, NULL); 2038 bond_select_active_slave(bond); 2039 unblock_netpoll_tx(); 2040 } 2041 /* either primary_slave or curr_active_slave might've changed */ 2042 synchronize_rcu(); 2043 slave_disable_netpoll(new_slave); 2044 2045 err_close: 2046 if (!netif_is_bond_master(slave_dev)) 2047 slave_dev->priv_flags &= ~IFF_BONDING; 2048 dev_close(slave_dev); 2049 2050 err_restore_mac: 2051 slave_dev->flags &= ~IFF_SLAVE; 2052 if (!bond->params.fail_over_mac || 2053 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2054 /* XXX TODO - fom follow mode needs to change master's 2055 * MAC if this slave's MAC is in use by the bond, or at 2056 * least print a warning. 2057 */ 2058 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2059 new_slave->dev->addr_len); 2060 ss.ss_family = slave_dev->type; 2061 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2062 } 2063 2064 err_restore_mtu: 2065 dev_set_mtu(slave_dev, new_slave->original_mtu); 2066 2067 err_free: 2068 kobject_put(&new_slave->kobj); 2069 2070 err_undo_flags: 2071 /* Enslave of first slave has failed and we need to fix master's mac */ 2072 if (!bond_has_slaves(bond)) { 2073 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2074 slave_dev->dev_addr)) 2075 eth_hw_addr_random(bond_dev); 2076 if (bond_dev->type != ARPHRD_ETHER) { 2077 dev_close(bond_dev); 2078 ether_setup(bond_dev); 2079 bond_dev->flags |= IFF_MASTER; 2080 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2081 } 2082 } 2083 2084 return res; 2085 } 2086 2087 /* Try to release the slave device <slave> from the bond device <master> 2088 * It is legal to access curr_active_slave without a lock because all the function 2089 * is RTNL-locked. If "all" is true it means that the function is being called 2090 * while destroying a bond interface and all slaves are being released. 2091 * 2092 * The rules for slave state should be: 2093 * for Active/Backup: 2094 * Active stays on all backups go down 2095 * for Bonded connections: 2096 * The first up interface should be left on and all others downed. 2097 */ 2098 static int __bond_release_one(struct net_device *bond_dev, 2099 struct net_device *slave_dev, 2100 bool all, bool unregister) 2101 { 2102 struct bonding *bond = netdev_priv(bond_dev); 2103 struct slave *slave, *oldcurrent; 2104 struct sockaddr_storage ss; 2105 int old_flags = bond_dev->flags; 2106 netdev_features_t old_features = bond_dev->features; 2107 2108 /* slave is not a slave or master is not master of this slave */ 2109 if (!(slave_dev->flags & IFF_SLAVE) || 2110 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2111 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2112 return -EINVAL; 2113 } 2114 2115 block_netpoll_tx(); 2116 2117 slave = bond_get_slave_by_dev(bond, slave_dev); 2118 if (!slave) { 2119 /* not a slave of this bond */ 2120 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2121 unblock_netpoll_tx(); 2122 return -EINVAL; 2123 } 2124 2125 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2126 2127 bond_sysfs_slave_del(slave); 2128 2129 /* recompute stats just before removing the slave */ 2130 bond_get_stats(bond->dev, &bond->bond_stats); 2131 2132 bond_upper_dev_unlink(bond, slave); 2133 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2134 * for this slave anymore. 2135 */ 2136 netdev_rx_handler_unregister(slave_dev); 2137 2138 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2139 bond_3ad_unbind_slave(slave); 2140 2141 if (bond_mode_can_use_xmit_hash(bond)) 2142 bond_update_slave_arr(bond, slave); 2143 2144 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2145 bond_is_active_slave(slave) ? "active" : "backup"); 2146 2147 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2148 2149 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2150 2151 if (!all && (!bond->params.fail_over_mac || 2152 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2153 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2154 bond_has_slaves(bond)) 2155 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", 2156 slave->perm_hwaddr); 2157 } 2158 2159 if (rtnl_dereference(bond->primary_slave) == slave) 2160 RCU_INIT_POINTER(bond->primary_slave, NULL); 2161 2162 if (oldcurrent == slave) 2163 bond_change_active_slave(bond, NULL); 2164 2165 if (bond_is_lb(bond)) { 2166 /* Must be called only after the slave has been 2167 * detached from the list and the curr_active_slave 2168 * has been cleared (if our_slave == old_current), 2169 * but before a new active slave is selected. 2170 */ 2171 bond_alb_deinit_slave(bond, slave); 2172 } 2173 2174 if (all) { 2175 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2176 } else if (oldcurrent == slave) { 2177 /* Note that we hold RTNL over this sequence, so there 2178 * is no concern that another slave add/remove event 2179 * will interfere. 2180 */ 2181 bond_select_active_slave(bond); 2182 } 2183 2184 if (!bond_has_slaves(bond)) { 2185 bond_set_carrier(bond); 2186 eth_hw_addr_random(bond_dev); 2187 } 2188 2189 unblock_netpoll_tx(); 2190 synchronize_rcu(); 2191 bond->slave_cnt--; 2192 2193 if (!bond_has_slaves(bond)) { 2194 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2195 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2196 } 2197 2198 bond_compute_features(bond); 2199 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2200 (old_features & NETIF_F_VLAN_CHALLENGED)) 2201 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2202 2203 vlan_vids_del_by_dev(slave_dev, bond_dev); 2204 2205 /* If the mode uses primary, then this case was handled above by 2206 * bond_change_active_slave(..., NULL) 2207 */ 2208 if (!bond_uses_primary(bond)) { 2209 /* unset promiscuity level from slave 2210 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2211 * of the IFF_PROMISC flag in the bond_dev, but we need the 2212 * value of that flag before that change, as that was the value 2213 * when this slave was attached, so we cache at the start of the 2214 * function and use it here. Same goes for ALLMULTI below 2215 */ 2216 if (old_flags & IFF_PROMISC) 2217 dev_set_promiscuity(slave_dev, -1); 2218 2219 /* unset allmulti level from slave */ 2220 if (old_flags & IFF_ALLMULTI) 2221 dev_set_allmulti(slave_dev, -1); 2222 2223 bond_hw_addr_flush(bond_dev, slave_dev); 2224 } 2225 2226 slave_disable_netpoll(slave); 2227 2228 /* close slave before restoring its mac address */ 2229 dev_close(slave_dev); 2230 2231 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2232 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2233 /* restore original ("permanent") mac address */ 2234 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2235 slave->dev->addr_len); 2236 ss.ss_family = slave_dev->type; 2237 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2238 } 2239 2240 if (unregister) 2241 __dev_set_mtu(slave_dev, slave->original_mtu); 2242 else 2243 dev_set_mtu(slave_dev, slave->original_mtu); 2244 2245 if (!netif_is_bond_master(slave_dev)) 2246 slave_dev->priv_flags &= ~IFF_BONDING; 2247 2248 kobject_put(&slave->kobj); 2249 2250 return 0; 2251 } 2252 2253 /* A wrapper used because of ndo_del_link */ 2254 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2255 { 2256 return __bond_release_one(bond_dev, slave_dev, false, false); 2257 } 2258 2259 /* First release a slave and then destroy the bond if no more slaves are left. 2260 * Must be under rtnl_lock when this function is called. 2261 */ 2262 static int bond_release_and_destroy(struct net_device *bond_dev, 2263 struct net_device *slave_dev) 2264 { 2265 struct bonding *bond = netdev_priv(bond_dev); 2266 int ret; 2267 2268 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2269 if (ret == 0 && !bond_has_slaves(bond) && 2270 bond_dev->reg_state != NETREG_UNREGISTERING) { 2271 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2272 netdev_info(bond_dev, "Destroying bond\n"); 2273 bond_remove_proc_entry(bond); 2274 unregister_netdevice(bond_dev); 2275 } 2276 return ret; 2277 } 2278 2279 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2280 { 2281 struct bonding *bond = netdev_priv(bond_dev); 2282 2283 bond_fill_ifbond(bond, info); 2284 } 2285 2286 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2287 { 2288 struct bonding *bond = netdev_priv(bond_dev); 2289 struct list_head *iter; 2290 int i = 0, res = -ENODEV; 2291 struct slave *slave; 2292 2293 bond_for_each_slave(bond, slave, iter) { 2294 if (i++ == (int)info->slave_id) { 2295 res = 0; 2296 bond_fill_ifslave(slave, info); 2297 break; 2298 } 2299 } 2300 2301 return res; 2302 } 2303 2304 /*-------------------------------- Monitoring -------------------------------*/ 2305 2306 /* called with rcu_read_lock() */ 2307 static int bond_miimon_inspect(struct bonding *bond) 2308 { 2309 int link_state, commit = 0; 2310 struct list_head *iter; 2311 struct slave *slave; 2312 bool ignore_updelay; 2313 2314 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2315 2316 bond_for_each_slave_rcu(bond, slave, iter) { 2317 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2318 2319 link_state = bond_check_dev_link(bond, slave->dev, 0); 2320 2321 switch (slave->link) { 2322 case BOND_LINK_UP: 2323 if (link_state) 2324 continue; 2325 2326 bond_propose_link_state(slave, BOND_LINK_FAIL); 2327 commit++; 2328 slave->delay = bond->params.downdelay; 2329 if (slave->delay) { 2330 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2331 (BOND_MODE(bond) == 2332 BOND_MODE_ACTIVEBACKUP) ? 2333 (bond_is_active_slave(slave) ? 2334 "active " : "backup ") : "", 2335 bond->params.downdelay * bond->params.miimon); 2336 } 2337 fallthrough; 2338 case BOND_LINK_FAIL: 2339 if (link_state) { 2340 /* recovered before downdelay expired */ 2341 bond_propose_link_state(slave, BOND_LINK_UP); 2342 slave->last_link_up = jiffies; 2343 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2344 (bond->params.downdelay - slave->delay) * 2345 bond->params.miimon); 2346 commit++; 2347 continue; 2348 } 2349 2350 if (slave->delay <= 0) { 2351 bond_propose_link_state(slave, BOND_LINK_DOWN); 2352 commit++; 2353 continue; 2354 } 2355 2356 slave->delay--; 2357 break; 2358 2359 case BOND_LINK_DOWN: 2360 if (!link_state) 2361 continue; 2362 2363 bond_propose_link_state(slave, BOND_LINK_BACK); 2364 commit++; 2365 slave->delay = bond->params.updelay; 2366 2367 if (slave->delay) { 2368 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2369 ignore_updelay ? 0 : 2370 bond->params.updelay * 2371 bond->params.miimon); 2372 } 2373 fallthrough; 2374 case BOND_LINK_BACK: 2375 if (!link_state) { 2376 bond_propose_link_state(slave, BOND_LINK_DOWN); 2377 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2378 (bond->params.updelay - slave->delay) * 2379 bond->params.miimon); 2380 commit++; 2381 continue; 2382 } 2383 2384 if (ignore_updelay) 2385 slave->delay = 0; 2386 2387 if (slave->delay <= 0) { 2388 bond_propose_link_state(slave, BOND_LINK_UP); 2389 commit++; 2390 ignore_updelay = false; 2391 continue; 2392 } 2393 2394 slave->delay--; 2395 break; 2396 } 2397 } 2398 2399 return commit; 2400 } 2401 2402 static void bond_miimon_link_change(struct bonding *bond, 2403 struct slave *slave, 2404 char link) 2405 { 2406 switch (BOND_MODE(bond)) { 2407 case BOND_MODE_8023AD: 2408 bond_3ad_handle_link_change(slave, link); 2409 break; 2410 case BOND_MODE_TLB: 2411 case BOND_MODE_ALB: 2412 bond_alb_handle_link_change(bond, slave, link); 2413 break; 2414 case BOND_MODE_XOR: 2415 bond_update_slave_arr(bond, NULL); 2416 break; 2417 } 2418 } 2419 2420 static void bond_miimon_commit(struct bonding *bond) 2421 { 2422 struct list_head *iter; 2423 struct slave *slave, *primary; 2424 2425 bond_for_each_slave(bond, slave, iter) { 2426 switch (slave->link_new_state) { 2427 case BOND_LINK_NOCHANGE: 2428 /* For 802.3ad mode, check current slave speed and 2429 * duplex again in case its port was disabled after 2430 * invalid speed/duplex reporting but recovered before 2431 * link monitoring could make a decision on the actual 2432 * link status 2433 */ 2434 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2435 slave->link == BOND_LINK_UP) 2436 bond_3ad_adapter_speed_duplex_changed(slave); 2437 continue; 2438 2439 case BOND_LINK_UP: 2440 if (bond_update_speed_duplex(slave) && 2441 bond_needs_speed_duplex(bond)) { 2442 slave->link = BOND_LINK_DOWN; 2443 if (net_ratelimit()) 2444 slave_warn(bond->dev, slave->dev, 2445 "failed to get link speed/duplex\n"); 2446 continue; 2447 } 2448 bond_set_slave_link_state(slave, BOND_LINK_UP, 2449 BOND_SLAVE_NOTIFY_NOW); 2450 slave->last_link_up = jiffies; 2451 2452 primary = rtnl_dereference(bond->primary_slave); 2453 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2454 /* prevent it from being the active one */ 2455 bond_set_backup_slave(slave); 2456 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2457 /* make it immediately active */ 2458 bond_set_active_slave(slave); 2459 } 2460 2461 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2462 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2463 slave->duplex ? "full" : "half"); 2464 2465 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2466 2467 if (!bond->curr_active_slave || slave == primary) 2468 goto do_failover; 2469 2470 continue; 2471 2472 case BOND_LINK_DOWN: 2473 if (slave->link_failure_count < UINT_MAX) 2474 slave->link_failure_count++; 2475 2476 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2477 BOND_SLAVE_NOTIFY_NOW); 2478 2479 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2480 BOND_MODE(bond) == BOND_MODE_8023AD) 2481 bond_set_slave_inactive_flags(slave, 2482 BOND_SLAVE_NOTIFY_NOW); 2483 2484 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2485 2486 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2487 2488 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2489 goto do_failover; 2490 2491 continue; 2492 2493 default: 2494 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2495 slave->link_new_state); 2496 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2497 2498 continue; 2499 } 2500 2501 do_failover: 2502 block_netpoll_tx(); 2503 bond_select_active_slave(bond); 2504 unblock_netpoll_tx(); 2505 } 2506 2507 bond_set_carrier(bond); 2508 } 2509 2510 /* bond_mii_monitor 2511 * 2512 * Really a wrapper that splits the mii monitor into two phases: an 2513 * inspection, then (if inspection indicates something needs to be done) 2514 * an acquisition of appropriate locks followed by a commit phase to 2515 * implement whatever link state changes are indicated. 