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 int err; 1636 1637 type = bond_lag_tx_type(bond); 1638 lag_upper_info.tx_type = type; 1639 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1640 1641 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1642 &lag_upper_info, extack); 1643 if (err) 1644 return err; 1645 1646 slave->dev->flags |= IFF_SLAVE; 1647 return 0; 1648 } 1649 1650 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1651 { 1652 netdev_upper_dev_unlink(slave->dev, bond->dev); 1653 slave->dev->flags &= ~IFF_SLAVE; 1654 } 1655 1656 static void slave_kobj_release(struct kobject *kobj) 1657 { 1658 struct slave *slave = to_slave(kobj); 1659 struct bonding *bond = bond_get_bond_by_slave(slave); 1660 1661 cancel_delayed_work_sync(&slave->notify_work); 1662 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1663 kfree(SLAVE_AD_INFO(slave)); 1664 1665 kfree(slave); 1666 } 1667 1668 static struct kobj_type slave_ktype = { 1669 .release = slave_kobj_release, 1670 #ifdef CONFIG_SYSFS 1671 .sysfs_ops = &slave_sysfs_ops, 1672 #endif 1673 }; 1674 1675 static int bond_kobj_init(struct slave *slave) 1676 { 1677 int err; 1678 1679 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1680 &(slave->dev->dev.kobj), "bonding_slave"); 1681 if (err) 1682 kobject_put(&slave->kobj); 1683 1684 return err; 1685 } 1686 1687 static struct slave *bond_alloc_slave(struct bonding *bond, 1688 struct net_device *slave_dev) 1689 { 1690 struct slave *slave = NULL; 1691 1692 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1693 if (!slave) 1694 return NULL; 1695 1696 slave->bond = bond; 1697 slave->dev = slave_dev; 1698 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1699 1700 if (bond_kobj_init(slave)) 1701 return NULL; 1702 1703 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1704 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1705 GFP_KERNEL); 1706 if (!SLAVE_AD_INFO(slave)) { 1707 kobject_put(&slave->kobj); 1708 return NULL; 1709 } 1710 } 1711 1712 return slave; 1713 } 1714 1715 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1716 { 1717 info->bond_mode = BOND_MODE(bond); 1718 info->miimon = bond->params.miimon; 1719 info->num_slaves = bond->slave_cnt; 1720 } 1721 1722 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1723 { 1724 strcpy(info->slave_name, slave->dev->name); 1725 info->link = slave->link; 1726 info->state = bond_slave_state(slave); 1727 info->link_failure_count = slave->link_failure_count; 1728 } 1729 1730 static void bond_netdev_notify_work(struct work_struct *_work) 1731 { 1732 struct slave *slave = container_of(_work, struct slave, 1733 notify_work.work); 1734 1735 if (rtnl_trylock()) { 1736 struct netdev_bonding_info binfo; 1737 1738 bond_fill_ifslave(slave, &binfo.slave); 1739 bond_fill_ifbond(slave->bond, &binfo.master); 1740 netdev_bonding_info_change(slave->dev, &binfo); 1741 rtnl_unlock(); 1742 } else { 1743 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1744 } 1745 } 1746 1747 void bond_queue_slave_event(struct slave *slave) 1748 { 1749 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1750 } 1751 1752 void bond_lower_state_changed(struct slave *slave) 1753 { 1754 struct netdev_lag_lower_state_info info; 1755 1756 info.link_up = slave->link == BOND_LINK_UP || 1757 slave->link == BOND_LINK_FAIL; 1758 info.tx_enabled = bond_is_active_slave(slave); 1759 netdev_lower_state_changed(slave->dev, &info); 1760 } 1761 1762 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1763 if (extack) \ 1764 NL_SET_ERR_MSG(extack, errmsg); \ 1765 else \ 1766 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1767 } while (0) 1768 1769 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1770 if (extack) \ 1771 NL_SET_ERR_MSG(extack, errmsg); \ 1772 else \ 1773 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1774 } while (0) 1775 1776 /* enslave device <slave> to bond device <master> */ 1777 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1778 struct netlink_ext_ack *extack) 1779 { 1780 struct bonding *bond = netdev_priv(bond_dev); 1781 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1782 struct slave *new_slave = NULL, *prev_slave; 1783 struct sockaddr_storage ss; 1784 int link_reporting; 1785 int res = 0, i; 1786 1787 if (slave_dev->flags & IFF_MASTER && 1788 !netif_is_bond_master(slave_dev)) { 1789 BOND_NL_ERR(bond_dev, extack, 1790 "Device type (master device) cannot be enslaved"); 1791 return -EPERM; 1792 } 1793 1794 if (!bond->params.use_carrier && 1795 slave_dev->ethtool_ops->get_link == NULL && 1796 slave_ops->ndo_eth_ioctl == NULL) { 1797 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1798 } 1799 1800 /* already in-use? */ 1801 if (netdev_is_rx_handler_busy(slave_dev)) { 1802 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1803 "Device is in use and cannot be enslaved"); 1804 return -EBUSY; 1805 } 1806 1807 if (bond_dev == slave_dev) { 1808 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1809 return -EPERM; 1810 } 1811 1812 /* vlan challenged mutual exclusion */ 1813 /* no need to lock since we're protected by rtnl_lock */ 1814 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1815 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1816 if (vlan_uses_dev(bond_dev)) { 1817 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1818 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1819 return -EPERM; 1820 } else { 1821 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1822 } 1823 } else { 1824 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1825 } 1826 1827 if (slave_dev->features & NETIF_F_HW_ESP) 1828 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1829 1830 /* Old ifenslave binaries are no longer supported. These can 1831 * be identified with moderate accuracy by the state of the slave: 1832 * the current ifenslave will set the interface down prior to 1833 * enslaving it; the old ifenslave will not. 1834 */ 1835 if (slave_dev->flags & IFF_UP) { 1836 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1837 "Device can not be enslaved while up"); 1838 return -EPERM; 1839 } 1840 1841 /* set bonding device ether type by slave - bonding netdevices are 1842 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1843 * there is a need to override some of the type dependent attribs/funcs. 1844 * 1845 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1846 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1847 */ 1848 if (!bond_has_slaves(bond)) { 1849 if (bond_dev->type != slave_dev->type) { 1850 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1851 bond_dev->type, slave_dev->type); 1852 1853 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1854 bond_dev); 1855 res = notifier_to_errno(res); 1856 if (res) { 1857 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1858 return -EBUSY; 1859 } 1860 1861 /* Flush unicast and multicast addresses */ 1862 dev_uc_flush(bond_dev); 1863 dev_mc_flush(bond_dev); 1864 1865 if (slave_dev->type != ARPHRD_ETHER) 1866 bond_setup_by_slave(bond_dev, slave_dev); 1867 else { 1868 ether_setup(bond_dev); 1869 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1870 } 1871 1872 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1873 bond_dev); 1874 } 1875 } else if (bond_dev->type != slave_dev->type) { 1876 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1877 "Device type is different from other slaves"); 1878 return -EINVAL; 1879 } 1880 1881 if (slave_dev->type == ARPHRD_INFINIBAND && 1882 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1883 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1884 "Only active-backup mode is supported for infiniband slaves"); 1885 res = -EOPNOTSUPP; 1886 goto err_undo_flags; 1887 } 1888 1889 if (!slave_ops->ndo_set_mac_address || 1890 slave_dev->type == ARPHRD_INFINIBAND) { 1891 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1892 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1893 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1894 if (!bond_has_slaves(bond)) { 1895 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1896 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1897 } else { 1898 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1899 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1900 res = -EOPNOTSUPP; 1901 goto err_undo_flags; 1902 } 1903 } 1904 } 1905 1906 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1907 1908 /* If this is the first slave, then we need to set the master's hardware 1909 * address to be the same as the slave's. 1910 */ 1911 if (!bond_has_slaves(bond) && 1912 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1913 res = bond_set_dev_addr(bond->dev, slave_dev); 1914 if (res) 1915 goto err_undo_flags; 1916 } 1917 1918 new_slave = bond_alloc_slave(bond, slave_dev); 1919 if (!new_slave) { 1920 res = -ENOMEM; 1921 goto err_undo_flags; 1922 } 1923 1924 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1925 * is set via sysfs or module option if desired. 1926 */ 1927 new_slave->queue_id = 0; 1928 1929 /* Save slave's original mtu and then set it to match the bond */ 1930 new_slave->original_mtu = slave_dev->mtu; 1931 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1932 if (res) { 1933 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1934 goto err_free; 1935 } 1936 1937 /* Save slave's original ("permanent") mac address for modes 1938 * that need it, and for restoring it upon release, and then 1939 * set it to the master's address 1940 */ 1941 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1942 slave_dev->addr_len); 1943 1944 if (!bond->params.fail_over_mac || 1945 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1946 /* Set slave to master's mac address. The application already 1947 * set the master's mac address to that of the first slave 1948 */ 1949 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 1950 ss.ss_family = slave_dev->type; 1951 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 1952 extack); 1953 if (res) { 1954 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 1955 goto err_restore_mtu; 1956 } 1957 } 1958 1959 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 1960 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 1961 1962 /* open the slave since the application closed it */ 1963 res = dev_open(slave_dev, extack); 1964 if (res) { 1965 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 1966 goto err_restore_mac; 1967 } 1968 1969 slave_dev->priv_flags |= IFF_BONDING; 1970 /* initialize slave stats */ 1971 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 1972 1973 if (bond_is_lb(bond)) { 1974 /* bond_alb_init_slave() must be called before all other stages since 1975 * it might fail and we do not want to have to undo everything 1976 */ 1977 res = bond_alb_init_slave(bond, new_slave); 1978 if (res) 1979 goto err_close; 1980 } 1981 1982 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 1983 if (res) { 1984 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 1985 goto err_close; 1986 } 1987 1988 prev_slave = bond_last_slave(bond); 1989 1990 new_slave->delay = 0; 1991 new_slave->link_failure_count = 0; 1992 1993 if (bond_update_speed_duplex(new_slave) && 1994 bond_needs_speed_duplex(bond)) 1995 new_slave->link = BOND_LINK_DOWN; 1996 1997 new_slave->last_rx = jiffies - 1998 (msecs_to_jiffies(bond->params.arp_interval) + 1); 1999 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2000 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2001 2002 new_slave->last_tx = new_slave->last_rx; 2003 2004 if (bond->params.miimon && !bond->params.use_carrier) { 2005 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2006 2007 if ((link_reporting == -1) && !bond->params.arp_interval) { 2008 /* miimon is set but a bonded network driver 2009 * does not support ETHTOOL/MII and 2010 * arp_interval is not set. Note: if 2011 * use_carrier is enabled, we will never go 2012 * here (because netif_carrier is always 2013 * supported); thus, we don't need to change 2014 * the messages for netif_carrier. 2015 */ 2016 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n"); 2017 } else if (link_reporting == -1) { 2018 /* unable get link status using mii/ethtool */ 2019 slave_warn(bond_dev, slave_dev, "can't get link status from slave; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n"); 2020 } 2021 } 2022 2023 /* check for initial state */ 2024 new_slave->link = BOND_LINK_NOCHANGE; 2025 if (bond->params.miimon) { 2026 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2027 if (bond->params.updelay) { 2028 bond_set_slave_link_state(new_slave, 2029 BOND_LINK_BACK, 2030 BOND_SLAVE_NOTIFY_NOW); 2031 new_slave->delay = bond->params.updelay; 2032 } else { 2033 bond_set_slave_link_state(new_slave, 2034 BOND_LINK_UP, 2035 BOND_SLAVE_NOTIFY_NOW); 2036 } 2037 } else { 2038 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2039 BOND_SLAVE_NOTIFY_NOW); 2040 } 2041 } else if (bond->params.arp_interval) { 2042 bond_set_slave_link_state(new_slave, 2043 (netif_carrier_ok(slave_dev) ? 2044 BOND_LINK_UP : BOND_LINK_DOWN), 2045 BOND_SLAVE_NOTIFY_NOW); 2046 } else { 2047 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2048 BOND_SLAVE_NOTIFY_NOW); 2049 } 2050 2051 if (new_slave->link != BOND_LINK_DOWN) 2052 new_slave->last_link_up = jiffies; 2053 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2054 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2055 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2056 2057 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2058 /* if there is a primary slave, remember it */ 2059 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2060 rcu_assign_pointer(bond->primary_slave, new_slave); 2061 bond->force_primary = true; 2062 } 2063 } 2064 2065 switch (BOND_MODE(bond)) { 2066 case BOND_MODE_ACTIVEBACKUP: 2067 bond_set_slave_inactive_flags(new_slave, 2068 BOND_SLAVE_NOTIFY_NOW); 2069 break; 2070 case BOND_MODE_8023AD: 2071 /* in 802.3ad mode, the internal mechanism 2072 * will activate the slaves in the selected 2073 * aggregator 2074 */ 2075 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2076 /* if this is the first slave */ 2077 if (!prev_slave) { 2078 SLAVE_AD_INFO(new_slave)->id = 1; 2079 /* Initialize AD with the number of times that the AD timer is called in 1 second 2080 * can be called only after the mac address of the bond is set 2081 */ 2082 bond_3ad_initialize(bond); 2083 } else { 2084 SLAVE_AD_INFO(new_slave)->id = 2085 SLAVE_AD_INFO(prev_slave)->id + 1; 2086 } 2087 2088 bond_3ad_bind_slave(new_slave); 2089 break; 2090 case BOND_MODE_TLB: 2091 case BOND_MODE_ALB: 2092 bond_set_active_slave(new_slave); 2093 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2094 break; 2095 default: 2096 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2097 2098 /* always active in trunk mode */ 2099 bond_set_active_slave(new_slave); 2100 2101 /* In trunking mode there is little meaning to curr_active_slave 2102 * anyway (it holds no special properties of the bond device), 2103 * so we can change it without calling change_active_interface() 2104 */ 2105 if (!rcu_access_pointer(bond->curr_active_slave) && 2106 new_slave->link == BOND_LINK_UP) 2107 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2108 2109 break; 2110 } /* switch(bond_mode) */ 2111 2112 #ifdef CONFIG_NET_POLL_CONTROLLER 2113 if (bond->dev->npinfo) { 2114 if (slave_enable_netpoll(new_slave)) { 2115 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2116 res = -EBUSY; 2117 goto err_detach; 2118 } 2119 } 2120 #endif 2121 2122 if (!(bond_dev->features & NETIF_F_LRO)) 2123 dev_disable_lro(slave_dev); 2124 2125 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2126 new_slave); 2127 if (res) { 2128 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2129 goto err_detach; 2130 } 2131 2132 res = bond_master_upper_dev_link(bond, new_slave, extack); 2133 if (res) { 2134 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2135 goto err_unregister; 2136 } 2137 2138 bond_lower_state_changed(new_slave); 2139 2140 res = bond_sysfs_slave_add(new_slave); 2141 if (res) { 2142 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2143 goto err_upper_unlink; 2144 } 2145 2146 /* If the mode uses primary, then the following is handled by 2147 * bond_change_active_slave(). 