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/filter.h> 39 #include <linux/interrupt.h> 40 #include <linux/ptrace.h> 41 #include <linux/ioport.h> 42 #include <linux/in.h> 43 #include <net/ip.h> 44 #include <linux/ip.h> 45 #include <linux/icmp.h> 46 #include <linux/icmpv6.h> 47 #include <linux/tcp.h> 48 #include <linux/udp.h> 49 #include <linux/slab.h> 50 #include <linux/string.h> 51 #include <linux/init.h> 52 #include <linux/timer.h> 53 #include <linux/socket.h> 54 #include <linux/ctype.h> 55 #include <linux/inet.h> 56 #include <linux/bitops.h> 57 #include <linux/io.h> 58 #include <asm/dma.h> 59 #include <linux/uaccess.h> 60 #include <linux/errno.h> 61 #include <linux/netdevice.h> 62 #include <linux/inetdevice.h> 63 #include <linux/igmp.h> 64 #include <linux/etherdevice.h> 65 #include <linux/skbuff.h> 66 #include <net/sock.h> 67 #include <linux/rtnetlink.h> 68 #include <linux/smp.h> 69 #include <linux/if_ether.h> 70 #include <net/arp.h> 71 #include <linux/mii.h> 72 #include <linux/ethtool.h> 73 #include <linux/if_vlan.h> 74 #include <linux/if_bonding.h> 75 #include <linux/phy.h> 76 #include <linux/jiffies.h> 77 #include <linux/preempt.h> 78 #include <net/route.h> 79 #include <net/net_namespace.h> 80 #include <net/netns/generic.h> 81 #include <net/pkt_sched.h> 82 #include <linux/rculist.h> 83 #include <net/flow_dissector.h> 84 #include <net/xfrm.h> 85 #include <net/bonding.h> 86 #include <net/bond_3ad.h> 87 #include <net/bond_alb.h> 88 #if IS_ENABLED(CONFIG_TLS_DEVICE) 89 #include <net/tls.h> 90 #endif 91 #include <net/ip6_route.h> 92 93 #include "bonding_priv.h" 94 95 /*---------------------------- Module parameters ----------------------------*/ 96 97 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 98 99 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 100 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 101 static int num_peer_notif = 1; 102 static int miimon; 103 static int updelay; 104 static int downdelay; 105 static int use_carrier = 1; 106 static char *mode; 107 static char *primary; 108 static char *primary_reselect; 109 static char *lacp_rate; 110 static int min_links; 111 static char *ad_select; 112 static char *xmit_hash_policy; 113 static int arp_interval; 114 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 115 static char *arp_validate; 116 static char *arp_all_targets; 117 static char *fail_over_mac; 118 static int all_slaves_active; 119 static struct bond_params bonding_defaults; 120 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 121 static int packets_per_slave = 1; 122 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 123 124 module_param(max_bonds, int, 0); 125 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 126 module_param(tx_queues, int, 0); 127 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 128 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 129 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 130 "failover event (alias of num_unsol_na)"); 131 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 132 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 133 "failover event (alias of num_grat_arp)"); 134 module_param(miimon, int, 0); 135 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 136 module_param(updelay, int, 0); 137 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 138 module_param(downdelay, int, 0); 139 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 140 "in milliseconds"); 141 module_param(use_carrier, int, 0); 142 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 143 "0 for off, 1 for on (default)"); 144 module_param(mode, charp, 0); 145 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 146 "1 for active-backup, 2 for balance-xor, " 147 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 148 "6 for balance-alb"); 149 module_param(primary, charp, 0); 150 MODULE_PARM_DESC(primary, "Primary network device to use"); 151 module_param(primary_reselect, charp, 0); 152 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 153 "once it comes up; " 154 "0 for always (default), " 155 "1 for only if speed of primary is " 156 "better, " 157 "2 for only on active slave " 158 "failure"); 159 module_param(lacp_rate, charp, 0); 160 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 161 "0 for slow, 1 for fast"); 162 module_param(ad_select, charp, 0); 163 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 164 "0 for stable (default), 1 for bandwidth, " 165 "2 for count"); 166 module_param(min_links, int, 0); 167 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 168 169 module_param(xmit_hash_policy, charp, 0); 170 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 171 "0 for layer 2 (default), 1 for layer 3+4, " 172 "2 for layer 2+3, 3 for encap layer 2+3, " 173 "4 for encap layer 3+4, 5 for vlan+srcmac"); 174 module_param(arp_interval, int, 0); 175 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 176 module_param_array(arp_ip_target, charp, NULL, 0); 177 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 178 module_param(arp_validate, charp, 0); 179 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 180 "0 for none (default), 1 for active, " 181 "2 for backup, 3 for all"); 182 module_param(arp_all_targets, charp, 0); 183 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 184 module_param(fail_over_mac, charp, 0); 185 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 186 "the same MAC; 0 for none (default), " 187 "1 for active, 2 for follow"); 188 module_param(all_slaves_active, int, 0); 189 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 190 "by setting active flag for all slaves; " 191 "0 for never (default), 1 for always."); 192 module_param(resend_igmp, int, 0); 193 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 194 "link failure"); 195 module_param(packets_per_slave, int, 0); 196 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 197 "mode; 0 for a random slave, 1 packet per " 198 "slave (default), >1 packets per slave."); 199 module_param(lp_interval, uint, 0); 200 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 201 "the bonding driver sends learning packets to " 202 "each slaves peer switch. The default is 1."); 203 204 /*----------------------------- Global variables ----------------------------*/ 205 206 #ifdef CONFIG_NET_POLL_CONTROLLER 207 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 208 #endif 209 210 unsigned int bond_net_id __read_mostly; 211 212 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 213 { 214 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 215 .offset = offsetof(struct flow_keys, control), 216 }, 217 { 218 .key_id = FLOW_DISSECTOR_KEY_BASIC, 219 .offset = offsetof(struct flow_keys, basic), 220 }, 221 { 222 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 223 .offset = offsetof(struct flow_keys, addrs.v4addrs), 224 }, 225 { 226 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 227 .offset = offsetof(struct flow_keys, addrs.v6addrs), 228 }, 229 { 230 .key_id = FLOW_DISSECTOR_KEY_TIPC, 231 .offset = offsetof(struct flow_keys, addrs.tipckey), 232 }, 233 { 234 .key_id = FLOW_DISSECTOR_KEY_PORTS, 235 .offset = offsetof(struct flow_keys, ports), 236 }, 237 { 238 .key_id = FLOW_DISSECTOR_KEY_ICMP, 239 .offset = offsetof(struct flow_keys, icmp), 240 }, 241 { 242 .key_id = FLOW_DISSECTOR_KEY_VLAN, 243 .offset = offsetof(struct flow_keys, vlan), 244 }, 245 { 246 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 247 .offset = offsetof(struct flow_keys, tags), 248 }, 249 { 250 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 251 .offset = offsetof(struct flow_keys, keyid), 252 }, 253 }; 254 255 static struct flow_dissector flow_keys_bonding __read_mostly; 256 257 /*-------------------------- Forward declarations ---------------------------*/ 258 259 static int bond_init(struct net_device *bond_dev); 260 static void bond_uninit(struct net_device *bond_dev); 261 static void bond_get_stats(struct net_device *bond_dev, 262 struct rtnl_link_stats64 *stats); 263 static void bond_slave_arr_handler(struct work_struct *work); 264 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 265 int mod); 266 static void bond_netdev_notify_work(struct work_struct *work); 267 268 /*---------------------------- General routines -----------------------------*/ 269 270 const char *bond_mode_name(int mode) 271 { 272 static const char *names[] = { 273 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 274 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 275 [BOND_MODE_XOR] = "load balancing (xor)", 276 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 277 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 278 [BOND_MODE_TLB] = "transmit load balancing", 279 [BOND_MODE_ALB] = "adaptive load balancing", 280 }; 281 282 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 283 return "unknown"; 284 285 return names[mode]; 286 } 287 288 /** 289 * bond_dev_queue_xmit - Prepare skb for xmit. 290 * 291 * @bond: bond device that got this skb for tx. 292 * @skb: hw accel VLAN tagged skb to transmit 293 * @slave_dev: slave that is supposed to xmit this skbuff 294 */ 295 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 296 struct net_device *slave_dev) 297 { 298 skb->dev = slave_dev; 299 300 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 301 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 302 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 303 304 if (unlikely(netpoll_tx_running(bond->dev))) 305 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 306 307 return dev_queue_xmit(skb); 308 } 309 310 static bool bond_sk_check(struct bonding *bond) 311 { 312 switch (BOND_MODE(bond)) { 313 case BOND_MODE_8023AD: 314 case BOND_MODE_XOR: 315 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 316 return true; 317 fallthrough; 318 default: 319 return false; 320 } 321 } 322 323 static bool bond_xdp_check(struct bonding *bond) 324 { 325 switch (BOND_MODE(bond)) { 326 case BOND_MODE_ROUNDROBIN: 327 case BOND_MODE_ACTIVEBACKUP: 328 return true; 329 case BOND_MODE_8023AD: 330 case BOND_MODE_XOR: 331 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 332 * payload is not in the packet due to hardware offload. 333 */ 334 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 335 return true; 336 fallthrough; 337 default: 338 return false; 339 } 340 } 341 342 /*---------------------------------- VLAN -----------------------------------*/ 343 344 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 345 * We don't protect the slave list iteration with a lock because: 346 * a. This operation is performed in IOCTL context, 347 * b. The operation is protected by the RTNL semaphore in the 8021q code, 348 * c. Holding a lock with BH disabled while directly calling a base driver 349 * entry point is generally a BAD idea. 350 * 351 * The design of synchronization/protection for this operation in the 8021q 352 * module is good for one or more VLAN devices over a single physical device 353 * and cannot be extended for a teaming solution like bonding, so there is a 354 * potential race condition here where a net device from the vlan group might 355 * be referenced (either by a base driver or the 8021q code) while it is being 356 * removed from the system. However, it turns out we're not making matters 357 * worse, and if it works for regular VLAN usage it will work here too. 358 */ 359 360 /** 361 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 362 * @bond_dev: bonding net device that got called 363 * @proto: network protocol ID 364 * @vid: vlan id being added 365 */ 366 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 367 __be16 proto, u16 vid) 368 { 369 struct bonding *bond = netdev_priv(bond_dev); 370 struct slave *slave, *rollback_slave; 371 struct list_head *iter; 372 int res; 373 374 bond_for_each_slave(bond, slave, iter) { 375 res = vlan_vid_add(slave->dev, proto, vid); 376 if (res) 377 goto unwind; 378 } 379 380 return 0; 381 382 unwind: 383 /* unwind to the slave that failed */ 384 bond_for_each_slave(bond, rollback_slave, iter) { 385 if (rollback_slave == slave) 386 break; 387 388 vlan_vid_del(rollback_slave->dev, proto, vid); 389 } 390 391 return res; 392 } 393 394 /** 395 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 396 * @bond_dev: bonding net device that got called 397 * @proto: network protocol ID 398 * @vid: vlan id being removed 399 */ 400 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 401 __be16 proto, u16 vid) 402 { 403 struct bonding *bond = netdev_priv(bond_dev); 404 struct list_head *iter; 405 struct slave *slave; 406 407 bond_for_each_slave(bond, slave, iter) 408 vlan_vid_del(slave->dev, proto, vid); 409 410 if (bond_is_lb(bond)) 411 bond_alb_clear_vlan(bond, vid); 412 413 return 0; 414 } 415 416 /*---------------------------------- XFRM -----------------------------------*/ 417 418 #ifdef CONFIG_XFRM_OFFLOAD 419 /** 420 * bond_ipsec_add_sa - program device with a security association 421 * @xs: pointer to transformer state struct 422 **/ 423 static int bond_ipsec_add_sa(struct xfrm_state *xs) 424 { 425 struct net_device *bond_dev = xs->xso.dev; 426 struct bond_ipsec *ipsec; 427 struct bonding *bond; 428 struct slave *slave; 429 int err; 430 431 if (!bond_dev) 432 return -EINVAL; 433 434 rcu_read_lock(); 435 bond = netdev_priv(bond_dev); 436 slave = rcu_dereference(bond->curr_active_slave); 437 if (!slave) { 438 rcu_read_unlock(); 439 return -ENODEV; 440 } 441 442 if (!slave->dev->xfrmdev_ops || 443 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 444 netif_is_bond_master(slave->dev)) { 445 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n"); 446 rcu_read_unlock(); 447 return -EINVAL; 448 } 449 450 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC); 451 if (!ipsec) { 452 rcu_read_unlock(); 453 return -ENOMEM; 454 } 455 xs->xso.real_dev = slave->dev; 456 457 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs); 458 if (!err) { 459 ipsec->xs = xs; 460 INIT_LIST_HEAD(&ipsec->list); 461 spin_lock_bh(&bond->ipsec_lock); 462 list_add(&ipsec->list, &bond->ipsec_list); 463 spin_unlock_bh(&bond->ipsec_lock); 464 } else { 465 kfree(ipsec); 466 } 467 rcu_read_unlock(); 468 return err; 469 } 470 471 static void bond_ipsec_add_sa_all(struct bonding *bond) 472 { 473 struct net_device *bond_dev = bond->dev; 474 struct bond_ipsec *ipsec; 475 struct slave *slave; 476 477 rcu_read_lock(); 478 slave = rcu_dereference(bond->curr_active_slave); 479 if (!slave) 480 goto out; 481 482 if (!slave->dev->xfrmdev_ops || 483 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 484 netif_is_bond_master(slave->dev)) { 485 spin_lock_bh(&bond->ipsec_lock); 486 if (!list_empty(&bond->ipsec_list)) 487 slave_warn(bond_dev, slave->dev, 488 "%s: no slave xdo_dev_state_add\n", 489 __func__); 490 spin_unlock_bh(&bond->ipsec_lock); 491 goto out; 492 } 493 494 spin_lock_bh(&bond->ipsec_lock); 495 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 496 ipsec->xs->xso.real_dev = slave->dev; 497 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) { 498 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__); 499 ipsec->xs->xso.real_dev = NULL; 500 } 501 } 502 spin_unlock_bh(&bond->ipsec_lock); 503 out: 504 rcu_read_unlock(); 505 } 506 507 /** 508 * bond_ipsec_del_sa - clear out this specific SA 509 * @xs: pointer to transformer state struct 510 **/ 511 static void bond_ipsec_del_sa(struct xfrm_state *xs) 512 { 513 struct net_device *bond_dev = xs->xso.dev; 514 struct bond_ipsec *ipsec; 515 struct bonding *bond; 516 struct slave *slave; 517 518 if (!bond_dev) 519 return; 520 521 rcu_read_lock(); 522 bond = netdev_priv(bond_dev); 523 slave = rcu_dereference(bond->curr_active_slave); 524 525 if (!slave) 526 goto out; 527 528 if (!xs->xso.real_dev) 529 goto out; 530 531 WARN_ON(xs->xso.real_dev != slave->dev); 532 533 if (!slave->dev->xfrmdev_ops || 534 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 535 netif_is_bond_master(slave->dev)) { 536 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__); 537 goto out; 538 } 539 540 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs); 541 out: 542 spin_lock_bh(&bond->ipsec_lock); 543 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 544 if (ipsec->xs == xs) { 545 list_del(&ipsec->list); 546 kfree(ipsec); 547 break; 548 } 549 } 550 spin_unlock_bh(&bond->ipsec_lock); 551 rcu_read_unlock(); 552 } 553 554 static void bond_ipsec_del_sa_all(struct bonding *bond) 555 { 556 struct net_device *bond_dev = bond->dev; 557 struct bond_ipsec *ipsec; 558 struct slave *slave; 559 560 rcu_read_lock(); 561 slave = rcu_dereference(bond->curr_active_slave); 562 if (!slave) { 563 rcu_read_unlock(); 564 return; 565 } 566 567 spin_lock_bh(&bond->ipsec_lock); 568 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 569 if (!ipsec->xs->xso.real_dev) 570 continue; 571 572 if (!slave->dev->xfrmdev_ops || 573 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 574 netif_is_bond_master(slave->dev)) { 575 slave_warn(bond_dev, slave->dev, 576 "%s: no slave xdo_dev_state_delete\n", 577 __func__); 578 } else { 579 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 580 } 581 ipsec->xs->xso.real_dev = NULL; 582 } 583 spin_unlock_bh(&bond->ipsec_lock); 584 rcu_read_unlock(); 585 } 586 587 /** 588 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 589 * @skb: current data packet 590 * @xs: pointer to transformer state struct 591 **/ 592 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 593 { 594 struct net_device *bond_dev = xs->xso.dev; 595 struct net_device *real_dev; 596 struct slave *curr_active; 597 struct bonding *bond; 598 int err; 599 600 bond = netdev_priv(bond_dev); 601 rcu_read_lock(); 602 curr_active = rcu_dereference(bond->curr_active_slave); 603 real_dev = curr_active->dev; 604 605 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 606 err = false; 607 goto out; 608 } 609 610 if (!xs->xso.real_dev) { 611 err = false; 612 goto out; 613 } 614 615 if (!real_dev->xfrmdev_ops || 616 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 617 netif_is_bond_master(real_dev)) { 618 err = false; 619 goto out; 620 } 621 622 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 623 out: 624 rcu_read_unlock(); 625 return err; 626 } 627 628 static const struct xfrmdev_ops bond_xfrmdev_ops = { 629 .xdo_dev_state_add = bond_ipsec_add_sa, 630 .xdo_dev_state_delete = bond_ipsec_del_sa, 631 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 632 }; 633 #endif /* CONFIG_XFRM_OFFLOAD */ 634 635 /*------------------------------- Link status -------------------------------*/ 636 637 /* Set the carrier state for the master according to the state of its 638 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 639 * do special 802.3ad magic. 640 * 641 * Returns zero if carrier state does not change, nonzero if it does. 642 */ 643 int bond_set_carrier(struct bonding *bond) 644 { 645 struct list_head *iter; 646 struct slave *slave; 647 648 if (!bond_has_slaves(bond)) 649 goto down; 650 651 if (BOND_MODE(bond) == BOND_MODE_8023AD) 652 return bond_3ad_set_carrier(bond); 653 654 bond_for_each_slave(bond, slave, iter) { 655 if (slave->link == BOND_LINK_UP) { 656 if (!netif_carrier_ok(bond->dev)) { 657 netif_carrier_on(bond->dev); 658 return 1; 659 } 660 return 0; 661 } 662 } 663 664 down: 665 if (netif_carrier_ok(bond->dev)) { 666 netif_carrier_off(bond->dev); 667 return 1; 668 } 669 return 0; 670 } 671 672 /* Get link speed and duplex from the slave's base driver 673 * using ethtool. If for some reason the call fails or the 674 * values are invalid, set speed and duplex to -1, 675 * and return. Return 1 if speed or duplex settings are 676 * UNKNOWN; 0 otherwise. 677 */ 678 static int bond_update_speed_duplex(struct slave *slave) 679 { 680 struct net_device *slave_dev = slave->dev; 681 struct ethtool_link_ksettings ecmd; 682 int res; 683 684 slave->speed = SPEED_UNKNOWN; 685 slave->duplex = DUPLEX_UNKNOWN; 686 687 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 688 if (res < 0) 689 return 1; 690 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 691 return 1; 692 switch (ecmd.base.duplex) { 693 case DUPLEX_FULL: 694 case DUPLEX_HALF: 695 break; 696 default: 697 return 1; 698 } 699 700 slave->speed = ecmd.base.speed; 701 slave->duplex = ecmd.base.duplex; 702 703 return 0; 704 } 705 706 const char *bond_slave_link_status(s8 link) 707 { 708 switch (link) { 709 case BOND_LINK_UP: 710 return "up"; 711 case BOND_LINK_FAIL: 712 return "going down"; 713 case BOND_LINK_DOWN: 714 return "down"; 715 case BOND_LINK_BACK: 716 return "going back"; 717 default: 718 return "unknown"; 719 } 720 } 721 722 /* if <dev> supports MII link status reporting, check its link status. 723 * 724 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 725 * depending upon the setting of the use_carrier parameter. 726 * 727 * Return either BMSR_LSTATUS, meaning that the link is up (or we 728 * can't tell and just pretend it is), or 0, meaning that the link is 729 * down. 730 * 731 * If reporting is non-zero, instead of faking link up, return -1 if 732 * both ETHTOOL and MII ioctls fail (meaning the device does not 733 * support them). If use_carrier is set, return whatever it says. 734 * It'd be nice if there was a good way to tell if a driver supports 735 * netif_carrier, but there really isn't. 736 */ 737 static int bond_check_dev_link(struct bonding *bond, 738 struct net_device *slave_dev, int reporting) 739 { 740 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 741 int (*ioctl)(struct net_device *, struct ifreq *, int); 742 struct ifreq ifr; 743 struct mii_ioctl_data *mii; 744 745 if (!reporting && !netif_running(slave_dev)) 746 return 0; 747 748 if (bond->params.use_carrier) 749 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 750 751 /* Try to get link status using Ethtool first. */ 752 if (slave_dev->ethtool_ops->get_link) 753 return slave_dev->ethtool_ops->get_link(slave_dev) ? 754 BMSR_LSTATUS : 0; 755 756 /* Ethtool can't be used, fallback to MII ioctls. */ 757 ioctl = slave_ops->ndo_eth_ioctl; 758 if (ioctl) { 759 /* TODO: set pointer to correct ioctl on a per team member 760 * bases to make this more efficient. that is, once 761 * we determine the correct ioctl, we will always 762 * call it and not the others for that team 763 * member. 764 */ 765 766 /* We cannot assume that SIOCGMIIPHY will also read a 767 * register; not all network drivers (e.g., e100) 768 * support that. 769 */ 770 771 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 772 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 773 mii = if_mii(&ifr); 774 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 775 mii->reg_num = MII_BMSR; 776 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 777 return mii->val_out & BMSR_LSTATUS; 778 } 779 } 780 781 /* If reporting, report that either there's no ndo_eth_ioctl, 782 * or both SIOCGMIIREG and get_link failed (meaning that we 783 * cannot report link status). If not reporting, pretend 784 * we're ok. 785 */ 786 return reporting ? -1 : BMSR_LSTATUS; 787 } 788 789 /*----------------------------- Multicast list ------------------------------*/ 790 791 /* Push the promiscuity flag down to appropriate slaves */ 792 static int bond_set_promiscuity(struct bonding *bond, int inc) 793 { 794 struct list_head *iter; 795 int err = 0; 796 797 if (bond_uses_primary(bond)) { 798 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 799 800 if (curr_active) 801 err = dev_set_promiscuity(curr_active->dev, inc); 802 } else { 803 struct slave *slave; 804 805 bond_for_each_slave(bond, slave, iter) { 806 err = dev_set_promiscuity(slave->dev, inc); 807 if (err) 808 return err; 809 } 810 } 811 return err; 812 } 813 814 /* Push the allmulti flag down to all slaves */ 815 static int bond_set_allmulti(struct bonding *bond, int inc) 816 { 817 struct list_head *iter; 818 int err = 0; 819 820 if (bond_uses_primary(bond)) { 821 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 822 823 if (curr_active) 824 err = dev_set_allmulti(curr_active->dev, inc); 825 } else { 826 struct slave *slave; 827 828 bond_for_each_slave(bond, slave, iter) { 829 err = dev_set_allmulti(slave->dev, inc); 830 if (err) 831 return err; 832 } 833 } 834 return err; 835 } 836 837 /* Retrieve the list of registered multicast addresses for the bonding 838 * device and retransmit an IGMP JOIN request to the current active 839 * slave. 