1 // SPDX-License-Identifier: GPL-1.0+ 2 /* 3 * originally based on the dummy device. 4 * 5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 6 * Based on dummy.c, and eql.c devices. 7 * 8 * bonding.c: an Ethernet Bonding driver 9 * 10 * This is useful to talk to a Cisco EtherChannel compatible equipment: 11 * Cisco 5500 12 * Sun Trunking (Solaris) 13 * Alteon AceDirector Trunks 14 * Linux Bonding 15 * and probably many L2 switches ... 16 * 17 * How it works: 18 * ifconfig bond0 ipaddress netmask up 19 * will setup a network device, with an ip address. No mac address 20 * will be assigned at this time. The hw mac address will come from 21 * the first slave bonded to the channel. All slaves will then use 22 * this hw mac address. 23 * 24 * ifconfig bond0 down 25 * will release all slaves, marking them as down. 26 * 27 * ifenslave bond0 eth0 28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 29 * a: be used as initial mac address 30 * b: if a hw mac address already is there, eth0's hw mac address 31 * will then be set from bond0. 32 * 33 */ 34 35 #include <linux/kernel.h> 36 #include <linux/module.h> 37 #include <linux/types.h> 38 #include <linux/fcntl.h> 39 #include <linux/filter.h> 40 #include <linux/interrupt.h> 41 #include <linux/ptrace.h> 42 #include <linux/ioport.h> 43 #include <linux/in.h> 44 #include <net/ip.h> 45 #include <linux/ip.h> 46 #include <linux/icmp.h> 47 #include <linux/icmpv6.h> 48 #include <linux/tcp.h> 49 #include <linux/udp.h> 50 #include <linux/slab.h> 51 #include <linux/string.h> 52 #include <linux/init.h> 53 #include <linux/timer.h> 54 #include <linux/socket.h> 55 #include <linux/ctype.h> 56 #include <linux/inet.h> 57 #include <linux/bitops.h> 58 #include <linux/io.h> 59 #include <asm/dma.h> 60 #include <linux/uaccess.h> 61 #include <linux/errno.h> 62 #include <linux/netdevice.h> 63 #include <linux/inetdevice.h> 64 #include <linux/igmp.h> 65 #include <linux/etherdevice.h> 66 #include <linux/skbuff.h> 67 #include <net/sock.h> 68 #include <linux/rtnetlink.h> 69 #include <linux/smp.h> 70 #include <linux/if_ether.h> 71 #include <net/arp.h> 72 #include <linux/mii.h> 73 #include <linux/ethtool.h> 74 #include <linux/if_vlan.h> 75 #include <linux/if_bonding.h> 76 #include <linux/phy.h> 77 #include <linux/jiffies.h> 78 #include <linux/preempt.h> 79 #include <net/route.h> 80 #include <net/net_namespace.h> 81 #include <net/netns/generic.h> 82 #include <net/pkt_sched.h> 83 #include <linux/rculist.h> 84 #include <net/flow_dissector.h> 85 #include <net/xfrm.h> 86 #include <net/bonding.h> 87 #include <net/bond_3ad.h> 88 #include <net/bond_alb.h> 89 #if IS_ENABLED(CONFIG_TLS_DEVICE) 90 #include <net/tls.h> 91 #endif 92 #include <net/ip6_route.h> 93 #include <net/netdev_lock.h> 94 #include <net/xdp.h> 95 96 #include "bonding_priv.h" 97 98 /*---------------------------- Module parameters ----------------------------*/ 99 100 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 101 102 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 103 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 104 static int num_peer_notif = 1; 105 static int miimon; 106 static int updelay; 107 static int downdelay; 108 static int use_carrier = 1; 109 static char *mode; 110 static char *primary; 111 static char *primary_reselect; 112 static char *lacp_rate; 113 static int min_links; 114 static char *ad_select; 115 static char *xmit_hash_policy; 116 static int arp_interval; 117 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 118 static char *arp_validate; 119 static char *arp_all_targets; 120 static char *fail_over_mac; 121 static int all_slaves_active; 122 static struct bond_params bonding_defaults; 123 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 124 static int packets_per_slave = 1; 125 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 126 127 module_param(max_bonds, int, 0); 128 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 129 module_param(tx_queues, int, 0); 130 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 131 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 132 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 133 "failover event (alias of num_unsol_na)"); 134 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 135 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 136 "failover event (alias of num_grat_arp)"); 137 module_param(miimon, int, 0); 138 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 139 module_param(updelay, int, 0); 140 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 141 module_param(downdelay, int, 0); 142 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 143 "in milliseconds"); 144 module_param(use_carrier, int, 0); 145 MODULE_PARM_DESC(use_carrier, "option obsolete, use_carrier cannot be disabled"); 146 module_param(mode, charp, 0); 147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 148 "1 for active-backup, 2 for balance-xor, " 149 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 150 "6 for balance-alb"); 151 module_param(primary, charp, 0); 152 MODULE_PARM_DESC(primary, "Primary network device to use"); 153 module_param(primary_reselect, charp, 0); 154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 155 "once it comes up; " 156 "0 for always (default), " 157 "1 for only if speed of primary is " 158 "better, " 159 "2 for only on active slave " 160 "failure"); 161 module_param(lacp_rate, charp, 0); 162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 163 "0 for slow, 1 for fast"); 164 module_param(ad_select, charp, 0); 165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 166 "0 for stable (default), 1 for bandwidth, " 167 "2 for count"); 168 module_param(min_links, int, 0); 169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 170 171 module_param(xmit_hash_policy, charp, 0); 172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 173 "0 for layer 2 (default), 1 for layer 3+4, " 174 "2 for layer 2+3, 3 for encap layer 2+3, " 175 "4 for encap layer 3+4, 5 for vlan+srcmac"); 176 module_param(arp_interval, int, 0); 177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 178 module_param_array(arp_ip_target, charp, NULL, 0); 179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 180 module_param(arp_validate, charp, 0); 181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 182 "0 for none (default), 1 for active, " 183 "2 for backup, 3 for all"); 184 module_param(arp_all_targets, charp, 0); 185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 186 module_param(fail_over_mac, charp, 0); 187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 188 "the same MAC; 0 for none (default), " 189 "1 for active, 2 for follow"); 190 module_param(all_slaves_active, int, 0); 191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 192 "by setting active flag for all slaves; " 193 "0 for never (default), 1 for always."); 194 module_param(resend_igmp, int, 0); 195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 196 "link failure"); 197 module_param(packets_per_slave, int, 0); 198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 199 "mode; 0 for a random slave, 1 packet per " 200 "slave (default), >1 packets per slave."); 201 module_param(lp_interval, uint, 0); 202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 203 "the bonding driver sends learning packets to " 204 "each slaves peer switch. The default is 1."); 205 206 /*----------------------------- Global variables ----------------------------*/ 207 208 #ifdef CONFIG_NET_POLL_CONTROLLER 209 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 210 #endif 211 212 unsigned int bond_net_id __read_mostly; 213 214 DEFINE_STATIC_KEY_FALSE(bond_bcast_neigh_enabled); 215 216 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 217 { 218 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 219 .offset = offsetof(struct flow_keys, control), 