2148 */ 2149 if (!bond_uses_primary(bond)) { 2150 /* set promiscuity level to new slave */ 2151 if (bond_dev->flags & IFF_PROMISC) { 2152 res = dev_set_promiscuity(slave_dev, 1); 2153 if (res) 2154 goto err_sysfs_del; 2155 } 2156 2157 /* set allmulti level to new slave */ 2158 if (bond_dev->flags & IFF_ALLMULTI) { 2159 res = dev_set_allmulti(slave_dev, 1); 2160 if (res) { 2161 if (bond_dev->flags & IFF_PROMISC) 2162 dev_set_promiscuity(slave_dev, -1); 2163 goto err_sysfs_del; 2164 } 2165 } 2166 2167 if (bond_dev->flags & IFF_UP) { 2168 netif_addr_lock_bh(bond_dev); 2169 dev_mc_sync_multiple(slave_dev, bond_dev); 2170 dev_uc_sync_multiple(slave_dev, bond_dev); 2171 netif_addr_unlock_bh(bond_dev); 2172 2173 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2174 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2175 } 2176 } 2177 2178 bond->slave_cnt++; 2179 bond_compute_features(bond); 2180 bond_set_carrier(bond); 2181 2182 if (bond_uses_primary(bond)) { 2183 block_netpoll_tx(); 2184 bond_select_active_slave(bond); 2185 unblock_netpoll_tx(); 2186 } 2187 2188 if (bond_mode_can_use_xmit_hash(bond)) 2189 bond_update_slave_arr(bond, NULL); 2190 2191 2192 if (!slave_dev->netdev_ops->ndo_bpf || 2193 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2194 if (bond->xdp_prog) { 2195 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2196 "Slave does not support XDP"); 2197 res = -EOPNOTSUPP; 2198 goto err_sysfs_del; 2199 } 2200 } else if (bond->xdp_prog) { 2201 struct netdev_bpf xdp = { 2202 .command = XDP_SETUP_PROG, 2203 .flags = 0, 2204 .prog = bond->xdp_prog, 2205 .extack = extack, 2206 }; 2207 2208 if (dev_xdp_prog_count(slave_dev) > 0) { 2209 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2210 "Slave has XDP program loaded, please unload before enslaving"); 2211 res = -EOPNOTSUPP; 2212 goto err_sysfs_del; 2213 } 2214 2215 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 2216 if (res < 0) { 2217 /* ndo_bpf() sets extack error message */ 2218 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2219 goto err_sysfs_del; 2220 } 2221 if (bond->xdp_prog) 2222 bpf_prog_inc(bond->xdp_prog); 2223 } 2224 2225 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2226 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2227 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2228 2229 /* enslave is successful */ 2230 bond_queue_slave_event(new_slave); 2231 return 0; 2232 2233 /* Undo stages on error */ 2234 err_sysfs_del: 2235 bond_sysfs_slave_del(new_slave); 2236 2237 err_upper_unlink: 2238 bond_upper_dev_unlink(bond, new_slave); 2239 2240 err_unregister: 2241 netdev_rx_handler_unregister(slave_dev); 2242 2243 err_detach: 2244 vlan_vids_del_by_dev(slave_dev, bond_dev); 2245 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2246 RCU_INIT_POINTER(bond->primary_slave, NULL); 2247 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2248 block_netpoll_tx(); 2249 bond_change_active_slave(bond, NULL); 2250 bond_select_active_slave(bond); 2251 unblock_netpoll_tx(); 2252 } 2253 /* either primary_slave or curr_active_slave might've changed */ 2254 synchronize_rcu(); 2255 slave_disable_netpoll(new_slave); 2256 2257 err_close: 2258 if (!netif_is_bond_master(slave_dev)) 2259 slave_dev->priv_flags &= ~IFF_BONDING; 2260 dev_close(slave_dev); 2261 2262 err_restore_mac: 2263 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2264 if (!bond->params.fail_over_mac || 2265 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2266 /* XXX TODO - fom follow mode needs to change master's 2267 * MAC if this slave's MAC is in use by the bond, or at 2268 * least print a warning. 2269 */ 2270 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2271 new_slave->dev->addr_len); 2272 ss.ss_family = slave_dev->type; 2273 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2274 } 2275 2276 err_restore_mtu: 2277 dev_set_mtu(slave_dev, new_slave->original_mtu); 2278 2279 err_free: 2280 kobject_put(&new_slave->kobj); 2281 2282 err_undo_flags: 2283 /* Enslave of first slave has failed and we need to fix master's mac */ 2284 if (!bond_has_slaves(bond)) { 2285 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2286 slave_dev->dev_addr)) 2287 eth_hw_addr_random(bond_dev); 2288 if (bond_dev->type != ARPHRD_ETHER) { 2289 dev_close(bond_dev); 2290 ether_setup(bond_dev); 2291 bond_dev->flags |= IFF_MASTER; 2292 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 2293 } 2294 } 2295 2296 return res; 2297 } 2298 2299 /* Try to release the slave device <slave> from the bond device <master> 2300 * It is legal to access curr_active_slave without a lock because all the function 2301 * is RTNL-locked. If "all" is true it means that the function is being called 2302 * while destroying a bond interface and all slaves are being released. 2303 * 2304 * The rules for slave state should be: 2305 * for Active/Backup: 2306 * Active stays on all backups go down 2307 * for Bonded connections: 2308 * The first up interface should be left on and all others downed. 2309 */ 2310 static int __bond_release_one(struct net_device *bond_dev, 2311 struct net_device *slave_dev, 2312 bool all, bool unregister) 2313 { 2314 struct bonding *bond = netdev_priv(bond_dev); 2315 struct slave *slave, *oldcurrent; 2316 struct sockaddr_storage ss; 2317 int old_flags = bond_dev->flags; 2318 netdev_features_t old_features = bond_dev->features; 2319 2320 /* slave is not a slave or master is not master of this slave */ 2321 if (!(slave_dev->flags & IFF_SLAVE) || 2322 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2323 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2324 return -EINVAL; 2325 } 2326 2327 block_netpoll_tx(); 2328 2329 slave = bond_get_slave_by_dev(bond, slave_dev); 2330 if (!slave) { 2331 /* not a slave of this bond */ 2332 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2333 unblock_netpoll_tx(); 2334 return -EINVAL; 2335 } 2336 2337 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2338 2339 bond_sysfs_slave_del(slave); 2340 2341 /* recompute stats just before removing the slave */ 2342 bond_get_stats(bond->dev, &bond->bond_stats); 2343 2344 if (bond->xdp_prog) { 2345 struct netdev_bpf xdp = { 2346 .command = XDP_SETUP_PROG, 2347 .flags = 0, 2348 .prog = NULL, 2349 .extack = NULL, 2350 }; 2351 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp)) 2352 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2353 } 2354 2355 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2356 * for this slave anymore. 2357 */ 2358 netdev_rx_handler_unregister(slave_dev); 2359 2360 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2361 bond_3ad_unbind_slave(slave); 2362 2363 bond_upper_dev_unlink(bond, slave); 2364 2365 if (bond_mode_can_use_xmit_hash(bond)) 2366 bond_update_slave_arr(bond, slave); 2367 2368 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2369 bond_is_active_slave(slave) ? "active" : "backup"); 2370 2371 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2372 2373 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2374 2375 if (!all && (!bond->params.fail_over_mac || 2376 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2377 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2378 bond_has_slaves(bond)) 2379 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", 2380 slave->perm_hwaddr); 2381 } 2382 2383 if (rtnl_dereference(bond->primary_slave) == slave) 2384 RCU_INIT_POINTER(bond->primary_slave, NULL); 2385 2386 if (oldcurrent == slave) 2387 bond_change_active_slave(bond, NULL); 2388 2389 if (bond_is_lb(bond)) { 2390 /* Must be called only after the slave has been 2391 * detached from the list and the curr_active_slave 2392 * has been cleared (if our_slave == old_current), 2393 * but before a new active slave is selected. 2394 */ 2395 bond_alb_deinit_slave(bond, slave); 2396 } 2397 2398 if (all) { 2399 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2400 } else if (oldcurrent == slave) { 2401 /* Note that we hold RTNL over this sequence, so there 2402 * is no concern that another slave add/remove event 2403 * will interfere. 2404 */ 2405 bond_select_active_slave(bond); 2406 } 2407 2408 bond_set_carrier(bond); 2409 if (!bond_has_slaves(bond)) 2410 eth_hw_addr_random(bond_dev); 2411 2412 unblock_netpoll_tx(); 2413 synchronize_rcu(); 2414 bond->slave_cnt--; 2415 2416 if (!bond_has_slaves(bond)) { 2417 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2418 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2419 } 2420 2421 bond_compute_features(bond); 2422 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2423 (old_features & NETIF_F_VLAN_CHALLENGED)) 2424 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2425 2426 vlan_vids_del_by_dev(slave_dev, bond_dev); 2427 2428 /* If the mode uses primary, then this case was handled above by 2429 * bond_change_active_slave(..., NULL) 2430 */ 2431 if (!bond_uses_primary(bond)) { 2432 /* unset promiscuity level from slave 2433 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2434 * of the IFF_PROMISC flag in the bond_dev, but we need the 2435 * value of that flag before that change, as that was the value 2436 * when this slave was attached, so we cache at the start of the 2437 * function and use it here. Same goes for ALLMULTI below 2438 */ 2439 if (old_flags & IFF_PROMISC) 2440 dev_set_promiscuity(slave_dev, -1); 2441 2442 /* unset allmulti level from slave */ 2443 if (old_flags & IFF_ALLMULTI) 2444 dev_set_allmulti(slave_dev, -1); 2445 2446 if (old_flags & IFF_UP) 2447 bond_hw_addr_flush(bond_dev, slave_dev); 2448 } 2449 2450 slave_disable_netpoll(slave); 2451 2452 /* close slave before restoring its mac address */ 2453 dev_close(slave_dev); 2454 2455 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2456 2457 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2458 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2459 /* restore original ("permanent") mac address */ 2460 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2461 slave->dev->addr_len); 2462 ss.ss_family = slave_dev->type; 2463 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2464 } 2465 2466 if (unregister) 2467 __dev_set_mtu(slave_dev, slave->original_mtu); 2468 else 2469 dev_set_mtu(slave_dev, slave->original_mtu); 2470 2471 if (!netif_is_bond_master(slave_dev)) 2472 slave_dev->priv_flags &= ~IFF_BONDING; 2473 2474 kobject_put(&slave->kobj); 2475 2476 return 0; 2477 } 2478 2479 /* A wrapper used because of ndo_del_link */ 2480 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2481 { 2482 return __bond_release_one(bond_dev, slave_dev, false, false); 2483 } 2484 2485 /* First release a slave and then destroy the bond if no more slaves are left. 2486 * Must be under rtnl_lock when this function is called. 2487 */ 2488 static int bond_release_and_destroy(struct net_device *bond_dev, 2489 struct net_device *slave_dev) 2490 { 2491 struct bonding *bond = netdev_priv(bond_dev); 2492 int ret; 2493 2494 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2495 if (ret == 0 && !bond_has_slaves(bond) && 2496 bond_dev->reg_state != NETREG_UNREGISTERING) { 2497 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2498 netdev_info(bond_dev, "Destroying bond\n"); 2499 bond_remove_proc_entry(bond); 2500 unregister_netdevice(bond_dev); 2501 } 2502 return ret; 2503 } 2504 2505 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2506 { 2507 struct bonding *bond = netdev_priv(bond_dev); 2508 2509 bond_fill_ifbond(bond, info); 2510 } 2511 2512 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2513 { 2514 struct bonding *bond = netdev_priv(bond_dev); 2515 struct list_head *iter; 2516 int i = 0, res = -ENODEV; 2517 struct slave *slave; 2518 2519 bond_for_each_slave(bond, slave, iter) { 2520 if (i++ == (int)info->slave_id) { 2521 res = 0; 2522 bond_fill_ifslave(slave, info); 2523 break; 2524 } 2525 } 2526 2527 return res; 2528 } 2529 2530 /*-------------------------------- Monitoring -------------------------------*/ 2531 2532 /* called with rcu_read_lock() */ 2533 static int bond_miimon_inspect(struct bonding *bond) 2534 { 2535 bool ignore_updelay = false; 2536 int link_state, commit = 0; 2537 struct list_head *iter; 2538 struct slave *slave; 2539 2540 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2541 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2542 } else { 2543 struct bond_up_slave *usable_slaves; 2544 2545 usable_slaves = rcu_dereference(bond->usable_slaves); 2546 2547 if (usable_slaves && usable_slaves->count == 0) 2548 ignore_updelay = true; 2549 } 2550 2551 bond_for_each_slave_rcu(bond, slave, iter) { 2552 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2553 2554 link_state = bond_check_dev_link(bond, slave->dev, 0); 2555 2556 switch (slave->link) { 2557 case BOND_LINK_UP: 2558 if (link_state) 2559 continue; 2560 2561 bond_propose_link_state(slave, BOND_LINK_FAIL); 2562 commit++; 2563 slave->delay = bond->params.downdelay; 2564 if (slave->delay) { 2565 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2566 (BOND_MODE(bond) == 2567 BOND_MODE_ACTIVEBACKUP) ? 2568 (bond_is_active_slave(slave) ? 2569 "active " : "backup ") : "", 2570 bond->params.downdelay * bond->params.miimon); 2571 } 2572 fallthrough; 2573 case BOND_LINK_FAIL: 2574 if (link_state) { 2575 /* recovered before downdelay expired */ 2576 bond_propose_link_state(slave, BOND_LINK_UP); 2577 slave->last_link_up = jiffies; 2578 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2579 (bond->params.downdelay - slave->delay) * 2580 bond->params.miimon); 2581 commit++; 2582 continue; 2583 } 2584 2585 if (slave->delay <= 0) { 2586 bond_propose_link_state(slave, BOND_LINK_DOWN); 2587 commit++; 2588 continue; 2589 } 2590 2591 slave->delay--; 2592 break; 2593 2594 case BOND_LINK_DOWN: 2595 if (!link_state) 2596 continue; 2597 2598 bond_propose_link_state(slave, BOND_LINK_BACK); 2599 commit++; 2600 slave->delay = bond->params.updelay; 2601 2602 if (slave->delay) { 2603 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2604 ignore_updelay ? 0 : 2605 bond->params.updelay * 2606 bond->params.miimon); 2607 } 2608 fallthrough; 2609 case BOND_LINK_BACK: 2610 if (!link_state) { 2611 bond_propose_link_state(slave, BOND_LINK_DOWN); 2612 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2613 (bond->params.updelay - slave->delay) * 2614 bond->params.miimon); 2615 commit++; 2616 continue; 2617 } 2618 2619 if (ignore_updelay) 2620 slave->delay = 0; 2621 2622 if (slave->delay <= 0) { 2623 bond_propose_link_state(slave, BOND_LINK_UP); 2624 commit++; 2625 ignore_updelay = false; 2626 continue; 2627 } 2628 2629 slave->delay--; 2630 break; 2631 } 2632 } 2633 2634 return commit; 2635 } 2636 2637 static void bond_miimon_link_change(struct bonding *bond, 2638 struct slave *slave, 2639 char link) 2640 { 2641 switch (BOND_MODE(bond)) { 2642 case BOND_MODE_8023AD: 2643 bond_3ad_handle_link_change(slave, link); 2644 break; 2645 case BOND_MODE_TLB: 2646 case BOND_MODE_ALB: 2647 bond_alb_handle_link_change(bond, slave, link); 2648 break; 2649 case BOND_MODE_XOR: 2650 bond_update_slave_arr(bond, NULL); 2651 break; 2652 } 2653 } 2654 2655 static void bond_miimon_commit(struct bonding *bond) 2656 { 2657 struct list_head *iter; 2658 struct slave *slave, *primary; 2659 2660 bond_for_each_slave(bond, slave, iter) { 2661 switch (slave->link_new_state) { 2662 case BOND_LINK_NOCHANGE: 2663 /* For 802.3ad mode, check current slave speed and 2664 * duplex again in case its port was disabled after 2665 * invalid speed/duplex reporting but recovered before 2666 * link monitoring could make a decision on the actual 2667 * link status 2668 */ 2669 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2670 slave->link == BOND_LINK_UP) 2671 bond_3ad_adapter_speed_duplex_changed(slave); 2672 continue; 2673 2674 case BOND_LINK_UP: 2675 if (bond_update_speed_duplex(slave) && 2676 bond_needs_speed_duplex(bond)) { 2677 slave->link = BOND_LINK_DOWN; 2678 if (net_ratelimit()) 2679 slave_warn(bond->dev, slave->dev, 2680 "failed to get link speed/duplex\n"); 2681 continue; 2682 } 2683 bond_set_slave_link_state(slave, BOND_LINK_UP, 2684 BOND_SLAVE_NOTIFY_NOW); 2685 slave->last_link_up = jiffies; 2686 2687 primary = rtnl_dereference(bond->primary_slave); 2688 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2689 /* prevent it from being the active one */ 2690 bond_set_backup_slave(slave); 2691 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2692 /* make it immediately active */ 2693 bond_set_active_slave(slave); 2694 } 2695 2696 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2697 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2698 slave->duplex ? "full" : "half"); 2699 2700 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2701 2702 if (!bond->curr_active_slave || slave == primary) 2703 goto do_failover; 2704 2705 continue; 2706 2707 case BOND_LINK_DOWN: 2708 if (slave->link_failure_count < UINT_MAX) 2709 slave->link_failure_count++; 2710 2711 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2712 BOND_SLAVE_NOTIFY_NOW); 2713 2714 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2715 BOND_MODE(bond) == BOND_MODE_8023AD) 2716 bond_set_slave_inactive_flags(slave, 2717 BOND_SLAVE_NOTIFY_NOW); 2718 2719 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2720 2721 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2722 2723 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2724 goto do_failover; 2725 2726 continue; 2727 2728 default: 2729 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2730 slave->link_new_state); 2731 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2732 2733 continue; 2734 } 2735 2736 do_failover: 2737 block_netpoll_tx(); 2738 bond_select_active_slave(bond); 2739 unblock_netpoll_tx(); 2740 } 2741 2742 bond_set_carrier(bond); 2743 } 2744 2745 /* bond_mii_monitor 2746 * 2747 * Really a wrapper that splits the mii monitor into two phases: an 2748 * inspection, then (if inspection indicates something needs to be done) 2749 * an acquisition of appropriate locks followed by a commit phase to 2750 * implement whatever link state changes are indicated. 