840 */ 841 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 842 { 843 struct bonding *bond = container_of(work, struct bonding, 844 mcast_work.work); 845 846 if (!rtnl_trylock()) { 847 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 848 return; 849 } 850 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 851 852 if (bond->igmp_retrans > 1) { 853 bond->igmp_retrans--; 854 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 855 } 856 rtnl_unlock(); 857 } 858 859 /* Flush bond's hardware addresses from slave */ 860 static void bond_hw_addr_flush(struct net_device *bond_dev, 861 struct net_device *slave_dev) 862 { 863 struct bonding *bond = netdev_priv(bond_dev); 864 865 dev_uc_unsync(slave_dev, bond_dev); 866 dev_mc_unsync(slave_dev, bond_dev); 867 868 if (BOND_MODE(bond) == BOND_MODE_8023AD) 869 dev_mc_del(slave_dev, lacpdu_mcast_addr); 870 } 871 872 /*--------------------------- Active slave change ---------------------------*/ 873 874 /* Update the hardware address list and promisc/allmulti for the new and 875 * old active slaves (if any). Modes that are not using primary keep all 876 * slaves up date at all times; only the modes that use primary need to call 877 * this function to swap these settings during a failover. 878 */ 879 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 880 struct slave *old_active) 881 { 882 if (old_active) { 883 if (bond->dev->flags & IFF_PROMISC) 884 dev_set_promiscuity(old_active->dev, -1); 885 886 if (bond->dev->flags & IFF_ALLMULTI) 887 dev_set_allmulti(old_active->dev, -1); 888 889 if (bond->dev->flags & IFF_UP) 890 bond_hw_addr_flush(bond->dev, old_active->dev); 891 } 892 893 if (new_active) { 894 /* FIXME: Signal errors upstream. */ 895 if (bond->dev->flags & IFF_PROMISC) 896 dev_set_promiscuity(new_active->dev, 1); 897 898 if (bond->dev->flags & IFF_ALLMULTI) 899 dev_set_allmulti(new_active->dev, 1); 900 901 if (bond->dev->flags & IFF_UP) { 902 netif_addr_lock_bh(bond->dev); 903 dev_uc_sync(new_active->dev, bond->dev); 904 dev_mc_sync(new_active->dev, bond->dev); 905 netif_addr_unlock_bh(bond->dev); 906 } 907 } 908 } 909 910 /** 911 * bond_set_dev_addr - clone slave's address to bond 912 * @bond_dev: bond net device 913 * @slave_dev: slave net device 914 * 915 * Should be called with RTNL held. 916 */ 917 static int bond_set_dev_addr(struct net_device *bond_dev, 918 struct net_device *slave_dev) 919 { 920 int err; 921 922 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 923 bond_dev, slave_dev, slave_dev->addr_len); 924 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 925 if (err) 926 return err; 927 928 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 929 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 930 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 931 return 0; 932 } 933 934 static struct slave *bond_get_old_active(struct bonding *bond, 935 struct slave *new_active) 936 { 937 struct slave *slave; 938 struct list_head *iter; 939 940 bond_for_each_slave(bond, slave, iter) { 941 if (slave == new_active) 942 continue; 943 944 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 945 return slave; 946 } 947 948 return NULL; 949 } 950 951 /* bond_do_fail_over_mac 952 * 953 * Perform special MAC address swapping for fail_over_mac settings 954 * 955 * Called with RTNL 956 */ 957 static void bond_do_fail_over_mac(struct bonding *bond, 958 struct slave *new_active, 959 struct slave *old_active) 960 { 961 u8 tmp_mac[MAX_ADDR_LEN]; 962 struct sockaddr_storage ss; 963 int rv; 964 965 switch (bond->params.fail_over_mac) { 966 case BOND_FOM_ACTIVE: 967 if (new_active) { 968 rv = bond_set_dev_addr(bond->dev, new_active->dev); 969 if (rv) 970 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 971 -rv); 972 } 973 break; 974 case BOND_FOM_FOLLOW: 975 /* if new_active && old_active, swap them 976 * if just old_active, do nothing (going to no active slave) 977 * if just new_active, set new_active to bond's MAC 978 */ 979 if (!new_active) 980 return; 981 982 if (!old_active) 983 old_active = bond_get_old_active(bond, new_active); 984 985 if (old_active) { 986 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 987 new_active->dev->addr_len); 988 bond_hw_addr_copy(ss.__data, 989 old_active->dev->dev_addr, 990 old_active->dev->addr_len); 991 ss.ss_family = new_active->dev->type; 992 } else { 993 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 994 bond->dev->addr_len); 995 ss.ss_family = bond->dev->type; 996 } 997 998 rv = dev_set_mac_address(new_active->dev, 999 (struct sockaddr *)&ss, NULL); 1000 if (rv) { 1001 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1002 -rv); 1003 goto out; 1004 } 1005 1006 if (!old_active) 1007 goto out; 1008 1009 bond_hw_addr_copy(ss.__data, tmp_mac, 1010 new_active->dev->addr_len); 1011 ss.ss_family = old_active->dev->type; 1012 1013 rv = dev_set_mac_address(old_active->dev, 1014 (struct sockaddr *)&ss, NULL); 1015 if (rv) 1016 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1017 -rv); 1018 out: 1019 break; 1020 default: 1021 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1022 bond->params.fail_over_mac); 1023 break; 1024 } 1025 1026 } 1027 1028 /** 1029 * bond_choose_primary_or_current - select the primary or high priority slave 1030 * @bond: our bonding struct 1031 * 1032 * - Check if there is a primary link. If the primary link was set and is up, 1033 * go on and do link reselection. 1034 * 1035 * - If primary link is not set or down, find the highest priority link. 1036 * If the highest priority link is not current slave, set it as primary 1037 * link and do link reselection. 1038 */ 1039 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1040 { 1041 struct slave *prim = rtnl_dereference(bond->primary_slave); 1042 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1043 struct slave *slave, *hprio = NULL; 1044 struct list_head *iter; 1045 1046 if (!prim || prim->link != BOND_LINK_UP) { 1047 bond_for_each_slave(bond, slave, iter) { 1048 if (slave->link == BOND_LINK_UP) { 1049 hprio = hprio ?: slave; 1050 if (slave->prio > hprio->prio) 1051 hprio = slave; 1052 } 1053 } 1054 1055 if (hprio && hprio != curr) { 1056 prim = hprio; 1057 goto link_reselect; 1058 } 1059 1060 if (!curr || curr->link != BOND_LINK_UP) 1061 return NULL; 1062 return curr; 1063 } 1064 1065 if (bond->force_primary) { 1066 bond->force_primary = false; 1067 return prim; 1068 } 1069 1070 link_reselect: 1071 if (!curr || curr->link != BOND_LINK_UP) 1072 return prim; 1073 1074 /* At this point, prim and curr are both up */ 1075 switch (bond->params.primary_reselect) { 1076 case BOND_PRI_RESELECT_ALWAYS: 1077 return prim; 1078 case BOND_PRI_RESELECT_BETTER: 1079 if (prim->speed < curr->speed) 1080 return curr; 1081 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1082 return curr; 1083 return prim; 1084 case BOND_PRI_RESELECT_FAILURE: 1085 return curr; 1086 default: 1087 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1088 bond->params.primary_reselect); 1089 return curr; 1090 } 1091 } 1092 1093 /** 1094 * bond_find_best_slave - select the best available slave to be the active one 1095 * @bond: our bonding struct 1096 */ 1097 static struct slave *bond_find_best_slave(struct bonding *bond) 1098 { 1099 struct slave *slave, *bestslave = NULL; 1100 struct list_head *iter; 1101 int mintime = bond->params.updelay; 1102 1103 slave = bond_choose_primary_or_current(bond); 1104 if (slave) 1105 return slave; 1106 1107 bond_for_each_slave(bond, slave, iter) { 1108 if (slave->link == BOND_LINK_UP) 1109 return slave; 1110 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1111 slave->delay < mintime) { 1112 mintime = slave->delay; 1113 bestslave = slave; 1114 } 1115 } 1116 1117 return bestslave; 1118 } 1119 1120 static bool bond_should_notify_peers(struct bonding *bond) 1121 { 1122 struct slave *slave; 1123 1124 rcu_read_lock(); 1125 slave = rcu_dereference(bond->curr_active_slave); 1126 rcu_read_unlock(); 1127 1128 if (!slave || !bond->send_peer_notif || 1129 bond->send_peer_notif % 1130 max(1, bond->params.peer_notif_delay) != 0 || 1131 !netif_carrier_ok(bond->dev) || 1132 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1133 return false; 1134 1135 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1136 slave ? slave->dev->name : "NULL"); 1137 1138 return true; 1139 } 1140 1141 /** 1142 * bond_change_active_slave - change the active slave into the specified one 1143 * @bond: our bonding struct 1144 * @new_active: the new slave to make the active one 1145 * 1146 * Set the new slave to the bond's settings and unset them on the old 1147 * curr_active_slave. 1148 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1149 * 1150 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1151 * because it is apparently the best available slave we have, even though its 1152 * updelay hasn't timed out yet. 1153 * 1154 * Caller must hold RTNL. 1155 */ 1156 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1157 { 1158 struct slave *old_active; 1159 1160 ASSERT_RTNL(); 1161 1162 old_active = rtnl_dereference(bond->curr_active_slave); 1163 1164 if (old_active == new_active) 1165 return; 1166 1167 #ifdef CONFIG_XFRM_OFFLOAD 1168 bond_ipsec_del_sa_all(bond); 1169 #endif /* CONFIG_XFRM_OFFLOAD */ 1170 1171 if (new_active) { 1172 new_active->last_link_up = jiffies; 1173 1174 if (new_active->link == BOND_LINK_BACK) { 1175 if (bond_uses_primary(bond)) { 1176 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1177 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1178 } 1179 1180 new_active->delay = 0; 1181 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1182 BOND_SLAVE_NOTIFY_NOW); 1183 1184 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1185 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1186 1187 if (bond_is_lb(bond)) 1188 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1189 } else { 1190 if (bond_uses_primary(bond)) 1191 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1192 } 1193 } 1194 1195 if (bond_uses_primary(bond)) 1196 bond_hw_addr_swap(bond, new_active, old_active); 1197 1198 if (bond_is_lb(bond)) { 1199 bond_alb_handle_active_change(bond, new_active); 1200 if (old_active) 1201 bond_set_slave_inactive_flags(old_active, 1202 BOND_SLAVE_NOTIFY_NOW); 1203 if (new_active) 1204 bond_set_slave_active_flags(new_active, 1205 BOND_SLAVE_NOTIFY_NOW); 1206 } else { 1207 rcu_assign_pointer(bond->curr_active_slave, new_active); 1208 } 1209 1210 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1211 if (old_active) 1212 bond_set_slave_inactive_flags(old_active, 1213 BOND_SLAVE_NOTIFY_NOW); 1214 1215 if (new_active) { 1216 bool should_notify_peers = false; 1217 1218 bond_set_slave_active_flags(new_active, 1219 BOND_SLAVE_NOTIFY_NOW); 1220 1221 if (bond->params.fail_over_mac) 1222 bond_do_fail_over_mac(bond, new_active, 1223 old_active); 1224 1225 if (netif_running(bond->dev)) { 1226 bond->send_peer_notif = 1227 bond->params.num_peer_notif * 1228 max(1, bond->params.peer_notif_delay); 1229 should_notify_peers = 1230 bond_should_notify_peers(bond); 1231 } 1232 1233 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1234 if (should_notify_peers) { 1235 bond->send_peer_notif--; 1236 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1237 bond->dev); 1238 } 1239 } 1240 } 1241 1242 #ifdef CONFIG_XFRM_OFFLOAD 1243 bond_ipsec_add_sa_all(bond); 1244 #endif /* CONFIG_XFRM_OFFLOAD */ 1245 1246 /* resend IGMP joins since active slave has changed or 1247 * all were sent on curr_active_slave. 1248 * resend only if bond is brought up with the affected 1249 * bonding modes and the retransmission is enabled 1250 */ 1251 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1252 ((bond_uses_primary(bond) && new_active) || 1253 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1254 bond->igmp_retrans = bond->params.resend_igmp; 1255 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1256 } 1257 } 1258 1259 /** 1260 * bond_select_active_slave - select a new active slave, if needed 1261 * @bond: our bonding struct 1262 * 1263 * This functions should be called when one of the following occurs: 1264 * - The old curr_active_slave has been released or lost its link. 1265 * - The primary_slave has got its link back. 1266 * - A slave has got its link back and there's no old curr_active_slave. 1267 * 1268 * Caller must hold RTNL. 1269 */ 1270 void bond_select_active_slave(struct bonding *bond) 1271 { 1272 struct slave *best_slave; 1273 int rv; 1274 1275 ASSERT_RTNL(); 1276 1277 best_slave = bond_find_best_slave(bond); 1278 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1279 bond_change_active_slave(bond, best_slave); 1280 rv = bond_set_carrier(bond); 1281 if (!rv) 1282 return; 1283 1284 if (netif_carrier_ok(bond->dev)) 1285 netdev_info(bond->dev, "active interface up!\n"); 1286 else 1287 netdev_info(bond->dev, "now running without any active interface!\n"); 1288 } 1289 } 1290 1291 #ifdef CONFIG_NET_POLL_CONTROLLER 1292 static inline int slave_enable_netpoll(struct slave *slave) 1293 { 1294 struct netpoll *np; 1295 int err = 0; 1296 1297 np = kzalloc(sizeof(*np), GFP_KERNEL); 1298 err = -ENOMEM; 1299 if (!np) 1300 goto out; 1301 1302 err = __netpoll_setup(np, slave->dev); 1303 if (err) { 1304 kfree(np); 1305 goto out; 1306 } 1307 slave->np = np; 1308 out: 1309 return err; 1310 } 1311 static inline void slave_disable_netpoll(struct slave *slave) 1312 { 1313 struct netpoll *np = slave->np; 1314 1315 if (!np) 1316 return; 1317 1318 slave->np = NULL; 1319 1320 __netpoll_free(np); 1321 } 1322 1323 static void bond_poll_controller(struct net_device *bond_dev) 1324 { 1325 struct bonding *bond = netdev_priv(bond_dev); 1326 struct slave *slave = NULL; 1327 struct list_head *iter; 1328 struct ad_info ad_info; 1329 1330 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1331 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1332 return; 1333 1334 bond_for_each_slave_rcu(bond, slave, iter) { 1335 if (!bond_slave_is_up(slave)) 1336 continue; 1337 1338 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1339 struct aggregator *agg = 1340 SLAVE_AD_INFO(slave)->port.aggregator; 1341 1342 if (agg && 1343 agg->aggregator_identifier != ad_info.aggregator_id) 1344 continue; 1345 } 1346 1347 netpoll_poll_dev(slave->dev); 1348 } 1349 } 1350 1351 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1352 { 1353 struct bonding *bond = netdev_priv(bond_dev); 1354 struct list_head *iter; 1355 struct slave *slave; 1356 1357 bond_for_each_slave(bond, slave, iter) 1358 if (bond_slave_is_up(slave)) 1359 slave_disable_netpoll(slave); 1360 } 1361 1362 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1363 { 1364 struct bonding *bond = netdev_priv(dev); 1365 struct list_head *iter; 1366 struct slave *slave; 1367 int err = 0; 1368 1369 bond_for_each_slave(bond, slave, iter) { 1370 err = slave_enable_netpoll(slave); 1371 if (err) { 1372 bond_netpoll_cleanup(dev); 1373 break; 1374 } 1375 } 1376 return err; 1377 } 1378 #else 1379 static inline int slave_enable_netpoll(struct slave *slave) 1380 { 1381 return 0; 1382 } 1383 static inline void slave_disable_netpoll(struct slave *slave) 1384 { 1385 } 1386 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1387 { 1388 } 1389 #endif 1390 1391 /*---------------------------------- IOCTL ----------------------------------*/ 1392 1393 static netdev_features_t bond_fix_features(struct net_device *dev, 1394 netdev_features_t features) 1395 { 1396 struct bonding *bond = netdev_priv(dev); 1397 struct list_head *iter; 1398 netdev_features_t mask; 1399 struct slave *slave; 1400 1401 mask = features; 1402 1403 features &= ~NETIF_F_ONE_FOR_ALL; 1404 features |= NETIF_F_ALL_FOR_ALL; 1405 1406 bond_for_each_slave(bond, slave, iter) { 1407 features = netdev_increment_features(features, 1408 slave->dev->features, 1409 mask); 1410 } 1411 features = netdev_add_tso_features(features, mask); 1412 1413 return features; 1414 } 1415 1416 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1417 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1418 NETIF_F_HIGHDMA | NETIF_F_LRO) 1419 1420 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1421 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1422 1423 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1424 NETIF_F_GSO_SOFTWARE) 1425 1426 1427 static void bond_compute_features(struct bonding *bond) 1428 { 1429 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1430 IFF_XMIT_DST_RELEASE_PERM; 1431 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1432 netdev_features_t enc_features = BOND_ENC_FEATURES; 1433 #ifdef CONFIG_XFRM_OFFLOAD 1434 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1435 #endif /* CONFIG_XFRM_OFFLOAD */ 1436 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1437 struct net_device *bond_dev = bond->dev; 1438 struct list_head *iter; 1439 struct slave *slave; 1440 unsigned short max_hard_header_len = ETH_HLEN; 1441 unsigned int tso_max_size = TSO_MAX_SIZE; 1442 u16 tso_max_segs = TSO_MAX_SEGS; 1443 1444 if (!bond_has_slaves(bond)) 1445 goto done; 1446 vlan_features &= NETIF_F_ALL_FOR_ALL; 1447 mpls_features &= NETIF_F_ALL_FOR_ALL; 1448 1449 bond_for_each_slave(bond, slave, iter) { 1450 vlan_features = netdev_increment_features(vlan_features, 1451 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1452 1453 enc_features = netdev_increment_features(enc_features, 1454 slave->dev->hw_enc_features, 1455 BOND_ENC_FEATURES); 1456 1457 #ifdef CONFIG_XFRM_OFFLOAD 1458 xfrm_features = netdev_increment_features(xfrm_features, 1459 slave->dev->hw_enc_features, 1460 BOND_XFRM_FEATURES); 1461 #endif /* CONFIG_XFRM_OFFLOAD */ 1462 1463 mpls_features = netdev_increment_features(mpls_features, 1464 slave->dev->mpls_features, 1465 BOND_MPLS_FEATURES); 1466 1467 dst_release_flag &= slave->dev->priv_flags; 1468 if (slave->dev->hard_header_len > max_hard_header_len) 1469 max_hard_header_len = slave->dev->hard_header_len; 1470 1471 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1472 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1473 } 1474 bond_dev->hard_header_len = max_hard_header_len; 1475 1476 done: 1477 bond_dev->vlan_features = vlan_features; 1478 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1479 NETIF_F_HW_VLAN_CTAG_TX | 1480 NETIF_F_HW_VLAN_STAG_TX; 1481 #ifdef CONFIG_XFRM_OFFLOAD 1482 bond_dev->hw_enc_features |= xfrm_features; 1483 #endif /* CONFIG_XFRM_OFFLOAD */ 1484 bond_dev->mpls_features = mpls_features; 1485 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1486 netif_set_tso_max_size(bond_dev, tso_max_size); 1487 1488 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1489 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1490 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1491 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1492 1493 netdev_change_features(bond_dev); 1494 } 1495 1496 static void bond_setup_by_slave(struct net_device *bond_dev, 1497 struct net_device *slave_dev) 1498 { 1499 bond_dev->header_ops = slave_dev->header_ops; 1500 1501 bond_dev->type = slave_dev->type; 1502 bond_dev->hard_header_len = slave_dev->hard_header_len; 1503 bond_dev->needed_headroom = slave_dev->needed_headroom; 1504 bond_dev->addr_len = slave_dev->addr_len; 1505 1506 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1507 slave_dev->addr_len); 1508 } 1509 1510 /* On bonding slaves other than the currently active slave, suppress 1511 * duplicates except for alb non-mcast/bcast. 1512 */ 1513 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1514 struct slave *slave, 1515 struct bonding *bond) 1516 { 1517 if (bond_is_slave_inactive(slave)) { 1518 if (BOND_MODE(bond) == BOND_MODE_ALB && 1519 skb->pkt_type != PACKET_BROADCAST && 1520 skb->pkt_type != PACKET_MULTICAST) 1521 return false; 1522 return true; 1523 } 1524 return false; 1525 } 1526 1527 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1528 { 1529 struct sk_buff *skb = *pskb; 1530 struct slave *slave; 1531 struct bonding *bond; 1532 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1533 struct slave *); 1534 int ret = RX_HANDLER_ANOTHER; 1535 1536 skb = skb_share_check(skb, GFP_ATOMIC); 1537 if (unlikely(!skb)) 1538 return RX_HANDLER_CONSUMED; 1539 1540 *pskb = skb; 1541 1542 slave = bond_slave_get_rcu(skb->dev); 1543 bond = slave->bond; 1544 1545 recv_probe = READ_ONCE(bond->recv_probe); 1546 if (recv_probe) { 1547 ret = recv_probe(skb, bond, slave); 1548 if (ret == RX_HANDLER_CONSUMED) { 1549 consume_skb(skb); 1550 return ret; 1551 } 1552 } 1553 1554 /* 1555 * For packets determined by bond_should_deliver_exact_match() call to 1556 * be suppressed we want to make an exception for link-local packets. 1557 * This is necessary for e.g. LLDP daemons to be able to monitor 1558 * inactive slave links without being forced to bind to them 1559 * explicitly. 1560 * 1561 * At the same time, packets that are passed to the bonding master 1562 * (including link-local ones) can have their originating interface 1563 * determined via PACKET_ORIGDEV socket option. 