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 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 /* del lacpdu mc addr from mc list */ 870 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 871 872 dev_mc_del(slave_dev, lacpdu_multicast); 873 } 874 } 875 876 /*--------------------------- Active slave change ---------------------------*/ 877 878 /* Update the hardware address list and promisc/allmulti for the new and 879 * old active slaves (if any). Modes that are not using primary keep all 880 * slaves up date at all times; only the modes that use primary need to call 881 * this function to swap these settings during a failover. 882 */ 883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 884 struct slave *old_active) 885 { 886 if (old_active) { 887 if (bond->dev->flags & IFF_PROMISC) 888 dev_set_promiscuity(old_active->dev, -1); 889 890 if (bond->dev->flags & IFF_ALLMULTI) 891 dev_set_allmulti(old_active->dev, -1); 892 893 bond_hw_addr_flush(bond->dev, old_active->dev); 894 } 895 896 if (new_active) { 897 /* FIXME: Signal errors upstream. */ 898 if (bond->dev->flags & IFF_PROMISC) 899 dev_set_promiscuity(new_active->dev, 1); 900 901 if (bond->dev->flags & IFF_ALLMULTI) 902 dev_set_allmulti(new_active->dev, 1); 903 904 netif_addr_lock_bh(bond->dev); 905 dev_uc_sync(new_active->dev, bond->dev); 906 dev_mc_sync(new_active->dev, bond->dev); 907 netif_addr_unlock_bh(bond->dev); 908 } 909 } 910 911 /** 912 * bond_set_dev_addr - clone slave's address to bond 913 * @bond_dev: bond net device 914 * @slave_dev: slave net device 915 * 916 * Should be called with RTNL held. 917 */ 918 static int bond_set_dev_addr(struct net_device *bond_dev, 919 struct net_device *slave_dev) 920 { 921 int err; 922 923 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 924 bond_dev, slave_dev, slave_dev->addr_len); 925 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 926 if (err) 927 return err; 928 929 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 930 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 931 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 932 return 0; 933 } 934 935 static struct slave *bond_get_old_active(struct bonding *bond, 936 struct slave *new_active) 937 { 938 struct slave *slave; 939 struct list_head *iter; 940 941 bond_for_each_slave(bond, slave, iter) { 942 if (slave == new_active) 943 continue; 944 945 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 946 return slave; 947 } 948 949 return NULL; 950 } 951 952 /* bond_do_fail_over_mac 953 * 954 * Perform special MAC address swapping for fail_over_mac settings 955 * 956 * Called with RTNL 957 */ 958 static void bond_do_fail_over_mac(struct bonding *bond, 959 struct slave *new_active, 960 struct slave *old_active) 961 { 962 u8 tmp_mac[MAX_ADDR_LEN]; 963 struct sockaddr_storage ss; 964 int rv; 965 966 switch (bond->params.fail_over_mac) { 967 case BOND_FOM_ACTIVE: 968 if (new_active) { 969 rv = bond_set_dev_addr(bond->dev, new_active->dev); 970 if (rv) 971 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 972 -rv); 973 } 974 break; 975 case BOND_FOM_FOLLOW: 976 /* if new_active && old_active, swap them 977 * if just old_active, do nothing (going to no active slave) 978 * if just new_active, set new_active to bond's MAC 979 */ 980 if (!new_active) 981 return; 982 983 if (!old_active) 984 old_active = bond_get_old_active(bond, new_active); 985 986 if (old_active) { 987 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 988 new_active->dev->addr_len); 989 bond_hw_addr_copy(ss.__data, 990 old_active->dev->dev_addr, 991 old_active->dev->addr_len); 992 ss.ss_family = new_active->dev->type; 993 } else { 994 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 995 bond->dev->addr_len); 996 ss.ss_family = bond->dev->type; 997 } 998 999 rv = dev_set_mac_address(new_active->dev, 1000 (struct sockaddr *)&ss, NULL); 1001 if (rv) { 1002 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1003 -rv); 1004 goto out; 1005 } 1006 1007 if (!old_active) 1008 goto out; 1009 1010 bond_hw_addr_copy(ss.__data, tmp_mac, 1011 new_active->dev->addr_len); 1012 ss.ss_family = old_active->dev->type; 1013 1014 rv = dev_set_mac_address(old_active->dev, 1015 (struct sockaddr *)&ss, NULL); 1016 if (rv) 1017 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1018 -rv); 1019 out: 1020 break; 1021 default: 1022 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1023 bond->params.fail_over_mac); 1024 break; 1025 } 1026 1027 } 1028 1029 /** 1030 * bond_choose_primary_or_current - select the primary or high priority slave 1031 * @bond: our bonding struct 1032 * 1033 * - Check if there is a primary link. If the primary link was set and is up, 1034 * go on and do link reselection. 1035 * 1036 * - If primary link is not set or down, find the highest priority link. 1037 * If the highest priority link is not current slave, set it as primary 1038 * link and do link reselection. 1039 */ 1040 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1041 { 1042 struct slave *prim = rtnl_dereference(bond->primary_slave); 1043 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1044 struct slave *slave, *hprio = NULL; 1045 struct list_head *iter; 1046 1047 if (!prim || prim->link != BOND_LINK_UP) { 1048 bond_for_each_slave(bond, slave, iter) { 1049 if (slave->link == BOND_LINK_UP) { 1050 hprio = hprio ?: slave; 1051 if (slave->prio > hprio->prio) 1052 hprio = slave; 1053 } 1054 } 1055 1056 if (hprio && hprio != curr) { 1057 prim = hprio; 1058 goto link_reselect; 1059 } 1060 1061 if (!curr || curr->link != BOND_LINK_UP) 1062 return NULL; 1063 return curr; 1064 } 1065 1066 if (bond->force_primary) { 1067 bond->force_primary = false; 1068 return prim; 1069 } 1070 1071 link_reselect: 1072 if (!curr || curr->link != BOND_LINK_UP) 1073 return prim; 1074 1075 /* At this point, prim and curr are both up */ 1076 switch (bond->params.primary_reselect) { 1077 case BOND_PRI_RESELECT_ALWAYS: 1078 return prim; 1079 case BOND_PRI_RESELECT_BETTER: 1080 if (prim->speed < curr->speed) 1081 return curr; 1082 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1083 return curr; 1084 return prim; 1085 case BOND_PRI_RESELECT_FAILURE: 1086 return curr; 1087 default: 1088 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1089 bond->params.primary_reselect); 1090 return curr; 1091 } 1092 } 1093 1094 /** 1095 * bond_find_best_slave - select the best available slave to be the active one 1096 * @bond: our bonding struct 1097 */ 1098 static struct slave *bond_find_best_slave(struct bonding *bond) 1099 { 1100 struct slave *slave, *bestslave = NULL; 1101 struct list_head *iter; 1102 int mintime = bond->params.updelay; 1103 1104 slave = bond_choose_primary_or_current(bond); 1105 if (slave) 1106 return slave; 1107 1108 bond_for_each_slave(bond, slave, iter) { 1109 if (slave->link == BOND_LINK_UP) 1110 return slave; 1111 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1112 slave->delay < mintime) { 1113 mintime = slave->delay; 1114 bestslave = slave; 1115 } 1116 } 1117 1118 return bestslave; 1119 } 1120 1121 static bool bond_should_notify_peers(struct bonding *bond) 1122 { 1123 struct slave *slave; 1124 1125 rcu_read_lock(); 1126 slave = rcu_dereference(bond->curr_active_slave); 1127 rcu_read_unlock(); 1128 1129 if (!slave || !bond->send_peer_notif || 1130 bond->send_peer_notif % 1131 max(1, bond->params.peer_notif_delay) != 0 || 1132 !netif_carrier_ok(bond->dev) || 1133 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1134 return false; 1135 1136 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1137 slave ? slave->dev->name : "NULL"); 1138 1139 return true; 1140 } 1141 1142 /** 1143 * bond_change_active_slave - change the active slave into the specified one 1144 * @bond: our bonding struct 1145 * @new_active: the new slave to make the active one 1146 * 1147 * Set the new slave to the bond's settings and unset them on the old 1148 * curr_active_slave. 1149 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1150 * 1151 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1152 * because it is apparently the best available slave we have, even though its 1153 * updelay hasn't timed out yet. 1154 * 1155 * Caller must hold RTNL. 1156 */ 1157 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1158 { 1159 struct slave *old_active; 1160 1161 ASSERT_RTNL(); 1162 1163 old_active = rtnl_dereference(bond->curr_active_slave); 1164 1165 if (old_active == new_active) 1166 return; 1167 1168 #ifdef CONFIG_XFRM_OFFLOAD 1169 bond_ipsec_del_sa_all(bond); 1170 #endif /* CONFIG_XFRM_OFFLOAD */ 1171 1172 if (new_active) { 1173 new_active->last_link_up = jiffies; 1174 1175 if (new_active->link == BOND_LINK_BACK) { 1176 if (bond_uses_primary(bond)) { 1177 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1178 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1179 } 1180 1181 new_active->delay = 0; 1182 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1183 BOND_SLAVE_NOTIFY_NOW); 1184 1185 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1186 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1187 1188 if (bond_is_lb(bond)) 1189 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1190 } else { 1191 if (bond_uses_primary(bond)) 1192 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1193 } 1194 } 1195 1196 if (bond_uses_primary(bond)) 1197 bond_hw_addr_swap(bond, new_active, old_active); 1198 1199 if (bond_is_lb(bond)) { 1200 bond_alb_handle_active_change(bond, new_active); 1201 if (old_active) 1202 bond_set_slave_inactive_flags(old_active, 1203 BOND_SLAVE_NOTIFY_NOW); 1204 if (new_active) 1205 bond_set_slave_active_flags(new_active, 1206 BOND_SLAVE_NOTIFY_NOW); 1207 } else { 1208 rcu_assign_pointer(bond->curr_active_slave, new_active); 1209 } 1210 1211 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1212 if (old_active) 1213 bond_set_slave_inactive_flags(old_active, 1214 BOND_SLAVE_NOTIFY_NOW); 1215 1216 if (new_active) { 1217 bool should_notify_peers = false; 1218 1219 bond_set_slave_active_flags(new_active, 1220 BOND_SLAVE_NOTIFY_NOW); 1221 1222 if (bond->params.fail_over_mac) 1223 bond_do_fail_over_mac(bond, new_active, 1224 old_active); 1225 1226 if (netif_running(bond->dev)) { 1227 bond->send_peer_notif = 1228 bond->params.num_peer_notif * 1229 max(1, bond->params.peer_notif_delay); 1230 should_notify_peers = 1231 bond_should_notify_peers(bond); 1232 } 1233 1234 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1235 if (should_notify_peers) { 1236 bond->send_peer_notif--; 1237 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1238 bond->dev); 1239 } 1240 } 1241 } 1242 1243 #ifdef CONFIG_XFRM_OFFLOAD 1244 bond_ipsec_add_sa_all(bond); 1245 #endif /* CONFIG_XFRM_OFFLOAD */ 1246 1247 /* resend IGMP joins since active slave has changed or 1248 * all were sent on curr_active_slave. 1249 * resend only if bond is brought up with the affected 1250 * bonding modes and the retransmission is enabled 1251 */ 1252 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1253 ((bond_uses_primary(bond) && new_active) || 1254 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1255 bond->igmp_retrans = bond->params.resend_igmp; 1256 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1257 } 1258 } 1259 1260 /** 1261 * bond_select_active_slave - select a new active slave, if needed 1262 * @bond: our bonding struct 1263 * 1264 * This functions should be called when one of the following occurs: 1265 * - The old curr_active_slave has been released or lost its link. 1266 * - The primary_slave has got its link back. 1267 * - A slave has got its link back and there's no old curr_active_slave. 1268 * 1269 * Caller must hold RTNL. 1270 */ 1271 void bond_select_active_slave(struct bonding *bond) 1272 { 1273 struct slave *best_slave; 1274 int rv; 1275 1276 ASSERT_RTNL(); 1277 1278 best_slave = bond_find_best_slave(bond); 1279 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1280 bond_change_active_slave(bond, best_slave); 1281 rv = bond_set_carrier(bond); 1282 if (!rv) 1283 return; 1284 1285 if (netif_carrier_ok(bond->dev)) 1286 netdev_info(bond->dev, "active interface up!\n"); 1287 else 1288 netdev_info(bond->dev, "now running without any active interface!\n"); 1289 } 1290 } 1291 1292 #ifdef CONFIG_NET_POLL_CONTROLLER 1293 static inline int slave_enable_netpoll(struct slave *slave) 1294 { 1295 struct netpoll *np; 1296 int err = 0; 1297 1298 np = kzalloc(sizeof(*np), GFP_KERNEL); 1299 err = -ENOMEM; 1300 if (!np) 1301 goto out; 1302 1303 err = __netpoll_setup(np, slave->dev); 1304 if (err) { 1305 kfree(np); 1306 goto out; 1307 } 1308 slave->np = np; 1309 out: 1310 return err; 1311 } 1312 static inline void slave_disable_netpoll(struct slave *slave) 1313 { 1314 struct netpoll *np = slave->np; 1315 1316 if (!np) 1317 return; 1318 1319 slave->np = NULL; 1320 1321 __netpoll_free(np); 1322 } 1323 1324 static void bond_poll_controller(struct net_device *bond_dev) 1325 { 1326 struct bonding *bond = netdev_priv(bond_dev); 1327 struct slave *slave = NULL; 1328 struct list_head *iter; 1329 struct ad_info ad_info; 1330 1331 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1332 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1333 return; 1334 1335 bond_for_each_slave_rcu(bond, slave, iter) { 1336 if (!bond_slave_is_up(slave)) 1337 continue; 1338 1339 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1340 struct aggregator *agg = 1341 SLAVE_AD_INFO(slave)->port.aggregator; 1342 1343 if (agg && 1344 agg->aggregator_identifier != ad_info.aggregator_id) 1345 continue; 1346 } 1347 1348 netpoll_poll_dev(slave->dev); 1349 } 1350 } 1351 1352 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1353 { 1354 struct bonding *bond = netdev_priv(bond_dev); 1355 struct list_head *iter; 1356 struct slave *slave; 1357 1358 bond_for_each_slave(bond, slave, iter) 1359 if (bond_slave_is_up(slave)) 1360 slave_disable_netpoll(slave); 1361 } 1362 1363 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1364 { 1365 struct bonding *bond = netdev_priv(dev); 1366 struct list_head *iter; 1367 struct slave *slave; 1368 int err = 0; 1369 1370 bond_for_each_slave(bond, slave, iter) { 1371 err = slave_enable_netpoll(slave); 1372 if (err) { 1373 bond_netpoll_cleanup(dev); 1374 break; 1375 } 1376 } 1377 return err; 1378 } 1379 #else 1380 static inline int slave_enable_netpoll(struct slave *slave) 1381 { 1382 return 0; 1383 } 1384 static inline void slave_disable_netpoll(struct slave *slave) 1385 { 1386 } 1387 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1388 { 1389 } 1390 #endif 1391 1392 /*---------------------------------- IOCTL ----------------------------------*/ 1393 1394 static netdev_features_t bond_fix_features(struct net_device *dev, 1395 netdev_features_t features) 1396 { 1397 struct bonding *bond = netdev_priv(dev); 1398 struct list_head *iter; 1399 netdev_features_t mask; 1400 struct slave *slave; 1401 1402 #if IS_ENABLED(CONFIG_TLS_DEVICE) 1403 if (bond_sk_check(bond)) 1404 features |= BOND_TLS_FEATURES; 1405 else 1406 features &= ~BOND_TLS_FEATURES; 1407 #endif 1408 1409 mask = features; 1410 1411 features &= ~NETIF_F_ONE_FOR_ALL; 1412 features |= NETIF_F_ALL_FOR_ALL; 1413 1414 bond_for_each_slave(bond, slave, iter) { 1415 features = netdev_increment_features(features, 1416 slave->dev->features, 1417 mask); 1418 } 1419 features = netdev_add_tso_features(features, mask); 1420 1421 return features; 1422 } 1423 1424 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1425 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1426 NETIF_F_HIGHDMA | NETIF_F_LRO) 1427 1428 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1429 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1430 1431 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1432 NETIF_F_GSO_SOFTWARE) 1433 1434 1435 static void bond_compute_features(struct bonding *bond) 1436 { 1437 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1438 IFF_XMIT_DST_RELEASE_PERM; 1439 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1440 netdev_features_t enc_features = BOND_ENC_FEATURES; 1441 #ifdef CONFIG_XFRM_OFFLOAD 1442 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1443 #endif /* CONFIG_XFRM_OFFLOAD */ 1444 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1445 struct net_device *bond_dev = bond->dev; 1446 struct list_head *iter; 1447 struct slave *slave; 1448 unsigned short max_hard_header_len = ETH_HLEN; 1449 unsigned int tso_max_size = TSO_MAX_SIZE; 1450 u16 tso_max_segs = TSO_MAX_SEGS; 1451 1452 if (!bond_has_slaves(bond)) 1453 goto done; 1454 vlan_features &= NETIF_F_ALL_FOR_ALL; 1455 mpls_features &= NETIF_F_ALL_FOR_ALL; 1456 1457 bond_for_each_slave(bond, slave, iter) { 1458 vlan_features = netdev_increment_features(vlan_features, 1459 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1460 1461 enc_features = netdev_increment_features(enc_features, 1462 slave->dev->hw_enc_features, 1463 BOND_ENC_FEATURES); 1464 1465 #ifdef CONFIG_XFRM_OFFLOAD 1466 xfrm_features = netdev_increment_features(xfrm_features, 1467 slave->dev->hw_enc_features, 1468 BOND_XFRM_FEATURES); 1469 #endif /* CONFIG_XFRM_OFFLOAD */ 1470 1471 mpls_features = netdev_increment_features(mpls_features, 1472 slave->dev->mpls_features, 1473 BOND_MPLS_FEATURES); 1474 1475 dst_release_flag &= slave->dev->priv_flags; 1476 if (slave->dev->hard_header_len > max_hard_header_len) 1477 max_hard_header_len = slave->dev->hard_header_len; 1478 1479 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1480 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1481 } 1482 bond_dev->hard_header_len = max_hard_header_len; 1483 1484 done: 1485 bond_dev->vlan_features = vlan_features; 1486 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1487 NETIF_F_HW_VLAN_CTAG_TX | 1488 NETIF_F_HW_VLAN_STAG_TX; 1489 #ifdef CONFIG_XFRM_OFFLOAD 1490 bond_dev->hw_enc_features |= xfrm_features; 1491 #endif /* CONFIG_XFRM_OFFLOAD */ 1492 bond_dev->mpls_features = mpls_features; 1493 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1494 netif_set_tso_max_size(bond_dev, tso_max_size); 1495 1496 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1497 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1498 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1499 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1500 1501 netdev_change_features(bond_dev); 1502 } 1503 1504 static void bond_setup_by_slave(struct net_device *bond_dev, 1505 struct net_device *slave_dev) 1506 { 1507 bond_dev->header_ops = slave_dev->header_ops; 1508 1509 bond_dev->type = slave_dev->type; 1510 bond_dev->hard_header_len = slave_dev->hard_header_len; 1511 bond_dev->needed_headroom = slave_dev->needed_headroom; 1512 bond_dev->addr_len = slave_dev->addr_len; 1513 1514 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1515 slave_dev->addr_len); 1516 } 1517 1518 /* On bonding slaves other than the currently active slave, suppress 1519 * duplicates except for alb non-mcast/bcast. 1520 */ 1521 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1522 struct slave *slave, 1523 struct bonding *bond) 1524 { 1525 if (bond_is_slave_inactive(slave)) { 1526 if (BOND_MODE(bond) == BOND_MODE_ALB && 1527 skb->pkt_type != PACKET_BROADCAST && 1528 skb->pkt_type != PACKET_MULTICAST) 1529 return false; 1530 return true; 1531 } 1532 return false; 1533 } 1534 1535 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1536 { 1537 struct sk_buff *skb = *pskb; 1538 struct slave *slave; 1539 struct bonding *bond; 1540 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1541 struct slave *); 1542 int ret = RX_HANDLER_ANOTHER; 1543 1544 skb = skb_share_check(skb, GFP_ATOMIC); 1545 if (unlikely(!skb)) 1546 return RX_HANDLER_CONSUMED; 1547 1548 *pskb = skb; 1549 1550 slave = bond_slave_get_rcu(skb->dev); 1551 bond = slave->bond; 1552 1553 recv_probe = READ_ONCE(bond->recv_probe); 1554 if (recv_probe) { 1555 ret = recv_probe(skb, bond, slave); 1556 if (ret == RX_HANDLER_CONSUMED) { 1557 consume_skb(skb); 1558 return ret; 1559 } 1560 } 1561 1562 /* 1563 * For packets determined by bond_should_deliver_exact_match() call to 1564 * be suppressed we want to make an exception for link-local packets. 1565 * This is necessary for e.g. LLDP daemons to be able to monitor 1566 * inactive slave links without being forced to bind to them 1567 * explicitly. 