1 /* 2 * originally based on the dummy device. 3 * 4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 5 * Licensed under the GPL. Based on dummy.c, and eql.c devices. 6 * 7 * bonding.c: an Ethernet Bonding driver 8 * 9 * This is useful to talk to a Cisco EtherChannel compatible equipment: 10 * Cisco 5500 11 * Sun Trunking (Solaris) 12 * Alteon AceDirector Trunks 13 * Linux Bonding 14 * and probably many L2 switches ... 15 * 16 * How it works: 17 * ifconfig bond0 ipaddress netmask up 18 * will setup a network device, with an ip address. No mac address 19 * will be assigned at this time. The hw mac address will come from 20 * the first slave bonded to the channel. All slaves will then use 21 * this hw mac address. 22 * 23 * ifconfig bond0 down 24 * will release all slaves, marking them as down. 25 * 26 * ifenslave bond0 eth0 27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 28 * a: be used as initial mac address 29 * b: if a hw mac address already is there, eth0's hw mac address 30 * will then be set from bond0. 31 * 32 */ 33 34 #include <linux/kernel.h> 35 #include <linux/module.h> 36 #include <linux/types.h> 37 #include <linux/fcntl.h> 38 #include <linux/filter.h> 39 #include <linux/interrupt.h> 40 #include <linux/ptrace.h> 41 #include <linux/ioport.h> 42 #include <linux/in.h> 43 #include <net/ip.h> 44 #include <linux/ip.h> 45 #include <linux/icmp.h> 46 #include <linux/icmpv6.h> 47 #include <linux/tcp.h> 48 #include <linux/udp.h> 49 #include <linux/slab.h> 50 #include <linux/string.h> 51 #include <linux/init.h> 52 #include <linux/timer.h> 53 #include <linux/socket.h> 54 #include <linux/ctype.h> 55 #include <linux/inet.h> 56 #include <linux/bitops.h> 57 #include <linux/io.h> 58 #include <asm/dma.h> 59 #include <linux/uaccess.h> 60 #include <linux/errno.h> 61 #include <linux/netdevice.h> 62 #include <linux/inetdevice.h> 63 #include <linux/igmp.h> 64 #include <linux/etherdevice.h> 65 #include <linux/skbuff.h> 66 #include <net/sock.h> 67 #include <linux/rtnetlink.h> 68 #include <linux/smp.h> 69 #include <linux/if_ether.h> 70 #include <net/arp.h> 71 #include <linux/mii.h> 72 #include <linux/ethtool.h> 73 #include <linux/if_vlan.h> 74 #include <linux/if_bonding.h> 75 #include <linux/phy.h> 76 #include <linux/jiffies.h> 77 #include <linux/preempt.h> 78 #include <net/route.h> 79 #include <net/net_namespace.h> 80 #include <net/netns/generic.h> 81 #include <net/pkt_sched.h> 82 #include <linux/rculist.h> 83 #include <net/flow_dissector.h> 84 #include <net/xfrm.h> 85 #include <net/bonding.h> 86 #include <net/bond_3ad.h> 87 #include <net/bond_alb.h> 88 #if IS_ENABLED(CONFIG_TLS_DEVICE) 89 #include <net/tls.h> 90 #endif 91 #include <net/ip6_route.h> 92 93 #include "bonding_priv.h" 94 95 /*---------------------------- Module parameters ----------------------------*/ 96 97 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 98 99 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 100 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 101 static int num_peer_notif = 1; 102 static int miimon; 103 static int updelay; 104 static int downdelay; 105 static int use_carrier = 1; 106 static char *mode; 107 static char *primary; 108 static char *primary_reselect; 109 static char *lacp_rate; 110 static int min_links; 111 static char *ad_select; 112 static char *xmit_hash_policy; 113 static int arp_interval; 114 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 115 static char *arp_validate; 116 static char *arp_all_targets; 117 static char *fail_over_mac; 118 static int all_slaves_active; 119 static struct bond_params bonding_defaults; 120 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 121 static int packets_per_slave = 1; 122 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 123 124 module_param(max_bonds, int, 0); 125 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 126 module_param(tx_queues, int, 0); 127 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 128 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 129 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 130 "failover event (alias of num_unsol_na)"); 131 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 132 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 133 "failover event (alias of num_grat_arp)"); 134 module_param(miimon, int, 0); 135 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 136 module_param(updelay, int, 0); 137 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 138 module_param(downdelay, int, 0); 139 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 140 "in milliseconds"); 141 module_param(use_carrier, int, 0); 142 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 143 "0 for off, 1 for on (default)"); 144 module_param(mode, charp, 0); 145 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 146 "1 for active-backup, 2 for balance-xor, " 147 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 148 "6 for balance-alb"); 149 module_param(primary, charp, 0); 150 MODULE_PARM_DESC(primary, "Primary network device to use"); 151 module_param(primary_reselect, charp, 0); 152 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 153 "once it comes up; " 154 "0 for always (default), " 155 "1 for only if speed of primary is " 156 "better, " 157 "2 for only on active slave " 158 "failure"); 159 module_param(lacp_rate, charp, 0); 160 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 161 "0 for slow, 1 for fast"); 162 module_param(ad_select, charp, 0); 163 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 164 "0 for stable (default), 1 for bandwidth, " 165 "2 for count"); 166 module_param(min_links, int, 0); 167 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 168 169 module_param(xmit_hash_policy, charp, 0); 170 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 171 "0 for layer 2 (default), 1 for layer 3+4, " 172 "2 for layer 2+3, 3 for encap layer 2+3, " 173 "4 for encap layer 3+4, 5 for vlan+srcmac"); 174 module_param(arp_interval, int, 0); 175 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 176 module_param_array(arp_ip_target, charp, NULL, 0); 177 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 178 module_param(arp_validate, charp, 0); 179 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 180 "0 for none (default), 1 for active, " 181 "2 for backup, 3 for all"); 182 module_param(arp_all_targets, charp, 0); 183 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 184 module_param(fail_over_mac, charp, 0); 185 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 186 "the same MAC; 0 for none (default), " 187 "1 for active, 2 for follow"); 188 module_param(all_slaves_active, int, 0); 189 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 190 "by setting active flag for all slaves; " 191 "0 for never (default), 1 for always."); 192 module_param(resend_igmp, int, 0); 193 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 194 "link failure"); 195 module_param(packets_per_slave, int, 0); 196 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 197 "mode; 0 for a random slave, 1 packet per " 198 "slave (default), >1 packets per slave."); 199 module_param(lp_interval, uint, 0); 200 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 201 "the bonding driver sends learning packets to " 202 "each slaves peer switch. The default is 1."); 203 204 /*----------------------------- Global variables ----------------------------*/ 205 206 #ifdef CONFIG_NET_POLL_CONTROLLER 207 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 208 #endif 209 210 unsigned int bond_net_id __read_mostly; 211 212 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 213 { 214 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 215 .offset = offsetof(struct flow_keys, control), 216 }, 217 { 218 .key_id = FLOW_DISSECTOR_KEY_BASIC, 219 .offset = offsetof(struct flow_keys, basic), 220 }, 221 { 222 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 223 .offset = offsetof(struct flow_keys, addrs.v4addrs), 224 }, 225 { 226 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 227 .offset = offsetof(struct flow_keys, addrs.v6addrs), 228 }, 229 { 230 .key_id = FLOW_DISSECTOR_KEY_TIPC, 231 .offset = offsetof(struct flow_keys, addrs.tipckey), 232 }, 233 { 234 .key_id = FLOW_DISSECTOR_KEY_PORTS, 235 .offset = offsetof(struct flow_keys, ports), 236 }, 237 { 238 .key_id = FLOW_DISSECTOR_KEY_ICMP, 239 .offset = offsetof(struct flow_keys, icmp), 240 }, 241 { 242 .key_id = FLOW_DISSECTOR_KEY_VLAN, 243 .offset = offsetof(struct flow_keys, vlan), 244 }, 245 { 246 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 247 .offset = offsetof(struct flow_keys, tags), 248 }, 249 { 250 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 251 .offset = offsetof(struct flow_keys, keyid), 252 }, 253 }; 254 255 static struct flow_dissector flow_keys_bonding __read_mostly; 256 257 /*-------------------------- Forward declarations ---------------------------*/ 258 259 static int bond_init(struct net_device *bond_dev); 260 static void bond_uninit(struct net_device *bond_dev); 261 static void bond_get_stats(struct net_device *bond_dev, 262 struct rtnl_link_stats64 *stats); 263 static void bond_slave_arr_handler(struct work_struct *work); 264 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 265 int mod); 266 static void bond_netdev_notify_work(struct work_struct *work); 267 268 /*---------------------------- General routines -----------------------------*/ 269 270 const char *bond_mode_name(int mode) 271 { 272 static const char *names[] = { 273 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 274 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 275 [BOND_MODE_XOR] = "load balancing (xor)", 276 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 277 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 278 [BOND_MODE_TLB] = "transmit load balancing", 279 [BOND_MODE_ALB] = "adaptive load balancing", 280 }; 281 282 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 283 return "unknown"; 284 285 return names[mode]; 286 } 287 288 /** 289 * bond_dev_queue_xmit - Prepare skb for xmit. 290 * 291 * @bond: bond device that got this skb for tx. 292 * @skb: hw accel VLAN tagged skb to transmit 293 * @slave_dev: slave that is supposed to xmit this skbuff 294 */ 295 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 296 struct net_device *slave_dev) 297 { 298 skb->dev = slave_dev; 299 300 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 301 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 302 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 303 304 if (unlikely(netpoll_tx_running(bond->dev))) 305 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 306 307 return dev_queue_xmit(skb); 308 } 309 310 static bool bond_sk_check(struct bonding *bond) 311 { 312 switch (BOND_MODE(bond)) { 313 case BOND_MODE_8023AD: 314 case BOND_MODE_XOR: 315 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 316 return true; 317 fallthrough; 318 default: 319 return false; 320 } 321 } 322 323 static bool bond_xdp_check(struct bonding *bond) 324 { 325 switch (BOND_MODE(bond)) { 326 case BOND_MODE_ROUNDROBIN: 327 case BOND_MODE_ACTIVEBACKUP: 328 return true; 329 case BOND_MODE_8023AD: 330 case BOND_MODE_XOR: 331 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 332 * payload is not in the packet due to hardware offload. 333 */ 334 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 335 return true; 336 fallthrough; 337 default: 338 return false; 339 } 340 } 341 342 /*---------------------------------- VLAN -----------------------------------*/ 343 344 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 345 * We don't protect the slave list iteration with a lock because: 346 * a. This operation is performed in IOCTL context, 347 * b. The operation is protected by the RTNL semaphore in the 8021q code, 348 * c. Holding a lock with BH disabled while directly calling a base driver 349 * entry point is generally a BAD idea. 350 * 351 * The design of synchronization/protection for this operation in the 8021q 352 * module is good for one or more VLAN devices over a single physical device 353 * and cannot be extended for a teaming solution like bonding, so there is a 354 * potential race condition here where a net device from the vlan group might 355 * be referenced (either by a base driver or the 8021q code) while it is being 356 * removed from the system. However, it turns out we're not making matters 357 * worse, and if it works for regular VLAN usage it will work here too. 358 */ 359 360 /** 361 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 362 * @bond_dev: bonding net device that got called 363 * @proto: network protocol ID 364 * @vid: vlan id being added 365 */ 366 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 367 __be16 proto, u16 vid) 368 { 369 struct bonding *bond = netdev_priv(bond_dev); 370 struct slave *slave, *rollback_slave; 371 struct list_head *iter; 372 int res; 373 374 bond_for_each_slave(bond, slave, iter) { 375 res = vlan_vid_add(slave->dev, proto, vid); 376 if (res) 377 goto unwind; 378 } 379 380 return 0; 381 382 unwind: 383 /* unwind to the slave that failed */ 384 bond_for_each_slave(bond, rollback_slave, iter) { 385 if (rollback_slave == slave) 386 break; 387 388 vlan_vid_del(rollback_slave->dev, proto, vid); 389 } 390 391 return res; 392 } 393 394 /** 395 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 396 * @bond_dev: bonding net device that got called 397 * @proto: network protocol ID 398 * @vid: vlan id being removed 399 */ 400 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 401 __be16 proto, u16 vid) 402 { 403 struct bonding *bond = netdev_priv(bond_dev); 404 struct list_head *iter; 405 struct slave *slave; 406 407 bond_for_each_slave(bond, slave, iter) 408 vlan_vid_del(slave->dev, proto, vid); 409 410 if (bond_is_lb(bond)) 411 bond_alb_clear_vlan(bond, vid); 412 413 return 0; 414 } 415 416 /*---------------------------------- XFRM -----------------------------------*/ 417 418 #ifdef CONFIG_XFRM_OFFLOAD 419 /** 420 * bond_ipsec_add_sa - program device with a security association 421 * @xs: pointer to transformer state struct 422 **/ 423 static int bond_ipsec_add_sa(struct xfrm_state *xs) 424 { 425 struct net_device *bond_dev = xs->xso.dev; 426 struct bond_ipsec *ipsec; 427 struct bonding *bond; 428 struct slave *slave; 429 int err; 430 431 if (!bond_dev) 432 return -EINVAL; 433 434 rcu_read_lock(); 435 bond = netdev_priv(bond_dev); 436 slave = rcu_dereference(bond->curr_active_slave); 437 if (!slave) { 438 rcu_read_unlock(); 439 return -ENODEV; 440 } 441 442 if (!slave->dev->xfrmdev_ops || 443 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 444 netif_is_bond_master(slave->dev)) { 445 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n"); 446 rcu_read_unlock(); 447 return -EINVAL; 448 } 449 450 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC); 451 if (!ipsec) { 452 rcu_read_unlock(); 453 return -ENOMEM; 454 } 455 xs->xso.real_dev = slave->dev; 456 457 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs); 458 if (!err) { 459 ipsec->xs = xs; 460 INIT_LIST_HEAD(&ipsec->list); 461 spin_lock_bh(&bond->ipsec_lock); 462 list_add(&ipsec->list, &bond->ipsec_list); 463 spin_unlock_bh(&bond->ipsec_lock); 464 } else { 465 kfree(ipsec); 466 } 467 rcu_read_unlock(); 468 return err; 469 } 470 471 static void bond_ipsec_add_sa_all(struct bonding *bond) 472 { 473 struct net_device *bond_dev = bond->dev; 474 struct bond_ipsec *ipsec; 475 struct slave *slave; 476 477 rcu_read_lock(); 478 slave = rcu_dereference(bond->curr_active_slave); 479 if (!slave) 480 goto out; 481 482 if (!slave->dev->xfrmdev_ops || 483 !slave->dev->xfrmdev_ops->xdo_dev_state_add || 484 netif_is_bond_master(slave->dev)) { 485 spin_lock_bh(&bond->ipsec_lock); 486 if (!list_empty(&bond->ipsec_list)) 487 slave_warn(bond_dev, slave->dev, 488 "%s: no slave xdo_dev_state_add\n", 489 __func__); 490 spin_unlock_bh(&bond->ipsec_lock); 491 goto out; 492 } 493 494 spin_lock_bh(&bond->ipsec_lock); 495 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 496 ipsec->xs->xso.real_dev = slave->dev; 497 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) { 498 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__); 499 ipsec->xs->xso.real_dev = NULL; 500 } 501 } 502 spin_unlock_bh(&bond->ipsec_lock); 503 out: 504 rcu_read_unlock(); 505 } 506 507 /** 508 * bond_ipsec_del_sa - clear out this specific SA 509 * @xs: pointer to transformer state struct 510 **/ 511 static void bond_ipsec_del_sa(struct xfrm_state *xs) 512 { 513 struct net_device *bond_dev = xs->xso.dev; 514 struct bond_ipsec *ipsec; 515 struct bonding *bond; 516 struct slave *slave; 517 518 if (!bond_dev) 519 return; 520 521 rcu_read_lock(); 522 bond = netdev_priv(bond_dev); 523 slave = rcu_dereference(bond->curr_active_slave); 524 525 if (!slave) 526 goto out; 527 528 if (!xs->xso.real_dev) 529 goto out; 530 531 WARN_ON(xs->xso.real_dev != slave->dev); 532 533 if (!slave->dev->xfrmdev_ops || 534 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 535 netif_is_bond_master(slave->dev)) { 536 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__); 537 goto out; 538 } 539 540 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs); 541 out: 542 spin_lock_bh(&bond->ipsec_lock); 543 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 544 if (ipsec->xs == xs) { 545 list_del(&ipsec->list); 546 kfree(ipsec); 547 break; 548 } 549 } 550 spin_unlock_bh(&bond->ipsec_lock); 551 rcu_read_unlock(); 552 } 553 554 static void bond_ipsec_del_sa_all(struct bonding *bond) 555 { 556 struct net_device *bond_dev = bond->dev; 557 struct bond_ipsec *ipsec; 558 struct slave *slave; 559 560 rcu_read_lock(); 561 slave = rcu_dereference(bond->curr_active_slave); 562 if (!slave) { 563 rcu_read_unlock(); 564 return; 565 } 566 567 spin_lock_bh(&bond->ipsec_lock); 568 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 569 if (!ipsec->xs->xso.real_dev) 570 continue; 571 572 if (!slave->dev->xfrmdev_ops || 573 !slave->dev->xfrmdev_ops->xdo_dev_state_delete || 574 netif_is_bond_master(slave->dev)) { 575 slave_warn(bond_dev, slave->dev, 576 "%s: no slave xdo_dev_state_delete\n", 577 __func__); 578 } else { 579 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 580 } 581 ipsec->xs->xso.real_dev = NULL; 582 } 583 spin_unlock_bh(&bond->ipsec_lock); 584 rcu_read_unlock(); 585 } 586 587 /** 588 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 589 * @skb: current data packet 590 * @xs: pointer to transformer state struct 591 **/ 592 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 593 { 594 struct net_device *bond_dev = xs->xso.dev; 595 struct net_device *real_dev; 596 struct slave *curr_active; 597 struct bonding *bond; 598 int err; 599 600 bond = netdev_priv(bond_dev); 601 rcu_read_lock(); 602 curr_active = rcu_dereference(bond->curr_active_slave); 603 real_dev = curr_active->dev; 604 605 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 606 err = false; 607 goto out; 608 } 609 610 if (!xs->xso.real_dev) { 611 err = false; 612 goto out; 613 } 614 615 if (!real_dev->xfrmdev_ops || 616 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 617 netif_is_bond_master(real_dev)) { 618 err = false; 619 goto out; 620 } 621 622 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 623 out: 624 rcu_read_unlock(); 625 return err; 626 } 627 628 static const struct xfrmdev_ops bond_xfrmdev_ops = { 629 .xdo_dev_state_add = bond_ipsec_add_sa, 630 .xdo_dev_state_delete = bond_ipsec_del_sa, 631 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 632 }; 633 #endif /* CONFIG_XFRM_OFFLOAD */ 634 635 /*------------------------------- Link status -------------------------------*/ 636 637 /* Set the carrier state for the master according to the state of its 638 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 639 * do special 802.3ad magic. 640 * 641 * Returns zero if carrier state does not change, nonzero if it does. 642 */ 643 int bond_set_carrier(struct bonding *bond) 644 { 645 struct list_head *iter; 646 struct slave *slave; 647 648 if (!bond_has_slaves(bond)) 649 goto down; 650 651 if (BOND_MODE(bond) == BOND_MODE_8023AD) 652 return bond_3ad_set_carrier(bond); 653 654 bond_for_each_slave(bond, slave, iter) { 655 if (slave->link == BOND_LINK_UP) { 656 if (!netif_carrier_ok(bond->dev)) { 657 netif_carrier_on(bond->dev); 658 return 1; 659 } 660 return 0; 661 } 662 } 663 664 down: 665 if (netif_carrier_ok(bond->dev)) { 666 netif_carrier_off(bond->dev); 667 return 1; 668 } 669 return 0; 670 } 671 672 /* Get link speed and duplex from the slave's base driver 673 * using ethtool. If for some reason the call fails or the 674 * values are invalid, set speed and duplex to -1, 675 * and return. Return 1 if speed or duplex settings are 676 * UNKNOWN; 0 otherwise. 677 */ 678 static int bond_update_speed_duplex(struct slave *slave) 679 { 680 struct net_device *slave_dev = slave->dev; 681 struct ethtool_link_ksettings ecmd; 682 int res; 683 684 slave->speed = SPEED_UNKNOWN; 685 slave->duplex = DUPLEX_UNKNOWN; 686 687 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 688 if (res < 0) 689 return 1; 690 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 691 return 1; 692 switch (ecmd.base.duplex) { 693 case DUPLEX_FULL: 694 case DUPLEX_HALF: 695 break; 696 default: 697 return 1; 698 } 699 700 slave->speed = ecmd.base.speed; 701 slave->duplex = ecmd.base.duplex; 702 703 return 0; 704 } 705 706 const char *bond_slave_link_status(s8 link) 707 { 708 switch (link) { 709 case BOND_LINK_UP: 710 return "up"; 711 case BOND_LINK_FAIL: 712 return "going down"; 713 case BOND_LINK_DOWN: 714 return "down"; 715 case BOND_LINK_BACK: 716 return "going back"; 717 default: 718 return "unknown"; 719 } 720 } 721 722 /* if <dev> supports MII link status reporting, check its link status. 723 * 724 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 725 * depending upon the setting of the use_carrier parameter. 726 * 727 * Return either BMSR_LSTATUS, meaning that the link is up (or we 728 * can't tell and just pretend it is), or 0, meaning that the link is 729 * down. 730 * 731 * If reporting is non-zero, instead of faking link up, return -1 if 732 * both ETHTOOL and MII ioctls fail (meaning the device does not 733 * support them). If use_carrier is set, return whatever it says. 734 * It'd be nice if there was a good way to tell if a driver supports 735 * netif_carrier, but there really isn't. 736 */ 737 static int bond_check_dev_link(struct bonding *bond, 738 struct net_device *slave_dev, int reporting) 739 { 740 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 741 int (*ioctl)(struct net_device *, struct ifreq *, int); 742 struct ifreq ifr; 743 struct mii_ioctl_data *mii; 744 745 if (!reporting && !netif_running(slave_dev)) 746 return 0; 747 748 if (bond->params.use_carrier) 749 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 750 751 /* Try to get link status using Ethtool first. */ 752 if (slave_dev->ethtool_ops->get_link) 753 return slave_dev->ethtool_ops->get_link(slave_dev) ? 754 BMSR_LSTATUS : 0; 755 756 /* Ethtool can't be used, fallback to MII ioctls. */ 757 ioctl = slave_ops->ndo_eth_ioctl; 758 if (ioctl) { 759 /* TODO: set pointer to correct ioctl on a per team member 760 * bases to make this more efficient. that is, once 761 * we determine the correct ioctl, we will always 762 * call it and not the others for that team 763 * member. 