1564 */ 1565 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1566 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1567 return RX_HANDLER_PASS; 1568 return RX_HANDLER_EXACT; 1569 } 1570 1571 skb->dev = bond->dev; 1572 1573 if (BOND_MODE(bond) == BOND_MODE_ALB && 1574 netif_is_bridge_port(bond->dev) && 1575 skb->pkt_type == PACKET_HOST) { 1576 1577 if (unlikely(skb_cow_head(skb, 1578 skb->data - skb_mac_header(skb)))) { 1579 kfree_skb(skb); 1580 return RX_HANDLER_CONSUMED; 1581 } 1582 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1583 bond->dev->addr_len); 1584 } 1585 1586 return ret; 1587 } 1588 1589 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1590 { 1591 switch (BOND_MODE(bond)) { 1592 case BOND_MODE_ROUNDROBIN: 1593 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1594 case BOND_MODE_ACTIVEBACKUP: 1595 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1596 case BOND_MODE_BROADCAST: 1597 return NETDEV_LAG_TX_TYPE_BROADCAST; 1598 case BOND_MODE_XOR: 1599 case BOND_MODE_8023AD: 1600 return NETDEV_LAG_TX_TYPE_HASH; 1601 default: 1602 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1603 } 1604 } 1605 1606 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1607 enum netdev_lag_tx_type type) 1608 { 1609 if (type != NETDEV_LAG_TX_TYPE_HASH) 1610 return NETDEV_LAG_HASH_NONE; 1611 1612 switch (bond->params.xmit_policy) { 1613 case BOND_XMIT_POLICY_LAYER2: 1614 return NETDEV_LAG_HASH_L2; 1615 case BOND_XMIT_POLICY_LAYER34: 1616 return NETDEV_LAG_HASH_L34; 1617 case BOND_XMIT_POLICY_LAYER23: 1618 return NETDEV_LAG_HASH_L23; 1619 case BOND_XMIT_POLICY_ENCAP23: 1620 return NETDEV_LAG_HASH_E23; 1621 case BOND_XMIT_POLICY_ENCAP34: 1622 return NETDEV_LAG_HASH_E34; 1623 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1624 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1625 default: 1626 return NETDEV_LAG_HASH_UNKNOWN; 1627 } 1628 } 1629 1630 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1631 struct netlink_ext_ack *extack) 1632 { 1633 struct netdev_lag_upper_info lag_upper_info; 1634 enum netdev_lag_tx_type type; 1635 1636 type = bond_lag_tx_type(bond); 1637 lag_upper_info.tx_type = type; 1638 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1639 1640 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1641 &lag_upper_info, extack); 1642 } 1643 1644 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1645 { 1646 netdev_upper_dev_unlink(slave->dev, bond->dev); 1647 slave->dev->flags &= ~IFF_SLAVE; 1648 } 1649 1650 static void slave_kobj_release(struct kobject *kobj) 1651 { 1652 struct slave *slave = to_slave(kobj); 1653 struct bonding *bond = bond_get_bond_by_slave(slave); 1654 1655 cancel_delayed_work_sync(&slave->notify_work); 1656 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1657 kfree(SLAVE_AD_INFO(slave)); 1658 1659 kfree(slave); 1660 } 1661 1662 static struct kobj_type slave_ktype = { 1663 .release = slave_kobj_release, 1664 #ifdef CONFIG_SYSFS 1665 .sysfs_ops = &slave_sysfs_ops, 1666 #endif 1667 }; 1668 1669 static int bond_kobj_init(struct slave *slave) 1670 { 1671 int err; 1672 1673 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1674 &(slave->dev->dev.kobj), "bonding_slave"); 1675 if (err) 1676 kobject_put(&slave->kobj); 1677 1678 return err; 1679 } 1680 1681 static struct slave *bond_alloc_slave(struct bonding *bond, 1682 struct net_device *slave_dev) 1683 { 1684 struct slave *slave = NULL; 1685 1686 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1687 if (!slave) 1688 return NULL; 1689 1690 slave->bond = bond; 1691 slave->dev = slave_dev; 1692 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1693 1694 if (bond_kobj_init(slave)) 1695 return NULL; 1696 1697 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1698 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1699 GFP_KERNEL); 1700 if (!SLAVE_AD_INFO(slave)) { 1701 kobject_put(&slave->kobj); 1702 return NULL; 1703 } 1704 } 1705 1706 return slave; 1707 } 1708 1709 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1710 { 1711 info->bond_mode = BOND_MODE(bond); 1712 info->miimon = bond->params.miimon; 1713 info->num_slaves = bond->slave_cnt; 1714 } 1715 1716 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1717 { 1718 strcpy(info->slave_name, slave->dev->name); 1719 info->link = slave->link; 1720 info->state = bond_slave_state(slave); 1721 info->link_failure_count = slave->link_failure_count; 1722 } 1723 1724 static void bond_netdev_notify_work(struct work_struct *_work) 1725 { 1726 struct slave *slave = container_of(_work, struct slave, 1727 notify_work.work); 1728 1729 if (rtnl_trylock()) { 1730 struct netdev_bonding_info binfo; 1731 1732 bond_fill_ifslave(slave, &binfo.slave); 1733 bond_fill_ifbond(slave->bond, &binfo.master); 1734 netdev_bonding_info_change(slave->dev, &binfo); 1735 rtnl_unlock(); 1736 } else { 1737 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1738 } 1739 } 1740 1741 void bond_queue_slave_event(struct slave *slave) 1742 { 1743 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1744 } 1745 1746 void bond_lower_state_changed(struct slave *slave) 1747 { 1748 struct netdev_lag_lower_state_info info; 1749 1750 info.link_up = slave->link == BOND_LINK_UP || 1751 slave->link == BOND_LINK_FAIL; 1752 info.tx_enabled = bond_is_active_slave(slave); 1753 netdev_lower_state_changed(slave->dev, &info); 1754 } 1755 1756 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1757 if (extack) \ 1758 NL_SET_ERR_MSG(extack, errmsg); \ 1759 else \ 1760 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1761 } while (0) 1762 1763 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1764 if (extack) \ 1765 NL_SET_ERR_MSG(extack, errmsg); \ 1766 else \ 1767 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1768 } while (0) 1769 1770 /* enslave device <slave> to bond device <master> */ 1771 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1772 struct netlink_ext_ack *extack) 1773 { 1774 struct bonding *bond = netdev_priv(bond_dev); 1775 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1776 struct slave *new_slave = NULL, *prev_slave; 1777 struct sockaddr_storage ss; 1778 int link_reporting; 1779 int res = 0, i; 1780 1781 if (slave_dev->flags & IFF_MASTER && 1782 !netif_is_bond_master(slave_dev)) { 1783 BOND_NL_ERR(bond_dev, extack, 1784 "Device type (master device) cannot be enslaved"); 1785 return -EPERM; 1786 } 1787 1788 if (!bond->params.use_carrier && 1789 slave_dev->ethtool_ops->get_link == NULL && 1790 slave_ops->ndo_eth_ioctl == NULL) { 1791 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1792 } 1793 1794 /* already in-use? */ 1795 if (netdev_is_rx_handler_busy(slave_dev)) { 1796 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1797 "Device is in use and cannot be enslaved"); 1798 return -EBUSY; 1799 } 1800 1801 if (bond_dev == slave_dev) { 1802 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1803 return -EPERM; 1804 } 1805 1806 /* vlan challenged mutual exclusion */ 1807 /* no need to lock since we're protected by rtnl_lock */ 1808 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1809 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1810 if (vlan_uses_dev(bond_dev)) { 1811 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1812 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1813 return -EPERM; 1814 } else { 1815 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1816 } 1817 } else { 1818 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1819 } 1820 1821 if (slave_dev->features & NETIF_F_HW_ESP) 1822 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1823 1824 /* Old ifenslave binaries are no longer supported. These can 1825 * be identified with moderate accuracy by the state of the slave: 1826 * the current ifenslave will set the interface down prior to 1827 * enslaving it; the old ifenslave will not. 1828 */ 1829 if (slave_dev->flags & IFF_UP) { 1830 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1831 "Device can not be enslaved while up"); 1832 return -EPERM; 1833 } 1834 1835 /* set bonding device ether type by slave - bonding netdevices are 1836 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1837 * there is a need to override some of the type dependent attribs/funcs. 1838 * 1839 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1840 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1841 */ 1842 if (!bond_has_slaves(bond)) { 1843 if (bond_dev->type != slave_dev->type) { 1844 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1845 bond_dev->type, slave_dev->type); 1846 1847 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1848 bond_dev); 1849 res = notifier_to_errno(res); 1850 if (res) { 1851 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1852 return -EBUSY; 1853 } 1854 1855 /* Flush unicast and multicast addresses */ 1856 dev_uc_flush(bond_dev); 1857 dev_mc_flush(bond_dev); 1858 1859 if (slave_dev->type != ARPHRD_ETHER) 1860 bond_setup_by_slave(bond_dev, slave_dev); 1861 else { 1862 ether_setup(bond_dev); 1863 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1864 } 1865 1866 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1867 bond_dev); 1868 } 1869 } else if (bond_dev->type != slave_dev->type) { 1870 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1871 "Device type is different from other slaves"); 1872 return -EINVAL; 1873 } 1874 1875 if (slave_dev->type == ARPHRD_INFINIBAND && 1876 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1877 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1878 "Only active-backup mode is supported for infiniband slaves"); 1879 res = -EOPNOTSUPP; 1880 goto err_undo_flags; 1881 } 1882 1883 if (!slave_ops->ndo_set_mac_address || 1884 slave_dev->type == ARPHRD_INFINIBAND) { 1885 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1886 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1887 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1888 if (!bond_has_slaves(bond)) { 1889 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1890 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1891 } else { 1892 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1893 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1894 res = -EOPNOTSUPP; 1895 goto err_undo_flags; 1896 } 1897 } 1898 } 1899 1900 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1901 1902 /* If this is the first slave, then we need to set the master's hardware 1903 * address to be the same as the slave's. 1904 */ 1905 if (!bond_has_slaves(bond) && 1906 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1907 res = bond_set_dev_addr(bond->dev, slave_dev); 1908 if (res) 1909 goto err_undo_flags; 1910 } 1911 1912 new_slave = bond_alloc_slave(bond, slave_dev); 1913 if (!new_slave) { 1914 res = -ENOMEM; 1915 goto err_undo_flags; 1916 } 1917 1918 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1919 * is set via sysfs or module option if desired. 1920 */ 1921 new_slave->queue_id = 0; 1922 1923 /* Save slave's original mtu and then set it to match the bond */ 1924 new_slave->original_mtu = slave_dev->mtu; 1925 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1926 if (res) { 1927 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1928 goto err_free; 1929 } 1930 1931 /* Save slave's original ("permanent") mac address for modes 1932 * that need it, and for restoring it upon release, and then 1933 * set it to the master's address 1934 */ 1935 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1936 slave_dev->addr_len); 1937 1938 if (!bond->params.fail_over_mac || 1939 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1940 /* Set slave to master's mac address. The application already 1941 * set the master's mac address to that of the first slave 1942 */ 1943 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1944 ss.ss_family = slave_dev->type; 1945 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1946 extack); 1947 if (res) { 1948 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1949 goto err_restore_mtu; 1950 } 1951 } 1952 1953 /* set slave flag before open to prevent IPv6 addrconf */ 1954 slave_dev->flags |= IFF_SLAVE; 1955 1956 /* open the slave since the application closed it */ 1957 res = dev_open(slave_dev, extack); 1958 if (res) { 1959 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1960 goto err_restore_mac; 1961 } 1962 1963 slave_dev->priv_flags |= IFF_BONDING; 1964 /* initialize slave stats */ 1965 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1966 1967 if (bond_is_lb(bond)) { 1968 /* bond_alb_init_slave() must be called before all other stages since 1969 * it might fail and we do not want to have to undo everything 1970 */ 1971 res = bond_alb_init_slave(bond, new_slave); 1972 if (res) 1973 goto err_close; 1974 } 1975 1976 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1977 if (res) { 1978 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1979 goto err_close; 1980 } 1981 1982 prev_slave = bond_last_slave(bond); 1983 1984 new_slave->delay = 0; 1985 new_slave->link_failure_count = 0; 1986 1987 if (bond_update_speed_duplex(new_slave) && 1988 bond_needs_speed_duplex(bond)) 1989 new_slave->link = BOND_LINK_DOWN; 1990 1991 new_slave->last_rx = jiffies - 1992 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1993 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1994 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1995 1996 new_slave->last_tx = new_slave->last_rx; 1997 1998 if (bond->params.miimon && !bond->params.use_carrier) { 1999 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2000 2001 if ((link_reporting == -1) && !bond->params.arp_interval) { 2002 /* miimon is set but a bonded network driver 2003 * does not support ETHTOOL/MII and 2004 * arp_interval is not set. Note: if 2005 * use_carrier is enabled, we will never go 2006 * here (because netif_carrier is always 2007 * supported); thus, we don't need to change 2008 * the messages for netif_carrier. 2009 */ 2010 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"); 2011 } else if (link_reporting == -1) { 2012 /* unable get link status using mii/ethtool */ 2013 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"); 2014 } 2015 } 2016 2017 /* check for initial state */ 2018 new_slave->link = BOND_LINK_NOCHANGE; 2019 if (bond->params.miimon) { 2020 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2021 if (bond->params.updelay) { 2022 bond_set_slave_link_state(new_slave, 2023 BOND_LINK_BACK, 2024 BOND_SLAVE_NOTIFY_NOW); 2025 new_slave->delay = bond->params.updelay; 2026 } else { 2027 bond_set_slave_link_state(new_slave, 2028 BOND_LINK_UP, 2029 BOND_SLAVE_NOTIFY_NOW); 2030 } 2031 } else { 2032 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2033 BOND_SLAVE_NOTIFY_NOW); 2034 } 2035 } else if (bond->params.arp_interval) { 2036 bond_set_slave_link_state(new_slave, 2037 (netif_carrier_ok(slave_dev) ? 2038 BOND_LINK_UP : BOND_LINK_DOWN), 2039 BOND_SLAVE_NOTIFY_NOW); 2040 } else { 2041 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2042 BOND_SLAVE_NOTIFY_NOW); 2043 } 2044 2045 if (new_slave->link != BOND_LINK_DOWN) 2046 new_slave->last_link_up = jiffies; 2047 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2048 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2049 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2050 2051 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2052 /* if there is a primary slave, remember it */ 2053 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2054 rcu_assign_pointer(bond->primary_slave, new_slave); 2055 bond->force_primary = true; 2056 } 2057 } 2058 2059 switch (BOND_MODE(bond)) { 2060 case BOND_MODE_ACTIVEBACKUP: 2061 bond_set_slave_inactive_flags(new_slave, 2062 BOND_SLAVE_NOTIFY_NOW); 2063 break; 2064 case BOND_MODE_8023AD: 2065 /* in 802.3ad mode, the internal mechanism 2066 * will activate the slaves in the selected 2067 * aggregator 2068 */ 2069 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2070 /* if this is the first slave */ 2071 if (!prev_slave) { 2072 SLAVE_AD_INFO(new_slave)->id = 1; 2073 /* Initialize AD with the number of times that the AD timer is called in 1 second 2074 * can be called only after the mac address of the bond is set 2075 */ 2076 bond_3ad_initialize(bond); 2077 } else { 2078 SLAVE_AD_INFO(new_slave)->id = 2079 SLAVE_AD_INFO(prev_slave)->id + 1; 2080 } 2081 2082 bond_3ad_bind_slave(new_slave); 2083 break; 2084 case BOND_MODE_TLB: 2085 case BOND_MODE_ALB: 2086 bond_set_active_slave(new_slave); 2087 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2088 break; 2089 default: 2090 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2091 2092 /* always active in trunk mode */ 2093 bond_set_active_slave(new_slave); 2094 2095 /* In trunking mode there is little meaning to curr_active_slave 2096 * anyway (it holds no special properties of the bond device), 2097 * so we can change it without calling change_active_interface() 2098 */ 2099 if (!rcu_access_pointer(bond->curr_active_slave) && 2100 new_slave->link == BOND_LINK_UP) 2101 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2102 2103 break; 2104 } /* switch(bond_mode) */ 2105 2106 #ifdef CONFIG_NET_POLL_CONTROLLER 2107 if (bond->dev->npinfo) { 2108 if (slave_enable_netpoll(new_slave)) { 2109 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2110 res = -EBUSY; 2111 goto err_detach; 2112 } 2113 } 2114 #endif 2115 2116 if (!(bond_dev->features & NETIF_F_LRO)) 2117 dev_disable_lro(slave_dev); 2118 2119 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2120 new_slave); 2121 if (res) { 2122 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2123 goto err_detach; 2124 } 2125 2126 res = bond_master_upper_dev_link(bond, new_slave, extack); 2127 if (res) { 2128 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2129 goto err_unregister; 2130 } 2131 2132 bond_lower_state_changed(new_slave); 2133 2134 res = bond_sysfs_slave_add(new_slave); 2135 if (res) { 2136 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2137 goto err_upper_unlink; 2138 } 2139 2140 /* If the mode uses primary, then the following is handled by 2141 * bond_change_active_slave(). 2142 */ 2143 if (!bond_uses_primary(bond)) { 2144 /* set promiscuity level to new slave */ 2145 if (bond_dev->flags & IFF_PROMISC) { 2146 res = dev_set_promiscuity(slave_dev, 1); 2147 if (res) 2148 goto err_sysfs_del; 2149 } 2150 2151 /* set allmulti level to new slave */ 2152 if (bond_dev->flags & IFF_ALLMULTI) { 2153 res = dev_set_allmulti(slave_dev, 1); 2154 if (res) { 2155 if (bond_dev->flags & IFF_PROMISC) 2156 dev_set_promiscuity(slave_dev, -1); 2157 goto err_sysfs_del; 2158 } 2159 } 2160 2161 if (bond_dev->flags & IFF_UP) { 2162 netif_addr_lock_bh(bond_dev); 2163 dev_mc_sync_multiple(slave_dev, bond_dev); 2164 dev_uc_sync_multiple(slave_dev, bond_dev); 2165 netif_addr_unlock_bh(bond_dev); 2166 2167 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2168 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2169 } 2170 } 2171 2172 bond->slave_cnt++; 2173 bond_compute_features(bond); 2174 bond_set_carrier(bond); 2175 2176 if (bond_uses_primary(bond)) { 2177 block_netpoll_tx(); 2178 bond_select_active_slave(bond); 2179 unblock_netpoll_tx(); 2180 } 2181 2182 if (bond_mode_can_use_xmit_hash(bond)) 2183 bond_update_slave_arr(bond, NULL); 2184 2185 2186 if (!slave_dev->netdev_ops->ndo_bpf || 2187 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2188 if (bond->xdp_prog) { 2189 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2190 "Slave does not support XDP"); 2191 res = -EOPNOTSUPP; 2192 goto err_sysfs_del; 2193 } 2194 } else if (bond->xdp_prog) { 2195 struct netdev_bpf xdp = { 2196 .command = XDP_SETUP_PROG, 2197 .flags = 0, 2198 .prog = bond->xdp_prog, 2199 .extack = extack, 2200 }; 2201 2202 if (dev_xdp_prog_count(slave_dev) > 0) { 2203 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2204 "Slave has XDP program loaded, please unload before enslaving"); 2205 res = -EOPNOTSUPP; 2206 goto err_sysfs_del; 2207 } 2208 2209 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 2210 if (res < 0) { 2211 /* ndo_bpf() sets extack error message */ 2212 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2213 goto err_sysfs_del; 2214 } 2215 if (bond->xdp_prog) 2216 bpf_prog_inc(bond->xdp_prog); 2217 } 2218 2219 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2220 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2221 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2222 2223 /* enslave is successful */ 2224 bond_queue_slave_event(new_slave); 2225 return 0; 2226 2227 /* Undo stages on error */ 2228 err_sysfs_del: 2229 bond_sysfs_slave_del(new_slave); 2230 2231 err_upper_unlink: 2232 bond_upper_dev_unlink(bond, new_slave); 2233 2234 err_unregister: 2235 netdev_rx_handler_unregister(slave_dev); 2236 2237 err_detach: 2238 vlan_vids_del_by_dev(slave_dev, bond_dev); 2239 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2240 RCU_INIT_POINTER(bond->primary_slave, NULL); 2241 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2242 block_netpoll_tx(); 2243 bond_change_active_slave(bond, NULL); 2244 bond_select_active_slave(bond); 2245 unblock_netpoll_tx(); 2246 } 2247 /* either primary_slave or curr_active_slave might've changed */ 2248 synchronize_rcu(); 2249 slave_disable_netpoll(new_slave); 2250 2251 err_close: 2252 if (!netif_is_bond_master(slave_dev)) 2253 slave_dev->priv_flags &= ~IFF_BONDING; 2254 dev_close(slave_dev); 2255 2256 err_restore_mac: 2257 slave_dev->flags &= ~IFF_SLAVE; 2258 if (!bond->params.fail_over_mac || 2259 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2260 /* XXX TODO - fom follow mode needs to change master's 2261 * MAC if this slave's MAC is in use by the bond, or at 2262 * least print a warning. 2263 */ 2264 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2265 new_slave->dev->addr_len); 2266 ss.ss_family = slave_dev->type; 2267 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2268 } 2269 2270 err_restore_mtu: 2271 dev_set_mtu(slave_dev, new_slave->original_mtu); 2272 2273 err_free: 2274 kobject_put(&new_slave->kobj); 2275 2276 err_undo_flags: 2277 /* Enslave of first slave has failed and we need to fix master's mac */ 2278 if (!bond_has_slaves(bond)) { 2279 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2280 slave_dev->dev_addr)) 2281 eth_hw_addr_random(bond_dev); 2282 if (bond_dev->type != ARPHRD_ETHER) { 2283 dev_close(bond_dev); 2284 ether_setup(bond_dev); 2285 bond_dev->flags |= IFF_MASTER; 2286 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2287 } 2288 } 2289 2290 return res; 2291 } 2292 2293 /* Try to release the slave device <slave> from the bond device <master> 2294 * It is legal to access curr_active_slave without a lock because all the function 2295 * is RTNL-locked. If "all" is true it means that the function is being called 2296 * while destroying a bond interface and all slaves are being released. 2297 * 2298 * The rules for slave state should be: 2299 * for Active/Backup: 2300 * Active stays on all backups go down 2301 * for Bonded connections: 2302 * The first up interface should be left on and all others downed. 2303 */ 2304 static int __bond_release_one(struct net_device *bond_dev, 2305 struct net_device *slave_dev, 2306 bool all, bool unregister) 2307 { 2308 struct bonding *bond = netdev_priv(bond_dev); 2309 struct slave *slave, *oldcurrent; 2310 struct sockaddr_storage ss; 2311 int old_flags = bond_dev->flags; 2312 netdev_features_t old_features = bond_dev->features; 2313 2314 /* slave is not a slave or master is not master of this slave */ 2315 if (!(slave_dev->flags & IFF_SLAVE) || 2316 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2317 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2318 return -EINVAL; 2319 } 2320 2321 block_netpoll_tx(); 2322 2323 slave = bond_get_slave_by_dev(bond, slave_dev); 2324 if (!slave) { 2325 /* not a slave of this bond */ 2326 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2327 unblock_netpoll_tx(); 2328 return -EINVAL; 2329 } 2330 2331 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2332 2333 bond_sysfs_slave_del(slave); 2334 2335 /* recompute stats just before removing the slave */ 2336 bond_get_stats(bond->dev, &bond->bond_stats); 2337 2338 if (bond->xdp_prog) { 2339 struct netdev_bpf xdp = { 2340 .command = XDP_SETUP_PROG, 2341 .flags = 0, 2342 .prog = NULL, 2343 .extack = NULL, 2344 }; 2345 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp)) 2346 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2347 } 2348 2349 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2350 * for this slave anymore. 2351 */ 2352 netdev_rx_handler_unregister(slave_dev); 2353 2354 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2355 bond_3ad_unbind_slave(slave); 2356 2357 bond_upper_dev_unlink(bond, slave); 2358 2359 if (bond_mode_can_use_xmit_hash(bond)) 2360 bond_update_slave_arr(bond, slave); 2361 2362 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2363 bond_is_active_slave(slave) ? "active" : "backup"); 2364 2365 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2366 2367 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2368 2369 if (!all && (!bond->params.fail_over_mac || 2370 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2371 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2372 bond_has_slaves(bond)) 2373 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", 2374 slave->perm_hwaddr); 2375 } 2376 2377 if (rtnl_dereference(bond->primary_slave) == slave) 2378 RCU_INIT_POINTER(bond->primary_slave, NULL); 2379 2380 if (oldcurrent == slave) 2381 bond_change_active_slave(bond, NULL); 2382 2383 if (bond_is_lb(bond)) { 2384 /* Must be called only after the slave has been 2385 * detached from the list and the curr_active_slave 2386 * has been cleared (if our_slave == old_current), 2387 * but before a new active slave is selected. 