1568 * 1569 * At the same time, packets that are passed to the bonding master 1570 * (including link-local ones) can have their originating interface 1571 * determined via PACKET_ORIGDEV socket option. 1572 */ 1573 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1574 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1575 return RX_HANDLER_PASS; 1576 return RX_HANDLER_EXACT; 1577 } 1578 1579 skb->dev = bond->dev; 1580 1581 if (BOND_MODE(bond) == BOND_MODE_ALB && 1582 netif_is_bridge_port(bond->dev) && 1583 skb->pkt_type == PACKET_HOST) { 1584 1585 if (unlikely(skb_cow_head(skb, 1586 skb->data - skb_mac_header(skb)))) { 1587 kfree_skb(skb); 1588 return RX_HANDLER_CONSUMED; 1589 } 1590 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1591 bond->dev->addr_len); 1592 } 1593 1594 return ret; 1595 } 1596 1597 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1598 { 1599 switch (BOND_MODE(bond)) { 1600 case BOND_MODE_ROUNDROBIN: 1601 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1602 case BOND_MODE_ACTIVEBACKUP: 1603 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1604 case BOND_MODE_BROADCAST: 1605 return NETDEV_LAG_TX_TYPE_BROADCAST; 1606 case BOND_MODE_XOR: 1607 case BOND_MODE_8023AD: 1608 return NETDEV_LAG_TX_TYPE_HASH; 1609 default: 1610 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1611 } 1612 } 1613 1614 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1615 enum netdev_lag_tx_type type) 1616 { 1617 if (type != NETDEV_LAG_TX_TYPE_HASH) 1618 return NETDEV_LAG_HASH_NONE; 1619 1620 switch (bond->params.xmit_policy) { 1621 case BOND_XMIT_POLICY_LAYER2: 1622 return NETDEV_LAG_HASH_L2; 1623 case BOND_XMIT_POLICY_LAYER34: 1624 return NETDEV_LAG_HASH_L34; 1625 case BOND_XMIT_POLICY_LAYER23: 1626 return NETDEV_LAG_HASH_L23; 1627 case BOND_XMIT_POLICY_ENCAP23: 1628 return NETDEV_LAG_HASH_E23; 1629 case BOND_XMIT_POLICY_ENCAP34: 1630 return NETDEV_LAG_HASH_E34; 1631 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1632 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1633 default: 1634 return NETDEV_LAG_HASH_UNKNOWN; 1635 } 1636 } 1637 1638 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1639 struct netlink_ext_ack *extack) 1640 { 1641 struct netdev_lag_upper_info lag_upper_info; 1642 enum netdev_lag_tx_type type; 1643 1644 type = bond_lag_tx_type(bond); 1645 lag_upper_info.tx_type = type; 1646 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1647 1648 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1649 &lag_upper_info, extack); 1650 } 1651 1652 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1653 { 1654 netdev_upper_dev_unlink(slave->dev, bond->dev); 1655 slave->dev->flags &= ~IFF_SLAVE; 1656 } 1657 1658 static void slave_kobj_release(struct kobject *kobj) 1659 { 1660 struct slave *slave = to_slave(kobj); 1661 struct bonding *bond = bond_get_bond_by_slave(slave); 1662 1663 cancel_delayed_work_sync(&slave->notify_work); 1664 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1665 kfree(SLAVE_AD_INFO(slave)); 1666 1667 kfree(slave); 1668 } 1669 1670 static struct kobj_type slave_ktype = { 1671 .release = slave_kobj_release, 1672 #ifdef CONFIG_SYSFS 1673 .sysfs_ops = &slave_sysfs_ops, 1674 #endif 1675 }; 1676 1677 static int bond_kobj_init(struct slave *slave) 1678 { 1679 int err; 1680 1681 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1682 &(slave->dev->dev.kobj), "bonding_slave"); 1683 if (err) 1684 kobject_put(&slave->kobj); 1685 1686 return err; 1687 } 1688 1689 static struct slave *bond_alloc_slave(struct bonding *bond, 1690 struct net_device *slave_dev) 1691 { 1692 struct slave *slave = NULL; 1693 1694 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1695 if (!slave) 1696 return NULL; 1697 1698 slave->bond = bond; 1699 slave->dev = slave_dev; 1700 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1701 1702 if (bond_kobj_init(slave)) 1703 return NULL; 1704 1705 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1706 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1707 GFP_KERNEL); 1708 if (!SLAVE_AD_INFO(slave)) { 1709 kobject_put(&slave->kobj); 1710 return NULL; 1711 } 1712 } 1713 1714 return slave; 1715 } 1716 1717 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1718 { 1719 info->bond_mode = BOND_MODE(bond); 1720 info->miimon = bond->params.miimon; 1721 info->num_slaves = bond->slave_cnt; 1722 } 1723 1724 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1725 { 1726 strcpy(info->slave_name, slave->dev->name); 1727 info->link = slave->link; 1728 info->state = bond_slave_state(slave); 1729 info->link_failure_count = slave->link_failure_count; 1730 } 1731 1732 static void bond_netdev_notify_work(struct work_struct *_work) 1733 { 1734 struct slave *slave = container_of(_work, struct slave, 1735 notify_work.work); 1736 1737 if (rtnl_trylock()) { 1738 struct netdev_bonding_info binfo; 1739 1740 bond_fill_ifslave(slave, &binfo.slave); 1741 bond_fill_ifbond(slave->bond, &binfo.master); 1742 netdev_bonding_info_change(slave->dev, &binfo); 1743 rtnl_unlock(); 1744 } else { 1745 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1746 } 1747 } 1748 1749 void bond_queue_slave_event(struct slave *slave) 1750 { 1751 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1752 } 1753 1754 void bond_lower_state_changed(struct slave *slave) 1755 { 1756 struct netdev_lag_lower_state_info info; 1757 1758 info.link_up = slave->link == BOND_LINK_UP || 1759 slave->link == BOND_LINK_FAIL; 1760 info.tx_enabled = bond_is_active_slave(slave); 1761 netdev_lower_state_changed(slave->dev, &info); 1762 } 1763 1764 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1765 if (extack) \ 1766 NL_SET_ERR_MSG(extack, errmsg); \ 1767 else \ 1768 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1769 } while (0) 1770 1771 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1772 if (extack) \ 1773 NL_SET_ERR_MSG(extack, errmsg); \ 1774 else \ 1775 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1776 } while (0) 1777 1778 /* enslave device <slave> to bond device <master> */ 1779 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1780 struct netlink_ext_ack *extack) 1781 { 1782 struct bonding *bond = netdev_priv(bond_dev); 1783 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1784 struct slave *new_slave = NULL, *prev_slave; 1785 struct sockaddr_storage ss; 1786 int link_reporting; 1787 int res = 0, i; 1788 1789 if (slave_dev->flags & IFF_MASTER && 1790 !netif_is_bond_master(slave_dev)) { 1791 BOND_NL_ERR(bond_dev, extack, 1792 "Device type (master device) cannot be enslaved"); 1793 return -EPERM; 1794 } 1795 1796 if (!bond->params.use_carrier && 1797 slave_dev->ethtool_ops->get_link == NULL && 1798 slave_ops->ndo_eth_ioctl == NULL) { 1799 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1800 } 1801 1802 /* already in-use? */ 1803 if (netdev_is_rx_handler_busy(slave_dev)) { 1804 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1805 "Device is in use and cannot be enslaved"); 1806 return -EBUSY; 1807 } 1808 1809 if (bond_dev == slave_dev) { 1810 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1811 return -EPERM; 1812 } 1813 1814 /* vlan challenged mutual exclusion */ 1815 /* no need to lock since we're protected by rtnl_lock */ 1816 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1817 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1818 if (vlan_uses_dev(bond_dev)) { 1819 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1820 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1821 return -EPERM; 1822 } else { 1823 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1824 } 1825 } else { 1826 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1827 } 1828 1829 if (slave_dev->features & NETIF_F_HW_ESP) 1830 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1831 1832 /* Old ifenslave binaries are no longer supported. These can 1833 * be identified with moderate accuracy by the state of the slave: 1834 * the current ifenslave will set the interface down prior to 1835 * enslaving it; the old ifenslave will not. 1836 */ 1837 if (slave_dev->flags & IFF_UP) { 1838 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1839 "Device can not be enslaved while up"); 1840 return -EPERM; 1841 } 1842 1843 /* set bonding device ether type by slave - bonding netdevices are 1844 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1845 * there is a need to override some of the type dependent attribs/funcs. 1846 * 1847 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1848 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1849 */ 1850 if (!bond_has_slaves(bond)) { 1851 if (bond_dev->type != slave_dev->type) { 1852 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1853 bond_dev->type, slave_dev->type); 1854 1855 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1856 bond_dev); 1857 res = notifier_to_errno(res); 1858 if (res) { 1859 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1860 return -EBUSY; 1861 } 1862 1863 /* Flush unicast and multicast addresses */ 1864 dev_uc_flush(bond_dev); 1865 dev_mc_flush(bond_dev); 1866 1867 if (slave_dev->type != ARPHRD_ETHER) 1868 bond_setup_by_slave(bond_dev, slave_dev); 1869 else { 1870 ether_setup(bond_dev); 1871 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1872 } 1873 1874 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1875 bond_dev); 1876 } 1877 } else if (bond_dev->type != slave_dev->type) { 1878 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1879 "Device type is different from other slaves"); 1880 return -EINVAL; 1881 } 1882 1883 if (slave_dev->type == ARPHRD_INFINIBAND && 1884 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1885 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1886 "Only active-backup mode is supported for infiniband slaves"); 1887 res = -EOPNOTSUPP; 1888 goto err_undo_flags; 1889 } 1890 1891 if (!slave_ops->ndo_set_mac_address || 1892 slave_dev->type == ARPHRD_INFINIBAND) { 1893 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1894 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1895 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1896 if (!bond_has_slaves(bond)) { 1897 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1898 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1899 } else { 1900 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1901 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1902 res = -EOPNOTSUPP; 1903 goto err_undo_flags; 1904 } 1905 } 1906 } 1907 1908 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1909 1910 /* If this is the first slave, then we need to set the master's hardware 1911 * address to be the same as the slave's. 1912 */ 1913 if (!bond_has_slaves(bond) && 1914 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1915 res = bond_set_dev_addr(bond->dev, slave_dev); 1916 if (res) 1917 goto err_undo_flags; 1918 } 1919 1920 new_slave = bond_alloc_slave(bond, slave_dev); 1921 if (!new_slave) { 1922 res = -ENOMEM; 1923 goto err_undo_flags; 1924 } 1925 1926 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1927 * is set via sysfs or module option if desired. 1928 */ 1929 new_slave->queue_id = 0; 1930 1931 /* Save slave's original mtu and then set it to match the bond */ 1932 new_slave->original_mtu = slave_dev->mtu; 1933 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1934 if (res) { 1935 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1936 goto err_free; 1937 } 1938 1939 /* Save slave's original ("permanent") mac address for modes 1940 * that need it, and for restoring it upon release, and then 1941 * set it to the master's address 1942 */ 1943 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1944 slave_dev->addr_len); 1945 1946 if (!bond->params.fail_over_mac || 1947 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1948 /* Set slave to master's mac address. The application already 1949 * set the master's mac address to that of the first slave 1950 */ 1951 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1952 ss.ss_family = slave_dev->type; 1953 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1954 extack); 1955 if (res) { 1956 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1957 goto err_restore_mtu; 1958 } 1959 } 1960 1961 /* set slave flag before open to prevent IPv6 addrconf */ 1962 slave_dev->flags |= IFF_SLAVE; 1963 1964 /* open the slave since the application closed it */ 1965 res = dev_open(slave_dev, extack); 1966 if (res) { 1967 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1968 goto err_restore_mac; 1969 } 1970 1971 slave_dev->priv_flags |= IFF_BONDING; 1972 /* initialize slave stats */ 1973 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1974 1975 if (bond_is_lb(bond)) { 1976 /* bond_alb_init_slave() must be called before all other stages since 1977 * it might fail and we do not want to have to undo everything 1978 */ 1979 res = bond_alb_init_slave(bond, new_slave); 1980 if (res) 1981 goto err_close; 1982 } 1983 1984 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1985 if (res) { 1986 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1987 goto err_close; 1988 } 1989 1990 prev_slave = bond_last_slave(bond); 1991 1992 new_slave->delay = 0; 1993 new_slave->link_failure_count = 0; 1994 1995 if (bond_update_speed_duplex(new_slave) && 1996 bond_needs_speed_duplex(bond)) 1997 new_slave->link = BOND_LINK_DOWN; 1998 1999 new_slave->last_rx = jiffies - 2000 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2001 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2002 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2003 2004 if (bond->params.miimon && !bond->params.use_carrier) { 2005 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2006 2007 if ((link_reporting == -1) && !bond->params.arp_interval) { 2008 /* miimon is set but a bonded network driver 2009 * does not support ETHTOOL/MII and 2010 * arp_interval is not set. Note: if 2011 * use_carrier is enabled, we will never go 2012 * here (because netif_carrier is always 2013 * supported); thus, we don't need to change 2014 * the messages for netif_carrier. 2015 */ 2016 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"); 2017 } else if (link_reporting == -1) { 2018 /* unable get link status using mii/ethtool */ 2019 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"); 2020 } 2021 } 2022 2023 /* check for initial state */ 2024 new_slave->link = BOND_LINK_NOCHANGE; 2025 if (bond->params.miimon) { 2026 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2027 if (bond->params.updelay) { 2028 bond_set_slave_link_state(new_slave, 2029 BOND_LINK_BACK, 2030 BOND_SLAVE_NOTIFY_NOW); 2031 new_slave->delay = bond->params.updelay; 2032 } else { 2033 bond_set_slave_link_state(new_slave, 2034 BOND_LINK_UP, 2035 BOND_SLAVE_NOTIFY_NOW); 2036 } 2037 } else { 2038 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2039 BOND_SLAVE_NOTIFY_NOW); 2040 } 2041 } else if (bond->params.arp_interval) { 2042 bond_set_slave_link_state(new_slave, 2043 (netif_carrier_ok(slave_dev) ? 2044 BOND_LINK_UP : BOND_LINK_DOWN), 2045 BOND_SLAVE_NOTIFY_NOW); 2046 } else { 2047 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2048 BOND_SLAVE_NOTIFY_NOW); 2049 } 2050 2051 if (new_slave->link != BOND_LINK_DOWN) 2052 new_slave->last_link_up = jiffies; 2053 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2054 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2055 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2056 2057 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2058 /* if there is a primary slave, remember it */ 2059 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2060 rcu_assign_pointer(bond->primary_slave, new_slave); 2061 bond->force_primary = true; 2062 } 2063 } 2064 2065 switch (BOND_MODE(bond)) { 2066 case BOND_MODE_ACTIVEBACKUP: 2067 bond_set_slave_inactive_flags(new_slave, 2068 BOND_SLAVE_NOTIFY_NOW); 2069 break; 2070 case BOND_MODE_8023AD: 2071 /* in 802.3ad mode, the internal mechanism 2072 * will activate the slaves in the selected 2073 * aggregator 2074 */ 2075 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2076 /* if this is the first slave */ 2077 if (!prev_slave) { 2078 SLAVE_AD_INFO(new_slave)->id = 1; 2079 /* Initialize AD with the number of times that the AD timer is called in 1 second 2080 * can be called only after the mac address of the bond is set 2081 */ 2082 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL); 2083 } else { 2084 SLAVE_AD_INFO(new_slave)->id = 2085 SLAVE_AD_INFO(prev_slave)->id + 1; 2086 } 2087 2088 bond_3ad_bind_slave(new_slave); 2089 break; 2090 case BOND_MODE_TLB: 2091 case BOND_MODE_ALB: 2092 bond_set_active_slave(new_slave); 2093 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2094 break; 2095 default: 2096 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2097 2098 /* always active in trunk mode */ 2099 bond_set_active_slave(new_slave); 2100 2101 /* In trunking mode there is little meaning to curr_active_slave 2102 * anyway (it holds no special properties of the bond device), 2103 * so we can change it without calling change_active_interface() 2104 */ 2105 if (!rcu_access_pointer(bond->curr_active_slave) && 2106 new_slave->link == BOND_LINK_UP) 2107 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2108 2109 break; 2110 } /* switch(bond_mode) */ 2111 2112 #ifdef CONFIG_NET_POLL_CONTROLLER 2113 if (bond->dev->npinfo) { 2114 if (slave_enable_netpoll(new_slave)) { 2115 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2116 res = -EBUSY; 2117 goto err_detach; 2118 } 2119 } 2120 #endif 2121 2122 if (!(bond_dev->features & NETIF_F_LRO)) 2123 dev_disable_lro(slave_dev); 2124 2125 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2126 new_slave); 2127 if (res) { 2128 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2129 goto err_detach; 2130 } 2131 2132 res = bond_master_upper_dev_link(bond, new_slave, extack); 2133 if (res) { 2134 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2135 goto err_unregister; 2136 } 2137 2138 bond_lower_state_changed(new_slave); 2139 2140 res = bond_sysfs_slave_add(new_slave); 2141 if (res) { 2142 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2143 goto err_upper_unlink; 2144 } 2145 2146 /* If the mode uses primary, then the following is handled by 2147 * bond_change_active_slave(). 2148 */ 2149 if (!bond_uses_primary(bond)) { 2150 /* set promiscuity level to new slave */ 2151 if (bond_dev->flags & IFF_PROMISC) { 2152 res = dev_set_promiscuity(slave_dev, 1); 2153 if (res) 2154 goto err_sysfs_del; 2155 } 2156 2157 /* set allmulti level to new slave */ 2158 if (bond_dev->flags & IFF_ALLMULTI) { 2159 res = dev_set_allmulti(slave_dev, 1); 2160 if (res) { 2161 if (bond_dev->flags & IFF_PROMISC) 2162 dev_set_promiscuity(slave_dev, -1); 2163 goto err_sysfs_del; 2164 } 2165 } 2166 2167 netif_addr_lock_bh(bond_dev); 2168 dev_mc_sync_multiple(slave_dev, bond_dev); 2169 dev_uc_sync_multiple(slave_dev, bond_dev); 2170 netif_addr_unlock_bh(bond_dev); 2171 2172 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2173 /* add lacpdu mc addr to mc list */ 2174 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR; 2175 2176 dev_mc_add(slave_dev, lacpdu_multicast); 2177 } 2178 } 2179 2180 bond->slave_cnt++; 2181 bond_compute_features(bond); 2182 bond_set_carrier(bond); 2183 2184 if (bond_uses_primary(bond)) { 2185 block_netpoll_tx(); 2186 bond_select_active_slave(bond); 2187 unblock_netpoll_tx(); 2188 } 2189 2190 if (bond_mode_can_use_xmit_hash(bond)) 2191 bond_update_slave_arr(bond, NULL); 2192 2193 2194 if (!slave_dev->netdev_ops->ndo_bpf || 2195 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2196 if (bond->xdp_prog) { 2197 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2198 "Slave does not support XDP"); 2199 res = -EOPNOTSUPP; 2200 goto err_sysfs_del; 2201 } 2202 } else if (bond->xdp_prog) { 2203 struct netdev_bpf xdp = { 2204 .command = XDP_SETUP_PROG, 2205 .flags = 0, 2206 .prog = bond->xdp_prog, 2207 .extack = extack, 2208 }; 2209 2210 if (dev_xdp_prog_count(slave_dev) > 0) { 2211 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2212 "Slave has XDP program loaded, please unload before enslaving"); 2213 res = -EOPNOTSUPP; 2214 goto err_sysfs_del; 2215 } 2216 2217 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 2218 if (res < 0) { 2219 /* ndo_bpf() sets extack error message */ 2220 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2221 goto err_sysfs_del; 2222 } 2223 if (bond->xdp_prog) 2224 bpf_prog_inc(bond->xdp_prog); 2225 } 2226 2227 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2228 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2229 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2230 2231 /* enslave is successful */ 2232 bond_queue_slave_event(new_slave); 2233 return 0; 2234 2235 /* Undo stages on error */ 2236 err_sysfs_del: 2237 bond_sysfs_slave_del(new_slave); 2238 2239 err_upper_unlink: 2240 bond_upper_dev_unlink(bond, new_slave); 2241 2242 err_unregister: 2243 netdev_rx_handler_unregister(slave_dev); 2244 2245 err_detach: 2246 vlan_vids_del_by_dev(slave_dev, bond_dev); 2247 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2248 RCU_INIT_POINTER(bond->primary_slave, NULL); 2249 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2250 block_netpoll_tx(); 2251 bond_change_active_slave(bond, NULL); 2252 bond_select_active_slave(bond); 2253 unblock_netpoll_tx(); 2254 } 2255 /* either primary_slave or curr_active_slave might've changed */ 2256 synchronize_rcu(); 2257 slave_disable_netpoll(new_slave); 2258 2259 err_close: 2260 if (!netif_is_bond_master(slave_dev)) 2261 slave_dev->priv_flags &= ~IFF_BONDING; 2262 dev_close(slave_dev); 2263 2264 err_restore_mac: 2265 slave_dev->flags &= ~IFF_SLAVE; 2266 if (!bond->params.fail_over_mac || 2267 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2268 /* XXX TODO - fom follow mode needs to change master's 2269 * MAC if this slave's MAC is in use by the bond, or at 2270 * least print a warning. 