2516 */ 2517 static void bond_mii_monitor(struct work_struct *work) 2518 { 2519 struct bonding *bond = container_of(work, struct bonding, 2520 mii_work.work); 2521 bool should_notify_peers = false; 2522 bool commit; 2523 unsigned long delay; 2524 struct slave *slave; 2525 struct list_head *iter; 2526 2527 delay = msecs_to_jiffies(bond->params.miimon); 2528 2529 if (!bond_has_slaves(bond)) 2530 goto re_arm; 2531 2532 rcu_read_lock(); 2533 should_notify_peers = bond_should_notify_peers(bond); 2534 commit = !!bond_miimon_inspect(bond); 2535 if (bond->send_peer_notif) { 2536 rcu_read_unlock(); 2537 if (rtnl_trylock()) { 2538 bond->send_peer_notif--; 2539 rtnl_unlock(); 2540 } 2541 } else { 2542 rcu_read_unlock(); 2543 } 2544 2545 if (commit) { 2546 /* Race avoidance with bond_close cancel of workqueue */ 2547 if (!rtnl_trylock()) { 2548 delay = 1; 2549 should_notify_peers = false; 2550 goto re_arm; 2551 } 2552 2553 bond_for_each_slave(bond, slave, iter) { 2554 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2555 } 2556 bond_miimon_commit(bond); 2557 2558 rtnl_unlock(); /* might sleep, hold no other locks */ 2559 } 2560 2561 re_arm: 2562 if (bond->params.miimon) 2563 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2564 2565 if (should_notify_peers) { 2566 if (!rtnl_trylock()) 2567 return; 2568 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2569 rtnl_unlock(); 2570 } 2571 } 2572 2573 static int bond_upper_dev_walk(struct net_device *upper, 2574 struct netdev_nested_priv *priv) 2575 { 2576 __be32 ip = *(__be32 *)priv->data; 2577 2578 return ip == bond_confirm_addr(upper, 0, ip); 2579 } 2580 2581 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2582 { 2583 struct netdev_nested_priv priv = { 2584 .data = (void *)&ip, 2585 }; 2586 bool ret = false; 2587 2588 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2589 return true; 2590 2591 rcu_read_lock(); 2592 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2593 ret = true; 2594 rcu_read_unlock(); 2595 2596 return ret; 2597 } 2598 2599 /* We go to the (large) trouble of VLAN tagging ARP frames because 2600 * switches in VLAN mode (especially if ports are configured as 2601 * "native" to a VLAN) might not pass non-tagged frames. 2602 */ 2603 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2604 __be32 src_ip, struct bond_vlan_tag *tags) 2605 { 2606 struct sk_buff *skb; 2607 struct bond_vlan_tag *outer_tag = tags; 2608 struct net_device *slave_dev = slave->dev; 2609 struct net_device *bond_dev = slave->bond->dev; 2610 2611 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2612 arp_op, &dest_ip, &src_ip); 2613 2614 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2615 NULL, slave_dev->dev_addr, NULL); 2616 2617 if (!skb) { 2618 net_err_ratelimited("ARP packet allocation failed\n"); 2619 return; 2620 } 2621 2622 if (!tags || tags->vlan_proto == VLAN_N_VID) 2623 goto xmit; 2624 2625 tags++; 2626 2627 /* Go through all the tags backwards and add them to the packet */ 2628 while (tags->vlan_proto != VLAN_N_VID) { 2629 if (!tags->vlan_id) { 2630 tags++; 2631 continue; 2632 } 2633 2634 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2635 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2636 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2637 tags->vlan_id); 2638 if (!skb) { 2639 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2640 return; 2641 } 2642 2643 tags++; 2644 } 2645 /* Set the outer tag */ 2646 if (outer_tag->vlan_id) { 2647 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2648 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2649 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2650 outer_tag->vlan_id); 2651 } 2652 2653 xmit: 2654 arp_xmit(skb); 2655 } 2656 2657 /* Validate the device path between the @start_dev and the @end_dev. 2658 * The path is valid if the @end_dev is reachable through device 2659 * stacking. 2660 * When the path is validated, collect any vlan information in the 2661 * path. 2662 */ 2663 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2664 struct net_device *end_dev, 2665 int level) 2666 { 2667 struct bond_vlan_tag *tags; 2668 struct net_device *upper; 2669 struct list_head *iter; 2670 2671 if (start_dev == end_dev) { 2672 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2673 if (!tags) 2674 return ERR_PTR(-ENOMEM); 2675 tags[level].vlan_proto = VLAN_N_VID; 2676 return tags; 2677 } 2678 2679 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2680 tags = bond_verify_device_path(upper, end_dev, level + 1); 2681 if (IS_ERR_OR_NULL(tags)) { 2682 if (IS_ERR(tags)) 2683 return tags; 2684 continue; 2685 } 2686 if (is_vlan_dev(upper)) { 2687 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2688 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2689 } 2690 2691 return tags; 2692 } 2693 2694 return NULL; 2695 } 2696 2697 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2698 { 2699 struct rtable *rt; 2700 struct bond_vlan_tag *tags; 2701 __be32 *targets = bond->params.arp_targets, addr; 2702 int i; 2703 2704 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2705 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2706 __func__, &targets[i]); 2707 tags = NULL; 2708 2709 /* Find out through which dev should the packet go */ 2710 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2711 RTO_ONLINK, 0); 2712 if (IS_ERR(rt)) { 2713 /* there's no route to target - try to send arp 2714 * probe to generate any traffic (arp_validate=0) 2715 */ 2716 if (bond->params.arp_validate) 2717 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2718 bond->dev->name, 2719 &targets[i]); 2720 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2721 0, tags); 2722 continue; 2723 } 2724 2725 /* bond device itself */ 2726 if (rt->dst.dev == bond->dev) 2727 goto found; 2728 2729 rcu_read_lock(); 2730 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2731 rcu_read_unlock(); 2732 2733 if (!IS_ERR_OR_NULL(tags)) 2734 goto found; 2735 2736 /* Not our device - skip */ 2737 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2738 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2739 2740 ip_rt_put(rt); 2741 continue; 2742 2743 found: 2744 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2745 ip_rt_put(rt); 2746 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2747 kfree(tags); 2748 } 2749 } 2750 2751 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2752 { 2753 int i; 2754 2755 if (!sip || !bond_has_this_ip(bond, tip)) { 2756 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2757 __func__, &sip, &tip); 2758 return; 2759 } 2760 2761 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2762 if (i == -1) { 2763 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2764 __func__, &sip); 2765 return; 2766 } 2767 slave->last_rx = jiffies; 2768 slave->target_last_arp_rx[i] = jiffies; 2769 } 2770 2771 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 2772 struct slave *slave) 2773 { 2774 struct arphdr *arp = (struct arphdr *)skb->data; 2775 struct slave *curr_active_slave, *curr_arp_slave; 2776 unsigned char *arp_ptr; 2777 __be32 sip, tip; 2778 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 2779 unsigned int alen; 2780 2781 if (!slave_do_arp_validate(bond, slave)) { 2782 if ((slave_do_arp_validate_only(bond) && is_arp) || 2783 !slave_do_arp_validate_only(bond)) 2784 slave->last_rx = jiffies; 2785 return RX_HANDLER_ANOTHER; 2786 } else if (!is_arp) { 2787 return RX_HANDLER_ANOTHER; 2788 } 2789 2790 alen = arp_hdr_len(bond->dev); 2791 2792 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 2793 __func__, skb->dev->name); 2794 2795 if (alen > skb_headlen(skb)) { 2796 arp = kmalloc(alen, GFP_ATOMIC); 2797 if (!arp) 2798 goto out_unlock; 2799 if (skb_copy_bits(skb, 0, arp, alen) < 0) 2800 goto out_unlock; 2801 } 2802 2803 if (arp->ar_hln != bond->dev->addr_len || 2804 skb->pkt_type == PACKET_OTHERHOST || 2805 skb->pkt_type == PACKET_LOOPBACK || 2806 arp->ar_hrd != htons(ARPHRD_ETHER) || 2807 arp->ar_pro != htons(ETH_P_IP) || 2808 arp->ar_pln != 4) 2809 goto out_unlock; 2810 2811 arp_ptr = (unsigned char *)(arp + 1); 2812 arp_ptr += bond->dev->addr_len; 2813 memcpy(&sip, arp_ptr, 4); 2814 arp_ptr += 4 + bond->dev->addr_len; 2815 memcpy(&tip, arp_ptr, 4); 2816 2817 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 2818 __func__, slave->dev->name, bond_slave_state(slave), 2819 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 2820 &sip, &tip); 2821 2822 curr_active_slave = rcu_dereference(bond->curr_active_slave); 2823 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 2824 2825 /* We 'trust' the received ARP enough to validate it if: 2826 * 2827 * (a) the slave receiving the ARP is active (which includes the 2828 * current ARP slave, if any), or 2829 * 2830 * (b) the receiving slave isn't active, but there is a currently 2831 * active slave and it received valid arp reply(s) after it became 2832 * the currently active slave, or 2833 * 2834 * (c) there is an ARP slave that sent an ARP during the prior ARP 2835 * interval, and we receive an ARP reply on any slave. We accept 2836 * these because switch FDB update delays may deliver the ARP 2837 * reply to a slave other than the sender of the ARP request. 2838 * 2839 * Note: for (b), backup slaves are receiving the broadcast ARP 2840 * request, not a reply. This request passes from the sending 2841 * slave through the L2 switch(es) to the receiving slave. Since 2842 * this is checking the request, sip/tip are swapped for 2843 * validation. 2844 * 2845 * This is done to avoid endless looping when we can't reach the 2846 * arp_ip_target and fool ourselves with our own arp requests. 2847 */ 2848 if (bond_is_active_slave(slave)) 2849 bond_validate_arp(bond, slave, sip, tip); 2850 else if (curr_active_slave && 2851 time_after(slave_last_rx(bond, curr_active_slave), 2852 curr_active_slave->last_link_up)) 2853 bond_validate_arp(bond, slave, tip, sip); 2854 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 2855 bond_time_in_interval(bond, 2856 dev_trans_start(curr_arp_slave->dev), 1)) 2857 bond_validate_arp(bond, slave, sip, tip); 2858 2859 out_unlock: 2860 if (arp != (struct arphdr *)skb->data) 2861 kfree(arp); 2862 return RX_HANDLER_ANOTHER; 2863 } 2864 2865 /* function to verify if we're in the arp_interval timeslice, returns true if 2866 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 2867 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 2868 */ 2869 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 2870 int mod) 2871 { 2872 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 2873 2874 return time_in_range(jiffies, 2875 last_act - delta_in_ticks, 2876 last_act + mod * delta_in_ticks + delta_in_ticks/2); 2877 } 2878 2879 /* This function is called regularly to monitor each slave's link 2880 * ensuring that traffic is being sent and received when arp monitoring 2881 * is used in load-balancing mode. if the adapter has been dormant, then an 2882 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 2883 * arp monitoring in active backup mode. 2884 */ 2885 static void bond_loadbalance_arp_mon(struct bonding *bond) 2886 { 2887 struct slave *slave, *oldcurrent; 2888 struct list_head *iter; 2889 int do_failover = 0, slave_state_changed = 0; 2890 2891 if (!bond_has_slaves(bond)) 2892 goto re_arm; 2893 2894 rcu_read_lock(); 2895 2896 oldcurrent = rcu_dereference(bond->curr_active_slave); 2897 /* see if any of the previous devices are up now (i.e. they have 2898 * xmt and rcv traffic). the curr_active_slave does not come into 2899 * the picture unless it is null. also, slave->last_link_up is not 2900 * needed here because we send an arp on each slave and give a slave 2901 * as long as it needs to get the tx/rx within the delta. 