2388 */ 2389 bond_alb_deinit_slave(bond, slave); 2390 } 2391 2392 if (all) { 2393 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2394 } else if (oldcurrent == slave) { 2395 /* Note that we hold RTNL over this sequence, so there 2396 * is no concern that another slave add/remove event 2397 * will interfere. 2398 */ 2399 bond_select_active_slave(bond); 2400 } 2401 2402 bond_set_carrier(bond); 2403 if (!bond_has_slaves(bond)) 2404 eth_hw_addr_random(bond_dev); 2405 2406 unblock_netpoll_tx(); 2407 synchronize_rcu(); 2408 bond->slave_cnt--; 2409 2410 if (!bond_has_slaves(bond)) { 2411 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2412 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2413 } 2414 2415 bond_compute_features(bond); 2416 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2417 (old_features & NETIF_F_VLAN_CHALLENGED)) 2418 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2419 2420 vlan_vids_del_by_dev(slave_dev, bond_dev); 2421 2422 /* If the mode uses primary, then this case was handled above by 2423 * bond_change_active_slave(..., NULL) 2424 */ 2425 if (!bond_uses_primary(bond)) { 2426 /* unset promiscuity level from slave 2427 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2428 * of the IFF_PROMISC flag in the bond_dev, but we need the 2429 * value of that flag before that change, as that was the value 2430 * when this slave was attached, so we cache at the start of the 2431 * function and use it here. Same goes for ALLMULTI below 2432 */ 2433 if (old_flags & IFF_PROMISC) 2434 dev_set_promiscuity(slave_dev, -1); 2435 2436 /* unset allmulti level from slave */ 2437 if (old_flags & IFF_ALLMULTI) 2438 dev_set_allmulti(slave_dev, -1); 2439 2440 if (old_flags & IFF_UP) 2441 bond_hw_addr_flush(bond_dev, slave_dev); 2442 } 2443 2444 slave_disable_netpoll(slave); 2445 2446 /* close slave before restoring its mac address */ 2447 dev_close(slave_dev); 2448 2449 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2450 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2451 /* restore original ("permanent") mac address */ 2452 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2453 slave->dev->addr_len); 2454 ss.ss_family = slave_dev->type; 2455 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2456 } 2457 2458 if (unregister) 2459 __dev_set_mtu(slave_dev, slave->original_mtu); 2460 else 2461 dev_set_mtu(slave_dev, slave->original_mtu); 2462 2463 if (!netif_is_bond_master(slave_dev)) 2464 slave_dev->priv_flags &= ~IFF_BONDING; 2465 2466 kobject_put(&slave->kobj); 2467 2468 return 0; 2469 } 2470 2471 /* A wrapper used because of ndo_del_link */ 2472 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2473 { 2474 return __bond_release_one(bond_dev, slave_dev, false, false); 2475 } 2476 2477 /* First release a slave and then destroy the bond if no more slaves are left. 2478 * Must be under rtnl_lock when this function is called. 2479 */ 2480 static int bond_release_and_destroy(struct net_device *bond_dev, 2481 struct net_device *slave_dev) 2482 { 2483 struct bonding *bond = netdev_priv(bond_dev); 2484 int ret; 2485 2486 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2487 if (ret == 0 && !bond_has_slaves(bond) && 2488 bond_dev->reg_state != NETREG_UNREGISTERING) { 2489 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2490 netdev_info(bond_dev, "Destroying bond\n"); 2491 bond_remove_proc_entry(bond); 2492 unregister_netdevice(bond_dev); 2493 } 2494 return ret; 2495 } 2496 2497 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2498 { 2499 struct bonding *bond = netdev_priv(bond_dev); 2500 2501 bond_fill_ifbond(bond, info); 2502 } 2503 2504 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2505 { 2506 struct bonding *bond = netdev_priv(bond_dev); 2507 struct list_head *iter; 2508 int i = 0, res = -ENODEV; 2509 struct slave *slave; 2510 2511 bond_for_each_slave(bond, slave, iter) { 2512 if (i++ == (int)info->slave_id) { 2513 res = 0; 2514 bond_fill_ifslave(slave, info); 2515 break; 2516 } 2517 } 2518 2519 return res; 2520 } 2521 2522 /*-------------------------------- Monitoring -------------------------------*/ 2523 2524 /* called with rcu_read_lock() */ 2525 static int bond_miimon_inspect(struct bonding *bond) 2526 { 2527 int link_state, commit = 0; 2528 struct list_head *iter; 2529 struct slave *slave; 2530 bool ignore_updelay; 2531 2532 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2533 2534 bond_for_each_slave_rcu(bond, slave, iter) { 2535 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2536 2537 link_state = bond_check_dev_link(bond, slave->dev, 0); 2538 2539 switch (slave->link) { 2540 case BOND_LINK_UP: 2541 if (link_state) 2542 continue; 2543 2544 bond_propose_link_state(slave, BOND_LINK_FAIL); 2545 commit++; 2546 slave->delay = bond->params.downdelay; 2547 if (slave->delay) { 2548 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2549 (BOND_MODE(bond) == 2550 BOND_MODE_ACTIVEBACKUP) ? 2551 (bond_is_active_slave(slave) ? 2552 "active " : "backup ") : "", 2553 bond->params.downdelay * bond->params.miimon); 2554 } 2555 fallthrough; 2556 case BOND_LINK_FAIL: 2557 if (link_state) { 2558 /* recovered before downdelay expired */ 2559 bond_propose_link_state(slave, BOND_LINK_UP); 2560 slave->last_link_up = jiffies; 2561 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2562 (bond->params.downdelay - slave->delay) * 2563 bond->params.miimon); 2564 commit++; 2565 continue; 2566 } 2567 2568 if (slave->delay <= 0) { 2569 bond_propose_link_state(slave, BOND_LINK_DOWN); 2570 commit++; 2571 continue; 2572 } 2573 2574 slave->delay--; 2575 break; 2576 2577 case BOND_LINK_DOWN: 2578 if (!link_state) 2579 continue; 2580 2581 bond_propose_link_state(slave, BOND_LINK_BACK); 2582 commit++; 2583 slave->delay = bond->params.updelay; 2584 2585 if (slave->delay) { 2586 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2587 ignore_updelay ? 0 : 2588 bond->params.updelay * 2589 bond->params.miimon); 2590 } 2591 fallthrough; 2592 case BOND_LINK_BACK: 2593 if (!link_state) { 2594 bond_propose_link_state(slave, BOND_LINK_DOWN); 2595 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2596 (bond->params.updelay - slave->delay) * 2597 bond->params.miimon); 2598 commit++; 2599 continue; 2600 } 2601 2602 if (ignore_updelay) 2603 slave->delay = 0; 2604 2605 if (slave->delay <= 0) { 2606 bond_propose_link_state(slave, BOND_LINK_UP); 2607 commit++; 2608 ignore_updelay = false; 2609 continue; 2610 } 2611 2612 slave->delay--; 2613 break; 2614 } 2615 } 2616 2617 return commit; 2618 } 2619 2620 static void bond_miimon_link_change(struct bonding *bond, 2621 struct slave *slave, 2622 char link) 2623 { 2624 switch (BOND_MODE(bond)) { 2625 case BOND_MODE_8023AD: 2626 bond_3ad_handle_link_change(slave, link); 2627 break; 2628 case BOND_MODE_TLB: 2629 case BOND_MODE_ALB: 2630 bond_alb_handle_link_change(bond, slave, link); 2631 break; 2632 case BOND_MODE_XOR: 2633 bond_update_slave_arr(bond, NULL); 2634 break; 2635 } 2636 } 2637 2638 static void bond_miimon_commit(struct bonding *bond) 2639 { 2640 struct list_head *iter; 2641 struct slave *slave, *primary; 2642 2643 bond_for_each_slave(bond, slave, iter) { 2644 switch (slave->link_new_state) { 2645 case BOND_LINK_NOCHANGE: 2646 /* For 802.3ad mode, check current slave speed and 2647 * duplex again in case its port was disabled after 2648 * invalid speed/duplex reporting but recovered before 2649 * link monitoring could make a decision on the actual 2650 * link status 2651 */ 2652 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2653 slave->link == BOND_LINK_UP) 2654 bond_3ad_adapter_speed_duplex_changed(slave); 2655 continue; 2656 2657 case BOND_LINK_UP: 2658 if (bond_update_speed_duplex(slave) && 2659 bond_needs_speed_duplex(bond)) { 2660 slave->link = BOND_LINK_DOWN; 2661 if (net_ratelimit()) 2662 slave_warn(bond->dev, slave->dev, 2663 "failed to get link speed/duplex\n"); 2664 continue; 2665 } 2666 bond_set_slave_link_state(slave, BOND_LINK_UP, 2667 BOND_SLAVE_NOTIFY_NOW); 2668 slave->last_link_up = jiffies; 2669 2670 primary = rtnl_dereference(bond->primary_slave); 2671 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2672 /* prevent it from being the active one */ 2673 bond_set_backup_slave(slave); 2674 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2675 /* make it immediately active */ 2676 bond_set_active_slave(slave); 2677 } 2678 2679 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2680 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2681 slave->duplex ? "full" : "half"); 2682 2683 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2684 2685 if (!bond->curr_active_slave || slave == primary) 2686 goto do_failover; 2687 2688 continue; 2689 2690 case BOND_LINK_DOWN: 2691 if (slave->link_failure_count < UINT_MAX) 2692 slave->link_failure_count++; 2693 2694 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2695 BOND_SLAVE_NOTIFY_NOW); 2696 2697 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2698 BOND_MODE(bond) == BOND_MODE_8023AD) 2699 bond_set_slave_inactive_flags(slave, 2700 BOND_SLAVE_NOTIFY_NOW); 2701 2702 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2703 2704 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2705 2706 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2707 goto do_failover; 2708 2709 continue; 2710 2711 default: 2712 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2713 slave->link_new_state); 2714 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2715 2716 continue; 2717 } 2718 2719 do_failover: 2720 block_netpoll_tx(); 2721 bond_select_active_slave(bond); 2722 unblock_netpoll_tx(); 2723 } 2724 2725 bond_set_carrier(bond); 2726 } 2727 2728 /* bond_mii_monitor 2729 * 2730 * Really a wrapper that splits the mii monitor into two phases: an 2731 * inspection, then (if inspection indicates something needs to be done) 2732 * an acquisition of appropriate locks followed by a commit phase to 2733 * implement whatever link state changes are indicated. 2734 */ 2735 static void bond_mii_monitor(struct work_struct *work) 2736 { 2737 struct bonding *bond = container_of(work, struct bonding, 2738 mii_work.work); 2739 bool should_notify_peers = false; 2740 bool commit; 2741 unsigned long delay; 2742 struct slave *slave; 2743 struct list_head *iter; 2744 2745 delay = msecs_to_jiffies(bond->params.miimon); 2746 2747 if (!bond_has_slaves(bond)) 2748 goto re_arm; 2749 2750 rcu_read_lock(); 2751 should_notify_peers = bond_should_notify_peers(bond); 2752 commit = !!bond_miimon_inspect(bond); 2753 if (bond->send_peer_notif) { 2754 rcu_read_unlock(); 2755 if (rtnl_trylock()) { 2756 bond->send_peer_notif--; 2757 rtnl_unlock(); 2758 } 2759 } else { 2760 rcu_read_unlock(); 2761 } 2762 2763 if (commit) { 2764 /* Race avoidance with bond_close cancel of workqueue */ 2765 if (!rtnl_trylock()) { 2766 delay = 1; 2767 should_notify_peers = false; 2768 goto re_arm; 2769 } 2770 2771 bond_for_each_slave(bond, slave, iter) { 2772 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2773 } 2774 bond_miimon_commit(bond); 2775 2776 rtnl_unlock(); /* might sleep, hold no other locks */ 2777 } 2778 2779 re_arm: 2780 if (bond->params.miimon) 2781 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2782 2783 if (should_notify_peers) { 2784 if (!rtnl_trylock()) 2785 return; 2786 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2787 rtnl_unlock(); 2788 } 2789 } 2790 2791 static int bond_upper_dev_walk(struct net_device *upper, 2792 struct netdev_nested_priv *priv) 2793 { 2794 __be32 ip = *(__be32 *)priv->data; 2795 2796 return ip == bond_confirm_addr(upper, 0, ip); 2797 } 2798 2799 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2800 { 2801 struct netdev_nested_priv priv = { 2802 .data = (void *)&ip, 2803 }; 2804 bool ret = false; 2805 2806 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2807 return true; 2808 2809 rcu_read_lock(); 2810 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2811 ret = true; 2812 rcu_read_unlock(); 2813 2814 return ret; 2815 } 2816 2817 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2818 struct sk_buff *skb) 2819 { 2820 struct net_device *bond_dev = slave->bond->dev; 2821 struct net_device *slave_dev = slave->dev; 2822 struct bond_vlan_tag *outer_tag = tags; 2823 2824 if (!tags || tags->vlan_proto == VLAN_N_VID) 2825 return true; 2826 2827 tags++; 2828 2829 /* Go through all the tags backwards and add them to the packet */ 2830 while (tags->vlan_proto != VLAN_N_VID) { 2831 if (!tags->vlan_id) { 2832 tags++; 2833 continue; 2834 } 2835 2836 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2837 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2838 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2839 tags->vlan_id); 2840 if (!skb) { 2841 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2842 return false; 2843 } 2844 2845 tags++; 2846 } 2847 /* Set the outer tag */ 2848 if (outer_tag->vlan_id) { 2849 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2850 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2851 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2852 outer_tag->vlan_id); 2853 } 2854 2855 return true; 2856 } 2857 2858 /* We go to the (large) trouble of VLAN tagging ARP frames because 2859 * switches in VLAN mode (especially if ports are configured as 2860 * "native" to a VLAN) might not pass non-tagged frames. 2861 */ 2862 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2863 __be32 src_ip, struct bond_vlan_tag *tags) 2864 { 2865 struct net_device *bond_dev = slave->bond->dev; 2866 struct net_device *slave_dev = slave->dev; 2867 struct sk_buff *skb; 2868 2869 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2870 arp_op, &dest_ip, &src_ip); 2871 2872 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2873 NULL, slave_dev->dev_addr, NULL); 2874 2875 if (!skb) { 2876 net_err_ratelimited("ARP packet allocation failed\n"); 2877 return; 2878 } 2879 2880 if (bond_handle_vlan(slave, tags, skb)) { 2881 slave_update_last_tx(slave); 2882 arp_xmit(skb); 2883 } 2884 2885 return; 2886 } 2887 2888 /* Validate the device path between the @start_dev and the @end_dev. 2889 * The path is valid if the @end_dev is reachable through device 2890 * stacking. 2891 * When the path is validated, collect any vlan information in the 2892 * path. 2893 */ 2894 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2895 struct net_device *end_dev, 2896 int level) 2897 { 2898 struct bond_vlan_tag *tags; 2899 struct net_device *upper; 2900 struct list_head *iter; 2901 2902 if (start_dev == end_dev) { 2903 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2904 if (!tags) 2905 return ERR_PTR(-ENOMEM); 2906 tags[level].vlan_proto = VLAN_N_VID; 2907 return tags; 2908 } 2909 2910 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2911 tags = bond_verify_device_path(upper, end_dev, level + 1); 2912 if (IS_ERR_OR_NULL(tags)) { 2913 if (IS_ERR(tags)) 2914 return tags; 2915 continue; 2916 } 2917 if (is_vlan_dev(upper)) { 2918 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2919 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2920 } 2921 2922 return tags; 2923 } 2924 2925 return NULL; 2926 } 2927 2928 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2929 { 2930 struct rtable *rt; 2931 struct bond_vlan_tag *tags; 2932 __be32 *targets = bond->params.arp_targets, addr; 2933 int i; 2934 2935 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2936 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2937 __func__, &targets[i]); 2938 tags = NULL; 2939 2940 /* Find out through which dev should the packet go */ 2941 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2942 RTO_ONLINK, 0); 2943 if (IS_ERR(rt)) { 2944 /* there's no route to target - try to send arp 2945 * probe to generate any traffic (arp_validate=0) 2946 */ 2947 if (bond->params.arp_validate) 2948 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2949 bond->dev->name, 2950 &targets[i]); 2951 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2952 0, tags); 2953 continue; 2954 } 2955 2956 /* bond device itself */ 2957 if (rt->dst.dev == bond->dev) 2958 goto found; 2959 2960 rcu_read_lock(); 2961 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2962 rcu_read_unlock(); 2963 2964 if (!IS_ERR_OR_NULL(tags)) 2965 goto found; 2966 2967 /* Not our device - skip */ 2968 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2969 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2970 2971 ip_rt_put(rt); 2972 continue; 2973 2974 found: 2975 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2976 ip_rt_put(rt); 2977 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2978 kfree(tags); 2979 } 2980 } 2981 2982 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2983 { 2984 int i; 2985 2986 if (!sip || !bond_has_this_ip(bond, tip)) { 2987 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2988 __func__, &sip, &tip); 2989 return; 2990 } 2991 2992 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2993 if (i == -1) { 2994 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2995 __func__, &sip); 2996 return; 2997 } 2998 slave->last_rx = jiffies; 2999 slave->target_last_arp_rx[i] = jiffies; 3000 } 3001 3002 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3003 struct slave *slave) 3004 { 3005 struct arphdr *arp = (struct arphdr *)skb->data; 3006 struct slave *curr_active_slave, *curr_arp_slave; 3007 unsigned char *arp_ptr; 3008 __be32 sip, tip; 3009 unsigned int alen; 3010 3011 alen = arp_hdr_len(bond->dev); 3012 3013 if (alen > skb_headlen(skb)) { 3014 arp = kmalloc(alen, GFP_ATOMIC); 3015 if (!arp) 3016 goto out_unlock; 3017 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3018 goto out_unlock; 3019 } 3020 3021 if (arp->ar_hln != bond->dev->addr_len || 3022 skb->pkt_type == PACKET_OTHERHOST || 3023 skb->pkt_type == PACKET_LOOPBACK || 3024 arp->ar_hrd != htons(ARPHRD_ETHER) || 3025 arp->ar_pro != htons(ETH_P_IP) || 3026 arp->ar_pln != 4) 3027 goto out_unlock; 3028 3029 arp_ptr = (unsigned char *)(arp + 1); 3030 arp_ptr += bond->dev->addr_len; 3031 memcpy(&sip, arp_ptr, 4); 3032 arp_ptr += 4 + bond->dev->addr_len; 3033 memcpy(&tip, arp_ptr, 4); 3034 3035 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3036 __func__, slave->dev->name, bond_slave_state(slave), 3037 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3038 &sip, &tip); 3039 3040 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3041 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3042 3043 /* We 'trust' the received ARP enough to validate it if: 3044 * 3045 * (a) the slave receiving the ARP is active (which includes the 3046 * current ARP slave, if any), or 3047 * 3048 * (b) the receiving slave isn't active, but there is a currently 3049 * active slave and it received valid arp reply(s) after it became 3050 * the currently active slave, or 3051 * 3052 * (c) there is an ARP slave that sent an ARP during the prior ARP 3053 * interval, and we receive an ARP reply on any slave. We accept 3054 * these because switch FDB update delays may deliver the ARP 3055 * reply to a slave other than the sender of the ARP request. 3056 * 3057 * Note: for (b), backup slaves are receiving the broadcast ARP 3058 * request, not a reply. This request passes from the sending 3059 * slave through the L2 switch(es) to the receiving slave. Since 3060 * this is checking the request, sip/tip are swapped for 3061 * validation. 3062 * 3063 * This is done to avoid endless looping when we can't reach the 3064 * arp_ip_target and fool ourselves with our own arp requests. 3065 */ 3066 if (bond_is_active_slave(slave)) 3067 bond_validate_arp(bond, slave, sip, tip); 3068 else if (curr_active_slave && 3069 time_after(slave_last_rx(bond, curr_active_slave), 3070 curr_active_slave->last_link_up)) 3071 bond_validate_arp(bond, slave, tip, sip); 3072 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3073 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3074 bond_validate_arp(bond, slave, sip, tip); 3075 3076 out_unlock: 3077 if (arp != (struct arphdr *)skb->data) 3078 kfree(arp); 3079 return RX_HANDLER_ANOTHER; 3080 } 3081 3082 #if IS_ENABLED(CONFIG_IPV6) 3083 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3084 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3085 { 3086 struct net_device *bond_dev = slave->bond->dev; 3087 struct net_device *slave_dev = slave->dev; 3088 struct in6_addr mcaddr; 3089 struct sk_buff *skb; 3090 3091 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3092 daddr, saddr); 3093 3094 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3095 if (!skb) { 3096 net_err_ratelimited("NS packet allocation failed\n"); 3097 return; 3098 } 3099 3100 addrconf_addr_solict_mult(daddr, &mcaddr); 3101 if (bond_handle_vlan(slave, tags, skb)) { 3102 slave_update_last_tx(slave); 3103 ndisc_send_skb(skb, &mcaddr, saddr); 3104 } 3105 } 3106 3107 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3108 { 3109 struct in6_addr *targets = bond->params.ns_targets; 3110 struct bond_vlan_tag *tags; 3111 struct dst_entry *dst; 3112 struct in6_addr saddr; 3113 struct flowi6 fl6; 3114 int i; 3115 3116 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3117 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3118 __func__, &targets[i]); 3119 tags = NULL; 3120 3121 /* Find out through which dev should the packet go */ 3122 memset(&fl6, 0, sizeof(struct flowi6)); 3123 fl6.daddr = targets[i]; 3124 fl6.flowi6_oif = bond->dev->ifindex; 3125 3126 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3127 if (dst->error) { 3128 dst_release(dst); 3129 /* there's no route to target - try to send arp 3130 * probe to generate any traffic (arp_validate=0) 3131 */ 3132 if (bond->params.arp_validate) 3133 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3134 bond->dev->name, 3135 &targets[i]); 3136 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3137 continue; 3138 } 3139 3140 /* bond device itself */ 3141 if (dst->dev == bond->dev) 3142 goto found; 3143 3144 rcu_read_lock(); 3145 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3146 rcu_read_unlock(); 3147 3148 if (!IS_ERR_OR_NULL(tags)) 3149 goto found; 3150 3151 /* Not our device - skip */ 3152 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3153 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3154 3155 dst_release(dst); 3156 continue; 3157 3158 found: 3159 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3160 bond_ns_send(slave, &targets[i], &saddr, tags); 3161 else 3162 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3163 3164 dst_release(dst); 3165 kfree(tags); 3166 } 3167 } 3168 3169 static int bond_confirm_addr6(struct net_device *dev, 3170 struct netdev_nested_priv *priv) 3171 { 3172 struct in6_addr *addr = (struct in6_addr *)priv->data; 3173 3174 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3175 } 3176 3177 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3178 { 3179 struct netdev_nested_priv priv = { 3180 .data = addr, 3181 }; 3182 int ret = false; 3183 3184 if (bond_confirm_addr6(bond->dev, &priv)) 3185 return true; 3186 3187 rcu_read_lock(); 3188 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3189 ret = true; 3190 rcu_read_unlock(); 3191 3192 return ret; 3193 } 3194 3195 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3196 struct in6_addr *saddr, struct in6_addr *daddr) 3197 { 3198 int i; 3199 3200 /* Ignore NAs that: 3201 * 1. Source address is unspecified address. 