764 */ 765 766 /* We cannot assume that SIOCGMIIPHY will also read a 767 * register; not all network drivers (e.g., e100) 768 * support that. 769 */ 770 771 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 772 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 773 mii = if_mii(&ifr); 774 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 775 mii->reg_num = MII_BMSR; 776 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 777 return mii->val_out & BMSR_LSTATUS; 778 } 779 } 780 781 /* If reporting, report that either there's no ndo_eth_ioctl, 782 * or both SIOCGMIIREG and get_link failed (meaning that we 783 * cannot report link status). If not reporting, pretend 784 * we're ok. 785 */ 786 return reporting ? -1 : BMSR_LSTATUS; 787 } 788 789 /*----------------------------- Multicast list ------------------------------*/ 790 791 /* Push the promiscuity flag down to appropriate slaves */ 792 static int bond_set_promiscuity(struct bonding *bond, int inc) 793 { 794 struct list_head *iter; 795 int err = 0; 796 797 if (bond_uses_primary(bond)) { 798 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 799 800 if (curr_active) 801 err = dev_set_promiscuity(curr_active->dev, inc); 802 } else { 803 struct slave *slave; 804 805 bond_for_each_slave(bond, slave, iter) { 806 err = dev_set_promiscuity(slave->dev, inc); 807 if (err) 808 return err; 809 } 810 } 811 return err; 812 } 813 814 /* Push the allmulti flag down to all slaves */ 815 static int bond_set_allmulti(struct bonding *bond, int inc) 816 { 817 struct list_head *iter; 818 int err = 0; 819 820 if (bond_uses_primary(bond)) { 821 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 822 823 if (curr_active) 824 err = dev_set_allmulti(curr_active->dev, inc); 825 } else { 826 struct slave *slave; 827 828 bond_for_each_slave(bond, slave, iter) { 829 err = dev_set_allmulti(slave->dev, inc); 830 if (err) 831 return err; 832 } 833 } 834 return err; 835 } 836 837 /* Retrieve the list of registered multicast addresses for the bonding 838 * device and retransmit an IGMP JOIN request to the current active 839 * slave. 840 */ 841 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 842 { 843 struct bonding *bond = container_of(work, struct bonding, 844 mcast_work.work); 845 846 if (!rtnl_trylock()) { 847 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 848 return; 849 } 850 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 851 852 if (bond->igmp_retrans > 1) { 853 bond->igmp_retrans--; 854 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 855 } 856 rtnl_unlock(); 857 } 858 859 /* Flush bond's hardware addresses from slave */ 860 static void bond_hw_addr_flush(struct net_device *bond_dev, 861 struct net_device *slave_dev) 862 { 863 struct bonding *bond = netdev_priv(bond_dev); 864 865 dev_uc_unsync(slave_dev, bond_dev); 866 dev_mc_unsync(slave_dev, bond_dev); 867 868 if (BOND_MODE(bond) == BOND_MODE_8023AD) 869 dev_mc_del(slave_dev, lacpdu_mcast_addr); 870 } 871 872 /*--------------------------- Active slave change ---------------------------*/ 873 874 /* Update the hardware address list and promisc/allmulti for the new and 875 * old active slaves (if any). Modes that are not using primary keep all 876 * slaves up date at all times; only the modes that use primary need to call 877 * this function to swap these settings during a failover. 878 */ 879 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 880 struct slave *old_active) 881 { 882 if (old_active) { 883 if (bond->dev->flags & IFF_PROMISC) 884 dev_set_promiscuity(old_active->dev, -1); 885 886 if (bond->dev->flags & IFF_ALLMULTI) 887 dev_set_allmulti(old_active->dev, -1); 888 889 if (bond->dev->flags & IFF_UP) 890 bond_hw_addr_flush(bond->dev, old_active->dev); 891 } 892 893 if (new_active) { 894 /* FIXME: Signal errors upstream. */ 895 if (bond->dev->flags & IFF_PROMISC) 896 dev_set_promiscuity(new_active->dev, 1); 897 898 if (bond->dev->flags & IFF_ALLMULTI) 899 dev_set_allmulti(new_active->dev, 1); 900 901 if (bond->dev->flags & IFF_UP) { 902 netif_addr_lock_bh(bond->dev); 903 dev_uc_sync(new_active->dev, bond->dev); 904 dev_mc_sync(new_active->dev, bond->dev); 905 netif_addr_unlock_bh(bond->dev); 906 } 907 } 908 } 909 910 /** 911 * bond_set_dev_addr - clone slave's address to bond 912 * @bond_dev: bond net device 913 * @slave_dev: slave net device 914 * 915 * Should be called with RTNL held. 916 */ 917 static int bond_set_dev_addr(struct net_device *bond_dev, 918 struct net_device *slave_dev) 919 { 920 int err; 921 922 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 923 bond_dev, slave_dev, slave_dev->addr_len); 924 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 925 if (err) 926 return err; 927 928 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 929 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 930 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 931 return 0; 932 } 933 934 static struct slave *bond_get_old_active(struct bonding *bond, 935 struct slave *new_active) 936 { 937 struct slave *slave; 938 struct list_head *iter; 939 940 bond_for_each_slave(bond, slave, iter) { 941 if (slave == new_active) 942 continue; 943 944 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 945 return slave; 946 } 947 948 return NULL; 949 } 950 951 /* bond_do_fail_over_mac 952 * 953 * Perform special MAC address swapping for fail_over_mac settings 954 * 955 * Called with RTNL 956 */ 957 static void bond_do_fail_over_mac(struct bonding *bond, 958 struct slave *new_active, 959 struct slave *old_active) 960 { 961 u8 tmp_mac[MAX_ADDR_LEN]; 962 struct sockaddr_storage ss; 963 int rv; 964 965 switch (bond->params.fail_over_mac) { 966 case BOND_FOM_ACTIVE: 967 if (new_active) { 968 rv = bond_set_dev_addr(bond->dev, new_active->dev); 969 if (rv) 970 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 971 -rv); 972 } 973 break; 974 case BOND_FOM_FOLLOW: 975 /* if new_active && old_active, swap them 976 * if just old_active, do nothing (going to no active slave) 977 * if just new_active, set new_active to bond's MAC 978 */ 979 if (!new_active) 980 return; 981 982 if (!old_active) 983 old_active = bond_get_old_active(bond, new_active); 984 985 if (old_active) { 986 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 987 new_active->dev->addr_len); 988 bond_hw_addr_copy(ss.__data, 989 old_active->dev->dev_addr, 990 old_active->dev->addr_len); 991 ss.ss_family = new_active->dev->type; 992 } else { 993 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 994 bond->dev->addr_len); 995 ss.ss_family = bond->dev->type; 996 } 997 998 rv = dev_set_mac_address(new_active->dev, 999 (struct sockaddr *)&ss, NULL); 1000 if (rv) { 1001 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1002 -rv); 1003 goto out; 1004 } 1005 1006 if (!old_active) 1007 goto out; 1008 1009 bond_hw_addr_copy(ss.__data, tmp_mac, 1010 new_active->dev->addr_len); 1011 ss.ss_family = old_active->dev->type; 1012 1013 rv = dev_set_mac_address(old_active->dev, 1014 (struct sockaddr *)&ss, NULL); 1015 if (rv) 1016 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1017 -rv); 1018 out: 1019 break; 1020 default: 1021 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1022 bond->params.fail_over_mac); 1023 break; 1024 } 1025 1026 } 1027 1028 /** 1029 * bond_choose_primary_or_current - select the primary or high priority slave 1030 * @bond: our bonding struct 1031 * 1032 * - Check if there is a primary link. If the primary link was set and is up, 1033 * go on and do link reselection. 1034 * 1035 * - If primary link is not set or down, find the highest priority link. 1036 * If the highest priority link is not current slave, set it as primary 1037 * link and do link reselection. 1038 */ 1039 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1040 { 1041 struct slave *prim = rtnl_dereference(bond->primary_slave); 1042 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1043 struct slave *slave, *hprio = NULL; 1044 struct list_head *iter; 1045 1046 if (!prim || prim->link != BOND_LINK_UP) { 1047 bond_for_each_slave(bond, slave, iter) { 1048 if (slave->link == BOND_LINK_UP) { 1049 hprio = hprio ?: slave; 1050 if (slave->prio > hprio->prio) 1051 hprio = slave; 1052 } 1053 } 1054 1055 if (hprio && hprio != curr) { 1056 prim = hprio; 1057 goto link_reselect; 1058 } 1059 1060 if (!curr || curr->link != BOND_LINK_UP) 1061 return NULL; 1062 return curr; 1063 } 1064 1065 if (bond->force_primary) { 1066 bond->force_primary = false; 1067 return prim; 1068 } 1069 1070 link_reselect: 1071 if (!curr || curr->link != BOND_LINK_UP) 1072 return prim; 1073 1074 /* At this point, prim and curr are both up */ 1075 switch (bond->params.primary_reselect) { 1076 case BOND_PRI_RESELECT_ALWAYS: 1077 return prim; 1078 case BOND_PRI_RESELECT_BETTER: 1079 if (prim->speed < curr->speed) 1080 return curr; 1081 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1082 return curr; 1083 return prim; 1084 case BOND_PRI_RESELECT_FAILURE: 1085 return curr; 1086 default: 1087 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1088 bond->params.primary_reselect); 1089 return curr; 1090 } 1091 } 1092 1093 /** 1094 * bond_find_best_slave - select the best available slave to be the active one 1095 * @bond: our bonding struct 1096 */ 1097 static struct slave *bond_find_best_slave(struct bonding *bond) 1098 { 1099 struct slave *slave, *bestslave = NULL; 1100 struct list_head *iter; 1101 int mintime = bond->params.updelay; 1102 1103 slave = bond_choose_primary_or_current(bond); 1104 if (slave) 1105 return slave; 1106 1107 bond_for_each_slave(bond, slave, iter) { 1108 if (slave->link == BOND_LINK_UP) 1109 return slave; 1110 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1111 slave->delay < mintime) { 1112 mintime = slave->delay; 1113 bestslave = slave; 1114 } 1115 } 1116 1117 return bestslave; 1118 } 1119 1120 static bool bond_should_notify_peers(struct bonding *bond) 1121 { 1122 struct slave *slave; 1123 1124 rcu_read_lock(); 1125 slave = rcu_dereference(bond->curr_active_slave); 1126 rcu_read_unlock(); 1127 1128 if (!slave || !bond->send_peer_notif || 1129 bond->send_peer_notif % 1130 max(1, bond->params.peer_notif_delay) != 0 || 1131 !netif_carrier_ok(bond->dev) || 1132 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1133 return false; 1134 1135 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1136 slave ? slave->dev->name : "NULL"); 1137 1138 return true; 1139 } 1140 1141 /** 1142 * bond_change_active_slave - change the active slave into the specified one 1143 * @bond: our bonding struct 1144 * @new_active: the new slave to make the active one 1145 * 1146 * Set the new slave to the bond's settings and unset them on the old 1147 * curr_active_slave. 1148 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1149 * 1150 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1151 * because it is apparently the best available slave we have, even though its 1152 * updelay hasn't timed out yet. 1153 * 1154 * Caller must hold RTNL. 1155 */ 1156 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1157 { 1158 struct slave *old_active; 1159 1160 ASSERT_RTNL(); 1161 1162 old_active = rtnl_dereference(bond->curr_active_slave); 1163 1164 if (old_active == new_active) 1165 return; 1166 1167 #ifdef CONFIG_XFRM_OFFLOAD 1168 bond_ipsec_del_sa_all(bond); 1169 #endif /* CONFIG_XFRM_OFFLOAD */ 1170 1171 if (new_active) { 1172 new_active->last_link_up = jiffies; 1173 1174 if (new_active->link == BOND_LINK_BACK) { 1175 if (bond_uses_primary(bond)) { 1176 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1177 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1178 } 1179 1180 new_active->delay = 0; 1181 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1182 BOND_SLAVE_NOTIFY_NOW); 1183 1184 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1185 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1186 1187 if (bond_is_lb(bond)) 1188 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1189 } else { 1190 if (bond_uses_primary(bond)) 1191 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1192 } 1193 } 1194 1195 if (bond_uses_primary(bond)) 1196 bond_hw_addr_swap(bond, new_active, old_active); 1197 1198 if (bond_is_lb(bond)) { 1199 bond_alb_handle_active_change(bond, new_active); 1200 if (old_active) 1201 bond_set_slave_inactive_flags(old_active, 1202 BOND_SLAVE_NOTIFY_NOW); 1203 if (new_active) 1204 bond_set_slave_active_flags(new_active, 1205 BOND_SLAVE_NOTIFY_NOW); 1206 } else { 1207 rcu_assign_pointer(bond->curr_active_slave, new_active); 1208 } 1209 1210 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1211 if (old_active) 1212 bond_set_slave_inactive_flags(old_active, 1213 BOND_SLAVE_NOTIFY_NOW); 1214 1215 if (new_active) { 1216 bool should_notify_peers = false; 1217 1218 bond_set_slave_active_flags(new_active, 1219 BOND_SLAVE_NOTIFY_NOW); 1220 1221 if (bond->params.fail_over_mac) 1222 bond_do_fail_over_mac(bond, new_active, 1223 old_active); 1224 1225 if (netif_running(bond->dev)) { 1226 bond->send_peer_notif = 1227 bond->params.num_peer_notif * 1228 max(1, bond->params.peer_notif_delay); 1229 should_notify_peers = 1230 bond_should_notify_peers(bond); 1231 } 1232 1233 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1234 if (should_notify_peers) { 1235 bond->send_peer_notif--; 1236 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1237 bond->dev); 1238 } 1239 } 1240 } 1241 1242 #ifdef CONFIG_XFRM_OFFLOAD 1243 bond_ipsec_add_sa_all(bond); 1244 #endif /* CONFIG_XFRM_OFFLOAD */ 1245 1246 /* resend IGMP joins since active slave has changed or 1247 * all were sent on curr_active_slave. 1248 * resend only if bond is brought up with the affected 1249 * bonding modes and the retransmission is enabled 1250 */ 1251 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1252 ((bond_uses_primary(bond) && new_active) || 1253 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1254 bond->igmp_retrans = bond->params.resend_igmp; 1255 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1256 } 1257 } 1258 1259 /** 1260 * bond_select_active_slave - select a new active slave, if needed 1261 * @bond: our bonding struct 1262 * 1263 * This functions should be called when one of the following occurs: 1264 * - The old curr_active_slave has been released or lost its link. 1265 * - The primary_slave has got its link back. 1266 * - A slave has got its link back and there's no old curr_active_slave. 1267 * 1268 * Caller must hold RTNL. 1269 */ 1270 void bond_select_active_slave(struct bonding *bond) 1271 { 1272 struct slave *best_slave; 1273 int rv; 1274 1275 ASSERT_RTNL(); 1276 1277 best_slave = bond_find_best_slave(bond); 1278 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1279 bond_change_active_slave(bond, best_slave); 1280 rv = bond_set_carrier(bond); 1281 if (!rv) 1282 return; 1283 1284 if (netif_carrier_ok(bond->dev)) 1285 netdev_info(bond->dev, "active interface up!\n"); 1286 else 1287 netdev_info(bond->dev, "now running without any active interface!\n"); 1288 } 1289 } 1290 1291 #ifdef CONFIG_NET_POLL_CONTROLLER 1292 static inline int slave_enable_netpoll(struct slave *slave) 1293 { 1294 struct netpoll *np; 1295 int err = 0; 1296 1297 np = kzalloc(sizeof(*np), GFP_KERNEL); 1298 err = -ENOMEM; 1299 if (!np) 1300 goto out; 1301 1302 err = __netpoll_setup(np, slave->dev); 1303 if (err) { 1304 kfree(np); 1305 goto out; 1306 } 1307 slave->np = np; 1308 out: 1309 return err; 1310 } 1311 static inline void slave_disable_netpoll(struct slave *slave) 1312 { 1313 struct netpoll *np = slave->np; 1314 1315 if (!np) 1316 return; 1317 1318 slave->np = NULL; 1319 1320 __netpoll_free(np); 1321 } 1322 1323 static void bond_poll_controller(struct net_device *bond_dev) 1324 { 1325 struct bonding *bond = netdev_priv(bond_dev); 1326 struct slave *slave = NULL; 1327 struct list_head *iter; 1328 struct ad_info ad_info; 1329 1330 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1331 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1332 return; 1333 1334 bond_for_each_slave_rcu(bond, slave, iter) { 1335 if (!bond_slave_is_up(slave)) 1336 continue; 1337 1338 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1339 struct aggregator *agg = 1340 SLAVE_AD_INFO(slave)->port.aggregator; 1341 1342 if (agg && 1343 agg->aggregator_identifier != ad_info.aggregator_id) 1344 continue; 1345 } 1346 1347 netpoll_poll_dev(slave->dev); 1348 } 1349 } 1350 1351 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1352 { 1353 struct bonding *bond = netdev_priv(bond_dev); 1354 struct list_head *iter; 1355 struct slave *slave; 1356 1357 bond_for_each_slave(bond, slave, iter) 1358 if (bond_slave_is_up(slave)) 1359 slave_disable_netpoll(slave); 1360 } 1361 1362 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1363 { 1364 struct bonding *bond = netdev_priv(dev); 1365 struct list_head *iter; 1366 struct slave *slave; 1367 int err = 0; 1368 1369 bond_for_each_slave(bond, slave, iter) { 1370 err = slave_enable_netpoll(slave); 1371 if (err) { 1372 bond_netpoll_cleanup(dev); 1373 break; 1374 } 1375 } 1376 return err; 1377 } 1378 #else 1379 static inline int slave_enable_netpoll(struct slave *slave) 1380 { 1381 return 0; 1382 } 1383 static inline void slave_disable_netpoll(struct slave *slave) 1384 { 1385 } 1386 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1387 { 1388 } 1389 #endif 1390 1391 /*---------------------------------- IOCTL ----------------------------------*/ 1392 1393 static netdev_features_t bond_fix_features(struct net_device *dev, 1394 netdev_features_t features) 1395 { 1396 struct bonding *bond = netdev_priv(dev); 1397 struct list_head *iter; 1398 netdev_features_t mask; 1399 struct slave *slave; 1400 1401 mask = features; 1402 1403 features &= ~NETIF_F_ONE_FOR_ALL; 1404 features |= NETIF_F_ALL_FOR_ALL; 1405 1406 bond_for_each_slave(bond, slave, iter) { 1407 features = netdev_increment_features(features, 1408 slave->dev->features, 1409 mask); 1410 } 1411 features = netdev_add_tso_features(features, mask); 1412 1413 return features; 1414 } 1415 1416 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1417 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1418 NETIF_F_HIGHDMA | NETIF_F_LRO) 1419 1420 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1421 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1422 1423 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1424 NETIF_F_GSO_SOFTWARE) 1425 1426 1427 static void bond_compute_features(struct bonding *bond) 1428 { 1429 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1430 IFF_XMIT_DST_RELEASE_PERM; 1431 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1432 netdev_features_t enc_features = BOND_ENC_FEATURES; 1433 #ifdef CONFIG_XFRM_OFFLOAD 1434 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1435 #endif /* CONFIG_XFRM_OFFLOAD */ 1436 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1437 struct net_device *bond_dev = bond->dev; 1438 struct list_head *iter; 1439 struct slave *slave; 1440 unsigned short max_hard_header_len = ETH_HLEN; 1441 unsigned int tso_max_size = TSO_MAX_SIZE; 1442 u16 tso_max_segs = TSO_MAX_SEGS; 1443 1444 if (!bond_has_slaves(bond)) 1445 goto done; 1446 vlan_features &= NETIF_F_ALL_FOR_ALL; 1447 mpls_features &= NETIF_F_ALL_FOR_ALL; 1448 1449 bond_for_each_slave(bond, slave, iter) { 1450 vlan_features = netdev_increment_features(vlan_features, 1451 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1452 1453 enc_features = netdev_increment_features(enc_features, 1454 slave->dev->hw_enc_features, 1455 BOND_ENC_FEATURES); 1456 1457 #ifdef CONFIG_XFRM_OFFLOAD 1458 xfrm_features = netdev_increment_features(xfrm_features, 1459 slave->dev->hw_enc_features, 1460 BOND_XFRM_FEATURES); 1461 #endif /* CONFIG_XFRM_OFFLOAD */ 1462 1463 mpls_features = netdev_increment_features(mpls_features, 1464 slave->dev->mpls_features, 1465 BOND_MPLS_FEATURES); 1466 1467 dst_release_flag &= slave->dev->priv_flags; 1468 if (slave->dev->hard_header_len > max_hard_header_len) 1469 max_hard_header_len = slave->dev->hard_header_len; 1470 1471 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1472 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1473 } 1474 bond_dev->hard_header_len = max_hard_header_len; 1475 1476 done: 1477 bond_dev->vlan_features = vlan_features; 1478 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1479 NETIF_F_HW_VLAN_CTAG_TX | 1480 NETIF_F_HW_VLAN_STAG_TX; 1481 #ifdef CONFIG_XFRM_OFFLOAD 1482 bond_dev->hw_enc_features |= xfrm_features; 1483 #endif /* CONFIG_XFRM_OFFLOAD */ 1484 bond_dev->mpls_features = mpls_features; 1485 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1486 netif_set_tso_max_size(bond_dev, tso_max_size); 1487 1488 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1489 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1490 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1491 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1492 1493 netdev_change_features(bond_dev); 1494 } 1495 1496 static void bond_setup_by_slave(struct net_device *bond_dev, 1497 struct net_device *slave_dev) 1498 { 1499 bond_dev->header_ops = slave_dev->header_ops; 1500 1501 bond_dev->type = slave_dev->type; 1502 bond_dev->hard_header_len = slave_dev->hard_header_len; 1503 bond_dev->needed_headroom = slave_dev->needed_headroom; 1504 bond_dev->addr_len = slave_dev->addr_len; 1505 1506 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1507 slave_dev->addr_len); 1508 } 1509 1510 /* On bonding slaves other than the currently active slave, suppress 1511 * duplicates except for alb non-mcast/bcast. 1512 */ 1513 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1514 struct slave *slave, 1515 struct bonding *bond) 1516 { 1517 if (bond_is_slave_inactive(slave)) { 1518 if (BOND_MODE(bond) == BOND_MODE_ALB && 1519 skb->pkt_type != PACKET_BROADCAST && 1520 skb->pkt_type != PACKET_MULTICAST) 1521 return false; 1522 return true; 1523 } 1524 return false; 1525 } 1526 1527 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1528 { 1529 struct sk_buff *skb = *pskb; 1530 struct slave *slave; 1531 struct bonding *bond; 1532 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1533 struct slave *); 1534 int ret = RX_HANDLER_ANOTHER; 1535 1536 skb = skb_share_check(skb, GFP_ATOMIC); 1537 if (unlikely(!skb)) 1538 return RX_HANDLER_CONSUMED; 1539 1540 *pskb = skb; 1541 1542 slave = bond_slave_get_rcu(skb->dev); 1543 bond = slave->bond; 1544 1545 recv_probe = READ_ONCE(bond->recv_probe); 1546 if (recv_probe) { 1547 ret = recv_probe(skb, bond, slave); 1548 if (ret == RX_HANDLER_CONSUMED) { 1549 consume_skb(skb); 1550 return ret; 1551 } 1552 } 1553 1554 /* 1555 * For packets determined by bond_should_deliver_exact_match() call to 1556 * be suppressed we want to make an exception for link-local packets. 1557 * This is necessary for e.g. LLDP daemons to be able to monitor 1558 * inactive slave links without being forced to bind to them 1559 * explicitly. 1560 * 1561 * At the same time, packets that are passed to the bonding master 1562 * (including link-local ones) can have their originating interface 1563 * determined via PACKET_ORIGDEV socket option. 