220 }, 221 { 222 .key_id = FLOW_DISSECTOR_KEY_BASIC, 223 .offset = offsetof(struct flow_keys, basic), 224 }, 225 { 226 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 227 .offset = offsetof(struct flow_keys, addrs.v4addrs), 228 }, 229 { 230 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 231 .offset = offsetof(struct flow_keys, addrs.v6addrs), 232 }, 233 { 234 .key_id = FLOW_DISSECTOR_KEY_TIPC, 235 .offset = offsetof(struct flow_keys, addrs.tipckey), 236 }, 237 { 238 .key_id = FLOW_DISSECTOR_KEY_PORTS, 239 .offset = offsetof(struct flow_keys, ports), 240 }, 241 { 242 .key_id = FLOW_DISSECTOR_KEY_ICMP, 243 .offset = offsetof(struct flow_keys, icmp), 244 }, 245 { 246 .key_id = FLOW_DISSECTOR_KEY_VLAN, 247 .offset = offsetof(struct flow_keys, vlan), 248 }, 249 { 250 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 251 .offset = offsetof(struct flow_keys, tags), 252 }, 253 { 254 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 255 .offset = offsetof(struct flow_keys, keyid), 256 }, 257 }; 258 259 static struct flow_dissector flow_keys_bonding __read_mostly; 260 261 /*-------------------------- Forward declarations ---------------------------*/ 262 263 static int bond_init(struct net_device *bond_dev); 264 static void bond_uninit(struct net_device *bond_dev); 265 static void bond_get_stats(struct net_device *bond_dev, 266 struct rtnl_link_stats64 *stats); 267 static void bond_slave_arr_handler(struct work_struct *work); 268 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 269 int mod); 270 static void bond_netdev_notify_work(struct work_struct *work); 271 272 /*---------------------------- General routines -----------------------------*/ 273 274 const char *bond_mode_name(int mode) 275 { 276 static const char *names[] = { 277 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 278 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 279 [BOND_MODE_XOR] = "load balancing (xor)", 280 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 281 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 282 [BOND_MODE_TLB] = "transmit load balancing", 283 [BOND_MODE_ALB] = "adaptive load balancing", 284 }; 285 286 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 287 return "unknown"; 288 289 return names[mode]; 290 } 291 292 /** 293 * bond_dev_queue_xmit - Prepare skb for xmit. 294 * 295 * @bond: bond device that got this skb for tx. 296 * @skb: hw accel VLAN tagged skb to transmit 297 * @slave_dev: slave that is supposed to xmit this skbuff 298 */ 299 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 300 struct net_device *slave_dev) 301 { 302 skb->dev = slave_dev; 303 304 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 305 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 306 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 307 308 if (unlikely(netpoll_tx_running(bond->dev))) 309 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 310 311 return dev_queue_xmit(skb); 312 } 313 314 static bool bond_sk_check(struct bonding *bond) 315 { 316 switch (BOND_MODE(bond)) { 317 case BOND_MODE_8023AD: 318 case BOND_MODE_XOR: 319 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 320 return true; 321 fallthrough; 322 default: 323 return false; 324 } 325 } 326 327 bool bond_xdp_check(struct bonding *bond, int mode) 328 { 329 switch (mode) { 330 case BOND_MODE_ROUNDROBIN: 331 case BOND_MODE_ACTIVEBACKUP: 332 return true; 333 case BOND_MODE_8023AD: 334 case BOND_MODE_XOR: 335 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 336 * payload is not in the packet due to hardware offload. 337 */ 338 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 339 return true; 340 fallthrough; 341 default: 342 return false; 343 } 344 } 345 346 /*---------------------------------- VLAN -----------------------------------*/ 347 348 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 349 * We don't protect the slave list iteration with a lock because: 350 * a. This operation is performed in IOCTL context, 351 * b. The operation is protected by the RTNL semaphore in the 8021q code, 352 * c. Holding a lock with BH disabled while directly calling a base driver 353 * entry point is generally a BAD idea. 354 * 355 * The design of synchronization/protection for this operation in the 8021q 356 * module is good for one or more VLAN devices over a single physical device 357 * and cannot be extended for a teaming solution like bonding, so there is a 358 * potential race condition here where a net device from the vlan group might 359 * be referenced (either by a base driver or the 8021q code) while it is being 360 * removed from the system. However, it turns out we're not making matters 361 * worse, and if it works for regular VLAN usage it will work here too. 362 */ 363 364 /** 365 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 366 * @bond_dev: bonding net device that got called 367 * @proto: network protocol ID 368 * @vid: vlan id being added 369 */ 370 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 371 __be16 proto, u16 vid) 372 { 373 struct bonding *bond = netdev_priv(bond_dev); 374 struct slave *slave, *rollback_slave; 375 struct list_head *iter; 376 int res; 377 378 bond_for_each_slave(bond, slave, iter) { 379 res = vlan_vid_add(slave->dev, proto, vid); 380 if (res) 381 goto unwind; 382 } 383 384 return 0; 385 386 unwind: 387 /* unwind to the slave that failed */ 388 bond_for_each_slave(bond, rollback_slave, iter) { 389 if (rollback_slave == slave) 390 break; 391 392 vlan_vid_del(rollback_slave->dev, proto, vid); 393 } 394 395 return res; 396 } 397 398 /** 399 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 400 * @bond_dev: bonding net device that got called 401 * @proto: network protocol ID 402 * @vid: vlan id being removed 403 */ 404 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 405 __be16 proto, u16 vid) 406 { 407 struct bonding *bond = netdev_priv(bond_dev); 408 struct list_head *iter; 409 struct slave *slave; 410 411 bond_for_each_slave(bond, slave, iter) 412 vlan_vid_del(slave->dev, proto, vid); 413 414 if (bond_is_lb(bond)) 415 bond_alb_clear_vlan(bond, vid); 416 417 return 0; 418 } 419 420 /*---------------------------------- XFRM -----------------------------------*/ 421 422 #ifdef CONFIG_XFRM_OFFLOAD 423 /** 424 * bond_ipsec_dev - Get active device for IPsec offload 425 * @xs: pointer to transformer state struct 426 * 427 * Context: caller must hold rcu_read_lock. 428 * 429 * Return: the device for ipsec offload, or NULL if not exist. 430 **/ 431 static struct net_device *bond_ipsec_dev(struct xfrm_state *xs) 432 { 433 struct net_device *bond_dev = xs->xso.dev; 434 struct bonding *bond; 435 struct slave *slave; 436 437 bond = netdev_priv(bond_dev); 438 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 439 return NULL; 440 441 slave = rcu_dereference(bond->curr_active_slave); 442 if (!slave) 443 return NULL; 444 445 if (!xs->xso.real_dev) 446 return NULL; 447 448 if (xs->xso.real_dev != slave->dev) 449 pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n", 450 bond_dev->name, slave->dev->name, xs->xso.real_dev->name); 451 452 return slave->dev; 453 } 454 455 /** 456 * bond_ipsec_add_sa - program device with a security association 457 * @bond_dev: pointer to the bond net device 458 * @xs: pointer to transformer state struct 459 * @extack: extack point to fill failure reason 460 **/ 461 static int bond_ipsec_add_sa(struct net_device *bond_dev, 462 struct xfrm_state *xs, 463 struct netlink_ext_ack *extack) 464 { 465 struct net_device *real_dev; 466 netdevice_tracker tracker; 467 struct bond_ipsec *ipsec; 468 struct bonding *bond; 469 struct slave *slave; 470 int err; 471 472 if (!bond_dev) 473 return -EINVAL; 474 475 rcu_read_lock(); 476 bond = netdev_priv(bond_dev); 477 slave = rcu_dereference(bond->curr_active_slave); 478 real_dev = slave ? slave->dev : NULL; 479 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 480 rcu_read_unlock(); 481 if (!real_dev) { 482 err = -ENODEV; 483 goto out; 484 } 485 486 if (!real_dev->xfrmdev_ops || 487 !real_dev->xfrmdev_ops->xdo_dev_state_add || 488 netif_is_bond_master(real_dev)) { 489 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload"); 490 err = -EINVAL; 491 goto out; 492 } 493 494 ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL); 495 if (!ipsec) { 496 err = -ENOMEM; 497 goto out; 498 } 499 500 err = real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev, xs, extack); 501 if (!err) { 502 xs->xso.real_dev = real_dev; 503 ipsec->xs = xs; 504 INIT_LIST_HEAD(&ipsec->list); 505 mutex_lock(&bond->ipsec_lock); 506 list_add(&ipsec->list, &bond->ipsec_list); 507 mutex_unlock(&bond->ipsec_lock); 508 } else { 509 kfree(ipsec); 510 } 511 out: 512 netdev_put(real_dev, &tracker); 513 return err; 514 } 515 516 static void bond_ipsec_add_sa_all(struct bonding *bond) 517 { 518 struct net_device *bond_dev = bond->dev; 519 struct net_device *real_dev; 520 struct bond_ipsec *ipsec; 521 struct slave *slave; 522 523 slave = rtnl_dereference(bond->curr_active_slave); 524 real_dev = slave ? slave->dev : NULL; 525 if (!real_dev) 526 return; 527 528 mutex_lock(&bond->ipsec_lock); 529 if (!real_dev->xfrmdev_ops || 530 !real_dev->xfrmdev_ops->xdo_dev_state_add || 531 netif_is_bond_master(real_dev)) { 532 if (!list_empty(&bond->ipsec_list)) 533 slave_warn(bond_dev, real_dev, 534 "%s: no slave xdo_dev_state_add\n", 535 __func__); 536 goto out; 537 } 538 539 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 540 /* If new state is added before ipsec_lock acquired */ 541 if (ipsec->xs->xso.real_dev == real_dev) 542 continue; 543 544 if (real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev, 545 ipsec->xs, NULL)) { 546 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__); 547 continue; 548 } 549 550 spin_lock_bh(&ipsec->xs->lock); 551 /* xs might have been killed by the user during the migration 552 * to the new dev, but bond_ipsec_del_sa() should have done 553 * nothing, as xso.real_dev is NULL. 554 * Delete it from the device we just added it to. The pending 555 * bond_ipsec_free_sa() call will do the rest of the cleanup. 556 */ 557 if (ipsec->xs->km.state == XFRM_STATE_DEAD && 558 real_dev->xfrmdev_ops->xdo_dev_state_delete) 559 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, 560 ipsec->xs); 561 ipsec->xs->xso.real_dev = real_dev; 562 spin_unlock_bh(&ipsec->xs->lock); 563 } 564 out: 565 mutex_unlock(&bond->ipsec_lock); 566 } 567 568 /** 569 * bond_ipsec_del_sa - clear out this specific SA 570 * @bond_dev: pointer to the bond net device 571 * @xs: pointer to transformer state struct 572 **/ 573 static void bond_ipsec_del_sa(struct net_device *bond_dev, 574 struct xfrm_state *xs) 575 { 576 struct net_device *real_dev; 577 578 if (!bond_dev || !xs->xso.real_dev) 579 return; 580 581 real_dev = xs->xso.real_dev; 582 583 if (!real_dev->xfrmdev_ops || 584 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 585 netif_is_bond_master(real_dev)) { 586 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__); 587 return; 588 } 589 590 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, xs); 591 } 592 593 static void bond_ipsec_del_sa_all(struct bonding *bond) 594 { 595 struct net_device *bond_dev = bond->dev; 596 struct net_device *real_dev; 597 struct bond_ipsec *ipsec; 598 struct slave *slave; 599 600 slave = rtnl_dereference(bond->curr_active_slave); 601 real_dev = slave ? slave->dev : NULL; 602 if (!real_dev) 603 return; 604 605 mutex_lock(&bond->ipsec_lock); 606 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 607 if (!ipsec->xs->xso.real_dev) 608 continue; 609 610 if (!real_dev->xfrmdev_ops || 611 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 612 netif_is_bond_master(real_dev)) { 613 slave_warn(bond_dev, real_dev, 614 "%s: no slave xdo_dev_state_delete\n", 615 __func__); 616 continue; 617 } 618 619 spin_lock_bh(&ipsec->xs->lock); 620 ipsec->xs->xso.real_dev = NULL; 621 /* Don't double delete states killed by the user. */ 622 if (ipsec->xs->km.state != XFRM_STATE_DEAD) 623 real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, 624 ipsec->xs); 625 spin_unlock_bh(&ipsec->xs->lock); 626 627 if (real_dev->xfrmdev_ops->xdo_dev_state_free) 628 real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev, 629 ipsec->xs); 630 } 631 mutex_unlock(&bond->ipsec_lock); 632 } 633 634 static void bond_ipsec_free_sa(struct net_device *bond_dev, 635 struct xfrm_state *xs) 636 { 637 struct net_device *real_dev; 638 struct bond_ipsec *ipsec; 639 struct bonding *bond; 640 641 if (!bond_dev) 642 return; 643 644 bond = netdev_priv(bond_dev); 645 646 mutex_lock(&bond->ipsec_lock); 647 if (!xs->xso.real_dev) 648 goto out; 649 650 real_dev = xs->xso.real_dev; 651 652 xs->xso.real_dev = NULL; 653 if (real_dev->xfrmdev_ops && 654 real_dev->xfrmdev_ops->xdo_dev_state_free) 655 real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev, xs); 656 out: 657 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 658 if (ipsec->xs == xs) { 659 list_del(&ipsec->list); 660 kfree(ipsec); 661 break; 662 } 663 } 664 mutex_unlock(&bond->ipsec_lock); 665 } 666 667 /** 668 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 669 * @skb: current data packet 670 * @xs: pointer to transformer state struct 671 **/ 672 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 673 { 674 struct net_device *real_dev; 675 676 rcu_read_lock(); 677 real_dev = bond_ipsec_dev(xs); 678 if (!real_dev || netif_is_bond_master(real_dev)) { 679 rcu_read_unlock(); 680 return false; 681 } 682 683 rcu_read_unlock(); 684 return true; 685 } 686 687 /** 688 * bond_advance_esn_state - ESN support for IPSec HW offload 689 * @xs: pointer to transformer state struct 690 **/ 691 static void bond_advance_esn_state(struct xfrm_state *xs) 692 { 693 struct net_device *real_dev; 694 695 rcu_read_lock(); 696 real_dev = bond_ipsec_dev(xs); 697 if (!real_dev) 698 goto out; 699 700 if (!real_dev->xfrmdev_ops || 701 !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 702 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name); 703 goto out; 704 } 705 706 real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs); 707 out: 708 rcu_read_unlock(); 709 } 710 711 /** 712 * bond_xfrm_update_stats - Update xfrm state 713 * @xs: pointer to transformer state struct 714 **/ 715 static void bond_xfrm_update_stats(struct xfrm_state *xs) 716 { 717 struct net_device *real_dev; 718 719 rcu_read_lock(); 720 real_dev = bond_ipsec_dev(xs); 721 if (!real_dev) 722 goto out; 723 724 if (!real_dev->xfrmdev_ops || 725 !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) { 726 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name); 727 goto out; 728 } 729 730 real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs); 731 out: 732 rcu_read_unlock(); 733 } 734 735 static const struct xfrmdev_ops bond_xfrmdev_ops = { 736 .xdo_dev_state_add = bond_ipsec_add_sa, 737 .xdo_dev_state_delete = bond_ipsec_del_sa, 738 .xdo_dev_state_free = bond_ipsec_free_sa, 739 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 740 .xdo_dev_state_advance_esn = bond_advance_esn_state, 741 .xdo_dev_state_update_stats = bond_xfrm_update_stats, 742 }; 743 #endif /* CONFIG_XFRM_OFFLOAD */ 744 745 /*------------------------------- Link status -------------------------------*/ 746 747 /* Set the carrier state for the master according to the state of its 748 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 749 * do special 802.3ad magic. 