2902 * TODO: what about up/down delay in arp mode? it wasn't here before 2903 * so it can wait 2904 */ 2905 bond_for_each_slave_rcu(bond, slave, iter) { 2906 unsigned long trans_start = dev_trans_start(slave->dev); 2907 2908 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2909 2910 if (slave->link != BOND_LINK_UP) { 2911 if (bond_time_in_interval(bond, trans_start, 1) && 2912 bond_time_in_interval(bond, slave->last_rx, 1)) { 2913 2914 bond_propose_link_state(slave, BOND_LINK_UP); 2915 slave_state_changed = 1; 2916 2917 /* primary_slave has no meaning in round-robin 2918 * mode. the window of a slave being up and 2919 * curr_active_slave being null after enslaving 2920 * is closed. 2921 */ 2922 if (!oldcurrent) { 2923 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 2924 do_failover = 1; 2925 } else { 2926 slave_info(bond->dev, slave->dev, "interface is now up\n"); 2927 } 2928 } 2929 } else { 2930 /* slave->link == BOND_LINK_UP */ 2931 2932 /* not all switches will respond to an arp request 2933 * when the source ip is 0, so don't take the link down 2934 * if we don't know our ip yet 2935 */ 2936 if (!bond_time_in_interval(bond, trans_start, 2) || 2937 !bond_time_in_interval(bond, slave->last_rx, 2)) { 2938 2939 bond_propose_link_state(slave, BOND_LINK_DOWN); 2940 slave_state_changed = 1; 2941 2942 if (slave->link_failure_count < UINT_MAX) 2943 slave->link_failure_count++; 2944 2945 slave_info(bond->dev, slave->dev, "interface is now down\n"); 2946 2947 if (slave == oldcurrent) 2948 do_failover = 1; 2949 } 2950 } 2951 2952 /* note: if switch is in round-robin mode, all links 2953 * must tx arp to ensure all links rx an arp - otherwise 2954 * links may oscillate or not come up at all; if switch is 2955 * in something like xor mode, there is nothing we can 2956 * do - all replies will be rx'ed on same link causing slaves 2957 * to be unstable during low/no traffic periods 2958 */ 2959 if (bond_slave_is_up(slave)) 2960 bond_arp_send_all(bond, slave); 2961 } 2962 2963 rcu_read_unlock(); 2964 2965 if (do_failover || slave_state_changed) { 2966 if (!rtnl_trylock()) 2967 goto re_arm; 2968 2969 bond_for_each_slave(bond, slave, iter) { 2970 if (slave->link_new_state != BOND_LINK_NOCHANGE) 2971 slave->link = slave->link_new_state; 2972 } 2973 2974 if (slave_state_changed) { 2975 bond_slave_state_change(bond); 2976 if (BOND_MODE(bond) == BOND_MODE_XOR) 2977 bond_update_slave_arr(bond, NULL); 2978 } 2979 if (do_failover) { 2980 block_netpoll_tx(); 2981 bond_select_active_slave(bond); 2982 unblock_netpoll_tx(); 2983 } 2984 rtnl_unlock(); 2985 } 2986 2987 re_arm: 2988 if (bond->params.arp_interval) 2989 queue_delayed_work(bond->wq, &bond->arp_work, 2990 msecs_to_jiffies(bond->params.arp_interval)); 2991 } 2992 2993 /* Called to inspect slaves for active-backup mode ARP monitor link state 2994 * changes. Sets proposed link state in slaves to specify what action 2995 * should take place for the slave. Returns 0 if no changes are found, >0 2996 * if changes to link states must be committed. 2997 * 2998 * Called with rcu_read_lock held. 2999 */ 3000 static int bond_ab_arp_inspect(struct bonding *bond) 3001 { 3002 unsigned long trans_start, last_rx; 3003 struct list_head *iter; 3004 struct slave *slave; 3005 int commit = 0; 3006 3007 bond_for_each_slave_rcu(bond, slave, iter) { 3008 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3009 last_rx = slave_last_rx(bond, slave); 3010 3011 if (slave->link != BOND_LINK_UP) { 3012 if (bond_time_in_interval(bond, last_rx, 1)) { 3013 bond_propose_link_state(slave, BOND_LINK_UP); 3014 commit++; 3015 } else if (slave->link == BOND_LINK_BACK) { 3016 bond_propose_link_state(slave, BOND_LINK_FAIL); 3017 commit++; 3018 } 3019 continue; 3020 } 3021 3022 /* Give slaves 2*delta after being enslaved or made 3023 * active. This avoids bouncing, as the last receive 3024 * times need a full ARP monitor cycle to be updated. 3025 */ 3026 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3027 continue; 3028 3029 /* Backup slave is down if: 3030 * - No current_arp_slave AND 3031 * - more than 3*delta since last receive AND 3032 * - the bond has an IP address 3033 * 3034 * Note: a non-null current_arp_slave indicates 3035 * the curr_active_slave went down and we are 3036 * searching for a new one; under this condition 3037 * we only take the curr_active_slave down - this 3038 * gives each slave a chance to tx/rx traffic 3039 * before being taken out 3040 */ 3041 if (!bond_is_active_slave(slave) && 3042 !rcu_access_pointer(bond->current_arp_slave) && 3043 !bond_time_in_interval(bond, last_rx, 3)) { 3044 bond_propose_link_state(slave, BOND_LINK_DOWN); 3045 commit++; 3046 } 3047 3048 /* Active slave is down if: 3049 * - more than 2*delta since transmitting OR 3050 * - (more than 2*delta since receive AND 3051 * the bond has an IP address) 3052 */ 3053 trans_start = dev_trans_start(slave->dev); 3054 if (bond_is_active_slave(slave) && 3055 (!bond_time_in_interval(bond, trans_start, 2) || 3056 !bond_time_in_interval(bond, last_rx, 2))) { 3057 bond_propose_link_state(slave, BOND_LINK_DOWN); 3058 commit++; 3059 } 3060 } 3061 3062 return commit; 3063 } 3064 3065 /* Called to commit link state changes noted by inspection step of 3066 * active-backup mode ARP monitor. 3067 * 3068 * Called with RTNL hold. 3069 */ 3070 static void bond_ab_arp_commit(struct bonding *bond) 3071 { 3072 unsigned long trans_start; 3073 struct list_head *iter; 3074 struct slave *slave; 3075 3076 bond_for_each_slave(bond, slave, iter) { 3077 switch (slave->link_new_state) { 3078 case BOND_LINK_NOCHANGE: 3079 continue; 3080 3081 case BOND_LINK_UP: 3082 trans_start = dev_trans_start(slave->dev); 3083 if (rtnl_dereference(bond->curr_active_slave) != slave || 3084 (!rtnl_dereference(bond->curr_active_slave) && 3085 bond_time_in_interval(bond, trans_start, 1))) { 3086 struct slave *current_arp_slave; 3087 3088 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3089 bond_set_slave_link_state(slave, BOND_LINK_UP, 3090 BOND_SLAVE_NOTIFY_NOW); 3091 if (current_arp_slave) { 3092 bond_set_slave_inactive_flags( 3093 current_arp_slave, 3094 BOND_SLAVE_NOTIFY_NOW); 3095 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3096 } 3097 3098 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3099 3100 if (!rtnl_dereference(bond->curr_active_slave) || 3101 slave == rtnl_dereference(bond->primary_slave)) 3102 goto do_failover; 3103 3104 } 3105 3106 continue; 3107 3108 case BOND_LINK_DOWN: 3109 if (slave->link_failure_count < UINT_MAX) 3110 slave->link_failure_count++; 3111 3112 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3113 BOND_SLAVE_NOTIFY_NOW); 3114 bond_set_slave_inactive_flags(slave, 3115 BOND_SLAVE_NOTIFY_NOW); 3116 3117 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3118 3119 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3120 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3121 goto do_failover; 3122 } 3123 3124 continue; 3125 3126 case BOND_LINK_FAIL: 3127 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3128 BOND_SLAVE_NOTIFY_NOW); 3129 bond_set_slave_inactive_flags(slave, 3130 BOND_SLAVE_NOTIFY_NOW); 3131 3132 /* A slave has just been enslaved and has become 3133 * the current active slave. 3134 */ 3135 if (rtnl_dereference(bond->curr_active_slave)) 3136 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3137 continue; 3138 3139 default: 3140 slave_err(bond->dev, slave->dev, 3141 "impossible: link_new_state %d on slave\n", 3142 slave->link_new_state); 3143 continue; 3144 } 3145 3146 do_failover: 3147 block_netpoll_tx(); 3148 bond_select_active_slave(bond); 3149 unblock_netpoll_tx(); 3150 } 3151 3152 bond_set_carrier(bond); 3153 } 3154 3155 /* Send ARP probes for active-backup mode ARP monitor. 3156 * 3157 * Called with rcu_read_lock held. 3158 */ 3159 static bool bond_ab_arp_probe(struct bonding *bond) 3160 { 3161 struct slave *slave, *before = NULL, *new_slave = NULL, 3162 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3163 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3164 struct list_head *iter; 3165 bool found = false; 3166 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3167 3168 if (curr_arp_slave && curr_active_slave) 3169 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3170 curr_arp_slave->dev->name, 3171 curr_active_slave->dev->name); 3172 3173 if (curr_active_slave) { 3174 bond_arp_send_all(bond, curr_active_slave); 3175 return should_notify_rtnl; 3176 } 3177 3178 /* if we don't have a curr_active_slave, search for the next available 3179 * backup slave from the current_arp_slave and make it the candidate 3180 * for becoming the curr_active_slave 3181 */ 3182 3183 if (!curr_arp_slave) { 3184 curr_arp_slave = bond_first_slave_rcu(bond); 3185 if (!curr_arp_slave) 3186 return should_notify_rtnl; 3187 } 3188 3189 bond_for_each_slave_rcu(bond, slave, iter) { 3190 if (!found && !before && bond_slave_is_up(slave)) 3191 before = slave; 3192 3193 if (found && !new_slave && bond_slave_is_up(slave)) 3194 new_slave = slave; 3195 /* if the link state is up at this point, we 3196 * mark it down - this can happen if we have 3197 * simultaneous link failures and 3198 * reselect_active_interface doesn't make this 3199 * one the current slave so it is still marked 3200 * up when it is actually down 3201 */ 3202 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3203 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3204 BOND_SLAVE_NOTIFY_LATER); 3205 if (slave->link_failure_count < UINT_MAX) 3206 slave->link_failure_count++; 3207 3208 bond_set_slave_inactive_flags(slave, 3209 BOND_SLAVE_NOTIFY_LATER); 3210 3211 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3212 } 3213 if (slave == curr_arp_slave) 3214 found = true; 3215 } 3216 3217 if (!new_slave && before) 3218 new_slave = before; 3219 3220 if (!new_slave) 3221 goto check_state; 3222 3223 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3224 BOND_SLAVE_NOTIFY_LATER); 3225 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3226 bond_arp_send_all(bond, new_slave); 3227 new_slave->last_link_up = jiffies; 3228 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3229 3230 check_state: 3231 bond_for_each_slave_rcu(bond, slave, iter) { 3232 if (slave->should_notify || slave->should_notify_link) { 3233 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3234 break; 3235 } 3236 } 3237 return should_notify_rtnl; 3238 } 3239 3240 static void bond_activebackup_arp_mon(struct bonding *bond) 3241 { 3242 bool should_notify_peers = false; 3243 bool should_notify_rtnl = false; 3244 int delta_in_ticks; 3245 3246 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3247 3248 if (!bond_has_slaves(bond)) 3249 goto re_arm; 3250 3251 rcu_read_lock(); 3252 3253 should_notify_peers = bond_should_notify_peers(bond); 3254 3255 if (bond_ab_arp_inspect(bond)) { 3256 rcu_read_unlock(); 3257 3258 /* Race avoidance with bond_close flush of workqueue */ 3259 if (!rtnl_trylock()) { 3260 delta_in_ticks = 1; 3261 should_notify_peers = false; 3262 goto re_arm; 3263 } 3264 3265 bond_ab_arp_commit(bond); 3266 3267 rtnl_unlock(); 3268 rcu_read_lock(); 3269 } 3270 3271 should_notify_rtnl = bond_ab_arp_probe(bond); 3272 rcu_read_unlock(); 3273 3274 re_arm: 3275 if (bond->params.arp_interval) 3276 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3277 3278 if (should_notify_peers || should_notify_rtnl) { 3279 if (!rtnl_trylock()) 3280 return; 3281 3282 if (should_notify_peers) 3283 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3284 bond->dev); 3285 if (should_notify_rtnl) { 3286 bond_slave_state_notify(bond); 3287 bond_slave_link_notify(bond); 3288 } 3289 3290 rtnl_unlock(); 3291 } 3292 } 3293 3294 static void bond_arp_monitor(struct work_struct *work) 3295 { 3296 struct bonding *bond = container_of(work, struct bonding, 3297 arp_work.work); 3298 3299 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3300 bond_activebackup_arp_mon(bond); 3301 else 3302 bond_loadbalance_arp_mon(bond); 3303 } 3304 3305 /*-------------------------- netdev event handling --------------------------*/ 3306 3307 /* Change device name */ 3308 static int bond_event_changename(struct bonding *bond) 3309 { 3310 bond_remove_proc_entry(bond); 3311 bond_create_proc_entry(bond); 3312 3313 bond_debug_reregister(bond); 3314 3315 return NOTIFY_DONE; 3316 } 3317 3318 static int bond_master_netdev_event(unsigned long event, 3319 struct net_device *bond_dev) 3320 { 3321 struct bonding *event_bond = netdev_priv(bond_dev); 3322 3323 netdev_dbg(bond_dev, "%s called\n", __func__); 3324 3325 switch (event) { 3326 case NETDEV_CHANGENAME: 3327 return bond_event_changename(event_bond); 3328 case NETDEV_UNREGISTER: 3329 bond_remove_proc_entry(event_bond); 3330 break; 3331 case NETDEV_REGISTER: 3332 bond_create_proc_entry(event_bond); 3333 break; 3334 default: 3335 break; 3336 } 3337 3338 return NOTIFY_DONE; 3339 } 3340 3341 static int bond_slave_netdev_event(unsigned long event, 3342 struct net_device *slave_dev) 3343 { 3344 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3345 struct bonding *bond; 3346 struct net_device *bond_dev; 3347 3348 /* A netdev event can be generated while enslaving a device 3349 * before netdev_rx_handler_register is called in which case 3350 * slave will be NULL 3351 */ 3352 if (!slave) { 3353 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3354 return NOTIFY_DONE; 3355 } 3356 3357 bond_dev = slave->bond->dev; 3358 bond = slave->bond; 3359 primary = rtnl_dereference(bond->primary_slave); 3360 3361 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3362 3363 switch (event) { 3364 case NETDEV_UNREGISTER: 3365 if (bond_dev->type != ARPHRD_ETHER) 3366 bond_release_and_destroy(bond_dev, slave_dev); 3367 else 3368 __bond_release_one(bond_dev, slave_dev, false, true); 3369 break; 3370 case NETDEV_UP: 3371 case NETDEV_CHANGE: 3372 /* For 802.3ad mode only: 3373 * Getting invalid Speed/Duplex values here will put slave 3374 * in weird state. Mark it as link-fail if the link was 3375 * previously up or link-down if it hasn't yet come up, and 3376 * let link-monitoring (miimon) set it right when correct 3377 * speeds/duplex are available. 