1564 */ 1565 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1566 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1567 return RX_HANDLER_PASS; 1568 return RX_HANDLER_EXACT; 1569 } 1570 1571 skb->dev = bond->dev; 1572 1573 if (BOND_MODE(bond) == BOND_MODE_ALB && 1574 netif_is_bridge_port(bond->dev) && 1575 skb->pkt_type == PACKET_HOST) { 1576 1577 if (unlikely(skb_cow_head(skb, 1578 skb->data - skb_mac_header(skb)))) { 1579 kfree_skb(skb); 1580 return RX_HANDLER_CONSUMED; 1581 } 1582 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1583 bond->dev->addr_len); 1584 } 1585 1586 return ret; 1587 } 1588 1589 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1590 { 1591 switch (BOND_MODE(bond)) { 1592 case BOND_MODE_ROUNDROBIN: 1593 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1594 case BOND_MODE_ACTIVEBACKUP: 1595 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1596 case BOND_MODE_BROADCAST: 1597 return NETDEV_LAG_TX_TYPE_BROADCAST; 1598 case BOND_MODE_XOR: 1599 case BOND_MODE_8023AD: 1600 return NETDEV_LAG_TX_TYPE_HASH; 1601 default: 1602 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1603 } 1604 } 1605 1606 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1607 enum netdev_lag_tx_type type) 1608 { 1609 if (type != NETDEV_LAG_TX_TYPE_HASH) 1610 return NETDEV_LAG_HASH_NONE; 1611 1612 switch (bond->params.xmit_policy) { 1613 case BOND_XMIT_POLICY_LAYER2: 1614 return NETDEV_LAG_HASH_L2; 1615 case BOND_XMIT_POLICY_LAYER34: 1616 return NETDEV_LAG_HASH_L34; 1617 case BOND_XMIT_POLICY_LAYER23: 1618 return NETDEV_LAG_HASH_L23; 1619 case BOND_XMIT_POLICY_ENCAP23: 1620 return NETDEV_LAG_HASH_E23; 1621 case BOND_XMIT_POLICY_ENCAP34: 1622 return NETDEV_LAG_HASH_E34; 1623 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1624 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1625 default: 1626 return NETDEV_LAG_HASH_UNKNOWN; 1627 } 1628 } 1629 1630 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1631 struct netlink_ext_ack *extack) 1632 { 1633 struct netdev_lag_upper_info lag_upper_info; 1634 enum netdev_lag_tx_type type; 1635 1636 type = bond_lag_tx_type(bond); 1637 lag_upper_info.tx_type = type; 1638 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1639 1640 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1641 &lag_upper_info, extack); 1642 } 1643 1644 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1645 { 1646 netdev_upper_dev_unlink(slave->dev, bond->dev); 1647 slave->dev->flags &= ~IFF_SLAVE; 1648 } 1649 1650 static void slave_kobj_release(struct kobject *kobj) 1651 { 1652 struct slave *slave = to_slave(kobj); 1653 struct bonding *bond = bond_get_bond_by_slave(slave); 1654 1655 cancel_delayed_work_sync(&slave->notify_work); 1656 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1657 kfree(SLAVE_AD_INFO(slave)); 1658 1659 kfree(slave); 1660 } 1661 1662 static struct kobj_type slave_ktype = { 1663 .release = slave_kobj_release, 1664 #ifdef CONFIG_SYSFS 1665 .sysfs_ops = &slave_sysfs_ops, 1666 #endif 1667 }; 1668 1669 static int bond_kobj_init(struct slave *slave) 1670 { 1671 int err; 1672 1673 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1674 &(slave->dev->dev.kobj), "bonding_slave"); 1675 if (err) 1676 kobject_put(&slave->kobj); 1677 1678 return err; 1679 } 1680 1681 static struct slave *bond_alloc_slave(struct bonding *bond, 1682 struct net_device *slave_dev) 1683 { 1684 struct slave *slave = NULL; 1685 1686 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1687 if (!slave) 1688 return NULL; 1689 1690 slave->bond = bond; 1691 slave->dev = slave_dev; 1692 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1693 1694 if (bond_kobj_init(slave)) 1695 return NULL; 1696 1697 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1698 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1699 GFP_KERNEL); 1700 if (!SLAVE_AD_INFO(slave)) { 1701 kobject_put(&slave->kobj); 1702 return NULL; 1703 } 1704 } 1705 1706 return slave; 1707 } 1708 1709 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1710 { 1711 info->bond_mode = BOND_MODE(bond); 1712 info->miimon = bond->params.miimon; 1713 info->num_slaves = bond->slave_cnt; 1714 } 1715 1716 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1717 { 1718 strcpy(info->slave_name, slave->dev->name); 1719 info->link = slave->link; 1720 info->state = bond_slave_state(slave); 1721 info->link_failure_count = slave->link_failure_count; 1722 } 1723 1724 static void bond_netdev_notify_work(struct work_struct *_work) 1725 { 1726 struct slave *slave = container_of(_work, struct slave, 1727 notify_work.work); 1728 1729 if (rtnl_trylock()) { 1730 struct netdev_bonding_info binfo; 1731 1732 bond_fill_ifslave(slave, &binfo.slave); 1733 bond_fill_ifbond(slave->bond, &binfo.master); 1734 netdev_bonding_info_change(slave->dev, &binfo); 1735 rtnl_unlock(); 1736 } else { 1737 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1738 } 1739 } 1740 1741 void bond_queue_slave_event(struct slave *slave) 1742 { 1743 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1744 } 1745 1746 void bond_lower_state_changed(struct slave *slave) 1747 { 1748 struct netdev_lag_lower_state_info info; 1749 1750 info.link_up = slave->link == BOND_LINK_UP || 1751 slave->link == BOND_LINK_FAIL; 1752 info.tx_enabled = bond_is_active_slave(slave); 1753 netdev_lower_state_changed(slave->dev, &info); 1754 } 1755 1756 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1757 if (extack) \ 1758 NL_SET_ERR_MSG(extack, errmsg); \ 1759 else \ 1760 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1761 } while (0) 1762 1763 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1764 if (extack) \ 1765 NL_SET_ERR_MSG(extack, errmsg); \ 1766 else \ 1767 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1768 } while (0) 1769 1770 /* enslave device <slave> to bond device <master> */ 1771 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1772 struct netlink_ext_ack *extack) 1773 { 1774 struct bonding *bond = netdev_priv(bond_dev); 1775 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1776 struct slave *new_slave = NULL, *prev_slave; 1777 struct sockaddr_storage ss; 1778 int link_reporting; 1779 int res = 0, i; 1780 1781 if (slave_dev->flags & IFF_MASTER && 1782 !netif_is_bond_master(slave_dev)) { 1783 BOND_NL_ERR(bond_dev, extack, 1784 "Device type (master device) cannot be enslaved"); 1785 return -EPERM; 1786 } 1787 1788 if (!bond->params.use_carrier && 1789 slave_dev->ethtool_ops->get_link == NULL && 1790 slave_ops->ndo_eth_ioctl == NULL) { 1791 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1792 } 1793 1794 /* already in-use? */ 1795 if (netdev_is_rx_handler_busy(slave_dev)) { 1796 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1797 "Device is in use and cannot be enslaved"); 1798 return -EBUSY; 1799 } 1800 1801 if (bond_dev == slave_dev) { 1802 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1803 return -EPERM; 1804 } 1805 1806 /* vlan challenged mutual exclusion */ 1807 /* no need to lock since we're protected by rtnl_lock */ 1808 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1809 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1810 if (vlan_uses_dev(bond_dev)) { 1811 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1812 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1813 return -EPERM; 1814 } else { 1815 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1816 } 1817 } else { 1818 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1819 } 1820 1821 if (slave_dev->features & NETIF_F_HW_ESP) 1822 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1823 1824 /* Old ifenslave binaries are no longer supported. These can 1825 * be identified with moderate accuracy by the state of the slave: 1826 * the current ifenslave will set the interface down prior to 1827 * enslaving it; the old ifenslave will not. 1828 */ 1829 if (slave_dev->flags & IFF_UP) { 1830 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1831 "Device can not be enslaved while up"); 1832 return -EPERM; 1833 } 1834 1835 /* set bonding device ether type by slave - bonding netdevices are 1836 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1837 * there is a need to override some of the type dependent attribs/funcs. 1838 * 1839 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1840 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1841 */ 1842 if (!bond_has_slaves(bond)) { 1843 if (bond_dev->type != slave_dev->type) { 1844 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1845 bond_dev->type, slave_dev->type); 1846 1847 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1848 bond_dev); 1849 res = notifier_to_errno(res); 1850 if (res) { 1851 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1852 return -EBUSY; 1853 } 1854 1855 /* Flush unicast and multicast addresses */ 1856 dev_uc_flush(bond_dev); 1857 dev_mc_flush(bond_dev); 1858 1859 if (slave_dev->type != ARPHRD_ETHER) 1860 bond_setup_by_slave(bond_dev, slave_dev); 1861 else { 1862 ether_setup(bond_dev); 1863 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1864 } 1865 1866 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1867 bond_dev); 1868 } 1869 } else if (bond_dev->type != slave_dev->type) { 1870 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1871 "Device type is different from other slaves"); 1872 return -EINVAL; 1873 } 1874 1875 if (slave_dev->type == ARPHRD_INFINIBAND && 1876 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1877 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1878 "Only active-backup mode is supported for infiniband slaves"); 1879 res = -EOPNOTSUPP; 1880 goto err_undo_flags; 1881 } 1882 1883 if (!slave_ops->ndo_set_mac_address || 1884 slave_dev->type == ARPHRD_INFINIBAND) { 1885 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1886 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1887 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1888 if (!bond_has_slaves(bond)) { 1889 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1890 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1891 } else { 1892 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1893 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1894 res = -EOPNOTSUPP; 1895 goto err_undo_flags; 1896 } 1897 } 1898 } 1899 1900 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1901 1902 /* If this is the first slave, then we need to set the master's hardware 1903 * address to be the same as the slave's. 1904 */ 1905 if (!bond_has_slaves(bond) && 1906 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1907 res = bond_set_dev_addr(bond->dev, slave_dev); 1908 if (res) 1909 goto err_undo_flags; 1910 } 1911 1912 new_slave = bond_alloc_slave(bond, slave_dev); 1913 if (!new_slave) { 1914 res = -ENOMEM; 1915 goto err_undo_flags; 1916 } 1917 1918 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1919 * is set via sysfs or module option if desired. 1920 */ 1921 new_slave->queue_id = 0; 1922 1923 /* Save slave's original mtu and then set it to match the bond */ 1924 new_slave->original_mtu = slave_dev->mtu; 1925 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1926 if (res) { 1927 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1928 goto err_free; 1929 } 1930 1931 /* Save slave's original ("permanent") mac address for modes 1932 * that need it, and for restoring it upon release, and then 1933 * set it to the master's address 1934 */ 1935 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1936 slave_dev->addr_len); 1937 1938 if (!bond->params.fail_over_mac || 1939 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1940 /* Set slave to master's mac address. The application already 1941 * set the master's mac address to that of the first slave 1942 */ 1943 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1944 ss.ss_family = slave_dev->type; 1945 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1946 extack); 1947 if (res) { 1948 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1949 goto err_restore_mtu; 1950 } 1951 } 1952 1953 /* set slave flag before open to prevent IPv6 addrconf */ 1954 slave_dev->flags |= IFF_SLAVE; 1955 1956 /* open the slave since the application closed it */ 1957 res = dev_open(slave_dev, extack); 1958 if (res) { 1959 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1960 goto err_restore_mac; 1961 } 1962 1963 slave_dev->priv_flags |= IFF_BONDING; 1964 /* initialize slave stats */ 1965 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1966 1967 if (bond_is_lb(bond)) { 1968 /* bond_alb_init_slave() must be called before all other stages since 1969 * it might fail and we do not want to have to undo everything 1970 */ 1971 res = bond_alb_init_slave(bond, new_slave); 1972 if (res) 1973 goto err_close; 1974 } 1975 1976 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1977 if (res) { 1978 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1979 goto err_close; 1980 } 1981 1982 prev_slave = bond_last_slave(bond); 1983 1984 new_slave->delay = 0; 1985 new_slave->link_failure_count = 0; 1986 1987 if (bond_update_speed_duplex(new_slave) && 1988 bond_needs_speed_duplex(bond)) 1989 new_slave->link = BOND_LINK_DOWN; 1990 1991 new_slave->last_rx = jiffies - 1992 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1993 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 1994 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 1995 1996 new_slave->last_tx = new_slave->last_rx; 1997 1998 if (bond->params.miimon && !bond->params.use_carrier) { 1999 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2000 2001 if ((link_reporting == -1) && !bond->params.arp_interval) { 2002 /* miimon is set but a bonded network driver 2003 * does not support ETHTOOL/MII and 2004 * arp_interval is not set. Note: if 2005 * use_carrier is enabled, we will never go 2006 * here (because netif_carrier is always 2007 * supported); thus, we don't need to change 2008 * the messages for netif_carrier. 2009 */ 2010 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n"); 2011 } else if (link_reporting == -1) { 2012 /* unable get link status using mii/ethtool */ 2013 slave_warn(bond_dev, slave_dev, "can't get link status from slave; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n"); 2014 } 2015 } 2016 2017 /* check for initial state */ 2018 new_slave->link = BOND_LINK_NOCHANGE; 2019 if (bond->params.miimon) { 2020 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2021 if (bond->params.updelay) { 2022 bond_set_slave_link_state(new_slave, 2023 BOND_LINK_BACK, 2024 BOND_SLAVE_NOTIFY_NOW); 2025 new_slave->delay = bond->params.updelay; 2026 } else { 2027 bond_set_slave_link_state(new_slave, 2028 BOND_LINK_UP, 2029 BOND_SLAVE_NOTIFY_NOW); 2030 } 2031 } else { 2032 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2033 BOND_SLAVE_NOTIFY_NOW); 2034 } 2035 } else if (bond->params.arp_interval) { 2036 bond_set_slave_link_state(new_slave, 2037 (netif_carrier_ok(slave_dev) ? 2038 BOND_LINK_UP : BOND_LINK_DOWN), 2039 BOND_SLAVE_NOTIFY_NOW); 2040 } else { 2041 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2042 BOND_SLAVE_NOTIFY_NOW); 2043 } 2044 2045 if (new_slave->link != BOND_LINK_DOWN) 2046 new_slave->last_link_up = jiffies; 2047 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2048 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2049 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2050 2051 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2052 /* if there is a primary slave, remember it */ 2053 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2054 rcu_assign_pointer(bond->primary_slave, new_slave); 2055 bond->force_primary = true; 2056 } 2057 } 2058 2059 switch (BOND_MODE(bond)) { 2060 case BOND_MODE_ACTIVEBACKUP: 2061 bond_set_slave_inactive_flags(new_slave, 2062 BOND_SLAVE_NOTIFY_NOW); 2063 break; 2064 case BOND_MODE_8023AD: 2065 /* in 802.3ad mode, the internal mechanism 2066 * will activate the slaves in the selected 2067 * aggregator 2068 */ 2069 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2070 /* if this is the first slave */ 2071 if (!prev_slave) { 2072 SLAVE_AD_INFO(new_slave)->id = 1; 2073 /* Initialize AD with the number of times that the AD timer is called in 1 second 2074 * can be called only after the mac address of the bond is set 2075 */ 2076 bond_3ad_initialize(bond); 2077 } else { 2078 SLAVE_AD_INFO(new_slave)->id = 2079 SLAVE_AD_INFO(prev_slave)->id + 1; 2080 } 2081 2082 bond_3ad_bind_slave(new_slave); 2083 break; 2084 case BOND_MODE_TLB: 2085 case BOND_MODE_ALB: 2086 bond_set_active_slave(new_slave); 2087 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2088 break; 2089 default: 2090 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2091 2092 /* always active in trunk mode */ 2093 bond_set_active_slave(new_slave); 2094 2095 /* In trunking mode there is little meaning to curr_active_slave 2096 * anyway (it holds no special properties of the bond device), 2097 * so we can change it without calling change_active_interface() 2098 */ 2099 if (!rcu_access_pointer(bond->curr_active_slave) && 2100 new_slave->link == BOND_LINK_UP) 2101 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2102 2103 break; 2104 } /* switch(bond_mode) */ 2105 2106 #ifdef CONFIG_NET_POLL_CONTROLLER 2107 if (bond->dev->npinfo) { 2108 if (slave_enable_netpoll(new_slave)) { 2109 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2110 res = -EBUSY; 2111 goto err_detach; 2112 } 2113 } 2114 #endif 2115 2116 if (!(bond_dev->features & NETIF_F_LRO)) 2117 dev_disable_lro(slave_dev); 2118 2119 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2120 new_slave); 2121 if (res) { 2122 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2123 goto err_detach; 2124 } 2125 2126 res = bond_master_upper_dev_link(bond, new_slave, extack); 2127 if (res) { 2128 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2129 goto err_unregister; 2130 } 2131 2132 bond_lower_state_changed(new_slave); 2133 2134 res = bond_sysfs_slave_add(new_slave); 2135 if (res) { 2136 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2137 goto err_upper_unlink; 2138 } 2139 2140 /* If the mode uses primary, then the following is handled by 2141 * bond_change_active_slave(). 2142 */ 2143 if (!bond_uses_primary(bond)) { 2144 /* set promiscuity level to new slave */ 2145 if (bond_dev->flags & IFF_PROMISC) { 2146 res = dev_set_promiscuity(slave_dev, 1); 2147 if (res) 2148 goto err_sysfs_del; 2149 } 2150 2151 /* set allmulti level to new slave */ 2152 if (bond_dev->flags & IFF_ALLMULTI) { 2153 res = dev_set_allmulti(slave_dev, 1); 2154 if (res) { 2155 if (bond_dev->flags & IFF_PROMISC) 2156 dev_set_promiscuity(slave_dev, -1); 2157 goto err_sysfs_del; 2158 } 2159 } 2160 2161 if (bond_dev->flags & IFF_UP) { 2162 netif_addr_lock_bh(bond_dev); 2163 dev_mc_sync_multiple(slave_dev, bond_dev); 2164 dev_uc_sync_multiple(slave_dev, bond_dev); 2165 netif_addr_unlock_bh(bond_dev); 2166 2167 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2168 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2169 } 2170 } 2171 2172 bond->slave_cnt++; 2173 bond_compute_features(bond); 2174 bond_set_carrier(bond); 2175 2176 if (bond_uses_primary(bond)) { 2177 block_netpoll_tx(); 2178 bond_select_active_slave(bond); 2179 unblock_netpoll_tx(); 2180 } 2181 2182 if (bond_mode_can_use_xmit_hash(bond)) 2183 bond_update_slave_arr(bond, NULL); 2184 2185 2186 if (!slave_dev->netdev_ops->ndo_bpf || 2187 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2188 if (bond->xdp_prog) { 2189 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2190 "Slave does not support XDP"); 2191 res = -EOPNOTSUPP; 2192 goto err_sysfs_del; 2193 } 2194 } else if (bond->xdp_prog) { 2195 struct netdev_bpf xdp = { 2196 .command = XDP_SETUP_PROG, 2197 .flags = 0, 2198 .prog = bond->xdp_prog, 2199 .extack = extack, 2200 }; 2201 2202 if (dev_xdp_prog_count(slave_dev) > 0) { 2203 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2204 "Slave has XDP program loaded, please unload before enslaving"); 2205 res = -EOPNOTSUPP; 2206 goto err_sysfs_del; 2207 } 2208 2209 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 2210 if (res < 0) { 2211 /* ndo_bpf() sets extack error message */ 2212 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2213 goto err_sysfs_del; 2214 } 2215 if (bond->xdp_prog) 2216 bpf_prog_inc(bond->xdp_prog); 2217 } 2218 2219 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2220 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2221 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2222 2223 /* enslave is successful */ 2224 bond_queue_slave_event(new_slave); 2225 return 0; 2226 2227 /* Undo stages on error */ 2228 err_sysfs_del: 2229 bond_sysfs_slave_del(new_slave); 2230 2231 err_upper_unlink: 2232 bond_upper_dev_unlink(bond, new_slave); 2233 2234 err_unregister: 2235 netdev_rx_handler_unregister(slave_dev); 2236 2237 err_detach: 2238 vlan_vids_del_by_dev(slave_dev, bond_dev); 2239 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2240 RCU_INIT_POINTER(bond->primary_slave, NULL); 2241 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2242 block_netpoll_tx(); 2243 bond_change_active_slave(bond, NULL); 2244 bond_select_active_slave(bond); 2245 unblock_netpoll_tx(); 2246 } 2247 /* either primary_slave or curr_active_slave might've changed */ 2248 synchronize_rcu(); 2249 slave_disable_netpoll(new_slave); 2250 2251 err_close: 2252 if (!netif_is_bond_master(slave_dev)) 2253 slave_dev->priv_flags &= ~IFF_BONDING; 2254 dev_close(slave_dev); 2255 2256 err_restore_mac: 2257 slave_dev->flags &= ~IFF_SLAVE; 2258 if (!bond->params.fail_over_mac || 2259 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2260 /* XXX TODO - fom follow mode needs to change master's 2261 * MAC if this slave's MAC is in use by the bond, or at 2262 * least print a warning. 2263 */ 2264 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2265 new_slave->dev->addr_len); 2266 ss.ss_family = slave_dev->type; 2267 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2268 } 2269 2270 err_restore_mtu: 2271 dev_set_mtu(slave_dev, new_slave->original_mtu); 2272 2273 err_free: 2274 kobject_put(&new_slave->kobj); 2275 2276 err_undo_flags: 2277 /* Enslave of first slave has failed and we need to fix master's mac */ 2278 if (!bond_has_slaves(bond)) { 2279 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2280 slave_dev->dev_addr)) 2281 eth_hw_addr_random(bond_dev); 2282 if (bond_dev->type != ARPHRD_ETHER) { 2283 dev_close(bond_dev); 2284 ether_setup(bond_dev); 2285 bond_dev->flags |= IFF_MASTER; 2286 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2287 } 2288 } 2289 2290 return res; 2291 } 2292 2293 /* Try to release the slave device <slave> from the bond device <master> 2294 * It is legal to access curr_active_slave without a lock because all the function 2295 * is RTNL-locked. If "all" is true it means that the function is being called 2296 * while destroying a bond interface and all slaves are being released. 2297 * 2298 * The rules for slave state should be: 2299 * for Active/Backup: 2300 * Active stays on all backups go down 2301 * for Bonded connections: 2302 * The first up interface should be left on and all others downed. 2303 */ 2304 static int __bond_release_one(struct net_device *bond_dev, 2305 struct net_device *slave_dev, 2306 bool all, bool unregister) 2307 { 2308 struct bonding *bond = netdev_priv(bond_dev); 2309 struct slave *slave, *oldcurrent; 2310 struct sockaddr_storage ss; 2311 int old_flags = bond_dev->flags; 2312 netdev_features_t old_features = bond_dev->features; 2313 2314 /* slave is not a slave or master is not master of this slave */ 2315 if (!(slave_dev->flags & IFF_SLAVE) || 2316 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2317 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2318 return -EINVAL; 2319 } 2320 2321 block_netpoll_tx(); 2322 2323 slave = bond_get_slave_by_dev(bond, slave_dev); 2324 if (!slave) { 2325 /* not a slave of this bond */ 2326 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2327 unblock_netpoll_tx(); 2328 return -EINVAL; 2329 } 2330 2331 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2332 2333 bond_sysfs_slave_del(slave); 2334 2335 /* recompute stats just before removing the slave */ 2336 bond_get_stats(bond->dev, &bond->bond_stats); 2337 2338 if (bond->xdp_prog) { 2339 struct netdev_bpf xdp = { 2340 .command = XDP_SETUP_PROG, 2341 .flags = 0, 2342 .prog = NULL, 2343 .extack = NULL, 2344 }; 2345 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp)) 2346 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2347 } 2348 2349 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2350 * for this slave anymore. 