750 * 751 * Returns zero if carrier state does not change, nonzero if it does. 752 */ 753 int bond_set_carrier(struct bonding *bond) 754 { 755 struct list_head *iter; 756 struct slave *slave; 757 758 if (!bond_has_slaves(bond)) 759 goto down; 760 761 if (BOND_MODE(bond) == BOND_MODE_8023AD) 762 return bond_3ad_set_carrier(bond); 763 764 bond_for_each_slave(bond, slave, iter) { 765 if (slave->link == BOND_LINK_UP) { 766 if (!netif_carrier_ok(bond->dev)) { 767 netif_carrier_on(bond->dev); 768 return 1; 769 } 770 return 0; 771 } 772 } 773 774 down: 775 if (netif_carrier_ok(bond->dev)) { 776 netif_carrier_off(bond->dev); 777 return 1; 778 } 779 return 0; 780 } 781 782 /* Get link speed and duplex from the slave's base driver 783 * using ethtool. If for some reason the call fails or the 784 * values are invalid, set speed and duplex to -1, 785 * and return. Return 1 if speed or duplex settings are 786 * UNKNOWN; 0 otherwise. 787 */ 788 static int bond_update_speed_duplex(struct slave *slave) 789 { 790 struct net_device *slave_dev = slave->dev; 791 struct ethtool_link_ksettings ecmd; 792 int res; 793 794 slave->speed = SPEED_UNKNOWN; 795 slave->duplex = DUPLEX_UNKNOWN; 796 797 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 798 if (res < 0) 799 return 1; 800 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 801 return 1; 802 switch (ecmd.base.duplex) { 803 case DUPLEX_FULL: 804 case DUPLEX_HALF: 805 break; 806 default: 807 return 1; 808 } 809 810 slave->speed = ecmd.base.speed; 811 slave->duplex = ecmd.base.duplex; 812 813 return 0; 814 } 815 816 const char *bond_slave_link_status(s8 link) 817 { 818 switch (link) { 819 case BOND_LINK_UP: 820 return "up"; 821 case BOND_LINK_FAIL: 822 return "going down"; 823 case BOND_LINK_DOWN: 824 return "down"; 825 case BOND_LINK_BACK: 826 return "going back"; 827 default: 828 return "unknown"; 829 } 830 } 831 832 /*----------------------------- Multicast list ------------------------------*/ 833 834 /* Push the promiscuity flag down to appropriate slaves */ 835 static int bond_set_promiscuity(struct bonding *bond, int inc) 836 { 837 struct list_head *iter; 838 int err = 0; 839 840 if (bond_uses_primary(bond)) { 841 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 842 843 if (curr_active) 844 err = dev_set_promiscuity(curr_active->dev, inc); 845 } else { 846 struct slave *slave; 847 848 bond_for_each_slave(bond, slave, iter) { 849 err = dev_set_promiscuity(slave->dev, inc); 850 if (err) 851 return err; 852 } 853 } 854 return err; 855 } 856 857 /* Push the allmulti flag down to all slaves */ 858 static int bond_set_allmulti(struct bonding *bond, int inc) 859 { 860 struct list_head *iter; 861 int err = 0; 862 863 if (bond_uses_primary(bond)) { 864 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 865 866 if (curr_active) 867 err = dev_set_allmulti(curr_active->dev, inc); 868 } else { 869 struct slave *slave; 870 871 bond_for_each_slave(bond, slave, iter) { 872 err = dev_set_allmulti(slave->dev, inc); 873 if (err) 874 return err; 875 } 876 } 877 return err; 878 } 879 880 /* Retrieve the list of registered multicast addresses for the bonding 881 * device and retransmit an IGMP JOIN request to the current active 882 * slave. 883 */ 884 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 885 { 886 struct bonding *bond = container_of(work, struct bonding, 887 mcast_work.work); 888 889 if (!rtnl_trylock()) { 890 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 891 return; 892 } 893 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 894 895 if (bond->igmp_retrans > 1) { 896 bond->igmp_retrans--; 897 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 898 } 899 rtnl_unlock(); 900 } 901 902 /* Flush bond's hardware addresses from slave */ 903 static void bond_hw_addr_flush(struct net_device *bond_dev, 904 struct net_device *slave_dev) 905 { 906 struct bonding *bond = netdev_priv(bond_dev); 907 908 dev_uc_unsync(slave_dev, bond_dev); 909 dev_mc_unsync(slave_dev, bond_dev); 910 911 if (BOND_MODE(bond) == BOND_MODE_8023AD) 912 dev_mc_del(slave_dev, lacpdu_mcast_addr); 913 } 914 915 /*--------------------------- Active slave change ---------------------------*/ 916 917 /* Update the hardware address list and promisc/allmulti for the new and 918 * old active slaves (if any). Modes that are not using primary keep all 919 * slaves up date at all times; only the modes that use primary need to call 920 * this function to swap these settings during a failover. 921 */ 922 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 923 struct slave *old_active) 924 { 925 if (old_active) { 926 if (bond->dev->flags & IFF_PROMISC) 927 dev_set_promiscuity(old_active->dev, -1); 928 929 if (bond->dev->flags & IFF_ALLMULTI) 930 dev_set_allmulti(old_active->dev, -1); 931 932 if (bond->dev->flags & IFF_UP) 933 bond_hw_addr_flush(bond->dev, old_active->dev); 934 935 bond_slave_ns_maddrs_add(bond, old_active); 936 } 937 938 if (new_active) { 939 /* FIXME: Signal errors upstream. */ 940 if (bond->dev->flags & IFF_PROMISC) 941 dev_set_promiscuity(new_active->dev, 1); 942 943 if (bond->dev->flags & IFF_ALLMULTI) 944 dev_set_allmulti(new_active->dev, 1); 945 946 if (bond->dev->flags & IFF_UP) { 947 netif_addr_lock_bh(bond->dev); 948 dev_uc_sync(new_active->dev, bond->dev); 949 dev_mc_sync(new_active->dev, bond->dev); 950 netif_addr_unlock_bh(bond->dev); 951 } 952 953 bond_slave_ns_maddrs_del(bond, new_active); 954 } 955 } 956 957 /** 958 * bond_set_dev_addr - clone slave's address to bond 959 * @bond_dev: bond net device 960 * @slave_dev: slave net device 961 * 962 * Should be called with RTNL held. 963 */ 964 static int bond_set_dev_addr(struct net_device *bond_dev, 965 struct net_device *slave_dev) 966 { 967 int err; 968 969 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 970 bond_dev, slave_dev, slave_dev->addr_len); 971 err = netif_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 972 if (err) 973 return err; 974 975 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 976 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 977 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 978 return 0; 979 } 980 981 static struct slave *bond_get_old_active(struct bonding *bond, 982 struct slave *new_active) 983 { 984 struct slave *slave; 985 struct list_head *iter; 986 987 bond_for_each_slave(bond, slave, iter) { 988 if (slave == new_active) 989 continue; 990 991 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 992 return slave; 993 } 994 995 return NULL; 996 } 997 998 /* bond_do_fail_over_mac 999 * 1000 * Perform special MAC address swapping for fail_over_mac settings 1001 * 1002 * Called with RTNL 1003 */ 1004 static void bond_do_fail_over_mac(struct bonding *bond, 1005 struct slave *new_active, 1006 struct slave *old_active) 1007 { 1008 u8 tmp_mac[MAX_ADDR_LEN]; 1009 struct sockaddr_storage ss; 1010 int rv; 1011 1012 switch (bond->params.fail_over_mac) { 1013 case BOND_FOM_ACTIVE: 1014 if (new_active) { 1015 rv = bond_set_dev_addr(bond->dev, new_active->dev); 1016 if (rv) 1017 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 1018 -rv); 1019 } 1020 break; 1021 case BOND_FOM_FOLLOW: 1022 /* if new_active && old_active, swap them 1023 * if just old_active, do nothing (going to no active slave) 1024 * if just new_active, set new_active to bond's MAC 1025 */ 1026 if (!new_active) 1027 return; 1028 1029 if (!old_active) 1030 old_active = bond_get_old_active(bond, new_active); 1031 1032 if (old_active) { 1033 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 1034 new_active->dev->addr_len); 1035 bond_hw_addr_copy(ss.__data, 1036 old_active->dev->dev_addr, 1037 old_active->dev->addr_len); 1038 ss.ss_family = new_active->dev->type; 1039 } else { 1040 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 1041 bond->dev->addr_len); 1042 ss.ss_family = bond->dev->type; 1043 } 1044 1045 rv = dev_set_mac_address(new_active->dev, &ss, NULL); 1046 if (rv) { 1047 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1048 -rv); 1049 goto out; 1050 } 1051 1052 if (!old_active) 1053 goto out; 1054 1055 bond_hw_addr_copy(ss.