2751 */ 2752 static void bond_mii_monitor(struct work_struct *work) 2753 { 2754 struct bonding *bond = container_of(work, struct bonding, 2755 mii_work.work); 2756 bool should_notify_peers = false; 2757 bool commit; 2758 unsigned long delay; 2759 struct slave *slave; 2760 struct list_head *iter; 2761 2762 delay = msecs_to_jiffies(bond->params.miimon); 2763 2764 if (!bond_has_slaves(bond)) 2765 goto re_arm; 2766 2767 rcu_read_lock(); 2768 should_notify_peers = bond_should_notify_peers(bond); 2769 commit = !!bond_miimon_inspect(bond); 2770 if (bond->send_peer_notif) { 2771 rcu_read_unlock(); 2772 if (rtnl_trylock()) { 2773 bond->send_peer_notif--; 2774 rtnl_unlock(); 2775 } 2776 } else { 2777 rcu_read_unlock(); 2778 } 2779 2780 if (commit) { 2781 /* Race avoidance with bond_close cancel of workqueue */ 2782 if (!rtnl_trylock()) { 2783 delay = 1; 2784 should_notify_peers = false; 2785 goto re_arm; 2786 } 2787 2788 bond_for_each_slave(bond, slave, iter) { 2789 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2790 } 2791 bond_miimon_commit(bond); 2792 2793 rtnl_unlock(); /* might sleep, hold no other locks */ 2794 } 2795 2796 re_arm: 2797 if (bond->params.miimon) 2798 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2799 2800 if (should_notify_peers) { 2801 if (!rtnl_trylock()) 2802 return; 2803 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2804 rtnl_unlock(); 2805 } 2806 } 2807 2808 static int bond_upper_dev_walk(struct net_device *upper, 2809 struct netdev_nested_priv *priv) 2810 { 2811 __be32 ip = *(__be32 *)priv->data; 2812 2813 return ip == bond_confirm_addr(upper, 0, ip); 2814 } 2815 2816 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2817 { 2818 struct netdev_nested_priv priv = { 2819 .data = (void *)&ip, 2820 }; 2821 bool ret = false; 2822 2823 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2824 return true; 2825 2826 rcu_read_lock(); 2827 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2828 ret = true; 2829 rcu_read_unlock(); 2830 2831 return ret; 2832 } 2833 2834 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2835 struct sk_buff *skb) 2836 { 2837 struct net_device *bond_dev = slave->bond->dev; 2838 struct net_device *slave_dev = slave->dev; 2839 struct bond_vlan_tag *outer_tag = tags; 2840 2841 if (!tags || tags->vlan_proto == VLAN_N_VID) 2842 return true; 2843 2844 tags++; 2845 2846 /* Go through all the tags backwards and add them to the packet */ 2847 while (tags->vlan_proto != VLAN_N_VID) { 2848 if (!tags->vlan_id) { 2849 tags++; 2850 continue; 2851 } 2852 2853 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2854 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2855 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2856 tags->vlan_id); 2857 if (!skb) { 2858 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2859 return false; 2860 } 2861 2862 tags++; 2863 } 2864 /* Set the outer tag */ 2865 if (outer_tag->vlan_id) { 2866 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2867 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2868 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2869 outer_tag->vlan_id); 2870 } 2871 2872 return true; 2873 } 2874 2875 /* We go to the (large) trouble of VLAN tagging ARP frames because 2876 * switches in VLAN mode (especially if ports are configured as 2877 * "native" to a VLAN) might not pass non-tagged frames. 2878 */ 2879 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2880 __be32 src_ip, struct bond_vlan_tag *tags) 2881 { 2882 struct net_device *bond_dev = slave->bond->dev; 2883 struct net_device *slave_dev = slave->dev; 2884 struct sk_buff *skb; 2885 2886 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2887 arp_op, &dest_ip, &src_ip); 2888 2889 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2890 NULL, slave_dev->dev_addr, NULL); 2891 2892 if (!skb) { 2893 net_err_ratelimited("ARP packet allocation failed\n"); 2894 return; 2895 } 2896 2897 if (bond_handle_vlan(slave, tags, skb)) { 2898 slave_update_last_tx(slave); 2899 arp_xmit(skb); 2900 } 2901 2902 return; 2903 } 2904 2905 /* Validate the device path between the @start_dev and the @end_dev. 2906 * The path is valid if the @end_dev is reachable through device 2907 * stacking. 2908 * When the path is validated, collect any vlan information in the 2909 * path. 2910 */ 2911 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2912 struct net_device *end_dev, 2913 int level) 2914 { 2915 struct bond_vlan_tag *tags; 2916 struct net_device *upper; 2917 struct list_head *iter; 2918 2919 if (start_dev == end_dev) { 2920 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2921 if (!tags) 2922 return ERR_PTR(-ENOMEM); 2923 tags[level].vlan_proto = VLAN_N_VID; 2924 return tags; 2925 } 2926 2927 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2928 tags = bond_verify_device_path(upper, end_dev, level + 1); 2929 if (IS_ERR_OR_NULL(tags)) { 2930 if (IS_ERR(tags)) 2931 return tags; 2932 continue; 2933 } 2934 if (is_vlan_dev(upper)) { 2935 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2936 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2937 } 2938 2939 return tags; 2940 } 2941 2942 return NULL; 2943 } 2944 2945 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 2946 { 2947 struct rtable *rt; 2948 struct bond_vlan_tag *tags; 2949 __be32 *targets = bond->params.arp_targets, addr; 2950 int i; 2951 2952 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 2953 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 2954 __func__, &targets[i]); 2955 tags = NULL; 2956 2957 /* Find out through which dev should the packet go */ 2958 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 2959 RTO_ONLINK, 0); 2960 if (IS_ERR(rt)) { 2961 /* there's no route to target - try to send arp 2962 * probe to generate any traffic (arp_validate=0) 2963 */ 2964 if (bond->params.arp_validate) 2965 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 2966 bond->dev->name, 2967 &targets[i]); 2968 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 2969 0, tags); 2970 continue; 2971 } 2972 2973 /* bond device itself */ 2974 if (rt->dst.dev == bond->dev) 2975 goto found; 2976 2977 rcu_read_lock(); 2978 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 2979 rcu_read_unlock(); 2980 2981 if (!IS_ERR_OR_NULL(tags)) 2982 goto found; 2983 2984 /* Not our device - skip */ 2985 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 2986 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 2987 2988 ip_rt_put(rt); 2989 continue; 2990 2991 found: 2992 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 2993 ip_rt_put(rt); 2994 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 2995 kfree(tags); 2996 } 2997 } 2998 2999 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3000 { 3001 int i; 3002 3003 if (!sip || !bond_has_this_ip(bond, tip)) { 3004 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3005 __func__, &sip, &tip); 3006 return; 3007 } 3008 3009 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3010 if (i == -1) { 3011 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3012 __func__, &sip); 3013 return; 3014 } 3015 slave->last_rx = jiffies; 3016 slave->target_last_arp_rx[i] = jiffies; 3017 } 3018 3019 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3020 struct slave *slave) 3021 { 3022 struct arphdr *arp = (struct arphdr *)skb->data; 3023 struct slave *curr_active_slave, *curr_arp_slave; 3024 unsigned char *arp_ptr; 3025 __be32 sip, tip; 3026 unsigned int alen; 3027 3028 alen = arp_hdr_len(bond->dev); 3029 3030 if (alen > skb_headlen(skb)) { 3031 arp = kmalloc(alen, GFP_ATOMIC); 3032 if (!arp) 3033 goto out_unlock; 3034 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3035 goto out_unlock; 3036 } 3037 3038 if (arp->ar_hln != bond->dev->addr_len || 3039 skb->pkt_type == PACKET_OTHERHOST || 3040 skb->pkt_type == PACKET_LOOPBACK || 3041 arp->ar_hrd != htons(ARPHRD_ETHER) || 3042 arp->ar_pro != htons(ETH_P_IP) || 3043 arp->ar_pln != 4) 3044 goto out_unlock; 3045 3046 arp_ptr = (unsigned char *)(arp + 1); 3047 arp_ptr += bond->dev->addr_len; 3048 memcpy(&sip, arp_ptr, 4); 3049 arp_ptr += 4 + bond->dev->addr_len; 3050 memcpy(&tip, arp_ptr, 4); 3051 3052 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3053 __func__, slave->dev->name, bond_slave_state(slave), 3054 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3055 &sip, &tip); 3056 3057 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3058 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3059 3060 /* We 'trust' the received ARP enough to validate it if: 3061 * 3062 * (a) the slave receiving the ARP is active (which includes the 3063 * current ARP slave, if any), or 3064 * 3065 * (b) the receiving slave isn't active, but there is a currently 3066 * active slave and it received valid arp reply(s) after it became 3067 * the currently active slave, or 3068 * 3069 * (c) there is an ARP slave that sent an ARP during the prior ARP 3070 * interval, and we receive an ARP reply on any slave. We accept 3071 * these because switch FDB update delays may deliver the ARP 3072 * reply to a slave other than the sender of the ARP request. 3073 * 3074 * Note: for (b), backup slaves are receiving the broadcast ARP 3075 * request, not a reply. This request passes from the sending 3076 * slave through the L2 switch(es) to the receiving slave. Since 3077 * this is checking the request, sip/tip are swapped for 3078 * validation. 3079 * 3080 * This is done to avoid endless looping when we can't reach the 3081 * arp_ip_target and fool ourselves with our own arp requests. 3082 */ 3083 if (bond_is_active_slave(slave)) 3084 bond_validate_arp(bond, slave, sip, tip); 3085 else if (curr_active_slave && 3086 time_after(slave_last_rx(bond, curr_active_slave), 3087 curr_active_slave->last_link_up)) 3088 bond_validate_arp(bond, slave, tip, sip); 3089 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3090 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3091 bond_validate_arp(bond, slave, sip, tip); 3092 3093 out_unlock: 3094 if (arp != (struct arphdr *)skb->data) 3095 kfree(arp); 3096 return RX_HANDLER_ANOTHER; 3097 } 3098 3099 #if IS_ENABLED(CONFIG_IPV6) 3100 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3101 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3102 { 3103 struct net_device *bond_dev = slave->bond->dev; 3104 struct net_device *slave_dev = slave->dev; 3105 struct in6_addr mcaddr; 3106 struct sk_buff *skb; 3107 3108 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3109 daddr, saddr); 3110 3111 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3112 if (!skb) { 3113 net_err_ratelimited("NS packet allocation failed\n"); 3114 return; 3115 } 3116 3117 addrconf_addr_solict_mult(daddr, &mcaddr); 3118 if (bond_handle_vlan(slave, tags, skb)) { 3119 slave_update_last_tx(slave); 3120 ndisc_send_skb(skb, &mcaddr, saddr); 3121 } 3122 } 3123 3124 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3125 { 3126 struct in6_addr *targets = bond->params.ns_targets; 3127 struct bond_vlan_tag *tags; 3128 struct dst_entry *dst; 3129 struct in6_addr saddr; 3130 struct flowi6 fl6; 3131 int i; 3132 3133 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3134 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3135 __func__, &targets[i]); 3136 tags = NULL; 3137 3138 /* Find out through which dev should the packet go */ 3139 memset(&fl6, 0, sizeof(struct flowi6)); 3140 fl6.daddr = targets[i]; 3141 fl6.flowi6_oif = bond->dev->ifindex; 3142 3143 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3144 if (dst->error) { 3145 dst_release(dst); 3146 /* there's no route to target - try to send arp 3147 * probe to generate any traffic (arp_validate=0) 3148 */ 3149 if (bond->params.arp_validate) 3150 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3151 bond->dev->name, 3152 &targets[i]); 3153 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3154 continue; 3155 } 3156 3157 /* bond device itself */ 3158 if (dst->dev == bond->dev) 3159 goto found; 3160 3161 rcu_read_lock(); 3162 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3163 rcu_read_unlock(); 3164 3165 if (!IS_ERR_OR_NULL(tags)) 3166 goto found; 3167 3168 /* Not our device - skip */ 3169 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3170 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3171 3172 dst_release(dst); 3173 continue; 3174 3175 found: 3176 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3177 bond_ns_send(slave, &targets[i], &saddr, tags); 3178 else 3179 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3180 3181 dst_release(dst); 3182 kfree(tags); 3183 } 3184 } 3185 3186 static int bond_confirm_addr6(struct net_device *dev, 3187 struct netdev_nested_priv *priv) 3188 { 3189 struct in6_addr *addr = (struct in6_addr *)priv->data; 3190 3191 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3192 } 3193 3194 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3195 { 3196 struct netdev_nested_priv priv = { 3197 .data = addr, 3198 }; 3199 int ret = false; 3200 3201 if (bond_confirm_addr6(bond->dev, &priv)) 3202 return true; 3203 3204 rcu_read_lock(); 3205 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3206 ret = true; 3207 rcu_read_unlock(); 3208 3209 return ret; 3210 } 3211 3212 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3213 struct in6_addr *saddr, struct in6_addr *daddr) 3214 { 3215 int i; 3216 3217 /* Ignore NAs that: 3218 * 1. Source address is unspecified address. 3219 * 2. Dest address is neither all-nodes multicast address nor 3220 * exist on bond interface. 3221 */ 3222 if (ipv6_addr_any(saddr) || 3223 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3224 !bond_has_this_ip6(bond, daddr))) { 3225 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3226 __func__, saddr, daddr); 3227 return; 3228 } 3229 3230 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3231 if (i == -1) { 3232 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3233 __func__, saddr); 3234 return; 3235 } 3236 slave->last_rx = jiffies; 3237 slave->target_last_arp_rx[i] = jiffies; 3238 } 3239 3240 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3241 struct slave *slave) 3242 { 3243 struct slave *curr_active_slave, *curr_arp_slave; 3244 struct in6_addr *saddr, *daddr; 3245 struct { 3246 struct ipv6hdr ip6; 3247 struct icmp6hdr icmp6; 3248 } *combined, _combined; 3249 3250 if (skb->pkt_type == PACKET_OTHERHOST || 3251 skb->pkt_type == PACKET_LOOPBACK) 3252 goto out; 3253 3254 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3255 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3256 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT) 3257 goto out; 3258 3259 saddr = &combined->ip6.saddr; 3260 daddr = &combined->ip6.daddr; 3261 3262 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3263 __func__, slave->dev->name, bond_slave_state(slave), 3264 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3265 saddr, daddr); 3266 3267 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3268 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3269 3270 /* We 'trust' the received ARP enough to validate it if: 3271 * see bond_arp_rcv(). 3272 */ 3273 if (bond_is_active_slave(slave)) 3274 bond_validate_na(bond, slave, saddr, daddr); 3275 else if (curr_active_slave && 3276 time_after(slave_last_rx(bond, curr_active_slave), 3277 curr_active_slave->last_link_up)) 3278 bond_validate_na(bond, slave, saddr, daddr); 3279 else if (curr_arp_slave && 3280 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3281 bond_validate_na(bond, slave, saddr, daddr); 3282 3283 out: 3284 return RX_HANDLER_ANOTHER; 3285 } 3286 #endif 3287 3288 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3289 struct slave *slave) 3290 { 3291 #if IS_ENABLED(CONFIG_IPV6) 3292 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3293 #endif 3294 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3295 3296 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3297 __func__, skb->dev->name); 3298 3299 /* Use arp validate logic for both ARP and NS */ 3300 if (!slave_do_arp_validate(bond, slave)) { 3301 if ((slave_do_arp_validate_only(bond) && is_arp) || 3302 #if IS_ENABLED(CONFIG_IPV6) 3303 (slave_do_arp_validate_only(bond) && is_ipv6) || 3304 #endif 3305 !slave_do_arp_validate_only(bond)) 3306 slave->last_rx = jiffies; 3307 return RX_HANDLER_ANOTHER; 3308 } else if (is_arp) { 3309 return bond_arp_rcv(skb, bond, slave); 3310 #if IS_ENABLED(CONFIG_IPV6) 3311 } else if (is_ipv6) { 3312 return bond_na_rcv(skb, bond, slave); 3313 #endif 3314 } else { 3315 return RX_HANDLER_ANOTHER; 3316 } 3317 } 3318 3319 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3320 { 3321 bond_arp_send_all(bond, slave); 3322 #if IS_ENABLED(CONFIG_IPV6) 3323 bond_ns_send_all(bond, slave); 3324 #endif 3325 } 3326 3327 /* function to verify if we're in the arp_interval timeslice, returns true if 3328 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3329 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3330 */ 3331 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3332 int mod) 3333 { 3334 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3335 3336 return time_in_range(jiffies, 3337 last_act - delta_in_ticks, 3338 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3339 } 3340 3341 /* This function is called regularly to monitor each slave's link 3342 * ensuring that traffic is being sent and received when arp monitoring 3343 * is used in load-balancing mode. if the adapter has been dormant, then an 3344 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3345 * arp monitoring in active backup mode. 3346 */ 3347 static void bond_loadbalance_arp_mon(struct bonding *bond) 3348 { 3349 struct slave *slave, *oldcurrent; 3350 struct list_head *iter; 3351 int do_failover = 0, slave_state_changed = 0; 3352 3353 if (!bond_has_slaves(bond)) 3354 goto re_arm; 3355 3356 rcu_read_lock(); 3357 3358 oldcurrent = rcu_dereference(bond->curr_active_slave); 3359 /* see if any of the previous devices are up now (i.e. they have 3360 * xmt and rcv traffic). the curr_active_slave does not come into 3361 * the picture unless it is null. also, slave->last_link_up is not 3362 * needed here because we send an arp on each slave and give a slave 3363 * as long as it needs to get the tx/rx within the delta. 3364 * TODO: what about up/down delay in arp mode? it wasn't here before 3365 * so it can wait 3366 */ 3367 bond_for_each_slave_rcu(bond, slave, iter) { 3368 unsigned long last_tx = slave_last_tx(slave); 3369 3370 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3371 3372 if (slave->link != BOND_LINK_UP) { 3373 if (bond_time_in_interval(bond, last_tx, 1) && 3374 bond_time_in_interval(bond, slave->last_rx, 1)) { 3375 3376 bond_propose_link_state(slave, BOND_LINK_UP); 3377 slave_state_changed = 1; 3378 3379 /* primary_slave has no meaning in round-robin 3380 * mode. the window of a slave being up and 3381 * curr_active_slave being null after enslaving 3382 * is closed. 