2351 */ 2352 netdev_rx_handler_unregister(slave_dev); 2353 2354 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2355 bond_3ad_unbind_slave(slave); 2356 2357 bond_upper_dev_unlink(bond, slave); 2358 2359 if (bond_mode_can_use_xmit_hash(bond)) 2360 bond_update_slave_arr(bond, slave); 2361 2362 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2363 bond_is_active_slave(slave) ? "active" : "backup"); 2364 2365 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2366 2367 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2368 2369 if (!all && (!bond->params.fail_over_mac || 2370 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2371 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2372 bond_has_slaves(bond)) 2373 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n", 2374 slave->perm_hwaddr); 2375 } 2376 2377 if (rtnl_dereference(bond->primary_slave) == slave) 2378 RCU_INIT_POINTER(bond->primary_slave, NULL); 2379 2380 if (oldcurrent == slave) 2381 bond_change_active_slave(bond, NULL); 2382 2383 if (bond_is_lb(bond)) { 2384 /* Must be called only after the slave has been 2385 * detached from the list and the curr_active_slave 2386 * has been cleared (if our_slave == old_current), 2387 * but before a new active slave is selected. 2388 */ 2389 bond_alb_deinit_slave(bond, slave); 2390 } 2391 2392 if (all) { 2393 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2394 } else if (oldcurrent == slave) { 2395 /* Note that we hold RTNL over this sequence, so there 2396 * is no concern that another slave add/remove event 2397 * will interfere. 2398 */ 2399 bond_select_active_slave(bond); 2400 } 2401 2402 bond_set_carrier(bond); 2403 if (!bond_has_slaves(bond)) 2404 eth_hw_addr_random(bond_dev); 2405 2406 unblock_netpoll_tx(); 2407 synchronize_rcu(); 2408 bond->slave_cnt--; 2409 2410 if (!bond_has_slaves(bond)) { 2411 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2412 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2413 } 2414 2415 bond_compute_features(bond); 2416 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2417 (old_features & NETIF_F_VLAN_CHALLENGED)) 2418 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2419 2420 vlan_vids_del_by_dev(slave_dev, bond_dev); 2421 2422 /* If the mode uses primary, then this case was handled above by 2423 * bond_change_active_slave(..., NULL) 2424 */ 2425 if (!bond_uses_primary(bond)) { 2426 /* unset promiscuity level from slave 2427 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2428 * of the IFF_PROMISC flag in the bond_dev, but we need the 2429 * value of that flag before that change, as that was the value 2430 * when this slave was attached, so we cache at the start of the 2431 * function and use it here. Same goes for ALLMULTI below 2432 */ 2433 if (old_flags & IFF_PROMISC) 2434 dev_set_promiscuity(slave_dev, -1); 2435 2436 /* unset allmulti level from slave */ 2437 if (old_flags & IFF_ALLMULTI) 2438 dev_set_allmulti(slave_dev, -1); 2439 2440 if (old_flags & IFF_UP) 2441 bond_hw_addr_flush(bond_dev, slave_dev); 2442 } 2443 2444 slave_disable_netpoll(slave); 2445 2446 /* close slave before restoring its mac address */ 2447 dev_close(slave_dev); 2448 2449 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2450 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2451 /* restore original ("permanent") mac address */ 2452 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2453 slave->dev->addr_len); 2454 ss.ss_family = slave_dev->type; 2455 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2456 } 2457 2458 if (unregister) 2459 __dev_set_mtu(slave_dev, slave->original_mtu); 2460 else 2461 dev_set_mtu(slave_dev, slave->original_mtu); 2462 2463 if (!netif_is_bond_master(slave_dev)) 2464 slave_dev->priv_flags &= ~IFF_BONDING; 2465 2466 kobject_put(&slave->kobj); 2467 2468 return 0; 2469 } 2470 2471 /* A wrapper used because of ndo_del_link */ 2472 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2473 { 2474 return __bond_release_one(bond_dev, slave_dev, false, false); 2475 } 2476 2477 /* First release a slave and then destroy the bond if no more slaves are left. 2478 * Must be under rtnl_lock when this function is called. 2479 */ 2480 static int bond_release_and_destroy(struct net_device *bond_dev, 2481 struct net_device *slave_dev) 2482 { 2483 struct bonding *bond = netdev_priv(bond_dev); 2484 int ret; 2485 2486 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2487 if (ret == 0 && !bond_has_slaves(bond) && 2488 bond_dev->reg_state != NETREG_UNREGISTERING) { 2489 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2490 netdev_info(bond_dev, "Destroying bond\n"); 2491 bond_remove_proc_entry(bond); 2492 unregister_netdevice(bond_dev); 2493 } 2494 return ret; 2495 } 2496 2497 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2498 { 2499 struct bonding *bond = netdev_priv(bond_dev); 2500 2501 bond_fill_ifbond(bond, info); 2502 } 2503 2504 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2505 { 2506 struct bonding *bond = netdev_priv(bond_dev); 2507 struct list_head *iter; 2508 int i = 0, res = -ENODEV; 2509 struct slave *slave; 2510 2511 bond_for_each_slave(bond, slave, iter) { 2512 if (i++ == (int)info->slave_id) { 2513 res = 0; 2514 bond_fill_ifslave(slave, info); 2515 break; 2516 } 2517 } 2518 2519 return res; 2520 } 2521 2522 /*-------------------------------- Monitoring -------------------------------*/ 2523 2524 /* called with rcu_read_lock() */ 2525 static int bond_miimon_inspect(struct bonding *bond) 2526 { 2527 int link_state, commit = 0; 2528 struct list_head *iter; 2529 struct slave *slave; 2530 bool ignore_updelay; 2531 2532 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2533 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2534 } else { 2535 struct bond_up_slave *usable_slaves; 2536 2537 usable_slaves = rcu_dereference(bond->usable_slaves); 2538 2539 if (usable_slaves && usable_slaves->count == 0) 2540 ignore_updelay = true; 2541 } 2542 2543 bond_for_each_slave_rcu(bond, slave, iter) { 2544 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2545 2546 link_state = bond_check_dev_link(bond, slave->dev, 0); 2547 2548 switch (slave->link) { 2549 case BOND_LINK_UP: 2550 if (link_state) 2551 continue; 2552 2553 bond_propose_link_state(slave, BOND_LINK_FAIL); 2554 commit++; 2555 slave->delay = bond->params.downdelay; 2556 if (slave->delay) { 2557 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2558 (BOND_MODE(bond) == 2559 BOND_MODE_ACTIVEBACKUP) ? 2560 (bond_is_active_slave(slave) ? 2561 "active " : "backup ") : "", 2562 bond->params.downdelay * bond->params.miimon); 2563 } 2564 fallthrough; 2565 case BOND_LINK_FAIL: 2566 if (link_state) { 2567 /* recovered before downdelay expired */ 2568 bond_propose_link_state(slave, BOND_LINK_UP); 2569 slave->last_link_up = jiffies; 2570 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2571 (bond->params.downdelay - slave->delay) * 2572 bond->params.miimon); 2573 commit++; 2574 continue; 2575 } 2576 2577 if (slave->delay <= 0) { 2578 bond_propose_link_state(slave, BOND_LINK_DOWN); 2579 commit++; 2580 continue; 2581 } 2582 2583 slave->delay--; 2584 break; 2585 2586 case BOND_LINK_DOWN: 2587 if (!link_state) 2588 continue; 2589 2590 bond_propose_link_state(slave, BOND_LINK_BACK); 2591 commit++; 2592 slave->delay = bond->params.updelay; 2593 2594 if (slave->delay) { 2595 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2596 ignore_updelay ? 0 : 2597 bond->params.updelay * 2598 bond->params.miimon); 2599 } 2600 fallthrough; 2601 case BOND_LINK_BACK: 2602 if (!link_state) { 2603 bond_propose_link_state(slave, BOND_LINK_DOWN); 2604 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2605 (bond->params.updelay - slave->delay) * 2606 bond->params.miimon); 2607 commit++; 2608 continue; 2609 } 2610 2611 if (ignore_updelay) 2612 slave->delay = 0; 2613 2614 if (slave->delay <= 0) { 2615 bond_propose_link_state(slave, BOND_LINK_UP); 2616 commit++; 2617 ignore_updelay = false; 2618 continue; 2619 } 2620 2621 slave->delay--; 2622 break; 2623 } 2624 } 2625 2626 return commit; 2627 } 2628 2629 static void bond_miimon_link_change(struct bonding *bond, 2630 struct slave *slave, 2631 char link) 2632 { 2633 switch (BOND_MODE(bond)) { 2634 case BOND_MODE_8023AD: 2635 bond_3ad_handle_link_change(slave, link); 2636 break; 2637 case BOND_MODE_TLB: 2638 case BOND_MODE_ALB: 2639 bond_alb_handle_link_change(bond, slave, link); 2640 break; 2641 case BOND_MODE_XOR: 2642 bond_update_slave_arr(bond, NULL); 2643 break; 2644 } 2645 } 2646 2647 static void bond_miimon_commit(struct bonding *bond) 2648 { 2649 struct list_head *iter; 2650 struct slave *slave, *primary; 2651 2652 bond_for_each_slave(bond, slave, iter) { 2653 switch (slave->link_new_state) { 2654 case BOND_LINK_NOCHANGE: 2655 /* For 802.3ad mode, check current slave speed and 2656 * duplex again in case its port was disabled after 2657 * invalid speed/duplex reporting but recovered before 2658 * link monitoring could make a decision on the actual 2659 * link status 2660 */ 2661 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2662 slave->link == BOND_LINK_UP) 2663 bond_3ad_adapter_speed_duplex_changed(slave); 2664 continue; 2665 2666 case BOND_LINK_UP: 2667 if (bond_update_speed_duplex(slave) && 2668 bond_needs_speed_duplex(bond)) { 2669 slave->link = BOND_LINK_DOWN; 2670 if (net_ratelimit()) 2671 slave_warn(bond->dev, slave->dev, 2672 "failed to get link speed/duplex\n"); 2673 continue; 2674 } 2675 bond_set_slave_link_state(slave, BOND_LINK_UP, 2676 BOND_SLAVE_NOTIFY_NOW); 2677 slave->last_link_up = jiffies; 2678 2679 primary = rtnl_dereference(bond->primary_slave); 2680 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2681 /* prevent it from being the active one */ 2682 bond_set_backup_slave(slave); 2683 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2684 /* make it immediately active */ 2685 bond_set_active_slave(slave); 2686 } 2687 2688 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2689 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2690 slave->duplex ? "full" : "half"); 2691 2692 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2693 2694 if (!bond->curr_active_slave || slave == primary) 2695 goto do_failover; 2696 2697 continue; 2698 2699 case BOND_LINK_DOWN: 2700 if (slave->link_failure_count < UINT_MAX) 2701 slave->link_failure_count++; 2702 2703 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2704 BOND_SLAVE_NOTIFY_NOW); 2705 2706 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2707 BOND_MODE(bond) == BOND_MODE_8023AD) 2708 bond_set_slave_inactive_flags(slave, 2709 BOND_SLAVE_NOTIFY_NOW); 2710 2711 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2712 2713 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2714 2715 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2716 goto do_failover; 2717 2718 continue; 2719 2720 default: 2721 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2722 slave->link_new_state); 2723 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2724 2725 continue; 2726 } 2727 2728 do_failover: 2729 block_netpoll_tx(); 2730 bond_select_active_slave(bond); 2731 unblock_netpoll_tx(); 2732 } 2733 2734 bond_set_carrier(bond); 2735 } 2736 2737 /* bond_mii_monitor 2738 * 2739 * Really a wrapper that splits the mii monitor into two phases: an 2740 * inspection, then (if inspection indicates something needs to be done) 2741 * an acquisition of appropriate locks followed by a commit phase to 2742 * implement whatever link state changes are indicated. 2743 */ 2744 static void bond_mii_monitor(struct work_struct *work) 2745 { 2746 struct bonding *bond = container_of(work, struct bonding, 2747 mii_work.work); 2748 bool should_notify_peers = false; 2749 bool commit; 2750 unsigned long delay; 2751 struct slave *slave; 2752 struct list_head *iter; 2753 2754 delay = msecs_to_jiffies(bond->params.miimon); 2755 2756 if (!bond_has_slaves(bond)) 2757 goto re_arm; 2758 2759 rcu_read_lock(); 2760 should_notify_peers = bond_should_notify_peers(bond); 2761 commit = !!bond_miimon_inspect(bond); 2762 if (bond->send_peer_notif) { 2763 rcu_read_unlock(); 2764 if (rtnl_trylock()) { 2765 bond->send_peer_notif--; 2766 rtnl_unlock(); 2767 } 2768 } else { 2769 rcu_read_unlock(); 2770 } 2771 2772 if (commit) { 2773 /* Race avoidance with bond_close cancel of workqueue */ 2774 if (!rtnl_trylock()) { 2775 delay = 1; 2776 should_notify_peers = false; 2777 goto re_arm; 2778 } 2779 2780 bond_for_each_slave(bond, slave, iter) { 2781 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2782 } 2783 bond_miimon_commit(bond); 2784 2785 rtnl_unlock(); /* might sleep, hold no other locks */ 2786 } 2787 2788 re_arm: 2789 if (bond->params.miimon) 2790 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2791 2792 if (should_notify_peers) { 2793 if (!rtnl_trylock()) 2794 return; 2795 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2796 rtnl_unlock(); 2797 } 2798 } 2799 2800 static int bond_upper_dev_walk(struct net_device *upper, 2801 struct netdev_nested_priv *priv) 2802 { 2803 __be32 ip = *(__be32 *)priv->data; 2804 2805 return ip == bond_confirm_addr(upper, 0, ip); 2806 } 2807 2808 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2809 { 2810 struct netdev_nested_priv priv = { 2811 .data = (void *)&ip, 2812 }; 2813 bool ret = false; 2814 2815 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2816 return true; 2817 2818 rcu_read_lock(); 2819 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2820 ret = true; 2821 rcu_read_unlock(); 2822 2823 return ret; 2824 } 2825 2826 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2827 struct sk_buff *skb) 2828 { 2829 struct net_device *bond_dev = slave->bond->dev; 2830 struct net_device *slave_dev = slave->dev; 2831 struct bond_vlan_tag *outer_tag = tags; 2832 2833 if (!tags || tags->vlan_proto == VLAN_N_VID) 2834 return true; 2835 2836 tags++; 2837 2838 /* Go through all the tags backwards and add them to the packet */ 2839 while (tags->vlan_proto != VLAN_N_VID) { 2840 if (!tags->vlan_id) { 2841 tags++; 2842 continue; 2843 } 2844 2845 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2846 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2847 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2848 tags->vlan_id); 2849 if (!skb) { 2850 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2851 return false; 2852 } 2853 2854 tags++; 2855 } 2856 /* Set the outer tag */ 2857 if (outer_tag->vlan_id) { 2858 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2859 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2860 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2861 outer_tag->vlan_id); 2862 } 2863 2864 return true; 2865 } 2866 2867 /* We go to the (large) trouble of VLAN tagging ARP frames because 2868 * switches in VLAN mode (especially if ports are configured as 2869 * "native" to a VLAN) might not pass non-tagged frames. 2870 */ 2871 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2872 __be32 src_ip, struct bond_vlan_tag *tags) 2873 { 2874 struct net_device *bond_dev = slave->bond->dev; 2875 struct net_device *slave_dev = slave->dev; 2876 struct sk_buff *skb; 2877 2878 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2879 arp_op, &dest_ip, &src_ip); 2880 2881 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2882 NULL, slave_dev->dev_addr, NULL); 2883 2884 if (!skb) { 2885 net_err_ratelimited("ARP packet allocation failed\n"); 2886 return; 2887 } 2888 2889 if (bond_handle_vlan(slave, tags, skb)) { 2890 slave_update_last_tx(slave); 2891 arp_xmit(skb); 2892 } 2893 2894 return; 2895 } 2896 2897 /* Validate the device path between the @start_dev and the @end_dev. 2898 * The path is valid if the @end_dev is reachable through device 2899 * stacking. 2900 * When the path is validated, collect any vlan information in the 2901 * path. 2902 */ 2903 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2904 struct net_device *end_dev, 2905 int level) 2906 { 2907 struct bond_vlan_tag *tags; 2908 struct net_device *upper; 2909 struct list_head *iter; 2910 2911 if (start_dev == end_dev) { 2912 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2913 if (!tags) 2914 return ERR_PTR(-ENOMEM); 2915 tags[level].vlan_proto = VLAN_N_VID; 2916 return tags; 2917 } 2918 2919 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2920 tags = bond_verify_device_path(upper, end_dev, level + 1); 2921 if (IS_ERR_OR_NULL(tags)) { 2922 if (IS_ERR(tags)) 2923 return tags; 2924 continue; 2925 } 2926 if (is_vlan_dev(upper)) { 2927 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2928 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2929 } 2930 2931 return tags; 2932 } 2933 2934 return NULL; 2935 } 2936 2937 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2938 { 2939 struct rtable *rt; 2940 struct bond_vlan_tag *tags; 2941 __be32 *targets = bond->params.arp_targets, addr; 2942 int i; 2943 2944 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2945 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2946 __func__, &targets[i]); 2947 tags = NULL; 2948 2949 /* Find out through which dev should the packet go */ 2950 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2951 RTO_ONLINK, 0); 2952 if (IS_ERR(rt)) { 2953 /* there's no route to target - try to send arp 2954 * probe to generate any traffic (arp_validate=0) 2955 */ 2956 if (bond->params.arp_validate) 2957 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2958 bond->dev->name, 2959 &targets[i]); 2960 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2961 0, tags); 2962 continue; 2963 } 2964 2965 /* bond device itself */ 2966 if (rt->dst.dev == bond->dev) 2967 goto found; 2968 2969 rcu_read_lock(); 2970 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2971 rcu_read_unlock(); 2972 2973 if (!IS_ERR_OR_NULL(tags)) 2974 goto found; 2975 2976 /* Not our device - skip */ 2977 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2978 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2979 2980 ip_rt_put(rt); 2981 continue; 2982 2983 found: 2984 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2985 ip_rt_put(rt); 2986 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2987 kfree(tags); 2988 } 2989 } 2990 2991 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 2992 { 2993 int i; 2994 2995 if (!sip || !bond_has_this_ip(bond, tip)) { 2996 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 2997 __func__, &sip, &tip); 2998 return; 2999 } 3000 3001 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3002 if (i == -1) { 3003 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3004 __func__, &sip); 3005 return; 3006 } 3007 slave->last_rx = jiffies; 3008 slave->target_last_arp_rx[i] = jiffies; 3009 } 3010 3011 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3012 struct slave *slave) 3013 { 3014 struct arphdr *arp = (struct arphdr *)skb->data; 3015 struct slave *curr_active_slave, *curr_arp_slave; 3016 unsigned char *arp_ptr; 3017 __be32 sip, tip; 3018 unsigned int alen; 3019 3020 alen = arp_hdr_len(bond->dev); 3021 3022 if (alen > skb_headlen(skb)) { 3023 arp = kmalloc(alen, GFP_ATOMIC); 3024 if (!arp) 3025 goto out_unlock; 3026 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3027 goto out_unlock; 3028 } 3029 3030 if (arp->ar_hln != bond->dev->addr_len || 3031 skb->pkt_type == PACKET_OTHERHOST || 3032 skb->pkt_type == PACKET_LOOPBACK || 3033 arp->ar_hrd != htons(ARPHRD_ETHER) || 3034 arp->ar_pro != htons(ETH_P_IP) || 3035 arp->ar_pln != 4) 3036 goto out_unlock; 3037 3038 arp_ptr = (unsigned char *)(arp + 1); 3039 arp_ptr += bond->dev->addr_len; 3040 memcpy(&sip, arp_ptr, 4); 3041 arp_ptr += 4 + bond->dev->addr_len; 3042 memcpy(&tip, arp_ptr, 4); 3043 3044 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3045 __func__, slave->dev->name, bond_slave_state(slave), 3046 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3047 &sip, &tip); 3048 3049 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3050 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3051 3052 /* We 'trust' the received ARP enough to validate it if: 3053 * 3054 * (a) the slave receiving the ARP is active (which includes the 3055 * current ARP slave, if any), or 3056 * 3057 * (b) the receiving slave isn't active, but there is a currently 3058 * active slave and it received valid arp reply(s) after it became 3059 * the currently active slave, or 3060 * 3061 * (c) there is an ARP slave that sent an ARP during the prior ARP 3062 * interval, and we receive an ARP reply on any slave. We accept 3063 * these because switch FDB update delays may deliver the ARP 3064 * reply to a slave other than the sender of the ARP request. 3065 * 3066 * Note: for (b), backup slaves are receiving the broadcast ARP 3067 * request, not a reply. This request passes from the sending 3068 * slave through the L2 switch(es) to the receiving slave. Since 3069 * this is checking the request, sip/tip are swapped for 3070 * validation. 3071 * 3072 * This is done to avoid endless looping when we can't reach the 3073 * arp_ip_target and fool ourselves with our own arp requests. 3074 */ 3075 if (bond_is_active_slave(slave)) 3076 bond_validate_arp(bond, slave, sip, tip); 3077 else if (curr_active_slave && 3078 time_after(slave_last_rx(bond, curr_active_slave), 3079 curr_active_slave->last_link_up)) 3080 bond_validate_arp(bond, slave, tip, sip); 3081 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3082 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3083 bond_validate_arp(bond, slave, sip, tip); 3084 3085 out_unlock: 3086 if (arp != (struct arphdr *)skb->data) 3087 kfree(arp); 3088 return RX_HANDLER_ANOTHER; 3089 } 3090 3091 #if IS_ENABLED(CONFIG_IPV6) 3092 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3093 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3094 { 3095 struct net_device *bond_dev = slave->bond->dev; 3096 struct net_device *slave_dev = slave->dev; 3097 struct in6_addr mcaddr; 3098 struct sk_buff *skb; 3099 3100 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3101 daddr, saddr); 3102 3103 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3104 if (!skb) { 3105 net_err_ratelimited("NS packet allocation failed\n"); 3106 return; 3107 } 3108 3109 addrconf_addr_solict_mult(daddr, &mcaddr); 3110 if (bond_handle_vlan(slave, tags, skb)) { 3111 slave_update_last_tx(slave); 3112 ndisc_send_skb(skb, &mcaddr, saddr); 3113 } 3114 } 3115 3116 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3117 { 3118 struct in6_addr *targets = bond->params.ns_targets; 3119 struct bond_vlan_tag *tags; 3120 struct dst_entry *dst; 3121 struct in6_addr saddr; 3122 struct flowi6 fl6; 3123 int i; 3124 3125 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3126 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3127 __func__, &targets[i]); 3128 tags = NULL; 3129 3130 /* Find out through which dev should the packet go */ 3131 memset(&fl6, 0, sizeof(struct flowi6)); 3132 fl6.daddr = targets[i]; 3133 fl6.flowi6_oif = bond->dev->ifindex; 3134 3135 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3136 if (dst->error) { 3137 dst_release(dst); 3138 /* there's no route to target - try to send arp 3139 * probe to generate any traffic (arp_validate=0) 3140 */ 3141 if (bond->params.arp_validate) 3142 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3143 bond->dev->name, 3144 &targets[i]); 3145 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3146 continue; 3147 } 3148 3149 /* bond device itself */ 3150 if (dst->dev == bond->dev) 3151 goto found; 3152 3153 rcu_read_lock(); 3154 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3155 rcu_read_unlock(); 3156 3157 if (!IS_ERR_OR_NULL(tags)) 3158 goto found; 3159 3160 /* Not our device - skip */ 3161 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3162 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3163 3164 dst_release(dst); 3165 continue; 3166 3167 found: 3168 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3169 bond_ns_send(slave, &targets[i], &saddr, tags); 3170 else 3171 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3172 3173 dst_release(dst); 3174 kfree(tags); 3175 } 3176 } 3177 3178 static int bond_confirm_addr6(struct net_device *dev, 3179 struct netdev_nested_priv *priv) 3180 { 3181 struct in6_addr *addr = (struct in6_addr *)priv->data; 3182 3183 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3184 } 3185 3186 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3187 { 3188 struct netdev_nested_priv priv = { 3189 .data = addr, 3190 }; 3191 int ret = false; 3192 3193 if (bond_confirm_addr6(bond->dev, &priv)) 3194 return true; 3195 3196 rcu_read_lock(); 3197 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3198 ret = true; 3199 rcu_read_unlock(); 3200 3201 return ret; 3202 } 3203 3204 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3205 struct in6_addr *saddr, struct in6_addr *daddr) 3206 { 3207 int i; 3208 3209 /* Ignore NAs that: 3210 * 1. Source address is unspecified address. 