2271 */ 2272 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2273 new_slave->dev->addr_len); 2274 ss.ss_family = slave_dev->type; 2275 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2276 } 2277 2278 err_restore_mtu: 2279 dev_set_mtu(slave_dev, new_slave->original_mtu); 2280 2281 err_free: 2282 kobject_put(&new_slave->kobj); 2283 2284 err_undo_flags: 2285 /* Enslave of first slave has failed and we need to fix master's mac */ 2286 if (!bond_has_slaves(bond)) { 2287 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2288 slave_dev->dev_addr)) 2289 eth_hw_addr_random(bond_dev); 2290 if (bond_dev->type != ARPHRD_ETHER) { 2291 dev_close(bond_dev); 2292 ether_setup(bond_dev); 2293 bond_dev->flags |= IFF_MASTER; 2294 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2295 } 2296 } 2297 2298 return res; 2299 } 2300 2301 /* Try to release the slave device <slave> from the bond device <master> 2302 * It is legal to access curr_active_slave without a lock because all the function 2303 * is RTNL-locked. If "all" is true it means that the function is being called 2304 * while destroying a bond interface and all slaves are being released. 2305 * 2306 * The rules for slave state should be: 2307 * for Active/Backup: 2308 * Active stays on all backups go down 2309 * for Bonded connections: 2310 * The first up interface should be left on and all others downed. 2311 */ 2312 static int __bond_release_one(struct net_device *bond_dev, 2313 struct net_device *slave_dev, 2314 bool all, bool unregister) 2315 { 2316 struct bonding *bond = netdev_priv(bond_dev); 2317 struct slave *slave, *oldcurrent; 2318 struct sockaddr_storage ss; 2319 int old_flags = bond_dev->flags; 2320 netdev_features_t old_features = bond_dev->features; 2321 2322 /* slave is not a slave or master is not master of this slave */ 2323 if (!(slave_dev->flags & IFF_SLAVE) || 2324 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2325 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2326 return -EINVAL; 2327 } 2328 2329 block_netpoll_tx(); 2330 2331 slave = bond_get_slave_by_dev(bond, slave_dev); 2332 if (!slave) { 2333 /* not a slave of this bond */ 2334 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2335 unblock_netpoll_tx(); 2336 return -EINVAL; 2337 } 2338 2339 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2340 2341 bond_sysfs_slave_del(slave); 2342 2343 /* recompute stats just before removing the slave */ 2344 bond_get_stats(bond->dev, &bond->bond_stats); 2345 2346 if (bond->xdp_prog) { 2347 struct netdev_bpf xdp = { 2348 .command = XDP_SETUP_PROG, 2349 .flags = 0, 2350 .prog = NULL, 2351 .extack = NULL, 2352 }; 2353 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp)) 2354 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2355 } 2356 2357 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2358 * for this slave anymore. 2359 */ 2360 netdev_rx_handler_unregister(slave_dev); 2361 2362 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2363 bond_3ad_unbind_slave(slave); 2364 2365 bond_upper_dev_unlink(bond, slave); 2366 2367 if (bond_mode_can_use_xmit_hash(bond)) 2368 bond_update_slave_arr(bond, slave); 2369 2370 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2371 bond_is_active_slave(slave) ? "active" : "backup"); 2372 2373 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2374 2375 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2376 2377 if (!all && (!bond->params.fail_over_mac || 2378 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2379 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2380 bond_has_slaves(bond)) 2381 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", 2382 slave->perm_hwaddr); 2383 } 2384 2385 if (rtnl_dereference(bond->primary_slave) == slave) 2386 RCU_INIT_POINTER(bond->primary_slave, NULL); 2387 2388 if (oldcurrent == slave) 2389 bond_change_active_slave(bond, NULL); 2390 2391 if (bond_is_lb(bond)) { 2392 /* Must be called only after the slave has been 2393 * detached from the list and the curr_active_slave 2394 * has been cleared (if our_slave == old_current), 2395 * but before a new active slave is selected. 2396 */ 2397 bond_alb_deinit_slave(bond, slave); 2398 } 2399 2400 if (all) { 2401 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2402 } else if (oldcurrent == slave) { 2403 /* Note that we hold RTNL over this sequence, so there 2404 * is no concern that another slave add/remove event 2405 * will interfere. 2406 */ 2407 bond_select_active_slave(bond); 2408 } 2409 2410 bond_set_carrier(bond); 2411 if (!bond_has_slaves(bond)) 2412 eth_hw_addr_random(bond_dev); 2413 2414 unblock_netpoll_tx(); 2415 synchronize_rcu(); 2416 bond->slave_cnt--; 2417 2418 if (!bond_has_slaves(bond)) { 2419 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2420 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2421 } 2422 2423 bond_compute_features(bond); 2424 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2425 (old_features & NETIF_F_VLAN_CHALLENGED)) 2426 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2427 2428 vlan_vids_del_by_dev(slave_dev, bond_dev); 2429 2430 /* If the mode uses primary, then this case was handled above by 2431 * bond_change_active_slave(..., NULL) 2432 */ 2433 if (!bond_uses_primary(bond)) { 2434 /* unset promiscuity level from slave 2435 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2436 * of the IFF_PROMISC flag in the bond_dev, but we need the 2437 * value of that flag before that change, as that was the value 2438 * when this slave was attached, so we cache at the start of the 2439 * function and use it here. Same goes for ALLMULTI below 2440 */ 2441 if (old_flags & IFF_PROMISC) 2442 dev_set_promiscuity(slave_dev, -1); 2443 2444 /* unset allmulti level from slave */ 2445 if (old_flags & IFF_ALLMULTI) 2446 dev_set_allmulti(slave_dev, -1); 2447 2448 bond_hw_addr_flush(bond_dev, slave_dev); 2449 } 2450 2451 slave_disable_netpoll(slave); 2452 2453 /* close slave before restoring its mac address */ 2454 dev_close(slave_dev); 2455 2456 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2457 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2458 /* restore original ("permanent") mac address */ 2459 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2460 slave->dev->addr_len); 2461 ss.ss_family = slave_dev->type; 2462 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2463 } 2464 2465 if (unregister) 2466 __dev_set_mtu(slave_dev, slave->original_mtu); 2467 else 2468 dev_set_mtu(slave_dev, slave->original_mtu); 2469 2470 if (!netif_is_bond_master(slave_dev)) 2471 slave_dev->priv_flags &= ~IFF_BONDING; 2472 2473 kobject_put(&slave->kobj); 2474 2475 return 0; 2476 } 2477 2478 /* A wrapper used because of ndo_del_link */ 2479 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2480 { 2481 return __bond_release_one(bond_dev, slave_dev, false, false); 2482 } 2483 2484 /* First release a slave and then destroy the bond if no more slaves are left. 2485 * Must be under rtnl_lock when this function is called. 2486 */ 2487 static int bond_release_and_destroy(struct net_device *bond_dev, 2488 struct net_device *slave_dev) 2489 { 2490 struct bonding *bond = netdev_priv(bond_dev); 2491 int ret; 2492 2493 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2494 if (ret == 0 && !bond_has_slaves(bond) && 2495 bond_dev->reg_state != NETREG_UNREGISTERING) { 2496 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2497 netdev_info(bond_dev, "Destroying bond\n"); 2498 bond_remove_proc_entry(bond); 2499 unregister_netdevice(bond_dev); 2500 } 2501 return ret; 2502 } 2503 2504 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2505 { 2506 struct bonding *bond = netdev_priv(bond_dev); 2507 2508 bond_fill_ifbond(bond, info); 2509 } 2510 2511 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2512 { 2513 struct bonding *bond = netdev_priv(bond_dev); 2514 struct list_head *iter; 2515 int i = 0, res = -ENODEV; 2516 struct slave *slave; 2517 2518 bond_for_each_slave(bond, slave, iter) { 2519 if (i++ == (int)info->slave_id) { 2520 res = 0; 2521 bond_fill_ifslave(slave, info); 2522 break; 2523 } 2524 } 2525 2526 return res; 2527 } 2528 2529 /*-------------------------------- Monitoring -------------------------------*/ 2530 2531 /* called with rcu_read_lock() */ 2532 static int bond_miimon_inspect(struct bonding *bond) 2533 { 2534 int link_state, commit = 0; 2535 struct list_head *iter; 2536 struct slave *slave; 2537 bool ignore_updelay; 2538 2539 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2540 2541 bond_for_each_slave_rcu(bond, slave, iter) { 2542 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2543 2544 link_state = bond_check_dev_link(bond, slave->dev, 0); 2545 2546 switch (slave->link) { 2547 case BOND_LINK_UP: 2548 if (link_state) 2549 continue; 2550 2551 bond_propose_link_state(slave, BOND_LINK_FAIL); 2552 commit++; 2553 slave->delay = bond->params.downdelay; 2554 if (slave->delay) { 2555 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2556 (BOND_MODE(bond) == 2557 BOND_MODE_ACTIVEBACKUP) ? 2558 (bond_is_active_slave(slave) ? 2559 "active " : "backup ") : "", 2560 bond->params.downdelay * bond->params.miimon); 2561 } 2562 fallthrough; 2563 case BOND_LINK_FAIL: 2564 if (link_state) { 2565 /* recovered before downdelay expired */ 2566 bond_propose_link_state(slave, BOND_LINK_UP); 2567 slave->last_link_up = jiffies; 2568 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2569 (bond->params.downdelay - slave->delay) * 2570 bond->params.miimon); 2571 commit++; 2572 continue; 2573 } 2574 2575 if (slave->delay <= 0) { 2576 bond_propose_link_state(slave, BOND_LINK_DOWN); 2577 commit++; 2578 continue; 2579 } 2580 2581 slave->delay--; 2582 break; 2583 2584 case BOND_LINK_DOWN: 2585 if (!link_state) 2586 continue; 2587 2588 bond_propose_link_state(slave, BOND_LINK_BACK); 2589 commit++; 2590 slave->delay = bond->params.updelay; 2591 2592 if (slave->delay) { 2593 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2594 ignore_updelay ? 0 : 2595 bond->params.updelay * 2596 bond->params.miimon); 2597 } 2598 fallthrough; 2599 case BOND_LINK_BACK: 2600 if (!link_state) { 2601 bond_propose_link_state(slave, BOND_LINK_DOWN); 2602 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2603 (bond->params.updelay - slave->delay) * 2604 bond->params.miimon); 2605 commit++; 2606 continue; 2607 } 2608 2609 if (ignore_updelay) 2610 slave->delay = 0; 2611 2612 if (slave->delay <= 0) { 2613 bond_propose_link_state(slave, BOND_LINK_UP); 2614 commit++; 2615 ignore_updelay = false; 2616 continue; 2617 } 2618 2619 slave->delay--; 2620 break; 2621 } 2622 } 2623 2624 return commit; 2625 } 2626 2627 static void bond_miimon_link_change(struct bonding *bond, 2628 struct slave *slave, 2629 char link) 2630 { 2631 switch (BOND_MODE(bond)) { 2632 case BOND_MODE_8023AD: 2633 bond_3ad_handle_link_change(slave, link); 2634 break; 2635 case BOND_MODE_TLB: 2636 case BOND_MODE_ALB: 2637 bond_alb_handle_link_change(bond, slave, link); 2638 break; 2639 case BOND_MODE_XOR: 2640 bond_update_slave_arr(bond, NULL); 2641 break; 2642 } 2643 } 2644 2645 static void bond_miimon_commit(struct bonding *bond) 2646 { 2647 struct list_head *iter; 2648 struct slave *slave, *primary; 2649 2650 bond_for_each_slave(bond, slave, iter) { 2651 switch (slave->link_new_state) { 2652 case BOND_LINK_NOCHANGE: 2653 /* For 802.3ad mode, check current slave speed and 2654 * duplex again in case its port was disabled after 2655 * invalid speed/duplex reporting but recovered before 2656 * link monitoring could make a decision on the actual 2657 * link status 2658 */ 2659 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2660 slave->link == BOND_LINK_UP) 2661 bond_3ad_adapter_speed_duplex_changed(slave); 2662 continue; 2663 2664 case BOND_LINK_UP: 2665 if (bond_update_speed_duplex(slave) && 2666 bond_needs_speed_duplex(bond)) { 2667 slave->link = BOND_LINK_DOWN; 2668 if (net_ratelimit()) 2669 slave_warn(bond->dev, slave->dev, 2670 "failed to get link speed/duplex\n"); 2671 continue; 2672 } 2673 bond_set_slave_link_state(slave, BOND_LINK_UP, 2674 BOND_SLAVE_NOTIFY_NOW); 2675 slave->last_link_up = jiffies; 2676 2677 primary = rtnl_dereference(bond->primary_slave); 2678 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2679 /* prevent it from being the active one */ 2680 bond_set_backup_slave(slave); 2681 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2682 /* make it immediately active */ 2683 bond_set_active_slave(slave); 2684 } 2685 2686 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2687 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2688 slave->duplex ? "full" : "half"); 2689 2690 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2691 2692 if (!bond->curr_active_slave || slave == primary) 2693 goto do_failover; 2694 2695 continue; 2696 2697 case BOND_LINK_DOWN: 2698 if (slave->link_failure_count < UINT_MAX) 2699 slave->link_failure_count++; 2700 2701 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2702 BOND_SLAVE_NOTIFY_NOW); 2703 2704 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2705 BOND_MODE(bond) == BOND_MODE_8023AD) 2706 bond_set_slave_inactive_flags(slave, 2707 BOND_SLAVE_NOTIFY_NOW); 2708 2709 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2710 2711 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2712 2713 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2714 goto do_failover; 2715 2716 continue; 2717 2718 default: 2719 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2720 slave->link_new_state); 2721 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2722 2723 continue; 2724 } 2725 2726 do_failover: 2727 block_netpoll_tx(); 2728 bond_select_active_slave(bond); 2729 unblock_netpoll_tx(); 2730 } 2731 2732 bond_set_carrier(bond); 2733 } 2734 2735 /* bond_mii_monitor 2736 * 2737 * Really a wrapper that splits the mii monitor into two phases: an 2738 * inspection, then (if inspection indicates something needs to be done) 2739 * an acquisition of appropriate locks followed by a commit phase to 2740 * implement whatever link state changes are indicated. 2741 */ 2742 static void bond_mii_monitor(struct work_struct *work) 2743 { 2744 struct bonding *bond = container_of(work, struct bonding, 2745 mii_work.work); 2746 bool should_notify_peers = false; 2747 bool commit; 2748 unsigned long delay; 2749 struct slave *slave; 2750 struct list_head *iter; 2751 2752 delay = msecs_to_jiffies(bond->params.miimon); 2753 2754 if (!bond_has_slaves(bond)) 2755 goto re_arm; 2756 2757 rcu_read_lock(); 2758 should_notify_peers = bond_should_notify_peers(bond); 2759 commit = !!bond_miimon_inspect(bond); 2760 if (bond->send_peer_notif) { 2761 rcu_read_unlock(); 2762 if (rtnl_trylock()) { 2763 bond->send_peer_notif--; 2764 rtnl_unlock(); 2765 } 2766 } else { 2767 rcu_read_unlock(); 2768 } 2769 2770 if (commit) { 2771 /* Race avoidance with bond_close cancel of workqueue */ 2772 if (!rtnl_trylock()) { 2773 delay = 1; 2774 should_notify_peers = false; 2775 goto re_arm; 2776 } 2777 2778 bond_for_each_slave(bond, slave, iter) { 2779 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2780 } 2781 bond_miimon_commit(bond); 2782 2783 rtnl_unlock(); /* might sleep, hold no other locks */ 2784 } 2785 2786 re_arm: 2787 if (bond->params.miimon) 2788 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2789 2790 if (should_notify_peers) { 2791 if (!rtnl_trylock()) 2792 return; 2793 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2794 rtnl_unlock(); 2795 } 2796 } 2797 2798 static int bond_upper_dev_walk(struct net_device *upper, 2799 struct netdev_nested_priv *priv) 2800 { 2801 __be32 ip = *(__be32 *)priv->data; 2802 2803 return ip == bond_confirm_addr(upper, 0, ip); 2804 } 2805 2806 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2807 { 2808 struct netdev_nested_priv priv = { 2809 .data = (void *)&ip, 2810 }; 2811 bool ret = false; 2812 2813 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2814 return true; 2815 2816 rcu_read_lock(); 2817 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2818 ret = true; 2819 rcu_read_unlock(); 2820 2821 return ret; 2822 } 2823 2824 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2825 struct sk_buff *skb) 2826 { 2827 struct net_device *bond_dev = slave->bond->dev; 2828 struct net_device *slave_dev = slave->dev; 2829 struct bond_vlan_tag *outer_tag = tags; 2830 2831 if (!tags || tags->vlan_proto == VLAN_N_VID) 2832 return true; 2833 2834 tags++; 2835 2836 /* Go through all the tags backwards and add them to the packet */ 2837 while (tags->vlan_proto != VLAN_N_VID) { 2838 if (!tags->vlan_id) { 2839 tags++; 2840 continue; 2841 } 2842 2843 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2844 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2845 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2846 tags->vlan_id); 2847 if (!skb) { 2848 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2849 return false; 2850 } 2851 2852 tags++; 2853 } 2854 /* Set the outer tag */ 2855 if (outer_tag->vlan_id) { 2856 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2857 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2858 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2859 outer_tag->vlan_id); 2860 } 2861 2862 return true; 2863 } 2864 2865 /* We go to the (large) trouble of VLAN tagging ARP frames because 2866 * switches in VLAN mode (especially if ports are configured as 2867 * "native" to a VLAN) might not pass non-tagged frames. 2868 */ 2869 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2870 __be32 src_ip, struct bond_vlan_tag *tags) 2871 { 2872 struct net_device *bond_dev = slave->bond->dev; 2873 struct net_device *slave_dev = slave->dev; 2874 struct sk_buff *skb; 2875 2876 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2877 arp_op, &dest_ip, &src_ip); 2878 2879 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2880 NULL, slave_dev->dev_addr, NULL); 2881 2882 if (!skb) { 2883 net_err_ratelimited("ARP packet allocation failed\n"); 2884 return; 2885 } 2886 2887 if (bond_handle_vlan(slave, tags, skb)) 2888 arp_xmit(skb); 2889 return; 2890 } 2891 2892 /* Validate the device path between the @start_dev and the @end_dev. 2893 * The path is valid if the @end_dev is reachable through device 2894 * stacking. 