__data, tmp_mac, 1056 new_active->dev->addr_len); 1057 ss.ss_family = old_active->dev->type; 1058 1059 rv = dev_set_mac_address(old_active->dev, &ss, NULL); 1060 if (rv) 1061 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1062 -rv); 1063 out: 1064 break; 1065 default: 1066 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1067 bond->params.fail_over_mac); 1068 break; 1069 } 1070 1071 } 1072 1073 /** 1074 * bond_choose_primary_or_current - select the primary or high priority slave 1075 * @bond: our bonding struct 1076 * 1077 * - Check if there is a primary link. If the primary link was set and is up, 1078 * go on and do link reselection. 1079 * 1080 * - If primary link is not set or down, find the highest priority link. 1081 * If the highest priority link is not current slave, set it as primary 1082 * link and do link reselection. 1083 */ 1084 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1085 { 1086 struct slave *prim = rtnl_dereference(bond->primary_slave); 1087 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1088 struct slave *slave, *hprio = NULL; 1089 struct list_head *iter; 1090 1091 if (!prim || prim->link != BOND_LINK_UP) { 1092 bond_for_each_slave(bond, slave, iter) { 1093 if (slave->link == BOND_LINK_UP) { 1094 hprio = hprio ?: slave; 1095 if (slave->prio > hprio->prio) 1096 hprio = slave; 1097 } 1098 } 1099 1100 if (hprio && hprio != curr) { 1101 prim = hprio; 1102 goto link_reselect; 1103 } 1104 1105 if (!curr || curr->link != BOND_LINK_UP) 1106 return NULL; 1107 return curr; 1108 } 1109 1110 if (bond->force_primary) { 1111 bond->force_primary = false; 1112 return prim; 1113 } 1114 1115 link_reselect: 1116 if (!curr || curr->link != BOND_LINK_UP) 1117 return prim; 1118 1119 /* At this point, prim and curr are both up */ 1120 switch (bond->params.primary_reselect) { 1121 case BOND_PRI_RESELECT_ALWAYS: 1122 return prim; 1123 case BOND_PRI_RESELECT_BETTER: 1124 if (prim->speed < curr->speed) 1125 return curr; 1126 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1127 return curr; 1128 return prim; 1129 case BOND_PRI_RESELECT_FAILURE: 1130 return curr; 1131 default: 1132 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1133 bond->params.primary_reselect); 1134 return curr; 1135 } 1136 } 1137 1138 /** 1139 * bond_find_best_slave - select the best available slave to be the active one 1140 * @bond: our bonding struct 1141 */ 1142 static struct slave *bond_find_best_slave(struct bonding *bond) 1143 { 1144 struct slave *slave, *bestslave = NULL; 1145 struct list_head *iter; 1146 int mintime = bond->params.updelay; 1147 1148 slave = bond_choose_primary_or_current(bond); 1149 if (slave) 1150 return slave; 1151 1152 bond_for_each_slave(bond, slave, iter) { 1153 if (slave->link == BOND_LINK_UP) 1154 return slave; 1155 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1156 slave->delay < mintime) { 1157 mintime = slave->delay; 1158 bestslave = slave; 1159 } 1160 } 1161 1162 return bestslave; 1163 } 1164 1165 /* must be called in RCU critical section or with RTNL held */ 1166 static bool bond_should_notify_peers(struct bonding *bond) 1167 { 1168 struct bond_up_slave *usable; 1169 struct slave *slave = NULL; 1170 1171 if (!bond->send_peer_notif || 1172 bond->send_peer_notif % 1173 max(1, bond->params.peer_notif_delay) != 0 || 1174 !netif_carrier_ok(bond->dev)) 1175 return false; 1176 1177 /* The send_peer_notif is set by active-backup or 8023ad 1178 * mode, and cleared in bond_close() when changing mode. 1179 * It is safe to only check bond mode here. 1180 */ 1181 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1182 usable = rcu_dereference_rtnl(bond->usable_slaves); 1183 if (!usable || !READ_ONCE(usable->count)) 1184 return false; 1185 } else { 1186 slave = rcu_dereference_rtnl(bond->curr_active_slave); 1187 if (!slave || test_bit(__LINK_STATE_LINKWATCH_PENDING, 1188 &slave->dev->state)) 1189 return false; 1190 } 1191 1192 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1193 slave ? slave->dev->name : "all"); 1194 1195 return true; 1196 } 1197 1198 /* Use this to update send_peer_notif when RTNL may be held in other places. */ 1199 void bond_peer_notify_work_rearm(struct bonding *bond, unsigned long delay) 1200 { 1201 queue_delayed_work(bond->wq, &bond->peer_notify_work, delay); 1202 } 1203 1204 /* Peer notify update handler. Holds only RTNL */ 1205 static void bond_peer_notify_reset(struct bonding *bond) 1206 { 1207 WRITE_ONCE(bond->send_peer_notif, 1208 bond->params.num_peer_notif * 1209 max(1, bond->params.peer_notif_delay)); 1210 } 1211 1212 static void bond_peer_notify_handler(struct work_struct *work) 1213 { 1214 struct bonding *bond = container_of(work, struct bonding, 1215 peer_notify_work.work); 1216 1217 if (!rtnl_trylock()) { 1218 bond_peer_notify_work_rearm(bond, 1); 1219 return; 1220 } 1221 1222 bond_peer_notify_reset(bond); 1223 1224 rtnl_unlock(); 1225 } 1226 1227 /* Peer notify events post. Holds only RTNL */ 1228 static void bond_peer_notify_may_events(struct bonding *bond, bool force) 1229 { 1230 bool notified = false; 1231 1232 if (bond_should_notify_peers(bond)) { 1233 notified = true; 1234 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 1235 } 1236 1237 if (notified || force) 1238 bond->send_peer_notif--; 1239 } 1240 1241 /** 1242 * bond_change_active_slave - change the active slave into the specified one 1243 * @bond: our bonding struct 1244 * @new_active: the new slave to make the active one 1245 * 1246 * Set the new slave to the bond's settings and unset them on the old 1247 * curr_active_slave. 1248 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1249 * 1250 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1251 * because it is apparently the best available slave we have, even though its 1252 * updelay hasn't timed out yet. 1253 * 1254 * Caller must hold RTNL. 1255 */ 1256 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1257 { 1258 struct slave *old_active; 1259 1260 ASSERT_RTNL(); 1261 1262 old_active = rtnl_dereference(bond->curr_active_slave); 1263 1264 if (old_active == new_active) 1265 return; 1266 1267 #ifdef CONFIG_XFRM_OFFLOAD 1268 bond_ipsec_del_sa_all(bond); 1269 #endif /* CONFIG_XFRM_OFFLOAD */ 1270 1271 if (new_active) { 1272 new_active->last_link_up = jiffies; 1273 1274 if (new_active->link == BOND_LINK_BACK) { 1275 if (bond_uses_primary(bond)) { 1276 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1277 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1278 } 1279 1280 new_active->delay = 0; 1281 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1282 BOND_SLAVE_NOTIFY_NOW); 1283 1284 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1285 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1286 1287 if (bond_is_lb(bond)) 1288 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1289 } else { 1290 if (bond_uses_primary(bond)) 1291 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1292 } 1293 } 1294 1295 if (bond_uses_primary(bond)) 1296 bond_hw_addr_swap(bond, new_active, old_active); 1297 1298 if (bond_is_lb(bond)) { 1299 bond_alb_handle_active_change(bond, new_active); 1300 if (old_active) 1301 bond_set_slave_inactive_flags(old_active, 1302 BOND_SLAVE_NOTIFY_NOW); 1303 if (new_active) 1304 bond_set_slave_active_flags(new_active, 1305 BOND_SLAVE_NOTIFY_NOW); 1306 } else { 1307 rcu_assign_pointer(bond->curr_active_slave, new_active); 1308 } 1309 1310 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1311 if (old_active) 1312 bond_set_slave_inactive_flags(old_active, 1313 BOND_SLAVE_NOTIFY_NOW); 1314 1315 if (new_active) { 1316 bond_set_slave_active_flags(new_active, 1317 BOND_SLAVE_NOTIFY_NOW); 1318 1319 if (bond->params.fail_over_mac) 1320 bond_do_fail_over_mac(bond, new_active, 1321 old_active); 1322 1323 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1324 1325 if (netif_running(bond->dev)) { 1326 bond_peer_notify_reset(bond); 1327 bond_peer_notify_may_events(bond, false); 1328 } 1329 } 1330 } 1331 1332 #ifdef CONFIG_XFRM_OFFLOAD 1333 bond_ipsec_add_sa_all(bond); 1334 #endif /* CONFIG_XFRM_OFFLOAD */ 1335 1336 /* resend IGMP joins since active slave has changed or 1337 * all were sent on curr_active_slave. 