3202 * 2. Dest address is neither all-nodes multicast address nor 3203 * exist on bond interface. 3204 */ 3205 if (ipv6_addr_any(saddr) || 3206 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3207 !bond_has_this_ip6(bond, daddr))) { 3208 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3209 __func__, saddr, daddr); 3210 return; 3211 } 3212 3213 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3214 if (i == -1) { 3215 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3216 __func__, saddr); 3217 return; 3218 } 3219 slave->last_rx = jiffies; 3220 slave->target_last_arp_rx[i] = jiffies; 3221 } 3222 3223 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3224 struct slave *slave) 3225 { 3226 struct slave *curr_active_slave, *curr_arp_slave; 3227 struct icmp6hdr *hdr = icmp6_hdr(skb); 3228 struct in6_addr *saddr, *daddr; 3229 3230 if (skb->pkt_type == PACKET_OTHERHOST || 3231 skb->pkt_type == PACKET_LOOPBACK || 3232 hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT) 3233 goto out; 3234 3235 saddr = &ipv6_hdr(skb)->saddr; 3236 daddr = &ipv6_hdr(skb)->daddr; 3237 3238 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3239 __func__, slave->dev->name, bond_slave_state(slave), 3240 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3241 saddr, daddr); 3242 3243 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3244 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3245 3246 /* We 'trust' the received ARP enough to validate it if: 3247 * see bond_arp_rcv(). 3248 */ 3249 if (bond_is_active_slave(slave)) 3250 bond_validate_na(bond, slave, saddr, daddr); 3251 else if (curr_active_slave && 3252 time_after(slave_last_rx(bond, curr_active_slave), 3253 curr_active_slave->last_link_up)) 3254 bond_validate_na(bond, slave, saddr, daddr); 3255 else if (curr_arp_slave && 3256 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3257 bond_validate_na(bond, slave, saddr, daddr); 3258 3259 out: 3260 return RX_HANDLER_ANOTHER; 3261 } 3262 #endif 3263 3264 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3265 struct slave *slave) 3266 { 3267 #if IS_ENABLED(CONFIG_IPV6) 3268 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3269 #endif 3270 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3271 3272 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3273 __func__, skb->dev->name); 3274 3275 /* Use arp validate logic for both ARP and NS */ 3276 if (!slave_do_arp_validate(bond, slave)) { 3277 if ((slave_do_arp_validate_only(bond) && is_arp) || 3278 #if IS_ENABLED(CONFIG_IPV6) 3279 (slave_do_arp_validate_only(bond) && is_ipv6) || 3280 #endif 3281 !slave_do_arp_validate_only(bond)) 3282 slave->last_rx = jiffies; 3283 return RX_HANDLER_ANOTHER; 3284 } else if (is_arp) { 3285 return bond_arp_rcv(skb, bond, slave); 3286 #if IS_ENABLED(CONFIG_IPV6) 3287 } else if (is_ipv6) { 3288 return bond_na_rcv(skb, bond, slave); 3289 #endif 3290 } else { 3291 return RX_HANDLER_ANOTHER; 3292 } 3293 } 3294 3295 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3296 { 3297 bond_arp_send_all(bond, slave); 3298 #if IS_ENABLED(CONFIG_IPV6) 3299 bond_ns_send_all(bond, slave); 3300 #endif 3301 } 3302 3303 /* function to verify if we're in the arp_interval timeslice, returns true if 3304 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3305 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3306 */ 3307 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3308 int mod) 3309 { 3310 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3311 3312 return time_in_range(jiffies, 3313 last_act - delta_in_ticks, 3314 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3315 } 3316 3317 /* This function is called regularly to monitor each slave's link 3318 * ensuring that traffic is being sent and received when arp monitoring 3319 * is used in load-balancing mode. if the adapter has been dormant, then an 3320 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3321 * arp monitoring in active backup mode. 3322 */ 3323 static void bond_loadbalance_arp_mon(struct bonding *bond) 3324 { 3325 struct slave *slave, *oldcurrent; 3326 struct list_head *iter; 3327 int do_failover = 0, slave_state_changed = 0; 3328 3329 if (!bond_has_slaves(bond)) 3330 goto re_arm; 3331 3332 rcu_read_lock(); 3333 3334 oldcurrent = rcu_dereference(bond->curr_active_slave); 3335 /* see if any of the previous devices are up now (i.e. they have 3336 * xmt and rcv traffic). the curr_active_slave does not come into 3337 * the picture unless it is null. also, slave->last_link_up is not 3338 * needed here because we send an arp on each slave and give a slave 3339 * as long as it needs to get the tx/rx within the delta. 3340 * TODO: what about up/down delay in arp mode? it wasn't here before 3341 * so it can wait 3342 */ 3343 bond_for_each_slave_rcu(bond, slave, iter) { 3344 unsigned long last_tx = slave_last_tx(slave); 3345 3346 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3347 3348 if (slave->link != BOND_LINK_UP) { 3349 if (bond_time_in_interval(bond, last_tx, 1) && 3350 bond_time_in_interval(bond, slave->last_rx, 1)) { 3351 3352 bond_propose_link_state(slave, BOND_LINK_UP); 3353 slave_state_changed = 1; 3354 3355 /* primary_slave has no meaning in round-robin 3356 * mode. the window of a slave being up and 3357 * curr_active_slave being null after enslaving 3358 * is closed. 3359 */ 3360 if (!oldcurrent) { 3361 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3362 do_failover = 1; 3363 } else { 3364 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3365 } 3366 } 3367 } else { 3368 /* slave->link == BOND_LINK_UP */ 3369 3370 /* not all switches will respond to an arp request 3371 * when the source ip is 0, so don't take the link down 3372 * if we don't know our ip yet 3373 */ 3374 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3375 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3376 3377 bond_propose_link_state(slave, BOND_LINK_DOWN); 3378 slave_state_changed = 1; 3379 3380 if (slave->link_failure_count < UINT_MAX) 3381 slave->link_failure_count++; 3382 3383 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3384 3385 if (slave == oldcurrent) 3386 do_failover = 1; 3387 } 3388 } 3389 3390 /* note: if switch is in round-robin mode, all links 3391 * must tx arp to ensure all links rx an arp - otherwise 3392 * links may oscillate or not come up at all; if switch is 3393 * in something like xor mode, there is nothing we can 3394 * do - all replies will be rx'ed on same link causing slaves 3395 * to be unstable during low/no traffic periods 3396 */ 3397 if (bond_slave_is_up(slave)) 3398 bond_send_validate(bond, slave); 3399 } 3400 3401 rcu_read_unlock(); 3402 3403 if (do_failover || slave_state_changed) { 3404 if (!rtnl_trylock()) 3405 goto re_arm; 3406 3407 bond_for_each_slave(bond, slave, iter) { 3408 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3409 slave->link = slave->link_new_state; 3410 } 3411 3412 if (slave_state_changed) { 3413 bond_slave_state_change(bond); 3414 if (BOND_MODE(bond) == BOND_MODE_XOR) 3415 bond_update_slave_arr(bond, NULL); 3416 } 3417 if (do_failover) { 3418 block_netpoll_tx(); 3419 bond_select_active_slave(bond); 3420 unblock_netpoll_tx(); 3421 } 3422 rtnl_unlock(); 3423 } 3424 3425 re_arm: 3426 if (bond->params.arp_interval) 3427 queue_delayed_work(bond->wq, &bond->arp_work, 3428 msecs_to_jiffies(bond->params.arp_interval)); 3429 } 3430 3431 /* Called to inspect slaves for active-backup mode ARP monitor link state 3432 * changes. Sets proposed link state in slaves to specify what action 3433 * should take place for the slave. Returns 0 if no changes are found, >0 3434 * if changes to link states must be committed. 3435 * 3436 * Called with rcu_read_lock held. 3437 */ 3438 static int bond_ab_arp_inspect(struct bonding *bond) 3439 { 3440 unsigned long last_tx, last_rx; 3441 struct list_head *iter; 3442 struct slave *slave; 3443 int commit = 0; 3444 3445 bond_for_each_slave_rcu(bond, slave, iter) { 3446 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3447 last_rx = slave_last_rx(bond, slave); 3448 3449 if (slave->link != BOND_LINK_UP) { 3450 if (bond_time_in_interval(bond, last_rx, 1)) { 3451 bond_propose_link_state(slave, BOND_LINK_UP); 3452 commit++; 3453 } else if (slave->link == BOND_LINK_BACK) { 3454 bond_propose_link_state(slave, BOND_LINK_FAIL); 3455 commit++; 3456 } 3457 continue; 3458 } 3459 3460 /* Give slaves 2*delta after being enslaved or made 3461 * active. This avoids bouncing, as the last receive 3462 * times need a full ARP monitor cycle to be updated. 3463 */ 3464 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3465 continue; 3466 3467 /* Backup slave is down if: 3468 * - No current_arp_slave AND 3469 * - more than (missed_max+1)*delta since last receive AND 3470 * - the bond has an IP address 3471 * 3472 * Note: a non-null current_arp_slave indicates 3473 * the curr_active_slave went down and we are 3474 * searching for a new one; under this condition 3475 * we only take the curr_active_slave down - this 3476 * gives each slave a chance to tx/rx traffic 3477 * before being taken out 3478 */ 3479 if (!bond_is_active_slave(slave) && 3480 !rcu_access_pointer(bond->current_arp_slave) && 3481 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3482 bond_propose_link_state(slave, BOND_LINK_DOWN); 3483 commit++; 3484 } 3485 3486 /* Active slave is down if: 3487 * - more than missed_max*delta since transmitting OR 3488 * - (more than missed_max*delta since receive AND 3489 * the bond has an IP address) 3490 */ 3491 last_tx = slave_last_tx(slave); 3492 if (bond_is_active_slave(slave) && 3493 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3494 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3495 bond_propose_link_state(slave, BOND_LINK_DOWN); 3496 commit++; 3497 } 3498 } 3499 3500 return commit; 3501 } 3502 3503 /* Called to commit link state changes noted by inspection step of 3504 * active-backup mode ARP monitor. 3505 * 3506 * Called with RTNL hold. 3507 */ 3508 static void bond_ab_arp_commit(struct bonding *bond) 3509 { 3510 struct list_head *iter; 3511 unsigned long last_tx; 3512 struct slave *slave; 3513 3514 bond_for_each_slave(bond, slave, iter) { 3515 switch (slave->link_new_state) { 3516 case BOND_LINK_NOCHANGE: 3517 continue; 3518 3519 case BOND_LINK_UP: 3520 last_tx = slave_last_tx(slave); 3521 if (rtnl_dereference(bond->curr_active_slave) != slave || 3522 (!rtnl_dereference(bond->curr_active_slave) && 3523 bond_time_in_interval(bond, last_tx, 1))) { 3524 struct slave *current_arp_slave; 3525 3526 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3527 bond_set_slave_link_state(slave, BOND_LINK_UP, 3528 BOND_SLAVE_NOTIFY_NOW); 3529 if (current_arp_slave) { 3530 bond_set_slave_inactive_flags( 3531 current_arp_slave, 3532 BOND_SLAVE_NOTIFY_NOW); 3533 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3534 } 3535 3536 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3537 3538 if (!rtnl_dereference(bond->curr_active_slave) || 3539 slave == rtnl_dereference(bond->primary_slave)) 3540 goto do_failover; 3541 3542 } 3543 3544 continue; 3545 3546 case BOND_LINK_DOWN: 3547 if (slave->link_failure_count < UINT_MAX) 3548 slave->link_failure_count++; 3549 3550 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3551 BOND_SLAVE_NOTIFY_NOW); 3552 bond_set_slave_inactive_flags(slave, 3553 BOND_SLAVE_NOTIFY_NOW); 3554 3555 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3556 3557 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3558 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3559 goto do_failover; 3560 } 3561 3562 continue; 3563 3564 case BOND_LINK_FAIL: 3565 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3566 BOND_SLAVE_NOTIFY_NOW); 3567 bond_set_slave_inactive_flags(slave, 3568 BOND_SLAVE_NOTIFY_NOW); 3569 3570 /* A slave has just been enslaved and has become 3571 * the current active slave. 3572 */ 3573 if (rtnl_dereference(bond->curr_active_slave)) 3574 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3575 continue; 3576 3577 default: 3578 slave_err(bond->dev, slave->dev, 3579 "impossible: link_new_state %d on slave\n", 3580 slave->link_new_state); 3581 continue; 3582 } 3583 3584 do_failover: 3585 block_netpoll_tx(); 3586 bond_select_active_slave(bond); 3587 unblock_netpoll_tx(); 3588 } 3589 3590 bond_set_carrier(bond); 3591 } 3592 3593 /* Send ARP probes for active-backup mode ARP monitor. 3594 * 3595 * Called with rcu_read_lock held. 3596 */ 3597 static bool bond_ab_arp_probe(struct bonding *bond) 3598 { 3599 struct slave *slave, *before = NULL, *new_slave = NULL, 3600 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3601 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3602 struct list_head *iter; 3603 bool found = false; 3604 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3605 3606 if (curr_arp_slave && curr_active_slave) 3607 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3608 curr_arp_slave->dev->name, 3609 curr_active_slave->dev->name); 3610 3611 if (curr_active_slave) { 3612 bond_send_validate(bond, curr_active_slave); 3613 return should_notify_rtnl; 3614 } 3615 3616 /* if we don't have a curr_active_slave, search for the next available 3617 * backup slave from the current_arp_slave and make it the candidate 3618 * for becoming the curr_active_slave 3619 */ 3620 3621 if (!curr_arp_slave) { 3622 curr_arp_slave = bond_first_slave_rcu(bond); 3623 if (!curr_arp_slave) 3624 return should_notify_rtnl; 3625 } 3626 3627 bond_for_each_slave_rcu(bond, slave, iter) { 3628 if (!found && !before && bond_slave_is_up(slave)) 3629 before = slave; 3630 3631 if (found && !new_slave && bond_slave_is_up(slave)) 3632 new_slave = slave; 3633 /* if the link state is up at this point, we 3634 * mark it down - this can happen if we have 3635 * simultaneous link failures and 3636 * reselect_active_interface doesn't make this 3637 * one the current slave so it is still marked 3638 * up when it is actually down 3639 */ 3640 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3641 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3642 BOND_SLAVE_NOTIFY_LATER); 3643 if (slave->link_failure_count < UINT_MAX) 3644 slave->link_failure_count++; 3645 3646 bond_set_slave_inactive_flags(slave, 3647 BOND_SLAVE_NOTIFY_LATER); 3648 3649 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3650 } 3651 if (slave == curr_arp_slave) 3652 found = true; 3653 } 3654 3655 if (!new_slave && before) 3656 new_slave = before; 3657 3658 if (!new_slave) 3659 goto check_state; 3660 3661 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3662 BOND_SLAVE_NOTIFY_LATER); 3663 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3664 bond_send_validate(bond, new_slave); 3665 new_slave->last_link_up = jiffies; 3666 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3667 3668 check_state: 3669 bond_for_each_slave_rcu(bond, slave, iter) { 3670 if (slave->should_notify || slave->should_notify_link) { 3671 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3672 break; 3673 } 3674 } 3675 return should_notify_rtnl; 3676 } 3677 3678 static void bond_activebackup_arp_mon(struct bonding *bond) 3679 { 3680 bool should_notify_peers = false; 3681 bool should_notify_rtnl = false; 3682 int delta_in_ticks; 3683 3684 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3685 3686 if (!bond_has_slaves(bond)) 3687 goto re_arm; 3688 3689 rcu_read_lock(); 3690 3691 should_notify_peers = bond_should_notify_peers(bond); 3692 3693 if (bond_ab_arp_inspect(bond)) { 3694 rcu_read_unlock(); 3695 3696 /* Race avoidance with bond_close flush of workqueue */ 3697 if (!rtnl_trylock()) { 3698 delta_in_ticks = 1; 3699 should_notify_peers = false; 3700 goto re_arm; 3701 } 3702 3703 bond_ab_arp_commit(bond); 3704 3705 rtnl_unlock(); 3706 rcu_read_lock(); 3707 } 3708 3709 should_notify_rtnl = bond_ab_arp_probe(bond); 3710 rcu_read_unlock(); 3711 3712 re_arm: 3713 if (bond->params.arp_interval) 3714 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3715 3716 if (should_notify_peers || should_notify_rtnl) { 3717 if (!rtnl_trylock()) 3718 return; 3719 3720 if (should_notify_peers) { 3721 bond->send_peer_notif--; 3722 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3723 bond->dev); 3724 } 3725 if (should_notify_rtnl) { 3726 bond_slave_state_notify(bond); 3727 bond_slave_link_notify(bond); 3728 } 3729 3730 rtnl_unlock(); 3731 } 3732 } 3733 3734 static void bond_arp_monitor(struct work_struct *work) 3735 { 3736 struct bonding *bond = container_of(work, struct bonding, 3737 arp_work.work); 3738 3739 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3740 bond_activebackup_arp_mon(bond); 3741 else 3742 bond_loadbalance_arp_mon(bond); 3743 } 3744 3745 /*-------------------------- netdev event handling --------------------------*/ 3746 3747 /* Change device name */ 3748 static int bond_event_changename(struct bonding *bond) 3749 { 3750 bond_remove_proc_entry(bond); 3751 bond_create_proc_entry(bond); 3752 3753 bond_debug_reregister(bond); 3754 3755 return NOTIFY_DONE; 3756 } 3757 3758 static int bond_master_netdev_event(unsigned long event, 3759 struct net_device *bond_dev) 3760 { 3761 struct bonding *event_bond = netdev_priv(bond_dev); 3762 3763 netdev_dbg(bond_dev, "%s called\n", __func__); 3764 3765 switch (event) { 3766 case NETDEV_CHANGENAME: 3767 return bond_event_changename(event_bond); 3768 case NETDEV_UNREGISTER: 3769 bond_remove_proc_entry(event_bond); 3770 #ifdef CONFIG_XFRM_OFFLOAD 3771 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3772 #endif /* CONFIG_XFRM_OFFLOAD */ 3773 break; 3774 case NETDEV_REGISTER: 3775 bond_create_proc_entry(event_bond); 3776 break; 3777 default: 3778 break; 3779 } 3780 3781 return NOTIFY_DONE; 3782 } 3783 3784 static int bond_slave_netdev_event(unsigned long event, 3785 struct net_device *slave_dev) 3786 { 3787 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3788 struct bonding *bond; 3789 struct net_device *bond_dev; 3790 3791 /* A netdev event can be generated while enslaving a device 3792 * before netdev_rx_handler_register is called in which case 3793 * slave will be NULL 3794 */ 3795 if (!slave) { 3796 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3797 return NOTIFY_DONE; 3798 } 3799 3800 bond_dev = slave->bond->dev; 3801 bond = slave->bond; 3802 primary = rtnl_dereference(bond->primary_slave); 3803 3804 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3805 3806 switch (event) { 3807 case NETDEV_UNREGISTER: 3808 if (bond_dev->type != ARPHRD_ETHER) 3809 bond_release_and_destroy(bond_dev, slave_dev); 3810 else 3811 __bond_release_one(bond_dev, slave_dev, false, true); 3812 break; 3813 case NETDEV_UP: 3814 case NETDEV_CHANGE: 3815 /* For 802.3ad mode only: 3816 * Getting invalid Speed/Duplex values here will put slave 3817 * in weird state. Mark it as link-fail if the link was 3818 * previously up or link-down if it hasn't yet come up, and 3819 * let link-monitoring (miimon) set it right when correct 3820 * speeds/duplex are available. 3821 */ 3822 if (bond_update_speed_duplex(slave) && 3823 BOND_MODE(bond) == BOND_MODE_8023AD) { 3824 if (slave->last_link_up) 3825 slave->link = BOND_LINK_FAIL; 3826 else 3827 slave->link = BOND_LINK_DOWN; 3828 } 3829 3830 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3831 bond_3ad_adapter_speed_duplex_changed(slave); 3832 fallthrough; 3833 case NETDEV_DOWN: 3834 /* Refresh slave-array if applicable! 3835 * If the setup does not use miimon or arpmon (mode-specific!), 3836 * then these events will not cause the slave-array to be 3837 * refreshed. This will cause xmit to use a slave that is not 3838 * usable. Avoid such situation by refeshing the array at these 3839 * events. If these (miimon/arpmon) parameters are configured 3840 * then array gets refreshed twice and that should be fine! 3841 */ 3842 if (bond_mode_can_use_xmit_hash(bond)) 3843 bond_update_slave_arr(bond, NULL); 3844 break; 3845 case NETDEV_CHANGEMTU: 3846 /* TODO: Should slaves be allowed to 3847 * independently alter their MTU? For 3848 * an active-backup bond, slaves need 3849 * not be the same type of device, so 3850 * MTUs may vary. For other modes, 3851 * slaves arguably should have the 3852 * same MTUs. To do this, we'd need to 3853 * take over the slave's change_mtu 3854 * function for the duration of their 3855 * servitude. 3856 */ 3857 break; 3858 case NETDEV_CHANGENAME: 3859 /* we don't care if we don't have primary set */ 3860 if (!bond_uses_primary(bond) || 3861 !bond->params.primary[0]) 3862 break; 3863 3864 if (slave == primary) { 3865 /* slave's name changed - he's no longer primary */ 3866 RCU_INIT_POINTER(bond->primary_slave, NULL); 3867 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3868 /* we have a new primary slave */ 3869 rcu_assign_pointer(bond->primary_slave, slave); 3870 } else { /* we didn't change primary - exit */ 3871 break; 3872 } 3873 3874 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3875 primary ? slave_dev->name : "none"); 3876 3877 block_netpoll_tx(); 3878 bond_select_active_slave(bond); 3879 unblock_netpoll_tx(); 3880 break; 3881 case NETDEV_FEAT_CHANGE: 3882 bond_compute_features(bond); 3883 break; 3884 case NETDEV_RESEND_IGMP: 3885 /* Propagate to master device */ 3886 call_netdevice_notifiers(event, slave->bond->dev); 3887 break; 3888 default: 3889 break; 3890 } 3891 3892 return NOTIFY_DONE; 3893 } 3894 3895 /* bond_netdev_event: handle netdev notifier chain events. 3896 * 3897 * This function receives events for the netdev chain. The caller (an 3898 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3899 * locks for us to safely manipulate the slave devices (RTNL lock, 3900 * dev_probe_lock). 3901 */ 3902 static int bond_netdev_event(struct notifier_block *this, 3903 unsigned long event, void *ptr) 3904 { 3905 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3906 3907 netdev_dbg(event_dev, "%s received %s\n", 3908 __func__, netdev_cmd_to_name(event)); 3909 3910 if (!