2895 * When the path is validated, collect any vlan information in the 2896 * path. 2897 */ 2898 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2899 struct net_device *end_dev, 2900 int level) 2901 { 2902 struct bond_vlan_tag *tags; 2903 struct net_device *upper; 2904 struct list_head *iter; 2905 2906 if (start_dev == end_dev) { 2907 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2908 if (!tags) 2909 return ERR_PTR(-ENOMEM); 2910 tags[level].vlan_proto = VLAN_N_VID; 2911 return tags; 2912 } 2913 2914 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2915 tags = bond_verify_device_path(upper, end_dev, level + 1); 2916 if (IS_ERR_OR_NULL(tags)) { 2917 if (IS_ERR(tags)) 2918 return tags; 2919 continue; 2920 } 2921 if (is_vlan_dev(upper)) { 2922 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2923 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2924 } 2925 2926 return tags; 2927 } 2928 2929 return NULL; 2930 } 2931 2932 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2933 { 2934 struct rtable *rt; 2935 struct bond_vlan_tag *tags; 2936 __be32 *targets = bond->params.arp_targets, addr; 2937 int i; 2938 2939 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2940 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2941 __func__, &targets[i]); 2942 tags = NULL; 2943 2944 /* Find out through which dev should the packet go */ 2945 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2946 RTO_ONLINK, 0); 2947 if (IS_ERR(rt)) { 2948 /* there's no route to target - try to send arp 2949 * probe to generate any traffic (arp_validate=0) 2950 */ 2951 if (bond->params.arp_validate) 2952 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2953 bond->dev->name, 2954 &targets[i]); 2955 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2956 0, tags); 2957 continue; 2958 } 2959 2960 /* bond device itself */ 2961 if (rt->dst.dev == bond->dev) 2962 goto found; 2963 2964 rcu_read_lock(); 2965 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2966 rcu_read_unlock(); 2967 2968 if (!IS_ERR_OR_NULL(tags)) 2969 goto found; 2970 2971 /* Not our device - skip */ 2972 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2973 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2974 2975 ip_rt_put(rt); 2976 continue; 2977 2978 found: 2979 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2980 ip_rt_put(rt); 2981 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2982 kfree(tags); 2983 } 2984 } 2985 2986 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2987 { 2988 int i; 2989 2990 if (!sip || !bond_has_this_ip(bond, tip)) { 2991 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2992 __func__, &sip, &tip); 2993 return; 2994 } 2995 2996 i = bond_get_targets_ip(bond->params.arp_targets, sip); 2997 if (i == -1) { 2998 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 2999 __func__, &sip); 3000 return; 3001 } 3002 slave->last_rx = jiffies; 3003 slave->target_last_arp_rx[i] = jiffies; 3004 } 3005 3006 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3007 struct slave *slave) 3008 { 3009 struct arphdr *arp = (struct arphdr *)skb->data; 3010 struct slave *curr_active_slave, *curr_arp_slave; 3011 unsigned char *arp_ptr; 3012 __be32 sip, tip; 3013 unsigned int alen; 3014 3015 alen = arp_hdr_len(bond->dev); 3016 3017 if (alen > skb_headlen(skb)) { 3018 arp = kmalloc(alen, GFP_ATOMIC); 3019 if (!arp) 3020 goto out_unlock; 3021 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3022 goto out_unlock; 3023 } 3024 3025 if (arp->ar_hln != bond->dev->addr_len || 3026 skb->pkt_type == PACKET_OTHERHOST || 3027 skb->pkt_type == PACKET_LOOPBACK || 3028 arp->ar_hrd != htons(ARPHRD_ETHER) || 3029 arp->ar_pro != htons(ETH_P_IP) || 3030 arp->ar_pln != 4) 3031 goto out_unlock; 3032 3033 arp_ptr = (unsigned char *)(arp + 1); 3034 arp_ptr += bond->dev->addr_len; 3035 memcpy(&sip, arp_ptr, 4); 3036 arp_ptr += 4 + bond->dev->addr_len; 3037 memcpy(&tip, arp_ptr, 4); 3038 3039 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3040 __func__, slave->dev->name, bond_slave_state(slave), 3041 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3042 &sip, &tip); 3043 3044 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3045 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3046 3047 /* We 'trust' the received ARP enough to validate it if: 3048 * 3049 * (a) the slave receiving the ARP is active (which includes the 3050 * current ARP slave, if any), or 3051 * 3052 * (b) the receiving slave isn't active, but there is a currently 3053 * active slave and it received valid arp reply(s) after it became 3054 * the currently active slave, or 3055 * 3056 * (c) there is an ARP slave that sent an ARP during the prior ARP 3057 * interval, and we receive an ARP reply on any slave. We accept 3058 * these because switch FDB update delays may deliver the ARP 3059 * reply to a slave other than the sender of the ARP request. 3060 * 3061 * Note: for (b), backup slaves are receiving the broadcast ARP 3062 * request, not a reply. This request passes from the sending 3063 * slave through the L2 switch(es) to the receiving slave. Since 3064 * this is checking the request, sip/tip are swapped for 3065 * validation. 3066 * 3067 * This is done to avoid endless looping when we can't reach the 3068 * arp_ip_target and fool ourselves with our own arp requests. 3069 */ 3070 if (bond_is_active_slave(slave)) 3071 bond_validate_arp(bond, slave, sip, tip); 3072 else if (curr_active_slave && 3073 time_after(slave_last_rx(bond, curr_active_slave), 3074 curr_active_slave->last_link_up)) 3075 bond_validate_arp(bond, slave, tip, sip); 3076 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3077 bond_time_in_interval(bond, 3078 dev_trans_start(curr_arp_slave->dev), 1)) 3079 bond_validate_arp(bond, slave, sip, tip); 3080 3081 out_unlock: 3082 if (arp != (struct arphdr *)skb->data) 3083 kfree(arp); 3084 return RX_HANDLER_ANOTHER; 3085 } 3086 3087 #if IS_ENABLED(CONFIG_IPV6) 3088 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3089 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3090 { 3091 struct net_device *bond_dev = slave->bond->dev; 3092 struct net_device *slave_dev = slave->dev; 3093 struct in6_addr mcaddr; 3094 struct sk_buff *skb; 3095 3096 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3097 daddr, saddr); 3098 3099 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3100 if (!skb) { 3101 net_err_ratelimited("NS packet allocation failed\n"); 3102 return; 3103 } 3104 3105 addrconf_addr_solict_mult(daddr, &mcaddr); 3106 if (bond_handle_vlan(slave, tags, skb)) 3107 ndisc_send_skb(skb, &mcaddr, saddr); 3108 } 3109 3110 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3111 { 3112 struct in6_addr *targets = bond->params.ns_targets; 3113 struct bond_vlan_tag *tags; 3114 struct dst_entry *dst; 3115 struct in6_addr saddr; 3116 struct flowi6 fl6; 3117 int i; 3118 3119 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3120 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3121 __func__, &targets[i]); 3122 tags = NULL; 3123 3124 /* Find out through which dev should the packet go */ 3125 memset(&fl6, 0, sizeof(struct flowi6)); 3126 fl6.daddr = targets[i]; 3127 fl6.flowi6_oif = bond->dev->ifindex; 3128 3129 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3130 if (dst->error) { 3131 dst_release(dst); 3132 /* there's no route to target - try to send arp 3133 * probe to generate any traffic (arp_validate=0) 3134 */ 3135 if (bond->params.arp_validate) 3136 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3137 bond->dev->name, 3138 &targets[i]); 3139 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3140 continue; 3141 } 3142 3143 /* bond device itself */ 3144 if (dst->dev == bond->dev) 3145 goto found; 3146 3147 rcu_read_lock(); 3148 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3149 rcu_read_unlock(); 3150 3151 if (!IS_ERR_OR_NULL(tags)) 3152 goto found; 3153 3154 /* Not our device - skip */ 3155 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3156 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3157 3158 dst_release(dst); 3159 continue; 3160 3161 found: 3162 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3163 bond_ns_send(slave, &targets[i], &saddr, tags); 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_ns(struct bonding *bond, struct slave *slave, 3196 struct in6_addr *saddr, struct in6_addr *daddr) 3197 { 3198 int i; 3199 3200 if (ipv6_addr_any(saddr) || !bond_has_this_ip6(bond, daddr)) { 3201 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3202 __func__, saddr, daddr); 3203 return; 3204 } 3205 3206 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3207 if (i == -1) { 3208 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3209 __func__, saddr); 3210 return; 3211 } 3212 slave->last_rx = jiffies; 3213 slave->target_last_arp_rx[i] = jiffies; 3214 } 3215 3216 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3217 struct slave *slave) 3218 { 3219 struct slave *curr_active_slave, *curr_arp_slave; 3220 struct icmp6hdr *hdr = icmp6_hdr(skb); 3221 struct in6_addr *saddr, *daddr; 3222 3223 if (skb->pkt_type == PACKET_OTHERHOST || 3224 skb->pkt_type == PACKET_LOOPBACK || 3225 hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT) 3226 goto out; 3227 3228 saddr = &ipv6_hdr(skb)->saddr; 3229 daddr = &ipv6_hdr(skb)->daddr; 3230 3231 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3232 __func__, slave->dev->name, bond_slave_state(slave), 3233 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3234 saddr, daddr); 3235 3236 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3237 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3238 3239 /* We 'trust' the received ARP enough to validate it if: 3240 * see bond_arp_rcv(). 3241 */ 3242 if (bond_is_active_slave(slave)) 3243 bond_validate_ns(bond, slave, saddr, daddr); 3244 else if (curr_active_slave && 3245 time_after(slave_last_rx(bond, curr_active_slave), 3246 curr_active_slave->last_link_up)) 3247 bond_validate_ns(bond, slave, saddr, daddr); 3248 else if (curr_arp_slave && 3249 bond_time_in_interval(bond, 3250 dev_trans_start(curr_arp_slave->dev), 1)) 3251 bond_validate_ns(bond, slave, saddr, daddr); 3252 3253 out: 3254 return RX_HANDLER_ANOTHER; 3255 } 3256 #endif 3257 3258 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3259 struct slave *slave) 3260 { 3261 #if IS_ENABLED(CONFIG_IPV6) 3262 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3263 #endif 3264 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3265 3266 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3267 __func__, skb->dev->name); 3268 3269 /* Use arp validate logic for both ARP and NS */ 3270 if (!slave_do_arp_validate(bond, slave)) { 3271 if ((slave_do_arp_validate_only(bond) && is_arp) || 3272 #if IS_ENABLED(CONFIG_IPV6) 3273 (slave_do_arp_validate_only(bond) && is_ipv6) || 3274 #endif 3275 !slave_do_arp_validate_only(bond)) 3276 slave->last_rx = jiffies; 3277 return RX_HANDLER_ANOTHER; 3278 } else if (is_arp) { 3279 return bond_arp_rcv(skb, bond, slave); 3280 #if IS_ENABLED(CONFIG_IPV6) 3281 } else if (is_ipv6) { 3282 return bond_na_rcv(skb, bond, slave); 3283 #endif 3284 } else { 3285 return RX_HANDLER_ANOTHER; 3286 } 3287 } 3288 3289 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3290 { 3291 bond_arp_send_all(bond, slave); 3292 #if IS_ENABLED(CONFIG_IPV6) 3293 bond_ns_send_all(bond, slave); 3294 #endif 3295 } 3296 3297 /* function to verify if we're in the arp_interval timeslice, returns true if 3298 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3299 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3300 */ 3301 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3302 int mod) 3303 { 3304 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3305 3306 return time_in_range(jiffies, 3307 last_act - delta_in_ticks, 3308 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3309 } 3310 3311 /* This function is called regularly to monitor each slave's link 3312 * ensuring that traffic is being sent and received when arp monitoring 3313 * is used in load-balancing mode. if the adapter has been dormant, then an 3314 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3315 * arp monitoring in active backup mode. 3316 */ 3317 static void bond_loadbalance_arp_mon(struct bonding *bond) 3318 { 3319 struct slave *slave, *oldcurrent; 3320 struct list_head *iter; 3321 int do_failover = 0, slave_state_changed = 0; 3322 3323 if (!bond_has_slaves(bond)) 3324 goto re_arm; 3325 3326 rcu_read_lock(); 3327 3328 oldcurrent = rcu_dereference(bond->curr_active_slave); 3329 /* see if any of the previous devices are up now (i.e. they have 3330 * xmt and rcv traffic). the curr_active_slave does not come into 3331 * the picture unless it is null. also, slave->last_link_up is not 3332 * needed here because we send an arp on each slave and give a slave 3333 * as long as it needs to get the tx/rx within the delta. 3334 * TODO: what about up/down delay in arp mode? it wasn't here before 3335 * so it can wait 3336 */ 3337 bond_for_each_slave_rcu(bond, slave, iter) { 3338 unsigned long trans_start = dev_trans_start(slave->dev); 3339 3340 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3341 3342 if (slave->link != BOND_LINK_UP) { 3343 if (bond_time_in_interval(bond, trans_start, 1) && 3344 bond_time_in_interval(bond, slave->last_rx, 1)) { 3345 3346 bond_propose_link_state(slave, BOND_LINK_UP); 3347 slave_state_changed = 1; 3348 3349 /* primary_slave has no meaning in round-robin 3350 * mode. the window of a slave being up and 3351 * curr_active_slave being null after enslaving 3352 * is closed. 3353 */ 3354 if (!oldcurrent) { 3355 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3356 do_failover = 1; 3357 } else { 3358 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3359 } 3360 } 3361 } else { 3362 /* slave->link == BOND_LINK_UP */ 3363 3364 /* not all switches will respond to an arp request 3365 * when the source ip is 0, so don't take the link down 3366 * if we don't know our ip yet 3367 */ 3368 if (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) || 3369 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3370 3371 bond_propose_link_state(slave, BOND_LINK_DOWN); 3372 slave_state_changed = 1; 3373 3374 if (slave->link_failure_count < UINT_MAX) 3375 slave->link_failure_count++; 3376 3377 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3378 3379 if (slave == oldcurrent) 3380 do_failover = 1; 3381 } 3382 } 3383 3384 /* note: if switch is in round-robin mode, all links 3385 * must tx arp to ensure all links rx an arp - otherwise 3386 * links may oscillate or not come up at all; if switch is 3387 * in something like xor mode, there is nothing we can 3388 * do - all replies will be rx'ed on same link causing slaves 3389 * to be unstable during low/no traffic periods 3390 */ 3391 if (bond_slave_is_up(slave)) 3392 bond_send_validate(bond, slave); 3393 } 3394 3395 rcu_read_unlock(); 3396 3397 if (do_failover || slave_state_changed) { 3398 if (!rtnl_trylock()) 3399 goto re_arm; 3400 3401 bond_for_each_slave(bond, slave, iter) { 3402 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3403 slave->link = slave->link_new_state; 3404 } 3405 3406 if (slave_state_changed) { 3407 bond_slave_state_change(bond); 3408 if (BOND_MODE(bond) == BOND_MODE_XOR) 3409 bond_update_slave_arr(bond, NULL); 3410 } 3411 if (do_failover) { 3412 block_netpoll_tx(); 3413 bond_select_active_slave(bond); 3414 unblock_netpoll_tx(); 3415 } 3416 rtnl_unlock(); 3417 } 3418 3419 re_arm: 3420 if (bond->params.arp_interval) 3421 queue_delayed_work(bond->wq, &bond->arp_work, 3422 msecs_to_jiffies(bond->params.arp_interval)); 3423 } 3424 3425 /* Called to inspect slaves for active-backup mode ARP monitor link state 3426 * changes. Sets proposed link state in slaves to specify what action 3427 * should take place for the slave. Returns 0 if no changes are found, >0 3428 * if changes to link states must be committed. 3429 * 3430 * Called with rcu_read_lock held. 3431 */ 3432 static int bond_ab_arp_inspect(struct bonding *bond) 3433 { 3434 unsigned long trans_start, last_rx; 3435 struct list_head *iter; 3436 struct slave *slave; 3437 int commit = 0; 3438 3439 bond_for_each_slave_rcu(bond, slave, iter) { 3440 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3441 last_rx = slave_last_rx(bond, slave); 3442 3443 if (slave->link != BOND_LINK_UP) { 3444 if (bond_time_in_interval(bond, last_rx, 1)) { 3445 bond_propose_link_state(slave, BOND_LINK_UP); 3446 commit++; 3447 } else if (slave->link == BOND_LINK_BACK) { 3448 bond_propose_link_state(slave, BOND_LINK_FAIL); 3449 commit++; 3450 } 3451 continue; 3452 } 3453 3454 /* Give slaves 2*delta after being enslaved or made 3455 * active. This avoids bouncing, as the last receive 3456 * times need a full ARP monitor cycle to be updated. 3457 */ 3458 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3459 continue; 3460 3461 /* Backup slave is down if: 3462 * - No current_arp_slave AND 3463 * - more than (missed_max+1)*delta since last receive AND 3464 * - the bond has an IP address 3465 * 3466 * Note: a non-null current_arp_slave indicates 3467 * the curr_active_slave went down and we are 3468 * searching for a new one; under this condition 3469 * we only take the curr_active_slave down - this 3470 * gives each slave a chance to tx/rx traffic 3471 * before being taken out 3472 */ 3473 if (!bond_is_active_slave(slave) && 3474 !rcu_access_pointer(bond->current_arp_slave) && 3475 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3476 bond_propose_link_state(slave, BOND_LINK_DOWN); 3477 commit++; 3478 } 3479 3480 /* Active slave is down if: 3481 * - more than missed_max*delta since transmitting OR 3482 * - (more than missed_max*delta since receive AND 3483 * the bond has an IP address) 3484 */ 3485 trans_start = dev_trans_start(slave->dev); 3486 if (bond_is_active_slave(slave) && 3487 (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) || 3488 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3489 bond_propose_link_state(slave, BOND_LINK_DOWN); 3490 commit++; 3491 } 3492 } 3493 3494 return commit; 3495 } 3496 3497 /* Called to commit link state changes noted by inspection step of 3498 * active-backup mode ARP monitor. 3499 * 3500 * Called with RTNL hold. 3501 */ 3502 static void bond_ab_arp_commit(struct bonding *bond) 3503 { 3504 unsigned long trans_start; 3505 struct list_head *iter; 3506 struct slave *slave; 3507 3508 bond_for_each_slave(bond, slave, iter) { 3509 switch (slave->link_new_state) { 3510 case BOND_LINK_NOCHANGE: 3511 continue; 3512 3513 case BOND_LINK_UP: 3514 trans_start = dev_trans_start(slave->dev); 3515 if (rtnl_dereference(bond->curr_active_slave) != slave || 3516 (!rtnl_dereference(bond->curr_active_slave) && 3517 bond_time_in_interval(bond, trans_start, 1))) { 3518 struct slave *current_arp_slave; 3519 3520 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3521 bond_set_slave_link_state(slave, BOND_LINK_UP, 3522 BOND_SLAVE_NOTIFY_NOW); 3523 if (current_arp_slave) { 3524 bond_set_slave_inactive_flags( 3525 current_arp_slave, 3526 BOND_SLAVE_NOTIFY_NOW); 3527 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3528 } 3529 3530 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3531 3532 if (!rtnl_dereference(bond->curr_active_slave) || 3533 slave == rtnl_dereference(bond->primary_slave)) 3534 goto do_failover; 3535 3536 } 3537 3538 continue; 3539 3540 case BOND_LINK_DOWN: 3541 if (slave->link_failure_count < UINT_MAX) 3542 slave->link_failure_count++; 3543 3544 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3545 BOND_SLAVE_NOTIFY_NOW); 3546 bond_set_slave_inactive_flags(slave, 3547 BOND_SLAVE_NOTIFY_NOW); 3548 3549 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3550 3551 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3552 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3553 goto do_failover; 3554 } 3555 3556 continue; 3557 3558 case BOND_LINK_FAIL: 3559 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3560 BOND_SLAVE_NOTIFY_NOW); 3561 bond_set_slave_inactive_flags(slave, 3562 BOND_SLAVE_NOTIFY_NOW); 3563 3564 /* A slave has just been enslaved and has become 3565 * the current active slave. 3566 */ 3567 if (rtnl_dereference(bond->curr_active_slave)) 3568 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3569 continue; 3570 3571 default: 3572 slave_err(bond->dev, slave->dev, 3573 "impossible: link_new_state %d on slave\n", 3574 slave->link_new_state); 3575 continue; 3576 } 3577 3578 do_failover: 3579 block_netpoll_tx(); 3580 bond_select_active_slave(bond); 3581 unblock_netpoll_tx(); 3582 } 3583 3584 bond_set_carrier(bond); 3585 } 3586 3587 /* Send ARP probes for active-backup mode ARP monitor. 