3378 */ 3379 if (bond_update_speed_duplex(slave) && 3380 BOND_MODE(bond) == BOND_MODE_8023AD) { 3381 if (slave->last_link_up) 3382 slave->link = BOND_LINK_FAIL; 3383 else 3384 slave->link = BOND_LINK_DOWN; 3385 } 3386 3387 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3388 bond_3ad_adapter_speed_duplex_changed(slave); 3389 fallthrough; 3390 case NETDEV_DOWN: 3391 /* Refresh slave-array if applicable! 3392 * If the setup does not use miimon or arpmon (mode-specific!), 3393 * then these events will not cause the slave-array to be 3394 * refreshed. This will cause xmit to use a slave that is not 3395 * usable. Avoid such situation by refeshing the array at these 3396 * events. If these (miimon/arpmon) parameters are configured 3397 * then array gets refreshed twice and that should be fine! 3398 */ 3399 if (bond_mode_can_use_xmit_hash(bond)) 3400 bond_update_slave_arr(bond, NULL); 3401 break; 3402 case NETDEV_CHANGEMTU: 3403 /* TODO: Should slaves be allowed to 3404 * independently alter their MTU? For 3405 * an active-backup bond, slaves need 3406 * not be the same type of device, so 3407 * MTUs may vary. For other modes, 3408 * slaves arguably should have the 3409 * same MTUs. To do this, we'd need to 3410 * take over the slave's change_mtu 3411 * function for the duration of their 3412 * servitude. 3413 */ 3414 break; 3415 case NETDEV_CHANGENAME: 3416 /* we don't care if we don't have primary set */ 3417 if (!bond_uses_primary(bond) || 3418 !bond->params.primary[0]) 3419 break; 3420 3421 if (slave == primary) { 3422 /* slave's name changed - he's no longer primary */ 3423 RCU_INIT_POINTER(bond->primary_slave, NULL); 3424 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3425 /* we have a new primary slave */ 3426 rcu_assign_pointer(bond->primary_slave, slave); 3427 } else { /* we didn't change primary - exit */ 3428 break; 3429 } 3430 3431 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3432 primary ? slave_dev->name : "none"); 3433 3434 block_netpoll_tx(); 3435 bond_select_active_slave(bond); 3436 unblock_netpoll_tx(); 3437 break; 3438 case NETDEV_FEAT_CHANGE: 3439 bond_compute_features(bond); 3440 break; 3441 case NETDEV_RESEND_IGMP: 3442 /* Propagate to master device */ 3443 call_netdevice_notifiers(event, slave->bond->dev); 3444 break; 3445 default: 3446 break; 3447 } 3448 3449 return NOTIFY_DONE; 3450 } 3451 3452 /* bond_netdev_event: handle netdev notifier chain events. 3453 * 3454 * This function receives events for the netdev chain. The caller (an 3455 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3456 * locks for us to safely manipulate the slave devices (RTNL lock, 3457 * dev_probe_lock). 3458 */ 3459 static int bond_netdev_event(struct notifier_block *this, 3460 unsigned long event, void *ptr) 3461 { 3462 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3463 3464 netdev_dbg(event_dev, "%s received %s\n", 3465 __func__, netdev_cmd_to_name(event)); 3466 3467 if (!(event_dev->priv_flags & IFF_BONDING)) 3468 return NOTIFY_DONE; 3469 3470 if (event_dev->flags & IFF_MASTER) { 3471 int ret; 3472 3473 ret = bond_master_netdev_event(event, event_dev); 3474 if (ret != NOTIFY_DONE) 3475 return ret; 3476 } 3477 3478 if (event_dev->flags & IFF_SLAVE) 3479 return bond_slave_netdev_event(event, event_dev); 3480 3481 return NOTIFY_DONE; 3482 } 3483 3484 static struct notifier_block bond_netdev_notifier = { 3485 .notifier_call = bond_netdev_event, 3486 }; 3487 3488 /*---------------------------- Hashing Policies -----------------------------*/ 3489 3490 /* L2 hash helper */ 3491 static inline u32 bond_eth_hash(struct sk_buff *skb) 3492 { 3493 struct ethhdr *ep, hdr_tmp; 3494 3495 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp); 3496 if (ep) 3497 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto; 3498 return 0; 3499 } 3500 3501 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, 3502 int *noff, int *proto, bool l34) 3503 { 3504 const struct ipv6hdr *iph6; 3505 const struct iphdr *iph; 3506 3507 if (skb->protocol == htons(ETH_P_IP)) { 3508 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph)))) 3509 return false; 3510 iph = (const struct iphdr *)(skb->data + *noff); 3511 iph_to_flow_copy_v4addrs(fk, iph); 3512 *noff += iph->ihl << 2; 3513 if (!ip_is_fragment(iph)) 3514 *proto = iph->protocol; 3515 } else if (skb->protocol == htons(ETH_P_IPV6)) { 3516 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6)))) 3517 return false; 3518 iph6 = (const struct ipv6hdr *)(skb->data + *noff); 3519 iph_to_flow_copy_v6addrs(fk, iph6); 3520 *noff += sizeof(*iph6); 3521 *proto = iph6->nexthdr; 3522 } else { 3523 return false; 3524 } 3525 3526 if (l34 && *proto >= 0) 3527 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto); 3528 3529 return true; 3530 } 3531 3532 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb) 3533 { 3534 struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb); 3535 u32 srcmac_vendor = 0, srcmac_dev = 0; 3536 u16 vlan; 3537 int i; 3538 3539 for (i = 0; i < 3; i++) 3540 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 3541 3542 for (i = 3; i < ETH_ALEN; i++) 3543 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 3544 3545 if (!skb_vlan_tag_present(skb)) 3546 return srcmac_vendor ^ srcmac_dev; 3547 3548 vlan = skb_vlan_tag_get(skb); 3549 3550 return vlan ^ srcmac_vendor ^ srcmac_dev; 3551 } 3552 3553 /* Extract the appropriate headers based on bond's xmit policy */ 3554 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, 3555 struct flow_keys *fk) 3556 { 3557 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 3558 int noff, proto = -1; 3559 3560 switch (bond->params.xmit_policy) { 3561 case BOND_XMIT_POLICY_ENCAP23: 3562 case BOND_XMIT_POLICY_ENCAP34: 3563 memset(fk, 0, sizeof(*fk)); 3564 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 3565 fk, NULL, 0, 0, 0, 0); 3566 default: 3567 break; 3568 } 3569 3570 fk->ports.ports = 0; 3571 memset(&fk->icmp, 0, sizeof(fk->icmp)); 3572 noff = skb_network_offset(skb); 3573 if (!bond_flow_ip(skb, fk, &noff, &proto, l34)) 3574 return false; 3575 3576 /* ICMP error packets contains at least 8 bytes of the header 3577 * of the packet which generated the error. Use this information 3578 * to correlate ICMP error packets within the same flow which 3579 * generated the error. 3580 */ 3581 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) { 3582 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data, 3583 skb_transport_offset(skb), 3584 skb_headlen(skb)); 3585 if (proto == IPPROTO_ICMP) { 3586 if (!icmp_is_err(fk->icmp.type)) 3587 return true; 3588 3589 noff += sizeof(struct icmphdr); 3590 } else if (proto == IPPROTO_ICMPV6) { 3591 if (!icmpv6_is_err(fk->icmp.type)) 3592 return true; 3593 3594 noff += sizeof(struct icmp6hdr); 3595 } 3596 return bond_flow_ip(skb, fk, &noff, &proto, l34); 3597 } 3598 3599 return true; 3600 } 3601 3602 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow) 3603 { 3604 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 3605 (__force u32)flow_get_u32_src(flow); 3606 hash ^= (hash >> 16); 3607 hash ^= (hash >> 8); 3608 /* discard lowest hash bit to deal with the common even ports pattern */ 3609 return hash >> 1; 3610 } 3611 3612 /** 3613 * bond_xmit_hash - generate a hash value based on the xmit policy 3614 * @bond: bonding device 3615 * @skb: buffer to use for headers 3616 * 3617 * This function will extract the necessary headers from the skb buffer and use 3618 * them to generate a hash based on the xmit_policy set in the bonding device 3619 */ 3620 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 3621 { 3622 struct flow_keys flow; 3623 u32 hash; 3624 3625 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 3626 skb->l4_hash) 3627 return skb->hash; 3628 3629 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 3630 return bond_vlan_srcmac_hash(skb); 3631 3632 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 3633 !bond_flow_dissect(bond, skb, &flow)) 3634 return bond_eth_hash(skb); 3635 3636 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 3637 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 3638 hash = bond_eth_hash(skb); 3639 } else { 3640 if (flow.icmp.id) 3641 memcpy(&hash, &flow.icmp, sizeof(hash)); 3642 else 3643 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 3644 } 3645 3646 return bond_ip_hash(hash, &flow); 3647 } 3648 3649 /*-------------------------- Device entry points ----------------------------*/ 3650 3651 void bond_work_init_all(struct bonding *bond) 3652 { 3653 INIT_DELAYED_WORK(&bond->mcast_work, 3654 bond_resend_igmp_join_requests_delayed); 3655 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 3656 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 3657 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 3658 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 3659 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 3660 } 3661 3662 static void bond_work_cancel_all(struct bonding *bond) 3663 { 3664 cancel_delayed_work_sync(&bond->mii_work); 3665 cancel_delayed_work_sync(&bond->arp_work); 3666 cancel_delayed_work_sync(&bond->alb_work); 3667 cancel_delayed_work_sync(&bond->ad_work); 3668 cancel_delayed_work_sync(&bond->mcast_work); 3669 cancel_delayed_work_sync(&bond->slave_arr_work); 3670 } 3671 3672 static int bond_open(struct net_device *bond_dev) 3673 { 3674 struct bonding *bond = netdev_priv(bond_dev); 3675 struct list_head *iter; 3676 struct slave *slave; 3677 3678 /* reset slave->backup and slave->inactive */ 3679 if (bond_has_slaves(bond)) { 3680 bond_for_each_slave(bond, slave, iter) { 3681 if (bond_uses_primary(bond) && 3682 slave != rcu_access_pointer(bond->curr_active_slave)) { 3683 bond_set_slave_inactive_flags(slave, 3684 BOND_SLAVE_NOTIFY_NOW); 3685 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 3686 bond_set_slave_active_flags(slave, 3687 BOND_SLAVE_NOTIFY_NOW); 3688 } 3689 } 3690 } 3691 3692 if (bond_is_lb(bond)) { 3693 /* bond_alb_initialize must be called before the timer 3694 * is started. 3695 */ 3696 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 3697 return -ENOMEM; 3698 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 3699 queue_delayed_work(bond->wq, &bond->alb_work, 0); 3700 } 3701 3702 if (bond->params.miimon) /* link check interval, in milliseconds. */ 3703 queue_delayed_work(bond->wq, &bond->mii_work, 0); 3704 3705 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 3706 queue_delayed_work(bond->wq, &bond->arp_work, 0); 3707 bond->recv_probe = bond_arp_rcv; 3708 } 3709 3710 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 3711 queue_delayed_work(bond->wq, &bond->ad_work, 0); 3712 /* register to receive LACPDUs */ 3713 bond->recv_probe = bond_3ad_lacpdu_recv; 3714 bond_3ad_initiate_agg_selection(bond, 1); 3715 } 3716 3717 if (bond_mode_can_use_xmit_hash(bond)) 3718 bond_update_slave_arr(bond, NULL); 3719 3720 return 0; 3721 } 3722 3723 static int bond_close(struct net_device *bond_dev) 3724 { 3725 struct bonding *bond = netdev_priv(bond_dev); 3726 3727 bond_work_cancel_all(bond); 3728 bond->send_peer_notif = 0; 3729 if (bond_is_lb(bond)) 3730 bond_alb_deinitialize(bond); 3731 bond->recv_probe = NULL; 3732 3733 return 0; 3734 } 3735 3736 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 3737 * that some drivers can provide 32bit values only. 3738 */ 3739 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 3740 const struct rtnl_link_stats64 *_new, 3741 const struct rtnl_link_stats64 *_old) 3742 { 3743 const u64 *new = (const u64 *)_new; 3744 const u64 *old = (const u64 *)_old; 3745 u64 *res = (u64 *)_res; 3746 int i; 3747 3748 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 3749 u64 nv = new[i]; 3750 u64 ov = old[i]; 3751 s64 delta = nv - ov; 3752 3753 /* detects if this particular field is 32bit only */ 3754 if (((nv | ov) >> 32) == 0) 3755 delta = (s64)(s32)((u32)nv - (u32)ov); 3756 3757 /* filter anomalies, some drivers reset their stats 3758 * at down/up events. 