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