3588 * 3589 * Called with rcu_read_lock held. 3590 */ 3591 static bool bond_ab_arp_probe(struct bonding *bond) 3592 { 3593 struct slave *slave, *before = NULL, *new_slave = NULL, 3594 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3595 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3596 struct list_head *iter; 3597 bool found = false; 3598 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3599 3600 if (curr_arp_slave && curr_active_slave) 3601 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3602 curr_arp_slave->dev->name, 3603 curr_active_slave->dev->name); 3604 3605 if (curr_active_slave) { 3606 bond_send_validate(bond, curr_active_slave); 3607 return should_notify_rtnl; 3608 } 3609 3610 /* if we don't have a curr_active_slave, search for the next available 3611 * backup slave from the current_arp_slave and make it the candidate 3612 * for becoming the curr_active_slave 3613 */ 3614 3615 if (!curr_arp_slave) { 3616 curr_arp_slave = bond_first_slave_rcu(bond); 3617 if (!curr_arp_slave) 3618 return should_notify_rtnl; 3619 } 3620 3621 bond_for_each_slave_rcu(bond, slave, iter) { 3622 if (!found && !before && bond_slave_is_up(slave)) 3623 before = slave; 3624 3625 if (found && !new_slave && bond_slave_is_up(slave)) 3626 new_slave = slave; 3627 /* if the link state is up at this point, we 3628 * mark it down - this can happen if we have 3629 * simultaneous link failures and 3630 * reselect_active_interface doesn't make this 3631 * one the current slave so it is still marked 3632 * up when it is actually down 3633 */ 3634 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3635 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3636 BOND_SLAVE_NOTIFY_LATER); 3637 if (slave->link_failure_count < UINT_MAX) 3638 slave->link_failure_count++; 3639 3640 bond_set_slave_inactive_flags(slave, 3641 BOND_SLAVE_NOTIFY_LATER); 3642 3643 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3644 } 3645 if (slave == curr_arp_slave) 3646 found = true; 3647 } 3648 3649 if (!new_slave && before) 3650 new_slave = before; 3651 3652 if (!new_slave) 3653 goto check_state; 3654 3655 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3656 BOND_SLAVE_NOTIFY_LATER); 3657 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3658 bond_send_validate(bond, new_slave); 3659 new_slave->last_link_up = jiffies; 3660 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3661 3662 check_state: 3663 bond_for_each_slave_rcu(bond, slave, iter) { 3664 if (slave->should_notify || slave->should_notify_link) { 3665 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3666 break; 3667 } 3668 } 3669 return should_notify_rtnl; 3670 } 3671 3672 static void bond_activebackup_arp_mon(struct bonding *bond) 3673 { 3674 bool should_notify_peers = false; 3675 bool should_notify_rtnl = false; 3676 int delta_in_ticks; 3677 3678 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3679 3680 if (!bond_has_slaves(bond)) 3681 goto re_arm; 3682 3683 rcu_read_lock(); 3684 3685 should_notify_peers = bond_should_notify_peers(bond); 3686 3687 if (bond_ab_arp_inspect(bond)) { 3688 rcu_read_unlock(); 3689 3690 /* Race avoidance with bond_close flush of workqueue */ 3691 if (!rtnl_trylock()) { 3692 delta_in_ticks = 1; 3693 should_notify_peers = false; 3694 goto re_arm; 3695 } 3696 3697 bond_ab_arp_commit(bond); 3698 3699 rtnl_unlock(); 3700 rcu_read_lock(); 3701 } 3702 3703 should_notify_rtnl = bond_ab_arp_probe(bond); 3704 rcu_read_unlock(); 3705 3706 re_arm: 3707 if (bond->params.arp_interval) 3708 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3709 3710 if (should_notify_peers || should_notify_rtnl) { 3711 if (!rtnl_trylock()) 3712 return; 3713 3714 if (should_notify_peers) { 3715 bond->send_peer_notif--; 3716 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3717 bond->dev); 3718 } 3719 if (should_notify_rtnl) { 3720 bond_slave_state_notify(bond); 3721 bond_slave_link_notify(bond); 3722 } 3723 3724 rtnl_unlock(); 3725 } 3726 } 3727 3728 static void bond_arp_monitor(struct work_struct *work) 3729 { 3730 struct bonding *bond = container_of(work, struct bonding, 3731 arp_work.work); 3732 3733 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3734 bond_activebackup_arp_mon(bond); 3735 else 3736 bond_loadbalance_arp_mon(bond); 3737 } 3738 3739 /*-------------------------- netdev event handling --------------------------*/ 3740 3741 /* Change device name */ 3742 static int bond_event_changename(struct bonding *bond) 3743 { 3744 bond_remove_proc_entry(bond); 3745 bond_create_proc_entry(bond); 3746 3747 bond_debug_reregister(bond); 3748 3749 return NOTIFY_DONE; 3750 } 3751 3752 static int bond_master_netdev_event(unsigned long event, 3753 struct net_device *bond_dev) 3754 { 3755 struct bonding *event_bond = netdev_priv(bond_dev); 3756 3757 netdev_dbg(bond_dev, "%s called\n", __func__); 3758 3759 switch (event) { 3760 case NETDEV_CHANGENAME: 3761 return bond_event_changename(event_bond); 3762 case NETDEV_UNREGISTER: 3763 bond_remove_proc_entry(event_bond); 3764 #ifdef CONFIG_XFRM_OFFLOAD 3765 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3766 #endif /* CONFIG_XFRM_OFFLOAD */ 3767 break; 3768 case NETDEV_REGISTER: 3769 bond_create_proc_entry(event_bond); 3770 break; 3771 default: 3772 break; 3773 } 3774 3775 return NOTIFY_DONE; 3776 } 3777 3778 static int bond_slave_netdev_event(unsigned long event, 3779 struct net_device *slave_dev) 3780 { 3781 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3782 struct bonding *bond; 3783 struct net_device *bond_dev; 3784 3785 /* A netdev event can be generated while enslaving a device 3786 * before netdev_rx_handler_register is called in which case 3787 * slave will be NULL 3788 */ 3789 if (!slave) { 3790 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3791 return NOTIFY_DONE; 3792 } 3793 3794 bond_dev = slave->bond->dev; 3795 bond = slave->bond; 3796 primary = rtnl_dereference(bond->primary_slave); 3797 3798 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3799 3800 switch (event) { 3801 case NETDEV_UNREGISTER: 3802 if (bond_dev->type != ARPHRD_ETHER) 3803 bond_release_and_destroy(bond_dev, slave_dev); 3804 else 3805 __bond_release_one(bond_dev, slave_dev, false, true); 3806 break; 3807 case NETDEV_UP: 3808 case NETDEV_CHANGE: 3809 /* For 802.3ad mode only: 3810 * Getting invalid Speed/Duplex values here will put slave 3811 * in weird state. Mark it as link-fail if the link was 3812 * previously up or link-down if it hasn't yet come up, and 3813 * let link-monitoring (miimon) set it right when correct 3814 * speeds/duplex are available. 3815 */ 3816 if (bond_update_speed_duplex(slave) && 3817 BOND_MODE(bond) == BOND_MODE_8023AD) { 3818 if (slave->last_link_up) 3819 slave->link = BOND_LINK_FAIL; 3820 else 3821 slave->link = BOND_LINK_DOWN; 3822 } 3823 3824 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3825 bond_3ad_adapter_speed_duplex_changed(slave); 3826 fallthrough; 3827 case NETDEV_DOWN: 3828 /* Refresh slave-array if applicable! 3829 * If the setup does not use miimon or arpmon (mode-specific!), 3830 * then these events will not cause the slave-array to be 3831 * refreshed. This will cause xmit to use a slave that is not 3832 * usable. Avoid such situation by refeshing the array at these 3833 * events. If these (miimon/arpmon) parameters are configured 3834 * then array gets refreshed twice and that should be fine! 3835 */ 3836 if (bond_mode_can_use_xmit_hash(bond)) 3837 bond_update_slave_arr(bond, NULL); 3838 break; 3839 case NETDEV_CHANGEMTU: 3840 /* TODO: Should slaves be allowed to 3841 * independently alter their MTU? For 3842 * an active-backup bond, slaves need 3843 * not be the same type of device, so 3844 * MTUs may vary. For other modes, 3845 * slaves arguably should have the 3846 * same MTUs. To do this, we'd need to 3847 * take over the slave's change_mtu 3848 * function for the duration of their 3849 * servitude. 3850 */ 3851 break; 3852 case NETDEV_CHANGENAME: 3853 /* we don't care if we don't have primary set */ 3854 if (!bond_uses_primary(bond) || 3855 !bond->params.primary[0]) 3856 break; 3857 3858 if (slave == primary) { 3859 /* slave's name changed - he's no longer primary */ 3860 RCU_INIT_POINTER(bond->primary_slave, NULL); 3861 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3862 /* we have a new primary slave */ 3863 rcu_assign_pointer(bond->primary_slave, slave); 3864 } else { /* we didn't change primary - exit */ 3865 break; 3866 } 3867 3868 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3869 primary ? slave_dev->name : "none"); 3870 3871 block_netpoll_tx(); 3872 bond_select_active_slave(bond); 3873 unblock_netpoll_tx(); 3874 break; 3875 case NETDEV_FEAT_CHANGE: 3876 bond_compute_features(bond); 3877 break; 3878 case NETDEV_RESEND_IGMP: 3879 /* Propagate to master device */ 3880 call_netdevice_notifiers(event, slave->bond->dev); 3881 break; 3882 default: 3883 break; 3884 } 3885 3886 return NOTIFY_DONE; 3887 } 3888 3889 /* bond_netdev_event: handle netdev notifier chain events. 3890 * 3891 * This function receives events for the netdev chain. The caller (an 3892 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3893 * locks for us to safely manipulate the slave devices (RTNL lock, 3894 * dev_probe_lock). 3895 */ 3896 static int bond_netdev_event(struct notifier_block *this, 3897 unsigned long event, void *ptr) 3898 { 3899 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3900 3901 netdev_dbg(event_dev, "%s received %s\n", 3902 __func__, netdev_cmd_to_name(event)); 3903 3904 if (!(event_dev->priv_flags & IFF_BONDING)) 3905 return NOTIFY_DONE; 3906 3907 if (event_dev->flags & IFF_MASTER) { 3908 int ret; 3909 3910 ret = bond_master_netdev_event(event, event_dev); 3911 if (ret != NOTIFY_DONE) 3912 return ret; 3913 } 3914 3915 if (event_dev->flags & IFF_SLAVE) 3916 return bond_slave_netdev_event(event, event_dev); 3917 3918 return NOTIFY_DONE; 3919 } 3920 3921 static struct notifier_block bond_netdev_notifier = { 3922 .notifier_call = bond_netdev_event, 3923 }; 3924 3925 /*---------------------------- Hashing Policies -----------------------------*/ 3926 3927 /* Helper to access data in a packet, with or without a backing skb. 3928 * If skb is given the data is linearized if necessary via pskb_may_pull. 3929 */ 3930 static inline const void *bond_pull_data(struct sk_buff *skb, 3931 const void *data, int hlen, int n) 3932 { 3933 if (likely(n <= hlen)) 3934 return data; 3935 else if (skb && likely(pskb_may_pull(skb, n))) 3936 return skb->head; 3937 3938 return NULL; 3939 } 3940 3941 /* L2 hash helper */ 3942 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3943 { 3944 struct ethhdr *ep; 3945 3946 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 3947 if (!data) 3948 return 0; 3949 3950 ep = (struct ethhdr *)(data + mhoff); 3951 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 3952 } 3953 3954 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 3955 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 3956 { 3957 const struct ipv6hdr *iph6; 3958 const struct iphdr *iph; 3959 3960 if (l2_proto == htons(ETH_P_IP)) { 3961 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 3962 if (!data) 3963 return false; 3964 3965 iph = (const struct iphdr *)(data + *nhoff); 3966 iph_to_flow_copy_v4addrs(fk, iph); 3967 *nhoff += iph->ihl << 2; 3968 if (!ip_is_fragment(iph)) 3969 *ip_proto = iph->protocol; 3970 } else if (l2_proto == htons(ETH_P_IPV6)) { 3971 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 3972 if (!data) 3973 return false; 3974 3975 iph6 = (const struct ipv6hdr *)(data + *nhoff); 3976 iph_to_flow_copy_v6addrs(fk, iph6); 3977 *nhoff += sizeof(*iph6); 3978 *ip_proto = iph6->nexthdr; 3979 } else { 3980 return false; 3981 } 3982 3983 if (l34 && *ip_proto >= 0) 3984 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 3985 3986 return true; 3987 } 3988 3989 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3990 { 3991 u32 srcmac_vendor = 0, srcmac_dev = 0; 3992 struct ethhdr *mac_hdr; 3993 u16 vlan = 0; 3994 int i; 3995 3996 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 3997 if (!data) 3998 return 0; 3999 mac_hdr = (struct ethhdr *)(data + mhoff); 4000 4001 for (i = 0; i < 3; i++) 4002 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4003 4004 for (i = 3; i < ETH_ALEN; i++) 4005 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4006 4007 if (skb && skb_vlan_tag_present(skb)) 4008 vlan = skb_vlan_tag_get(skb); 4009 4010 return vlan ^ srcmac_vendor ^ srcmac_dev; 4011 } 4012 4013 /* Extract the appropriate headers based on bond's xmit policy */ 4014 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4015 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4016 { 4017 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4018 int ip_proto = -1; 4019 4020 switch (bond->params.xmit_policy) { 4021 case BOND_XMIT_POLICY_ENCAP23: 4022 case BOND_XMIT_POLICY_ENCAP34: 4023 memset(fk, 0, sizeof(*fk)); 4024 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4025 fk, data, l2_proto, nhoff, hlen, 0); 4026 default: 4027 break; 4028 } 4029 4030 fk->ports.ports = 0; 4031 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4032 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4033 return false; 4034 4035 /* ICMP error packets contains at least 8 bytes of the header 4036 * of the packet which generated the error. Use this information 4037 * to correlate ICMP error packets within the same flow which 4038 * generated the error. 4039 */ 4040 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4041 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4042 if (ip_proto == IPPROTO_ICMP) { 4043 if (!icmp_is_err(fk->icmp.type)) 4044 return true; 4045 4046 nhoff += sizeof(struct icmphdr); 4047 } else if (ip_proto == IPPROTO_ICMPV6) { 4048 if (!icmpv6_is_err(fk->icmp.type)) 4049 return true; 4050 4051 nhoff += sizeof(struct icmp6hdr); 4052 } 4053 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4054 } 4055 4056 return true; 4057 } 4058 4059 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4060 { 4061 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4062 (__force u32)flow_get_u32_src(flow); 4063 hash ^= (hash >> 16); 4064 hash ^= (hash >> 8); 4065 4066 /* discard lowest hash bit to deal with the common even ports pattern */ 4067 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4068 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4069 return hash >> 1; 4070 4071 return hash; 4072 } 4073 4074 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4075 * the data as required, but this function can be used without it if the data is 4076 * known to be linear (e.g. with xdp_buff). 4077 */ 4078 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4079 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4080 { 4081 struct flow_keys flow; 4082 u32 hash; 4083 4084 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4085 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4086 4087 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4088 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4089 return bond_eth_hash(skb, data, mhoff, hlen); 4090 4091 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4092 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4093 hash = bond_eth_hash(skb, data, mhoff, hlen); 4094 } else { 4095 if (flow.icmp.id) 4096 memcpy(&hash, &flow.icmp, sizeof(hash)); 4097 else 4098 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4099 } 4100 4101 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4102 } 4103 4104 /** 4105 * bond_xmit_hash - generate a hash value based on the xmit policy 4106 * @bond: bonding device 4107 * @skb: buffer to use for headers 4108 * 4109 * This function will extract the necessary headers from the skb buffer and use 4110 * them to generate a hash based on the xmit_policy set in the bonding device 4111 */ 4112 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4113 { 4114 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4115 skb->l4_hash) 4116 return skb->hash; 4117 4118 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4119 skb_mac_offset(skb), skb_network_offset(skb), 4120 skb_headlen(skb)); 4121 } 4122 4123 /** 4124 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4125 * @bond: bonding device 4126 * @xdp: buffer to use for headers 4127 * 4128 * The XDP variant of bond_xmit_hash. 4129 */ 4130 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4131 { 4132 struct ethhdr *eth; 4133 4134 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4135 return 0; 4136 4137 eth = (struct ethhdr *)xdp->data; 4138 4139 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4140 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4141 } 4142 4143 /*-------------------------- Device entry points ----------------------------*/ 4144 4145 void bond_work_init_all(struct bonding *bond) 4146 { 4147 INIT_DELAYED_WORK(&bond->mcast_work, 4148 bond_resend_igmp_join_requests_delayed); 4149 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4150 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4151 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4152 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4153 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4154 } 4155 4156 static void bond_work_cancel_all(struct bonding *bond) 4157 { 4158 cancel_delayed_work_sync(&bond->mii_work); 4159 cancel_delayed_work_sync(&bond->arp_work); 4160 cancel_delayed_work_sync(&bond->alb_work); 4161 cancel_delayed_work_sync(&bond->ad_work); 4162 cancel_delayed_work_sync(&bond->mcast_work); 4163 cancel_delayed_work_sync(&bond->slave_arr_work); 4164 } 4165 4166 static int bond_open(struct net_device *bond_dev) 4167 { 4168 struct bonding *bond = netdev_priv(bond_dev); 4169 struct list_head *iter; 4170 struct slave *slave; 4171 4172 /* reset slave->backup and slave->inactive */ 4173 if (bond_has_slaves(bond)) { 4174 bond_for_each_slave(bond, slave, iter) { 4175 if (bond_uses_primary(bond) && 4176 slave != rcu_access_pointer(bond->curr_active_slave)) { 4177 bond_set_slave_inactive_flags(slave, 4178 BOND_SLAVE_NOTIFY_NOW); 4179 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4180 bond_set_slave_active_flags(slave, 4181 BOND_SLAVE_NOTIFY_NOW); 4182 } 4183 } 4184 } 4185 4186 if (bond_is_lb(bond)) { 4187 /* bond_alb_initialize must be called before the timer 4188 * is started. 4189 */ 4190 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4191 return -ENOMEM; 4192 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4193 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4194 } 4195 4196 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4197 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4198 4199 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4200 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4201 bond->recv_probe = bond_rcv_validate; 4202 } 4203 4204 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4205 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4206 /* register to receive LACPDUs */ 4207 bond->recv_probe = bond_3ad_lacpdu_recv; 4208 bond_3ad_initiate_agg_selection(bond, 1); 4209 } 4210 4211 if (bond_mode_can_use_xmit_hash(bond)) 4212 bond_update_slave_arr(bond, NULL); 4213 4214 return 0; 4215 } 4216 4217 static int bond_close(struct net_device *bond_dev) 4218 { 4219 struct bonding *bond = netdev_priv(bond_dev); 4220 4221 bond_work_cancel_all(bond); 4222 bond->send_peer_notif = 0; 4223 if (bond_is_lb(bond)) 4224 bond_alb_deinitialize(bond); 4225 bond->recv_probe = NULL; 4226 4227 return 0; 4228 } 4229 4230 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4231 * that some drivers can provide 32bit values only. 4232 */ 4233 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4234 const struct rtnl_link_stats64 *_new, 4235 const struct rtnl_link_stats64 *_old) 4236 { 4237 const u64 *new = (const u64 *)_new; 4238 const u64 *old = (const u64 *)_old; 4239 u64 *res = (u64 *)_res; 4240 int i; 4241 4242 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4243 u64 nv = new[i]; 4244 u64 ov = old[i]; 4245 s64 delta = nv - ov; 4246 4247 /* detects if this particular field is 32bit only */ 4248 if (((nv | ov) >> 32) == 0) 4249 delta = (s64)(s32)((u32)nv - (u32)ov); 4250 4251 /* filter anomalies, some drivers reset their stats 4252 * at down/up events. 