3759 */ 3760 if (delta > 0) 3761 res[i] += delta; 3762 } 3763 } 3764 3765 #ifdef CONFIG_LOCKDEP 3766 static int bond_get_lowest_level_rcu(struct net_device *dev) 3767 { 3768 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 3769 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 3770 int cur = 0, max = 0; 3771 3772 now = dev; 3773 iter = &dev->adj_list.lower; 3774 3775 while (1) { 3776 next = NULL; 3777 while (1) { 3778 ldev = netdev_next_lower_dev_rcu(now, &iter); 3779 if (!ldev) 3780 break; 3781 3782 next = ldev; 3783 niter = &ldev->adj_list.lower; 3784 dev_stack[cur] = now; 3785 iter_stack[cur++] = iter; 3786 if (max <= cur) 3787 max = cur; 3788 break; 3789 } 3790 3791 if (!next) { 3792 if (!cur) 3793 return max; 3794 next = dev_stack[--cur]; 3795 niter = iter_stack[cur]; 3796 } 3797 3798 now = next; 3799 iter = niter; 3800 } 3801 3802 return max; 3803 } 3804 #endif 3805 3806 static void bond_get_stats(struct net_device *bond_dev, 3807 struct rtnl_link_stats64 *stats) 3808 { 3809 struct bonding *bond = netdev_priv(bond_dev); 3810 struct rtnl_link_stats64 temp; 3811 struct list_head *iter; 3812 struct slave *slave; 3813 int nest_level = 0; 3814 3815 3816 rcu_read_lock(); 3817 #ifdef CONFIG_LOCKDEP 3818 nest_level = bond_get_lowest_level_rcu(bond_dev); 3819 #endif 3820 3821 spin_lock_nested(&bond->stats_lock, nest_level); 3822 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 3823 3824 bond_for_each_slave_rcu(bond, slave, iter) { 3825 const struct rtnl_link_stats64 *new = 3826 dev_get_stats(slave->dev, &temp); 3827 3828 bond_fold_stats(stats, new, &slave->slave_stats); 3829 3830 /* save off the slave stats for the next run */ 3831 memcpy(&slave->slave_stats, new, sizeof(*new)); 3832 } 3833 3834 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 3835 spin_unlock(&bond->stats_lock); 3836 rcu_read_unlock(); 3837 } 3838 3839 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 3840 { 3841 struct bonding *bond = netdev_priv(bond_dev); 3842 struct net_device *slave_dev = NULL; 3843 struct ifbond k_binfo; 3844 struct ifbond __user *u_binfo = NULL; 3845 struct ifslave k_sinfo; 3846 struct ifslave __user *u_sinfo = NULL; 3847 struct mii_ioctl_data *mii = NULL; 3848 struct bond_opt_value newval; 3849 struct net *net; 3850 int res = 0; 3851 3852 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 3853 3854 switch (cmd) { 3855 case SIOCGMIIPHY: 3856 mii = if_mii(ifr); 3857 if (!mii) 3858 return -EINVAL; 3859 3860 mii->phy_id = 0; 3861 fallthrough; 3862 case SIOCGMIIREG: 3863 /* We do this again just in case we were called by SIOCGMIIREG 3864 * instead of SIOCGMIIPHY. 3865 */ 3866 mii = if_mii(ifr); 3867 if (!mii) 3868 return -EINVAL; 3869 3870 if (mii->reg_num == 1) { 3871 mii->val_out = 0; 3872 if (netif_carrier_ok(bond->dev)) 3873 mii->val_out = BMSR_LSTATUS; 3874 } 3875 3876 return 0; 3877 case BOND_INFO_QUERY_OLD: 3878 case SIOCBONDINFOQUERY: 3879 u_binfo = (struct ifbond __user *)ifr->ifr_data; 3880 3881 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 3882 return -EFAULT; 3883 3884 bond_info_query(bond_dev, &k_binfo); 3885 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 3886 return -EFAULT; 3887 3888 return 0; 3889 case BOND_SLAVE_INFO_QUERY_OLD: 3890 case SIOCBONDSLAVEINFOQUERY: 3891 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 3892 3893 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 3894 return -EFAULT; 3895 3896 res = bond_slave_info_query(bond_dev, &k_sinfo); 3897 if (res == 0 && 3898 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 3899 return -EFAULT; 3900 3901 return res; 3902 default: 3903 break; 3904 } 3905 3906 net = dev_net(bond_dev); 3907 3908 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 3909 return -EPERM; 3910 3911 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 3912 3913 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 3914 3915 if (!slave_dev) 3916 return -ENODEV; 3917 3918 switch (cmd) { 3919 case BOND_ENSLAVE_OLD: 3920 case SIOCBONDENSLAVE: 3921 res = bond_enslave(bond_dev, slave_dev, NULL); 3922 break; 3923 case BOND_RELEASE_OLD: 3924 case SIOCBONDRELEASE: 3925 res = bond_release(bond_dev, slave_dev); 3926 break; 3927 case BOND_SETHWADDR_OLD: 3928 case SIOCBONDSETHWADDR: 3929 res = bond_set_dev_addr(bond_dev, slave_dev); 3930 break; 3931 case BOND_CHANGE_ACTIVE_OLD: 3932 case SIOCBONDCHANGEACTIVE: 3933 bond_opt_initstr(&newval, slave_dev->name); 3934 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 3935 &newval); 3936 break; 3937 default: 3938 res = -EOPNOTSUPP; 3939 } 3940 3941 return res; 3942 } 3943 3944 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 3945 { 3946 struct bonding *bond = netdev_priv(bond_dev); 3947 3948 if (change & IFF_PROMISC) 3949 bond_set_promiscuity(bond, 3950 bond_dev->flags & IFF_PROMISC ? 1 : -1); 3951 3952 if (change & IFF_ALLMULTI) 3953 bond_set_allmulti(bond, 3954 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 3955 } 3956 3957 static void bond_set_rx_mode(struct net_device *bond_dev) 3958 { 3959 struct bonding *bond = netdev_priv(bond_dev); 3960 struct list_head *iter; 3961 struct slave *slave; 3962 3963 rcu_read_lock(); 3964 if (bond_uses_primary(bond)) { 3965 slave = rcu_dereference(bond->curr_active_slave); 3966 if (slave) { 3967 dev_uc_sync(slave->dev, bond_dev); 3968 dev_mc_sync(slave->dev, bond_dev); 3969 } 3970 } else { 3971 bond_for_each_slave_rcu(bond, slave, iter) { 3972 dev_uc_sync_multiple(slave->dev, bond_dev); 3973 dev_mc_sync_multiple(slave->dev, bond_dev); 3974 } 3975 } 3976 rcu_read_unlock(); 3977 } 3978 3979 static int bond_neigh_init(struct neighbour *n) 3980 { 3981 struct bonding *bond = netdev_priv(n->dev); 3982 const struct net_device_ops *slave_ops; 3983 struct neigh_parms parms; 3984 struct slave *slave; 3985 int ret = 0; 3986 3987 rcu_read_lock(); 3988 slave = bond_first_slave_rcu(bond); 3989 if (!slave) 3990 goto out; 3991 slave_ops = slave->dev->netdev_ops; 3992 if (!slave_ops->ndo_neigh_setup) 3993 goto out; 3994 3995 /* TODO: find another way [1] to implement this. 3996 * Passing a zeroed structure is fragile, 3997 * but at least we do not pass garbage. 3998 * 3999 * [1] One way would be that ndo_neigh_setup() never touch 4000 * struct neigh_parms, but propagate the new neigh_setup() 4001 * back to ___neigh_create() / neigh_parms_alloc() 4002 */ 4003 memset(&parms, 0, sizeof(parms)); 4004 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4005 4006 if (ret) 4007 goto out; 4008 4009 if (parms.neigh_setup) 4010 ret = parms.neigh_setup(n); 4011 out: 4012 rcu_read_unlock(); 4013 return ret; 4014 } 4015 4016 /* The bonding ndo_neigh_setup is called at init time beofre any 4017 * slave exists. So we must declare proxy setup function which will 4018 * be used at run time to resolve the actual slave neigh param setup. 4019 * 4020 * It's also called by master devices (such as vlans) to setup their 4021 * underlying devices. In that case - do nothing, we're already set up from 4022 * our init. 4023 */ 4024 static int bond_neigh_setup(struct net_device *dev, 4025 struct neigh_parms *parms) 4026 { 4027 /* modify only our neigh_parms */ 4028 if (parms->dev == dev) 4029 parms->neigh_setup = bond_neigh_init; 4030 4031 return 0; 4032 } 4033 4034 /* Change the MTU of all of a master's slaves to match the master */ 4035 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4036 { 4037 struct bonding *bond = netdev_priv(bond_dev); 4038 struct slave *slave, *rollback_slave; 4039 struct list_head *iter; 4040 int res = 0; 4041 4042 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4043 4044 bond_for_each_slave(bond, slave, iter) { 4045 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4046 slave, slave->dev->netdev_ops->ndo_change_mtu); 4047 4048 res = dev_set_mtu(slave->dev, new_mtu); 4049 4050 if (res) { 4051 /* If we failed to set the slave's mtu to the new value 4052 * we must abort the operation even in ACTIVE_BACKUP 4053 * mode, because if we allow the backup slaves to have 4054 * different mtu values than the active slave we'll 4055 * need to change their mtu when doing a failover. That 4056 * means changing their mtu from timer context, which 4057 * is probably not a good idea. 4058 */ 4059 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4060 res, new_mtu); 4061 goto unwind; 4062 } 4063 } 4064 4065 bond_dev->mtu = new_mtu; 4066 4067 return 0; 4068 4069 unwind: 4070 /* unwind from head to the slave that failed */ 4071 bond_for_each_slave(bond, rollback_slave, iter) { 4072 int tmp_res; 4073 4074 if (rollback_slave == slave) 4075 break; 4076 4077 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4078 if (tmp_res) 4079 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4080 tmp_res); 4081 } 4082 4083 return res; 4084 } 4085 4086 /* Change HW address 4087 * 4088 * Note that many devices must be down to change the HW address, and 4089 * downing the master releases all slaves. We can make bonds full of 4090 * bonding devices to test this, however. 4091 */ 4092 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4093 { 4094 struct bonding *bond = netdev_priv(bond_dev); 4095 struct slave *slave, *rollback_slave; 4096 struct sockaddr_storage *ss = addr, tmp_ss; 4097 struct list_head *iter; 4098 int res = 0; 4099 4100 if (BOND_MODE(bond) == BOND_MODE_ALB) 4101 return bond_alb_set_mac_address(bond_dev, addr); 4102 4103 4104 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4105 4106 /* If fail_over_mac is enabled, do nothing and return success. 4107 * Returning an error causes ifenslave to fail. 4108 */ 4109 if (bond->params.fail_over_mac && 4110 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4111 return 0; 4112 4113 if (!is_valid_ether_addr(ss->__data)) 4114 return -EADDRNOTAVAIL; 4115 4116 bond_for_each_slave(bond, slave, iter) { 4117 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4118 __func__, slave); 4119 res = dev_set_mac_address(slave->dev, addr, NULL); 4120 if (res) { 4121 /* TODO: consider downing the slave 4122 * and retry ? 4123 * User should expect communications 4124 * breakage anyway until ARP finish 4125 * updating, so... 4126 */ 4127 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4128 __func__, res); 4129 goto unwind; 4130 } 4131 } 4132 4133 /* success */ 4134 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len); 4135 return 0; 4136 4137 unwind: 4138 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4139 tmp_ss.ss_family = bond_dev->type; 4140 4141 /* unwind from head to the slave that failed */ 4142 bond_for_each_slave(bond, rollback_slave, iter) { 4143 int tmp_res; 4144 4145 if (rollback_slave == slave) 4146 break; 4147 4148 tmp_res = dev_set_mac_address(rollback_slave->dev, 4149 (struct sockaddr *)&tmp_ss, NULL); 4150 if (tmp_res) { 4151 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4152 __func__, tmp_res); 4153 } 4154 } 4155 4156 return res; 4157 } 4158 4159 /** 4160 * bond_get_slave_by_id - get xmit slave with slave_id 4161 * @bond: bonding device that is transmitting 4162 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4163 * 4164 * This function tries to get slave with slave_id but in case 4165 * it fails, it tries to find the first available slave for transmission. 4166 */ 4167 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4168 int slave_id) 4169 { 4170 struct list_head *iter; 4171 struct slave *slave; 4172 int i = slave_id; 4173 4174 /* Here we start from the slave with slave_id */ 4175 bond_for_each_slave_rcu(bond, slave, iter) { 4176 if (--i < 0) { 4177 if (bond_slave_can_tx(slave)) 4178 return slave; 4179 } 4180 } 4181 4182 /* Here we start from the first slave up to slave_id */ 4183 i = slave_id; 4184 bond_for_each_slave_rcu(bond, slave, iter) { 4185 if (--i < 0) 4186 break; 4187 if (bond_slave_can_tx(slave)) 4188 return slave; 4189 } 4190 /* no slave that can tx has been found */ 4191 return NULL; 4192 } 4193 4194 /** 4195 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4196 * @bond: bonding device to use 4197 * 4198 * Based on the value of the bonding device's packets_per_slave parameter 4199 * this function generates a slave id, which is usually used as the next 4200 * slave to transmit through. 4201 */ 4202 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4203 { 4204 u32 slave_id; 4205 struct reciprocal_value reciprocal_packets_per_slave; 4206 int packets_per_slave = bond->params.packets_per_slave; 4207 4208 switch (packets_per_slave) { 4209 case 0: 4210 slave_id = prandom_u32(); 4211 break; 4212 case 1: 4213 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4214 break; 4215 default: 4216 reciprocal_packets_per_slave = 4217 bond->params.reciprocal_packets_per_slave; 4218 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4219 slave_id = reciprocal_divide(slave_id, 4220 reciprocal_packets_per_slave); 4221 break; 4222 } 4223 4224 return slave_id; 4225 } 4226 4227 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4228 struct sk_buff *skb) 4229 { 4230 struct slave *slave; 4231 int slave_cnt; 4232 u32 slave_id; 4233 4234 /* Start with the curr_active_slave that joined the bond as the 4235 * default for sending IGMP traffic. For failover purposes one 4236 * needs to maintain some consistency for the interface that will 4237 * send the join/membership reports. The curr_active_slave found 4238 * will send all of this type of traffic. 