(event_dev->priv_flags & IFF_BONDING)) 3911 return NOTIFY_DONE; 3912 3913 if (event_dev->flags & IFF_MASTER) { 3914 int ret; 3915 3916 ret = bond_master_netdev_event(event, event_dev); 3917 if (ret != NOTIFY_DONE) 3918 return ret; 3919 } 3920 3921 if (event_dev->flags & IFF_SLAVE) 3922 return bond_slave_netdev_event(event, event_dev); 3923 3924 return NOTIFY_DONE; 3925 } 3926 3927 static struct notifier_block bond_netdev_notifier = { 3928 .notifier_call = bond_netdev_event, 3929 }; 3930 3931 /*---------------------------- Hashing Policies -----------------------------*/ 3932 3933 /* Helper to access data in a packet, with or without a backing skb. 3934 * If skb is given the data is linearized if necessary via pskb_may_pull. 3935 */ 3936 static inline const void *bond_pull_data(struct sk_buff *skb, 3937 const void *data, int hlen, int n) 3938 { 3939 if (likely(n <= hlen)) 3940 return data; 3941 else if (skb && likely(pskb_may_pull(skb, n))) 3942 return skb->head; 3943 3944 return NULL; 3945 } 3946 3947 /* L2 hash helper */ 3948 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3949 { 3950 struct ethhdr *ep; 3951 3952 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 3953 if (!data) 3954 return 0; 3955 3956 ep = (struct ethhdr *)(data + mhoff); 3957 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 3958 } 3959 3960 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 3961 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 3962 { 3963 const struct ipv6hdr *iph6; 3964 const struct iphdr *iph; 3965 3966 if (l2_proto == htons(ETH_P_IP)) { 3967 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 3968 if (!data) 3969 return false; 3970 3971 iph = (const struct iphdr *)(data + *nhoff); 3972 iph_to_flow_copy_v4addrs(fk, iph); 3973 *nhoff += iph->ihl << 2; 3974 if (!ip_is_fragment(iph)) 3975 *ip_proto = iph->protocol; 3976 } else if (l2_proto == htons(ETH_P_IPV6)) { 3977 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 3978 if (!data) 3979 return false; 3980 3981 iph6 = (const struct ipv6hdr *)(data + *nhoff); 3982 iph_to_flow_copy_v6addrs(fk, iph6); 3983 *nhoff += sizeof(*iph6); 3984 *ip_proto = iph6->nexthdr; 3985 } else { 3986 return false; 3987 } 3988 3989 if (l34 && *ip_proto >= 0) 3990 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 3991 3992 return true; 3993 } 3994 3995 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3996 { 3997 u32 srcmac_vendor = 0, srcmac_dev = 0; 3998 struct ethhdr *mac_hdr; 3999 u16 vlan = 0; 4000 int i; 4001 4002 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4003 if (!data) 4004 return 0; 4005 mac_hdr = (struct ethhdr *)(data + mhoff); 4006 4007 for (i = 0; i < 3; i++) 4008 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4009 4010 for (i = 3; i < ETH_ALEN; i++) 4011 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4012 4013 if (skb && skb_vlan_tag_present(skb)) 4014 vlan = skb_vlan_tag_get(skb); 4015 4016 return vlan ^ srcmac_vendor ^ srcmac_dev; 4017 } 4018 4019 /* Extract the appropriate headers based on bond's xmit policy */ 4020 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4021 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4022 { 4023 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4024 int ip_proto = -1; 4025 4026 switch (bond->params.xmit_policy) { 4027 case BOND_XMIT_POLICY_ENCAP23: 4028 case BOND_XMIT_POLICY_ENCAP34: 4029 memset(fk, 0, sizeof(*fk)); 4030 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4031 fk, data, l2_proto, nhoff, hlen, 0); 4032 default: 4033 break; 4034 } 4035 4036 fk->ports.ports = 0; 4037 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4038 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4039 return false; 4040 4041 /* ICMP error packets contains at least 8 bytes of the header 4042 * of the packet which generated the error. Use this information 4043 * to correlate ICMP error packets within the same flow which 4044 * generated the error. 4045 */ 4046 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4047 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4048 if (ip_proto == IPPROTO_ICMP) { 4049 if (!icmp_is_err(fk->icmp.type)) 4050 return true; 4051 4052 nhoff += sizeof(struct icmphdr); 4053 } else if (ip_proto == IPPROTO_ICMPV6) { 4054 if (!icmpv6_is_err(fk->icmp.type)) 4055 return true; 4056 4057 nhoff += sizeof(struct icmp6hdr); 4058 } 4059 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4060 } 4061 4062 return true; 4063 } 4064 4065 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4066 { 4067 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4068 (__force u32)flow_get_u32_src(flow); 4069 hash ^= (hash >> 16); 4070 hash ^= (hash >> 8); 4071 4072 /* discard lowest hash bit to deal with the common even ports pattern */ 4073 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4074 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4075 return hash >> 1; 4076 4077 return hash; 4078 } 4079 4080 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4081 * the data as required, but this function can be used without it if the data is 4082 * known to be linear (e.g. with xdp_buff). 4083 */ 4084 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4085 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4086 { 4087 struct flow_keys flow; 4088 u32 hash; 4089 4090 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4091 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4092 4093 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4094 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4095 return bond_eth_hash(skb, data, mhoff, hlen); 4096 4097 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4098 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4099 hash = bond_eth_hash(skb, data, mhoff, hlen); 4100 } else { 4101 if (flow.icmp.id) 4102 memcpy(&hash, &flow.icmp, sizeof(hash)); 4103 else 4104 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4105 } 4106 4107 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4108 } 4109 4110 /** 4111 * bond_xmit_hash - generate a hash value based on the xmit policy 4112 * @bond: bonding device 4113 * @skb: buffer to use for headers 4114 * 4115 * This function will extract the necessary headers from the skb buffer and use 4116 * them to generate a hash based on the xmit_policy set in the bonding device 4117 */ 4118 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4119 { 4120 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4121 skb->l4_hash) 4122 return skb->hash; 4123 4124 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4125 skb_mac_offset(skb), skb_network_offset(skb), 4126 skb_headlen(skb)); 4127 } 4128 4129 /** 4130 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4131 * @bond: bonding device 4132 * @xdp: buffer to use for headers 4133 * 4134 * The XDP variant of bond_xmit_hash. 4135 */ 4136 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4137 { 4138 struct ethhdr *eth; 4139 4140 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4141 return 0; 4142 4143 eth = (struct ethhdr *)xdp->data; 4144 4145 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4146 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4147 } 4148 4149 /*-------------------------- Device entry points ----------------------------*/ 4150 4151 void bond_work_init_all(struct bonding *bond) 4152 { 4153 INIT_DELAYED_WORK(&bond->mcast_work, 4154 bond_resend_igmp_join_requests_delayed); 4155 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4156 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4157 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4158 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4159 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4160 } 4161 4162 static void bond_work_cancel_all(struct bonding *bond) 4163 { 4164 cancel_delayed_work_sync(&bond->mii_work); 4165 cancel_delayed_work_sync(&bond->arp_work); 4166 cancel_delayed_work_sync(&bond->alb_work); 4167 cancel_delayed_work_sync(&bond->ad_work); 4168 cancel_delayed_work_sync(&bond->mcast_work); 4169 cancel_delayed_work_sync(&bond->slave_arr_work); 4170 } 4171 4172 static int bond_open(struct net_device *bond_dev) 4173 { 4174 struct bonding *bond = netdev_priv(bond_dev); 4175 struct list_head *iter; 4176 struct slave *slave; 4177 4178 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4179 bond->rr_tx_counter = alloc_percpu(u32); 4180 if (!bond->rr_tx_counter) 4181 return -ENOMEM; 4182 } 4183 4184 /* reset slave->backup and slave->inactive */ 4185 if (bond_has_slaves(bond)) { 4186 bond_for_each_slave(bond, slave, iter) { 4187 if (bond_uses_primary(bond) && 4188 slave != rcu_access_pointer(bond->curr_active_slave)) { 4189 bond_set_slave_inactive_flags(slave, 4190 BOND_SLAVE_NOTIFY_NOW); 4191 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4192 bond_set_slave_active_flags(slave, 4193 BOND_SLAVE_NOTIFY_NOW); 4194 } 4195 } 4196 } 4197 4198 if (bond_is_lb(bond)) { 4199 /* bond_alb_initialize must be called before the timer 4200 * is started. 4201 */ 4202 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4203 return -ENOMEM; 4204 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4205 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4206 } 4207 4208 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4209 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4210 4211 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4212 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4213 bond->recv_probe = bond_rcv_validate; 4214 } 4215 4216 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4217 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4218 /* register to receive LACPDUs */ 4219 bond->recv_probe = bond_3ad_lacpdu_recv; 4220 bond_3ad_initiate_agg_selection(bond, 1); 4221 4222 bond_for_each_slave(bond, slave, iter) 4223 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4224 } 4225 4226 if (bond_mode_can_use_xmit_hash(bond)) 4227 bond_update_slave_arr(bond, NULL); 4228 4229 return 0; 4230 } 4231 4232 static int bond_close(struct net_device *bond_dev) 4233 { 4234 struct bonding *bond = netdev_priv(bond_dev); 4235 struct slave *slave; 4236 4237 bond_work_cancel_all(bond); 4238 bond->send_peer_notif = 0; 4239 if (bond_is_lb(bond)) 4240 bond_alb_deinitialize(bond); 4241 bond->recv_probe = NULL; 4242 4243 if (bond_uses_primary(bond)) { 4244 rcu_read_lock(); 4245 slave = rcu_dereference(bond->curr_active_slave); 4246 if (slave) 4247 bond_hw_addr_flush(bond_dev, slave->dev); 4248 rcu_read_unlock(); 4249 } else { 4250 struct list_head *iter; 4251 4252 bond_for_each_slave(bond, slave, iter) 4253 bond_hw_addr_flush(bond_dev, slave->dev); 4254 } 4255 4256 return 0; 4257 } 4258 4259 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4260 * that some drivers can provide 32bit values only. 4261 */ 4262 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4263 const struct rtnl_link_stats64 *_new, 4264 const struct rtnl_link_stats64 *_old) 4265 { 4266 const u64 *new = (const u64 *)_new; 4267 const u64 *old = (const u64 *)_old; 4268 u64 *res = (u64 *)_res; 4269 int i; 4270 4271 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4272 u64 nv = new[i]; 4273 u64 ov = old[i]; 4274 s64 delta = nv - ov; 4275 4276 /* detects if this particular field is 32bit only */ 4277 if (((nv | ov) >> 32) == 0) 4278 delta = (s64)(s32)((u32)nv - (u32)ov); 4279 4280 /* filter anomalies, some drivers reset their stats 4281 * at down/up events. 4282 */ 4283 if (delta > 0) 4284 res[i] += delta; 4285 } 4286 } 4287 4288 #ifdef CONFIG_LOCKDEP 4289 static int bond_get_lowest_level_rcu(struct net_device *dev) 4290 { 4291 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4292 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4293 int cur = 0, max = 0; 4294 4295 now = dev; 4296 iter = &dev->adj_list.lower; 4297 4298 while (1) { 4299 next = NULL; 4300 while (1) { 4301 ldev = netdev_next_lower_dev_rcu(now, &iter); 4302 if (!ldev) 4303 break; 4304 4305 next = ldev; 4306 niter = &ldev->adj_list.lower; 4307 dev_stack[cur] = now; 4308 iter_stack[cur++] = iter; 4309 if (max <= cur) 4310 max = cur; 4311 break; 4312 } 4313 4314 if (!next) { 4315 if (!cur) 4316 return max; 4317 next = dev_stack[--cur]; 4318 niter = iter_stack[cur]; 4319 } 4320 4321 now = next; 4322 iter = niter; 4323 } 4324 4325 return max; 4326 } 4327 #endif 4328 4329 static void bond_get_stats(struct net_device *bond_dev, 4330 struct rtnl_link_stats64 *stats) 4331 { 4332 struct bonding *bond = netdev_priv(bond_dev); 4333 struct rtnl_link_stats64 temp; 4334 struct list_head *iter; 4335 struct slave *slave; 4336 int nest_level = 0; 4337 4338 4339 rcu_read_lock(); 4340 #ifdef CONFIG_LOCKDEP 4341 nest_level = bond_get_lowest_level_rcu(bond_dev); 4342 #endif 4343 4344 spin_lock_nested(&bond->stats_lock, nest_level); 4345 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4346 4347 bond_for_each_slave_rcu(bond, slave, iter) { 4348 const struct rtnl_link_stats64 *new = 4349 dev_get_stats(slave->dev, &temp); 4350 4351 bond_fold_stats(stats, new, &slave->slave_stats); 4352 4353 /* save off the slave stats for the next run */ 4354 memcpy(&slave->slave_stats, new, sizeof(*new)); 4355 } 4356 4357 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4358 spin_unlock(&bond->stats_lock); 4359 rcu_read_unlock(); 4360 } 4361 4362 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4363 { 4364 struct bonding *bond = netdev_priv(bond_dev); 4365 struct mii_ioctl_data *mii = NULL; 4366 const struct net_device_ops *ops; 4367 struct net_device *real_dev; 4368 struct hwtstamp_config cfg; 4369 struct ifreq ifrr; 4370 int res = 0; 4371 4372 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4373 4374 switch (cmd) { 4375 case SIOCGMIIPHY: 4376 mii = if_mii(ifr); 4377 if (!mii) 4378 return -EINVAL; 4379 4380 mii->phy_id = 0; 4381 fallthrough; 4382 case SIOCGMIIREG: 4383 /* We do this again just in case we were called by SIOCGMIIREG 4384 * instead of SIOCGMIIPHY. 4385 */ 4386 mii = if_mii(ifr); 4387 if (!mii) 4388 return -EINVAL; 4389 4390 if (mii->reg_num == 1) { 4391 mii->val_out = 0; 4392 if (netif_carrier_ok(bond->dev)) 4393 mii->val_out = BMSR_LSTATUS; 4394 } 4395 4396 break; 4397 case SIOCSHWTSTAMP: 4398 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4399 return -EFAULT; 4400 4401 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 4402 return -EOPNOTSUPP; 4403 4404 fallthrough; 4405 case SIOCGHWTSTAMP: 4406 real_dev = bond_option_active_slave_get_rcu(bond); 4407 if (!real_dev) 4408 return -EOPNOTSUPP; 4409 4410 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ); 4411 ifrr.ifr_ifru = ifr->ifr_ifru; 4412 4413 ops = real_dev->netdev_ops; 4414 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) { 4415 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd); 4416 if (res) 4417 return res; 4418 4419 ifr->ifr_ifru = ifrr.ifr_ifru; 4420 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4421 return -EFAULT; 4422 4423 /* Set the BOND_PHC_INDEX flag to notify user space */ 4424 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 4425 4426 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ? 4427 -EFAULT : 0; 4428 } 4429 fallthrough; 4430 default: 4431 res = -EOPNOTSUPP; 4432 } 4433 4434 return res; 4435 } 4436 4437 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4438 { 4439 struct bonding *bond = netdev_priv(bond_dev); 4440 struct net_device *slave_dev = NULL; 4441 struct ifbond k_binfo; 4442 struct ifbond __user *u_binfo = NULL; 4443 struct ifslave k_sinfo; 4444 struct ifslave __user *u_sinfo = NULL; 4445 struct bond_opt_value newval; 4446 struct net *net; 4447 int res = 0; 4448 4449 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4450 4451 switch (cmd) { 4452 case SIOCBONDINFOQUERY: 4453 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4454 4455 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4456 return -EFAULT; 4457 4458 bond_info_query(bond_dev, &k_binfo); 4459 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4460 return -EFAULT; 4461 4462 return 0; 4463 case SIOCBONDSLAVEINFOQUERY: 4464 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4465 4466 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4467 return -EFAULT; 4468 4469 res = bond_slave_info_query(bond_dev, &k_sinfo); 4470 if (res == 0 && 4471 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4472 return -EFAULT; 4473 4474 return res; 4475 default: 4476 break; 4477 } 4478 4479 net = dev_net(bond_dev); 4480 4481 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4482 return -EPERM; 4483 4484 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4485 4486 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4487 4488 if (!slave_dev) 4489 return -ENODEV; 4490 4491 switch (cmd) { 4492 case SIOCBONDENSLAVE: 4493 res = bond_enslave(bond_dev, slave_dev, NULL); 4494 break; 4495 case SIOCBONDRELEASE: 4496 res = bond_release(bond_dev, slave_dev); 4497 break; 4498 case SIOCBONDSETHWADDR: 4499 res = bond_set_dev_addr(bond_dev, slave_dev); 4500 break; 4501 case SIOCBONDCHANGEACTIVE: 4502 bond_opt_initstr(&newval, slave_dev->name); 4503 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4504 &newval); 4505 break; 4506 default: 4507 res = -EOPNOTSUPP; 4508 } 4509 4510 return res; 4511 } 4512 4513 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4514 void __user *data, int cmd) 4515 { 4516 struct ifreq ifrdata = { .ifr_data = data }; 4517 4518 switch (cmd) { 4519 case BOND_INFO_QUERY_OLD: 4520 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4521 case BOND_SLAVE_INFO_QUERY_OLD: 4522 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4523 case BOND_ENSLAVE_OLD: 4524 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4525 case BOND_RELEASE_OLD: 4526 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4527 case BOND_SETHWADDR_OLD: 4528 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4529 case BOND_CHANGE_ACTIVE_OLD: 4530 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4531 } 4532 4533 return -EOPNOTSUPP; 4534 } 4535 4536 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4537 { 4538 struct bonding *bond = netdev_priv(bond_dev); 4539 4540 if (change & IFF_PROMISC) 4541 bond_set_promiscuity(bond, 4542 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4543 4544 if (change & IFF_ALLMULTI) 4545 bond_set_allmulti(bond, 4546 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4547 } 4548 4549 static void bond_set_rx_mode(struct net_device *bond_dev) 4550 { 4551 struct bonding *bond = netdev_priv(bond_dev); 4552 struct list_head *iter; 4553 struct slave *slave; 4554 4555 rcu_read_lock(); 4556 if (bond_uses_primary(bond)) { 4557 slave = rcu_dereference(bond->curr_active_slave); 4558 if (slave) { 4559 dev_uc_sync(slave->dev, bond_dev); 4560 dev_mc_sync(slave->dev, bond_dev); 4561 } 4562 } else { 4563 bond_for_each_slave_rcu(bond, slave, iter) { 4564 dev_uc_sync_multiple(slave->dev, bond_dev); 4565 dev_mc_sync_multiple(slave->dev, bond_dev); 4566 } 4567 } 4568 rcu_read_unlock(); 4569 } 4570 4571 static int bond_neigh_init(struct neighbour *n) 4572 { 4573 struct bonding *bond = netdev_priv(n->dev); 4574 const struct net_device_ops *slave_ops; 4575 struct neigh_parms parms; 4576 struct slave *slave; 4577 int ret = 0; 4578 4579 rcu_read_lock(); 4580 slave = bond_first_slave_rcu(bond); 4581 if (!slave) 4582 goto out; 4583 slave_ops = slave->dev->netdev_ops; 4584 if (!slave_ops->ndo_neigh_setup) 4585 goto out; 4586 4587 /* TODO: find another way [1] to implement this. 4588 * Passing a zeroed structure is fragile, 4589 * but at least we do not pass garbage. 4590 * 4591 * [1] One way would be that ndo_neigh_setup() never touch 4592 * struct neigh_parms, but propagate the new neigh_setup() 4593 * back to ___neigh_create() / neigh_parms_alloc() 4594 */ 4595 memset(&parms, 0, sizeof(parms)); 4596 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4597 4598 if (ret) 4599 goto out; 4600 4601 if (parms.neigh_setup) 4602 ret = parms.neigh_setup(n); 4603 out: 4604 rcu_read_unlock(); 4605 return ret; 4606 } 4607 4608 /* The bonding ndo_neigh_setup is called at init time beofre any 4609 * slave exists. So we must declare proxy setup function which will 4610 * be used at run time to resolve the actual slave neigh param setup. 4611 * 4612 * It's also called by master devices (such as vlans) to setup their 4613 * underlying devices. In that case - do nothing, we're already set up from 4614 * our init. 4615 */ 4616 static int bond_neigh_setup(struct net_device *dev, 4617 struct neigh_parms *parms) 4618 { 4619 /* modify only our neigh_parms */ 4620 if (parms->dev == dev) 4621 parms->neigh_setup = bond_neigh_init; 4622 4623 return 0; 4624 } 4625 4626 /* Change the MTU of all of a master's slaves to match the master */ 4627 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4628 { 4629 struct bonding *bond = netdev_priv(bond_dev); 4630 struct slave *slave, *rollback_slave; 4631 struct list_head *iter; 4632 int res = 0; 4633 4634 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4635 4636 bond_for_each_slave(bond, slave, iter) { 4637 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4638 slave, slave->dev->netdev_ops->ndo_change_mtu); 4639 4640 res = dev_set_mtu(slave->dev, new_mtu); 4641 4642 if (res) { 4643 /* If we failed to set the slave's mtu to the new value 4644 * we must abort the operation even in ACTIVE_BACKUP 4645 * mode, because if we allow the backup slaves to have 4646 * different mtu values than the active slave we'll 4647 * need to change their mtu when doing a failover. That 4648 * means changing their mtu from timer context, which 4649 * is probably not a good idea. 4650 */ 4651 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4652 res, new_mtu); 4653 goto unwind; 4654 } 4655 } 4656 4657 bond_dev->mtu = new_mtu; 4658 4659 return 0; 4660 4661 unwind: 4662 /* unwind from head to the slave that failed */ 4663 bond_for_each_slave(bond, rollback_slave, iter) { 4664 int tmp_res; 4665 4666 if (rollback_slave == slave) 4667 break; 4668 4669 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4670 if (tmp_res) 4671 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4672 tmp_res); 4673 } 4674 4675 return res; 4676 } 4677 4678 /* Change HW address 4679 * 4680 * Note that many devices must be down to change the HW address, and 4681 * downing the master releases all slaves. We can make bonds full of 4682 * bonding devices to test this, however. 4683 */ 4684 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4685 { 4686 struct bonding *bond = netdev_priv(bond_dev); 4687 struct slave *slave, *rollback_slave; 4688 struct sockaddr_storage *ss = addr, tmp_ss; 4689 struct list_head *iter; 4690 int res = 0; 4691 4692 if (BOND_MODE(bond) == BOND_MODE_ALB) 4693 return bond_alb_set_mac_address(bond_dev, addr); 4694 4695 4696 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4697 4698 /* If fail_over_mac is enabled, do nothing and return success. 4699 * Returning an error causes ifenslave to fail. 4700 */ 4701 if (bond->params.fail_over_mac && 4702 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4703 return 0; 4704 4705 if (!is_valid_ether_addr(ss->__data)) 4706 return -EADDRNOTAVAIL; 4707 4708 bond_for_each_slave(bond, slave, iter) { 4709 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4710 __func__, slave); 4711 res = dev_set_mac_address(slave->dev, addr, NULL); 4712 if (res) { 4713 /* TODO: consider downing the slave 4714 * and retry ? 4715 * User should expect communications 4716 * breakage anyway until ARP finish 4717 * updating, so... 4718 */ 4719 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4720 __func__, res); 4721 goto unwind; 4722 } 4723 } 4724 4725 /* success */ 4726 dev_addr_set(bond_dev, ss->__data); 4727 return 0; 4728 4729 unwind: 4730 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4731 tmp_ss.ss_family = bond_dev->type; 4732 4733 /* unwind from head to the slave that failed */ 4734 bond_for_each_slave(bond, rollback_slave, iter) { 4735 int tmp_res; 4736 4737 if (rollback_slave == slave) 4738 break; 4739 4740 tmp_res = dev_set_mac_address(rollback_slave->dev, 4741 (struct sockaddr *)&tmp_ss, NULL); 4742 if (tmp_res) { 4743 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4744 __func__, tmp_res); 4745 } 4746 } 4747 4748 return res; 4749 } 4750 4751 /** 4752 * bond_get_slave_by_id - get xmit slave with slave_id 4753 * @bond: bonding device that is transmitting 4754 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4755 * 4756 * This function tries to get slave with slave_id but in case 4757 * it fails, it tries to find the first available slave for transmission. 4758 */ 4759 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4760 int slave_id) 4761 { 4762 struct list_head *iter; 4763 struct slave *slave; 4764 int i = slave_id; 4765 4766 /* Here we start from the slave with slave_id */ 4767 bond_for_each_slave_rcu(bond, slave, iter) { 4768 if (--i < 0) { 4769 if (bond_slave_can_tx(slave)) 4770 return slave; 4771 } 4772 } 4773 4774 /* Here we start from the first slave up to slave_id */ 4775 i = slave_id; 4776 bond_for_each_slave_rcu(bond, slave, iter) { 4777 if (--i < 0) 4778 break; 4779 if (bond_slave_can_tx(slave)) 4780 return slave; 4781 } 4782 /* no slave that can tx has been found */ 4783 return NULL; 4784 } 4785 4786 /** 4787 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4788 * @bond: bonding device to use 4789 * 4790 * Based on the value of the bonding device's packets_per_slave parameter 4791 * this function generates a slave id, which is usually used as the next 4792 * slave to transmit through. 