4253 */ 4254 if (delta > 0) 4255 res[i] += delta; 4256 } 4257 } 4258 4259 #ifdef CONFIG_LOCKDEP 4260 static int bond_get_lowest_level_rcu(struct net_device *dev) 4261 { 4262 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4263 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4264 int cur = 0, max = 0; 4265 4266 now = dev; 4267 iter = &dev->adj_list.lower; 4268 4269 while (1) { 4270 next = NULL; 4271 while (1) { 4272 ldev = netdev_next_lower_dev_rcu(now, &iter); 4273 if (!ldev) 4274 break; 4275 4276 next = ldev; 4277 niter = &ldev->adj_list.lower; 4278 dev_stack[cur] = now; 4279 iter_stack[cur++] = iter; 4280 if (max <= cur) 4281 max = cur; 4282 break; 4283 } 4284 4285 if (!next) { 4286 if (!cur) 4287 return max; 4288 next = dev_stack[--cur]; 4289 niter = iter_stack[cur]; 4290 } 4291 4292 now = next; 4293 iter = niter; 4294 } 4295 4296 return max; 4297 } 4298 #endif 4299 4300 static void bond_get_stats(struct net_device *bond_dev, 4301 struct rtnl_link_stats64 *stats) 4302 { 4303 struct bonding *bond = netdev_priv(bond_dev); 4304 struct rtnl_link_stats64 temp; 4305 struct list_head *iter; 4306 struct slave *slave; 4307 int nest_level = 0; 4308 4309 4310 rcu_read_lock(); 4311 #ifdef CONFIG_LOCKDEP 4312 nest_level = bond_get_lowest_level_rcu(bond_dev); 4313 #endif 4314 4315 spin_lock_nested(&bond->stats_lock, nest_level); 4316 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4317 4318 bond_for_each_slave_rcu(bond, slave, iter) { 4319 const struct rtnl_link_stats64 *new = 4320 dev_get_stats(slave->dev, &temp); 4321 4322 bond_fold_stats(stats, new, &slave->slave_stats); 4323 4324 /* save off the slave stats for the next run */ 4325 memcpy(&slave->slave_stats, new, sizeof(*new)); 4326 } 4327 4328 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4329 spin_unlock(&bond->stats_lock); 4330 rcu_read_unlock(); 4331 } 4332 4333 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4334 { 4335 struct bonding *bond = netdev_priv(bond_dev); 4336 struct mii_ioctl_data *mii = NULL; 4337 const struct net_device_ops *ops; 4338 struct net_device *real_dev; 4339 struct hwtstamp_config cfg; 4340 struct ifreq ifrr; 4341 int res = 0; 4342 4343 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4344 4345 switch (cmd) { 4346 case SIOCGMIIPHY: 4347 mii = if_mii(ifr); 4348 if (!mii) 4349 return -EINVAL; 4350 4351 mii->phy_id = 0; 4352 fallthrough; 4353 case SIOCGMIIREG: 4354 /* We do this again just in case we were called by SIOCGMIIREG 4355 * instead of SIOCGMIIPHY. 4356 */ 4357 mii = if_mii(ifr); 4358 if (!mii) 4359 return -EINVAL; 4360 4361 if (mii->reg_num == 1) { 4362 mii->val_out = 0; 4363 if (netif_carrier_ok(bond->dev)) 4364 mii->val_out = BMSR_LSTATUS; 4365 } 4366 4367 break; 4368 case SIOCSHWTSTAMP: 4369 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4370 return -EFAULT; 4371 4372 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 4373 return -EOPNOTSUPP; 4374 4375 fallthrough; 4376 case SIOCGHWTSTAMP: 4377 real_dev = bond_option_active_slave_get_rcu(bond); 4378 if (!real_dev) 4379 return -EOPNOTSUPP; 4380 4381 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ); 4382 ifrr.ifr_ifru = ifr->ifr_ifru; 4383 4384 ops = real_dev->netdev_ops; 4385 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) { 4386 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd); 4387 if (res) 4388 return res; 4389 4390 ifr->ifr_ifru = ifrr.ifr_ifru; 4391 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4392 return -EFAULT; 4393 4394 /* Set the BOND_PHC_INDEX flag to notify user space */ 4395 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 4396 4397 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ? 4398 -EFAULT : 0; 4399 } 4400 fallthrough; 4401 default: 4402 res = -EOPNOTSUPP; 4403 } 4404 4405 return res; 4406 } 4407 4408 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4409 { 4410 struct bonding *bond = netdev_priv(bond_dev); 4411 struct net_device *slave_dev = NULL; 4412 struct ifbond k_binfo; 4413 struct ifbond __user *u_binfo = NULL; 4414 struct ifslave k_sinfo; 4415 struct ifslave __user *u_sinfo = NULL; 4416 struct bond_opt_value newval; 4417 struct net *net; 4418 int res = 0; 4419 4420 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4421 4422 switch (cmd) { 4423 case SIOCBONDINFOQUERY: 4424 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4425 4426 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4427 return -EFAULT; 4428 4429 bond_info_query(bond_dev, &k_binfo); 4430 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4431 return -EFAULT; 4432 4433 return 0; 4434 case SIOCBONDSLAVEINFOQUERY: 4435 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4436 4437 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4438 return -EFAULT; 4439 4440 res = bond_slave_info_query(bond_dev, &k_sinfo); 4441 if (res == 0 && 4442 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4443 return -EFAULT; 4444 4445 return res; 4446 default: 4447 break; 4448 } 4449 4450 net = dev_net(bond_dev); 4451 4452 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4453 return -EPERM; 4454 4455 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4456 4457 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4458 4459 if (!slave_dev) 4460 return -ENODEV; 4461 4462 switch (cmd) { 4463 case SIOCBONDENSLAVE: 4464 res = bond_enslave(bond_dev, slave_dev, NULL); 4465 break; 4466 case SIOCBONDRELEASE: 4467 res = bond_release(bond_dev, slave_dev); 4468 break; 4469 case SIOCBONDSETHWADDR: 4470 res = bond_set_dev_addr(bond_dev, slave_dev); 4471 break; 4472 case SIOCBONDCHANGEACTIVE: 4473 bond_opt_initstr(&newval, slave_dev->name); 4474 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4475 &newval); 4476 break; 4477 default: 4478 res = -EOPNOTSUPP; 4479 } 4480 4481 return res; 4482 } 4483 4484 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4485 void __user *data, int cmd) 4486 { 4487 struct ifreq ifrdata = { .ifr_data = data }; 4488 4489 switch (cmd) { 4490 case BOND_INFO_QUERY_OLD: 4491 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4492 case BOND_SLAVE_INFO_QUERY_OLD: 4493 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4494 case BOND_ENSLAVE_OLD: 4495 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4496 case BOND_RELEASE_OLD: 4497 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4498 case BOND_SETHWADDR_OLD: 4499 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4500 case BOND_CHANGE_ACTIVE_OLD: 4501 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4502 } 4503 4504 return -EOPNOTSUPP; 4505 } 4506 4507 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4508 { 4509 struct bonding *bond = netdev_priv(bond_dev); 4510 4511 if (change & IFF_PROMISC) 4512 bond_set_promiscuity(bond, 4513 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4514 4515 if (change & IFF_ALLMULTI) 4516 bond_set_allmulti(bond, 4517 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4518 } 4519 4520 static void bond_set_rx_mode(struct net_device *bond_dev) 4521 { 4522 struct bonding *bond = netdev_priv(bond_dev); 4523 struct list_head *iter; 4524 struct slave *slave; 4525 4526 rcu_read_lock(); 4527 if (bond_uses_primary(bond)) { 4528 slave = rcu_dereference(bond->curr_active_slave); 4529 if (slave) { 4530 dev_uc_sync(slave->dev, bond_dev); 4531 dev_mc_sync(slave->dev, bond_dev); 4532 } 4533 } else { 4534 bond_for_each_slave_rcu(bond, slave, iter) { 4535 dev_uc_sync_multiple(slave->dev, bond_dev); 4536 dev_mc_sync_multiple(slave->dev, bond_dev); 4537 } 4538 } 4539 rcu_read_unlock(); 4540 } 4541 4542 static int bond_neigh_init(struct neighbour *n) 4543 { 4544 struct bonding *bond = netdev_priv(n->dev); 4545 const struct net_device_ops *slave_ops; 4546 struct neigh_parms parms; 4547 struct slave *slave; 4548 int ret = 0; 4549 4550 rcu_read_lock(); 4551 slave = bond_first_slave_rcu(bond); 4552 if (!slave) 4553 goto out; 4554 slave_ops = slave->dev->netdev_ops; 4555 if (!slave_ops->ndo_neigh_setup) 4556 goto out; 4557 4558 /* TODO: find another way [1] to implement this. 4559 * Passing a zeroed structure is fragile, 4560 * but at least we do not pass garbage. 4561 * 4562 * [1] One way would be that ndo_neigh_setup() never touch 4563 * struct neigh_parms, but propagate the new neigh_setup() 4564 * back to ___neigh_create() / neigh_parms_alloc() 4565 */ 4566 memset(&parms, 0, sizeof(parms)); 4567 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4568 4569 if (ret) 4570 goto out; 4571 4572 if (parms.neigh_setup) 4573 ret = parms.neigh_setup(n); 4574 out: 4575 rcu_read_unlock(); 4576 return ret; 4577 } 4578 4579 /* The bonding ndo_neigh_setup is called at init time beofre any 4580 * slave exists. So we must declare proxy setup function which will 4581 * be used at run time to resolve the actual slave neigh param setup. 4582 * 4583 * It's also called by master devices (such as vlans) to setup their 4584 * underlying devices. In that case - do nothing, we're already set up from 4585 * our init. 4586 */ 4587 static int bond_neigh_setup(struct net_device *dev, 4588 struct neigh_parms *parms) 4589 { 4590 /* modify only our neigh_parms */ 4591 if (parms->dev == dev) 4592 parms->neigh_setup = bond_neigh_init; 4593 4594 return 0; 4595 } 4596 4597 /* Change the MTU of all of a master's slaves to match the master */ 4598 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4599 { 4600 struct bonding *bond = netdev_priv(bond_dev); 4601 struct slave *slave, *rollback_slave; 4602 struct list_head *iter; 4603 int res = 0; 4604 4605 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4606 4607 bond_for_each_slave(bond, slave, iter) { 4608 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4609 slave, slave->dev->netdev_ops->ndo_change_mtu); 4610 4611 res = dev_set_mtu(slave->dev, new_mtu); 4612 4613 if (res) { 4614 /* If we failed to set the slave's mtu to the new value 4615 * we must abort the operation even in ACTIVE_BACKUP 4616 * mode, because if we allow the backup slaves to have 4617 * different mtu values than the active slave we'll 4618 * need to change their mtu when doing a failover. That 4619 * means changing their mtu from timer context, which 4620 * is probably not a good idea. 4621 */ 4622 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4623 res, new_mtu); 4624 goto unwind; 4625 } 4626 } 4627 4628 bond_dev->mtu = new_mtu; 4629 4630 return 0; 4631 4632 unwind: 4633 /* unwind from head to the slave that failed */ 4634 bond_for_each_slave(bond, rollback_slave, iter) { 4635 int tmp_res; 4636 4637 if (rollback_slave == slave) 4638 break; 4639 4640 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4641 if (tmp_res) 4642 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4643 tmp_res); 4644 } 4645 4646 return res; 4647 } 4648 4649 /* Change HW address 4650 * 4651 * Note that many devices must be down to change the HW address, and 4652 * downing the master releases all slaves. We can make bonds full of 4653 * bonding devices to test this, however. 4654 */ 4655 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4656 { 4657 struct bonding *bond = netdev_priv(bond_dev); 4658 struct slave *slave, *rollback_slave; 4659 struct sockaddr_storage *ss = addr, tmp_ss; 4660 struct list_head *iter; 4661 int res = 0; 4662 4663 if (BOND_MODE(bond) == BOND_MODE_ALB) 4664 return bond_alb_set_mac_address(bond_dev, addr); 4665 4666 4667 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4668 4669 /* If fail_over_mac is enabled, do nothing and return success. 4670 * Returning an error causes ifenslave to fail. 4671 */ 4672 if (bond->params.fail_over_mac && 4673 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4674 return 0; 4675 4676 if (!is_valid_ether_addr(ss->__data)) 4677 return -EADDRNOTAVAIL; 4678 4679 bond_for_each_slave(bond, slave, iter) { 4680 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4681 __func__, slave); 4682 res = dev_set_mac_address(slave->dev, addr, NULL); 4683 if (res) { 4684 /* TODO: consider downing the slave 4685 * and retry ? 4686 * User should expect communications 4687 * breakage anyway until ARP finish 4688 * updating, so... 4689 */ 4690 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4691 __func__, res); 4692 goto unwind; 4693 } 4694 } 4695 4696 /* success */ 4697 dev_addr_set(bond_dev, ss->__data); 4698 return 0; 4699 4700 unwind: 4701 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4702 tmp_ss.ss_family = bond_dev->type; 4703 4704 /* unwind from head to the slave that failed */ 4705 bond_for_each_slave(bond, rollback_slave, iter) { 4706 int tmp_res; 4707 4708 if (rollback_slave == slave) 4709 break; 4710 4711 tmp_res = dev_set_mac_address(rollback_slave->dev, 4712 (struct sockaddr *)&tmp_ss, NULL); 4713 if (tmp_res) { 4714 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4715 __func__, tmp_res); 4716 } 4717 } 4718 4719 return res; 4720 } 4721 4722 /** 4723 * bond_get_slave_by_id - get xmit slave with slave_id 4724 * @bond: bonding device that is transmitting 4725 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4726 * 4727 * This function tries to get slave with slave_id but in case 4728 * it fails, it tries to find the first available slave for transmission. 4729 */ 4730 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4731 int slave_id) 4732 { 4733 struct list_head *iter; 4734 struct slave *slave; 4735 int i = slave_id; 4736 4737 /* Here we start from the slave with slave_id */ 4738 bond_for_each_slave_rcu(bond, slave, iter) { 4739 if (--i < 0) { 4740 if (bond_slave_can_tx(slave)) 4741 return slave; 4742 } 4743 } 4744 4745 /* Here we start from the first slave up to slave_id */ 4746 i = slave_id; 4747 bond_for_each_slave_rcu(bond, slave, iter) { 4748 if (--i < 0) 4749 break; 4750 if (bond_slave_can_tx(slave)) 4751 return slave; 4752 } 4753 /* no slave that can tx has been found */ 4754 return NULL; 4755 } 4756 4757 /** 4758 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4759 * @bond: bonding device to use 4760 * 4761 * Based on the value of the bonding device's packets_per_slave parameter 4762 * this function generates a slave id, which is usually used as the next 4763 * slave to transmit through. 4764 */ 4765 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4766 { 4767 u32 slave_id; 4768 struct reciprocal_value reciprocal_packets_per_slave; 4769 int packets_per_slave = bond->params.packets_per_slave; 4770 4771 switch (packets_per_slave) { 4772 case 0: 4773 slave_id = prandom_u32(); 4774 break; 4775 case 1: 4776 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4777 break; 4778 default: 4779 reciprocal_packets_per_slave = 4780 bond->params.reciprocal_packets_per_slave; 4781 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4782 slave_id = reciprocal_divide(slave_id, 4783 reciprocal_packets_per_slave); 4784 break; 4785 } 4786 4787 return slave_id; 4788 } 4789 4790 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4791 struct sk_buff *skb) 4792 { 4793 struct slave *slave; 4794 int slave_cnt; 4795 u32 slave_id; 4796 4797 /* Start with the curr_active_slave that joined the bond as the 4798 * default for sending IGMP traffic. For failover purposes one 4799 * needs to maintain some consistency for the interface that will 4800 * send the join/membership reports. The curr_active_slave found 4801 * will send all of this type of traffic. 4802 */ 4803 if (skb->protocol == htons(ETH_P_IP)) { 4804 int noff = skb_network_offset(skb); 4805 struct iphdr *iph; 4806 4807 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4808 goto non_igmp; 4809 4810 iph = ip_hdr(skb); 4811 if (iph->protocol == IPPROTO_IGMP) { 4812 slave = rcu_dereference(bond->curr_active_slave); 4813 if (slave) 4814 return slave; 4815 return bond_get_slave_by_id(bond, 0); 4816 } 4817 } 4818 4819 non_igmp: 4820 slave_cnt = READ_ONCE(bond->slave_cnt); 4821 if (likely(slave_cnt)) { 4822 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4823 return bond_get_slave_by_id(bond, slave_id); 4824 } 4825 return NULL; 4826 } 4827 4828 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4829 struct xdp_buff *xdp) 4830 { 4831 struct slave *slave; 4832 int slave_cnt; 4833 u32 slave_id; 4834 const struct ethhdr *eth; 4835 void *data = xdp->data; 4836 4837 if (data + sizeof(struct ethhdr) > xdp->data_end) 4838 goto non_igmp; 4839 4840 eth = (struct ethhdr *)data; 4841 data += sizeof(struct ethhdr); 4842 4843 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4844 if (eth->h_proto == htons(ETH_P_IP)) { 4845 const struct iphdr *iph; 4846 4847 if (data + sizeof(struct iphdr) > xdp->data_end) 4848 goto non_igmp; 4849 4850 iph = (struct iphdr *)data; 4851 4852 if (iph->protocol == IPPROTO_IGMP) { 4853 slave = rcu_dereference(bond->curr_active_slave); 4854 if (slave) 4855 return slave; 4856 return bond_get_slave_by_id(bond, 0); 4857 } 4858 } 4859 4860 non_igmp: 4861 slave_cnt = READ_ONCE(bond->slave_cnt); 4862 if (likely(slave_cnt)) { 4863 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4864 return bond_get_slave_by_id(bond, slave_id); 4865 } 4866 return NULL; 4867 } 4868 4869 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4870 struct net_device *bond_dev) 4871 { 4872 struct bonding *bond = netdev_priv(bond_dev); 4873 struct slave *slave; 4874 4875 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4876 if (likely(slave)) 4877 return bond_dev_queue_xmit(bond, skb, slave->dev); 4878 4879 return bond_tx_drop(bond_dev, skb); 4880 } 4881 4882 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4883 { 4884 return rcu_dereference(bond->curr_active_slave); 4885 } 4886 4887 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4888 * the bond has a usable interface. 4889 */ 4890 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4891 struct net_device *bond_dev) 4892 { 4893 struct bonding *bond = netdev_priv(bond_dev); 4894 struct slave *slave; 4895 4896 slave = bond_xmit_activebackup_slave_get(bond); 4897 if (slave) 4898 return bond_dev_queue_xmit(bond, skb, slave->dev); 4899 4900 return bond_tx_drop(bond_dev, skb); 4901 } 4902 4903 /* Use this to update slave_array when (a) it's not appropriate to update 4904 * slave_array right away (note that update_slave_array() may sleep) 4905 * and / or (b) RTNL is not held. 4906 */ 4907 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4908 { 4909 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4910 } 4911 4912 /* Slave array work handler. Holds only RTNL */ 4913 static void bond_slave_arr_handler(struct work_struct *work) 4914 { 4915 struct bonding *bond = container_of(work, struct bonding, 4916 slave_arr_work.work); 4917 int ret; 4918 4919 if (!rtnl_trylock()) 4920 goto err; 4921 4922 ret = bond_update_slave_arr(bond, NULL); 4923 rtnl_unlock(); 4924 if (ret) { 4925 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4926 goto err; 4927 } 4928 return; 4929 4930 err: 4931 bond_slave_arr_work_rearm(bond, 1); 4932 } 4933 4934 static void bond_skip_slave(struct bond_up_slave *slaves, 4935 struct slave *skipslave) 4936 { 4937 int idx; 4938 4939 /* Rare situation where caller has asked to skip a specific 4940 * slave but allocation failed (most likely!). BTW this is 4941 * only possible when the call is initiated from 4942 * __bond_release_one(). In this situation; overwrite the 4943 * skipslave entry in the array with the last entry from the 4944 * array to avoid a situation where the xmit path may choose 4945 * this to-be-skipped slave to send a packet out. 4946 */ 4947 for (idx = 0; slaves && idx < slaves->count; idx++) { 4948 if (skipslave == slaves->arr[idx]) { 4949 slaves->arr[idx] = 4950 slaves->arr[slaves->count - 1]; 4951 slaves->count--; 4952 break; 4953 } 4954 } 4955 } 4956 4957 static void bond_set_slave_arr(struct bonding *bond, 4958 struct bond_up_slave *usable_slaves, 4959 struct bond_up_slave *all_slaves) 4960 { 4961 struct bond_up_slave *usable, *all; 4962 4963 usable = rtnl_dereference(bond->usable_slaves); 4964 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 4965 kfree_rcu(usable, rcu); 4966 4967 all = rtnl_dereference(bond->all_slaves); 4968 rcu_assign_pointer(bond->all_slaves, all_slaves); 4969 kfree_rcu(all, rcu); 4970 } 4971 4972 static void bond_reset_slave_arr(struct bonding *bond) 4973 { 4974 struct bond_up_slave *usable, *all; 4975 4976 usable = rtnl_dereference(bond->usable_slaves); 4977 if (usable) { 4978 RCU_INIT_POINTER(bond->usable_slaves, NULL); 4979 kfree_rcu(usable, rcu); 4980 } 4981 4982 all = rtnl_dereference(bond->all_slaves); 4983 if (all) { 4984 RCU_INIT_POINTER(bond->all_slaves, NULL); 4985 kfree_rcu(all, rcu); 4986 } 4987 } 4988 4989 /* Build the usable slaves array in control path for modes that use xmit-hash 4990 * to determine the slave interface - 4991 * (a) BOND_MODE_8023AD 4992 * (b) BOND_MODE_XOR 4993 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 4994 * 4995 * The caller is expected to hold RTNL only and NO other lock! 4996 */ 4997 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 4998 { 4999 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5000 struct slave *slave; 5001 struct list_head *iter; 5002 int agg_id = 0; 5003 int ret = 0; 5004 5005 might_sleep(); 5006 5007 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5008 bond->slave_cnt), GFP_KERNEL); 5009 all_slaves = kzalloc(struct_size(all_slaves, arr, 5010 bond->slave_cnt), GFP_KERNEL); 5011 if (!usable_slaves || !all_slaves) { 5012 ret = -ENOMEM; 5013 goto out; 5014 } 5015 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5016 struct ad_info ad_info; 5017 5018 spin_lock_bh(&bond->mode_lock); 5019 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5020 spin_unlock_bh(&bond->mode_lock); 5021 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5022 /* No active aggragator means it's not safe to use 5023 * the previous array. 