4239 */ 4240 if (skb->protocol == htons(ETH_P_IP)) { 4241 int noff = skb_network_offset(skb); 4242 struct iphdr *iph; 4243 4244 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4245 goto non_igmp; 4246 4247 iph = ip_hdr(skb); 4248 if (iph->protocol == IPPROTO_IGMP) { 4249 slave = rcu_dereference(bond->curr_active_slave); 4250 if (slave) 4251 return slave; 4252 return bond_get_slave_by_id(bond, 0); 4253 } 4254 } 4255 4256 non_igmp: 4257 slave_cnt = READ_ONCE(bond->slave_cnt); 4258 if (likely(slave_cnt)) { 4259 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4260 return bond_get_slave_by_id(bond, slave_id); 4261 } 4262 return NULL; 4263 } 4264 4265 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4266 struct net_device *bond_dev) 4267 { 4268 struct bonding *bond = netdev_priv(bond_dev); 4269 struct slave *slave; 4270 4271 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4272 if (likely(slave)) 4273 return bond_dev_queue_xmit(bond, skb, slave->dev); 4274 4275 return bond_tx_drop(bond_dev, skb); 4276 } 4277 4278 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond, 4279 struct sk_buff *skb) 4280 { 4281 return rcu_dereference(bond->curr_active_slave); 4282 } 4283 4284 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4285 * the bond has a usable interface. 4286 */ 4287 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4288 struct net_device *bond_dev) 4289 { 4290 struct bonding *bond = netdev_priv(bond_dev); 4291 struct slave *slave; 4292 4293 slave = bond_xmit_activebackup_slave_get(bond, skb); 4294 if (slave) 4295 return bond_dev_queue_xmit(bond, skb, slave->dev); 4296 4297 return bond_tx_drop(bond_dev, skb); 4298 } 4299 4300 /* Use this to update slave_array when (a) it's not appropriate to update 4301 * slave_array right away (note that update_slave_array() may sleep) 4302 * and / or (b) RTNL is not held. 4303 */ 4304 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4305 { 4306 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4307 } 4308 4309 /* Slave array work handler. Holds only RTNL */ 4310 static void bond_slave_arr_handler(struct work_struct *work) 4311 { 4312 struct bonding *bond = container_of(work, struct bonding, 4313 slave_arr_work.work); 4314 int ret; 4315 4316 if (!rtnl_trylock()) 4317 goto err; 4318 4319 ret = bond_update_slave_arr(bond, NULL); 4320 rtnl_unlock(); 4321 if (ret) { 4322 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4323 goto err; 4324 } 4325 return; 4326 4327 err: 4328 bond_slave_arr_work_rearm(bond, 1); 4329 } 4330 4331 static void bond_skip_slave(struct bond_up_slave *slaves, 4332 struct slave *skipslave) 4333 { 4334 int idx; 4335 4336 /* Rare situation where caller has asked to skip a specific 4337 * slave but allocation failed (most likely!). BTW this is 4338 * only possible when the call is initiated from 4339 * __bond_release_one(). In this situation; overwrite the 4340 * skipslave entry in the array with the last entry from the 4341 * array to avoid a situation where the xmit path may choose 4342 * this to-be-skipped slave to send a packet out. 4343 */ 4344 for (idx = 0; slaves && idx < slaves->count; idx++) { 4345 if (skipslave == slaves->arr[idx]) { 4346 slaves->arr[idx] = 4347 slaves->arr[slaves->count - 1]; 4348 slaves->count--; 4349 break; 4350 } 4351 } 4352 } 4353 4354 static void bond_set_slave_arr(struct bonding *bond, 4355 struct bond_up_slave *usable_slaves, 4356 struct bond_up_slave *all_slaves) 4357 { 4358 struct bond_up_slave *usable, *all; 4359 4360 usable = rtnl_dereference(bond->usable_slaves); 4361 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 4362 kfree_rcu(usable, rcu); 4363 4364 all = rtnl_dereference(bond->all_slaves); 4365 rcu_assign_pointer(bond->all_slaves, all_slaves); 4366 kfree_rcu(all, rcu); 4367 } 4368 4369 static void bond_reset_slave_arr(struct bonding *bond) 4370 { 4371 struct bond_up_slave *usable, *all; 4372 4373 usable = rtnl_dereference(bond->usable_slaves); 4374 if (usable) { 4375 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4376 kfree_rcu(usable, rcu); 4377 } 4378 4379 all = rtnl_dereference(bond->all_slaves); 4380 if (all) { 4381 RCU_INIT_POINTER(bond->all_slaves, NULL); 4382 kfree_rcu(all, rcu); 4383 } 4384 } 4385 4386 /* Build the usable slaves array in control path for modes that use xmit-hash 4387 * to determine the slave interface - 4388 * (a) BOND_MODE_8023AD 4389 * (b) BOND_MODE_XOR 4390 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 4391 * 4392 * The caller is expected to hold RTNL only and NO other lock! 4393 */ 4394 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 4395 { 4396 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 4397 struct slave *slave; 4398 struct list_head *iter; 4399 int agg_id = 0; 4400 int ret = 0; 4401 4402 might_sleep(); 4403 4404 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 4405 bond->slave_cnt), GFP_KERNEL); 4406 all_slaves = kzalloc(struct_size(all_slaves, arr, 4407 bond->slave_cnt), GFP_KERNEL); 4408 if (!usable_slaves || !all_slaves) { 4409 ret = -ENOMEM; 4410 goto out; 4411 } 4412 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4413 struct ad_info ad_info; 4414 4415 spin_lock_bh(&bond->mode_lock); 4416 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 4417 spin_unlock_bh(&bond->mode_lock); 4418 pr_debug("bond_3ad_get_active_agg_info failed\n"); 4419 /* No active aggragator means it's not safe to use 4420 * the previous array. 4421 */ 4422 bond_reset_slave_arr(bond); 4423 goto out; 4424 } 4425 spin_unlock_bh(&bond->mode_lock); 4426 agg_id = ad_info.aggregator_id; 4427 } 4428 bond_for_each_slave(bond, slave, iter) { 4429 if (skipslave == slave) 4430 continue; 4431 4432 all_slaves->arr[all_slaves->count++] = slave; 4433 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4434 struct aggregator *agg; 4435 4436 agg = SLAVE_AD_INFO(slave)->port.aggregator; 4437 if (!agg || agg->aggregator_identifier != agg_id) 4438 continue; 4439 } 4440 if (!bond_slave_can_tx(slave)) 4441 continue; 4442 4443 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 4444 usable_slaves->count); 4445 4446 usable_slaves->arr[usable_slaves->count++] = slave; 4447 } 4448 4449 bond_set_slave_arr(bond, usable_slaves, all_slaves); 4450 return ret; 4451 out: 4452 if (ret != 0 && skipslave) { 4453 bond_skip_slave(rtnl_dereference(bond->all_slaves), 4454 skipslave); 4455 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 4456 skipslave); 4457 } 4458 kfree_rcu(all_slaves, rcu); 4459 kfree_rcu(usable_slaves, rcu); 4460 4461 return ret; 4462 } 4463 4464 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 4465 struct sk_buff *skb, 4466 struct bond_up_slave *slaves) 4467 { 4468 struct slave *slave; 4469 unsigned int count; 4470 u32 hash; 4471 4472 hash = bond_xmit_hash(bond, skb); 4473 count = slaves ? READ_ONCE(slaves->count) : 0; 4474 if (unlikely(!count)) 4475 return NULL; 4476 4477 slave = slaves->arr[hash % count]; 4478 return slave; 4479 } 4480 4481 /* Use this Xmit function for 3AD as well as XOR modes. The current 4482 * usable slave array is formed in the control path. The xmit function 4483 * just calculates hash and sends the packet out. 4484 */ 4485 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 4486 struct net_device *dev) 4487 { 4488 struct bonding *bond = netdev_priv(dev); 4489 struct bond_up_slave *slaves; 4490 struct slave *slave; 4491 4492 slaves = rcu_dereference(bond->usable_slaves); 4493 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4494 if (likely(slave)) 4495 return bond_dev_queue_xmit(bond, skb, slave->dev); 4496 4497 return bond_tx_drop(dev, skb); 4498 } 4499 4500 /* in broadcast mode, we send everything to all usable interfaces. */ 4501 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 4502 struct net_device *bond_dev) 4503 { 4504 struct bonding *bond = netdev_priv(bond_dev); 4505 struct slave *slave = NULL; 4506 struct list_head *iter; 4507 4508 bond_for_each_slave_rcu(bond, slave, iter) { 4509 if (bond_is_last_slave(bond, slave)) 4510 break; 4511 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 4512 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC); 4513 4514 if (!skb2) { 4515 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 4516 bond_dev->name, __func__); 4517 continue; 4518 } 4519 bond_dev_queue_xmit(bond, skb2, slave->dev); 4520 } 4521 } 4522 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) 4523 return bond_dev_queue_xmit(bond, skb, slave->dev); 4524 4525 return bond_tx_drop(bond_dev, skb); 4526 } 4527 4528 /*------------------------- Device initialization ---------------------------*/ 4529 4530 /* Lookup the slave that corresponds to a qid */ 4531 static inline int bond_slave_override(struct bonding *bond, 4532 struct sk_buff *skb) 4533 { 4534 struct slave *slave = NULL; 4535 struct list_head *iter; 4536 4537 if (!skb_rx_queue_recorded(skb)) 4538 return 1; 4539 4540 /* Find out if any slaves have the same mapping as this skb. */ 4541 bond_for_each_slave_rcu(bond, slave, iter) { 4542 if (slave->queue_id == skb_get_queue_mapping(skb)) { 4543 if (bond_slave_is_up(slave) && 4544 slave->link == BOND_LINK_UP) { 4545 bond_dev_queue_xmit(bond, skb, slave->dev); 4546 return 0; 4547 } 4548 /* If the slave isn't UP, use default transmit policy. */ 4549 break; 4550 } 4551 } 4552 4553 return 1; 4554 } 4555 4556 4557 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 4558 struct net_device *sb_dev) 4559 { 4560 /* This helper function exists to help dev_pick_tx get the correct 4561 * destination queue. Using a helper function skips a call to 4562 * skb_tx_hash and will put the skbs in the queue we expect on their 4563 * way down to the bonding driver. 4564 */ 4565 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 4566 4567 /* Save the original txq to restore before passing to the driver */ 4568 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 4569 4570 if (unlikely(txq >= dev->real_num_tx_queues)) { 4571 do { 4572 txq -= dev->real_num_tx_queues; 4573 } while (txq >= dev->real_num_tx_queues); 4574 } 4575 return txq; 4576 } 4577 4578 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 4579 struct sk_buff *skb, 4580 bool all_slaves) 4581 { 4582 struct bonding *bond = netdev_priv(master_dev); 4583 struct bond_up_slave *slaves; 4584 struct slave *slave = NULL; 4585 4586 switch (BOND_MODE(bond)) { 4587 case BOND_MODE_ROUNDROBIN: 4588 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4589 break; 4590 case BOND_MODE_ACTIVEBACKUP: 4591 slave = bond_xmit_activebackup_slave_get(bond, skb); 4592 break; 4593 case BOND_MODE_8023AD: 4594 case BOND_MODE_XOR: 4595 if (all_slaves) 4596 slaves = rcu_dereference(bond->all_slaves); 4597 else 4598 slaves = rcu_dereference(bond->usable_slaves); 4599 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 4600 break; 4601 case BOND_MODE_BROADCAST: 4602 break; 4603 case BOND_MODE_ALB: 4604 slave = bond_xmit_alb_slave_get(bond, skb); 4605 break; 4606 case BOND_MODE_TLB: 4607 slave = bond_xmit_tlb_slave_get(bond, skb); 4608 break; 4609 default: 4610 /* Should never happen, mode already checked */ 4611 WARN_ONCE(true, "Unknown bonding mode"); 4612 break; 4613 } 4614 4615 if (slave) 4616 return slave->dev; 4617 return NULL; 4618 } 4619 4620 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 4621 { 4622 switch (sk->sk_family) { 4623 #if IS_ENABLED(CONFIG_IPV6) 4624 case AF_INET6: 4625 if (sk->sk_ipv6only || 4626 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 4627 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 4628 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 4629 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 4630 break; 4631 } 4632 fallthrough; 4633 #endif 4634 default: /* AF_INET */ 4635 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 4636 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 4637 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 4638 break; 4639 } 4640 4641 flow->ports.src = inet_sk(sk)->inet_sport; 4642 flow->ports.dst = inet_sk(sk)->inet_dport; 4643 } 4644 4645 /** 4646 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 4647 * @sk: socket to use for headers 4648 * 4649 * This function will extract the necessary field from the socket and use 4650 * them to generate a hash based on the LAYER34 xmit_policy. 4651 * Assumes that sk is a TCP or UDP socket. 