3211 * 2. Dest address is neither all-nodes multicast address nor 3212 * exist on bond interface. 3213 */ 3214 if (ipv6_addr_any(saddr) || 3215 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3216 !bond_has_this_ip6(bond, daddr))) { 3217 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3218 __func__, saddr, daddr); 3219 return; 3220 } 3221 3222 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3223 if (i == -1) { 3224 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3225 __func__, saddr); 3226 return; 3227 } 3228 slave->last_rx = jiffies; 3229 slave->target_last_arp_rx[i] = jiffies; 3230 } 3231 3232 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3233 struct slave *slave) 3234 { 3235 struct slave *curr_active_slave, *curr_arp_slave; 3236 struct in6_addr *saddr, *daddr; 3237 struct { 3238 struct ipv6hdr ip6; 3239 struct icmp6hdr icmp6; 3240 } *combined, _combined; 3241 3242 if (skb->pkt_type == PACKET_OTHERHOST || 3243 skb->pkt_type == PACKET_LOOPBACK) 3244 goto out; 3245 3246 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3247 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3248 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT) 3249 goto out; 3250 3251 saddr = &combined->ip6.saddr; 3252 daddr = &combined->ip6.saddr; 3253 3254 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3255 __func__, slave->dev->name, bond_slave_state(slave), 3256 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3257 saddr, daddr); 3258 3259 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3260 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3261 3262 /* We 'trust' the received ARP enough to validate it if: 3263 * see bond_arp_rcv(). 3264 */ 3265 if (bond_is_active_slave(slave)) 3266 bond_validate_na(bond, slave, saddr, daddr); 3267 else if (curr_active_slave && 3268 time_after(slave_last_rx(bond, curr_active_slave), 3269 curr_active_slave->last_link_up)) 3270 bond_validate_na(bond, slave, saddr, daddr); 3271 else if (curr_arp_slave && 3272 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3273 bond_validate_na(bond, slave, saddr, daddr); 3274 3275 out: 3276 return RX_HANDLER_ANOTHER; 3277 } 3278 #endif 3279 3280 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3281 struct slave *slave) 3282 { 3283 #if IS_ENABLED(CONFIG_IPV6) 3284 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3285 #endif 3286 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3287 3288 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3289 __func__, skb->dev->name); 3290 3291 /* Use arp validate logic for both ARP and NS */ 3292 if (!slave_do_arp_validate(bond, slave)) { 3293 if ((slave_do_arp_validate_only(bond) && is_arp) || 3294 #if IS_ENABLED(CONFIG_IPV6) 3295 (slave_do_arp_validate_only(bond) && is_ipv6) || 3296 #endif 3297 !slave_do_arp_validate_only(bond)) 3298 slave->last_rx = jiffies; 3299 return RX_HANDLER_ANOTHER; 3300 } else if (is_arp) { 3301 return bond_arp_rcv(skb, bond, slave); 3302 #if IS_ENABLED(CONFIG_IPV6) 3303 } else if (is_ipv6) { 3304 return bond_na_rcv(skb, bond, slave); 3305 #endif 3306 } else { 3307 return RX_HANDLER_ANOTHER; 3308 } 3309 } 3310 3311 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3312 { 3313 bond_arp_send_all(bond, slave); 3314 #if IS_ENABLED(CONFIG_IPV6) 3315 bond_ns_send_all(bond, slave); 3316 #endif 3317 } 3318 3319 /* function to verify if we're in the arp_interval timeslice, returns true if 3320 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3321 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3322 */ 3323 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3324 int mod) 3325 { 3326 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3327 3328 return time_in_range(jiffies, 3329 last_act - delta_in_ticks, 3330 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3331 } 3332 3333 /* This function is called regularly to monitor each slave's link 3334 * ensuring that traffic is being sent and received when arp monitoring 3335 * is used in load-balancing mode. if the adapter has been dormant, then an 3336 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3337 * arp monitoring in active backup mode. 3338 */ 3339 static void bond_loadbalance_arp_mon(struct bonding *bond) 3340 { 3341 struct slave *slave, *oldcurrent; 3342 struct list_head *iter; 3343 int do_failover = 0, slave_state_changed = 0; 3344 3345 if (!bond_has_slaves(bond)) 3346 goto re_arm; 3347 3348 rcu_read_lock(); 3349 3350 oldcurrent = rcu_dereference(bond->curr_active_slave); 3351 /* see if any of the previous devices are up now (i.e. they have 3352 * xmt and rcv traffic). the curr_active_slave does not come into 3353 * the picture unless it is null. also, slave->last_link_up is not 3354 * needed here because we send an arp on each slave and give a slave 3355 * as long as it needs to get the tx/rx within the delta. 3356 * TODO: what about up/down delay in arp mode? it wasn't here before 3357 * so it can wait 3358 */ 3359 bond_for_each_slave_rcu(bond, slave, iter) { 3360 unsigned long last_tx = slave_last_tx(slave); 3361 3362 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3363 3364 if (slave->link != BOND_LINK_UP) { 3365 if (bond_time_in_interval(bond, last_tx, 1) && 3366 bond_time_in_interval(bond, slave->last_rx, 1)) { 3367 3368 bond_propose_link_state(slave, BOND_LINK_UP); 3369 slave_state_changed = 1; 3370 3371 /* primary_slave has no meaning in round-robin 3372 * mode. the window of a slave being up and 3373 * curr_active_slave being null after enslaving 3374 * is closed. 3375 */ 3376 if (!oldcurrent) { 3377 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3378 do_failover = 1; 3379 } else { 3380 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3381 } 3382 } 3383 } else { 3384 /* slave->link == BOND_LINK_UP */ 3385 3386 /* not all switches will respond to an arp request 3387 * when the source ip is 0, so don't take the link down 3388 * if we don't know our ip yet 3389 */ 3390 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3391 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3392 3393 bond_propose_link_state(slave, BOND_LINK_DOWN); 3394 slave_state_changed = 1; 3395 3396 if (slave->link_failure_count < UINT_MAX) 3397 slave->link_failure_count++; 3398 3399 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3400 3401 if (slave == oldcurrent) 3402 do_failover = 1; 3403 } 3404 } 3405 3406 /* note: if switch is in round-robin mode, all links 3407 * must tx arp to ensure all links rx an arp - otherwise 3408 * links may oscillate or not come up at all; if switch is 3409 * in something like xor mode, there is nothing we can 3410 * do - all replies will be rx'ed on same link causing slaves 3411 * to be unstable during low/no traffic periods 3412 */ 3413 if (bond_slave_is_up(slave)) 3414 bond_send_validate(bond, slave); 3415 } 3416 3417 rcu_read_unlock(); 3418 3419 if (do_failover || slave_state_changed) { 3420 if (!rtnl_trylock()) 3421 goto re_arm; 3422 3423 bond_for_each_slave(bond, slave, iter) { 3424 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3425 slave->link = slave->link_new_state; 3426 } 3427 3428 if (slave_state_changed) { 3429 bond_slave_state_change(bond); 3430 if (BOND_MODE(bond) == BOND_MODE_XOR) 3431 bond_update_slave_arr(bond, NULL); 3432 } 3433 if (do_failover) { 3434 block_netpoll_tx(); 3435 bond_select_active_slave(bond); 3436 unblock_netpoll_tx(); 3437 } 3438 rtnl_unlock(); 3439 } 3440 3441 re_arm: 3442 if (bond->params.arp_interval) 3443 queue_delayed_work(bond->wq, &bond->arp_work, 3444 msecs_to_jiffies(bond->params.arp_interval)); 3445 } 3446 3447 /* Called to inspect slaves for active-backup mode ARP monitor link state 3448 * changes. Sets proposed link state in slaves to specify what action 3449 * should take place for the slave. Returns 0 if no changes are found, >0 3450 * if changes to link states must be committed. 3451 * 3452 * Called with rcu_read_lock held. 3453 */ 3454 static int bond_ab_arp_inspect(struct bonding *bond) 3455 { 3456 unsigned long last_tx, last_rx; 3457 struct list_head *iter; 3458 struct slave *slave; 3459 int commit = 0; 3460 3461 bond_for_each_slave_rcu(bond, slave, iter) { 3462 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3463 last_rx = slave_last_rx(bond, slave); 3464 3465 if (slave->link != BOND_LINK_UP) { 3466 if (bond_time_in_interval(bond, last_rx, 1)) { 3467 bond_propose_link_state(slave, BOND_LINK_UP); 3468 commit++; 3469 } else if (slave->link == BOND_LINK_BACK) { 3470 bond_propose_link_state(slave, BOND_LINK_FAIL); 3471 commit++; 3472 } 3473 continue; 3474 } 3475 3476 /* Give slaves 2*delta after being enslaved or made 3477 * active. This avoids bouncing, as the last receive 3478 * times need a full ARP monitor cycle to be updated. 3479 */ 3480 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3481 continue; 3482 3483 /* Backup slave is down if: 3484 * - No current_arp_slave AND 3485 * - more than (missed_max+1)*delta since last receive AND 3486 * - the bond has an IP address 3487 * 3488 * Note: a non-null current_arp_slave indicates 3489 * the curr_active_slave went down and we are 3490 * searching for a new one; under this condition 3491 * we only take the curr_active_slave down - this 3492 * gives each slave a chance to tx/rx traffic 3493 * before being taken out 3494 */ 3495 if (!bond_is_active_slave(slave) && 3496 !rcu_access_pointer(bond->current_arp_slave) && 3497 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3498 bond_propose_link_state(slave, BOND_LINK_DOWN); 3499 commit++; 3500 } 3501 3502 /* Active slave is down if: 3503 * - more than missed_max*delta since transmitting OR 3504 * - (more than missed_max*delta since receive AND 3505 * the bond has an IP address) 3506 */ 3507 last_tx = slave_last_tx(slave); 3508 if (bond_is_active_slave(slave) && 3509 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3510 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3511 bond_propose_link_state(slave, BOND_LINK_DOWN); 3512 commit++; 3513 } 3514 } 3515 3516 return commit; 3517 } 3518 3519 /* Called to commit link state changes noted by inspection step of 3520 * active-backup mode ARP monitor. 3521 * 3522 * Called with RTNL hold. 3523 */ 3524 static void bond_ab_arp_commit(struct bonding *bond) 3525 { 3526 struct list_head *iter; 3527 unsigned long last_tx; 3528 struct slave *slave; 3529 3530 bond_for_each_slave(bond, slave, iter) { 3531 switch (slave->link_new_state) { 3532 case BOND_LINK_NOCHANGE: 3533 continue; 3534 3535 case BOND_LINK_UP: 3536 last_tx = slave_last_tx(slave); 3537 if (rtnl_dereference(bond->curr_active_slave) != slave || 3538 (!rtnl_dereference(bond->curr_active_slave) && 3539 bond_time_in_interval(bond, last_tx, 1))) { 3540 struct slave *current_arp_slave; 3541 3542 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3543 bond_set_slave_link_state(slave, BOND_LINK_UP, 3544 BOND_SLAVE_NOTIFY_NOW); 3545 if (current_arp_slave) { 3546 bond_set_slave_inactive_flags( 3547 current_arp_slave, 3548 BOND_SLAVE_NOTIFY_NOW); 3549 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3550 } 3551 3552 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3553 3554 if (!rtnl_dereference(bond->curr_active_slave) || 3555 slave == rtnl_dereference(bond->primary_slave)) 3556 goto do_failover; 3557 3558 } 3559 3560 continue; 3561 3562 case BOND_LINK_DOWN: 3563 if (slave->link_failure_count < UINT_MAX) 3564 slave->link_failure_count++; 3565 3566 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3567 BOND_SLAVE_NOTIFY_NOW); 3568 bond_set_slave_inactive_flags(slave, 3569 BOND_SLAVE_NOTIFY_NOW); 3570 3571 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3572 3573 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3574 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3575 goto do_failover; 3576 } 3577 3578 continue; 3579 3580 case BOND_LINK_FAIL: 3581 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3582 BOND_SLAVE_NOTIFY_NOW); 3583 bond_set_slave_inactive_flags(slave, 3584 BOND_SLAVE_NOTIFY_NOW); 3585 3586 /* A slave has just been enslaved and has become 3587 * the current active slave. 3588 */ 3589 if (rtnl_dereference(bond->curr_active_slave)) 3590 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3591 continue; 3592 3593 default: 3594 slave_err(bond->dev, slave->dev, 3595 "impossible: link_new_state %d on slave\n", 3596 slave->link_new_state); 3597 continue; 3598 } 3599 3600 do_failover: 3601 block_netpoll_tx(); 3602 bond_select_active_slave(bond); 3603 unblock_netpoll_tx(); 3604 } 3605 3606 bond_set_carrier(bond); 3607 } 3608 3609 /* Send ARP probes for active-backup mode ARP monitor. 3610 * 3611 * Called with rcu_read_lock held. 3612 */ 3613 static bool bond_ab_arp_probe(struct bonding *bond) 3614 { 3615 struct slave *slave, *before = NULL, *new_slave = NULL, 3616 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3617 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3618 struct list_head *iter; 3619 bool found = false; 3620 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3621 3622 if (curr_arp_slave && curr_active_slave) 3623 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3624 curr_arp_slave->dev->name, 3625 curr_active_slave->dev->name); 3626 3627 if (curr_active_slave) { 3628 bond_send_validate(bond, curr_active_slave); 3629 return should_notify_rtnl; 3630 } 3631 3632 /* if we don't have a curr_active_slave, search for the next available 3633 * backup slave from the current_arp_slave and make it the candidate 3634 * for becoming the curr_active_slave 3635 */ 3636 3637 if (!curr_arp_slave) { 3638 curr_arp_slave = bond_first_slave_rcu(bond); 3639 if (!curr_arp_slave) 3640 return should_notify_rtnl; 3641 } 3642 3643 bond_for_each_slave_rcu(bond, slave, iter) { 3644 if (!found && !before && bond_slave_is_up(slave)) 3645 before = slave; 3646 3647 if (found && !new_slave && bond_slave_is_up(slave)) 3648 new_slave = slave; 3649 /* if the link state is up at this point, we 3650 * mark it down - this can happen if we have 3651 * simultaneous link failures and 3652 * reselect_active_interface doesn't make this 3653 * one the current slave so it is still marked 3654 * up when it is actually down 3655 */ 3656 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3657 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3658 BOND_SLAVE_NOTIFY_LATER); 3659 if (slave->link_failure_count < UINT_MAX) 3660 slave->link_failure_count++; 3661 3662 bond_set_slave_inactive_flags(slave, 3663 BOND_SLAVE_NOTIFY_LATER); 3664 3665 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3666 } 3667 if (slave == curr_arp_slave) 3668 found = true; 3669 } 3670 3671 if (!new_slave && before) 3672 new_slave = before; 3673 3674 if (!new_slave) 3675 goto check_state; 3676 3677 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3678 BOND_SLAVE_NOTIFY_LATER); 3679 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3680 bond_send_validate(bond, new_slave); 3681 new_slave->last_link_up = jiffies; 3682 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3683 3684 check_state: 3685 bond_for_each_slave_rcu(bond, slave, iter) { 3686 if (slave->should_notify || slave->should_notify_link) { 3687 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3688 break; 3689 } 3690 } 3691 return should_notify_rtnl; 3692 } 3693 3694 static void bond_activebackup_arp_mon(struct bonding *bond) 3695 { 3696 bool should_notify_peers = false; 3697 bool should_notify_rtnl = false; 3698 int delta_in_ticks; 3699 3700 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3701 3702 if (!bond_has_slaves(bond)) 3703 goto re_arm; 3704 3705 rcu_read_lock(); 3706 3707 should_notify_peers = bond_should_notify_peers(bond); 3708 3709 if (bond_ab_arp_inspect(bond)) { 3710 rcu_read_unlock(); 3711 3712 /* Race avoidance with bond_close flush of workqueue */ 3713 if (!rtnl_trylock()) { 3714 delta_in_ticks = 1; 3715 should_notify_peers = false; 3716 goto re_arm; 3717 } 3718 3719 bond_ab_arp_commit(bond); 3720 3721 rtnl_unlock(); 3722 rcu_read_lock(); 3723 } 3724 3725 should_notify_rtnl = bond_ab_arp_probe(bond); 3726 rcu_read_unlock(); 3727 3728 re_arm: 3729 if (bond->params.arp_interval) 3730 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3731 3732 if (should_notify_peers || should_notify_rtnl) { 3733 if (!rtnl_trylock()) 3734 return; 3735 3736 if (should_notify_peers) { 3737 bond->send_peer_notif--; 3738 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3739 bond->dev); 3740 } 3741 if (should_notify_rtnl) { 3742 bond_slave_state_notify(bond); 3743 bond_slave_link_notify(bond); 3744 } 3745 3746 rtnl_unlock(); 3747 } 3748 } 3749 3750 static void bond_arp_monitor(struct work_struct *work) 3751 { 3752 struct bonding *bond = container_of(work, struct bonding, 3753 arp_work.work); 3754 3755 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3756 bond_activebackup_arp_mon(bond); 3757 else 3758 bond_loadbalance_arp_mon(bond); 3759 } 3760 3761 /*-------------------------- netdev event handling --------------------------*/ 3762 3763 /* Change device name */ 3764 static int bond_event_changename(struct bonding *bond) 3765 { 3766 bond_remove_proc_entry(bond); 3767 bond_create_proc_entry(bond); 3768 3769 bond_debug_reregister(bond); 3770 3771 return NOTIFY_DONE; 3772 } 3773 3774 static int bond_master_netdev_event(unsigned long event, 3775 struct net_device *bond_dev) 3776 { 3777 struct bonding *event_bond = netdev_priv(bond_dev); 3778 3779 netdev_dbg(bond_dev, "%s called\n", __func__); 3780 3781 switch (event) { 3782 case NETDEV_CHANGENAME: 3783 return bond_event_changename(event_bond); 3784 case NETDEV_UNREGISTER: 3785 bond_remove_proc_entry(event_bond); 3786 #ifdef CONFIG_XFRM_OFFLOAD 3787 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3788 #endif /* CONFIG_XFRM_OFFLOAD */ 3789 break; 3790 case NETDEV_REGISTER: 3791 bond_create_proc_entry(event_bond); 3792 break; 3793 default: 3794 break; 3795 } 3796 3797 return NOTIFY_DONE; 3798 } 3799 3800 static int bond_slave_netdev_event(unsigned long event, 3801 struct net_device *slave_dev) 3802 { 3803 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3804 struct bonding *bond; 3805 struct net_device *bond_dev; 3806 3807 /* A netdev event can be generated while enslaving a device 3808 * before netdev_rx_handler_register is called in which case 3809 * slave will be NULL 3810 */ 3811 if (!slave) { 3812 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3813 return NOTIFY_DONE; 3814 } 3815 3816 bond_dev = slave->bond->dev; 3817 bond = slave->bond; 3818 primary = rtnl_dereference(bond->primary_slave); 3819 3820 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3821 3822 switch (event) { 3823 case NETDEV_UNREGISTER: 3824 if (bond_dev->type != ARPHRD_ETHER) 3825 bond_release_and_destroy(bond_dev, slave_dev); 3826 else 3827 __bond_release_one(bond_dev, slave_dev, false, true); 3828 break; 3829 case NETDEV_UP: 3830 case NETDEV_CHANGE: 3831 /* For 802.3ad mode only: 3832 * Getting invalid Speed/Duplex values here will put slave 3833 * in weird state. Mark it as link-fail if the link was 3834 * previously up or link-down if it hasn't yet come up, and 3835 * let link-monitoring (miimon) set it right when correct 3836 * speeds/duplex are available. 3837 */ 3838 if (bond_update_speed_duplex(slave) && 3839 BOND_MODE(bond) == BOND_MODE_8023AD) { 3840 if (slave->last_link_up) 3841 slave->link = BOND_LINK_FAIL; 3842 else 3843 slave->link = BOND_LINK_DOWN; 3844 } 3845 3846 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3847 bond_3ad_adapter_speed_duplex_changed(slave); 3848 fallthrough; 3849 case NETDEV_DOWN: 3850 /* Refresh slave-array if applicable! 3851 * If the setup does not use miimon or arpmon (mode-specific!), 3852 * then these events will not cause the slave-array to be 3853 * refreshed. This will cause xmit to use a slave that is not 3854 * usable. Avoid such situation by refeshing the array at these 3855 * events. If these (miimon/arpmon) parameters are configured 3856 * then array gets refreshed twice and that should be fine! 3857 */ 3858 if (bond_mode_can_use_xmit_hash(bond)) 3859 bond_update_slave_arr(bond, NULL); 3860 break; 3861 case NETDEV_CHANGEMTU: 3862 /* TODO: Should slaves be allowed to 3863 * independently alter their MTU? For 3864 * an active-backup bond, slaves need 3865 * not be the same type of device, so 3866 * MTUs may vary. For other modes, 3867 * slaves arguably should have the 3868 * same MTUs. To do this, we'd need to 3869 * take over the slave's change_mtu 3870 * function for the duration of their 3871 * servitude. 3872 */ 3873 break; 3874 case NETDEV_CHANGENAME: 3875 /* we don't care if we don't have primary set */ 3876 if (!bond_uses_primary(bond) || 3877 !bond->params.primary[0]) 3878 break; 3879 3880 if (slave == primary) { 3881 /* slave's name changed - he's no longer primary */ 3882 RCU_INIT_POINTER(bond->primary_slave, NULL); 3883 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3884 /* we have a new primary slave */ 3885 rcu_assign_pointer(bond->primary_slave, slave); 3886 } else { /* we didn't change primary - exit */ 3887 break; 3888 } 3889 3890 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3891 primary ? slave_dev->name : "none"); 3892 3893 block_netpoll_tx(); 3894 bond_select_active_slave(bond); 3895 unblock_netpoll_tx(); 3896 break; 3897 case NETDEV_FEAT_CHANGE: 3898 bond_compute_features(bond); 3899 break; 3900 case NETDEV_RESEND_IGMP: 3901 /* Propagate to master device */ 3902 call_netdevice_notifiers(event, slave->bond->dev); 3903 break; 3904 default: 3905 break; 3906 } 3907 3908 return NOTIFY_DONE; 3909 } 3910 3911 /* bond_netdev_event: handle netdev notifier chain events. 3912 * 3913 * This function receives events for the netdev chain. The caller (an 3914 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3915 * locks for us to safely manipulate the slave devices (RTNL lock, 3916 * dev_probe_lock). 3917 */ 3918 static int bond_netdev_event(struct notifier_block *this, 3919 unsigned long event, void *ptr) 3920 { 3921 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3922 3923 netdev_dbg(event_dev, "%s received %s\n", 3924 __func__, netdev_cmd_to_name(event)); 3925 3926 if (!