5024 */ 5025 bond_reset_slave_arr(bond); 5026 goto out; 5027 } 5028 spin_unlock_bh(&bond->mode_lock); 5029 agg_id = ad_info.aggregator_id; 5030 } 5031 bond_for_each_slave(bond, slave, iter) { 5032 if (skipslave == slave) 5033 continue; 5034 5035 all_slaves->arr[all_slaves->count++] = slave; 5036 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5037 struct aggregator *agg; 5038 5039 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5040 if (!agg || agg->aggregator_identifier != agg_id) 5041 continue; 5042 } 5043 if (!bond_slave_can_tx(slave)) 5044 continue; 5045 5046 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5047 usable_slaves->count); 5048 5049 usable_slaves->arr[usable_slaves->count++] = slave; 5050 } 5051 5052 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5053 return ret; 5054 out: 5055 if (ret != 0 && skipslave) { 5056 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5057 skipslave); 5058 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5059 skipslave); 5060 } 5061 kfree_rcu(all_slaves, rcu); 5062 kfree_rcu(usable_slaves, rcu); 5063 5064 return ret; 5065 } 5066 5067 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5068 struct sk_buff *skb, 5069 struct bond_up_slave *slaves) 5070 { 5071 struct slave *slave; 5072 unsigned int count; 5073 u32 hash; 5074 5075 hash = bond_xmit_hash(bond, skb); 5076 count = slaves ? READ_ONCE(slaves->count) : 0; 5077 if (unlikely(!count)) 5078 return NULL; 5079 5080 slave = slaves->arr[hash % count]; 5081 return slave; 5082 } 5083 5084 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5085 struct xdp_buff *xdp) 5086 { 5087 struct bond_up_slave *slaves; 5088 unsigned int count; 5089 u32 hash; 5090 5091 hash = bond_xmit_hash_xdp(bond, xdp); 5092 slaves = rcu_dereference(bond->usable_slaves); 5093 count = slaves ? READ_ONCE(slaves->count) : 0; 5094 if (unlikely(!count)) 5095 return NULL; 5096 5097 return slaves->arr[hash % count]; 5098 } 5099 5100 /* Use this Xmit function for 3AD as well as XOR modes. The current 5101 * usable slave array is formed in the control path. The xmit function 5102 * just calculates hash and sends the packet out. 5103 */ 5104 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5105 struct net_device *dev) 5106 { 5107 struct bonding *bond = netdev_priv(dev); 5108 struct bond_up_slave *slaves; 5109 struct slave *slave; 5110 5111 slaves = rcu_dereference(bond->usable_slaves); 5112 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5113 if (likely(slave)) 5114 return bond_dev_queue_xmit(bond, skb, slave->dev); 5115 5116 return bond_tx_drop(dev, skb); 5117 } 5118 5119 /* in broadcast mode, we send everything to all usable interfaces. */ 5120 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5121 struct net_device *bond_dev) 5122 { 5123 struct bonding *bond = netdev_priv(bond_dev); 5124 struct slave *slave = NULL; 5125 struct list_head *iter; 5126 bool xmit_suc = false; 5127 bool skb_used = false; 5128 5129 bond_for_each_slave_rcu(bond, slave, iter) { 5130 struct sk_buff *skb2; 5131 5132 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5133 continue; 5134 5135 if (bond_is_last_slave(bond, slave)) { 5136 skb2 = skb; 5137 skb_used = true; 5138 } else { 5139 skb2 = skb_clone(skb, GFP_ATOMIC); 5140 if (!skb2) { 5141 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5142 bond_dev->name, __func__); 5143 continue; 5144 } 5145 } 5146 5147 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5148 xmit_suc = true; 5149 } 5150 5151 if (!skb_used) 5152 dev_kfree_skb_any(skb); 5153 5154 if (xmit_suc) 5155 return NETDEV_TX_OK; 5156 5157 dev_core_stats_tx_dropped_inc(bond_dev); 5158 return NET_XMIT_DROP; 5159 } 5160 5161 /*------------------------- Device initialization ---------------------------*/ 5162 5163 /* Lookup the slave that corresponds to a qid */ 5164 static inline int bond_slave_override(struct bonding *bond, 5165 struct sk_buff *skb) 5166 { 5167 struct slave *slave = NULL; 5168 struct list_head *iter; 5169 5170 if (!skb_rx_queue_recorded(skb)) 5171 return 1; 5172 5173 /* Find out if any slaves have the same mapping as this skb. */ 5174 bond_for_each_slave_rcu(bond, slave, iter) { 5175 if (slave->queue_id == skb_get_queue_mapping(skb)) { 5176 if (bond_slave_is_up(slave) && 5177 slave->link == BOND_LINK_UP) { 5178 bond_dev_queue_xmit(bond, skb, slave->dev); 5179 return 0; 5180 } 5181 /* If the slave isn't UP, use default transmit policy. */ 5182 break; 5183 } 5184 } 5185 5186 return 1; 5187 } 5188 5189 5190 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5191 struct net_device *sb_dev) 5192 { 5193 /* This helper function exists to help dev_pick_tx get the correct 5194 * destination queue. Using a helper function skips a call to 5195 * skb_tx_hash and will put the skbs in the queue we expect on their 5196 * way down to the bonding driver. 5197 */ 5198 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5199 5200 /* Save the original txq to restore before passing to the driver */ 5201 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5202 5203 if (unlikely(txq >= dev->real_num_tx_queues)) { 5204 do { 5205 txq -= dev->real_num_tx_queues; 5206 } while (txq >= dev->real_num_tx_queues); 5207 } 5208 return txq; 5209 } 5210 5211 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5212 struct sk_buff *skb, 5213 bool all_slaves) 5214 { 5215 struct bonding *bond = netdev_priv(master_dev); 5216 struct bond_up_slave *slaves; 5217 struct slave *slave = NULL; 5218 5219 switch (BOND_MODE(bond)) { 5220 case BOND_MODE_ROUNDROBIN: 5221 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5222 break; 5223 case BOND_MODE_ACTIVEBACKUP: 5224 slave = bond_xmit_activebackup_slave_get(bond); 5225 break; 5226 case BOND_MODE_8023AD: 5227 case BOND_MODE_XOR: 5228 if (all_slaves) 5229 slaves = rcu_dereference(bond->all_slaves); 5230 else 5231 slaves = rcu_dereference(bond->usable_slaves); 5232 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5233 break; 5234 case BOND_MODE_BROADCAST: 5235 break; 5236 case BOND_MODE_ALB: 5237 slave = bond_xmit_alb_slave_get(bond, skb); 5238 break; 5239 case BOND_MODE_TLB: 5240 slave = bond_xmit_tlb_slave_get(bond, skb); 5241 break; 5242 default: 5243 /* Should never happen, mode already checked */ 5244 WARN_ONCE(true, "Unknown bonding mode"); 5245 break; 5246 } 5247 5248 if (slave) 5249 return slave->dev; 5250 return NULL; 5251 } 5252 5253 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5254 { 5255 switch (sk->sk_family) { 5256 #if IS_ENABLED(CONFIG_IPV6) 5257 case AF_INET6: 5258 if (ipv6_only_sock(sk) || 5259 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5260 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5261 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5262 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5263 break; 5264 } 5265 fallthrough; 5266 #endif 5267 default: /* AF_INET */ 5268 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5269 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5270 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5271 break; 5272 } 5273 5274 flow->ports.src = inet_sk(sk)->inet_sport; 5275 flow->ports.dst = inet_sk(sk)->inet_dport; 5276 } 5277 5278 /** 5279 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5280 * @sk: socket to use for headers 5281 * 5282 * This function will extract the necessary field from the socket and use 5283 * them to generate a hash based on the LAYER34 xmit_policy. 5284 * Assumes that sk is a TCP or UDP socket. 5285 */ 5286 static u32 bond_sk_hash_l34(struct sock *sk) 5287 { 5288 struct flow_keys flow; 5289 u32 hash; 5290 5291 bond_sk_to_flow(sk, &flow); 5292 5293 /* L4 */ 5294 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5295 /* L3 */ 5296 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5297 } 5298 5299 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5300 struct sock *sk) 5301 { 5302 struct bond_up_slave *slaves; 5303 struct slave *slave; 5304 unsigned int count; 5305 u32 hash; 5306 5307 slaves = rcu_dereference(bond->usable_slaves); 5308 count = slaves ? READ_ONCE(slaves->count) : 0; 5309 if (unlikely(!count)) 5310 return NULL; 5311 5312 hash = bond_sk_hash_l34(sk); 5313 slave = slaves->arr[hash % count]; 5314 5315 return slave->dev; 5316 } 5317 5318 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5319 struct sock *sk) 5320 { 5321 struct bonding *bond = netdev_priv(dev); 5322 struct net_device *lower = NULL; 5323 5324 rcu_read_lock(); 5325 if (bond_sk_check(bond)) 5326 lower = __bond_sk_get_lower_dev(bond, sk); 5327 rcu_read_unlock(); 5328 5329 return lower; 5330 } 5331 5332 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5333 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5334 struct net_device *dev) 5335 { 5336 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev))) 5337 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev); 5338 return bond_tx_drop(dev, skb); 5339 } 5340 #endif 5341 5342 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5343 { 5344 struct bonding *bond = netdev_priv(dev); 5345 5346 if (bond_should_override_tx_queue(bond) && 5347 !bond_slave_override(bond, skb)) 5348 return NETDEV_TX_OK; 5349 5350 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5351 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk)) 5352 return bond_tls_device_xmit(bond, skb, dev); 5353 #endif 5354 5355 switch (BOND_MODE(bond)) { 5356 case BOND_MODE_ROUNDROBIN: 5357 return bond_xmit_roundrobin(skb, dev); 5358 case BOND_MODE_ACTIVEBACKUP: 5359 return bond_xmit_activebackup(skb, dev); 5360 case BOND_MODE_8023AD: 5361 case BOND_MODE_XOR: 5362 return bond_3ad_xor_xmit(skb, dev); 5363 case BOND_MODE_BROADCAST: 5364 return bond_xmit_broadcast(skb, dev); 5365 case BOND_MODE_ALB: 5366 return bond_alb_xmit(skb, dev); 5367 case BOND_MODE_TLB: 5368 return bond_tlb_xmit(skb, dev); 5369 default: 5370 /* Should never happen, mode already checked */ 5371 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5372 WARN_ON_ONCE(1); 5373 return bond_tx_drop(dev, skb); 5374 } 5375 } 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 netdev_tx_t ret = NETDEV_TX_OK; 5381 5382 /* If we risk deadlock from transmitting this in the 5383 * netpoll path, tell netpoll to queue the frame for later tx 5384 */ 5385 if (unlikely(is_netpoll_tx_blocked(dev))) 5386 return NETDEV_TX_BUSY; 5387 5388 rcu_read_lock(); 5389 if (bond_has_slaves(bond)) 5390 ret = __bond_start_xmit(skb, dev); 5391 else 5392 ret = bond_tx_drop(dev, skb); 5393 rcu_read_unlock(); 5394 5395 return ret; 5396 } 5397 5398 static struct net_device * 5399 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5400 { 5401 struct bonding *bond = netdev_priv(bond_dev); 5402 struct slave *slave; 5403 5404 /* Caller needs to hold rcu_read_lock() */ 5405 5406 switch (BOND_MODE(bond)) { 5407 case BOND_MODE_ROUNDROBIN: 5408 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5409 break; 5410 5411 case BOND_MODE_ACTIVEBACKUP: 5412 slave = bond_xmit_activebackup_slave_get(bond); 5413 break; 5414 5415 case BOND_MODE_8023AD: 5416 case BOND_MODE_XOR: 5417 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5418 break; 5419 5420 default: 5421 /* Should never happen. Mode guarded by bond_xdp_check() */ 5422 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond)); 5423 WARN_ON_ONCE(1); 5424 return NULL; 5425 } 5426 5427 if (slave) 5428 return slave->dev; 5429 5430 return NULL; 5431 } 5432 5433 static int bond_xdp_xmit(struct net_device *bond_dev, 5434 int n, struct xdp_frame **frames, u32 flags) 5435 { 5436 int nxmit, err = -ENXIO; 5437 5438 rcu_read_lock(); 5439 5440 for (nxmit = 0; nxmit < n; nxmit++) { 5441 struct xdp_frame *frame = frames[nxmit]; 5442 struct xdp_frame *frames1[] = {frame}; 5443 struct net_device *slave_dev; 5444 struct xdp_buff xdp; 5445 5446 xdp_convert_frame_to_buff(frame, &xdp); 5447 5448 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5449 if (!slave_dev) { 5450 err = -ENXIO; 5451 break; 5452 } 5453 5454 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5455 if (err < 1) 5456 break; 5457 } 5458 5459 rcu_read_unlock(); 5460 5461 /* If error happened on the first frame then we can pass the error up, otherwise 5462 * report the number of frames that were xmitted. 5463 */ 5464 if (err < 0) 5465 return (nxmit == 0 ? err : nxmit); 5466 5467 return nxmit; 5468 } 5469 5470 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5471 struct netlink_ext_ack *extack) 5472 { 5473 struct bonding *bond = netdev_priv(dev); 5474 struct list_head *iter; 5475 struct slave *slave, *rollback_slave; 5476 struct bpf_prog *old_prog; 5477 struct netdev_bpf xdp = { 5478 .command = XDP_SETUP_PROG, 5479 .flags = 0, 5480 .prog = prog, 5481 .extack = extack, 5482 }; 5483 int err; 5484 5485 ASSERT_RTNL(); 5486 5487 if (!bond_xdp_check(bond)) 5488 return -EOPNOTSUPP; 5489 5490 old_prog = bond->xdp_prog; 5491 bond->xdp_prog = prog; 5492 5493 bond_for_each_slave(bond, slave, iter) { 5494 struct net_device *slave_dev = slave->dev; 5495 5496 if (!slave_dev->netdev_ops->ndo_bpf || 5497 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5498 SLAVE_NL_ERR(dev, slave_dev, extack, 5499 "Slave device does not support XDP"); 5500 err = -EOPNOTSUPP; 5501 goto err; 5502 } 5503 5504 if (dev_xdp_prog_count(slave_dev) > 0) { 5505 SLAVE_NL_ERR(dev, slave_dev, extack, 5506 "Slave has XDP program loaded, please unload before enslaving"); 5507 err = -EOPNOTSUPP; 5508 goto err; 5509 } 5510 5511 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5512 if (err < 0) { 5513 /* ndo_bpf() sets extack error message */ 5514 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5515 goto err; 5516 } 5517 if (prog) 5518 bpf_prog_inc(prog); 5519 } 5520 5521 if (prog) { 5522 static_branch_inc(&bpf_master_redirect_enabled_key); 5523 } else if (old_prog) { 5524 bpf_prog_put(old_prog); 5525 static_branch_dec(&bpf_master_redirect_enabled_key); 5526 } 5527 5528 return 0; 5529 5530 err: 5531 /* unwind the program changes */ 5532 bond->xdp_prog = old_prog; 5533 xdp.prog = old_prog; 5534 xdp.extack = NULL; /* do not overwrite original error */ 5535 5536 bond_for_each_slave(bond, rollback_slave, iter) { 5537 struct net_device *slave_dev = rollback_slave->dev; 5538 int err_unwind; 5539 5540 if (slave == rollback_slave) 5541 break; 5542 5543 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5544 if (err_unwind < 0) 5545 slave_err(dev, slave_dev, 5546 "Error %d when unwinding XDP program change\n", err_unwind); 5547 else if (xdp.prog) 5548 bpf_prog_inc(xdp.prog); 5549 } 5550 return err; 5551 } 5552 5553 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5554 { 5555 switch (xdp->command) { 5556 case XDP_SETUP_PROG: 5557 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5558 default: 5559 return -EINVAL; 5560 } 5561 } 5562 5563 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5564 { 5565 if (speed == 0 || speed == SPEED_UNKNOWN) 5566 speed = slave->speed; 5567 else 5568 speed = min(speed, slave->speed); 5569 5570 return speed; 5571 } 5572 5573 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5574 struct ethtool_link_ksettings *cmd) 5575 { 5576 struct bonding *bond = netdev_priv(bond_dev); 5577 struct list_head *iter; 5578 struct slave *slave; 5579 u32 speed = 0; 5580 5581 cmd->base.duplex = DUPLEX_UNKNOWN; 5582 cmd->base.port = PORT_OTHER; 5583 5584 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5585 * do not need to check mode. Though link speed might not represent 5586 * the true receive or transmit bandwidth (not all modes are symmetric) 5587 * this is an accurate maximum. 5588 */ 5589 bond_for_each_slave(bond, slave, iter) { 5590 if (bond_slave_can_tx(slave)) { 5591 if (slave->speed != SPEED_UNKNOWN) { 5592 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5593 speed = bond_mode_bcast_speed(slave, 5594 speed); 5595 else 5596 speed += slave->speed; 5597 } 5598 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5599 slave->duplex != DUPLEX_UNKNOWN) 5600 cmd->base.duplex = slave->duplex; 5601 } 5602 } 5603 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5604 5605 return 0; 5606 } 5607 5608 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5609 struct ethtool_drvinfo *drvinfo) 5610 { 5611 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5612 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5613 BOND_ABI_VERSION); 5614 } 5615 5616 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5617 struct ethtool_ts_info *info) 5618 { 5619 struct bonding *bond = netdev_priv(bond_dev); 5620 const struct ethtool_ops *ops; 5621 struct net_device *real_dev; 5622 struct phy_device *phydev; 5623 int ret = 0; 5624 5625 rcu_read_lock(); 5626 real_dev = bond_option_active_slave_get_rcu(bond); 5627 dev_hold(real_dev); 5628 rcu_read_unlock(); 5629 5630 if (real_dev) { 5631 ops = real_dev->ethtool_ops; 5632 phydev = real_dev->phydev; 5633 5634 if (phy_has_tsinfo(phydev)) { 5635 ret = phy_ts_info(phydev, info); 5636 goto out; 5637 } else if (ops->get_ts_info) { 5638 ret = ops->get_ts_info(real_dev, info); 5639 goto out; 5640 } 5641 } 5642 5643 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE | 5644 SOF_TIMESTAMPING_SOFTWARE; 5645 info->phc_index = -1; 5646 5647 out: 5648 dev_put(real_dev); 5649 return ret; 5650 } 5651 5652 static const struct ethtool_ops bond_ethtool_ops = { 5653 .get_drvinfo = bond_ethtool_get_drvinfo, 5654 .get_link = ethtool_op_get_link, 5655 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5656 .get_ts_info = bond_ethtool_get_ts_info, 5657 }; 5658 5659 static const struct net_device_ops bond_netdev_ops = { 5660 .ndo_init = bond_init, 5661 .ndo_uninit = bond_uninit, 5662 .ndo_open = bond_open, 5663 .ndo_stop = bond_close, 5664 .ndo_start_xmit = bond_start_xmit, 5665 .ndo_select_queue = bond_select_queue, 5666 .ndo_get_stats64 = bond_get_stats, 5667 .ndo_eth_ioctl = bond_eth_ioctl, 5668 .ndo_siocbond = bond_do_ioctl, 5669 .ndo_siocdevprivate = bond_siocdevprivate, 5670 .ndo_change_rx_flags = bond_change_rx_flags, 5671 .ndo_set_rx_mode = bond_set_rx_mode, 5672 .ndo_change_mtu = bond_change_mtu, 5673 .ndo_set_mac_address = bond_set_mac_address, 5674 .ndo_neigh_setup = bond_neigh_setup, 5675 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5676 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5677 #ifdef CONFIG_NET_POLL_CONTROLLER 5678 .ndo_netpoll_setup = bond_netpoll_setup, 5679 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5680 .ndo_poll_controller = bond_poll_controller, 5681 #endif 5682 .ndo_add_slave = bond_enslave, 5683 .ndo_del_slave = bond_release, 5684 .ndo_fix_features = bond_fix_features, 5685 .ndo_features_check = passthru_features_check, 5686 .ndo_get_xmit_slave = bond_xmit_get_slave, 5687 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5688 .ndo_bpf = bond_xdp, 5689 .ndo_xdp_xmit = bond_xdp_xmit, 5690 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5691 }; 5692 5693 static const struct device_type bond_type = { 5694 .name = "bond", 5695 }; 5696 5697 static void bond_destructor(struct net_device *bond_dev) 5698 { 5699 struct bonding *bond = netdev_priv(bond_dev); 5700 5701 if (bond->wq) 5702 destroy_workqueue(bond->wq); 5703 5704 if (bond->rr_tx_counter) 5705 free_percpu(bond->rr_tx_counter); 5706 } 5707 5708 void bond_setup(struct net_device *bond_dev) 5709 { 5710 struct bonding *bond = netdev_priv(bond_dev); 5711 5712 spin_lock_init(&bond->mode_lock); 5713 bond->params = bonding_defaults; 5714 5715 /* Initialize pointers */ 5716 bond->dev = bond_dev; 5717 5718 /* Initialize the device entry points */ 5719 ether_setup(bond_dev); 5720 bond_dev->max_mtu = ETH_MAX_MTU; 5721 bond_dev->netdev_ops = &bond_netdev_ops; 5722 bond_dev->ethtool_ops = &bond_ethtool_ops; 5723 5724 bond_dev->needs_free_netdev = true; 5725 bond_dev->priv_destructor = bond_destructor; 5726 5727 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5728 5729 /* Initialize the device options */ 5730 bond_dev->flags |= IFF_MASTER; 5731 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5732 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5733 5734 #ifdef CONFIG_XFRM_OFFLOAD 5735 /* set up xfrm device ops (only supported in active-backup right now) */ 5736 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5737 INIT_LIST_HEAD(&bond->ipsec_list); 5738 spin_lock_init(&bond->ipsec_lock); 5739 #endif /* CONFIG_XFRM_OFFLOAD */ 5740 5741 /* don't acquire bond device's netif_tx_lock when transmitting */ 5742 bond_dev->features |= NETIF_F_LLTX; 5743 5744 /* By default, we declare the bond to be fully 5745 * VLAN hardware accelerated capable. Special 5746 * care is taken in the various xmit functions 5747 * when there are slaves that are not hw accel 5748 * capable 5749 */ 5750 5751 /* Don't allow bond devices to change network namespaces. */ 5752 bond_dev->features |= NETIF_F_NETNS_LOCAL; 5753 5754 bond_dev->hw_features = BOND_VLAN_FEATURES | 5755 NETIF_F_HW_VLAN_CTAG_RX | 5756 NETIF_F_HW_VLAN_CTAG_FILTER; 5757 5758 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5759 bond_dev->features |= bond_dev->hw_features; 5760 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5761 #ifdef CONFIG_XFRM_OFFLOAD 5762 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5763 /* Only enable XFRM features if this is an active-backup config */ 5764 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5765 bond_dev->features |= BOND_XFRM_FEATURES; 5766 #endif /* CONFIG_XFRM_OFFLOAD */ 5767 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5768 if (bond_sk_check(bond)) 5769 bond_dev->features |= BOND_TLS_FEATURES; 5770 #endif 5771 } 5772 5773 /* Destroy a bonding device. 