3383 */ 3384 if (!oldcurrent) { 3385 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3386 do_failover = 1; 3387 } else { 3388 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3389 } 3390 } 3391 } else { 3392 /* slave->link == BOND_LINK_UP */ 3393 3394 /* not all switches will respond to an arp request 3395 * when the source ip is 0, so don't take the link down 3396 * if we don't know our ip yet 3397 */ 3398 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3399 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3400 3401 bond_propose_link_state(slave, BOND_LINK_DOWN); 3402 slave_state_changed = 1; 3403 3404 if (slave->link_failure_count < UINT_MAX) 3405 slave->link_failure_count++; 3406 3407 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3408 3409 if (slave == oldcurrent) 3410 do_failover = 1; 3411 } 3412 } 3413 3414 /* note: if switch is in round-robin mode, all links 3415 * must tx arp to ensure all links rx an arp - otherwise 3416 * links may oscillate or not come up at all; if switch is 3417 * in something like xor mode, there is nothing we can 3418 * do - all replies will be rx'ed on same link causing slaves 3419 * to be unstable during low/no traffic periods 3420 */ 3421 if (bond_slave_is_up(slave)) 3422 bond_send_validate(bond, slave); 3423 } 3424 3425 rcu_read_unlock(); 3426 3427 if (do_failover || slave_state_changed) { 3428 if (!rtnl_trylock()) 3429 goto re_arm; 3430 3431 bond_for_each_slave(bond, slave, iter) { 3432 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3433 slave->link = slave->link_new_state; 3434 } 3435 3436 if (slave_state_changed) { 3437 bond_slave_state_change(bond); 3438 if (BOND_MODE(bond) == BOND_MODE_XOR) 3439 bond_update_slave_arr(bond, NULL); 3440 } 3441 if (do_failover) { 3442 block_netpoll_tx(); 3443 bond_select_active_slave(bond); 3444 unblock_netpoll_tx(); 3445 } 3446 rtnl_unlock(); 3447 } 3448 3449 re_arm: 3450 if (bond->params.arp_interval) 3451 queue_delayed_work(bond->wq, &bond->arp_work, 3452 msecs_to_jiffies(bond->params.arp_interval)); 3453 } 3454 3455 /* Called to inspect slaves for active-backup mode ARP monitor link state 3456 * changes. Sets proposed link state in slaves to specify what action 3457 * should take place for the slave. Returns 0 if no changes are found, >0 3458 * if changes to link states must be committed. 3459 * 3460 * Called with rcu_read_lock held. 3461 */ 3462 static int bond_ab_arp_inspect(struct bonding *bond) 3463 { 3464 unsigned long last_tx, last_rx; 3465 struct list_head *iter; 3466 struct slave *slave; 3467 int commit = 0; 3468 3469 bond_for_each_slave_rcu(bond, slave, iter) { 3470 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3471 last_rx = slave_last_rx(bond, slave); 3472 3473 if (slave->link != BOND_LINK_UP) { 3474 if (bond_time_in_interval(bond, last_rx, 1)) { 3475 bond_propose_link_state(slave, BOND_LINK_UP); 3476 commit++; 3477 } else if (slave->link == BOND_LINK_BACK) { 3478 bond_propose_link_state(slave, BOND_LINK_FAIL); 3479 commit++; 3480 } 3481 continue; 3482 } 3483 3484 /* Give slaves 2*delta after being enslaved or made 3485 * active. This avoids bouncing, as the last receive 3486 * times need a full ARP monitor cycle to be updated. 3487 */ 3488 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3489 continue; 3490 3491 /* Backup slave is down if: 3492 * - No current_arp_slave AND 3493 * - more than (missed_max+1)*delta since last receive AND 3494 * - the bond has an IP address 3495 * 3496 * Note: a non-null current_arp_slave indicates 3497 * the curr_active_slave went down and we are 3498 * searching for a new one; under this condition 3499 * we only take the curr_active_slave down - this 3500 * gives each slave a chance to tx/rx traffic 3501 * before being taken out 3502 */ 3503 if (!bond_is_active_slave(slave) && 3504 !rcu_access_pointer(bond->current_arp_slave) && 3505 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3506 bond_propose_link_state(slave, BOND_LINK_DOWN); 3507 commit++; 3508 } 3509 3510 /* Active slave is down if: 3511 * - more than missed_max*delta since transmitting OR 3512 * - (more than missed_max*delta since receive AND 3513 * the bond has an IP address) 3514 */ 3515 last_tx = slave_last_tx(slave); 3516 if (bond_is_active_slave(slave) && 3517 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3518 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3519 bond_propose_link_state(slave, BOND_LINK_DOWN); 3520 commit++; 3521 } 3522 } 3523 3524 return commit; 3525 } 3526 3527 /* Called to commit link state changes noted by inspection step of 3528 * active-backup mode ARP monitor. 3529 * 3530 * Called with RTNL hold. 3531 */ 3532 static void bond_ab_arp_commit(struct bonding *bond) 3533 { 3534 struct list_head *iter; 3535 unsigned long last_tx; 3536 struct slave *slave; 3537 3538 bond_for_each_slave(bond, slave, iter) { 3539 switch (slave->link_new_state) { 3540 case BOND_LINK_NOCHANGE: 3541 continue; 3542 3543 case BOND_LINK_UP: 3544 last_tx = slave_last_tx(slave); 3545 if (rtnl_dereference(bond->curr_active_slave) != slave || 3546 (!rtnl_dereference(bond->curr_active_slave) && 3547 bond_time_in_interval(bond, last_tx, 1))) { 3548 struct slave *current_arp_slave; 3549 3550 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3551 bond_set_slave_link_state(slave, BOND_LINK_UP, 3552 BOND_SLAVE_NOTIFY_NOW); 3553 if (current_arp_slave) { 3554 bond_set_slave_inactive_flags( 3555 current_arp_slave, 3556 BOND_SLAVE_NOTIFY_NOW); 3557 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3558 } 3559 3560 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3561 3562 if (!rtnl_dereference(bond->curr_active_slave) || 3563 slave == rtnl_dereference(bond->primary_slave)) 3564 goto do_failover; 3565 3566 } 3567 3568 continue; 3569 3570 case BOND_LINK_DOWN: 3571 if (slave->link_failure_count < UINT_MAX) 3572 slave->link_failure_count++; 3573 3574 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3575 BOND_SLAVE_NOTIFY_NOW); 3576 bond_set_slave_inactive_flags(slave, 3577 BOND_SLAVE_NOTIFY_NOW); 3578 3579 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3580 3581 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3582 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3583 goto do_failover; 3584 } 3585 3586 continue; 3587 3588 case BOND_LINK_FAIL: 3589 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3590 BOND_SLAVE_NOTIFY_NOW); 3591 bond_set_slave_inactive_flags(slave, 3592 BOND_SLAVE_NOTIFY_NOW); 3593 3594 /* A slave has just been enslaved and has become 3595 * the current active slave. 3596 */ 3597 if (rtnl_dereference(bond->curr_active_slave)) 3598 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3599 continue; 3600 3601 default: 3602 slave_err(bond->dev, slave->dev, 3603 "impossible: link_new_state %d on slave\n", 3604 slave->link_new_state); 3605 continue; 3606 } 3607 3608 do_failover: 3609 block_netpoll_tx(); 3610 bond_select_active_slave(bond); 3611 unblock_netpoll_tx(); 3612 } 3613 3614 bond_set_carrier(bond); 3615 } 3616 3617 /* Send ARP probes for active-backup mode ARP monitor. 3618 * 3619 * Called with rcu_read_lock held. 3620 */ 3621 static bool bond_ab_arp_probe(struct bonding *bond) 3622 { 3623 struct slave *slave, *before = NULL, *new_slave = NULL, 3624 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3625 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3626 struct list_head *iter; 3627 bool found = false; 3628 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3629 3630 if (curr_arp_slave && curr_active_slave) 3631 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3632 curr_arp_slave->dev->name, 3633 curr_active_slave->dev->name); 3634 3635 if (curr_active_slave) { 3636 bond_send_validate(bond, curr_active_slave); 3637 return should_notify_rtnl; 3638 } 3639 3640 /* if we don't have a curr_active_slave, search for the next available 3641 * backup slave from the current_arp_slave and make it the candidate 3642 * for becoming the curr_active_slave 3643 */ 3644 3645 if (!curr_arp_slave) { 3646 curr_arp_slave = bond_first_slave_rcu(bond); 3647 if (!curr_arp_slave) 3648 return should_notify_rtnl; 3649 } 3650 3651 bond_for_each_slave_rcu(bond, slave, iter) { 3652 if (!found && !before && bond_slave_is_up(slave)) 3653 before = slave; 3654 3655 if (found && !new_slave && bond_slave_is_up(slave)) 3656 new_slave = slave; 3657 /* if the link state is up at this point, we 3658 * mark it down - this can happen if we have 3659 * simultaneous link failures and 3660 * reselect_active_interface doesn't make this 3661 * one the current slave so it is still marked 3662 * up when it is actually down 3663 */ 3664 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3665 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3666 BOND_SLAVE_NOTIFY_LATER); 3667 if (slave->link_failure_count < UINT_MAX) 3668 slave->link_failure_count++; 3669 3670 bond_set_slave_inactive_flags(slave, 3671 BOND_SLAVE_NOTIFY_LATER); 3672 3673 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3674 } 3675 if (slave == curr_arp_slave) 3676 found = true; 3677 } 3678 3679 if (!new_slave && before) 3680 new_slave = before; 3681 3682 if (!new_slave) 3683 goto check_state; 3684 3685 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3686 BOND_SLAVE_NOTIFY_LATER); 3687 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3688 bond_send_validate(bond, new_slave); 3689 new_slave->last_link_up = jiffies; 3690 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3691 3692 check_state: 3693 bond_for_each_slave_rcu(bond, slave, iter) { 3694 if (slave->should_notify || slave->should_notify_link) { 3695 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3696 break; 3697 } 3698 } 3699 return should_notify_rtnl; 3700 } 3701 3702 static void bond_activebackup_arp_mon(struct bonding *bond) 3703 { 3704 bool should_notify_peers = false; 3705 bool should_notify_rtnl = false; 3706 int delta_in_ticks; 3707 3708 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3709 3710 if (!bond_has_slaves(bond)) 3711 goto re_arm; 3712 3713 rcu_read_lock(); 3714 3715 should_notify_peers = bond_should_notify_peers(bond); 3716 3717 if (bond_ab_arp_inspect(bond)) { 3718 rcu_read_unlock(); 3719 3720 /* Race avoidance with bond_close flush of workqueue */ 3721 if (!rtnl_trylock()) { 3722 delta_in_ticks = 1; 3723 should_notify_peers = false; 3724 goto re_arm; 3725 } 3726 3727 bond_ab_arp_commit(bond); 3728 3729 rtnl_unlock(); 3730 rcu_read_lock(); 3731 } 3732 3733 should_notify_rtnl = bond_ab_arp_probe(bond); 3734 rcu_read_unlock(); 3735 3736 re_arm: 3737 if (bond->params.arp_interval) 3738 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3739 3740 if (should_notify_peers || should_notify_rtnl) { 3741 if (!rtnl_trylock()) 3742 return; 3743 3744 if (should_notify_peers) { 3745 bond->send_peer_notif--; 3746 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3747 bond->dev); 3748 } 3749 if (should_notify_rtnl) { 3750 bond_slave_state_notify(bond); 3751 bond_slave_link_notify(bond); 3752 } 3753 3754 rtnl_unlock(); 3755 } 3756 } 3757 3758 static void bond_arp_monitor(struct work_struct *work) 3759 { 3760 struct bonding *bond = container_of(work, struct bonding, 3761 arp_work.work); 3762 3763 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3764 bond_activebackup_arp_mon(bond); 3765 else 3766 bond_loadbalance_arp_mon(bond); 3767 } 3768 3769 /*-------------------------- netdev event handling --------------------------*/ 3770 3771 /* Change device name */ 3772 static int bond_event_changename(struct bonding *bond) 3773 { 3774 bond_remove_proc_entry(bond); 3775 bond_create_proc_entry(bond); 3776 3777 bond_debug_reregister(bond); 3778 3779 return NOTIFY_DONE; 3780 } 3781 3782 static int bond_master_netdev_event(unsigned long event, 3783 struct net_device *bond_dev) 3784 { 3785 struct bonding *event_bond = netdev_priv(bond_dev); 3786 3787 netdev_dbg(bond_dev, "%s called\n", __func__); 3788 3789 switch (event) { 3790 case NETDEV_CHANGENAME: 3791 return bond_event_changename(event_bond); 3792 case NETDEV_UNREGISTER: 3793 bond_remove_proc_entry(event_bond); 3794 #ifdef CONFIG_XFRM_OFFLOAD 3795 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3796 #endif /* CONFIG_XFRM_OFFLOAD */ 3797 break; 3798 case NETDEV_REGISTER: 3799 bond_create_proc_entry(event_bond); 3800 break; 3801 default: 3802 break; 3803 } 3804 3805 return NOTIFY_DONE; 3806 } 3807 3808 static int bond_slave_netdev_event(unsigned long event, 3809 struct net_device *slave_dev) 3810 { 3811 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3812 struct bonding *bond; 3813 struct net_device *bond_dev; 3814 3815 /* A netdev event can be generated while enslaving a device 3816 * before netdev_rx_handler_register is called in which case 3817 * slave will be NULL 3818 */ 3819 if (!slave) { 3820 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3821 return NOTIFY_DONE; 3822 } 3823 3824 bond_dev = slave->bond->dev; 3825 bond = slave->bond; 3826 primary = rtnl_dereference(bond->primary_slave); 3827 3828 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3829 3830 switch (event) { 3831 case NETDEV_UNREGISTER: 3832 if (bond_dev->type != ARPHRD_ETHER) 3833 bond_release_and_destroy(bond_dev, slave_dev); 3834 else 3835 __bond_release_one(bond_dev, slave_dev, false, true); 3836 break; 3837 case NETDEV_UP: 3838 case NETDEV_CHANGE: 3839 /* For 802.3ad mode only: 3840 * Getting invalid Speed/Duplex values here will put slave 3841 * in weird state. Mark it as link-fail if the link was 3842 * previously up or link-down if it hasn't yet come up, and 3843 * let link-monitoring (miimon) set it right when correct 3844 * speeds/duplex are available. 3845 */ 3846 if (bond_update_speed_duplex(slave) && 3847 BOND_MODE(bond) == BOND_MODE_8023AD) { 3848 if (slave->last_link_up) 3849 slave->link = BOND_LINK_FAIL; 3850 else 3851 slave->link = BOND_LINK_DOWN; 3852 } 3853 3854 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3855 bond_3ad_adapter_speed_duplex_changed(slave); 3856 fallthrough; 3857 case NETDEV_DOWN: 3858 /* Refresh slave-array if applicable! 3859 * If the setup does not use miimon or arpmon (mode-specific!), 3860 * then these events will not cause the slave-array to be 3861 * refreshed. This will cause xmit to use a slave that is not 3862 * usable. Avoid such situation by refeshing the array at these 3863 * events. If these (miimon/arpmon) parameters are configured 3864 * then array gets refreshed twice and that should be fine! 3865 */ 3866 if (bond_mode_can_use_xmit_hash(bond)) 3867 bond_update_slave_arr(bond, NULL); 3868 break; 3869 case NETDEV_CHANGEMTU: 3870 /* TODO: Should slaves be allowed to 3871 * independently alter their MTU? For 3872 * an active-backup bond, slaves need 3873 * not be the same type of device, so 3874 * MTUs may vary. For other modes, 3875 * slaves arguably should have the 3876 * same MTUs. To do this, we'd need to 3877 * take over the slave's change_mtu 3878 * function for the duration of their 3879 * servitude. 3880 */ 3881 break; 3882 case NETDEV_CHANGENAME: 3883 /* we don't care if we don't have primary set */ 3884 if (!bond_uses_primary(bond) || 3885 !bond->params.primary[0]) 3886 break; 3887 3888 if (slave == primary) { 3889 /* slave's name changed - he's no longer primary */ 3890 RCU_INIT_POINTER(bond->primary_slave, NULL); 3891 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3892 /* we have a new primary slave */ 3893 rcu_assign_pointer(bond->primary_slave, slave); 3894 } else { /* we didn't change primary - exit */ 3895 break; 3896 } 3897 3898 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3899 primary ? slave_dev->name : "none"); 3900 3901 block_netpoll_tx(); 3902 bond_select_active_slave(bond); 3903 unblock_netpoll_tx(); 3904 break; 3905 case NETDEV_FEAT_CHANGE: 3906 bond_compute_features(bond); 3907 break; 3908 case NETDEV_RESEND_IGMP: 3909 /* Propagate to master device */ 3910 call_netdevice_notifiers(event, slave->bond->dev); 3911 break; 3912 default: 3913 break; 3914 } 3915 3916 return NOTIFY_DONE; 3917 } 3918 3919 /* bond_netdev_event: handle netdev notifier chain events. 3920 * 3921 * This function receives events for the netdev chain. The caller (an 3922 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3923 * locks for us to safely manipulate the slave devices (RTNL lock, 3924 * dev_probe_lock). 3925 */ 3926 static int bond_netdev_event(struct notifier_block *this, 3927 unsigned long event, void *ptr) 3928 { 3929 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 3930 3931 netdev_dbg(event_dev, "%s received %s\n", 3932 __func__, netdev_cmd_to_name(event)); 3933 3934 if (!