4793 */ 4794 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4795 { 4796 u32 slave_id; 4797 struct reciprocal_value reciprocal_packets_per_slave; 4798 int packets_per_slave = bond->params.packets_per_slave; 4799 4800 switch (packets_per_slave) { 4801 case 0: 4802 slave_id = get_random_u32(); 4803 break; 4804 case 1: 4805 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4806 break; 4807 default: 4808 reciprocal_packets_per_slave = 4809 bond->params.reciprocal_packets_per_slave; 4810 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4811 slave_id = reciprocal_divide(slave_id, 4812 reciprocal_packets_per_slave); 4813 break; 4814 } 4815 4816 return slave_id; 4817 } 4818 4819 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4820 struct sk_buff *skb) 4821 { 4822 struct slave *slave; 4823 int slave_cnt; 4824 u32 slave_id; 4825 4826 /* Start with the curr_active_slave that joined the bond as the 4827 * default for sending IGMP traffic. For failover purposes one 4828 * needs to maintain some consistency for the interface that will 4829 * send the join/membership reports. The curr_active_slave found 4830 * will send all of this type of traffic. 4831 */ 4832 if (skb->protocol == htons(ETH_P_IP)) { 4833 int noff = skb_network_offset(skb); 4834 struct iphdr *iph; 4835 4836 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4837 goto non_igmp; 4838 4839 iph = ip_hdr(skb); 4840 if (iph->protocol == IPPROTO_IGMP) { 4841 slave = rcu_dereference(bond->curr_active_slave); 4842 if (slave) 4843 return slave; 4844 return bond_get_slave_by_id(bond, 0); 4845 } 4846 } 4847 4848 non_igmp: 4849 slave_cnt = READ_ONCE(bond->slave_cnt); 4850 if (likely(slave_cnt)) { 4851 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4852 return bond_get_slave_by_id(bond, slave_id); 4853 } 4854 return NULL; 4855 } 4856 4857 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4858 struct xdp_buff *xdp) 4859 { 4860 struct slave *slave; 4861 int slave_cnt; 4862 u32 slave_id; 4863 const struct ethhdr *eth; 4864 void *data = xdp->data; 4865 4866 if (data + sizeof(struct ethhdr) > xdp->data_end) 4867 goto non_igmp; 4868 4869 eth = (struct ethhdr *)data; 4870 data += sizeof(struct ethhdr); 4871 4872 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4873 if (eth->h_proto == htons(ETH_P_IP)) { 4874 const struct iphdr *iph; 4875 4876 if (data + sizeof(struct iphdr) > xdp->data_end) 4877 goto non_igmp; 4878 4879 iph = (struct iphdr *)data; 4880 4881 if (iph->protocol == IPPROTO_IGMP) { 4882 slave = rcu_dereference(bond->curr_active_slave); 4883 if (slave) 4884 return slave; 4885 return bond_get_slave_by_id(bond, 0); 4886 } 4887 } 4888 4889 non_igmp: 4890 slave_cnt = READ_ONCE(bond->slave_cnt); 4891 if (likely(slave_cnt)) { 4892 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4893 return bond_get_slave_by_id(bond, slave_id); 4894 } 4895 return NULL; 4896 } 4897 4898 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4899 struct net_device *bond_dev) 4900 { 4901 struct bonding *bond = netdev_priv(bond_dev); 4902 struct slave *slave; 4903 4904 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4905 if (likely(slave)) 4906 return bond_dev_queue_xmit(bond, skb, slave->dev); 4907 4908 return bond_tx_drop(bond_dev, skb); 4909 } 4910 4911 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4912 { 4913 return rcu_dereference(bond->curr_active_slave); 4914 } 4915 4916 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4917 * the bond has a usable interface. 4918 */ 4919 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4920 struct net_device *bond_dev) 4921 { 4922 struct bonding *bond = netdev_priv(bond_dev); 4923 struct slave *slave; 4924 4925 slave = bond_xmit_activebackup_slave_get(bond); 4926 if (slave) 4927 return bond_dev_queue_xmit(bond, skb, slave->dev); 4928 4929 return bond_tx_drop(bond_dev, skb); 4930 } 4931 4932 /* Use this to update slave_array when (a) it's not appropriate to update 4933 * slave_array right away (note that update_slave_array() may sleep) 4934 * and / or (b) RTNL is not held. 4935 */ 4936 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4937 { 4938 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4939 } 4940 4941 /* Slave array work handler. Holds only RTNL */ 4942 static void bond_slave_arr_handler(struct work_struct *work) 4943 { 4944 struct bonding *bond = container_of(work, struct bonding, 4945 slave_arr_work.work); 4946 int ret; 4947 4948 if (!rtnl_trylock()) 4949 goto err; 4950 4951 ret = bond_update_slave_arr(bond, NULL); 4952 rtnl_unlock(); 4953 if (ret) { 4954 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4955 goto err; 4956 } 4957 return; 4958 4959 err: 4960 bond_slave_arr_work_rearm(bond, 1); 4961 } 4962 4963 static void bond_skip_slave(struct bond_up_slave *slaves, 4964 struct slave *skipslave) 4965 { 4966 int idx; 4967 4968 /* Rare situation where caller has asked to skip a specific 4969 * slave but allocation failed (most likely!). BTW this is 4970 * only possible when the call is initiated from 4971 * __bond_release_one(). In this situation; overwrite the 4972 * skipslave entry in the array with the last entry from the 4973 * array to avoid a situation where the xmit path may choose 4974 * this to-be-skipped slave to send a packet out. 4975 */ 4976 for (idx = 0; slaves && idx < slaves->count; idx++) { 4977 if (skipslave == slaves->arr[idx]) { 4978 slaves->arr[idx] = 4979 slaves->arr[slaves->count - 1]; 4980 slaves->count--; 4981 break; 4982 } 4983 } 4984 } 4985 4986 static void bond_set_slave_arr(struct bonding *bond, 4987 struct bond_up_slave *usable_slaves, 4988 struct bond_up_slave *all_slaves) 4989 { 4990 struct bond_up_slave *usable, *all; 4991 4992 usable = rtnl_dereference(bond->usable_slaves); 4993 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 4994 kfree_rcu(usable, rcu); 4995 4996 all = rtnl_dereference(bond->all_slaves); 4997 rcu_assign_pointer(bond->all_slaves, all_slaves); 4998 kfree_rcu(all, rcu); 4999 } 5000 5001 static void bond_reset_slave_arr(struct bonding *bond) 5002 { 5003 struct bond_up_slave *usable, *all; 5004 5005 usable = rtnl_dereference(bond->usable_slaves); 5006 if (usable) { 5007 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5008 kfree_rcu(usable, rcu); 5009 } 5010 5011 all = rtnl_dereference(bond->all_slaves); 5012 if (all) { 5013 RCU_INIT_POINTER(bond->all_slaves, NULL); 5014 kfree_rcu(all, rcu); 5015 } 5016 } 5017 5018 /* Build the usable slaves array in control path for modes that use xmit-hash 5019 * to determine the slave interface - 5020 * (a) BOND_MODE_8023AD 5021 * (b) BOND_MODE_XOR 5022 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5023 * 5024 * The caller is expected to hold RTNL only and NO other lock! 5025 */ 5026 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5027 { 5028 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5029 struct slave *slave; 5030 struct list_head *iter; 5031 int agg_id = 0; 5032 int ret = 0; 5033 5034 might_sleep(); 5035 5036 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5037 bond->slave_cnt), GFP_KERNEL); 5038 all_slaves = kzalloc(struct_size(all_slaves, arr, 5039 bond->slave_cnt), GFP_KERNEL); 5040 if (!usable_slaves || !all_slaves) { 5041 ret = -ENOMEM; 5042 goto out; 5043 } 5044 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5045 struct ad_info ad_info; 5046 5047 spin_lock_bh(&bond->mode_lock); 5048 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5049 spin_unlock_bh(&bond->mode_lock); 5050 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5051 /* No active aggragator means it's not safe to use 5052 * the previous array. 5053 */ 5054 bond_reset_slave_arr(bond); 5055 goto out; 5056 } 5057 spin_unlock_bh(&bond->mode_lock); 5058 agg_id = ad_info.aggregator_id; 5059 } 5060 bond_for_each_slave(bond, slave, iter) { 5061 if (skipslave == slave) 5062 continue; 5063 5064 all_slaves->arr[all_slaves->count++] = slave; 5065 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5066 struct aggregator *agg; 5067 5068 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5069 if (!agg || agg->aggregator_identifier != agg_id) 5070 continue; 5071 } 5072 if (!bond_slave_can_tx(slave)) 5073 continue; 5074 5075 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5076 usable_slaves->count); 5077 5078 usable_slaves->arr[usable_slaves->count++] = slave; 5079 } 5080 5081 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5082 return ret; 5083 out: 5084 if (ret != 0 && skipslave) { 5085 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5086 skipslave); 5087 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5088 skipslave); 5089 } 5090 kfree_rcu(all_slaves, rcu); 5091 kfree_rcu(usable_slaves, rcu); 5092 5093 return ret; 5094 } 5095 5096 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5097 struct sk_buff *skb, 5098 struct bond_up_slave *slaves) 5099 { 5100 struct slave *slave; 5101 unsigned int count; 5102 u32 hash; 5103 5104 hash = bond_xmit_hash(bond, skb); 5105 count = slaves ? READ_ONCE(slaves->count) : 0; 5106 if (unlikely(!count)) 5107 return NULL; 5108 5109 slave = slaves->arr[hash % count]; 5110 return slave; 5111 } 5112 5113 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5114 struct xdp_buff *xdp) 5115 { 5116 struct bond_up_slave *slaves; 5117 unsigned int count; 5118 u32 hash; 5119 5120 hash = bond_xmit_hash_xdp(bond, xdp); 5121 slaves = rcu_dereference(bond->usable_slaves); 5122 count = slaves ? READ_ONCE(slaves->count) : 0; 5123 if (unlikely(!count)) 5124 return NULL; 5125 5126 return slaves->arr[hash % count]; 5127 } 5128 5129 /* Use this Xmit function for 3AD as well as XOR modes. The current 5130 * usable slave array is formed in the control path. The xmit function 5131 * just calculates hash and sends the packet out. 5132 */ 5133 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5134 struct net_device *dev) 5135 { 5136 struct bonding *bond = netdev_priv(dev); 5137 struct bond_up_slave *slaves; 5138 struct slave *slave; 5139 5140 slaves = rcu_dereference(bond->usable_slaves); 5141 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5142 if (likely(slave)) 5143 return bond_dev_queue_xmit(bond, skb, slave->dev); 5144 5145 return bond_tx_drop(dev, skb); 5146 } 5147 5148 /* in broadcast mode, we send everything to all usable interfaces. */ 5149 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5150 struct net_device *bond_dev) 5151 { 5152 struct bonding *bond = netdev_priv(bond_dev); 5153 struct slave *slave = NULL; 5154 struct list_head *iter; 5155 bool xmit_suc = false; 5156 bool skb_used = false; 5157 5158 bond_for_each_slave_rcu(bond, slave, iter) { 5159 struct sk_buff *skb2; 5160 5161 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5162 continue; 5163 5164 if (bond_is_last_slave(bond, slave)) { 5165 skb2 = skb; 5166 skb_used = true; 5167 } else { 5168 skb2 = skb_clone(skb, GFP_ATOMIC); 5169 if (!skb2) { 5170 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5171 bond_dev->name, __func__); 5172 continue; 5173 } 5174 } 5175 5176 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5177 xmit_suc = true; 5178 } 5179 5180 if (!skb_used) 5181 dev_kfree_skb_any(skb); 5182 5183 if (xmit_suc) 5184 return NETDEV_TX_OK; 5185 5186 dev_core_stats_tx_dropped_inc(bond_dev); 5187 return NET_XMIT_DROP; 5188 } 5189 5190 /*------------------------- Device initialization ---------------------------*/ 5191 5192 /* Lookup the slave that corresponds to a qid */ 5193 static inline int bond_slave_override(struct bonding *bond, 5194 struct sk_buff *skb) 5195 { 5196 struct slave *slave = NULL; 5197 struct list_head *iter; 5198 5199 if (!skb_rx_queue_recorded(skb)) 5200 return 1; 5201 5202 /* Find out if any slaves have the same mapping as this skb. */ 5203 bond_for_each_slave_rcu(bond, slave, iter) { 5204 if (slave->queue_id == skb_get_queue_mapping(skb)) { 5205 if (bond_slave_is_up(slave) && 5206 slave->link == BOND_LINK_UP) { 5207 bond_dev_queue_xmit(bond, skb, slave->dev); 5208 return 0; 5209 } 5210 /* If the slave isn't UP, use default transmit policy. */ 5211 break; 5212 } 5213 } 5214 5215 return 1; 5216 } 5217 5218 5219 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5220 struct net_device *sb_dev) 5221 { 5222 /* This helper function exists to help dev_pick_tx get the correct 5223 * destination queue. Using a helper function skips a call to 5224 * skb_tx_hash and will put the skbs in the queue we expect on their 5225 * way down to the bonding driver. 5226 */ 5227 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5228 5229 /* Save the original txq to restore before passing to the driver */ 5230 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5231 5232 if (unlikely(txq >= dev->real_num_tx_queues)) { 5233 do { 5234 txq -= dev->real_num_tx_queues; 5235 } while (txq >= dev->real_num_tx_queues); 5236 } 5237 return txq; 5238 } 5239 5240 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5241 struct sk_buff *skb, 5242 bool all_slaves) 5243 { 5244 struct bonding *bond = netdev_priv(master_dev); 5245 struct bond_up_slave *slaves; 5246 struct slave *slave = NULL; 5247 5248 switch (BOND_MODE(bond)) { 5249 case BOND_MODE_ROUNDROBIN: 5250 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5251 break; 5252 case BOND_MODE_ACTIVEBACKUP: 5253 slave = bond_xmit_activebackup_slave_get(bond); 5254 break; 5255 case BOND_MODE_8023AD: 5256 case BOND_MODE_XOR: 5257 if (all_slaves) 5258 slaves = rcu_dereference(bond->all_slaves); 5259 else 5260 slaves = rcu_dereference(bond->usable_slaves); 5261 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5262 break; 5263 case BOND_MODE_BROADCAST: 5264 break; 5265 case BOND_MODE_ALB: 5266 slave = bond_xmit_alb_slave_get(bond, skb); 5267 break; 5268 case BOND_MODE_TLB: 5269 slave = bond_xmit_tlb_slave_get(bond, skb); 5270 break; 5271 default: 5272 /* Should never happen, mode already checked */ 5273 WARN_ONCE(true, "Unknown bonding mode"); 5274 break; 5275 } 5276 5277 if (slave) 5278 return slave->dev; 5279 return NULL; 5280 } 5281 5282 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5283 { 5284 switch (sk->sk_family) { 5285 #if IS_ENABLED(CONFIG_IPV6) 5286 case AF_INET6: 5287 if (ipv6_only_sock(sk) || 5288 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5289 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5290 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5291 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5292 break; 5293 } 5294 fallthrough; 5295 #endif 5296 default: /* AF_INET */ 5297 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5298 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5299 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5300 break; 5301 } 5302 5303 flow->ports.src = inet_sk(sk)->inet_sport; 5304 flow->ports.dst = inet_sk(sk)->inet_dport; 5305 } 5306 5307 /** 5308 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5309 * @sk: socket to use for headers 5310 * 5311 * This function will extract the necessary field from the socket and use 5312 * them to generate a hash based on the LAYER34 xmit_policy. 5313 * Assumes that sk is a TCP or UDP socket. 5314 */ 5315 static u32 bond_sk_hash_l34(struct sock *sk) 5316 { 5317 struct flow_keys flow; 5318 u32 hash; 5319 5320 bond_sk_to_flow(sk, &flow); 5321 5322 /* L4 */ 5323 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5324 /* L3 */ 5325 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5326 } 5327 5328 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5329 struct sock *sk) 5330 { 5331 struct bond_up_slave *slaves; 5332 struct slave *slave; 5333 unsigned int count; 5334 u32 hash; 5335 5336 slaves = rcu_dereference(bond->usable_slaves); 5337 count = slaves ? READ_ONCE(slaves->count) : 0; 5338 if (unlikely(!count)) 5339 return NULL; 5340 5341 hash = bond_sk_hash_l34(sk); 5342 slave = slaves->arr[hash % count]; 5343 5344 return slave->dev; 5345 } 5346 5347 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5348 struct sock *sk) 5349 { 5350 struct bonding *bond = netdev_priv(dev); 5351 struct net_device *lower = NULL; 5352 5353 rcu_read_lock(); 5354 if (bond_sk_check(bond)) 5355 lower = __bond_sk_get_lower_dev(bond, sk); 5356 rcu_read_unlock(); 5357 5358 return lower; 5359 } 5360 5361 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5362 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5363 struct net_device *dev) 5364 { 5365 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5366 5367 /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded 5368 * was true, if tls_device_down is running in parallel, but it's OK, 5369 * because bond_get_slave_by_dev has a NULL check. 5370 */ 5371 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5372 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5373 return bond_tx_drop(dev, skb); 5374 } 5375 #endif 5376 5377 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5378 { 5379 struct bonding *bond = netdev_priv(dev); 5380 5381 if (bond_should_override_tx_queue(bond) && 5382 !bond_slave_override(bond, skb)) 5383 return NETDEV_TX_OK; 5384 5385 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5386 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk)) 5387 return bond_tls_device_xmit(bond, skb, dev); 5388 #endif 5389 5390 switch (BOND_MODE(bond)) { 5391 case BOND_MODE_ROUNDROBIN: 5392 return bond_xmit_roundrobin(skb, dev); 5393 case BOND_MODE_ACTIVEBACKUP: 5394 return bond_xmit_activebackup(skb, dev); 5395 case BOND_MODE_8023AD: 5396 case BOND_MODE_XOR: 5397 return bond_3ad_xor_xmit(skb, dev); 5398 case BOND_MODE_BROADCAST: 5399 return bond_xmit_broadcast(skb, dev); 5400 case BOND_MODE_ALB: 5401 return bond_alb_xmit(skb, dev); 5402 case BOND_MODE_TLB: 5403 return bond_tlb_xmit(skb, dev); 5404 default: 5405 /* Should never happen, mode already checked */ 5406 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5407 WARN_ON_ONCE(1); 5408 return bond_tx_drop(dev, skb); 5409 } 5410 } 5411 5412 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5413 { 5414 struct bonding *bond = netdev_priv(dev); 5415 netdev_tx_t ret = NETDEV_TX_OK; 5416 5417 /* If we risk deadlock from transmitting this in the 5418 * netpoll path, tell netpoll to queue the frame for later tx 5419 */ 5420 if (unlikely(is_netpoll_tx_blocked(dev))) 5421 return NETDEV_TX_BUSY; 5422 5423 rcu_read_lock(); 5424 if (bond_has_slaves(bond)) 5425 ret = __bond_start_xmit(skb, dev); 5426 else 5427 ret = bond_tx_drop(dev, skb); 5428 rcu_read_unlock(); 5429 5430 return ret; 5431 } 5432 5433 static struct net_device * 5434 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5435 { 5436 struct bonding *bond = netdev_priv(bond_dev); 5437 struct slave *slave; 5438 5439 /* Caller needs to hold rcu_read_lock() */ 5440 5441 switch (BOND_MODE(bond)) { 5442 case BOND_MODE_ROUNDROBIN: 5443 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5444 break; 5445 5446 case BOND_MODE_ACTIVEBACKUP: 5447 slave = bond_xmit_activebackup_slave_get(bond); 5448 break; 5449 5450 case BOND_MODE_8023AD: 5451 case BOND_MODE_XOR: 5452 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5453 break; 5454 5455 default: 5456 /* Should never happen. Mode guarded by bond_xdp_check() */ 5457 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond)); 5458 WARN_ON_ONCE(1); 5459 return NULL; 5460 } 5461 5462 if (slave) 5463 return slave->dev; 5464 5465 return NULL; 5466 } 5467 5468 static int bond_xdp_xmit(struct net_device *bond_dev, 5469 int n, struct xdp_frame **frames, u32 flags) 5470 { 5471 int nxmit, err = -ENXIO; 5472 5473 rcu_read_lock(); 5474 5475 for (nxmit = 0; nxmit < n; nxmit++) { 5476 struct xdp_frame *frame = frames[nxmit]; 5477 struct xdp_frame *frames1[] = {frame}; 5478 struct net_device *slave_dev; 5479 struct xdp_buff xdp; 5480 5481 xdp_convert_frame_to_buff(frame, &xdp); 5482 5483 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5484 if (!slave_dev) { 5485 err = -ENXIO; 5486 break; 5487 } 5488 5489 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5490 if (err < 1) 5491 break; 5492 } 5493 5494 rcu_read_unlock(); 5495 5496 /* If error happened on the first frame then we can pass the error up, otherwise 5497 * report the number of frames that were xmitted. 5498 */ 5499 if (err < 0) 5500 return (nxmit == 0 ? err : nxmit); 5501 5502 return nxmit; 5503 } 5504 5505 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5506 struct netlink_ext_ack *extack) 5507 { 5508 struct bonding *bond = netdev_priv(dev); 5509 struct list_head *iter; 5510 struct slave *slave, *rollback_slave; 5511 struct bpf_prog *old_prog; 5512 struct netdev_bpf xdp = { 5513 .command = XDP_SETUP_PROG, 5514 .flags = 0, 5515 .prog = prog, 5516 .extack = extack, 5517 }; 5518 int err; 5519 5520 ASSERT_RTNL(); 5521 5522 if (!bond_xdp_check(bond)) 5523 return -EOPNOTSUPP; 5524 5525 old_prog = bond->xdp_prog; 5526 bond->xdp_prog = prog; 5527 5528 bond_for_each_slave(bond, slave, iter) { 5529 struct net_device *slave_dev = slave->dev; 5530 5531 if (!slave_dev->netdev_ops->ndo_bpf || 5532 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5533 SLAVE_NL_ERR(dev, slave_dev, extack, 5534 "Slave device does not support XDP"); 5535 err = -EOPNOTSUPP; 5536 goto err; 5537 } 5538 5539 if (dev_xdp_prog_count(slave_dev) > 0) { 5540 SLAVE_NL_ERR(dev, slave_dev, extack, 5541 "Slave has XDP program loaded, please unload before enslaving"); 5542 err = -EOPNOTSUPP; 5543 goto err; 5544 } 5545 5546 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5547 if (err < 0) { 5548 /* ndo_bpf() sets extack error message */ 5549 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5550 goto err; 5551 } 5552 if (prog) 5553 bpf_prog_inc(prog); 5554 } 5555 5556 if (prog) { 5557 static_branch_inc(&bpf_master_redirect_enabled_key); 5558 } else if (old_prog) { 5559 bpf_prog_put(old_prog); 5560 static_branch_dec(&bpf_master_redirect_enabled_key); 5561 } 5562 5563 return 0; 5564 5565 err: 5566 /* unwind the program changes */ 5567 bond->xdp_prog = old_prog; 5568 xdp.prog = old_prog; 5569 xdp.extack = NULL; /* do not overwrite original error */ 5570 5571 bond_for_each_slave(bond, rollback_slave, iter) { 5572 struct net_device *slave_dev = rollback_slave->dev; 5573 int err_unwind; 5574 5575 if (slave == rollback_slave) 5576 break; 5577 5578 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5579 if (err_unwind < 0) 5580 slave_err(dev, slave_dev, 5581 "Error %d when unwinding XDP program change\n", err_unwind); 5582 else if (xdp.prog) 5583 bpf_prog_inc(xdp.prog); 5584 } 5585 return err; 5586 } 5587 5588 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5589 { 5590 switch (xdp->command) { 5591 case XDP_SETUP_PROG: 5592 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5593 default: 5594 return -EINVAL; 5595 } 5596 } 5597 5598 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5599 { 5600 if (speed == 0 || speed == SPEED_UNKNOWN) 5601 speed = slave->speed; 5602 else 5603 speed = min(speed, slave->speed); 5604 5605 return speed; 5606 } 5607 5608 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5609 struct ethtool_link_ksettings *cmd) 5610 { 5611 struct bonding *bond = netdev_priv(bond_dev); 5612 struct list_head *iter; 5613 struct slave *slave; 5614 u32 speed = 0; 5615 5616 cmd->base.duplex = DUPLEX_UNKNOWN; 5617 cmd->base.port = PORT_OTHER; 5618 5619 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5620 * do not need to check mode. Though link speed might not represent 5621 * the true receive or transmit bandwidth (not all modes are symmetric) 5622 * this is an accurate maximum. 