5774 * Must be under rtnl_lock when this function is called. 5775 */ 5776 static void bond_uninit(struct net_device *bond_dev) 5777 { 5778 struct bonding *bond = netdev_priv(bond_dev); 5779 struct bond_up_slave *usable, *all; 5780 struct list_head *iter; 5781 struct slave *slave; 5782 5783 bond_netpoll_cleanup(bond_dev); 5784 5785 /* Release the bonded slaves */ 5786 bond_for_each_slave(bond, slave, iter) 5787 __bond_release_one(bond_dev, slave->dev, true, true); 5788 netdev_info(bond_dev, "Released all slaves\n"); 5789 5790 usable = rtnl_dereference(bond->usable_slaves); 5791 if (usable) { 5792 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5793 kfree_rcu(usable, rcu); 5794 } 5795 5796 all = rtnl_dereference(bond->all_slaves); 5797 if (all) { 5798 RCU_INIT_POINTER(bond->all_slaves, NULL); 5799 kfree_rcu(all, rcu); 5800 } 5801 5802 list_del(&bond->bond_list); 5803 5804 bond_debug_unregister(bond); 5805 } 5806 5807 /*------------------------- Module initialization ---------------------------*/ 5808 5809 static int bond_check_params(struct bond_params *params) 5810 { 5811 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 5812 struct bond_opt_value newval; 5813 const struct bond_opt_value *valptr; 5814 int arp_all_targets_value = 0; 5815 u16 ad_actor_sys_prio = 0; 5816 u16 ad_user_port_key = 0; 5817 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 5818 int arp_ip_count; 5819 int bond_mode = BOND_MODE_ROUNDROBIN; 5820 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 5821 int lacp_fast = 0; 5822 int tlb_dynamic_lb; 5823 5824 /* Convert string parameters. */ 5825 if (mode) { 5826 bond_opt_initstr(&newval, mode); 5827 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5828 if (!valptr) { 5829 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5830 return -EINVAL; 5831 } 5832 bond_mode = valptr->value; 5833 } 5834 5835 if (xmit_hash_policy) { 5836 if (bond_mode == BOND_MODE_ROUNDROBIN || 5837 bond_mode == BOND_MODE_ACTIVEBACKUP || 5838 bond_mode == BOND_MODE_BROADCAST) { 5839 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 5840 bond_mode_name(bond_mode)); 5841 } else { 5842 bond_opt_initstr(&newval, xmit_hash_policy); 5843 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5844 &newval); 5845 if (!valptr) { 5846 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5847 xmit_hash_policy); 5848 return -EINVAL; 5849 } 5850 xmit_hashtype = valptr->value; 5851 } 5852 } 5853 5854 if (lacp_rate) { 5855 if (bond_mode != BOND_MODE_8023AD) { 5856 pr_info("lacp_rate param is irrelevant in mode %s\n", 5857 bond_mode_name(bond_mode)); 5858 } else { 5859 bond_opt_initstr(&newval, lacp_rate); 5860 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5861 &newval); 5862 if (!valptr) { 5863 pr_err("Error: Invalid lacp rate \"%s\"\n", 5864 lacp_rate); 5865 return -EINVAL; 5866 } 5867 lacp_fast = valptr->value; 5868 } 5869 } 5870 5871 if (ad_select) { 5872 bond_opt_initstr(&newval, ad_select); 5873 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5874 &newval); 5875 if (!valptr) { 5876 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5877 return -EINVAL; 5878 } 5879 params->ad_select = valptr->value; 5880 if (bond_mode != BOND_MODE_8023AD) 5881 pr_warn("ad_select param only affects 802.3ad mode\n"); 5882 } else { 5883 params->ad_select = BOND_AD_STABLE; 5884 } 5885 5886 if (max_bonds < 0) { 5887 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5888 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5889 max_bonds = BOND_DEFAULT_MAX_BONDS; 5890 } 5891 5892 if (miimon < 0) { 5893 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5894 miimon, INT_MAX); 5895 miimon = 0; 5896 } 5897 5898 if (updelay < 0) { 5899 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5900 updelay, INT_MAX); 5901 updelay = 0; 5902 } 5903 5904 if (downdelay < 0) { 5905 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5906 downdelay, INT_MAX); 5907 downdelay = 0; 5908 } 5909 5910 if ((use_carrier != 0) && (use_carrier != 1)) { 5911 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5912 use_carrier); 5913 use_carrier = 1; 5914 } 5915 5916 if (num_peer_notif < 0 || num_peer_notif > 255) { 5917 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5918 num_peer_notif); 5919 num_peer_notif = 1; 5920 } 5921 5922 /* reset values for 802.3ad/TLB/ALB */ 5923 if (!bond_mode_uses_arp(bond_mode)) { 5924 if (!miimon) { 5925 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"); 5926 pr_warn("Forcing miimon to 100msec\n"); 5927 miimon = BOND_DEFAULT_MIIMON; 5928 } 5929 } 5930 5931 if (tx_queues < 1 || tx_queues > 255) { 5932 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5933 tx_queues, BOND_DEFAULT_TX_QUEUES); 5934 tx_queues = BOND_DEFAULT_TX_QUEUES; 5935 } 5936 5937 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5938 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5939 all_slaves_active); 5940 all_slaves_active = 0; 5941 } 5942 5943 if (resend_igmp < 0 || resend_igmp > 255) { 5944 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 5945 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 5946 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 5947 } 5948 5949 bond_opt_initval(&newval, packets_per_slave); 5950 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 5951 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 5952 packets_per_slave, USHRT_MAX); 5953 packets_per_slave = 1; 5954 } 5955 5956 if (bond_mode == BOND_MODE_ALB) { 5957 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", 5958 updelay); 5959 } 5960 5961 if (!miimon) { 5962 if (updelay || downdelay) { 5963 /* just warn the user the up/down delay will have 5964 * no effect since miimon is zero... 5965 */ 5966 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", 5967 updelay, downdelay); 5968 } 5969 } else { 5970 /* don't allow arp monitoring */ 5971 if (arp_interval) { 5972 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 5973 miimon, arp_interval); 5974 arp_interval = 0; 5975 } 5976 5977 if ((updelay % miimon) != 0) { 5978 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 5979 updelay, miimon, (updelay / miimon) * miimon); 5980 } 5981 5982 updelay /= miimon; 5983 5984 if ((downdelay % miimon) != 0) { 5985 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 5986 downdelay, miimon, 5987 (downdelay / miimon) * miimon); 5988 } 5989 5990 downdelay /= miimon; 5991 } 5992 5993 if (arp_interval < 0) { 5994 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5995 arp_interval, INT_MAX); 5996 arp_interval = 0; 5997 } 5998 5999 for (arp_ip_count = 0, i = 0; 6000 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6001 __be32 ip; 6002 6003 /* not a complete check, but good enough to catch mistakes */ 6004 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6005 !bond_is_ip_target_ok(ip)) { 6006 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6007 arp_ip_target[i]); 6008 arp_interval = 0; 6009 } else { 6010 if (bond_get_targets_ip(arp_target, ip) == -1) 6011 arp_target[arp_ip_count++] = ip; 6012 else 6013 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6014 &ip); 6015 } 6016 } 6017 6018 if (arp_interval && !arp_ip_count) { 6019 /* don't allow arping if no arp_ip_target given... */ 6020 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6021 arp_interval); 6022 arp_interval = 0; 6023 } 6024 6025 if (arp_validate) { 6026 if (!arp_interval) { 6027 pr_err("arp_validate requires arp_interval\n"); 6028 return -EINVAL; 6029 } 6030 6031 bond_opt_initstr(&newval, arp_validate); 6032 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6033 &newval); 6034 if (!valptr) { 6035 pr_err("Error: invalid arp_validate \"%s\"\n", 6036 arp_validate); 6037 return -EINVAL; 6038 } 6039 arp_validate_value = valptr->value; 6040 } else { 6041 arp_validate_value = 0; 6042 } 6043 6044 if (arp_all_targets) { 6045 bond_opt_initstr(&newval, arp_all_targets); 6046 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6047 &newval); 6048 if (!valptr) { 6049 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6050 arp_all_targets); 6051 arp_all_targets_value = 0; 6052 } else { 6053 arp_all_targets_value = valptr->value; 6054 } 6055 } 6056 6057 if (miimon) { 6058 pr_info("MII link monitoring set to %d ms\n", miimon); 6059 } else if (arp_interval) { 6060 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6061 arp_validate_value); 6062 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6063 arp_interval, valptr->string, arp_ip_count); 6064 6065 for (i = 0; i < arp_ip_count; i++) 6066 pr_cont(" %s", arp_ip_target[i]); 6067 6068 pr_cont("\n"); 6069 6070 } else if (max_bonds) { 6071 /* miimon and arp_interval not set, we need one so things 6072 * work as expected, see bonding.txt for details 6073 */ 6074 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"); 6075 } 6076 6077 if (primary && !bond_mode_uses_primary(bond_mode)) { 6078 /* currently, using a primary only makes sense 6079 * in active backup, TLB or ALB modes 6080 */ 6081 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6082 primary, bond_mode_name(bond_mode)); 6083 primary = NULL; 6084 } 6085 6086 if (primary && primary_reselect) { 6087 bond_opt_initstr(&newval, primary_reselect); 6088 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6089 &newval); 6090 if (!valptr) { 6091 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6092 primary_reselect); 6093 return -EINVAL; 6094 } 6095 primary_reselect_value = valptr->value; 6096 } else { 6097 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6098 } 6099 6100 if (fail_over_mac) { 6101 bond_opt_initstr(&newval, fail_over_mac); 6102 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6103 &newval); 6104 if (!valptr) { 6105 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6106 fail_over_mac); 6107 return -EINVAL; 6108 } 6109 fail_over_mac_value = valptr->value; 6110 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6111 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6112 } else { 6113 fail_over_mac_value = BOND_FOM_NONE; 6114 } 6115 6116 bond_opt_initstr(&newval, "default"); 6117 valptr = bond_opt_parse( 6118 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6119 &newval); 6120 if (!valptr) { 6121 pr_err("Error: No ad_actor_sys_prio default value"); 6122 return -EINVAL; 6123 } 6124 ad_actor_sys_prio = valptr->value; 6125 6126 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6127 &newval); 6128 if (!valptr) { 6129 pr_err("Error: No ad_user_port_key default value"); 6130 return -EINVAL; 6131 } 6132 ad_user_port_key = valptr->value; 6133 6134 bond_opt_initstr(&newval, "default"); 6135 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6136 if (!valptr) { 6137 pr_err("Error: No tlb_dynamic_lb default value"); 6138 return -EINVAL; 6139 } 6140 tlb_dynamic_lb = valptr->value; 6141 6142 if (lp_interval == 0) { 6143 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6144 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6145 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6146 } 6147 6148 /* fill params struct with the proper values */ 6149 params->mode = bond_mode; 6150 params->xmit_policy = xmit_hashtype; 6151 params->miimon = miimon; 6152 params->num_peer_notif = num_peer_notif; 6153 params->arp_interval = arp_interval; 6154 params->arp_validate = arp_validate_value; 6155 params->arp_all_targets = arp_all_targets_value; 6156 params->missed_max = 2; 6157 params->updelay = updelay; 6158 params->downdelay = downdelay; 6159 params->peer_notif_delay = 0; 6160 params->use_carrier = use_carrier; 6161 params->lacp_active = 1; 6162 params->lacp_fast = lacp_fast; 6163 params->primary[0] = 0; 6164 params->primary_reselect = primary_reselect_value; 6165 params->fail_over_mac = fail_over_mac_value; 6166 params->tx_queues = tx_queues; 6167 params->all_slaves_active = all_slaves_active; 6168 params->resend_igmp = resend_igmp; 6169 params->min_links = min_links; 6170 params->lp_interval = lp_interval; 6171 params->packets_per_slave = packets_per_slave; 6172 params->tlb_dynamic_lb = tlb_dynamic_lb; 6173 params->ad_actor_sys_prio = ad_actor_sys_prio; 6174 eth_zero_addr(params->ad_actor_system); 6175 params->ad_user_port_key = ad_user_port_key; 6176 if (packets_per_slave > 0) { 6177 params->reciprocal_packets_per_slave = 6178 reciprocal_value(packets_per_slave); 6179 } else { 6180 /* reciprocal_packets_per_slave is unused if 6181 * packets_per_slave is 0 or 1, just initialize it 6182 */ 6183 params->reciprocal_packets_per_slave = 6184 (struct reciprocal_value) { 0 }; 6185 } 6186 6187 if (primary) 6188 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6189 6190 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6191 #if IS_ENABLED(CONFIG_IPV6) 6192 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6193 #endif 6194 6195 return 0; 6196 } 6197 6198 /* Called from registration process */ 6199 static int bond_init(struct net_device *bond_dev) 6200 { 6201 struct bonding *bond = netdev_priv(bond_dev); 6202 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6203 6204 netdev_dbg(bond_dev, "Begin bond_init\n"); 6205 6206 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 6207 if (!bond->wq) 6208 return -ENOMEM; 6209 6210 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) { 6211 bond->rr_tx_counter = alloc_percpu(u32); 6212 if (!bond->rr_tx_counter) { 6213 destroy_workqueue(bond->wq); 6214 bond->wq = NULL; 6215 return -ENOMEM; 6216 } 6217 } 6218 6219 spin_lock_init(&bond->stats_lock); 6220 netdev_lockdep_set_classes(bond_dev); 6221 6222 list_add_tail(&bond->bond_list, &bn->dev_list); 6223 6224 bond_prepare_sysfs_group(bond); 6225 6226 bond_debug_register(bond); 6227 6228 /* Ensure valid dev_addr */ 6229 if (is_zero_ether_addr(bond_dev->dev_addr) && 6230 bond_dev->addr_assign_type == NET_ADDR_PERM) 6231 eth_hw_addr_random(bond_dev); 6232 6233 return 0; 6234 } 6235 6236 unsigned int bond_get_num_tx_queues(void) 6237 { 6238 return tx_queues; 6239 } 6240 6241 /* Create a new bond based on the specified name and bonding parameters. 6242 * If name is NULL, obtain a suitable "bond%d" name for us. 6243 * Caller must NOT hold rtnl_lock; we need to release it here before we 6244 * set up our sysfs entries. 6245 */ 6246 int bond_create(struct net *net, const char *name) 6247 { 6248 struct net_device *bond_dev; 6249 struct bonding *bond; 6250 int res = -ENOMEM; 6251 6252 rtnl_lock(); 6253 6254 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6255 name ? name : "bond%d", NET_NAME_UNKNOWN, 6256 bond_setup, tx_queues); 6257 if (!bond_dev) 6258 goto out; 6259 6260 bond = netdev_priv(bond_dev); 6261 dev_net_set(bond_dev, net); 6262 bond_dev->rtnl_link_ops = &bond_link_ops; 6263 6264 res = register_netdevice(bond_dev); 6265 if (res < 0) { 6266 free_netdev(bond_dev); 6267 goto out; 6268 } 6269 6270 netif_carrier_off(bond_dev); 6271 6272 bond_work_init_all(bond); 6273 6274 out: 6275 rtnl_unlock(); 6276 return res; 6277 } 6278 6279 static int __net_init bond_net_init(struct net *net) 6280 { 6281 struct bond_net *bn = net_generic(net, bond_net_id); 6282 6283 bn->net = net; 6284 INIT_LIST_HEAD(&bn->dev_list); 6285 6286 bond_create_proc_dir(bn); 6287 bond_create_sysfs(bn); 6288 6289 return 0; 6290 } 6291 6292 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6293 { 6294 struct bond_net *bn; 6295 struct net *net; 6296 LIST_HEAD(list); 6297 6298 list_for_each_entry(net, net_list, exit_list) { 6299 bn = net_generic(net, bond_net_id); 6300 bond_destroy_sysfs(bn); 6301 } 6302 6303 /* Kill off any bonds created after unregistering bond rtnl ops */ 6304 rtnl_lock(); 6305 list_for_each_entry(net, net_list, exit_list) { 6306 struct bonding *bond, *tmp_bond; 6307 6308 bn = net_generic(net, bond_net_id); 6309 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6310 unregister_netdevice_queue(bond->dev, &list); 6311 } 6312 unregister_netdevice_many(&list); 6313 rtnl_unlock(); 6314 6315 list_for_each_entry(net, net_list, exit_list) { 6316 bn = net_generic(net, bond_net_id); 6317 bond_destroy_proc_dir(bn); 6318 } 6319 } 6320 6321 static struct pernet_operations bond_net_ops = { 6322 .init = bond_net_init, 6323 .exit_batch = bond_net_exit_batch, 6324 .id = &bond_net_id, 6325 .size = sizeof(struct bond_net), 6326 }; 6327 6328 static int __init bonding_init(void) 6329 { 6330 int i; 6331 int res; 6332 6333 res = bond_check_params(&bonding_defaults); 6334 if (res) 6335 goto out; 6336 6337 res = register_pernet_subsys(&bond_net_ops); 6338 if (res) 6339 goto out; 6340 6341 res = bond_netlink_init(); 6342 if (res) 6343 goto err_link; 6344 6345 bond_create_debugfs(); 6346 6347 for (i = 0; i < max_bonds; i++) { 6348 res = bond_create(&init_net, NULL); 6349 if (res) 6350 goto err; 6351 } 6352 6353 skb_flow_dissector_init(&flow_keys_bonding, 6354 flow_keys_bonding_keys, 6355 ARRAY_SIZE(flow_keys_bonding_keys)); 6356 6357 register_netdevice_notifier(&bond_netdev_notifier); 6358 out: 6359 return res; 6360 err: 6361 bond_destroy_debugfs(); 6362 bond_netlink_fini(); 6363 err_link: 6364 unregister_pernet_subsys(&bond_net_ops); 6365 goto out; 6366 6367 } 6368 6369 static void __exit bonding_exit(void) 6370 { 6371 unregister_netdevice_notifier(&bond_netdev_notifier); 6372 6373 bond_destroy_debugfs(); 6374 6375 bond_netlink_fini(); 6376 unregister_pernet_subsys(&bond_net_ops); 6377 6378 #ifdef CONFIG_NET_POLL_CONTROLLER 6379 /* Make sure we don't have an imbalance on our netpoll blocking */ 6380 WARN_ON(atomic_read(&netpoll_block_tx)); 6381 #endif 6382 } 6383 6384 module_init(bonding_init); 6385 module_exit(bonding_exit); 6386 MODULE_LICENSE("GPL"); 6387 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6388 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6389