(event_dev->priv_flags & IFF_BONDING)) 3927 return NOTIFY_DONE; 3928 3929 if (event_dev->flags & IFF_MASTER) { 3930 int ret; 3931 3932 ret = bond_master_netdev_event(event, event_dev); 3933 if (ret != NOTIFY_DONE) 3934 return ret; 3935 } 3936 3937 if (event_dev->flags & IFF_SLAVE) 3938 return bond_slave_netdev_event(event, event_dev); 3939 3940 return NOTIFY_DONE; 3941 } 3942 3943 static struct notifier_block bond_netdev_notifier = { 3944 .notifier_call = bond_netdev_event, 3945 }; 3946 3947 /*---------------------------- Hashing Policies -----------------------------*/ 3948 3949 /* Helper to access data in a packet, with or without a backing skb. 3950 * If skb is given the data is linearized if necessary via pskb_may_pull. 3951 */ 3952 static inline const void *bond_pull_data(struct sk_buff *skb, 3953 const void *data, int hlen, int n) 3954 { 3955 if (likely(n <= hlen)) 3956 return data; 3957 else if (skb && likely(pskb_may_pull(skb, n))) 3958 return skb->head; 3959 3960 return NULL; 3961 } 3962 3963 /* L2 hash helper */ 3964 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3965 { 3966 struct ethhdr *ep; 3967 3968 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 3969 if (!data) 3970 return 0; 3971 3972 ep = (struct ethhdr *)(data + mhoff); 3973 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 3974 } 3975 3976 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 3977 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 3978 { 3979 const struct ipv6hdr *iph6; 3980 const struct iphdr *iph; 3981 3982 if (l2_proto == htons(ETH_P_IP)) { 3983 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 3984 if (!data) 3985 return false; 3986 3987 iph = (const struct iphdr *)(data + *nhoff); 3988 iph_to_flow_copy_v4addrs(fk, iph); 3989 *nhoff += iph->ihl << 2; 3990 if (!ip_is_fragment(iph)) 3991 *ip_proto = iph->protocol; 3992 } else if (l2_proto == htons(ETH_P_IPV6)) { 3993 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 3994 if (!data) 3995 return false; 3996 3997 iph6 = (const struct ipv6hdr *)(data + *nhoff); 3998 iph_to_flow_copy_v6addrs(fk, iph6); 3999 *nhoff += sizeof(*iph6); 4000 *ip_proto = iph6->nexthdr; 4001 } else { 4002 return false; 4003 } 4004 4005 if (l34 && *ip_proto >= 0) 4006 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4007 4008 return true; 4009 } 4010 4011 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4012 { 4013 u32 srcmac_vendor = 0, srcmac_dev = 0; 4014 struct ethhdr *mac_hdr; 4015 u16 vlan = 0; 4016 int i; 4017 4018 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4019 if (!data) 4020 return 0; 4021 mac_hdr = (struct ethhdr *)(data + mhoff); 4022 4023 for (i = 0; i < 3; i++) 4024 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4025 4026 for (i = 3; i < ETH_ALEN; i++) 4027 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4028 4029 if (skb && skb_vlan_tag_present(skb)) 4030 vlan = skb_vlan_tag_get(skb); 4031 4032 return vlan ^ srcmac_vendor ^ srcmac_dev; 4033 } 4034 4035 /* Extract the appropriate headers based on bond's xmit policy */ 4036 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4037 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4038 { 4039 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4040 int ip_proto = -1; 4041 4042 switch (bond->params.xmit_policy) { 4043 case BOND_XMIT_POLICY_ENCAP23: 4044 case BOND_XMIT_POLICY_ENCAP34: 4045 memset(fk, 0, sizeof(*fk)); 4046 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4047 fk, data, l2_proto, nhoff, hlen, 0); 4048 default: 4049 break; 4050 } 4051 4052 fk->ports.ports = 0; 4053 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4054 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4055 return false; 4056 4057 /* ICMP error packets contains at least 8 bytes of the header 4058 * of the packet which generated the error. Use this information 4059 * to correlate ICMP error packets within the same flow which 4060 * generated the error. 4061 */ 4062 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4063 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4064 if (ip_proto == IPPROTO_ICMP) { 4065 if (!icmp_is_err(fk->icmp.type)) 4066 return true; 4067 4068 nhoff += sizeof(struct icmphdr); 4069 } else if (ip_proto == IPPROTO_ICMPV6) { 4070 if (!icmpv6_is_err(fk->icmp.type)) 4071 return true; 4072 4073 nhoff += sizeof(struct icmp6hdr); 4074 } 4075 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4076 } 4077 4078 return true; 4079 } 4080 4081 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4082 { 4083 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4084 (__force u32)flow_get_u32_src(flow); 4085 hash ^= (hash >> 16); 4086 hash ^= (hash >> 8); 4087 4088 /* discard lowest hash bit to deal with the common even ports pattern */ 4089 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4090 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4091 return hash >> 1; 4092 4093 return hash; 4094 } 4095 4096 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4097 * the data as required, but this function can be used without it if the data is 4098 * known to be linear (e.g. with xdp_buff). 4099 */ 4100 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4101 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4102 { 4103 struct flow_keys flow; 4104 u32 hash; 4105 4106 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4107 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4108 4109 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4110 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4111 return bond_eth_hash(skb, data, mhoff, hlen); 4112 4113 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4114 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4115 hash = bond_eth_hash(skb, data, mhoff, hlen); 4116 } else { 4117 if (flow.icmp.id) 4118 memcpy(&hash, &flow.icmp, sizeof(hash)); 4119 else 4120 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4121 } 4122 4123 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4124 } 4125 4126 /** 4127 * bond_xmit_hash - generate a hash value based on the xmit policy 4128 * @bond: bonding device 4129 * @skb: buffer to use for headers 4130 * 4131 * This function will extract the necessary headers from the skb buffer and use 4132 * them to generate a hash based on the xmit_policy set in the bonding device 4133 */ 4134 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4135 { 4136 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4137 skb->l4_hash) 4138 return skb->hash; 4139 4140 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4141 skb_mac_offset(skb), skb_network_offset(skb), 4142 skb_headlen(skb)); 4143 } 4144 4145 /** 4146 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4147 * @bond: bonding device 4148 * @xdp: buffer to use for headers 4149 * 4150 * The XDP variant of bond_xmit_hash. 4151 */ 4152 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4153 { 4154 struct ethhdr *eth; 4155 4156 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4157 return 0; 4158 4159 eth = (struct ethhdr *)xdp->data; 4160 4161 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4162 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4163 } 4164 4165 /*-------------------------- Device entry points ----------------------------*/ 4166 4167 void bond_work_init_all(struct bonding *bond) 4168 { 4169 INIT_DELAYED_WORK(&bond->mcast_work, 4170 bond_resend_igmp_join_requests_delayed); 4171 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4172 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4173 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4174 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4175 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4176 } 4177 4178 static void bond_work_cancel_all(struct bonding *bond) 4179 { 4180 cancel_delayed_work_sync(&bond->mii_work); 4181 cancel_delayed_work_sync(&bond->arp_work); 4182 cancel_delayed_work_sync(&bond->alb_work); 4183 cancel_delayed_work_sync(&bond->ad_work); 4184 cancel_delayed_work_sync(&bond->mcast_work); 4185 cancel_delayed_work_sync(&bond->slave_arr_work); 4186 } 4187 4188 static int bond_open(struct net_device *bond_dev) 4189 { 4190 struct bonding *bond = netdev_priv(bond_dev); 4191 struct list_head *iter; 4192 struct slave *slave; 4193 4194 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4195 bond->rr_tx_counter = alloc_percpu(u32); 4196 if (!bond->rr_tx_counter) 4197 return -ENOMEM; 4198 } 4199 4200 /* reset slave->backup and slave->inactive */ 4201 if (bond_has_slaves(bond)) { 4202 bond_for_each_slave(bond, slave, iter) { 4203 if (bond_uses_primary(bond) && 4204 slave != rcu_access_pointer(bond->curr_active_slave)) { 4205 bond_set_slave_inactive_flags(slave, 4206 BOND_SLAVE_NOTIFY_NOW); 4207 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4208 bond_set_slave_active_flags(slave, 4209 BOND_SLAVE_NOTIFY_NOW); 4210 } 4211 } 4212 } 4213 4214 if (bond_is_lb(bond)) { 4215 /* bond_alb_initialize must be called before the timer 4216 * is started. 4217 */ 4218 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4219 return -ENOMEM; 4220 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4221 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4222 } 4223 4224 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4225 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4226 4227 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4228 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4229 bond->recv_probe = bond_rcv_validate; 4230 } 4231 4232 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4233 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4234 /* register to receive LACPDUs */ 4235 bond->recv_probe = bond_3ad_lacpdu_recv; 4236 bond_3ad_initiate_agg_selection(bond, 1); 4237 4238 bond_for_each_slave(bond, slave, iter) 4239 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4240 } 4241 4242 if (bond_mode_can_use_xmit_hash(bond)) 4243 bond_update_slave_arr(bond, NULL); 4244 4245 return 0; 4246 } 4247 4248 static int bond_close(struct net_device *bond_dev) 4249 { 4250 struct bonding *bond = netdev_priv(bond_dev); 4251 struct slave *slave; 4252 4253 bond_work_cancel_all(bond); 4254 bond->send_peer_notif = 0; 4255 if (bond_is_lb(bond)) 4256 bond_alb_deinitialize(bond); 4257 bond->recv_probe = NULL; 4258 4259 if (bond_uses_primary(bond)) { 4260 rcu_read_lock(); 4261 slave = rcu_dereference(bond->curr_active_slave); 4262 if (slave) 4263 bond_hw_addr_flush(bond_dev, slave->dev); 4264 rcu_read_unlock(); 4265 } else { 4266 struct list_head *iter; 4267 4268 bond_for_each_slave(bond, slave, iter) 4269 bond_hw_addr_flush(bond_dev, slave->dev); 4270 } 4271 4272 return 0; 4273 } 4274 4275 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4276 * that some drivers can provide 32bit values only. 4277 */ 4278 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4279 const struct rtnl_link_stats64 *_new, 4280 const struct rtnl_link_stats64 *_old) 4281 { 4282 const u64 *new = (const u64 *)_new; 4283 const u64 *old = (const u64 *)_old; 4284 u64 *res = (u64 *)_res; 4285 int i; 4286 4287 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4288 u64 nv = new[i]; 4289 u64 ov = old[i]; 4290 s64 delta = nv - ov; 4291 4292 /* detects if this particular field is 32bit only */ 4293 if (((nv | ov) >> 32) == 0) 4294 delta = (s64)(s32)((u32)nv - (u32)ov); 4295 4296 /* filter anomalies, some drivers reset their stats 4297 * at down/up events. 4298 */ 4299 if (delta > 0) 4300 res[i] += delta; 4301 } 4302 } 4303 4304 #ifdef CONFIG_LOCKDEP 4305 static int bond_get_lowest_level_rcu(struct net_device *dev) 4306 { 4307 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4308 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4309 int cur = 0, max = 0; 4310 4311 now = dev; 4312 iter = &dev->adj_list.lower; 4313 4314 while (1) { 4315 next = NULL; 4316 while (1) { 4317 ldev = netdev_next_lower_dev_rcu(now, &iter); 4318 if (!ldev) 4319 break; 4320 4321 next = ldev; 4322 niter = &ldev->adj_list.lower; 4323 dev_stack[cur] = now; 4324 iter_stack[cur++] = iter; 4325 if (max <= cur) 4326 max = cur; 4327 break; 4328 } 4329 4330 if (!next) { 4331 if (!cur) 4332 return max; 4333 next = dev_stack[--cur]; 4334 niter = iter_stack[cur]; 4335 } 4336 4337 now = next; 4338 iter = niter; 4339 } 4340 4341 return max; 4342 } 4343 #endif 4344 4345 static void bond_get_stats(struct net_device *bond_dev, 4346 struct rtnl_link_stats64 *stats) 4347 { 4348 struct bonding *bond = netdev_priv(bond_dev); 4349 struct rtnl_link_stats64 temp; 4350 struct list_head *iter; 4351 struct slave *slave; 4352 int nest_level = 0; 4353 4354 4355 rcu_read_lock(); 4356 #ifdef CONFIG_LOCKDEP 4357 nest_level = bond_get_lowest_level_rcu(bond_dev); 4358 #endif 4359 4360 spin_lock_nested(&bond->stats_lock, nest_level); 4361 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4362 4363 bond_for_each_slave_rcu(bond, slave, iter) { 4364 const struct rtnl_link_stats64 *new = 4365 dev_get_stats(slave->dev, &temp); 4366 4367 bond_fold_stats(stats, new, &slave->slave_stats); 4368 4369 /* save off the slave stats for the next run */ 4370 memcpy(&slave->slave_stats, new, sizeof(*new)); 4371 } 4372 4373 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4374 spin_unlock(&bond->stats_lock); 4375 rcu_read_unlock(); 4376 } 4377 4378 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4379 { 4380 struct bonding *bond = netdev_priv(bond_dev); 4381 struct mii_ioctl_data *mii = NULL; 4382 const struct net_device_ops *ops; 4383 struct net_device *real_dev; 4384 struct hwtstamp_config cfg; 4385 struct ifreq ifrr; 4386 int res = 0; 4387 4388 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4389 4390 switch (cmd) { 4391 case SIOCGMIIPHY: 4392 mii = if_mii(ifr); 4393 if (!mii) 4394 return -EINVAL; 4395 4396 mii->phy_id = 0; 4397 fallthrough; 4398 case SIOCGMIIREG: 4399 /* We do this again just in case we were called by SIOCGMIIREG 4400 * instead of SIOCGMIIPHY. 4401 */ 4402 mii = if_mii(ifr); 4403 if (!mii) 4404 return -EINVAL; 4405 4406 if (mii->reg_num == 1) { 4407 mii->val_out = 0; 4408 if (netif_carrier_ok(bond->dev)) 4409 mii->val_out = BMSR_LSTATUS; 4410 } 4411 4412 break; 4413 case SIOCSHWTSTAMP: 4414 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4415 return -EFAULT; 4416 4417 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 4418 return -EOPNOTSUPP; 4419 4420 fallthrough; 4421 case SIOCGHWTSTAMP: 4422 real_dev = bond_option_active_slave_get_rcu(bond); 4423 if (!real_dev) 4424 return -EOPNOTSUPP; 4425 4426 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ); 4427 ifrr.ifr_ifru = ifr->ifr_ifru; 4428 4429 ops = real_dev->netdev_ops; 4430 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) { 4431 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd); 4432 if (res) 4433 return res; 4434 4435 ifr->ifr_ifru = ifrr.ifr_ifru; 4436 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4437 return -EFAULT; 4438 4439 /* Set the BOND_PHC_INDEX flag to notify user space */ 4440 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 4441 4442 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ? 4443 -EFAULT : 0; 4444 } 4445 fallthrough; 4446 default: 4447 res = -EOPNOTSUPP; 4448 } 4449 4450 return res; 4451 } 4452 4453 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4454 { 4455 struct bonding *bond = netdev_priv(bond_dev); 4456 struct net_device *slave_dev = NULL; 4457 struct ifbond k_binfo; 4458 struct ifbond __user *u_binfo = NULL; 4459 struct ifslave k_sinfo; 4460 struct ifslave __user *u_sinfo = NULL; 4461 struct bond_opt_value newval; 4462 struct net *net; 4463 int res = 0; 4464 4465 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4466 4467 switch (cmd) { 4468 case SIOCBONDINFOQUERY: 4469 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4470 4471 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4472 return -EFAULT; 4473 4474 bond_info_query(bond_dev, &k_binfo); 4475 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4476 return -EFAULT; 4477 4478 return 0; 4479 case SIOCBONDSLAVEINFOQUERY: 4480 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4481 4482 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4483 return -EFAULT; 4484 4485 res = bond_slave_info_query(bond_dev, &k_sinfo); 4486 if (res == 0 && 4487 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4488 return -EFAULT; 4489 4490 return res; 4491 default: 4492 break; 4493 } 4494 4495 net = dev_net(bond_dev); 4496 4497 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4498 return -EPERM; 4499 4500 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4501 4502 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4503 4504 if (!slave_dev) 4505 return -ENODEV; 4506 4507 switch (cmd) { 4508 case SIOCBONDENSLAVE: 4509 res = bond_enslave(bond_dev, slave_dev, NULL); 4510 break; 4511 case SIOCBONDRELEASE: 4512 res = bond_release(bond_dev, slave_dev); 4513 break; 4514 case SIOCBONDSETHWADDR: 4515 res = bond_set_dev_addr(bond_dev, slave_dev); 4516 break; 4517 case SIOCBONDCHANGEACTIVE: 4518 bond_opt_initstr(&newval, slave_dev->name); 4519 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4520 &newval); 4521 break; 4522 default: 4523 res = -EOPNOTSUPP; 4524 } 4525 4526 return res; 4527 } 4528 4529 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4530 void __user *data, int cmd) 4531 { 4532 struct ifreq ifrdata = { .ifr_data = data }; 4533 4534 switch (cmd) { 4535 case BOND_INFO_QUERY_OLD: 4536 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4537 case BOND_SLAVE_INFO_QUERY_OLD: 4538 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4539 case BOND_ENSLAVE_OLD: 4540 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4541 case BOND_RELEASE_OLD: 4542 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4543 case BOND_SETHWADDR_OLD: 4544 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4545 case BOND_CHANGE_ACTIVE_OLD: 4546 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4547 } 4548 4549 return -EOPNOTSUPP; 4550 } 4551 4552 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4553 { 4554 struct bonding *bond = netdev_priv(bond_dev); 4555 4556 if (change & IFF_PROMISC) 4557 bond_set_promiscuity(bond, 4558 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4559 4560 if (change & IFF_ALLMULTI) 4561 bond_set_allmulti(bond, 4562 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4563 } 4564 4565 static void bond_set_rx_mode(struct net_device *bond_dev) 4566 { 4567 struct bonding *bond = netdev_priv(bond_dev); 4568 struct list_head *iter; 4569 struct slave *slave; 4570 4571 rcu_read_lock(); 4572 if (bond_uses_primary(bond)) { 4573 slave = rcu_dereference(bond->curr_active_slave); 4574 if (slave) { 4575 dev_uc_sync(slave->dev, bond_dev); 4576 dev_mc_sync(slave->dev, bond_dev); 4577 } 4578 } else { 4579 bond_for_each_slave_rcu(bond, slave, iter) { 4580 dev_uc_sync_multiple(slave->dev, bond_dev); 4581 dev_mc_sync_multiple(slave->dev, bond_dev); 4582 } 4583 } 4584 rcu_read_unlock(); 4585 } 4586 4587 static int bond_neigh_init(struct neighbour *n) 4588 { 4589 struct bonding *bond = netdev_priv(n->dev); 4590 const struct net_device_ops *slave_ops; 4591 struct neigh_parms parms; 4592 struct slave *slave; 4593 int ret = 0; 4594 4595 rcu_read_lock(); 4596 slave = bond_first_slave_rcu(bond); 4597 if (!slave) 4598 goto out; 4599 slave_ops = slave->dev->netdev_ops; 4600 if (!slave_ops->ndo_neigh_setup) 4601 goto out; 4602 4603 /* TODO: find another way [1] to implement this. 4604 * Passing a zeroed structure is fragile, 4605 * but at least we do not pass garbage. 4606 * 4607 * [1] One way would be that ndo_neigh_setup() never touch 4608 * struct neigh_parms, but propagate the new neigh_setup() 4609 * back to ___neigh_create() / neigh_parms_alloc() 4610 */ 4611 memset(&parms, 0, sizeof(parms)); 4612 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4613 4614 if (ret) 4615 goto out; 4616 4617 if (parms.neigh_setup) 4618 ret = parms.neigh_setup(n); 4619 out: 4620 rcu_read_unlock(); 4621 return ret; 4622 } 4623 4624 /* The bonding ndo_neigh_setup is called at init time beofre any 4625 * slave exists. So we must declare proxy setup function which will 4626 * be used at run time to resolve the actual slave neigh param setup. 4627 * 4628 * It's also called by master devices (such as vlans) to setup their 4629 * underlying devices. In that case - do nothing, we're already set up from 4630 * our init. 4631 */ 4632 static int bond_neigh_setup(struct net_device *dev, 4633 struct neigh_parms *parms) 4634 { 4635 /* modify only our neigh_parms */ 4636 if (parms->dev == dev) 4637 parms->neigh_setup = bond_neigh_init; 4638 4639 return 0; 4640 } 4641 4642 /* Change the MTU of all of a master's slaves to match the master */ 4643 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4644 { 4645 struct bonding *bond = netdev_priv(bond_dev); 4646 struct slave *slave, *rollback_slave; 4647 struct list_head *iter; 4648 int res = 0; 4649 4650 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4651 4652 bond_for_each_slave(bond, slave, iter) { 4653 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4654 slave, slave->dev->netdev_ops->ndo_change_mtu); 4655 4656 res = dev_set_mtu(slave->dev, new_mtu); 4657 4658 if (res) { 4659 /* If we failed to set the slave's mtu to the new value 4660 * we must abort the operation even in ACTIVE_BACKUP 4661 * mode, because if we allow the backup slaves to have 4662 * different mtu values than the active slave we'll 4663 * need to change their mtu when doing a failover. That 4664 * means changing their mtu from timer context, which 4665 * is probably not a good idea. 4666 */ 4667 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4668 res, new_mtu); 4669 goto unwind; 4670 } 4671 } 4672 4673 bond_dev->mtu = new_mtu; 4674 4675 return 0; 4676 4677 unwind: 4678 /* unwind from head to the slave that failed */ 4679 bond_for_each_slave(bond, rollback_slave, iter) { 4680 int tmp_res; 4681 4682 if (rollback_slave == slave) 4683 break; 4684 4685 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4686 if (tmp_res) 4687 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4688 tmp_res); 4689 } 4690 4691 return res; 4692 } 4693 4694 /* Change HW address 4695 * 4696 * Note that many devices must be down to change the HW address, and 4697 * downing the master releases all slaves. We can make bonds full of 4698 * bonding devices to test this, however. 4699 */ 4700 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4701 { 4702 struct bonding *bond = netdev_priv(bond_dev); 4703 struct slave *slave, *rollback_slave; 4704 struct sockaddr_storage *ss = addr, tmp_ss; 4705 struct list_head *iter; 4706 int res = 0; 4707 4708 if (BOND_MODE(bond) == BOND_MODE_ALB) 4709 return bond_alb_set_mac_address(bond_dev, addr); 4710 4711 4712 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4713 4714 /* If fail_over_mac is enabled, do nothing and return success. 4715 * Returning an error causes ifenslave to fail. 4716 */ 4717 if (bond->params.fail_over_mac && 4718 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4719 return 0; 4720 4721 if (!is_valid_ether_addr(ss->__data)) 4722 return -EADDRNOTAVAIL; 4723 4724 bond_for_each_slave(bond, slave, iter) { 4725 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4726 __func__, slave); 4727 res = dev_set_mac_address(slave->dev, addr, NULL); 4728 if (res) { 4729 /* TODO: consider downing the slave 4730 * and retry ? 4731 * User should expect communications 4732 * breakage anyway until ARP finish 4733 * updating, so... 4734 */ 4735 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4736 __func__, res); 4737 goto unwind; 4738 } 4739 } 4740 4741 /* success */ 4742 dev_addr_set(bond_dev, ss->__data); 4743 return 0; 4744 4745 unwind: 4746 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4747 tmp_ss.ss_family = bond_dev->type; 4748 4749 /* unwind from head to the slave that failed */ 4750 bond_for_each_slave(bond, rollback_slave, iter) { 4751 int tmp_res; 4752 4753 if (rollback_slave == slave) 4754 break; 4755 4756 tmp_res = dev_set_mac_address(rollback_slave->dev, 4757 (struct sockaddr *)&tmp_ss, NULL); 4758 if (tmp_res) { 4759 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4760 __func__, tmp_res); 4761 } 4762 } 4763 4764 return res; 4765 } 4766 4767 /** 4768 * bond_get_slave_by_id - get xmit slave with slave_id 4769 * @bond: bonding device that is transmitting 4770 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4771 * 4772 * This function tries to get slave with slave_id but in case 4773 * it fails, it tries to find the first available slave for transmission. 