(event_dev->priv_flags & IFF_BONDING)) 3935 return NOTIFY_DONE; 3936 3937 if (event_dev->flags & IFF_MASTER) { 3938 int ret; 3939 3940 ret = bond_master_netdev_event(event, event_dev); 3941 if (ret != NOTIFY_DONE) 3942 return ret; 3943 } 3944 3945 if (event_dev->flags & IFF_SLAVE) 3946 return bond_slave_netdev_event(event, event_dev); 3947 3948 return NOTIFY_DONE; 3949 } 3950 3951 static struct notifier_block bond_netdev_notifier = { 3952 .notifier_call = bond_netdev_event, 3953 }; 3954 3955 /*---------------------------- Hashing Policies -----------------------------*/ 3956 3957 /* Helper to access data in a packet, with or without a backing skb. 3958 * If skb is given the data is linearized if necessary via pskb_may_pull. 3959 */ 3960 static inline const void *bond_pull_data(struct sk_buff *skb, 3961 const void *data, int hlen, int n) 3962 { 3963 if (likely(n <= hlen)) 3964 return data; 3965 else if (skb && likely(pskb_may_pull(skb, n))) 3966 return skb->head; 3967 3968 return NULL; 3969 } 3970 3971 /* L2 hash helper */ 3972 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 3973 { 3974 struct ethhdr *ep; 3975 3976 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 3977 if (!data) 3978 return 0; 3979 3980 ep = (struct ethhdr *)(data + mhoff); 3981 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 3982 } 3983 3984 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 3985 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 3986 { 3987 const struct ipv6hdr *iph6; 3988 const struct iphdr *iph; 3989 3990 if (l2_proto == htons(ETH_P_IP)) { 3991 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 3992 if (!data) 3993 return false; 3994 3995 iph = (const struct iphdr *)(data + *nhoff); 3996 iph_to_flow_copy_v4addrs(fk, iph); 3997 *nhoff += iph->ihl << 2; 3998 if (!ip_is_fragment(iph)) 3999 *ip_proto = iph->protocol; 4000 } else if (l2_proto == htons(ETH_P_IPV6)) { 4001 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4002 if (!data) 4003 return false; 4004 4005 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4006 iph_to_flow_copy_v6addrs(fk, iph6); 4007 *nhoff += sizeof(*iph6); 4008 *ip_proto = iph6->nexthdr; 4009 } else { 4010 return false; 4011 } 4012 4013 if (l34 && *ip_proto >= 0) 4014 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4015 4016 return true; 4017 } 4018 4019 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4020 { 4021 u32 srcmac_vendor = 0, srcmac_dev = 0; 4022 struct ethhdr *mac_hdr; 4023 u16 vlan = 0; 4024 int i; 4025 4026 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4027 if (!data) 4028 return 0; 4029 mac_hdr = (struct ethhdr *)(data + mhoff); 4030 4031 for (i = 0; i < 3; i++) 4032 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4033 4034 for (i = 3; i < ETH_ALEN; i++) 4035 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4036 4037 if (skb && skb_vlan_tag_present(skb)) 4038 vlan = skb_vlan_tag_get(skb); 4039 4040 return vlan ^ srcmac_vendor ^ srcmac_dev; 4041 } 4042 4043 /* Extract the appropriate headers based on bond's xmit policy */ 4044 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4045 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4046 { 4047 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4048 int ip_proto = -1; 4049 4050 switch (bond->params.xmit_policy) { 4051 case BOND_XMIT_POLICY_ENCAP23: 4052 case BOND_XMIT_POLICY_ENCAP34: 4053 memset(fk, 0, sizeof(*fk)); 4054 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4055 fk, data, l2_proto, nhoff, hlen, 0); 4056 default: 4057 break; 4058 } 4059 4060 fk->ports.ports = 0; 4061 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4062 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4063 return false; 4064 4065 /* ICMP error packets contains at least 8 bytes of the header 4066 * of the packet which generated the error. Use this information 4067 * to correlate ICMP error packets within the same flow which 4068 * generated the error. 4069 */ 4070 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4071 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4072 if (ip_proto == IPPROTO_ICMP) { 4073 if (!icmp_is_err(fk->icmp.type)) 4074 return true; 4075 4076 nhoff += sizeof(struct icmphdr); 4077 } else if (ip_proto == IPPROTO_ICMPV6) { 4078 if (!icmpv6_is_err(fk->icmp.type)) 4079 return true; 4080 4081 nhoff += sizeof(struct icmp6hdr); 4082 } 4083 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4084 } 4085 4086 return true; 4087 } 4088 4089 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4090 { 4091 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4092 (__force u32)flow_get_u32_src(flow); 4093 hash ^= (hash >> 16); 4094 hash ^= (hash >> 8); 4095 4096 /* discard lowest hash bit to deal with the common even ports pattern */ 4097 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4098 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4099 return hash >> 1; 4100 4101 return hash; 4102 } 4103 4104 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4105 * the data as required, but this function can be used without it if the data is 4106 * known to be linear (e.g. with xdp_buff). 4107 */ 4108 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4109 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4110 { 4111 struct flow_keys flow; 4112 u32 hash; 4113 4114 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4115 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4116 4117 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4118 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4119 return bond_eth_hash(skb, data, mhoff, hlen); 4120 4121 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4122 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4123 hash = bond_eth_hash(skb, data, mhoff, hlen); 4124 } else { 4125 if (flow.icmp.id) 4126 memcpy(&hash, &flow.icmp, sizeof(hash)); 4127 else 4128 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4129 } 4130 4131 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4132 } 4133 4134 /** 4135 * bond_xmit_hash - generate a hash value based on the xmit policy 4136 * @bond: bonding device 4137 * @skb: buffer to use for headers 4138 * 4139 * This function will extract the necessary headers from the skb buffer and use 4140 * them to generate a hash based on the xmit_policy set in the bonding device 4141 */ 4142 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4143 { 4144 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4145 skb->l4_hash) 4146 return skb->hash; 4147 4148 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4149 skb_mac_offset(skb), skb_network_offset(skb), 4150 skb_headlen(skb)); 4151 } 4152 4153 /** 4154 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4155 * @bond: bonding device 4156 * @xdp: buffer to use for headers 4157 * 4158 * The XDP variant of bond_xmit_hash. 4159 */ 4160 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4161 { 4162 struct ethhdr *eth; 4163 4164 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4165 return 0; 4166 4167 eth = (struct ethhdr *)xdp->data; 4168 4169 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4170 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4171 } 4172 4173 /*-------------------------- Device entry points ----------------------------*/ 4174 4175 void bond_work_init_all(struct bonding *bond) 4176 { 4177 INIT_DELAYED_WORK(&bond->mcast_work, 4178 bond_resend_igmp_join_requests_delayed); 4179 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4180 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4181 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4182 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4183 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4184 } 4185 4186 static void bond_work_cancel_all(struct bonding *bond) 4187 { 4188 cancel_delayed_work_sync(&bond->mii_work); 4189 cancel_delayed_work_sync(&bond->arp_work); 4190 cancel_delayed_work_sync(&bond->alb_work); 4191 cancel_delayed_work_sync(&bond->ad_work); 4192 cancel_delayed_work_sync(&bond->mcast_work); 4193 cancel_delayed_work_sync(&bond->slave_arr_work); 4194 } 4195 4196 static int bond_open(struct net_device *bond_dev) 4197 { 4198 struct bonding *bond = netdev_priv(bond_dev); 4199 struct list_head *iter; 4200 struct slave *slave; 4201 4202 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4203 bond->rr_tx_counter = alloc_percpu(u32); 4204 if (!bond->rr_tx_counter) 4205 return -ENOMEM; 4206 } 4207 4208 /* reset slave->backup and slave->inactive */ 4209 if (bond_has_slaves(bond)) { 4210 bond_for_each_slave(bond, slave, iter) { 4211 if (bond_uses_primary(bond) && 4212 slave != rcu_access_pointer(bond->curr_active_slave)) { 4213 bond_set_slave_inactive_flags(slave, 4214 BOND_SLAVE_NOTIFY_NOW); 4215 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4216 bond_set_slave_active_flags(slave, 4217 BOND_SLAVE_NOTIFY_NOW); 4218 } 4219 } 4220 } 4221 4222 if (bond_is_lb(bond)) { 4223 /* bond_alb_initialize must be called before the timer 4224 * is started. 4225 */ 4226 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4227 return -ENOMEM; 4228 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4229 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4230 } 4231 4232 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4233 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4234 4235 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4236 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4237 bond->recv_probe = bond_rcv_validate; 4238 } 4239 4240 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4241 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4242 /* register to receive LACPDUs */ 4243 bond->recv_probe = bond_3ad_lacpdu_recv; 4244 bond_3ad_initiate_agg_selection(bond, 1); 4245 4246 bond_for_each_slave(bond, slave, iter) 4247 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4248 } 4249 4250 if (bond_mode_can_use_xmit_hash(bond)) 4251 bond_update_slave_arr(bond, NULL); 4252 4253 return 0; 4254 } 4255 4256 static int bond_close(struct net_device *bond_dev) 4257 { 4258 struct bonding *bond = netdev_priv(bond_dev); 4259 struct slave *slave; 4260 4261 bond_work_cancel_all(bond); 4262 bond->send_peer_notif = 0; 4263 if (bond_is_lb(bond)) 4264 bond_alb_deinitialize(bond); 4265 bond->recv_probe = NULL; 4266 4267 if (bond_uses_primary(bond)) { 4268 rcu_read_lock(); 4269 slave = rcu_dereference(bond->curr_active_slave); 4270 if (slave) 4271 bond_hw_addr_flush(bond_dev, slave->dev); 4272 rcu_read_unlock(); 4273 } else { 4274 struct list_head *iter; 4275 4276 bond_for_each_slave(bond, slave, iter) 4277 bond_hw_addr_flush(bond_dev, slave->dev); 4278 } 4279 4280 return 0; 4281 } 4282 4283 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4284 * that some drivers can provide 32bit values only. 4285 */ 4286 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4287 const struct rtnl_link_stats64 *_new, 4288 const struct rtnl_link_stats64 *_old) 4289 { 4290 const u64 *new = (const u64 *)_new; 4291 const u64 *old = (const u64 *)_old; 4292 u64 *res = (u64 *)_res; 4293 int i; 4294 4295 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4296 u64 nv = new[i]; 4297 u64 ov = old[i]; 4298 s64 delta = nv - ov; 4299 4300 /* detects if this particular field is 32bit only */ 4301 if (((nv | ov) >> 32) == 0) 4302 delta = (s64)(s32)((u32)nv - (u32)ov); 4303 4304 /* filter anomalies, some drivers reset their stats 4305 * at down/up events. 4306 */ 4307 if (delta > 0) 4308 res[i] += delta; 4309 } 4310 } 4311 4312 #ifdef CONFIG_LOCKDEP 4313 static int bond_get_lowest_level_rcu(struct net_device *dev) 4314 { 4315 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4316 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4317 int cur = 0, max = 0; 4318 4319 now = dev; 4320 iter = &dev->adj_list.lower; 4321 4322 while (1) { 4323 next = NULL; 4324 while (1) { 4325 ldev = netdev_next_lower_dev_rcu(now, &iter); 4326 if (!ldev) 4327 break; 4328 4329 next = ldev; 4330 niter = &ldev->adj_list.lower; 4331 dev_stack[cur] = now; 4332 iter_stack[cur++] = iter; 4333 if (max <= cur) 4334 max = cur; 4335 break; 4336 } 4337 4338 if (!next) { 4339 if (!cur) 4340 return max; 4341 next = dev_stack[--cur]; 4342 niter = iter_stack[cur]; 4343 } 4344 4345 now = next; 4346 iter = niter; 4347 } 4348 4349 return max; 4350 } 4351 #endif 4352 4353 static void bond_get_stats(struct net_device *bond_dev, 4354 struct rtnl_link_stats64 *stats) 4355 { 4356 struct bonding *bond = netdev_priv(bond_dev); 4357 struct rtnl_link_stats64 temp; 4358 struct list_head *iter; 4359 struct slave *slave; 4360 int nest_level = 0; 4361 4362 4363 rcu_read_lock(); 4364 #ifdef CONFIG_LOCKDEP 4365 nest_level = bond_get_lowest_level_rcu(bond_dev); 4366 #endif 4367 4368 spin_lock_nested(&bond->stats_lock, nest_level); 4369 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4370 4371 bond_for_each_slave_rcu(bond, slave, iter) { 4372 const struct rtnl_link_stats64 *new = 4373 dev_get_stats(slave->dev, &temp); 4374 4375 bond_fold_stats(stats, new, &slave->slave_stats); 4376 4377 /* save off the slave stats for the next run */ 4378 memcpy(&slave->slave_stats, new, sizeof(*new)); 4379 } 4380 4381 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4382 spin_unlock(&bond->stats_lock); 4383 rcu_read_unlock(); 4384 } 4385 4386 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4387 { 4388 struct bonding *bond = netdev_priv(bond_dev); 4389 struct mii_ioctl_data *mii = NULL; 4390 const struct net_device_ops *ops; 4391 struct net_device *real_dev; 4392 struct hwtstamp_config cfg; 4393 struct ifreq ifrr; 4394 int res = 0; 4395 4396 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4397 4398 switch (cmd) { 4399 case SIOCGMIIPHY: 4400 mii = if_mii(ifr); 4401 if (!mii) 4402 return -EINVAL; 4403 4404 mii->phy_id = 0; 4405 fallthrough; 4406 case SIOCGMIIREG: 4407 /* We do this again just in case we were called by SIOCGMIIREG 4408 * instead of SIOCGMIIPHY. 4409 */ 4410 mii = if_mii(ifr); 4411 if (!mii) 4412 return -EINVAL; 4413 4414 if (mii->reg_num == 1) { 4415 mii->val_out = 0; 4416 if (netif_carrier_ok(bond->dev)) 4417 mii->val_out = BMSR_LSTATUS; 4418 } 4419 4420 break; 4421 case SIOCSHWTSTAMP: 4422 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4423 return -EFAULT; 4424 4425 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 4426 return -EOPNOTSUPP; 4427 4428 fallthrough; 4429 case SIOCGHWTSTAMP: 4430 real_dev = bond_option_active_slave_get_rcu(bond); 4431 if (!real_dev) 4432 return -EOPNOTSUPP; 4433 4434 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ); 4435 ifrr.ifr_ifru = ifr->ifr_ifru; 4436 4437 ops = real_dev->netdev_ops; 4438 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) { 4439 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd); 4440 if (res) 4441 return res; 4442 4443 ifr->ifr_ifru = ifrr.ifr_ifru; 4444 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 4445 return -EFAULT; 4446 4447 /* Set the BOND_PHC_INDEX flag to notify user space */ 4448 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 4449 4450 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ? 4451 -EFAULT : 0; 4452 } 4453 fallthrough; 4454 default: 4455 res = -EOPNOTSUPP; 4456 } 4457 4458 return res; 4459 } 4460 4461 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4462 { 4463 struct bonding *bond = netdev_priv(bond_dev); 4464 struct net_device *slave_dev = NULL; 4465 struct ifbond k_binfo; 4466 struct ifbond __user *u_binfo = NULL; 4467 struct ifslave k_sinfo; 4468 struct ifslave __user *u_sinfo = NULL; 4469 struct bond_opt_value newval; 4470 struct net *net; 4471 int res = 0; 4472 4473 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4474 4475 switch (cmd) { 4476 case SIOCBONDINFOQUERY: 4477 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4478 4479 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4480 return -EFAULT; 4481 4482 bond_info_query(bond_dev, &k_binfo); 4483 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4484 return -EFAULT; 4485 4486 return 0; 4487 case SIOCBONDSLAVEINFOQUERY: 4488 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4489 4490 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4491 return -EFAULT; 4492 4493 res = bond_slave_info_query(bond_dev, &k_sinfo); 4494 if (res == 0 && 4495 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4496 return -EFAULT; 4497 4498 return res; 4499 default: 4500 break; 4501 } 4502 4503 net = dev_net(bond_dev); 4504 4505 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4506 return -EPERM; 4507 4508 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4509 4510 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4511 4512 if (!slave_dev) 4513 return -ENODEV; 4514 4515 switch (cmd) { 4516 case SIOCBONDENSLAVE: 4517 res = bond_enslave(bond_dev, slave_dev, NULL); 4518 break; 4519 case SIOCBONDRELEASE: 4520 res = bond_release(bond_dev, slave_dev); 4521 break; 4522 case SIOCBONDSETHWADDR: 4523 res = bond_set_dev_addr(bond_dev, slave_dev); 4524 break; 4525 case SIOCBONDCHANGEACTIVE: 4526 bond_opt_initstr(&newval, slave_dev->name); 4527 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4528 &newval); 4529 break; 4530 default: 4531 res = -EOPNOTSUPP; 4532 } 4533 4534 return res; 4535 } 4536 4537 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4538 void __user *data, int cmd) 4539 { 4540 struct ifreq ifrdata = { .ifr_data = data }; 4541 4542 switch (cmd) { 4543 case BOND_INFO_QUERY_OLD: 4544 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4545 case BOND_SLAVE_INFO_QUERY_OLD: 4546 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4547 case BOND_ENSLAVE_OLD: 4548 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4549 case BOND_RELEASE_OLD: 4550 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4551 case BOND_SETHWADDR_OLD: 4552 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4553 case BOND_CHANGE_ACTIVE_OLD: 4554 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4555 } 4556 4557 return -EOPNOTSUPP; 4558 } 4559 4560 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4561 { 4562 struct bonding *bond = netdev_priv(bond_dev); 4563 4564 if (change & IFF_PROMISC) 4565 bond_set_promiscuity(bond, 4566 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4567 4568 if (change & IFF_ALLMULTI) 4569 bond_set_allmulti(bond, 4570 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4571 } 4572 4573 static void bond_set_rx_mode(struct net_device *bond_dev) 4574 { 4575 struct bonding *bond = netdev_priv(bond_dev); 4576 struct list_head *iter; 4577 struct slave *slave; 4578 4579 rcu_read_lock(); 4580 if (bond_uses_primary(bond)) { 4581 slave = rcu_dereference(bond->curr_active_slave); 4582 if (slave) { 4583 dev_uc_sync(slave->dev, bond_dev); 4584 dev_mc_sync(slave->dev, bond_dev); 4585 } 4586 } else { 4587 bond_for_each_slave_rcu(bond, slave, iter) { 4588 dev_uc_sync_multiple(slave->dev, bond_dev); 4589 dev_mc_sync_multiple(slave->dev, bond_dev); 4590 } 4591 } 4592 rcu_read_unlock(); 4593 } 4594 4595 static int bond_neigh_init(struct neighbour *n) 4596 { 4597 struct bonding *bond = netdev_priv(n->dev); 4598 const struct net_device_ops *slave_ops; 4599 struct neigh_parms parms; 4600 struct slave *slave; 4601 int ret = 0; 4602 4603 rcu_read_lock(); 4604 slave = bond_first_slave_rcu(bond); 4605 if (!slave) 4606 goto out; 4607 slave_ops = slave->dev->netdev_ops; 4608 if (!slave_ops->ndo_neigh_setup) 4609 goto out; 4610 4611 /* TODO: find another way [1] to implement this. 4612 * Passing a zeroed structure is fragile, 4613 * but at least we do not pass garbage. 