5623 */ 5624 bond_for_each_slave(bond, slave, iter) { 5625 if (bond_slave_can_tx(slave)) { 5626 if (slave->speed != SPEED_UNKNOWN) { 5627 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5628 speed = bond_mode_bcast_speed(slave, 5629 speed); 5630 else 5631 speed += slave->speed; 5632 } 5633 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5634 slave->duplex != DUPLEX_UNKNOWN) 5635 cmd->base.duplex = slave->duplex; 5636 } 5637 } 5638 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5639 5640 return 0; 5641 } 5642 5643 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5644 struct ethtool_drvinfo *drvinfo) 5645 { 5646 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5647 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5648 BOND_ABI_VERSION); 5649 } 5650 5651 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5652 struct ethtool_ts_info *info) 5653 { 5654 struct bonding *bond = netdev_priv(bond_dev); 5655 const struct ethtool_ops *ops; 5656 struct net_device *real_dev; 5657 struct phy_device *phydev; 5658 int ret = 0; 5659 5660 rcu_read_lock(); 5661 real_dev = bond_option_active_slave_get_rcu(bond); 5662 dev_hold(real_dev); 5663 rcu_read_unlock(); 5664 5665 if (real_dev) { 5666 ops = real_dev->ethtool_ops; 5667 phydev = real_dev->phydev; 5668 5669 if (phy_has_tsinfo(phydev)) { 5670 ret = phy_ts_info(phydev, info); 5671 goto out; 5672 } else if (ops->get_ts_info) { 5673 ret = ops->get_ts_info(real_dev, info); 5674 goto out; 5675 } 5676 } 5677 5678 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE | 5679 SOF_TIMESTAMPING_SOFTWARE; 5680 info->phc_index = -1; 5681 5682 out: 5683 dev_put(real_dev); 5684 return ret; 5685 } 5686 5687 static const struct ethtool_ops bond_ethtool_ops = { 5688 .get_drvinfo = bond_ethtool_get_drvinfo, 5689 .get_link = ethtool_op_get_link, 5690 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5691 .get_ts_info = bond_ethtool_get_ts_info, 5692 }; 5693 5694 static const struct net_device_ops bond_netdev_ops = { 5695 .ndo_init = bond_init, 5696 .ndo_uninit = bond_uninit, 5697 .ndo_open = bond_open, 5698 .ndo_stop = bond_close, 5699 .ndo_start_xmit = bond_start_xmit, 5700 .ndo_select_queue = bond_select_queue, 5701 .ndo_get_stats64 = bond_get_stats, 5702 .ndo_eth_ioctl = bond_eth_ioctl, 5703 .ndo_siocbond = bond_do_ioctl, 5704 .ndo_siocdevprivate = bond_siocdevprivate, 5705 .ndo_change_rx_flags = bond_change_rx_flags, 5706 .ndo_set_rx_mode = bond_set_rx_mode, 5707 .ndo_change_mtu = bond_change_mtu, 5708 .ndo_set_mac_address = bond_set_mac_address, 5709 .ndo_neigh_setup = bond_neigh_setup, 5710 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5711 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5712 #ifdef CONFIG_NET_POLL_CONTROLLER 5713 .ndo_netpoll_setup = bond_netpoll_setup, 5714 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5715 .ndo_poll_controller = bond_poll_controller, 5716 #endif 5717 .ndo_add_slave = bond_enslave, 5718 .ndo_del_slave = bond_release, 5719 .ndo_fix_features = bond_fix_features, 5720 .ndo_features_check = passthru_features_check, 5721 .ndo_get_xmit_slave = bond_xmit_get_slave, 5722 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5723 .ndo_bpf = bond_xdp, 5724 .ndo_xdp_xmit = bond_xdp_xmit, 5725 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5726 }; 5727 5728 static const struct device_type bond_type = { 5729 .name = "bond", 5730 }; 5731 5732 static void bond_destructor(struct net_device *bond_dev) 5733 { 5734 struct bonding *bond = netdev_priv(bond_dev); 5735 5736 if (bond->wq) 5737 destroy_workqueue(bond->wq); 5738 5739 if (bond->rr_tx_counter) 5740 free_percpu(bond->rr_tx_counter); 5741 } 5742 5743 void bond_setup(struct net_device *bond_dev) 5744 { 5745 struct bonding *bond = netdev_priv(bond_dev); 5746 5747 spin_lock_init(&bond->mode_lock); 5748 bond->params = bonding_defaults; 5749 5750 /* Initialize pointers */ 5751 bond->dev = bond_dev; 5752 5753 /* Initialize the device entry points */ 5754 ether_setup(bond_dev); 5755 bond_dev->max_mtu = ETH_MAX_MTU; 5756 bond_dev->netdev_ops = &bond_netdev_ops; 5757 bond_dev->ethtool_ops = &bond_ethtool_ops; 5758 5759 bond_dev->needs_free_netdev = true; 5760 bond_dev->priv_destructor = bond_destructor; 5761 5762 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5763 5764 /* Initialize the device options */ 5765 bond_dev->flags |= IFF_MASTER; 5766 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5767 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5768 5769 #ifdef CONFIG_XFRM_OFFLOAD 5770 /* set up xfrm device ops (only supported in active-backup right now) */ 5771 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5772 INIT_LIST_HEAD(&bond->ipsec_list); 5773 spin_lock_init(&bond->ipsec_lock); 5774 #endif /* CONFIG_XFRM_OFFLOAD */ 5775 5776 /* don't acquire bond device's netif_tx_lock when transmitting */ 5777 bond_dev->features |= NETIF_F_LLTX; 5778 5779 /* By default, we declare the bond to be fully 5780 * VLAN hardware accelerated capable. Special 5781 * care is taken in the various xmit functions 5782 * when there are slaves that are not hw accel 5783 * capable 5784 */ 5785 5786 /* Don't allow bond devices to change network namespaces. */ 5787 bond_dev->features |= NETIF_F_NETNS_LOCAL; 5788 5789 bond_dev->hw_features = BOND_VLAN_FEATURES | 5790 NETIF_F_HW_VLAN_CTAG_RX | 5791 NETIF_F_HW_VLAN_CTAG_FILTER; 5792 5793 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5794 bond_dev->features |= bond_dev->hw_features; 5795 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5796 #ifdef CONFIG_XFRM_OFFLOAD 5797 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5798 /* Only enable XFRM features if this is an active-backup config */ 5799 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5800 bond_dev->features |= BOND_XFRM_FEATURES; 5801 #endif /* CONFIG_XFRM_OFFLOAD */ 5802 } 5803 5804 /* Destroy a bonding device. 5805 * Must be under rtnl_lock when this function is called. 5806 */ 5807 static void bond_uninit(struct net_device *bond_dev) 5808 { 5809 struct bonding *bond = netdev_priv(bond_dev); 5810 struct bond_up_slave *usable, *all; 5811 struct list_head *iter; 5812 struct slave *slave; 5813 5814 bond_netpoll_cleanup(bond_dev); 5815 5816 /* Release the bonded slaves */ 5817 bond_for_each_slave(bond, slave, iter) 5818 __bond_release_one(bond_dev, slave->dev, true, true); 5819 netdev_info(bond_dev, "Released all slaves\n"); 5820 5821 usable = rtnl_dereference(bond->usable_slaves); 5822 if (usable) { 5823 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5824 kfree_rcu(usable, rcu); 5825 } 5826 5827 all = rtnl_dereference(bond->all_slaves); 5828 if (all) { 5829 RCU_INIT_POINTER(bond->all_slaves, NULL); 5830 kfree_rcu(all, rcu); 5831 } 5832 5833 list_del(&bond->bond_list); 5834 5835 bond_debug_unregister(bond); 5836 } 5837 5838 /*------------------------- Module initialization ---------------------------*/ 5839 5840 static int bond_check_params(struct bond_params *params) 5841 { 5842 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 5843 struct bond_opt_value newval; 5844 const struct bond_opt_value *valptr; 5845 int arp_all_targets_value = 0; 5846 u16 ad_actor_sys_prio = 0; 5847 u16 ad_user_port_key = 0; 5848 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 5849 int arp_ip_count; 5850 int bond_mode = BOND_MODE_ROUNDROBIN; 5851 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 5852 int lacp_fast = 0; 5853 int tlb_dynamic_lb; 5854 5855 /* Convert string parameters. */ 5856 if (mode) { 5857 bond_opt_initstr(&newval, mode); 5858 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5859 if (!valptr) { 5860 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5861 return -EINVAL; 5862 } 5863 bond_mode = valptr->value; 5864 } 5865 5866 if (xmit_hash_policy) { 5867 if (bond_mode == BOND_MODE_ROUNDROBIN || 5868 bond_mode == BOND_MODE_ACTIVEBACKUP || 5869 bond_mode == BOND_MODE_BROADCAST) { 5870 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 5871 bond_mode_name(bond_mode)); 5872 } else { 5873 bond_opt_initstr(&newval, xmit_hash_policy); 5874 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5875 &newval); 5876 if (!valptr) { 5877 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5878 xmit_hash_policy); 5879 return -EINVAL; 5880 } 5881 xmit_hashtype = valptr->value; 5882 } 5883 } 5884 5885 if (lacp_rate) { 5886 if (bond_mode != BOND_MODE_8023AD) { 5887 pr_info("lacp_rate param is irrelevant in mode %s\n", 5888 bond_mode_name(bond_mode)); 5889 } else { 5890 bond_opt_initstr(&newval, lacp_rate); 5891 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5892 &newval); 5893 if (!valptr) { 5894 pr_err("Error: Invalid lacp rate \"%s\"\n", 5895 lacp_rate); 5896 return -EINVAL; 5897 } 5898 lacp_fast = valptr->value; 5899 } 5900 } 5901 5902 if (ad_select) { 5903 bond_opt_initstr(&newval, ad_select); 5904 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5905 &newval); 5906 if (!valptr) { 5907 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5908 return -EINVAL; 5909 } 5910 params->ad_select = valptr->value; 5911 if (bond_mode != BOND_MODE_8023AD) 5912 pr_warn("ad_select param only affects 802.3ad mode\n"); 5913 } else { 5914 params->ad_select = BOND_AD_STABLE; 5915 } 5916 5917 if (max_bonds < 0) { 5918 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5919 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5920 max_bonds = BOND_DEFAULT_MAX_BONDS; 5921 } 5922 5923 if (miimon < 0) { 5924 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5925 miimon, INT_MAX); 5926 miimon = 0; 5927 } 5928 5929 if (updelay < 0) { 5930 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5931 updelay, INT_MAX); 5932 updelay = 0; 5933 } 5934 5935 if (downdelay < 0) { 5936 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5937 downdelay, INT_MAX); 5938 downdelay = 0; 5939 } 5940 5941 if ((use_carrier != 0) && (use_carrier != 1)) { 5942 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5943 use_carrier); 5944 use_carrier = 1; 5945 } 5946 5947 if (num_peer_notif < 0 || num_peer_notif > 255) { 5948 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5949 num_peer_notif); 5950 num_peer_notif = 1; 5951 } 5952 5953 /* reset values for 802.3ad/TLB/ALB */ 5954 if (!bond_mode_uses_arp(bond_mode)) { 5955 if (!miimon) { 5956 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"); 5957 pr_warn("Forcing miimon to 100msec\n"); 5958 miimon = BOND_DEFAULT_MIIMON; 5959 } 5960 } 5961 5962 if (tx_queues < 1 || tx_queues > 255) { 5963 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5964 tx_queues, BOND_DEFAULT_TX_QUEUES); 5965 tx_queues = BOND_DEFAULT_TX_QUEUES; 5966 } 5967 5968 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5969 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5970 all_slaves_active); 5971 all_slaves_active = 0; 5972 } 5973 5974 if (resend_igmp < 0 || resend_igmp > 255) { 5975 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 5976 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 5977 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 5978 } 5979 5980 bond_opt_initval(&newval, packets_per_slave); 5981 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 5982 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 5983 packets_per_slave, USHRT_MAX); 5984 packets_per_slave = 1; 5985 } 5986 5987 if (bond_mode == BOND_MODE_ALB) { 5988 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", 5989 updelay); 5990 } 5991 5992 if (!miimon) { 5993 if (updelay || downdelay) { 5994 /* just warn the user the up/down delay will have 5995 * no effect since miimon is zero... 5996 */ 5997 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", 5998 updelay, downdelay); 5999 } 6000 } else { 6001 /* don't allow arp monitoring */ 6002 if (arp_interval) { 6003 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6004 miimon, arp_interval); 6005 arp_interval = 0; 6006 } 6007 6008 if ((updelay % miimon) != 0) { 6009 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6010 updelay, miimon, (updelay / miimon) * miimon); 6011 } 6012 6013 updelay /= miimon; 6014 6015 if ((downdelay % miimon) != 0) { 6016 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6017 downdelay, miimon, 6018 (downdelay / miimon) * miimon); 6019 } 6020 6021 downdelay /= miimon; 6022 } 6023 6024 if (arp_interval < 0) { 6025 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6026 arp_interval, INT_MAX); 6027 arp_interval = 0; 6028 } 6029 6030 for (arp_ip_count = 0, i = 0; 6031 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6032 __be32 ip; 6033 6034 /* not a complete check, but good enough to catch mistakes */ 6035 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6036 !bond_is_ip_target_ok(ip)) { 6037 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6038 arp_ip_target[i]); 6039 arp_interval = 0; 6040 } else { 6041 if (bond_get_targets_ip(arp_target, ip) == -1) 6042 arp_target[arp_ip_count++] = ip; 6043 else 6044 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6045 &ip); 6046 } 6047 } 6048 6049 if (arp_interval && !arp_ip_count) { 6050 /* don't allow arping if no arp_ip_target given... */ 6051 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6052 arp_interval); 6053 arp_interval = 0; 6054 } 6055 6056 if (arp_validate) { 6057 if (!arp_interval) { 6058 pr_err("arp_validate requires arp_interval\n"); 6059 return -EINVAL; 6060 } 6061 6062 bond_opt_initstr(&newval, arp_validate); 6063 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6064 &newval); 6065 if (!valptr) { 6066 pr_err("Error: invalid arp_validate \"%s\"\n", 6067 arp_validate); 6068 return -EINVAL; 6069 } 6070 arp_validate_value = valptr->value; 6071 } else { 6072 arp_validate_value = 0; 6073 } 6074 6075 if (arp_all_targets) { 6076 bond_opt_initstr(&newval, arp_all_targets); 6077 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6078 &newval); 6079 if (!valptr) { 6080 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6081 arp_all_targets); 6082 arp_all_targets_value = 0; 6083 } else { 6084 arp_all_targets_value = valptr->value; 6085 } 6086 } 6087 6088 if (miimon) { 6089 pr_info("MII link monitoring set to %d ms\n", miimon); 6090 } else if (arp_interval) { 6091 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6092 arp_validate_value); 6093 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6094 arp_interval, valptr->string, arp_ip_count); 6095 6096 for (i = 0; i < arp_ip_count; i++) 6097 pr_cont(" %s", arp_ip_target[i]); 6098 6099 pr_cont("\n"); 6100 6101 } else if (max_bonds) { 6102 /* miimon and arp_interval not set, we need one so things 6103 * work as expected, see bonding.txt for details 6104 */ 6105 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"); 6106 } 6107 6108 if (primary && !bond_mode_uses_primary(bond_mode)) { 6109 /* currently, using a primary only makes sense 6110 * in active backup, TLB or ALB modes 6111 */ 6112 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6113 primary, bond_mode_name(bond_mode)); 6114 primary = NULL; 6115 } 6116 6117 if (primary && primary_reselect) { 6118 bond_opt_initstr(&newval, primary_reselect); 6119 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6120 &newval); 6121 if (!valptr) { 6122 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6123 primary_reselect); 6124 return -EINVAL; 6125 } 6126 primary_reselect_value = valptr->value; 6127 } else { 6128 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6129 } 6130 6131 if (fail_over_mac) { 6132 bond_opt_initstr(&newval, fail_over_mac); 6133 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6134 &newval); 6135 if (!valptr) { 6136 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6137 fail_over_mac); 6138 return -EINVAL; 6139 } 6140 fail_over_mac_value = valptr->value; 6141 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6142 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6143 } else { 6144 fail_over_mac_value = BOND_FOM_NONE; 6145 } 6146 6147 bond_opt_initstr(&newval, "default"); 6148 valptr = bond_opt_parse( 6149 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6150 &newval); 6151 if (!valptr) { 6152 pr_err("Error: No ad_actor_sys_prio default value"); 6153 return -EINVAL; 6154 } 6155 ad_actor_sys_prio = valptr->value; 6156 6157 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6158 &newval); 6159 if (!valptr) { 6160 pr_err("Error: No ad_user_port_key default value"); 6161 return -EINVAL; 6162 } 6163 ad_user_port_key = valptr->value; 6164 6165 bond_opt_initstr(&newval, "default"); 6166 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6167 if (!valptr) { 6168 pr_err("Error: No tlb_dynamic_lb default value"); 6169 return -EINVAL; 6170 } 6171 tlb_dynamic_lb = valptr->value; 6172 6173 if (lp_interval == 0) { 6174 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6175 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6176 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6177 } 6178 6179 /* fill params struct with the proper values */ 6180 params->mode = bond_mode; 6181 params->xmit_policy = xmit_hashtype; 6182 params->miimon = miimon; 6183 params->num_peer_notif = num_peer_notif; 6184 params->arp_interval = arp_interval; 6185 params->arp_validate = arp_validate_value; 6186 params->arp_all_targets = arp_all_targets_value; 6187 params->missed_max = 2; 6188 params->updelay = updelay; 6189 params->downdelay = downdelay; 6190 params->peer_notif_delay = 0; 6191 params->use_carrier = use_carrier; 6192 params->lacp_active = 1; 6193 params->lacp_fast = lacp_fast; 6194 params->primary[0] = 0; 6195 params->primary_reselect = primary_reselect_value; 6196 params->fail_over_mac = fail_over_mac_value; 6197 params->tx_queues = tx_queues; 6198 params->all_slaves_active = all_slaves_active; 6199 params->resend_igmp = resend_igmp; 6200 params->min_links = min_links; 6201 params->lp_interval = lp_interval; 6202 params->packets_per_slave = packets_per_slave; 6203 params->tlb_dynamic_lb = tlb_dynamic_lb; 6204 params->ad_actor_sys_prio = ad_actor_sys_prio; 6205 eth_zero_addr(params->ad_actor_system); 6206 params->ad_user_port_key = ad_user_port_key; 6207 if (packets_per_slave > 0) { 6208 params->reciprocal_packets_per_slave = 6209 reciprocal_value(packets_per_slave); 6210 } else { 6211 /* reciprocal_packets_per_slave is unused if 6212 * packets_per_slave is 0 or 1, just initialize it 6213 */ 6214 params->reciprocal_packets_per_slave = 6215 (struct reciprocal_value) { 0 }; 6216 } 6217 6218 if (primary) 6219 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6220 6221 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6222 #if IS_ENABLED(CONFIG_IPV6) 6223 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6224 #endif 6225 6226 return 0; 6227 } 6228 6229 /* Called from registration process */ 6230 static int bond_init(struct net_device *bond_dev) 6231 { 6232 struct bonding *bond = netdev_priv(bond_dev); 6233 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6234 6235 netdev_dbg(bond_dev, "Begin bond_init\n"); 6236 6237 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 6238 if (!bond->wq) 6239 return -ENOMEM; 6240 6241 spin_lock_init(&bond->stats_lock); 6242 netdev_lockdep_set_classes(bond_dev); 6243 6244 list_add_tail(&bond->bond_list, &bn->dev_list); 6245 6246 bond_prepare_sysfs_group(bond); 6247 6248 bond_debug_register(bond); 6249 6250 /* Ensure valid dev_addr */ 6251 if (is_zero_ether_addr(bond_dev->dev_addr) && 6252 bond_dev->addr_assign_type == NET_ADDR_PERM) 6253 eth_hw_addr_random(bond_dev); 6254 6255 return 0; 6256 } 6257 6258 unsigned int bond_get_num_tx_queues(void) 6259 { 6260 return tx_queues; 6261 } 6262 6263 /* Create a new bond based on the specified name and bonding parameters. 6264 * If name is NULL, obtain a suitable "bond%d" name for us. 6265 * Caller must NOT hold rtnl_lock; we need to release it here before we 6266 * set up our sysfs entries. 6267 */ 6268 int bond_create(struct net *net, const char *name) 6269 { 6270 struct net_device *bond_dev; 6271 struct bonding *bond; 6272 int res = -ENOMEM; 6273 6274 rtnl_lock(); 6275 6276 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6277 name ? name : "bond%d", NET_NAME_UNKNOWN, 6278 bond_setup, tx_queues); 6279 if (!bond_dev) 6280 goto out; 6281 6282 bond = netdev_priv(bond_dev); 6283 dev_net_set(bond_dev, net); 6284 bond_dev->rtnl_link_ops = &bond_link_ops; 6285 6286 res = register_netdevice(bond_dev); 6287 if (res < 0) { 6288 free_netdev(bond_dev); 6289 goto out; 6290 } 6291 6292 netif_carrier_off(bond_dev); 6293 6294 bond_work_init_all(bond); 6295 6296 out: 6297 rtnl_unlock(); 6298 return res; 6299 } 6300 6301 static int __net_init bond_net_init(struct net *net) 6302 { 6303 struct bond_net *bn = net_generic(net, bond_net_id); 6304 6305 bn->net = net; 6306 INIT_LIST_HEAD(&bn->dev_list); 6307 6308 bond_create_proc_dir(bn); 6309 bond_create_sysfs(bn); 6310 6311 return 0; 6312 } 6313 6314 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6315 { 6316 struct bond_net *bn; 6317 struct net *net; 6318 LIST_HEAD(list); 6319 6320 list_for_each_entry(net, net_list, exit_list) { 6321 bn = net_generic(net, bond_net_id); 6322 bond_destroy_sysfs(bn); 6323 } 6324 6325 /* Kill off any bonds created after unregistering bond rtnl ops */ 6326 rtnl_lock(); 6327 list_for_each_entry(net, net_list, exit_list) { 6328 struct bonding *bond, *tmp_bond; 6329 6330 bn = net_generic(net, bond_net_id); 6331 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6332 unregister_netdevice_queue(bond->dev, &list); 6333 } 6334 unregister_netdevice_many(&list); 6335 rtnl_unlock(); 6336 6337 list_for_each_entry(net, net_list, exit_list) { 6338 bn = net_generic(net, bond_net_id); 6339 bond_destroy_proc_dir(bn); 6340 } 6341 } 6342 6343 static struct pernet_operations bond_net_ops = { 6344 .init = bond_net_init, 6345 .exit_batch = bond_net_exit_batch, 6346 .id = &bond_net_id, 6347 .size = sizeof(struct bond_net), 6348 }; 6349 6350 static int __init bonding_init(void) 6351 { 6352 int i; 6353 int res; 6354 6355 res = bond_check_params(&bonding_defaults); 6356 if (res) 6357 goto out; 6358 6359 res = register_pernet_subsys(&bond_net_ops); 6360 if (res) 6361 goto out; 6362 6363 res = bond_netlink_init(); 6364 if (res) 6365 goto err_link; 6366 6367 bond_create_debugfs(); 6368 6369 for (i = 0; i < max_bonds; i++) { 6370 res = bond_create(&init_net, NULL); 6371 if (res) 6372 goto err; 6373 } 6374 6375 skb_flow_dissector_init(&flow_keys_bonding, 6376 flow_keys_bonding_keys, 6377 ARRAY_SIZE(flow_keys_bonding_keys)); 6378 6379 register_netdevice_notifier(&bond_netdev_notifier); 6380 out: 6381 return res; 6382 err: 6383 bond_destroy_debugfs(); 6384 bond_netlink_fini(); 6385 err_link: 6386 unregister_pernet_subsys(&bond_net_ops); 6387 goto out; 6388 6389 } 6390 6391 static void __exit bonding_exit(void) 6392 { 6393 unregister_netdevice_notifier(&bond_netdev_notifier); 6394 6395 bond_destroy_debugfs(); 6396 6397 bond_netlink_fini(); 6398 unregister_pernet_subsys(&bond_net_ops); 6399 6400 #ifdef CONFIG_NET_POLL_CONTROLLER 6401 /* Make sure we don't have an imbalance on our netpoll blocking */ 6402 WARN_ON(atomic_read(&netpoll_block_tx)); 6403 #endif 6404 } 6405 6406 module_init(bonding_init); 6407 module_exit(bonding_exit); 6408 MODULE_LICENSE("GPL"); 6409 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6410 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6411