4652 */ 4653 static u32 bond_sk_hash_l34(struct sock *sk) 4654 { 4655 struct flow_keys flow; 4656 u32 hash; 4657 4658 bond_sk_to_flow(sk, &flow); 4659 4660 /* L4 */ 4661 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4662 /* L3 */ 4663 return bond_ip_hash(hash, &flow); 4664 } 4665 4666 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 4667 struct sock *sk) 4668 { 4669 struct bond_up_slave *slaves; 4670 struct slave *slave; 4671 unsigned int count; 4672 u32 hash; 4673 4674 slaves = rcu_dereference(bond->usable_slaves); 4675 count = slaves ? READ_ONCE(slaves->count) : 0; 4676 if (unlikely(!count)) 4677 return NULL; 4678 4679 hash = bond_sk_hash_l34(sk); 4680 slave = slaves->arr[hash % count]; 4681 4682 return slave->dev; 4683 } 4684 4685 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 4686 struct sock *sk) 4687 { 4688 struct bonding *bond = netdev_priv(dev); 4689 struct net_device *lower = NULL; 4690 4691 rcu_read_lock(); 4692 if (bond_sk_check(bond)) 4693 lower = __bond_sk_get_lower_dev(bond, sk); 4694 rcu_read_unlock(); 4695 4696 return lower; 4697 } 4698 4699 #if IS_ENABLED(CONFIG_TLS_DEVICE) 4700 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 4701 struct net_device *dev) 4702 { 4703 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev))) 4704 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev); 4705 return bond_tx_drop(dev, skb); 4706 } 4707 #endif 4708 4709 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4710 { 4711 struct bonding *bond = netdev_priv(dev); 4712 4713 if (bond_should_override_tx_queue(bond) && 4714 !bond_slave_override(bond, skb)) 4715 return NETDEV_TX_OK; 4716 4717 #if IS_ENABLED(CONFIG_TLS_DEVICE) 4718 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk)) 4719 return bond_tls_device_xmit(bond, skb, dev); 4720 #endif 4721 4722 switch (BOND_MODE(bond)) { 4723 case BOND_MODE_ROUNDROBIN: 4724 return bond_xmit_roundrobin(skb, dev); 4725 case BOND_MODE_ACTIVEBACKUP: 4726 return bond_xmit_activebackup(skb, dev); 4727 case BOND_MODE_8023AD: 4728 case BOND_MODE_XOR: 4729 return bond_3ad_xor_xmit(skb, dev); 4730 case BOND_MODE_BROADCAST: 4731 return bond_xmit_broadcast(skb, dev); 4732 case BOND_MODE_ALB: 4733 return bond_alb_xmit(skb, dev); 4734 case BOND_MODE_TLB: 4735 return bond_tlb_xmit(skb, dev); 4736 default: 4737 /* Should never happen, mode already checked */ 4738 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 4739 WARN_ON_ONCE(1); 4740 return bond_tx_drop(dev, skb); 4741 } 4742 } 4743 4744 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 4745 { 4746 struct bonding *bond = netdev_priv(dev); 4747 netdev_tx_t ret = NETDEV_TX_OK; 4748 4749 /* If we risk deadlock from transmitting this in the 4750 * netpoll path, tell netpoll to queue the frame for later tx 4751 */ 4752 if (unlikely(is_netpoll_tx_blocked(dev))) 4753 return NETDEV_TX_BUSY; 4754 4755 rcu_read_lock(); 4756 if (bond_has_slaves(bond)) 4757 ret = __bond_start_xmit(skb, dev); 4758 else 4759 ret = bond_tx_drop(dev, skb); 4760 rcu_read_unlock(); 4761 4762 return ret; 4763 } 4764 4765 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 4766 { 4767 if (speed == 0 || speed == SPEED_UNKNOWN) 4768 speed = slave->speed; 4769 else 4770 speed = min(speed, slave->speed); 4771 4772 return speed; 4773 } 4774 4775 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 4776 struct ethtool_link_ksettings *cmd) 4777 { 4778 struct bonding *bond = netdev_priv(bond_dev); 4779 struct list_head *iter; 4780 struct slave *slave; 4781 u32 speed = 0; 4782 4783 cmd->base.duplex = DUPLEX_UNKNOWN; 4784 cmd->base.port = PORT_OTHER; 4785 4786 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 4787 * do not need to check mode. Though link speed might not represent 4788 * the true receive or transmit bandwidth (not all modes are symmetric) 4789 * this is an accurate maximum. 4790 */ 4791 bond_for_each_slave(bond, slave, iter) { 4792 if (bond_slave_can_tx(slave)) { 4793 if (slave->speed != SPEED_UNKNOWN) { 4794 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 4795 speed = bond_mode_bcast_speed(slave, 4796 speed); 4797 else 4798 speed += slave->speed; 4799 } 4800 if (cmd->base.duplex == DUPLEX_UNKNOWN && 4801 slave->duplex != DUPLEX_UNKNOWN) 4802 cmd->base.duplex = slave->duplex; 4803 } 4804 } 4805 cmd->base.speed = speed ? : SPEED_UNKNOWN; 4806 4807 return 0; 4808 } 4809 4810 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 4811 struct ethtool_drvinfo *drvinfo) 4812 { 4813 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 4814 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 4815 BOND_ABI_VERSION); 4816 } 4817 4818 static const struct ethtool_ops bond_ethtool_ops = { 4819 .get_drvinfo = bond_ethtool_get_drvinfo, 4820 .get_link = ethtool_op_get_link, 4821 .get_link_ksettings = bond_ethtool_get_link_ksettings, 4822 }; 4823 4824 static const struct net_device_ops bond_netdev_ops = { 4825 .ndo_init = bond_init, 4826 .ndo_uninit = bond_uninit, 4827 .ndo_open = bond_open, 4828 .ndo_stop = bond_close, 4829 .ndo_start_xmit = bond_start_xmit, 4830 .ndo_select_queue = bond_select_queue, 4831 .ndo_get_stats64 = bond_get_stats, 4832 .ndo_do_ioctl = bond_do_ioctl, 4833 .ndo_change_rx_flags = bond_change_rx_flags, 4834 .ndo_set_rx_mode = bond_set_rx_mode, 4835 .ndo_change_mtu = bond_change_mtu, 4836 .ndo_set_mac_address = bond_set_mac_address, 4837 .ndo_neigh_setup = bond_neigh_setup, 4838 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 4839 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 4840 #ifdef CONFIG_NET_POLL_CONTROLLER 4841 .ndo_netpoll_setup = bond_netpoll_setup, 4842 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 4843 .ndo_poll_controller = bond_poll_controller, 4844 #endif 4845 .ndo_add_slave = bond_enslave, 4846 .ndo_del_slave = bond_release, 4847 .ndo_fix_features = bond_fix_features, 4848 .ndo_features_check = passthru_features_check, 4849 .ndo_get_xmit_slave = bond_xmit_get_slave, 4850 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 4851 }; 4852 4853 static const struct device_type bond_type = { 4854 .name = "bond", 4855 }; 4856 4857 static void bond_destructor(struct net_device *bond_dev) 4858 { 4859 struct bonding *bond = netdev_priv(bond_dev); 4860 4861 if (bond->wq) 4862 destroy_workqueue(bond->wq); 4863 4864 if (bond->rr_tx_counter) 4865 free_percpu(bond->rr_tx_counter); 4866 } 4867 4868 void bond_setup(struct net_device *bond_dev) 4869 { 4870 struct bonding *bond = netdev_priv(bond_dev); 4871 4872 spin_lock_init(&bond->mode_lock); 4873 bond->params = bonding_defaults; 4874 4875 /* Initialize pointers */ 4876 bond->dev = bond_dev; 4877 4878 /* Initialize the device entry points */ 4879 ether_setup(bond_dev); 4880 bond_dev->max_mtu = ETH_MAX_MTU; 4881 bond_dev->netdev_ops = &bond_netdev_ops; 4882 bond_dev->ethtool_ops = &bond_ethtool_ops; 4883 4884 bond_dev->needs_free_netdev = true; 4885 bond_dev->priv_destructor = bond_destructor; 4886 4887 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 4888 4889 /* Initialize the device options */ 4890 bond_dev->flags |= IFF_MASTER; 4891 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 4892 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 4893 4894 #ifdef CONFIG_XFRM_OFFLOAD 4895 /* set up xfrm device ops (only supported in active-backup right now) */ 4896 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 4897 bond->xs = NULL; 4898 #endif /* CONFIG_XFRM_OFFLOAD */ 4899 4900 /* don't acquire bond device's netif_tx_lock when transmitting */ 4901 bond_dev->features |= NETIF_F_LLTX; 4902 4903 /* By default, we declare the bond to be fully 4904 * VLAN hardware accelerated capable. Special 4905 * care is taken in the various xmit functions 4906 * when there are slaves that are not hw accel 4907 * capable 4908 */ 4909 4910 /* Don't allow bond devices to change network namespaces. */ 4911 bond_dev->features |= NETIF_F_NETNS_LOCAL; 4912 4913 bond_dev->hw_features = BOND_VLAN_FEATURES | 4914 NETIF_F_HW_VLAN_CTAG_RX | 4915 NETIF_F_HW_VLAN_CTAG_FILTER; 4916 4917 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 4918 bond_dev->features |= bond_dev->hw_features; 4919 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 4920 #ifdef CONFIG_XFRM_OFFLOAD 4921 bond_dev->hw_features |= BOND_XFRM_FEATURES; 4922 /* Only enable XFRM features if this is an active-backup config */ 4923 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4924 bond_dev->features |= BOND_XFRM_FEATURES; 4925 #endif /* CONFIG_XFRM_OFFLOAD */ 4926 #if IS_ENABLED(CONFIG_TLS_DEVICE) 4927 if (bond_sk_check(bond)) 4928 bond_dev->features |= BOND_TLS_FEATURES; 4929 #endif 4930 } 4931 4932 /* Destroy a bonding device. 4933 * Must be under rtnl_lock when this function is called. 4934 */ 4935 static void bond_uninit(struct net_device *bond_dev) 4936 { 4937 struct bonding *bond = netdev_priv(bond_dev); 4938 struct bond_up_slave *usable, *all; 4939 struct list_head *iter; 4940 struct slave *slave; 4941 4942 bond_netpoll_cleanup(bond_dev); 4943 4944 /* Release the bonded slaves */ 4945 bond_for_each_slave(bond, slave, iter) 4946 __bond_release_one(bond_dev, slave->dev, true, true); 4947 netdev_info(bond_dev, "Released all slaves\n"); 4948 4949 usable = rtnl_dereference(bond->usable_slaves); 4950 if (usable) { 4951 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4952 kfree_rcu(usable, rcu); 4953 } 4954 4955 all = rtnl_dereference(bond->all_slaves); 4956 if (all) { 4957 RCU_INIT_POINTER(bond->all_slaves, NULL); 4958 kfree_rcu(all, rcu); 4959 } 4960 4961 list_del(&bond->bond_list); 4962 4963 bond_debug_unregister(bond); 4964 } 4965 4966 /*------------------------- Module initialization ---------------------------*/ 4967 4968 static int bond_check_params(struct bond_params *params) 4969 { 4970 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 4971 struct bond_opt_value newval; 4972 const struct bond_opt_value *valptr; 4973 int arp_all_targets_value = 0; 4974 u16 ad_actor_sys_prio = 0; 4975 u16 ad_user_port_key = 0; 4976 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 4977 int arp_ip_count; 4978 int bond_mode = BOND_MODE_ROUNDROBIN; 4979 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 4980 int lacp_fast = 0; 4981 int tlb_dynamic_lb; 4982 4983 /* Convert string parameters. */ 4984 if (mode) { 4985 bond_opt_initstr(&newval, mode); 4986 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 4987 if (!valptr) { 4988 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 4989 return -EINVAL; 4990 } 4991 bond_mode = valptr->value; 4992 } 4993 4994 if (xmit_hash_policy) { 4995 if (bond_mode == BOND_MODE_ROUNDROBIN || 4996 bond_mode == BOND_MODE_ACTIVEBACKUP || 4997 bond_mode == BOND_MODE_BROADCAST) { 4998 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 4999 bond_mode_name(bond_mode)); 5000 } else { 5001 bond_opt_initstr(&newval, xmit_hash_policy); 5002 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5003 &newval); 5004 if (!valptr) { 5005 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5006 xmit_hash_policy); 5007 return -EINVAL; 5008 } 5009 xmit_hashtype = valptr->value; 5010 } 5011 } 5012 5013 if (lacp_rate) { 5014 if (bond_mode != BOND_MODE_8023AD) { 5015 pr_info("lacp_rate param is irrelevant in mode %s\n", 5016 bond_mode_name(bond_mode)); 5017 } else { 5018 bond_opt_initstr(&newval, lacp_rate); 5019 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5020 &newval); 5021 if (!valptr) { 5022 pr_err("Error: Invalid lacp rate \"%s\"\n", 5023 lacp_rate); 5024 return -EINVAL; 5025 } 5026 lacp_fast = valptr->value; 5027 } 5028 } 5029 5030 if (ad_select) { 5031 bond_opt_initstr(&newval, ad_select); 5032 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5033 &newval); 5034 if (!valptr) { 5035 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5036 return -EINVAL; 5037 } 5038 params->ad_select = valptr->value; 5039 if (bond_mode != BOND_MODE_8023AD) 5040 pr_warn("ad_select param only affects 802.3ad mode\n"); 5041 } else { 5042 params->ad_select = BOND_AD_STABLE; 5043 } 5044 5045 if (max_bonds < 0) { 5046 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5047 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5048 max_bonds = BOND_DEFAULT_MAX_BONDS; 5049 } 5050 5051 if (miimon < 0) { 5052 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5053 miimon, INT_MAX); 5054 miimon = 0; 5055 } 5056 5057 if (updelay < 0) { 5058 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5059 updelay, INT_MAX); 5060 updelay = 0; 5061 } 5062 5063 if (downdelay < 0) { 5064 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5065 downdelay, INT_MAX); 5066 downdelay = 0; 5067 } 5068 5069 if ((use_carrier != 0) && (use_carrier != 1)) { 5070 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5071 use_carrier); 5072 use_carrier = 1; 5073 } 5074 5075 if (num_peer_notif < 0 || num_peer_notif > 255) { 5076 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5077 num_peer_notif); 5078 num_peer_notif = 1; 5079 } 5080 5081 /* reset values for 802.3ad/TLB/ALB */ 5082 if (!bond_mode_uses_arp(bond_mode)) { 5083 if (!miimon) { 5084 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"); 5085 pr_warn("Forcing miimon to 100msec\n"); 5086 miimon = BOND_DEFAULT_MIIMON; 5087 } 