4614 * 4615 * [1] One way would be that ndo_neigh_setup() never touch 4616 * struct neigh_parms, but propagate the new neigh_setup() 4617 * back to ___neigh_create() / neigh_parms_alloc() 4618 */ 4619 memset(&parms, 0, sizeof(parms)); 4620 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4621 4622 if (ret) 4623 goto out; 4624 4625 if (parms.neigh_setup) 4626 ret = parms.neigh_setup(n); 4627 out: 4628 rcu_read_unlock(); 4629 return ret; 4630 } 4631 4632 /* The bonding ndo_neigh_setup is called at init time beofre any 4633 * slave exists. So we must declare proxy setup function which will 4634 * be used at run time to resolve the actual slave neigh param setup. 4635 * 4636 * It's also called by master devices (such as vlans) to setup their 4637 * underlying devices. In that case - do nothing, we're already set up from 4638 * our init. 4639 */ 4640 static int bond_neigh_setup(struct net_device *dev, 4641 struct neigh_parms *parms) 4642 { 4643 /* modify only our neigh_parms */ 4644 if (parms->dev == dev) 4645 parms->neigh_setup = bond_neigh_init; 4646 4647 return 0; 4648 } 4649 4650 /* Change the MTU of all of a master's slaves to match the master */ 4651 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4652 { 4653 struct bonding *bond = netdev_priv(bond_dev); 4654 struct slave *slave, *rollback_slave; 4655 struct list_head *iter; 4656 int res = 0; 4657 4658 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4659 4660 bond_for_each_slave(bond, slave, iter) { 4661 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4662 slave, slave->dev->netdev_ops->ndo_change_mtu); 4663 4664 res = dev_set_mtu(slave->dev, new_mtu); 4665 4666 if (res) { 4667 /* If we failed to set the slave's mtu to the new value 4668 * we must abort the operation even in ACTIVE_BACKUP 4669 * mode, because if we allow the backup slaves to have 4670 * different mtu values than the active slave we'll 4671 * need to change their mtu when doing a failover. That 4672 * means changing their mtu from timer context, which 4673 * is probably not a good idea. 4674 */ 4675 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4676 res, new_mtu); 4677 goto unwind; 4678 } 4679 } 4680 4681 bond_dev->mtu = new_mtu; 4682 4683 return 0; 4684 4685 unwind: 4686 /* unwind from head to the slave that failed */ 4687 bond_for_each_slave(bond, rollback_slave, iter) { 4688 int tmp_res; 4689 4690 if (rollback_slave == slave) 4691 break; 4692 4693 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4694 if (tmp_res) 4695 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4696 tmp_res); 4697 } 4698 4699 return res; 4700 } 4701 4702 /* Change HW address 4703 * 4704 * Note that many devices must be down to change the HW address, and 4705 * downing the master releases all slaves. We can make bonds full of 4706 * bonding devices to test this, however. 4707 */ 4708 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4709 { 4710 struct bonding *bond = netdev_priv(bond_dev); 4711 struct slave *slave, *rollback_slave; 4712 struct sockaddr_storage *ss = addr, tmp_ss; 4713 struct list_head *iter; 4714 int res = 0; 4715 4716 if (BOND_MODE(bond) == BOND_MODE_ALB) 4717 return bond_alb_set_mac_address(bond_dev, addr); 4718 4719 4720 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4721 4722 /* If fail_over_mac is enabled, do nothing and return success. 4723 * Returning an error causes ifenslave to fail. 4724 */ 4725 if (bond->params.fail_over_mac && 4726 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4727 return 0; 4728 4729 if (!is_valid_ether_addr(ss->__data)) 4730 return -EADDRNOTAVAIL; 4731 4732 bond_for_each_slave(bond, slave, iter) { 4733 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4734 __func__, slave); 4735 res = dev_set_mac_address(slave->dev, addr, NULL); 4736 if (res) { 4737 /* TODO: consider downing the slave 4738 * and retry ? 4739 * User should expect communications 4740 * breakage anyway until ARP finish 4741 * updating, so... 4742 */ 4743 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4744 __func__, res); 4745 goto unwind; 4746 } 4747 } 4748 4749 /* success */ 4750 dev_addr_set(bond_dev, ss->__data); 4751 return 0; 4752 4753 unwind: 4754 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4755 tmp_ss.ss_family = bond_dev->type; 4756 4757 /* unwind from head to the slave that failed */ 4758 bond_for_each_slave(bond, rollback_slave, iter) { 4759 int tmp_res; 4760 4761 if (rollback_slave == slave) 4762 break; 4763 4764 tmp_res = dev_set_mac_address(rollback_slave->dev, 4765 (struct sockaddr *)&tmp_ss, NULL); 4766 if (tmp_res) { 4767 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4768 __func__, tmp_res); 4769 } 4770 } 4771 4772 return res; 4773 } 4774 4775 /** 4776 * bond_get_slave_by_id - get xmit slave with slave_id 4777 * @bond: bonding device that is transmitting 4778 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4779 * 4780 * This function tries to get slave with slave_id but in case 4781 * it fails, it tries to find the first available slave for transmission. 4782 */ 4783 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4784 int slave_id) 4785 { 4786 struct list_head *iter; 4787 struct slave *slave; 4788 int i = slave_id; 4789 4790 /* Here we start from the slave with slave_id */ 4791 bond_for_each_slave_rcu(bond, slave, iter) { 4792 if (--i < 0) { 4793 if (bond_slave_can_tx(slave)) 4794 return slave; 4795 } 4796 } 4797 4798 /* Here we start from the first slave up to slave_id */ 4799 i = slave_id; 4800 bond_for_each_slave_rcu(bond, slave, iter) { 4801 if (--i < 0) 4802 break; 4803 if (bond_slave_can_tx(slave)) 4804 return slave; 4805 } 4806 /* no slave that can tx has been found */ 4807 return NULL; 4808 } 4809 4810 /** 4811 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4812 * @bond: bonding device to use 4813 * 4814 * Based on the value of the bonding device's packets_per_slave parameter 4815 * this function generates a slave id, which is usually used as the next 4816 * slave to transmit through. 4817 */ 4818 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4819 { 4820 u32 slave_id; 4821 struct reciprocal_value reciprocal_packets_per_slave; 4822 int packets_per_slave = bond->params.packets_per_slave; 4823 4824 switch (packets_per_slave) { 4825 case 0: 4826 slave_id = get_random_u32(); 4827 break; 4828 case 1: 4829 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4830 break; 4831 default: 4832 reciprocal_packets_per_slave = 4833 bond->params.reciprocal_packets_per_slave; 4834 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4835 slave_id = reciprocal_divide(slave_id, 4836 reciprocal_packets_per_slave); 4837 break; 4838 } 4839 4840 return slave_id; 4841 } 4842 4843 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4844 struct sk_buff *skb) 4845 { 4846 struct slave *slave; 4847 int slave_cnt; 4848 u32 slave_id; 4849 4850 /* Start with the curr_active_slave that joined the bond as the 4851 * default for sending IGMP traffic. For failover purposes one 4852 * needs to maintain some consistency for the interface that will 4853 * send the join/membership reports. The curr_active_slave found 4854 * will send all of this type of traffic. 4855 */ 4856 if (skb->protocol == htons(ETH_P_IP)) { 4857 int noff = skb_network_offset(skb); 4858 struct iphdr *iph; 4859 4860 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4861 goto non_igmp; 4862 4863 iph = ip_hdr(skb); 4864 if (iph->protocol == IPPROTO_IGMP) { 4865 slave = rcu_dereference(bond->curr_active_slave); 4866 if (slave) 4867 return slave; 4868 return bond_get_slave_by_id(bond, 0); 4869 } 4870 } 4871 4872 non_igmp: 4873 slave_cnt = READ_ONCE(bond->slave_cnt); 4874 if (likely(slave_cnt)) { 4875 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4876 return bond_get_slave_by_id(bond, slave_id); 4877 } 4878 return NULL; 4879 } 4880 4881 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4882 struct xdp_buff *xdp) 4883 { 4884 struct slave *slave; 4885 int slave_cnt; 4886 u32 slave_id; 4887 const struct ethhdr *eth; 4888 void *data = xdp->data; 4889 4890 if (data + sizeof(struct ethhdr) > xdp->data_end) 4891 goto non_igmp; 4892 4893 eth = (struct ethhdr *)data; 4894 data += sizeof(struct ethhdr); 4895 4896 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4897 if (eth->h_proto == htons(ETH_P_IP)) { 4898 const struct iphdr *iph; 4899 4900 if (data + sizeof(struct iphdr) > xdp->data_end) 4901 goto non_igmp; 4902 4903 iph = (struct iphdr *)data; 4904 4905 if (iph->protocol == IPPROTO_IGMP) { 4906 slave = rcu_dereference(bond->curr_active_slave); 4907 if (slave) 4908 return slave; 4909 return bond_get_slave_by_id(bond, 0); 4910 } 4911 } 4912 4913 non_igmp: 4914 slave_cnt = READ_ONCE(bond->slave_cnt); 4915 if (likely(slave_cnt)) { 4916 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4917 return bond_get_slave_by_id(bond, slave_id); 4918 } 4919 return NULL; 4920 } 4921 4922 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4923 struct net_device *bond_dev) 4924 { 4925 struct bonding *bond = netdev_priv(bond_dev); 4926 struct slave *slave; 4927 4928 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4929 if (likely(slave)) 4930 return bond_dev_queue_xmit(bond, skb, slave->dev); 4931 4932 return bond_tx_drop(bond_dev, skb); 4933 } 4934 4935 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4936 { 4937 return rcu_dereference(bond->curr_active_slave); 4938 } 4939 4940 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4941 * the bond has a usable interface. 4942 */ 4943 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4944 struct net_device *bond_dev) 4945 { 4946 struct bonding *bond = netdev_priv(bond_dev); 4947 struct slave *slave; 4948 4949 slave = bond_xmit_activebackup_slave_get(bond); 4950 if (slave) 4951 return bond_dev_queue_xmit(bond, skb, slave->dev); 4952 4953 return bond_tx_drop(bond_dev, skb); 4954 } 4955 4956 /* Use this to update slave_array when (a) it's not appropriate to update 4957 * slave_array right away (note that update_slave_array() may sleep) 4958 * and / or (b) RTNL is not held. 4959 */ 4960 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 4961 { 4962 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 4963 } 4964 4965 /* Slave array work handler. Holds only RTNL */ 4966 static void bond_slave_arr_handler(struct work_struct *work) 4967 { 4968 struct bonding *bond = container_of(work, struct bonding, 4969 slave_arr_work.work); 4970 int ret; 4971 4972 if (!rtnl_trylock()) 4973 goto err; 4974 4975 ret = bond_update_slave_arr(bond, NULL); 4976 rtnl_unlock(); 4977 if (ret) { 4978 pr_warn_ratelimited("Failed to update slave array from WT\n"); 4979 goto err; 4980 } 4981 return; 4982 4983 err: 4984 bond_slave_arr_work_rearm(bond, 1); 4985 } 4986 4987 static void bond_skip_slave(struct bond_up_slave *slaves, 4988 struct slave *skipslave) 4989 { 4990 int idx; 4991 4992 /* Rare situation where caller has asked to skip a specific 4993 * slave but allocation failed (most likely!). BTW this is 4994 * only possible when the call is initiated from 4995 * __bond_release_one(). In this situation; overwrite the 4996 * skipslave entry in the array with the last entry from the 4997 * array to avoid a situation where the xmit path may choose 4998 * this to-be-skipped slave to send a packet out. 4999 */ 5000 for (idx = 0; slaves && idx < slaves->count; idx++) { 5001 if (skipslave == slaves->arr[idx]) { 5002 slaves->arr[idx] = 5003 slaves->arr[slaves->count - 1]; 5004 slaves->count--; 5005 break; 5006 } 5007 } 5008 } 5009 5010 static void bond_set_slave_arr(struct bonding *bond, 5011 struct bond_up_slave *usable_slaves, 5012 struct bond_up_slave *all_slaves) 5013 { 5014 struct bond_up_slave *usable, *all; 5015 5016 usable = rtnl_dereference(bond->usable_slaves); 5017 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5018 kfree_rcu(usable, rcu); 5019 5020 all = rtnl_dereference(bond->all_slaves); 5021 rcu_assign_pointer(bond->all_slaves, all_slaves); 5022 kfree_rcu(all, rcu); 5023 } 5024 5025 static void bond_reset_slave_arr(struct bonding *bond) 5026 { 5027 struct bond_up_slave *usable, *all; 5028 5029 usable = rtnl_dereference(bond->usable_slaves); 5030 if (usable) { 5031 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5032 kfree_rcu(usable, rcu); 5033 } 5034 5035 all = rtnl_dereference(bond->all_slaves); 5036 if (all) { 5037 RCU_INIT_POINTER(bond->all_slaves, NULL); 5038 kfree_rcu(all, rcu); 5039 } 5040 } 5041 5042 /* Build the usable slaves array in control path for modes that use xmit-hash 5043 * to determine the slave interface - 5044 * (a) BOND_MODE_8023AD 5045 * (b) BOND_MODE_XOR 5046 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5047 * 5048 * The caller is expected to hold RTNL only and NO other lock! 5049 */ 5050 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5051 { 5052 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5053 struct slave *slave; 5054 struct list_head *iter; 5055 int agg_id = 0; 5056 int ret = 0; 5057 5058 might_sleep(); 5059 5060 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5061 bond->slave_cnt), GFP_KERNEL); 5062 all_slaves = kzalloc(struct_size(all_slaves, arr, 5063 bond->slave_cnt), GFP_KERNEL); 5064 if (!usable_slaves || !all_slaves) { 5065 ret = -ENOMEM; 5066 goto out; 5067 } 5068 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5069 struct ad_info ad_info; 5070 5071 spin_lock_bh(&bond->mode_lock); 5072 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5073 spin_unlock_bh(&bond->mode_lock); 5074 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5075 /* No active aggragator means it's not safe to use 5076 * the previous array. 5077 */ 5078 bond_reset_slave_arr(bond); 5079 goto out; 5080 } 5081 spin_unlock_bh(&bond->mode_lock); 5082 agg_id = ad_info.aggregator_id; 5083 } 5084 bond_for_each_slave(bond, slave, iter) { 5085 if (skipslave == slave) 5086 continue; 5087 5088 all_slaves->arr[all_slaves->count++] = slave; 5089 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5090 struct aggregator *agg; 5091 5092 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5093 if (!agg || agg->aggregator_identifier != agg_id) 5094 continue; 5095 } 5096 if (!bond_slave_can_tx(slave)) 5097 continue; 5098 5099 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5100 usable_slaves->count); 5101 5102 usable_slaves->arr[usable_slaves->count++] = slave; 5103 } 5104 5105 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5106 return ret; 5107 out: 5108 if (ret != 0 && skipslave) { 5109 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5110 skipslave); 5111 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5112 skipslave); 5113 } 5114 kfree_rcu(all_slaves, rcu); 5115 kfree_rcu(usable_slaves, rcu); 5116 5117 return ret; 5118 } 5119 5120 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5121 struct sk_buff *skb, 5122 struct bond_up_slave *slaves) 5123 { 5124 struct slave *slave; 5125 unsigned int count; 5126 u32 hash; 5127 5128 hash = bond_xmit_hash(bond, skb); 5129 count = slaves ? READ_ONCE(slaves->count) : 0; 5130 if (unlikely(!count)) 5131 return NULL; 5132 5133 slave = slaves->arr[hash % count]; 5134 return slave; 5135 } 5136 5137 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5138 struct xdp_buff *xdp) 5139 { 5140 struct bond_up_slave *slaves; 5141 unsigned int count; 5142 u32 hash; 5143 5144 hash = bond_xmit_hash_xdp(bond, xdp); 5145 slaves = rcu_dereference(bond->usable_slaves); 5146 count = slaves ? READ_ONCE(slaves->count) : 0; 5147 if (unlikely(!count)) 5148 return NULL; 5149 5150 return slaves->arr[hash % count]; 5151 } 5152 5153 /* Use this Xmit function for 3AD as well as XOR modes. The current 5154 * usable slave array is formed in the control path. The xmit function 5155 * just calculates hash and sends the packet out. 5156 */ 5157 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5158 struct net_device *dev) 5159 { 5160 struct bonding *bond = netdev_priv(dev); 5161 struct bond_up_slave *slaves; 5162 struct slave *slave; 5163 5164 slaves = rcu_dereference(bond->usable_slaves); 5165 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5166 if (likely(slave)) 5167 return bond_dev_queue_xmit(bond, skb, slave->dev); 5168 5169 return bond_tx_drop(dev, skb); 5170 } 5171 5172 /* in broadcast mode, we send everything to all usable interfaces. */ 5173 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5174 struct net_device *bond_dev) 5175 { 5176 struct bonding *bond = netdev_priv(bond_dev); 5177 struct slave *slave = NULL; 5178 struct list_head *iter; 5179 bool xmit_suc = false; 5180 bool skb_used = false; 5181 5182 bond_for_each_slave_rcu(bond, slave, iter) { 5183 struct sk_buff *skb2; 5184 5185 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5186 continue; 5187 5188 if (bond_is_last_slave(bond, slave)) { 5189 skb2 = skb; 5190 skb_used = true; 5191 } else { 5192 skb2 = skb_clone(skb, GFP_ATOMIC); 5193 if (!skb2) { 5194 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5195 bond_dev->name, __func__); 5196 continue; 5197 } 5198 } 5199 5200 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5201 xmit_suc = true; 5202 } 5203 5204 if (!skb_used) 5205 dev_kfree_skb_any(skb); 5206 5207 if (xmit_suc) 5208 return NETDEV_TX_OK; 5209 5210 dev_core_stats_tx_dropped_inc(bond_dev); 5211 return NET_XMIT_DROP; 5212 } 5213 5214 /*------------------------- Device initialization ---------------------------*/ 5215 5216 /* Lookup the slave that corresponds to a qid */ 5217 static inline int bond_slave_override(struct bonding *bond, 5218 struct sk_buff *skb) 5219 { 5220 struct slave *slave = NULL; 5221 struct list_head *iter; 5222 5223 if (!skb_rx_queue_recorded(skb)) 5224 return 1; 5225 5226 /* Find out if any slaves have the same mapping as this skb. */ 5227 bond_for_each_slave_rcu(bond, slave, iter) { 5228 if (slave->queue_id == skb_get_queue_mapping(skb)) { 5229 if (bond_slave_is_up(slave) && 5230 slave->link == BOND_LINK_UP) { 5231 bond_dev_queue_xmit(bond, skb, slave->dev); 5232 return 0; 5233 } 5234 /* If the slave isn't UP, use default transmit policy. */ 5235 break; 5236 } 5237 } 5238 5239 return 1; 5240 } 5241 5242 5243 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5244 struct net_device *sb_dev) 5245 { 5246 /* This helper function exists to help dev_pick_tx get the correct 5247 * destination queue. Using a helper function skips a call to 5248 * skb_tx_hash and will put the skbs in the queue we expect on their 5249 * way down to the bonding driver. 