1338 * resend only if bond is brought up with the affected 1339 * bonding modes and the retransmission is enabled 1340 */ 1341 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1342 ((bond_uses_primary(bond) && new_active) || 1343 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1344 bond->igmp_retrans = bond->params.resend_igmp; 1345 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1346 } 1347 } 1348 1349 /** 1350 * bond_select_active_slave - select a new active slave, if needed 1351 * @bond: our bonding struct 1352 * 1353 * This functions should be called when one of the following occurs: 1354 * - The old curr_active_slave has been released or lost its link. 1355 * - The primary_slave has got its link back. 1356 * - A slave has got its link back and there's no old curr_active_slave. 1357 * 1358 * Caller must hold RTNL. 1359 */ 1360 void bond_select_active_slave(struct bonding *bond) 1361 { 1362 struct slave *best_slave; 1363 int rv; 1364 1365 ASSERT_RTNL(); 1366 1367 best_slave = bond_find_best_slave(bond); 1368 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1369 bond_change_active_slave(bond, best_slave); 1370 rv = bond_set_carrier(bond); 1371 if (!rv) 1372 return; 1373 1374 if (netif_carrier_ok(bond->dev)) 1375 netdev_info(bond->dev, "active interface up!\n"); 1376 else 1377 netdev_info(bond->dev, "now running without any active interface!\n"); 1378 } 1379 } 1380 1381 #ifdef CONFIG_NET_POLL_CONTROLLER 1382 static inline int slave_enable_netpoll(struct slave *slave) 1383 { 1384 struct netpoll *np; 1385 int err = 0; 1386 1387 np = kzalloc(sizeof(*np), GFP_KERNEL); 1388 err = -ENOMEM; 1389 if (!np) 1390 goto out; 1391 1392 err = __netpoll_setup(np, slave->dev); 1393 if (err) { 1394 kfree(np); 1395 goto out; 1396 } 1397 slave->np = np; 1398 out: 1399 return err; 1400 } 1401 static inline void slave_disable_netpoll(struct slave *slave) 1402 { 1403 struct netpoll *np = slave->np; 1404 1405 if (!np) 1406 return; 1407 1408 slave->np = NULL; 1409 1410 __netpoll_free(np); 1411 } 1412 1413 static void bond_poll_controller(struct net_device *bond_dev) 1414 { 1415 struct bonding *bond = netdev_priv(bond_dev); 1416 struct slave *slave = NULL; 1417 struct list_head *iter; 1418 struct ad_info ad_info; 1419 1420 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1421 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1422 return; 1423 1424 bond_for_each_slave_rcu(bond, slave, iter) { 1425 if (!bond_slave_is_up(slave)) 1426 continue; 1427 1428 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1429 struct aggregator *agg = 1430 SLAVE_AD_INFO(slave)->port.aggregator; 1431 1432 if (agg && 1433 agg->aggregator_identifier != ad_info.aggregator_id) 1434 continue; 1435 } 1436 1437 netpoll_poll_dev(slave->dev); 1438 } 1439 } 1440 1441 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1442 { 1443 struct bonding *bond = netdev_priv(bond_dev); 1444 struct list_head *iter; 1445 struct slave *slave; 1446 1447 bond_for_each_slave(bond, slave, iter) 1448 if (bond_slave_is_up(slave)) 1449 slave_disable_netpoll(slave); 1450 } 1451 1452 static int bond_netpoll_setup(struct net_device *dev) 1453 { 1454 struct bonding *bond = netdev_priv(dev); 1455 struct list_head *iter; 1456 struct slave *slave; 1457 int err = 0; 1458 1459 bond_for_each_slave(bond, slave, iter) { 1460 err = slave_enable_netpoll(slave); 1461 if (err) { 1462 bond_netpoll_cleanup(dev); 1463 break; 1464 } 1465 } 1466 return err; 1467 } 1468 #else 1469 static inline int slave_enable_netpoll(struct slave *slave) 1470 { 1471 return 0; 1472 } 1473 static inline void slave_disable_netpoll(struct slave *slave) 1474 { 1475 } 1476 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1477 { 1478 } 1479 #endif 1480 1481 /*---------------------------------- IOCTL ----------------------------------*/ 1482 1483 static netdev_features_t bond_fix_features(struct net_device *dev, 1484 netdev_features_t features) 1485 { 1486 struct bonding *bond = netdev_priv(dev); 1487 struct list_head *iter; 1488 netdev_features_t mask; 1489 struct slave *slave; 1490 1491 mask = features; 1492 features = netdev_base_features(features); 1493 1494 bond_for_each_slave(bond, slave, iter) { 1495 features = netdev_increment_features(features, 1496 slave->dev->features, 1497 mask); 1498 } 1499 features = netdev_add_tso_features(features, mask); 1500 1501 return features; 1502 } 1503 1504 static void bond_setup_by_slave(struct net_device *bond_dev, 1505 struct net_device *slave_dev) 1506 { 1507 bool was_up = !!(bond_dev->flags & IFF_UP); 1508 1509 dev_close(bond_dev); 1510 1511 bond_dev->header_ops = slave_dev->header_ops; 1512 1513 bond_dev->type = slave_dev->type; 1514 bond_dev->hard_header_len = slave_dev->hard_header_len; 1515 bond_dev->needed_headroom = slave_dev->needed_headroom; 1516 bond_dev->addr_len = slave_dev->addr_len; 1517 1518 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1519 slave_dev->addr_len); 1520 1521 if (slave_dev->flags & IFF_POINTOPOINT) { 1522 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1523 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1524 } 1525 if (was_up) 1526 dev_open(bond_dev, NULL); 1527 } 1528 1529 /* On bonding slaves other than the currently active slave, suppress 1530 * duplicates except for alb non-mcast/bcast. 1531 */ 1532 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1533 struct slave *slave, 1534 struct bonding *bond) 1535 { 1536 if (bond_is_slave_inactive(slave)) { 1537 if (BOND_MODE(bond) == BOND_MODE_ALB && 1538 skb->pkt_type != PACKET_BROADCAST && 1539 skb->pkt_type != PACKET_MULTICAST) 1540 return false; 1541 return true; 1542 } 1543 return false; 1544 } 1545 1546 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1547 { 1548 struct sk_buff *skb = *pskb; 1549 struct slave *slave; 1550 struct bonding *bond; 1551 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1552 struct slave *); 1553 int ret = RX_HANDLER_ANOTHER; 1554 1555 skb = skb_share_check(skb, GFP_ATOMIC); 1556 if (unlikely(!skb)) 1557 return RX_HANDLER_CONSUMED; 1558 1559 *pskb = skb; 1560 1561 slave = bond_slave_get_rcu(skb->dev); 1562 bond = slave->bond; 1563 1564 recv_probe = READ_ONCE(bond->recv_probe); 1565 if (recv_probe) { 1566 ret = recv_probe(skb, bond, slave); 1567 if (ret == RX_HANDLER_CONSUMED) { 1568 consume_skb(skb); 1569 return ret; 1570 } 1571 } 1572 1573 /* 1574 * For packets determined by bond_should_deliver_exact_match() call to 1575 * be suppressed we want to make an exception for link-local packets. 1576 * This is necessary for e.g. LLDP daemons to be able to monitor 1577 * inactive slave links without being forced to bind to them 1578 * explicitly. 1579 * 1580 * At the same time, packets that are passed to the bonding master 1581 * (including link-local ones) can have their originating interface 1582 * determined via PACKET_ORIGDEV socket option. 