4774 */ 4775 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4776 int slave_id) 4777 { 4778 struct list_head *iter; 4779 struct slave *slave; 4780 int i = slave_id; 4781 4782 /* Here we start from the slave with slave_id */ 4783 bond_for_each_slave_rcu(bond, slave, iter) { 4784 if (--i < 0) { 4785 if (bond_slave_can_tx(slave)) 4786 return slave; 4787 } 4788 } 4789 4790 /* Here we start from the first slave up to slave_id */ 4791 i = slave_id; 4792 bond_for_each_slave_rcu(bond, slave, iter) { 4793 if (--i < 0) 4794 break; 4795 if (bond_slave_can_tx(slave)) 4796 return slave; 4797 } 4798 /* no slave that can tx has been found */ 4799 return NULL; 4800 } 4801 4802 /** 4803 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4804 * @bond: bonding device to use 4805 * 4806 * Based on the value of the bonding device's packets_per_slave parameter 4807 * this function generates a slave id, which is usually used as the next 4808 * slave to transmit through. 4809 */ 4810 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4811 { 4812 u32 slave_id; 4813 struct reciprocal_value reciprocal_packets_per_slave; 4814 int packets_per_slave = bond->params.packets_per_slave; 4815 4816 switch (packets_per_slave) { 4817 case 0: 4818 slave_id = get_random_u32(); 4819 break; 4820 case 1: 4821 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4822 break; 4823 default: 4824 reciprocal_packets_per_slave = 4825 bond->params.reciprocal_packets_per_slave; 4826 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4827 slave_id = reciprocal_divide(slave_id, 4828 reciprocal_packets_per_slave); 4829 break; 4830 } 4831 4832 return slave_id; 4833 } 4834 4835 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4836 struct sk_buff *skb) 4837 { 4838 struct slave *slave; 4839 int slave_cnt; 4840 u32 slave_id; 4841 4842 /* Start with the curr_active_slave that joined the bond as the 4843 * default for sending IGMP traffic. For failover purposes one 4844 * needs to maintain some consistency for the interface that will 4845 * send the join/membership reports. The curr_active_slave found 4846 * will send all of this type of traffic. 4847 */ 4848 if (skb->protocol == htons(ETH_P_IP)) { 4849 int noff = skb_network_offset(skb); 4850 struct iphdr *iph; 4851 4852 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4853 goto non_igmp; 4854 4855 iph = ip_hdr(skb); 4856 if (iph->protocol == IPPROTO_IGMP) { 4857 slave = rcu_dereference(bond->curr_active_slave); 4858 if (slave) 4859 return slave; 4860 return bond_get_slave_by_id(bond, 0); 4861 } 4862 } 4863 4864 non_igmp: 4865 slave_cnt = READ_ONCE(bond->slave_cnt); 4866 if (likely(slave_cnt)) { 4867 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4868 return bond_get_slave_by_id(bond, slave_id); 4869 } 4870 return NULL; 4871 } 4872 4873 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4874 struct xdp_buff *xdp) 4875 { 4876 struct slave *slave; 4877 int slave_cnt; 4878 u32 slave_id; 4879 const struct ethhdr *eth; 4880 void *data = xdp->data; 4881 4882 if (data + sizeof(struct ethhdr) > xdp->data_end) 4883 goto non_igmp; 4884 4885 eth = (struct ethhdr *)data; 4886 data += sizeof(struct ethhdr); 4887 4888 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4889 if (eth->h_proto == htons(ETH_P_IP)) { 4890 const struct iphdr *iph; 4891 4892 if (data + sizeof(struct iphdr) > xdp->data_end) 4893 goto non_igmp; 4894 4895 iph = (struct iphdr *)data; 4896 4897 if (iph->protocol == IPPROTO_IGMP) { 4898 slave = rcu_dereference(bond->curr_active_slave); 4899 if (slave) 4900 return slave; 4901 return bond_get_slave_by_id(bond, 0); 4902 } 4903 } 4904 4905 non_igmp: 4906 slave_cnt = READ_ONCE(bond->slave_cnt); 4907 if (likely(slave_cnt)) { 4908 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4909 return bond_get_slave_by_id(bond, slave_id); 4910 } 4911 return NULL; 4912 } 4913 4914 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4915 struct net_device *bond_dev) 4916 { 4917 struct bonding *bond = netdev_priv(bond_dev); 4918 struct slave *slave; 4919 4920 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4921 if (likely(slave)) 4922 return bond_dev_queue_xmit(bond, skb, slave->dev); 4923 4924 return bond_tx_drop(bond_dev, skb); 4925 } 4926 4927 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4928 { 4929 return rcu_dereference(bond->curr_active_slave); 4930 } 4931 4932 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4933 * the bond has a usable interface. 4934 */ 4935 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4936 struct net_device *bond_dev) 4937 { 4938 struct bonding *bond = netdev_priv(bond_dev); 4939 struct slave *slave; 4940 4941 slave = bond_xmit_activebackup_slave_get(bond); 4942 if (slave) 4943 return bond_dev_queue_xmit(bond, skb, slave->dev); 4944 4945 return bond_tx_drop(bond_dev, skb); 4946 } 4947 4948 /* Use this to update slave_array when (a) it's not appropriate to update 4949 * slave_array right away (note that update_slave_array() may sleep) 4950 * and / or (b) RTNL is not held. 4951 */ 4952 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4953 { 4954 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4955 } 4956 4957 /* Slave array work handler. Holds only RTNL */ 4958 static void bond_slave_arr_handler(struct work_struct *work) 4959 { 4960 struct bonding *bond = container_of(work, struct bonding, 4961 slave_arr_work.work); 4962 int ret; 4963 4964 if (!rtnl_trylock()) 4965 goto err; 4966 4967 ret = bond_update_slave_arr(bond, NULL); 4968 rtnl_unlock(); 4969 if (ret) { 4970 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4971 goto err; 4972 } 4973 return; 4974 4975 err: 4976 bond_slave_arr_work_rearm(bond, 1); 4977 } 4978 4979 static void bond_skip_slave(struct bond_up_slave *slaves, 4980 struct slave *skipslave) 4981 { 4982 int idx; 4983 4984 /* Rare situation where caller has asked to skip a specific 4985 * slave but allocation failed (most likely!). BTW this is 4986 * only possible when the call is initiated from 4987 * __bond_release_one(). In this situation; overwrite the 4988 * skipslave entry in the array with the last entry from the 4989 * array to avoid a situation where the xmit path may choose 4990 * this to-be-skipped slave to send a packet out. 4991 */ 4992 for (idx = 0; slaves && idx < slaves->count; idx++) { 4993 if (skipslave == slaves->arr[idx]) { 4994 slaves->arr[idx] = 4995 slaves->arr[slaves->count - 1]; 4996 slaves->count--; 4997 break; 4998 } 4999 } 5000 } 5001 5002 static void bond_set_slave_arr(struct bonding *bond, 5003 struct bond_up_slave *usable_slaves, 5004 struct bond_up_slave *all_slaves) 5005 { 5006 struct bond_up_slave *usable, *all; 5007 5008 usable = rtnl_dereference(bond->usable_slaves); 5009 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5010 kfree_rcu(usable, rcu); 5011 5012 all = rtnl_dereference(bond->all_slaves); 5013 rcu_assign_pointer(bond->all_slaves, all_slaves); 5014 kfree_rcu(all, rcu); 5015 } 5016 5017 static void bond_reset_slave_arr(struct bonding *bond) 5018 { 5019 struct bond_up_slave *usable, *all; 5020 5021 usable = rtnl_dereference(bond->usable_slaves); 5022 if (usable) { 5023 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5024 kfree_rcu(usable, rcu); 5025 } 5026 5027 all = rtnl_dereference(bond->all_slaves); 5028 if (all) { 5029 RCU_INIT_POINTER(bond->all_slaves, NULL); 5030 kfree_rcu(all, rcu); 5031 } 5032 } 5033 5034 /* Build the usable slaves array in control path for modes that use xmit-hash 5035 * to determine the slave interface - 5036 * (a) BOND_MODE_8023AD 5037 * (b) BOND_MODE_XOR 5038 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5039 * 5040 * The caller is expected to hold RTNL only and NO other lock! 5041 */ 5042 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5043 { 5044 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5045 struct slave *slave; 5046 struct list_head *iter; 5047 int agg_id = 0; 5048 int ret = 0; 5049 5050 might_sleep(); 5051 5052 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5053 bond->slave_cnt), GFP_KERNEL); 5054 all_slaves = kzalloc(struct_size(all_slaves, arr, 5055 bond->slave_cnt), GFP_KERNEL); 5056 if (!usable_slaves || !all_slaves) { 5057 ret = -ENOMEM; 5058 goto out; 5059 } 5060 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5061 struct ad_info ad_info; 5062 5063 spin_lock_bh(&bond->mode_lock); 5064 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5065 spin_unlock_bh(&bond->mode_lock); 5066 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5067 /* No active aggragator means it's not safe to use 5068 * the previous array. 5069 */ 5070 bond_reset_slave_arr(bond); 5071 goto out; 5072 } 5073 spin_unlock_bh(&bond->mode_lock); 5074 agg_id = ad_info.aggregator_id; 5075 } 5076 bond_for_each_slave(bond, slave, iter) { 5077 if (skipslave == slave) 5078 continue; 5079 5080 all_slaves->arr[all_slaves->count++] = slave; 5081 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5082 struct aggregator *agg; 5083 5084 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5085 if (!agg || agg->aggregator_identifier != agg_id) 5086 continue; 5087 } 5088 if (!bond_slave_can_tx(slave)) 5089 continue; 5090 5091 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5092 usable_slaves->count); 5093 5094 usable_slaves->arr[usable_slaves->count++] = slave; 5095 } 5096 5097 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5098 return ret; 5099 out: 5100 if (ret != 0 && skipslave) { 5101 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5102 skipslave); 5103 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5104 skipslave); 5105 } 5106 kfree_rcu(all_slaves, rcu); 5107 kfree_rcu(usable_slaves, rcu); 5108 5109 return ret; 5110 } 5111 5112 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5113 struct sk_buff *skb, 5114 struct bond_up_slave *slaves) 5115 { 5116 struct slave *slave; 5117 unsigned int count; 5118 u32 hash; 5119 5120 hash = bond_xmit_hash(bond, skb); 5121 count = slaves ? READ_ONCE(slaves->count) : 0; 5122 if (unlikely(!count)) 5123 return NULL; 5124 5125 slave = slaves->arr[hash % count]; 5126 return slave; 5127 } 5128 5129 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5130 struct xdp_buff *xdp) 5131 { 5132 struct bond_up_slave *slaves; 5133 unsigned int count; 5134 u32 hash; 5135 5136 hash = bond_xmit_hash_xdp(bond, xdp); 5137 slaves = rcu_dereference(bond->usable_slaves); 5138 count = slaves ? READ_ONCE(slaves->count) : 0; 5139 if (unlikely(!count)) 5140 return NULL; 5141 5142 return slaves->arr[hash % count]; 5143 } 5144 5145 /* Use this Xmit function for 3AD as well as XOR modes. The current 5146 * usable slave array is formed in the control path. The xmit function 5147 * just calculates hash and sends the packet out. 5148 */ 5149 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5150 struct net_device *dev) 5151 { 5152 struct bonding *bond = netdev_priv(dev); 5153 struct bond_up_slave *slaves; 5154 struct slave *slave; 5155 5156 slaves = rcu_dereference(bond->usable_slaves); 5157 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5158 if (likely(slave)) 5159 return bond_dev_queue_xmit(bond, skb, slave->dev); 5160 5161 return bond_tx_drop(dev, skb); 5162 } 5163 5164 /* in broadcast mode, we send everything to all usable interfaces. */ 5165 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5166 struct net_device *bond_dev) 5167 { 5168 struct bonding *bond = netdev_priv(bond_dev); 5169 struct slave *slave = NULL; 5170 struct list_head *iter; 5171 bool xmit_suc = false; 5172 bool skb_used = false; 5173 5174 bond_for_each_slave_rcu(bond, slave, iter) { 5175 struct sk_buff *skb2; 5176 5177 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5178 continue; 5179 5180 if (bond_is_last_slave(bond, slave)) { 5181 skb2 = skb; 5182 skb_used = true; 5183 } else { 5184 skb2 = skb_clone(skb, GFP_ATOMIC); 5185 if (!skb2) { 5186 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5187 bond_dev->name, __func__); 5188 continue; 5189 } 5190 } 5191 5192 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5193 xmit_suc = true; 5194 } 5195 5196 if (!skb_used) 5197 dev_kfree_skb_any(skb); 5198 5199 if (xmit_suc) 5200 return NETDEV_TX_OK; 5201 5202 dev_core_stats_tx_dropped_inc(bond_dev); 5203 return NET_XMIT_DROP; 5204 } 5205 5206 /*------------------------- Device initialization ---------------------------*/ 5207 5208 /* Lookup the slave that corresponds to a qid */ 5209 static inline int bond_slave_override(struct bonding *bond, 5210 struct sk_buff *skb) 5211 { 5212 struct slave *slave = NULL; 5213 struct list_head *iter; 5214 5215 if (!skb_rx_queue_recorded(skb)) 5216 return 1; 5217 5218 /* Find out if any slaves have the same mapping as this skb. */ 5219 bond_for_each_slave_rcu(bond, slave, iter) { 5220 if (slave->queue_id == skb_get_queue_mapping(skb)) { 5221 if (bond_slave_is_up(slave) && 5222 slave->link == BOND_LINK_UP) { 5223 bond_dev_queue_xmit(bond, skb, slave->dev); 5224 return 0; 5225 } 5226 /* If the slave isn't UP, use default transmit policy. */ 5227 break; 5228 } 5229 } 5230 5231 return 1; 5232 } 5233 5234 5235 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5236 struct net_device *sb_dev) 5237 { 5238 /* This helper function exists to help dev_pick_tx get the correct 5239 * destination queue. Using a helper function skips a call to 5240 * skb_tx_hash and will put the skbs in the queue we expect on their 5241 * way down to the bonding driver. 5242 */ 5243 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5244 5245 /* Save the original txq to restore before passing to the driver */ 5246 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5247 5248 if (unlikely(txq >= dev->real_num_tx_queues)) { 5249 do { 5250 txq -= dev->real_num_tx_queues; 5251 } while (txq >= dev->real_num_tx_queues); 5252 } 5253 return txq; 5254 } 5255 5256 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5257 struct sk_buff *skb, 5258 bool all_slaves) 5259 { 5260 struct bonding *bond = netdev_priv(master_dev); 5261 struct bond_up_slave *slaves; 5262 struct slave *slave = NULL; 5263 5264 switch (BOND_MODE(bond)) { 5265 case BOND_MODE_ROUNDROBIN: 5266 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5267 break; 5268 case BOND_MODE_ACTIVEBACKUP: 5269 slave = bond_xmit_activebackup_slave_get(bond); 5270 break; 5271 case BOND_MODE_8023AD: 5272 case BOND_MODE_XOR: 5273 if (all_slaves) 5274 slaves = rcu_dereference(bond->all_slaves); 5275 else 5276 slaves = rcu_dereference(bond->usable_slaves); 5277 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5278 break; 5279 case BOND_MODE_BROADCAST: 5280 break; 5281 case BOND_MODE_ALB: 5282 slave = bond_xmit_alb_slave_get(bond, skb); 5283 break; 5284 case BOND_MODE_TLB: 5285 slave = bond_xmit_tlb_slave_get(bond, skb); 5286 break; 5287 default: 5288 /* Should never happen, mode already checked */ 5289 WARN_ONCE(true, "Unknown bonding mode"); 5290 break; 5291 } 5292 5293 if (slave) 5294 return slave->dev; 5295 return NULL; 5296 } 5297 5298 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5299 { 5300 switch (sk->sk_family) { 5301 #if IS_ENABLED(CONFIG_IPV6) 5302 case AF_INET6: 5303 if (ipv6_only_sock(sk) || 5304 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5305 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5306 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5307 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5308 break; 5309 } 5310 fallthrough; 5311 #endif 5312 default: /* AF_INET */ 5313 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5314 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5315 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5316 break; 5317 } 5318 5319 flow->ports.src = inet_sk(sk)->inet_sport; 5320 flow->ports.dst = inet_sk(sk)->inet_dport; 5321 } 5322 5323 /** 5324 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5325 * @sk: socket to use for headers 5326 * 5327 * This function will extract the necessary field from the socket and use 5328 * them to generate a hash based on the LAYER34 xmit_policy. 5329 * Assumes that sk is a TCP or UDP socket. 5330 */ 5331 static u32 bond_sk_hash_l34(struct sock *sk) 5332 { 5333 struct flow_keys flow; 5334 u32 hash; 5335 5336 bond_sk_to_flow(sk, &flow); 5337 5338 /* L4 */ 5339 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5340 /* L3 */ 5341 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5342 } 5343 5344 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5345 struct sock *sk) 5346 { 5347 struct bond_up_slave *slaves; 5348 struct slave *slave; 5349 unsigned int count; 5350 u32 hash; 5351 5352 slaves = rcu_dereference(bond->usable_slaves); 5353 count = slaves ? READ_ONCE(slaves->count) : 0; 5354 if (unlikely(!count)) 5355 return NULL; 5356 5357 hash = bond_sk_hash_l34(sk); 5358 slave = slaves->arr[hash % count]; 5359 5360 return slave->dev; 5361 } 5362 5363 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5364 struct sock *sk) 5365 { 5366 struct bonding *bond = netdev_priv(dev); 5367 struct net_device *lower = NULL; 5368 5369 rcu_read_lock(); 5370 if (bond_sk_check(bond)) 5371 lower = __bond_sk_get_lower_dev(bond, sk); 5372 rcu_read_unlock(); 5373 5374 return lower; 5375 } 5376 5377 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5378 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5379 struct net_device *dev) 5380 { 5381 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5382 5383 /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded 5384 * was true, if tls_device_down is running in parallel, but it's OK, 5385 * because bond_get_slave_by_dev has a NULL check. 5386 */ 5387 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5388 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5389 return bond_tx_drop(dev, skb); 5390 } 5391 #endif 5392 5393 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5394 { 5395 struct bonding *bond = netdev_priv(dev); 5396 5397 if (bond_should_override_tx_queue(bond) && 5398 !bond_slave_override(bond, skb)) 5399 return NETDEV_TX_OK; 5400 5401 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5402 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk)) 5403 return bond_tls_device_xmit(bond, skb, dev); 5404 #endif 5405 5406 switch (BOND_MODE(bond)) { 5407 case BOND_MODE_ROUNDROBIN: 5408 return bond_xmit_roundrobin(skb, dev); 5409 case BOND_MODE_ACTIVEBACKUP: 5410 return bond_xmit_activebackup(skb, dev); 5411 case BOND_MODE_8023AD: 5412 case BOND_MODE_XOR: 5413 return bond_3ad_xor_xmit(skb, dev); 5414 case BOND_MODE_BROADCAST: 5415 return bond_xmit_broadcast(skb, dev); 5416 case BOND_MODE_ALB: 5417 return bond_alb_xmit(skb, dev); 5418 case BOND_MODE_TLB: 5419 return bond_tlb_xmit(skb, dev); 5420 default: 5421 /* Should never happen, mode already checked */ 5422 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5423 WARN_ON_ONCE(1); 5424 return bond_tx_drop(dev, skb); 5425 } 5426 } 5427 5428 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5429 { 5430 struct bonding *bond = netdev_priv(dev); 5431 netdev_tx_t ret = NETDEV_TX_OK; 5432 5433 /* If we risk deadlock from transmitting this in the 5434 * netpoll path, tell netpoll to queue the frame for later tx 5435 */ 5436 if (unlikely(is_netpoll_tx_blocked(dev))) 5437 return NETDEV_TX_BUSY; 5438 5439 rcu_read_lock(); 5440 if (bond_has_slaves(bond)) 5441 ret = __bond_start_xmit(skb, dev); 5442 else 5443 ret = bond_tx_drop(dev, skb); 5444 rcu_read_unlock(); 5445 5446 return ret; 5447 } 5448 5449 static struct net_device * 5450 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5451 { 5452 struct bonding *bond = netdev_priv(bond_dev); 5453 struct slave *slave; 5454 5455 /* Caller needs to hold rcu_read_lock() */ 5456 5457 switch (BOND_MODE(bond)) { 5458 case BOND_MODE_ROUNDROBIN: 5459 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5460 break; 5461 5462 case BOND_MODE_ACTIVEBACKUP: 5463 slave = bond_xmit_activebackup_slave_get(bond); 5464 break; 5465 5466 case BOND_MODE_8023AD: 5467 case BOND_MODE_XOR: 5468 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5469 break; 5470 5471 default: 5472 /* Should never happen. Mode guarded by bond_xdp_check() */ 5473 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond)); 5474 WARN_ON_ONCE(1); 5475 return NULL; 5476 } 5477 5478 if (slave) 5479 return slave->dev; 5480 5481 return NULL; 5482 } 5483 5484 static int bond_xdp_xmit(struct net_device *bond_dev, 5485 int n, struct xdp_frame **frames, u32 flags) 5486 { 5487 int nxmit, err = -ENXIO; 5488 5489 rcu_read_lock(); 5490 5491 for (nxmit = 0; nxmit < n; nxmit++) { 5492 struct xdp_frame *frame = frames[nxmit]; 5493 struct xdp_frame *frames1[] = {frame}; 5494 struct net_device *slave_dev; 5495 struct xdp_buff xdp; 5496 5497 xdp_convert_frame_to_buff(frame, &xdp); 5498 5499 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5500 if (!slave_dev) { 5501 err = -ENXIO; 5502 break; 5503 } 5504 5505 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5506 if (err < 1) 5507 break; 5508 } 5509 5510 rcu_read_unlock(); 5511 5512 /* If error happened on the first frame then we can pass the error up, otherwise 5513 * report the number of frames that were xmitted. 5514 */ 5515 if (err < 0) 5516 return (nxmit == 0 ? err : nxmit); 5517 5518 return nxmit; 5519 } 5520 5521 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5522 struct netlink_ext_ack *extack) 5523 { 5524 struct bonding *bond = netdev_priv(dev); 5525 struct list_head *iter; 5526 struct slave *slave, *rollback_slave; 5527 struct bpf_prog *old_prog; 5528 struct netdev_bpf xdp = { 5529 .command = XDP_SETUP_PROG, 5530 .flags = 0, 5531 .prog = prog, 5532 .extack = extack, 5533 }; 5534 int err; 5535 5536 ASSERT_RTNL(); 5537 5538 if (!bond_xdp_check(bond)) 5539 return -EOPNOTSUPP; 5540 5541 old_prog = bond->xdp_prog; 5542 bond->xdp_prog = prog; 5543 5544 bond_for_each_slave(bond, slave, iter) { 5545 struct net_device *slave_dev = slave->dev; 5546 5547 if (!slave_dev->netdev_ops->ndo_bpf || 5548 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5549 SLAVE_NL_ERR(dev, slave_dev, extack, 5550 "Slave device does not support XDP"); 5551 err = -EOPNOTSUPP; 5552 goto err; 5553 } 5554 5555 if (dev_xdp_prog_count(slave_dev) > 0) { 5556 SLAVE_NL_ERR(dev, slave_dev, extack, 5557 "Slave has XDP program loaded, please unload before enslaving"); 5558 err = -EOPNOTSUPP; 5559 goto err; 5560 } 5561 5562 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5563 if (err < 0) { 5564 /* ndo_bpf() sets extack error message */ 5565 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5566 goto err; 5567 } 5568 if (prog) 5569 bpf_prog_inc(prog); 5570 } 5571 5572 if (prog) { 5573 static_branch_inc(&bpf_master_redirect_enabled_key); 5574 } else if (old_prog) { 5575 bpf_prog_put(old_prog); 5576 static_branch_dec(&bpf_master_redirect_enabled_key); 5577 } 5578 5579 return 0; 5580 5581 err: 5582 /* unwind the program changes */ 5583 bond->xdp_prog = old_prog; 5584 xdp.prog = old_prog; 5585 xdp.extack = NULL; /* do not overwrite original error */ 5586 5587 bond_for_each_slave(bond, rollback_slave, iter) { 5588 struct net_device *slave_dev = rollback_slave->dev; 5589 int err_unwind; 5590 5591 if (slave == rollback_slave) 5592 break; 5593 5594 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5595 if (err_unwind < 0) 5596 slave_err(dev, slave_dev, 5597 "Error %d when unwinding XDP program change\n", err_unwind); 5598 else if (xdp.prog) 5599 bpf_prog_inc(xdp.prog); 5600 } 5601 return err; 5602 } 5603 5604 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5605 { 5606 switch (xdp->command) { 5607 case XDP_SETUP_PROG: 5608 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5609 default: 5610 return -EINVAL; 5611 } 5612 } 5613 5614 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5615 { 5616 if (speed == 0 || speed == SPEED_UNKNOWN) 5617 speed = slave->speed; 5618 else 5619 speed = min(speed, slave->speed); 5620 5621 return speed; 5622 } 5623 5624 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5625 struct ethtool_link_ksettings *cmd) 5626 { 5627 struct bonding *bond = netdev_priv(bond_dev); 5628 struct list_head *iter; 5629 struct slave *slave; 5630 u32 speed = 0; 5631 5632 cmd->base.duplex = DUPLEX_UNKNOWN; 5633 cmd->base.port = PORT_OTHER; 5634 5635 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5636 * do not need to check mode. Though link speed might not represent 5637 * the true receive or transmit bandwidth (not all modes are symmetric) 5638 * this is an accurate maximum. 