5250 */ 5251 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5252 5253 /* Save the original txq to restore before passing to the driver */ 5254 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5255 5256 if (unlikely(txq >= dev->real_num_tx_queues)) { 5257 do { 5258 txq -= dev->real_num_tx_queues; 5259 } while (txq >= dev->real_num_tx_queues); 5260 } 5261 return txq; 5262 } 5263 5264 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5265 struct sk_buff *skb, 5266 bool all_slaves) 5267 { 5268 struct bonding *bond = netdev_priv(master_dev); 5269 struct bond_up_slave *slaves; 5270 struct slave *slave = NULL; 5271 5272 switch (BOND_MODE(bond)) { 5273 case BOND_MODE_ROUNDROBIN: 5274 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5275 break; 5276 case BOND_MODE_ACTIVEBACKUP: 5277 slave = bond_xmit_activebackup_slave_get(bond); 5278 break; 5279 case BOND_MODE_8023AD: 5280 case BOND_MODE_XOR: 5281 if (all_slaves) 5282 slaves = rcu_dereference(bond->all_slaves); 5283 else 5284 slaves = rcu_dereference(bond->usable_slaves); 5285 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5286 break; 5287 case BOND_MODE_BROADCAST: 5288 break; 5289 case BOND_MODE_ALB: 5290 slave = bond_xmit_alb_slave_get(bond, skb); 5291 break; 5292 case BOND_MODE_TLB: 5293 slave = bond_xmit_tlb_slave_get(bond, skb); 5294 break; 5295 default: 5296 /* Should never happen, mode already checked */ 5297 WARN_ONCE(true, "Unknown bonding mode"); 5298 break; 5299 } 5300 5301 if (slave) 5302 return slave->dev; 5303 return NULL; 5304 } 5305 5306 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5307 { 5308 switch (sk->sk_family) { 5309 #if IS_ENABLED(CONFIG_IPV6) 5310 case AF_INET6: 5311 if (ipv6_only_sock(sk) || 5312 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5313 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5314 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5315 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5316 break; 5317 } 5318 fallthrough; 5319 #endif 5320 default: /* AF_INET */ 5321 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5322 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5323 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5324 break; 5325 } 5326 5327 flow->ports.src = inet_sk(sk)->inet_sport; 5328 flow->ports.dst = inet_sk(sk)->inet_dport; 5329 } 5330 5331 /** 5332 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5333 * @sk: socket to use for headers 5334 * 5335 * This function will extract the necessary field from the socket and use 5336 * them to generate a hash based on the LAYER34 xmit_policy. 5337 * Assumes that sk is a TCP or UDP socket. 5338 */ 5339 static u32 bond_sk_hash_l34(struct sock *sk) 5340 { 5341 struct flow_keys flow; 5342 u32 hash; 5343 5344 bond_sk_to_flow(sk, &flow); 5345 5346 /* L4 */ 5347 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5348 /* L3 */ 5349 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5350 } 5351 5352 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5353 struct sock *sk) 5354 { 5355 struct bond_up_slave *slaves; 5356 struct slave *slave; 5357 unsigned int count; 5358 u32 hash; 5359 5360 slaves = rcu_dereference(bond->usable_slaves); 5361 count = slaves ? READ_ONCE(slaves->count) : 0; 5362 if (unlikely(!count)) 5363 return NULL; 5364 5365 hash = bond_sk_hash_l34(sk); 5366 slave = slaves->arr[hash % count]; 5367 5368 return slave->dev; 5369 } 5370 5371 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5372 struct sock *sk) 5373 { 5374 struct bonding *bond = netdev_priv(dev); 5375 struct net_device *lower = NULL; 5376 5377 rcu_read_lock(); 5378 if (bond_sk_check(bond)) 5379 lower = __bond_sk_get_lower_dev(bond, sk); 5380 rcu_read_unlock(); 5381 5382 return lower; 5383 } 5384 5385 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5386 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5387 struct net_device *dev) 5388 { 5389 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5390 5391 /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded 5392 * was true, if tls_device_down is running in parallel, but it's OK, 5393 * because bond_get_slave_by_dev has a NULL check. 5394 */ 5395 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5396 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5397 return bond_tx_drop(dev, skb); 5398 } 5399 #endif 5400 5401 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5402 { 5403 struct bonding *bond = netdev_priv(dev); 5404 5405 if (bond_should_override_tx_queue(bond) && 5406 !bond_slave_override(bond, skb)) 5407 return NETDEV_TX_OK; 5408 5409 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5410 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk)) 5411 return bond_tls_device_xmit(bond, skb, dev); 5412 #endif 5413 5414 switch (BOND_MODE(bond)) { 5415 case BOND_MODE_ROUNDROBIN: 5416 return bond_xmit_roundrobin(skb, dev); 5417 case BOND_MODE_ACTIVEBACKUP: 5418 return bond_xmit_activebackup(skb, dev); 5419 case BOND_MODE_8023AD: 5420 case BOND_MODE_XOR: 5421 return bond_3ad_xor_xmit(skb, dev); 5422 case BOND_MODE_BROADCAST: 5423 return bond_xmit_broadcast(skb, dev); 5424 case BOND_MODE_ALB: 5425 return bond_alb_xmit(skb, dev); 5426 case BOND_MODE_TLB: 5427 return bond_tlb_xmit(skb, dev); 5428 default: 5429 /* Should never happen, mode already checked */ 5430 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5431 WARN_ON_ONCE(1); 5432 return bond_tx_drop(dev, skb); 5433 } 5434 } 5435 5436 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5437 { 5438 struct bonding *bond = netdev_priv(dev); 5439 netdev_tx_t ret = NETDEV_TX_OK; 5440 5441 /* If we risk deadlock from transmitting this in the 5442 * netpoll path, tell netpoll to queue the frame for later tx 5443 */ 5444 if (unlikely(is_netpoll_tx_blocked(dev))) 5445 return NETDEV_TX_BUSY; 5446 5447 rcu_read_lock(); 5448 if (bond_has_slaves(bond)) 5449 ret = __bond_start_xmit(skb, dev); 5450 else 5451 ret = bond_tx_drop(dev, skb); 5452 rcu_read_unlock(); 5453 5454 return ret; 5455 } 5456 5457 static struct net_device * 5458 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5459 { 5460 struct bonding *bond = netdev_priv(bond_dev); 5461 struct slave *slave; 5462 5463 /* Caller needs to hold rcu_read_lock() */ 5464 5465 switch (BOND_MODE(bond)) { 5466 case BOND_MODE_ROUNDROBIN: 5467 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5468 break; 5469 5470 case BOND_MODE_ACTIVEBACKUP: 5471 slave = bond_xmit_activebackup_slave_get(bond); 5472 break; 5473 5474 case BOND_MODE_8023AD: 5475 case BOND_MODE_XOR: 5476 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5477 break; 5478 5479 default: 5480 /* Should never happen. Mode guarded by bond_xdp_check() */ 5481 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond)); 5482 WARN_ON_ONCE(1); 5483 return NULL; 5484 } 5485 5486 if (slave) 5487 return slave->dev; 5488 5489 return NULL; 5490 } 5491 5492 static int bond_xdp_xmit(struct net_device *bond_dev, 5493 int n, struct xdp_frame **frames, u32 flags) 5494 { 5495 int nxmit, err = -ENXIO; 5496 5497 rcu_read_lock(); 5498 5499 for (nxmit = 0; nxmit < n; nxmit++) { 5500 struct xdp_frame *frame = frames[nxmit]; 5501 struct xdp_frame *frames1[] = {frame}; 5502 struct net_device *slave_dev; 5503 struct xdp_buff xdp; 5504 5505 xdp_convert_frame_to_buff(frame, &xdp); 5506 5507 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5508 if (!slave_dev) { 5509 err = -ENXIO; 5510 break; 5511 } 5512 5513 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5514 if (err < 1) 5515 break; 5516 } 5517 5518 rcu_read_unlock(); 5519 5520 /* If error happened on the first frame then we can pass the error up, otherwise 5521 * report the number of frames that were xmitted. 5522 */ 5523 if (err < 0) 5524 return (nxmit == 0 ? err : nxmit); 5525 5526 return nxmit; 5527 } 5528 5529 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5530 struct netlink_ext_ack *extack) 5531 { 5532 struct bonding *bond = netdev_priv(dev); 5533 struct list_head *iter; 5534 struct slave *slave, *rollback_slave; 5535 struct bpf_prog *old_prog; 5536 struct netdev_bpf xdp = { 5537 .command = XDP_SETUP_PROG, 5538 .flags = 0, 5539 .prog = prog, 5540 .extack = extack, 5541 }; 5542 int err; 5543 5544 ASSERT_RTNL(); 5545 5546 if (!bond_xdp_check(bond)) 5547 return -EOPNOTSUPP; 5548 5549 old_prog = bond->xdp_prog; 5550 bond->xdp_prog = prog; 5551 5552 bond_for_each_slave(bond, slave, iter) { 5553 struct net_device *slave_dev = slave->dev; 5554 5555 if (!slave_dev->netdev_ops->ndo_bpf || 5556 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5557 SLAVE_NL_ERR(dev, slave_dev, extack, 5558 "Slave device does not support XDP"); 5559 err = -EOPNOTSUPP; 5560 goto err; 5561 } 5562 5563 if (dev_xdp_prog_count(slave_dev) > 0) { 5564 SLAVE_NL_ERR(dev, slave_dev, extack, 5565 "Slave has XDP program loaded, please unload before enslaving"); 5566 err = -EOPNOTSUPP; 5567 goto err; 5568 } 5569 5570 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5571 if (err < 0) { 5572 /* ndo_bpf() sets extack error message */ 5573 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5574 goto err; 5575 } 5576 if (prog) 5577 bpf_prog_inc(prog); 5578 } 5579 5580 if (prog) { 5581 static_branch_inc(&bpf_master_redirect_enabled_key); 5582 } else if (old_prog) { 5583 bpf_prog_put(old_prog); 5584 static_branch_dec(&bpf_master_redirect_enabled_key); 5585 } 5586 5587 return 0; 5588 5589 err: 5590 /* unwind the program changes */ 5591 bond->xdp_prog = old_prog; 5592 xdp.prog = old_prog; 5593 xdp.extack = NULL; /* do not overwrite original error */ 5594 5595 bond_for_each_slave(bond, rollback_slave, iter) { 5596 struct net_device *slave_dev = rollback_slave->dev; 5597 int err_unwind; 5598 5599 if (slave == rollback_slave) 5600 break; 5601 5602 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5603 if (err_unwind < 0) 5604 slave_err(dev, slave_dev, 5605 "Error %d when unwinding XDP program change\n", err_unwind); 5606 else if (xdp.prog) 5607 bpf_prog_inc(xdp.prog); 5608 } 5609 return err; 5610 } 5611 5612 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5613 { 5614 switch (xdp->command) { 5615 case XDP_SETUP_PROG: 5616 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5617 default: 5618 return -EINVAL; 5619 } 5620 } 5621 5622 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5623 { 5624 if (speed == 0 || speed == SPEED_UNKNOWN) 5625 speed = slave->speed; 5626 else 5627 speed = min(speed, slave->speed); 5628 5629 return speed; 5630 } 5631 5632 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5633 struct ethtool_link_ksettings *cmd) 5634 { 5635 struct bonding *bond = netdev_priv(bond_dev); 5636 struct list_head *iter; 5637 struct slave *slave; 5638 u32 speed = 0; 5639 5640 cmd->base.duplex = DUPLEX_UNKNOWN; 5641 cmd->base.port = PORT_OTHER; 5642 5643 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5644 * do not need to check mode. Though link speed might not represent 5645 * the true receive or transmit bandwidth (not all modes are symmetric) 5646 * this is an accurate maximum. 5647 */ 5648 bond_for_each_slave(bond, slave, iter) { 5649 if (bond_slave_can_tx(slave)) { 5650 if (slave->speed != SPEED_UNKNOWN) { 5651 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5652 speed = bond_mode_bcast_speed(slave, 5653 speed); 5654 else 5655 speed += slave->speed; 5656 } 5657 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5658 slave->duplex != DUPLEX_UNKNOWN) 5659 cmd->base.duplex = slave->duplex; 5660 } 5661 } 5662 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5663 5664 return 0; 5665 } 5666 5667 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5668 struct ethtool_drvinfo *drvinfo) 5669 { 5670 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5671 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5672 BOND_ABI_VERSION); 5673 } 5674 5675 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5676 struct ethtool_ts_info *info) 5677 { 5678 struct bonding *bond = netdev_priv(bond_dev); 5679 const struct ethtool_ops *ops; 5680 struct net_device *real_dev; 5681 struct phy_device *phydev; 5682 int ret = 0; 5683 5684 rcu_read_lock(); 5685 real_dev = bond_option_active_slave_get_rcu(bond); 5686 dev_hold(real_dev); 5687 rcu_read_unlock(); 5688 5689 if (real_dev) { 5690 ops = real_dev->ethtool_ops; 5691 phydev = real_dev->phydev; 5692 5693 if (phy_has_tsinfo(phydev)) { 5694 ret = phy_ts_info(phydev, info); 5695 goto out; 5696 } else if (ops->get_ts_info) { 5697 ret = ops->get_ts_info(real_dev, info); 5698 goto out; 5699 } 5700 } 5701 5702 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE | 5703 SOF_TIMESTAMPING_SOFTWARE; 5704 info->phc_index = -1; 5705 5706 out: 5707 dev_put(real_dev); 5708 return ret; 5709 } 5710 5711 static const struct ethtool_ops bond_ethtool_ops = { 5712 .get_drvinfo = bond_ethtool_get_drvinfo, 5713 .get_link = ethtool_op_get_link, 5714 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5715 .get_ts_info = bond_ethtool_get_ts_info, 5716 }; 5717 5718 static const struct net_device_ops bond_netdev_ops = { 5719 .ndo_init = bond_init, 5720 .ndo_uninit = bond_uninit, 5721 .ndo_open = bond_open, 5722 .ndo_stop = bond_close, 5723 .ndo_start_xmit = bond_start_xmit, 5724 .ndo_select_queue = bond_select_queue, 5725 .ndo_get_stats64 = bond_get_stats, 5726 .ndo_eth_ioctl = bond_eth_ioctl, 5727 .ndo_siocbond = bond_do_ioctl, 5728 .ndo_siocdevprivate = bond_siocdevprivate, 5729 .ndo_change_rx_flags = bond_change_rx_flags, 5730 .ndo_set_rx_mode = bond_set_rx_mode, 5731 .ndo_change_mtu = bond_change_mtu, 5732 .ndo_set_mac_address = bond_set_mac_address, 5733 .ndo_neigh_setup = bond_neigh_setup, 5734 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5735 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5736 #ifdef CONFIG_NET_POLL_CONTROLLER 5737 .ndo_netpoll_setup = bond_netpoll_setup, 5738 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5739 .ndo_poll_controller = bond_poll_controller, 5740 #endif 5741 .ndo_add_slave = bond_enslave, 5742 .ndo_del_slave = bond_release, 5743 .ndo_fix_features = bond_fix_features, 5744 .ndo_features_check = passthru_features_check, 5745 .ndo_get_xmit_slave = bond_xmit_get_slave, 5746 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5747 .ndo_bpf = bond_xdp, 5748 .ndo_xdp_xmit = bond_xdp_xmit, 5749 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5750 }; 5751 5752 static const struct device_type bond_type = { 5753 .name = "bond", 5754 }; 5755 5756 static void bond_destructor(struct net_device *bond_dev) 5757 { 5758 struct bonding *bond = netdev_priv(bond_dev); 5759 5760 if (bond->wq) 5761 destroy_workqueue(bond->wq); 5762 5763 if (bond->rr_tx_counter) 5764 free_percpu(bond->rr_tx_counter); 5765 } 5766 5767 void bond_setup(struct net_device *bond_dev) 5768 { 5769 struct bonding *bond = netdev_priv(bond_dev); 5770 5771 spin_lock_init(&bond->mode_lock); 5772 bond->params = bonding_defaults; 5773 5774 /* Initialize pointers */ 5775 bond->dev = bond_dev; 5776 5777 /* Initialize the device entry points */ 5778 ether_setup(bond_dev); 5779 bond_dev->max_mtu = ETH_MAX_MTU; 5780 bond_dev->netdev_ops = &bond_netdev_ops; 5781 bond_dev->ethtool_ops = &bond_ethtool_ops; 5782 5783 bond_dev->needs_free_netdev = true; 5784 bond_dev->priv_destructor = bond_destructor; 5785 5786 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5787 5788 /* Initialize the device options */ 5789 bond_dev->flags |= IFF_MASTER; 5790 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5791 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5792 5793 #ifdef CONFIG_XFRM_OFFLOAD 5794 /* set up xfrm device ops (only supported in active-backup right now) */ 5795 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5796 INIT_LIST_HEAD(&bond->ipsec_list); 5797 spin_lock_init(&bond->ipsec_lock); 5798 #endif /* CONFIG_XFRM_OFFLOAD */ 5799 5800 /* don't acquire bond device's netif_tx_lock when transmitting */ 5801 bond_dev->features |= NETIF_F_LLTX; 5802 5803 /* By default, we declare the bond to be fully 5804 * VLAN hardware accelerated capable. Special 5805 * care is taken in the various xmit functions 5806 * when there are slaves that are not hw accel 5807 * capable 5808 */ 5809 5810 /* Don't allow bond devices to change network namespaces. */ 5811 bond_dev->features |= NETIF_F_NETNS_LOCAL; 5812 5813 bond_dev->hw_features = BOND_VLAN_FEATURES | 5814 NETIF_F_HW_VLAN_CTAG_RX | 5815 NETIF_F_HW_VLAN_CTAG_FILTER; 5816 5817 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5818 bond_dev->features |= bond_dev->hw_features; 5819 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5820 #ifdef CONFIG_XFRM_OFFLOAD 5821 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5822 /* Only enable XFRM features if this is an active-backup config */ 5823 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5824 bond_dev->features |= BOND_XFRM_FEATURES; 5825 #endif /* CONFIG_XFRM_OFFLOAD */ 5826 } 5827 5828 /* Destroy a bonding device. 