5088 } 5089 5090 if (tx_queues < 1 || tx_queues > 255) { 5091 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5092 tx_queues, BOND_DEFAULT_TX_QUEUES); 5093 tx_queues = BOND_DEFAULT_TX_QUEUES; 5094 } 5095 5096 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5097 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5098 all_slaves_active); 5099 all_slaves_active = 0; 5100 } 5101 5102 if (resend_igmp < 0 || resend_igmp > 255) { 5103 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 5104 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 5105 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 5106 } 5107 5108 bond_opt_initval(&newval, packets_per_slave); 5109 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 5110 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 5111 packets_per_slave, USHRT_MAX); 5112 packets_per_slave = 1; 5113 } 5114 5115 if (bond_mode == BOND_MODE_ALB) { 5116 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", 5117 updelay); 5118 } 5119 5120 if (!miimon) { 5121 if (updelay || downdelay) { 5122 /* just warn the user the up/down delay will have 5123 * no effect since miimon is zero... 5124 */ 5125 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", 5126 updelay, downdelay); 5127 } 5128 } else { 5129 /* don't allow arp monitoring */ 5130 if (arp_interval) { 5131 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 5132 miimon, arp_interval); 5133 arp_interval = 0; 5134 } 5135 5136 if ((updelay % miimon) != 0) { 5137 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 5138 updelay, miimon, (updelay / miimon) * miimon); 5139 } 5140 5141 updelay /= miimon; 5142 5143 if ((downdelay % miimon) != 0) { 5144 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 5145 downdelay, miimon, 5146 (downdelay / miimon) * miimon); 5147 } 5148 5149 downdelay /= miimon; 5150 } 5151 5152 if (arp_interval < 0) { 5153 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5154 arp_interval, INT_MAX); 5155 arp_interval = 0; 5156 } 5157 5158 for (arp_ip_count = 0, i = 0; 5159 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 5160 __be32 ip; 5161 5162 /* not a complete check, but good enough to catch mistakes */ 5163 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 5164 !bond_is_ip_target_ok(ip)) { 5165 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 5166 arp_ip_target[i]); 5167 arp_interval = 0; 5168 } else { 5169 if (bond_get_targets_ip(arp_target, ip) == -1) 5170 arp_target[arp_ip_count++] = ip; 5171 else 5172 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 5173 &ip); 5174 } 5175 } 5176 5177 if (arp_interval && !arp_ip_count) { 5178 /* don't allow arping if no arp_ip_target given... */ 5179 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 5180 arp_interval); 5181 arp_interval = 0; 5182 } 5183 5184 if (arp_validate) { 5185 if (!arp_interval) { 5186 pr_err("arp_validate requires arp_interval\n"); 5187 return -EINVAL; 5188 } 5189 5190 bond_opt_initstr(&newval, arp_validate); 5191 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 5192 &newval); 5193 if (!valptr) { 5194 pr_err("Error: invalid arp_validate \"%s\"\n", 5195 arp_validate); 5196 return -EINVAL; 5197 } 5198 arp_validate_value = valptr->value; 5199 } else { 5200 arp_validate_value = 0; 5201 } 5202 5203 if (arp_all_targets) { 5204 bond_opt_initstr(&newval, arp_all_targets); 5205 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 5206 &newval); 5207 if (!valptr) { 5208 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 5209 arp_all_targets); 5210 arp_all_targets_value = 0; 5211 } else { 5212 arp_all_targets_value = valptr->value; 5213 } 5214 } 5215 5216 if (miimon) { 5217 pr_info("MII link monitoring set to %d ms\n", miimon); 5218 } else if (arp_interval) { 5219 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 5220 arp_validate_value); 5221 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 5222 arp_interval, valptr->string, arp_ip_count); 5223 5224 for (i = 0; i < arp_ip_count; i++) 5225 pr_cont(" %s", arp_ip_target[i]); 5226 5227 pr_cont("\n"); 5228 5229 } else if (max_bonds) { 5230 /* miimon and arp_interval not set, we need one so things 5231 * work as expected, see bonding.txt for details 5232 */ 5233 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"); 5234 } 5235 5236 if (primary && !bond_mode_uses_primary(bond_mode)) { 5237 /* currently, using a primary only makes sense 5238 * in active backup, TLB or ALB modes 5239 */ 5240 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 5241 primary, bond_mode_name(bond_mode)); 5242 primary = NULL; 5243 } 5244 5245 if (primary && primary_reselect) { 5246 bond_opt_initstr(&newval, primary_reselect); 5247 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 5248 &newval); 5249 if (!valptr) { 5250 pr_err("Error: Invalid primary_reselect \"%s\"\n", 5251 primary_reselect); 5252 return -EINVAL; 5253 } 5254 primary_reselect_value = valptr->value; 5255 } else { 5256 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 5257 } 5258 5259 if (fail_over_mac) { 5260 bond_opt_initstr(&newval, fail_over_mac); 5261 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 5262 &newval); 5263 if (!valptr) { 5264 pr_err("Error: invalid fail_over_mac \"%s\"\n", 5265 fail_over_mac); 5266 return -EINVAL; 5267 } 5268 fail_over_mac_value = valptr->value; 5269 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 5270 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 5271 } else { 5272 fail_over_mac_value = BOND_FOM_NONE; 5273 } 5274 5275 bond_opt_initstr(&newval, "default"); 5276 valptr = bond_opt_parse( 5277 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 5278 &newval); 5279 if (!valptr) { 5280 pr_err("Error: No ad_actor_sys_prio default value"); 5281 return -EINVAL; 5282 } 5283 ad_actor_sys_prio = valptr->value; 5284 5285 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 5286 &newval); 5287 if (!valptr) { 5288 pr_err("Error: No ad_user_port_key default value"); 5289 return -EINVAL; 5290 } 5291 ad_user_port_key = valptr->value; 5292 5293 bond_opt_initstr(&newval, "default"); 5294 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 5295 if (!valptr) { 5296 pr_err("Error: No tlb_dynamic_lb default value"); 5297 return -EINVAL; 5298 } 5299 tlb_dynamic_lb = valptr->value; 5300 5301 if (lp_interval == 0) { 5302 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 5303 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 5304 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 5305 } 5306 5307 /* fill params struct with the proper values */ 5308 params->mode = bond_mode; 5309 params->xmit_policy = xmit_hashtype; 5310 params->miimon = miimon; 5311 params->num_peer_notif = num_peer_notif; 5312 params->arp_interval = arp_interval; 5313 params->arp_validate = arp_validate_value; 5314 params->arp_all_targets = arp_all_targets_value; 5315 params->updelay = updelay; 5316 params->downdelay = downdelay; 5317 params->peer_notif_delay = 0; 5318 params->use_carrier = use_carrier; 5319 params->lacp_fast = lacp_fast; 5320 params->primary[0] = 0; 5321 params->primary_reselect = primary_reselect_value; 5322 params->fail_over_mac = fail_over_mac_value; 5323 params->tx_queues = tx_queues; 5324 params->all_slaves_active = all_slaves_active; 5325 params->resend_igmp = resend_igmp; 5326 params->min_links = min_links; 5327 params->lp_interval = lp_interval; 5328 params->packets_per_slave = packets_per_slave; 5329 params->tlb_dynamic_lb = tlb_dynamic_lb; 5330 params->ad_actor_sys_prio = ad_actor_sys_prio; 5331 eth_zero_addr(params->ad_actor_system); 5332 params->ad_user_port_key = ad_user_port_key; 5333 if (packets_per_slave > 0) { 5334 params->reciprocal_packets_per_slave = 5335 reciprocal_value(packets_per_slave); 5336 } else { 5337 /* reciprocal_packets_per_slave is unused if 5338 * packets_per_slave is 0 or 1, just initialize it 5339 */ 5340 params->reciprocal_packets_per_slave = 5341 (struct reciprocal_value) { 0 }; 5342 } 5343 5344 if (primary) 5345 strscpy_pad(params->primary, primary, sizeof(params->primary)); 5346 5347 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 5348 5349 return 0; 5350 } 5351 5352 /* Called from registration process */ 5353 static int bond_init(struct net_device *bond_dev) 5354 { 5355 struct bonding *bond = netdev_priv(bond_dev); 5356 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 5357 5358 netdev_dbg(bond_dev, "Begin bond_init\n"); 5359 5360 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 5361 if (!bond->wq) 5362 return -ENOMEM; 5363 5364 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) { 5365 bond->rr_tx_counter = alloc_percpu(u32); 5366 if (!bond->rr_tx_counter) { 5367 destroy_workqueue(bond->wq); 5368 bond->wq = NULL; 5369 return -ENOMEM; 5370 } 5371 } 5372 5373 spin_lock_init(&bond->stats_lock); 5374 netdev_lockdep_set_classes(bond_dev); 5375 5376 list_add_tail(&bond->bond_list, &bn->dev_list); 5377 5378 bond_prepare_sysfs_group(bond); 5379 5380 bond_debug_register(bond); 5381 5382 /* Ensure valid dev_addr */ 5383 if (is_zero_ether_addr(bond_dev->dev_addr) && 5384 bond_dev->addr_assign_type == NET_ADDR_PERM) 5385 eth_hw_addr_random(bond_dev); 5386 5387 return 0; 5388 } 5389 5390 unsigned int bond_get_num_tx_queues(void) 5391 { 5392 return tx_queues; 5393 } 5394 5395 /* Create a new bond based on the specified name and bonding parameters. 5396 * If name is NULL, obtain a suitable "bond%d" name for us. 5397 * Caller must NOT hold rtnl_lock; we need to release it here before we 5398 * set up our sysfs entries. 5399 */ 5400 int bond_create(struct net *net, const char *name) 5401 { 5402 struct net_device *bond_dev; 5403 struct bonding *bond; 5404 struct alb_bond_info *bond_info; 5405 int res; 5406 5407 rtnl_lock(); 5408 5409 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 5410 name ? name : "bond%d", NET_NAME_UNKNOWN, 5411 bond_setup, tx_queues); 5412 if (!bond_dev) { 5413 pr_err("%s: eek! can't alloc netdev!\n", name); 5414 rtnl_unlock(); 5415 return -ENOMEM; 5416 } 5417 5418 /* 5419 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX. 5420 * It is set to 0 by default which is wrong. 5421 */ 5422 bond = netdev_priv(bond_dev); 5423 bond_info = &(BOND_ALB_INFO(bond)); 5424 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX; 5425 5426 dev_net_set(bond_dev, net); 5427 bond_dev->rtnl_link_ops = &bond_link_ops; 5428 5429 res = register_netdevice(bond_dev); 5430 if (res < 0) { 5431 free_netdev(bond_dev); 5432 rtnl_unlock(); 5433 5434 return res; 5435 } 5436 5437 netif_carrier_off(bond_dev); 5438 5439 bond_work_init_all(bond); 5440 5441 rtnl_unlock(); 5442 return 0; 5443 } 5444 5445 static int __net_init bond_net_init(struct net *net) 5446 { 5447 struct bond_net *bn = net_generic(net, bond_net_id); 5448 5449 bn->net = net; 5450 INIT_LIST_HEAD(&bn->dev_list); 5451 5452 bond_create_proc_dir(bn); 5453 bond_create_sysfs(bn); 5454 5455 return 0; 5456 } 5457 5458 static void __net_exit bond_net_exit(struct net *net) 5459 { 5460 struct bond_net *bn = net_generic(net, bond_net_id); 5461 struct bonding *bond, *tmp_bond; 5462 LIST_HEAD(list); 5463 5464 bond_destroy_sysfs(bn); 5465 5466 /* Kill off any bonds created after unregistering bond rtnl ops */ 5467 rtnl_lock(); 5468 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 5469 unregister_netdevice_queue(bond->dev, &list); 5470 unregister_netdevice_many(&list); 5471 rtnl_unlock(); 5472 5473 bond_destroy_proc_dir(bn); 5474 } 5475 5476 static struct pernet_operations bond_net_ops = { 5477 .init = bond_net_init, 5478 .exit = bond_net_exit, 5479 .id = &bond_net_id, 5480 .size = sizeof(struct bond_net), 5481 }; 5482 5483 static int __init bonding_init(void) 5484 { 5485 int i; 5486 int res; 5487 5488 res = bond_check_params(&bonding_defaults); 5489 if (res) 5490 goto out; 5491 5492 res = register_pernet_subsys(&bond_net_ops); 5493 if (res) 5494 goto out; 5495 5496 res = bond_netlink_init(); 5497 if (res) 5498 goto err_link; 5499 5500 bond_create_debugfs(); 5501 5502 for (i = 0; i < max_bonds; i++) { 5503 res = bond_create(&init_net, NULL); 5504 if (res) 5505 goto err; 5506 } 5507 5508 skb_flow_dissector_init(&flow_keys_bonding, 5509 flow_keys_bonding_keys, 5510 ARRAY_SIZE(flow_keys_bonding_keys)); 5511 5512 register_netdevice_notifier(&bond_netdev_notifier); 5513 out: 5514 return res; 5515 err: 5516 bond_destroy_debugfs(); 5517 bond_netlink_fini(); 5518 err_link: 5519 unregister_pernet_subsys(&bond_net_ops); 5520 goto out; 5521 5522 } 5523 5524 static void __exit bonding_exit(void) 5525 { 5526 unregister_netdevice_notifier(&bond_netdev_notifier); 5527 5528 bond_destroy_debugfs(); 5529 5530 bond_netlink_fini(); 5531 unregister_pernet_subsys(&bond_net_ops); 5532 5533 #ifdef CONFIG_NET_POLL_CONTROLLER 5534 /* Make sure we don't have an imbalance on our netpoll blocking */ 5535 WARN_ON(atomic_read(&netpoll_block_tx)); 5536 #endif 5537 } 5538 5539 module_init(bonding_init); 5540 module_exit(bonding_exit); 5541 MODULE_LICENSE("GPL"); 5542 MODULE_DESCRIPTION(DRV_DESCRIPTION); 5543 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 5544