1583 */ 1584 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1585 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1586 return RX_HANDLER_PASS; 1587 return RX_HANDLER_EXACT; 1588 } 1589 1590 skb->dev = bond->dev; 1591 1592 if (BOND_MODE(bond) == BOND_MODE_ALB && 1593 netif_is_bridge_port(bond->dev) && 1594 skb->pkt_type == PACKET_HOST) { 1595 1596 if (unlikely(skb_cow_head(skb, 1597 skb->data - skb_mac_header(skb)))) { 1598 kfree_skb(skb); 1599 return RX_HANDLER_CONSUMED; 1600 } 1601 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1602 bond->dev->addr_len); 1603 } 1604 1605 return ret; 1606 } 1607 1608 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1609 { 1610 switch (BOND_MODE(bond)) { 1611 case BOND_MODE_ROUNDROBIN: 1612 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1613 case BOND_MODE_ACTIVEBACKUP: 1614 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1615 case BOND_MODE_BROADCAST: 1616 return NETDEV_LAG_TX_TYPE_BROADCAST; 1617 case BOND_MODE_XOR: 1618 case BOND_MODE_8023AD: 1619 return NETDEV_LAG_TX_TYPE_HASH; 1620 default: 1621 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1622 } 1623 } 1624 1625 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1626 enum netdev_lag_tx_type type) 1627 { 1628 if (type != NETDEV_LAG_TX_TYPE_HASH) 1629 return NETDEV_LAG_HASH_NONE; 1630 1631 switch (bond->params.xmit_policy) { 1632 case BOND_XMIT_POLICY_LAYER2: 1633 return NETDEV_LAG_HASH_L2; 1634 case BOND_XMIT_POLICY_LAYER34: 1635 return NETDEV_LAG_HASH_L34; 1636 case BOND_XMIT_POLICY_LAYER23: 1637 return NETDEV_LAG_HASH_L23; 1638 case BOND_XMIT_POLICY_ENCAP23: 1639 return NETDEV_LAG_HASH_E23; 1640 case BOND_XMIT_POLICY_ENCAP34: 1641 return NETDEV_LAG_HASH_E34; 1642 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1643 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1644 default: 1645 return NETDEV_LAG_HASH_UNKNOWN; 1646 } 1647 } 1648 1649 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1650 struct netlink_ext_ack *extack) 1651 { 1652 struct netdev_lag_upper_info lag_upper_info; 1653 enum netdev_lag_tx_type type; 1654 int err; 1655 1656 type = bond_lag_tx_type(bond); 1657 lag_upper_info.tx_type = type; 1658 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1659 1660 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1661 &lag_upper_info, extack); 1662 if (err) 1663 return err; 1664 1665 slave->dev->flags |= IFF_SLAVE; 1666 return 0; 1667 } 1668 1669 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1670 { 1671 netdev_upper_dev_unlink(slave->dev, bond->dev); 1672 slave->dev->flags &= ~IFF_SLAVE; 1673 } 1674 1675 static void slave_kobj_release(struct kobject *kobj) 1676 { 1677 struct slave *slave = to_slave(kobj); 1678 struct bonding *bond = bond_get_bond_by_slave(slave); 1679 1680 cancel_delayed_work_sync(&slave->notify_work); 1681 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1682 kfree(SLAVE_AD_INFO(slave)); 1683 1684 kfree(slave); 1685 } 1686 1687 static struct kobj_type slave_ktype = { 1688 .release = slave_kobj_release, 1689 #ifdef CONFIG_SYSFS 1690 .sysfs_ops = &slave_sysfs_ops, 1691 #endif 1692 }; 1693 1694 static int bond_kobj_init(struct slave *slave) 1695 { 1696 int err; 1697 1698 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1699 &(slave->dev->dev.kobj), "bonding_slave"); 1700 if (err) 1701 kobject_put(&slave->kobj); 1702 1703 return err; 1704 } 1705 1706 static struct slave *bond_alloc_slave(struct bonding *bond, 1707 struct net_device *slave_dev) 1708 { 1709 struct slave *slave = NULL; 1710 1711 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1712 if (!slave) 1713 return NULL; 1714 1715 slave->bond = bond; 1716 slave->dev = slave_dev; 1717 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1718 1719 if (bond_kobj_init(slave)) 1720 return NULL; 1721 1722 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1723 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1724 GFP_KERNEL); 1725 if (!SLAVE_AD_INFO(slave)) { 1726 kobject_put(&slave->kobj); 1727 return NULL; 1728 } 1729 } 1730 1731 return slave; 1732 } 1733 1734 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1735 { 1736 info->bond_mode = BOND_MODE(bond); 1737 info->miimon = bond->params.miimon; 1738 info->num_slaves = bond->slave_cnt; 1739 } 1740 1741 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1742 { 1743 strcpy(info->slave_name, slave->dev->name); 1744 info->link = slave->link; 1745 info->state = bond_slave_state(slave); 1746 info->link_failure_count = slave->link_failure_count; 1747 } 1748 1749 static void bond_netdev_notify_work(struct work_struct *_work) 1750 { 1751 struct slave *slave = container_of(_work, struct slave, 1752 notify_work.work); 1753 1754 if (rtnl_trylock()) { 1755 struct netdev_bonding_info binfo; 1756 1757 bond_fill_ifslave(slave, &binfo.slave); 1758 bond_fill_ifbond(slave->bond, &binfo.master); 1759 netdev_bonding_info_change(slave->dev, &binfo); 1760 rtnl_unlock(); 1761 } else { 1762 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1763 } 1764 } 1765 1766 void bond_queue_slave_event(struct slave *slave) 1767 { 1768 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1769 } 1770 1771 void bond_lower_state_changed(struct slave *slave) 1772 { 1773 struct netdev_lag_lower_state_info info; 1774 1775 info.link_up = slave->link == BOND_LINK_UP || 1776 slave->link == BOND_LINK_FAIL; 1777 info.tx_enabled = bond_is_active_slave(slave); 1778 netdev_lower_state_changed(slave->dev, &info); 1779 } 1780 1781 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1782 if (extack) \ 1783 NL_SET_ERR_MSG(extack, errmsg); \ 1784 else \ 1785 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1786 } while (0) 1787 1788 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1789 if (extack) \ 1790 NL_SET_ERR_MSG(extack, errmsg); \ 1791 else \ 1792 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1793 } while (0) 1794 1795 /* The bonding driver uses ether_setup() to convert a master bond device 1796 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always 1797 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP 1798 * if they were set 1799 */ 1800 static void bond_ether_setup(struct net_device *bond_dev) 1801 { 1802 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP); 1803 1804 ether_setup(bond_dev); 1805 bond_dev->flags |= IFF_MASTER | flags; 1806 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1807 } 1808 1809 void bond_xdp_set_features(struct net_device *bond_dev) 1810 { 1811 struct bonding *bond = netdev_priv(bond_dev); 1812 xdp_features_t val = NETDEV_XDP_ACT_MASK; 1813 struct list_head *iter; 1814 struct slave *slave; 1815 1816 ASSERT_RTNL(); 1817 1818 if (!bond_xdp_check(bond, BOND_MODE(bond)) || !bond_has_slaves(bond)) { 1819 xdp_clear_features_flag(bond_dev); 1820 return; 1821 } 1822 1823 bond_for_each_slave(bond, slave, iter) 1824 val &= slave->dev->xdp_features; 1825 1826 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY; 1827 1828 xdp_set_features_flag(bond_dev, val); 1829 } 1830 1831 /* enslave device <slave> to bond device <master> */ 1832 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1833 struct netlink_ext_ack *extack) 1834 { 1835 struct bonding *bond = netdev_priv(bond_dev); 1836 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1837 struct slave *new_slave = NULL, *prev_slave; 1838 struct sockaddr_storage ss; 1839 int res = 0, i; 1840 1841 if (slave_dev->flags & IFF_MASTER && 1842 !netif_is_bond_master(slave_dev)) { 1843 BOND_NL_ERR(bond_dev, extack, 1844 "Device type (master device) cannot be enslaved"); 1845 return -EPERM; 1846 } 1847 1848 /* already in-use? */ 1849 if (netdev_is_rx_handler_busy(slave_dev)) { 1850 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1851 "Device is in use and cannot be enslaved"); 1852 return -EBUSY; 1853 } 1854 1855 if (bond_dev == slave_dev) { 1856 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1857 return -EPERM; 1858 } 1859 1860 /* vlan challenged mutual exclusion */ 1861 /* no need to lock since we're protected by rtnl_lock */ 1862 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1863 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1864 if (vlan_uses_dev(bond_dev)) { 1865 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1866 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1867 return -EPERM; 1868 } else { 1869 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1870 } 1871 } else { 1872 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1873 } 1874 1875 if (slave_dev->features & NETIF_F_HW_ESP) 1876 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1877 1878 /* Old ifenslave binaries are no longer supported. These can 1879 * be identified with moderate accuracy by the state of the slave: 1880 * the current ifenslave will set the interface down prior to 1881 * enslaving it; the old ifenslave will not. 1882 */ 1883 if (slave_dev->flags & IFF_UP) { 1884 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1885 "Device can not be enslaved while up"); 1886 return -EPERM; 1887 } 1888 1889 /* set bonding device ether type by slave - bonding netdevices are 1890 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1891 * there is a need to override some of the type dependent attribs/funcs. 