5829 * Must be under rtnl_lock when this function is called. 5830 */ 5831 static void bond_uninit(struct net_device *bond_dev) 5832 { 5833 struct bonding *bond = netdev_priv(bond_dev); 5834 struct bond_up_slave *usable, *all; 5835 struct list_head *iter; 5836 struct slave *slave; 5837 5838 bond_netpoll_cleanup(bond_dev); 5839 5840 /* Release the bonded slaves */ 5841 bond_for_each_slave(bond, slave, iter) 5842 __bond_release_one(bond_dev, slave->dev, true, true); 5843 netdev_info(bond_dev, "Released all slaves\n"); 5844 5845 usable = rtnl_dereference(bond->usable_slaves); 5846 if (usable) { 5847 RCU_INIT_POINTER(bond->usable_slaves, NULL); 5848 kfree_rcu(usable, rcu); 5849 } 5850 5851 all = rtnl_dereference(bond->all_slaves); 5852 if (all) { 5853 RCU_INIT_POINTER(bond->all_slaves, NULL); 5854 kfree_rcu(all, rcu); 5855 } 5856 5857 list_del(&bond->bond_list); 5858 5859 bond_debug_unregister(bond); 5860 } 5861 5862 /*------------------------- Module initialization ---------------------------*/ 5863 5864 static int bond_check_params(struct bond_params *params) 5865 { 5866 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 5867 struct bond_opt_value newval; 5868 const struct bond_opt_value *valptr; 5869 int arp_all_targets_value = 0; 5870 u16 ad_actor_sys_prio = 0; 5871 u16 ad_user_port_key = 0; 5872 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 5873 int arp_ip_count; 5874 int bond_mode = BOND_MODE_ROUNDROBIN; 5875 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 5876 int lacp_fast = 0; 5877 int tlb_dynamic_lb; 5878 5879 /* Convert string parameters. */ 5880 if (mode) { 5881 bond_opt_initstr(&newval, mode); 5882 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 5883 if (!valptr) { 5884 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 5885 return -EINVAL; 5886 } 5887 bond_mode = valptr->value; 5888 } 5889 5890 if (xmit_hash_policy) { 5891 if (bond_mode == BOND_MODE_ROUNDROBIN || 5892 bond_mode == BOND_MODE_ACTIVEBACKUP || 5893 bond_mode == BOND_MODE_BROADCAST) { 5894 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 5895 bond_mode_name(bond_mode)); 5896 } else { 5897 bond_opt_initstr(&newval, xmit_hash_policy); 5898 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 5899 &newval); 5900 if (!valptr) { 5901 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 5902 xmit_hash_policy); 5903 return -EINVAL; 5904 } 5905 xmit_hashtype = valptr->value; 5906 } 5907 } 5908 5909 if (lacp_rate) { 5910 if (bond_mode != BOND_MODE_8023AD) { 5911 pr_info("lacp_rate param is irrelevant in mode %s\n", 5912 bond_mode_name(bond_mode)); 5913 } else { 5914 bond_opt_initstr(&newval, lacp_rate); 5915 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 5916 &newval); 5917 if (!valptr) { 5918 pr_err("Error: Invalid lacp rate \"%s\"\n", 5919 lacp_rate); 5920 return -EINVAL; 5921 } 5922 lacp_fast = valptr->value; 5923 } 5924 } 5925 5926 if (ad_select) { 5927 bond_opt_initstr(&newval, ad_select); 5928 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 5929 &newval); 5930 if (!valptr) { 5931 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 5932 return -EINVAL; 5933 } 5934 params->ad_select = valptr->value; 5935 if (bond_mode != BOND_MODE_8023AD) 5936 pr_warn("ad_select param only affects 802.3ad mode\n"); 5937 } else { 5938 params->ad_select = BOND_AD_STABLE; 5939 } 5940 5941 if (max_bonds < 0) { 5942 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 5943 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 5944 max_bonds = BOND_DEFAULT_MAX_BONDS; 5945 } 5946 5947 if (miimon < 0) { 5948 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5949 miimon, INT_MAX); 5950 miimon = 0; 5951 } 5952 5953 if (updelay < 0) { 5954 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5955 updelay, INT_MAX); 5956 updelay = 0; 5957 } 5958 5959 if (downdelay < 0) { 5960 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 5961 downdelay, INT_MAX); 5962 downdelay = 0; 5963 } 5964 5965 if ((use_carrier != 0) && (use_carrier != 1)) { 5966 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 5967 use_carrier); 5968 use_carrier = 1; 5969 } 5970 5971 if (num_peer_notif < 0 || num_peer_notif > 255) { 5972 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 5973 num_peer_notif); 5974 num_peer_notif = 1; 5975 } 5976 5977 /* reset values for 802.3ad/TLB/ALB */ 5978 if (!bond_mode_uses_arp(bond_mode)) { 5979 if (!miimon) { 5980 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"); 5981 pr_warn("Forcing miimon to 100msec\n"); 5982 miimon = BOND_DEFAULT_MIIMON; 5983 } 5984 } 5985 5986 if (tx_queues < 1 || tx_queues > 255) { 5987 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 5988 tx_queues, BOND_DEFAULT_TX_QUEUES); 5989 tx_queues = BOND_DEFAULT_TX_QUEUES; 5990 } 5991 5992 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 5993 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 5994 all_slaves_active); 5995 all_slaves_active = 0; 5996 } 5997 5998 if (resend_igmp < 0 || resend_igmp > 255) { 5999 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6000 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6001 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6002 } 6003 6004 bond_opt_initval(&newval, packets_per_slave); 6005 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6006 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6007 packets_per_slave, USHRT_MAX); 6008 packets_per_slave = 1; 6009 } 6010 6011 if (bond_mode == BOND_MODE_ALB) { 6012 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", 6013 updelay); 6014 } 6015 6016 if (!miimon) { 6017 if (updelay || downdelay) { 6018 /* just warn the user the up/down delay will have 6019 * no effect since miimon is zero... 6020 */ 6021 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", 6022 updelay, downdelay); 6023 } 6024 } else { 6025 /* don't allow arp monitoring */ 6026 if (arp_interval) { 6027 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6028 miimon, arp_interval); 6029 arp_interval = 0; 6030 } 6031 6032 if ((updelay % miimon) != 0) { 6033 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6034 updelay, miimon, (updelay / miimon) * miimon); 6035 } 6036 6037 updelay /= miimon; 6038 6039 if ((downdelay % miimon) != 0) { 6040 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6041 downdelay, miimon, 6042 (downdelay / miimon) * miimon); 6043 } 6044 6045 downdelay /= miimon; 6046 } 6047 6048 if (arp_interval < 0) { 6049 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6050 arp_interval, INT_MAX); 6051 arp_interval = 0; 6052 } 6053 6054 for (arp_ip_count = 0, i = 0; 6055 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6056 __be32 ip; 6057 6058 /* not a complete check, but good enough to catch mistakes */ 6059 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6060 !bond_is_ip_target_ok(ip)) { 6061 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6062 arp_ip_target[i]); 6063 arp_interval = 0; 6064 } else { 6065 if (bond_get_targets_ip(arp_target, ip) == -1) 6066 arp_target[arp_ip_count++] = ip; 6067 else 6068 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6069 &ip); 6070 } 6071 } 6072 6073 if (arp_interval && !arp_ip_count) { 6074 /* don't allow arping if no arp_ip_target given... */ 6075 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6076 arp_interval); 6077 arp_interval = 0; 6078 } 6079 6080 if (arp_validate) { 6081 if (!arp_interval) { 6082 pr_err("arp_validate requires arp_interval\n"); 6083 return -EINVAL; 6084 } 6085 6086 bond_opt_initstr(&newval, arp_validate); 6087 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6088 &newval); 6089 if (!valptr) { 6090 pr_err("Error: invalid arp_validate \"%s\"\n", 6091 arp_validate); 6092 return -EINVAL; 6093 } 6094 arp_validate_value = valptr->value; 6095 } else { 6096 arp_validate_value = 0; 6097 } 6098 6099 if (arp_all_targets) { 6100 bond_opt_initstr(&newval, arp_all_targets); 6101 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6102 &newval); 6103 if (!valptr) { 6104 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6105 arp_all_targets); 6106 arp_all_targets_value = 0; 6107 } else { 6108 arp_all_targets_value = valptr->value; 6109 } 6110 } 6111 6112 if (miimon) { 6113 pr_info("MII link monitoring set to %d ms\n", miimon); 6114 } else if (arp_interval) { 6115 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6116 arp_validate_value); 6117 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6118 arp_interval, valptr->string, arp_ip_count); 6119 6120 for (i = 0; i < arp_ip_count; i++) 6121 pr_cont(" %s", arp_ip_target[i]); 6122 6123 pr_cont("\n"); 6124 6125 } else if (max_bonds) { 6126 /* miimon and arp_interval not set, we need one so things 6127 * work as expected, see bonding.txt for details 6128 */ 6129 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"); 6130 } 6131 6132 if (primary && !bond_mode_uses_primary(bond_mode)) { 6133 /* currently, using a primary only makes sense 6134 * in active backup, TLB or ALB modes 6135 */ 6136 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6137 primary, bond_mode_name(bond_mode)); 6138 primary = NULL; 6139 } 6140 6141 if (primary && primary_reselect) { 6142 bond_opt_initstr(&newval, primary_reselect); 6143 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6144 &newval); 6145 if (!valptr) { 6146 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6147 primary_reselect); 6148 return -EINVAL; 6149 } 6150 primary_reselect_value = valptr->value; 6151 } else { 6152 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6153 } 6154 6155 if (fail_over_mac) { 6156 bond_opt_initstr(&newval, fail_over_mac); 6157 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6158 &newval); 6159 if (!valptr) { 6160 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6161 fail_over_mac); 6162 return -EINVAL; 6163 } 6164 fail_over_mac_value = valptr->value; 6165 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6166 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6167 } else { 6168 fail_over_mac_value = BOND_FOM_NONE; 6169 } 6170 6171 bond_opt_initstr(&newval, "default"); 6172 valptr = bond_opt_parse( 6173 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6174 &newval); 6175 if (!valptr) { 6176 pr_err("Error: No ad_actor_sys_prio default value"); 6177 return -EINVAL; 6178 } 6179 ad_actor_sys_prio = valptr->value; 6180 6181 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6182 &newval); 6183 if (!valptr) { 6184 pr_err("Error: No ad_user_port_key default value"); 6185 return -EINVAL; 6186 } 6187 ad_user_port_key = valptr->value; 6188 6189 bond_opt_initstr(&newval, "default"); 6190 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6191 if (!valptr) { 6192 pr_err("Error: No tlb_dynamic_lb default value"); 6193 return -EINVAL; 6194 } 6195 tlb_dynamic_lb = valptr->value; 6196 6197 if (lp_interval == 0) { 6198 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6199 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6200 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6201 } 6202 6203 /* fill params struct with the proper values */ 6204 params->mode = bond_mode; 6205 params->xmit_policy = xmit_hashtype; 6206 params->miimon = miimon; 6207 params->num_peer_notif = num_peer_notif; 6208 params->arp_interval = arp_interval; 6209 params->arp_validate = arp_validate_value; 6210 params->arp_all_targets = arp_all_targets_value; 6211 params->missed_max = 2; 6212 params->updelay = updelay; 6213 params->downdelay = downdelay; 6214 params->peer_notif_delay = 0; 6215 params->use_carrier = use_carrier; 6216 params->lacp_active = 1; 6217 params->lacp_fast = lacp_fast; 6218 params->primary[0] = 0; 6219 params->primary_reselect = primary_reselect_value; 6220 params->fail_over_mac = fail_over_mac_value; 6221 params->tx_queues = tx_queues; 6222 params->all_slaves_active = all_slaves_active; 6223 params->resend_igmp = resend_igmp; 6224 params->min_links = min_links; 6225 params->lp_interval = lp_interval; 6226 params->packets_per_slave = packets_per_slave; 6227 params->tlb_dynamic_lb = tlb_dynamic_lb; 6228 params->ad_actor_sys_prio = ad_actor_sys_prio; 6229 eth_zero_addr(params->ad_actor_system); 6230 params->ad_user_port_key = ad_user_port_key; 6231 if (packets_per_slave > 0) { 6232 params->reciprocal_packets_per_slave = 6233 reciprocal_value(packets_per_slave); 6234 } else { 6235 /* reciprocal_packets_per_slave is unused if 6236 * packets_per_slave is 0 or 1, just initialize it 6237 */ 6238 params->reciprocal_packets_per_slave = 6239 (struct reciprocal_value) { 0 }; 6240 } 6241 6242 if (primary) 6243 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6244 6245 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6246 #if IS_ENABLED(CONFIG_IPV6) 6247 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6248 #endif 6249 6250 return 0; 6251 } 6252 6253 /* Called from registration process */ 6254 static int bond_init(struct net_device *bond_dev) 6255 { 6256 struct bonding *bond = netdev_priv(bond_dev); 6257 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6258 6259 netdev_dbg(bond_dev, "Begin bond_init\n"); 6260 6261 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 6262 if (!bond->wq) 6263 return -ENOMEM; 6264 6265 spin_lock_init(&bond->stats_lock); 6266 netdev_lockdep_set_classes(bond_dev); 6267 6268 list_add_tail(&bond->bond_list, &bn->dev_list); 6269 6270 bond_prepare_sysfs_group(bond); 6271 6272 bond_debug_register(bond); 6273 6274 /* Ensure valid dev_addr */ 6275 if (is_zero_ether_addr(bond_dev->dev_addr) && 6276 bond_dev->addr_assign_type == NET_ADDR_PERM) 6277 eth_hw_addr_random(bond_dev); 6278 6279 return 0; 6280 } 6281 6282 unsigned int bond_get_num_tx_queues(void) 6283 { 6284 return tx_queues; 6285 } 6286 6287 /* Create a new bond based on the specified name and bonding parameters. 6288 * If name is NULL, obtain a suitable "bond%d" name for us. 6289 * Caller must NOT hold rtnl_lock; we need to release it here before we 6290 * set up our sysfs entries. 6291 */ 6292 int bond_create(struct net *net, const char *name) 6293 { 6294 struct net_device *bond_dev; 6295 struct bonding *bond; 6296 int res = -ENOMEM; 6297 6298 rtnl_lock(); 6299 6300 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6301 name ? name : "bond%d", NET_NAME_UNKNOWN, 6302 bond_setup, tx_queues); 6303 if (!bond_dev) 6304 goto out; 6305 6306 bond = netdev_priv(bond_dev); 6307 dev_net_set(bond_dev, net); 6308 bond_dev->rtnl_link_ops = &bond_link_ops; 6309 6310 res = register_netdevice(bond_dev); 6311 if (res < 0) { 6312 free_netdev(bond_dev); 6313 goto out; 6314 } 6315 6316 netif_carrier_off(bond_dev); 6317 6318 bond_work_init_all(bond); 6319 6320 out: 6321 rtnl_unlock(); 6322 return res; 6323 } 6324 6325 static int __net_init bond_net_init(struct net *net) 6326 { 6327 struct bond_net *bn = net_generic(net, bond_net_id); 6328 6329 bn->net = net; 6330 INIT_LIST_HEAD(&bn->dev_list); 6331 6332 bond_create_proc_dir(bn); 6333 bond_create_sysfs(bn); 6334 6335 return 0; 6336 } 6337 6338 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6339 { 6340 struct bond_net *bn; 6341 struct net *net; 6342 LIST_HEAD(list); 6343 6344 list_for_each_entry(net, net_list, exit_list) { 6345 bn = net_generic(net, bond_net_id); 6346 bond_destroy_sysfs(bn); 6347 } 6348 6349 /* Kill off any bonds created after unregistering bond rtnl ops */ 6350 rtnl_lock(); 6351 list_for_each_entry(net, net_list, exit_list) { 6352 struct bonding *bond, *tmp_bond; 6353 6354 bn = net_generic(net, bond_net_id); 6355 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6356 unregister_netdevice_queue(bond->dev, &list); 6357 } 6358 unregister_netdevice_many(&list); 6359 rtnl_unlock(); 6360 6361 list_for_each_entry(net, net_list, exit_list) { 6362 bn = net_generic(net, bond_net_id); 6363 bond_destroy_proc_dir(bn); 6364 } 6365 } 6366 6367 static struct pernet_operations bond_net_ops = { 6368 .init = bond_net_init, 6369 .exit_batch = bond_net_exit_batch, 6370 .id = &bond_net_id, 6371 .size = sizeof(struct bond_net), 6372 }; 6373 6374 static int __init bonding_init(void) 6375 { 6376 int i; 6377 int res; 6378 6379 res = bond_check_params(&bonding_defaults); 6380 if (res) 6381 goto out; 6382 6383 res = register_pernet_subsys(&bond_net_ops); 6384 if (res) 6385 goto out; 6386 6387 res = bond_netlink_init(); 6388 if (res) 6389 goto err_link; 6390 6391 bond_create_debugfs(); 6392 6393 for (i = 0; i < max_bonds; i++) { 6394 res = bond_create(&init_net, NULL); 6395 if (res) 6396 goto err; 6397 } 6398 6399 skb_flow_dissector_init(&flow_keys_bonding, 6400 flow_keys_bonding_keys, 6401 ARRAY_SIZE(flow_keys_bonding_keys)); 6402 6403 register_netdevice_notifier(&bond_netdev_notifier); 6404 out: 6405 return res; 6406 err: 6407 bond_destroy_debugfs(); 6408 bond_netlink_fini(); 6409 err_link: 6410 unregister_pernet_subsys(&bond_net_ops); 6411 goto out; 6412 6413 } 6414 6415 static void __exit bonding_exit(void) 6416 { 6417 unregister_netdevice_notifier(&bond_netdev_notifier); 6418 6419 bond_destroy_debugfs(); 6420 6421 bond_netlink_fini(); 6422 unregister_pernet_subsys(&bond_net_ops); 6423 6424 #ifdef CONFIG_NET_POLL_CONTROLLER 6425 /* Make sure we don't have an imbalance on our netpoll blocking */ 6426 WARN_ON(atomic_read(&netpoll_block_tx)); 6427 #endif 6428 } 6429 6430 module_init(bonding_init); 6431 module_exit(bonding_exit); 6432 MODULE_LICENSE("GPL"); 6433 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6434 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6435