5639 */ 5640 bond_for_each_slave(bond, slave, iter) { 5641 if (bond_slave_can_tx(slave)) { 5642 if (slave->speed != SPEED_UNKNOWN) { 5643 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5644 speed = bond_mode_bcast_speed(slave, 5645 speed); 5646 else 5647 speed += slave->speed; 5648 } 5649 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5650 slave->duplex != DUPLEX_UNKNOWN) 5651 cmd->base.duplex = slave->duplex; 5652 } 5653 } 5654 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5655 5656 return 0; 5657 } 5658 5659 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5660 struct ethtool_drvinfo *drvinfo) 5661 { 5662 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5663 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5664 BOND_ABI_VERSION); 5665 } 5666 5667 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5668 struct ethtool_ts_info *info) 5669 { 5670 struct bonding *bond = netdev_priv(bond_dev); 5671 const struct ethtool_ops *ops; 5672 struct net_device *real_dev; 5673 struct phy_device *phydev; 5674 int ret = 0; 5675 5676 rcu_read_lock(); 5677 real_dev = bond_option_active_slave_get_rcu(bond); 5678 dev_hold(real_dev); 5679 rcu_read_unlock(); 5680 5681 if (real_dev) { 5682 ops = real_dev->ethtool_ops; 5683 phydev = real_dev->phydev; 5684 5685 if (phy_has_tsinfo(phydev)) { 5686 ret = phy_ts_info(phydev, info); 5687 goto out; 5688 } else if (ops->get_ts_info) { 5689 ret = ops->get_ts_info(real_dev, info); 5690 goto out; 5691 } 5692 } 5693 5694 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE | 5695 SOF_TIMESTAMPING_SOFTWARE; 5696 info->phc_index = -1; 5697 5698 out: 5699 dev_put(real_dev); 5700 return ret; 5701 } 5702 5703 static const struct ethtool_ops bond_ethtool_ops = { 5704 .get_drvinfo = bond_ethtool_get_drvinfo, 5705 .get_link = ethtool_op_get_link, 5706 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5707 .get_ts_info = bond_ethtool_get_ts_info, 5708 }; 5709 5710 static const struct net_device_ops bond_netdev_ops = { 5711 .ndo_init = bond_init, 5712 .ndo_uninit = bond_uninit, 5713 .ndo_open = bond_open, 5714 .ndo_stop = bond_close, 5715 .ndo_start_xmit = bond_start_xmit, 5716 .ndo_select_queue = bond_select_queue, 5717 .ndo_get_stats64 = bond_get_stats, 5718 .ndo_eth_ioctl = bond_eth_ioctl, 5719 .ndo_siocbond = bond_do_ioctl, 5720 .ndo_siocdevprivate = bond_siocdevprivate, 5721 .ndo_change_rx_flags = bond_change_rx_flags, 5722 .ndo_set_rx_mode = bond_set_rx_mode, 5723 .ndo_change_mtu = bond_change_mtu, 5724 .ndo_set_mac_address = bond_set_mac_address, 5725 .ndo_neigh_setup = bond_neigh_setup, 5726 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5727 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5728 #ifdef CONFIG_NET_POLL_CONTROLLER 5729 .ndo_netpoll_setup = bond_netpoll_setup, 5730 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5731 .ndo_poll_controller = bond_poll_controller, 5732 #endif 5733 .ndo_add_slave = bond_enslave, 5734 .ndo_del_slave = bond_release, 5735 .ndo_fix_features = bond_fix_features, 5736 .ndo_features_check = passthru_features_check, 5737 .ndo_get_xmit_slave = bond_xmit_get_slave, 5738 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5739 .ndo_bpf = bond_xdp, 5740 .ndo_xdp_xmit = bond_xdp_xmit, 5741 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5742 }; 5743 5744 static const struct device_type bond_type = { 5745 .name = "bond", 5746 }; 5747 5748 static void bond_destructor(struct net_device *bond_dev) 5749 { 5750 struct bonding *bond = netdev_priv(bond_dev); 5751 5752 if (bond->wq) 5753 destroy_workqueue(bond->wq); 5754 5755 if (bond->rr_tx_counter) 5756 free_percpu(bond->rr_tx_counter); 5757 } 5758 5759 void bond_setup(struct net_device *bond_dev) 5760 { 5761 struct bonding *bond = netdev_priv(bond_dev); 5762 5763 spin_lock_init(&bond->mode_lock); 5764 bond->params = bonding_defaults; 5765 5766 /* Initialize pointers */ 5767 bond->dev = bond_dev; 5768 5769 /* Initialize the device entry points */ 5770 ether_setup(bond_dev); 5771 bond_dev->max_mtu = ETH_MAX_MTU; 5772 bond_dev->netdev_ops = &bond_netdev_ops; 5773 bond_dev->ethtool_ops = &bond_ethtool_ops; 5774 5775 bond_dev->needs_free_netdev = true; 5776 bond_dev->priv_destructor = bond_destructor; 5777 5778 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5779 5780 /* Initialize the device options */ 5781 bond_dev->flags |= IFF_MASTER; 5782 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5783 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5784 5785 #ifdef CONFIG_XFRM_OFFLOAD 5786 /* set up xfrm device ops (only supported in active-backup right now) */ 5787 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5788 INIT_LIST_HEAD(&bond->ipsec_list); 5789 spin_lock_init(&bond->ipsec_lock); 5790 #endif /* CONFIG_XFRM_OFFLOAD */ 5791 5792 /* don't acquire bond device's netif_tx_lock when transmitting */ 5793 bond_dev->features |= NETIF_F_LLTX; 5794 5795 /* By default, we declare the bond to be fully 5796 * VLAN hardware accelerated capable. Special 5797 * care is taken in the various xmit functions 5798 * when there are slaves that are not hw accel 5799 * capable 5800 */ 5801 5802 /* Don't allow bond devices to change network namespaces. */ 5803 bond_dev->features |= NETIF_F_NETNS_LOCAL; 5804 5805 bond_dev->hw_features = BOND_VLAN_FEATURES | 5806 NETIF_F_HW_VLAN_CTAG_RX | 5807 NETIF_F_HW_VLAN_CTAG_FILTER; 5808 5809 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5810 bond_dev->features |= bond_dev->hw_features; 5811 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5812 #ifdef CONFIG_XFRM_OFFLOAD 5813 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5814 /* Only enable XFRM features if this is an active-backup config */ 5815 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5816 bond_dev->features |= BOND_XFRM_FEATURES; 5817 #endif /* CONFIG_XFRM_OFFLOAD */ 5818 } 5819 5820 /* Destroy a bonding device. 5821 * Must be under rtnl_lock when this function is called. 5822 */ 5823 static void bond_uninit(struct net_device *bond_dev) 5824 { 5825 struct bonding *bond = netdev_priv(bond_dev); 5826 struct bond_up_slave *usable, *all; 5827 struct list_head *iter; 5828 struct slave *slave; 5829 5830 bond_netpoll_cleanup(bond_dev); 5831 5832 /* Release the bonded slaves */ 5833 bond_for_each_slave(bond, slave, iter) 5834 __bond_release_one(bond_dev, slave->dev, true, true); 5835 netdev_info(bond_dev, "Released all slaves\n"); 5836 5837 usable = rtnl_dereference(bond->usable_slaves); 5838 if (usable) { 5839 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5840 kfree_rcu(usable, rcu); 5841 } 5842 5843 all = rtnl_dereference(bond->all_slaves); 5844 if (all) { 5845 RCU_INIT_POINTER(bond->all_slaves, NULL); 5846 kfree_rcu(all, rcu); 5847 } 5848 5849 list_del(&bond->bond_list); 5850 5851 bond_debug_unregister(bond); 5852 } 5853 5854 /*------------------------- Module initialization ---------------------------*/ 5855 5856 static int bond_check_params(struct bond_params *params) 5857 { 5858 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 5859 struct bond_opt_value newval; 5860 const struct bond_opt_value *valptr; 5861 int arp_all_targets_value = 0; 5862 u16 ad_actor_sys_prio = 0; 5863 u16 ad_user_port_key = 0; 5864 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 5865 int arp_ip_count; 5866 int bond_mode = BOND_MODE_ROUNDROBIN; 5867 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 5868 int lacp_fast = 0; 5869 int tlb_dynamic_lb; 5870 5871 /* Convert string parameters. */ 5872 if (mode) { 5873 bond_opt_initstr(&newval, mode); 5874 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5875 if (!valptr) { 5876 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5877 return -EINVAL; 5878 } 5879 bond_mode = valptr->value; 5880 } 5881 5882 if (xmit_hash_policy) { 5883 if (bond_mode == BOND_MODE_ROUNDROBIN || 5884 bond_mode == BOND_MODE_ACTIVEBACKUP || 5885 bond_mode == BOND_MODE_BROADCAST) { 5886 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 5887 bond_mode_name(bond_mode)); 5888 } else { 5889 bond_opt_initstr(&newval, xmit_hash_policy); 5890 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5891 &newval); 5892 if (!valptr) { 5893 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5894 xmit_hash_policy); 5895 return -EINVAL; 5896 } 5897 xmit_hashtype = valptr->value; 5898 } 5899 } 5900 5901 if (lacp_rate) { 5902 if (bond_mode != BOND_MODE_8023AD) { 5903 pr_info("lacp_rate param is irrelevant in mode %s\n", 5904 bond_mode_name(bond_mode)); 5905 } else { 5906 bond_opt_initstr(&newval, lacp_rate); 5907 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5908 &newval); 5909 if (!valptr) { 5910 pr_err("Error: Invalid lacp rate \"%s\"\n", 5911 lacp_rate); 5912 return -EINVAL; 5913 } 5914 lacp_fast = valptr->value; 5915 } 5916 } 5917 5918 if (ad_select) { 5919 bond_opt_initstr(&newval, ad_select); 5920 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5921 &newval); 5922 if (!valptr) { 5923 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5924 return -EINVAL; 5925 } 5926 params->ad_select = valptr->value; 5927 if (bond_mode != BOND_MODE_8023AD) 5928 pr_warn("ad_select param only affects 802.3ad mode\n"); 5929 } else { 5930 params->ad_select = BOND_AD_STABLE; 5931 } 5932 5933 if (max_bonds < 0) { 5934 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5935 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5936 max_bonds = BOND_DEFAULT_MAX_BONDS; 5937 } 5938 5939 if (miimon < 0) { 5940 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5941 miimon, INT_MAX); 5942 miimon = 0; 5943 } 5944 5945 if (updelay < 0) { 5946 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5947 updelay, INT_MAX); 5948 updelay = 0; 5949 } 5950 5951 if (downdelay < 0) { 5952 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5953 downdelay, INT_MAX); 5954 downdelay = 0; 5955 } 5956 5957 if ((use_carrier != 0) && (use_carrier != 1)) { 5958 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5959 use_carrier); 5960 use_carrier = 1; 5961 } 5962 5963 if (num_peer_notif < 0 || num_peer_notif > 255) { 5964 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5965 num_peer_notif); 5966 num_peer_notif = 1; 5967 } 5968 5969 /* reset values for 802.3ad/TLB/ALB */ 5970 if (!bond_mode_uses_arp(bond_mode)) { 5971 if (!miimon) { 5972 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"); 5973 pr_warn("Forcing miimon to 100msec\n"); 5974 miimon = BOND_DEFAULT_MIIMON; 5975 } 5976 } 5977 5978 if (tx_queues < 1 || tx_queues > 255) { 5979 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5980 tx_queues, BOND_DEFAULT_TX_QUEUES); 5981 tx_queues = BOND_DEFAULT_TX_QUEUES; 5982 } 5983 5984 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5985 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5986 all_slaves_active); 5987 all_slaves_active = 0; 5988 } 5989 5990 if (resend_igmp < 0 || resend_igmp > 255) { 5991 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 5992 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 5993 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 5994 } 5995 5996 bond_opt_initval(&newval, packets_per_slave); 5997 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 5998 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 5999 packets_per_slave, USHRT_MAX); 6000 packets_per_slave = 1; 6001 } 6002 6003 if (bond_mode == BOND_MODE_ALB) { 6004 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", 6005 updelay); 6006 } 6007 6008 if (!miimon) { 6009 if (updelay || downdelay) { 6010 /* just warn the user the up/down delay will have 6011 * no effect since miimon is zero... 6012 */ 6013 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", 6014 updelay, downdelay); 6015 } 6016 } else { 6017 /* don't allow arp monitoring */ 6018 if (arp_interval) { 6019 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6020 miimon, arp_interval); 6021 arp_interval = 0; 6022 } 6023 6024 if ((updelay % miimon) != 0) { 6025 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6026 updelay, miimon, (updelay / miimon) * miimon); 6027 } 6028 6029 updelay /= miimon; 6030 6031 if ((downdelay % miimon) != 0) { 6032 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6033 downdelay, miimon, 6034 (downdelay / miimon) * miimon); 6035 } 6036 6037 downdelay /= miimon; 6038 } 6039 6040 if (arp_interval < 0) { 6041 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6042 arp_interval, INT_MAX); 6043 arp_interval = 0; 6044 } 6045 6046 for (arp_ip_count = 0, i = 0; 6047 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6048 __be32 ip; 6049 6050 /* not a complete check, but good enough to catch mistakes */ 6051 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6052 !bond_is_ip_target_ok(ip)) { 6053 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6054 arp_ip_target[i]); 6055 arp_interval = 0; 6056 } else { 6057 if (bond_get_targets_ip(arp_target, ip) == -1) 6058 arp_target[arp_ip_count++] = ip; 6059 else 6060 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6061 &ip); 6062 } 6063 } 6064 6065 if (arp_interval && !arp_ip_count) { 6066 /* don't allow arping if no arp_ip_target given... */ 6067 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6068 arp_interval); 6069 arp_interval = 0; 6070 } 6071 6072 if (arp_validate) { 6073 if (!arp_interval) { 6074 pr_err("arp_validate requires arp_interval\n"); 6075 return -EINVAL; 6076 } 6077 6078 bond_opt_initstr(&newval, arp_validate); 6079 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6080 &newval); 6081 if (!valptr) { 6082 pr_err("Error: invalid arp_validate \"%s\"\n", 6083 arp_validate); 6084 return -EINVAL; 6085 } 6086 arp_validate_value = valptr->value; 6087 } else { 6088 arp_validate_value = 0; 6089 } 6090 6091 if (arp_all_targets) { 6092 bond_opt_initstr(&newval, arp_all_targets); 6093 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6094 &newval); 6095 if (!valptr) { 6096 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6097 arp_all_targets); 6098 arp_all_targets_value = 0; 6099 } else { 6100 arp_all_targets_value = valptr->value; 6101 } 6102 } 6103 6104 if (miimon) { 6105 pr_info("MII link monitoring set to %d ms\n", miimon); 6106 } else if (arp_interval) { 6107 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6108 arp_validate_value); 6109 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6110 arp_interval, valptr->string, arp_ip_count); 6111 6112 for (i = 0; i < arp_ip_count; i++) 6113 pr_cont(" %s", arp_ip_target[i]); 6114 6115 pr_cont("\n"); 6116 6117 } else if (max_bonds) { 6118 /* miimon and arp_interval not set, we need one so things 6119 * work as expected, see bonding.txt for details 6120 */ 6121 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"); 6122 } 6123 6124 if (primary && !bond_mode_uses_primary(bond_mode)) { 6125 /* currently, using a primary only makes sense 6126 * in active backup, TLB or ALB modes 6127 */ 6128 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6129 primary, bond_mode_name(bond_mode)); 6130 primary = NULL; 6131 } 6132 6133 if (primary && primary_reselect) { 6134 bond_opt_initstr(&newval, primary_reselect); 6135 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6136 &newval); 6137 if (!valptr) { 6138 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6139 primary_reselect); 6140 return -EINVAL; 6141 } 6142 primary_reselect_value = valptr->value; 6143 } else { 6144 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6145 } 6146 6147 if (fail_over_mac) { 6148 bond_opt_initstr(&newval, fail_over_mac); 6149 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6150 &newval); 6151 if (!valptr) { 6152 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6153 fail_over_mac); 6154 return -EINVAL; 6155 } 6156 fail_over_mac_value = valptr->value; 6157 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6158 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6159 } else { 6160 fail_over_mac_value = BOND_FOM_NONE; 6161 } 6162 6163 bond_opt_initstr(&newval, "default"); 6164 valptr = bond_opt_parse( 6165 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6166 &newval); 6167 if (!valptr) { 6168 pr_err("Error: No ad_actor_sys_prio default value"); 6169 return -EINVAL; 6170 } 6171 ad_actor_sys_prio = valptr->value; 6172 6173 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6174 &newval); 6175 if (!valptr) { 6176 pr_err("Error: No ad_user_port_key default value"); 6177 return -EINVAL; 6178 } 6179 ad_user_port_key = valptr->value; 6180 6181 bond_opt_initstr(&newval, "default"); 6182 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6183 if (!valptr) { 6184 pr_err("Error: No tlb_dynamic_lb default value"); 6185 return -EINVAL; 6186 } 6187 tlb_dynamic_lb = valptr->value; 6188 6189 if (lp_interval == 0) { 6190 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6191 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6192 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6193 } 6194 6195 /* fill params struct with the proper values */ 6196 params->mode = bond_mode; 6197 params->xmit_policy = xmit_hashtype; 6198 params->miimon = miimon; 6199 params->num_peer_notif = num_peer_notif; 6200 params->arp_interval = arp_interval; 6201 params->arp_validate = arp_validate_value; 6202 params->arp_all_targets = arp_all_targets_value; 6203 params->missed_max = 2; 6204 params->updelay = updelay; 6205 params->downdelay = downdelay; 6206 params->peer_notif_delay = 0; 6207 params->use_carrier = use_carrier; 6208 params->lacp_active = 1; 6209 params->lacp_fast = lacp_fast; 6210 params->primary[0] = 0; 6211 params->primary_reselect = primary_reselect_value; 6212 params->fail_over_mac = fail_over_mac_value; 6213 params->tx_queues = tx_queues; 6214 params->all_slaves_active = all_slaves_active; 6215 params->resend_igmp = resend_igmp; 6216 params->min_links = min_links; 6217 params->lp_interval = lp_interval; 6218 params->packets_per_slave = packets_per_slave; 6219 params->tlb_dynamic_lb = tlb_dynamic_lb; 6220 params->ad_actor_sys_prio = ad_actor_sys_prio; 6221 eth_zero_addr(params->ad_actor_system); 6222 params->ad_user_port_key = ad_user_port_key; 6223 if (packets_per_slave > 0) { 6224 params->reciprocal_packets_per_slave = 6225 reciprocal_value(packets_per_slave); 6226 } else { 6227 /* reciprocal_packets_per_slave is unused if 6228 * packets_per_slave is 0 or 1, just initialize it 6229 */ 6230 params->reciprocal_packets_per_slave = 6231 (struct reciprocal_value) { 0 }; 6232 } 6233 6234 if (primary) 6235 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6236 6237 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6238 #if IS_ENABLED(CONFIG_IPV6) 6239 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6240 #endif 6241 6242 return 0; 6243 } 6244 6245 /* Called from registration process */ 6246 static int bond_init(struct net_device *bond_dev) 6247 { 6248 struct bonding *bond = netdev_priv(bond_dev); 6249 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6250 6251 netdev_dbg(bond_dev, "Begin bond_init\n"); 6252 6253 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 6254 if (!bond->wq) 6255 return -ENOMEM; 6256 6257 spin_lock_init(&bond->stats_lock); 6258 netdev_lockdep_set_classes(bond_dev); 6259 6260 list_add_tail(&bond->bond_list, &bn->dev_list); 6261 6262 bond_prepare_sysfs_group(bond); 6263 6264 bond_debug_register(bond); 6265 6266 /* Ensure valid dev_addr */ 6267 if (is_zero_ether_addr(bond_dev->dev_addr) && 6268 bond_dev->addr_assign_type == NET_ADDR_PERM) 6269 eth_hw_addr_random(bond_dev); 6270 6271 return 0; 6272 } 6273 6274 unsigned int bond_get_num_tx_queues(void) 6275 { 6276 return tx_queues; 6277 } 6278 6279 /* Create a new bond based on the specified name and bonding parameters. 6280 * If name is NULL, obtain a suitable "bond%d" name for us. 6281 * Caller must NOT hold rtnl_lock; we need to release it here before we 6282 * set up our sysfs entries. 6283 */ 6284 int bond_create(struct net *net, const char *name) 6285 { 6286 struct net_device *bond_dev; 6287 struct bonding *bond; 6288 int res = -ENOMEM; 6289 6290 rtnl_lock(); 6291 6292 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6293 name ? name : "bond%d", NET_NAME_UNKNOWN, 6294 bond_setup, tx_queues); 6295 if (!bond_dev) 6296 goto out; 6297 6298 bond = netdev_priv(bond_dev); 6299 dev_net_set(bond_dev, net); 6300 bond_dev->rtnl_link_ops = &bond_link_ops; 6301 6302 res = register_netdevice(bond_dev); 6303 if (res < 0) { 6304 free_netdev(bond_dev); 6305 goto out; 6306 } 6307 6308 netif_carrier_off(bond_dev); 6309 6310 bond_work_init_all(bond); 6311 6312 out: 6313 rtnl_unlock(); 6314 return res; 6315 } 6316 6317 static int __net_init bond_net_init(struct net *net) 6318 { 6319 struct bond_net *bn = net_generic(net, bond_net_id); 6320 6321 bn->net = net; 6322 INIT_LIST_HEAD(&bn->dev_list); 6323 6324 bond_create_proc_dir(bn); 6325 bond_create_sysfs(bn); 6326 6327 return 0; 6328 } 6329 6330 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6331 { 6332 struct bond_net *bn; 6333 struct net *net; 6334 LIST_HEAD(list); 6335 6336 list_for_each_entry(net, net_list, exit_list) { 6337 bn = net_generic(net, bond_net_id); 6338 bond_destroy_sysfs(bn); 6339 } 6340 6341 /* Kill off any bonds created after unregistering bond rtnl ops */ 6342 rtnl_lock(); 6343 list_for_each_entry(net, net_list, exit_list) { 6344 struct bonding *bond, *tmp_bond; 6345 6346 bn = net_generic(net, bond_net_id); 6347 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6348 unregister_netdevice_queue(bond->dev, &list); 6349 } 6350 unregister_netdevice_many(&list); 6351 rtnl_unlock(); 6352 6353 list_for_each_entry(net, net_list, exit_list) { 6354 bn = net_generic(net, bond_net_id); 6355 bond_destroy_proc_dir(bn); 6356 } 6357 } 6358 6359 static struct pernet_operations bond_net_ops = { 6360 .init = bond_net_init, 6361 .exit_batch = bond_net_exit_batch, 6362 .id = &bond_net_id, 6363 .size = sizeof(struct bond_net), 6364 }; 6365 6366 static int __init bonding_init(void) 6367 { 6368 int i; 6369 int res; 6370 6371 res = bond_check_params(&bonding_defaults); 6372 if (res) 6373 goto out; 6374 6375 res = register_pernet_subsys(&bond_net_ops); 6376 if (res) 6377 goto out; 6378 6379 res = bond_netlink_init(); 6380 if (res) 6381 goto err_link; 6382 6383 bond_create_debugfs(); 6384 6385 for (i = 0; i < max_bonds; i++) { 6386 res = bond_create(&init_net, NULL); 6387 if (res) 6388 goto err; 6389 } 6390 6391 skb_flow_dissector_init(&flow_keys_bonding, 6392 flow_keys_bonding_keys, 6393 ARRAY_SIZE(flow_keys_bonding_keys)); 6394 6395 register_netdevice_notifier(&bond_netdev_notifier); 6396 out: 6397 return res; 6398 err: 6399 bond_destroy_debugfs(); 6400 bond_netlink_fini(); 6401 err_link: 6402 unregister_pernet_subsys(&bond_net_ops); 6403 goto out; 6404 6405 } 6406 6407 static void __exit bonding_exit(void) 6408 { 6409 unregister_netdevice_notifier(&bond_netdev_notifier); 6410 6411 bond_destroy_debugfs(); 6412 6413 bond_netlink_fini(); 6414 unregister_pernet_subsys(&bond_net_ops); 6415 6416 #ifdef CONFIG_NET_POLL_CONTROLLER 6417 /* Make sure we don't have an imbalance on our netpoll blocking */ 6418 WARN_ON(atomic_read(&netpoll_block_tx)); 6419 #endif 6420 } 6421 6422 module_init(bonding_init); 6423 module_exit(bonding_exit); 6424 MODULE_LICENSE("GPL"); 6425 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6426 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6427