1892 * 1893 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1894 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1895 */ 1896 if (!bond_has_slaves(bond)) { 1897 if (bond_dev->type != slave_dev->type) { 1898 if (slave_dev->type != ARPHRD_ETHER && 1899 BOND_MODE(bond) == BOND_MODE_8023AD) { 1900 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1901 "8023AD mode requires Ethernet devices"); 1902 return -EINVAL; 1903 } 1904 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1905 bond_dev->type, slave_dev->type); 1906 1907 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1908 bond_dev); 1909 res = notifier_to_errno(res); 1910 if (res) { 1911 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1912 return -EBUSY; 1913 } 1914 1915 /* Flush unicast and multicast addresses */ 1916 dev_uc_flush(bond_dev); 1917 dev_mc_flush(bond_dev); 1918 1919 if (slave_dev->type != ARPHRD_ETHER) 1920 bond_setup_by_slave(bond_dev, slave_dev); 1921 else 1922 bond_ether_setup(bond_dev); 1923 1924 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1925 bond_dev); 1926 } 1927 } else if (bond_dev->type != slave_dev->type) { 1928 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1929 "Device type is different from other slaves"); 1930 return -EINVAL; 1931 } 1932 1933 if (slave_dev->type == ARPHRD_INFINIBAND && 1934 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1935 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1936 "Only active-backup mode is supported for infiniband slaves"); 1937 res = -EOPNOTSUPP; 1938 goto err_undo_flags; 1939 } 1940 1941 if (!slave_ops->ndo_set_mac_address || 1942 slave_dev->type == ARPHRD_INFINIBAND) { 1943 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1944 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1945 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1946 if (!bond_has_slaves(bond)) { 1947 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1948 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1949 } else { 1950 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1951 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1952 res = -EOPNOTSUPP; 1953 goto err_undo_flags; 1954 } 1955 } 1956 } 1957 1958 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 1959 1960 /* If this is the first slave, then we need to set the master's hardware 1961 * address to be the same as the slave's. 1962 */ 1963 if (!bond_has_slaves(bond) && 1964 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 1965 res = bond_set_dev_addr(bond->dev, slave_dev); 1966 if (res) 1967 goto err_undo_flags; 1968 } 1969 1970 new_slave = bond_alloc_slave(bond, slave_dev); 1971 if (!new_slave) { 1972 res = -ENOMEM; 1973 goto err_undo_flags; 1974 } 1975 1976 /* Set the new_slave's queue_id to be zero. Queue ID mapping 1977 * is set via sysfs or module option if desired. 1978 */ 1979 new_slave->queue_id = 0; 1980 1981 /* Save slave's original mtu and then set it to match the bond */ 1982 new_slave->original_mtu = slave_dev->mtu; 1983 res = dev_set_mtu(slave_dev, bond->dev->mtu); 1984 if (res) { 1985 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 1986 goto err_free; 1987 } 1988 1989 /* Save slave's original ("permanent") mac address for modes 1990 * that need it, and for restoring it upon release, and then 1991 * set it to the master's address 1992 */ 1993 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 1994 slave_dev->addr_len); 1995 1996 if (!bond->params.fail_over_mac || 1997 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1998 /* Set slave to master's mac address. The application already 1999 * set the master's mac address to that of the first slave 2000 */ 2001 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 2002 } else if (bond->params.fail_over_mac == BOND_FOM_FOLLOW && 2003 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 2004 bond_has_slaves(bond) && 2005 memcmp(slave_dev->dev_addr, bond_dev->dev_addr, bond_dev->addr_len) == 0) { 2006 /* Set slave to random address to avoid duplicate mac 2007 * address in later fail over. 2008 */ 2009 eth_random_addr(ss.__data); 2010 } else { 2011 goto skip_mac_set; 2012 } 2013 2014 ss.ss_family = slave_dev->type; 2015 res = dev_set_mac_address(slave_dev, &ss, extack); 2016 if (res) { 2017 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 2018 goto err_restore_mtu; 2019 } 2020 2021 skip_mac_set: 2022 2023 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 2024 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 2025 2026 /* open the slave since the application closed it */ 2027 res = dev_open(slave_dev, extack); 2028 if (res) { 2029 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 2030 goto err_restore_mac; 2031 } 2032 2033 slave_dev->priv_flags |= IFF_BONDING; 2034 /* initialize slave stats */ 2035 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 2036 2037 if (bond_is_lb(bond)) { 2038 /* bond_alb_init_slave() must be called before all other stages since 2039 * it might fail and we do not want to have to undo everything 2040 */ 2041 res = bond_alb_init_slave(bond, new_slave); 2042 if (res) 2043 goto err_close; 2044 } 2045 2046 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 2047 if (res) { 2048 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 2049 goto err_close; 2050 } 2051 2052 prev_slave = bond_last_slave(bond); 2053 2054 new_slave->delay = 0; 2055 new_slave->link_failure_count = 0; 2056 2057 if (bond_update_speed_duplex(new_slave) && 2058 bond_needs_speed_duplex(bond)) 2059 new_slave->link = BOND_LINK_DOWN; 2060 2061 new_slave->last_rx = jiffies - 2062 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2063 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2064 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2065 2066 new_slave->last_tx = new_slave->last_rx; 2067 2068 /* check for initial state */ 2069 new_slave->link = BOND_LINK_NOCHANGE; 2070 if (bond->params.miimon) { 2071 if (netif_running(slave_dev) && netif_carrier_ok(slave_dev)) { 2072 if (bond->params.updelay) { 2073 bond_set_slave_link_state(new_slave, 2074 BOND_LINK_BACK, 2075 BOND_SLAVE_NOTIFY_NOW); 2076 new_slave->delay = bond->params.updelay; 2077 } else { 2078 bond_set_slave_link_state(new_slave, 2079 BOND_LINK_UP, 2080 BOND_SLAVE_NOTIFY_NOW); 2081 } 2082 } else { 2083 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2084 BOND_SLAVE_NOTIFY_NOW); 2085 } 2086 } else if (bond->params.arp_interval) { 2087 bond_set_slave_link_state(new_slave, 2088 (netif_carrier_ok(slave_dev) ? 2089 BOND_LINK_UP : BOND_LINK_DOWN), 2090 BOND_SLAVE_NOTIFY_NOW); 2091 } else { 2092 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2093 BOND_SLAVE_NOTIFY_NOW); 2094 } 2095 2096 if (new_slave->link != BOND_LINK_DOWN) 2097 new_slave->last_link_up = jiffies; 2098 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2099 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2100 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2101 2102 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2103 /* if there is a primary slave, remember it */ 2104 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2105 rcu_assign_pointer(bond->primary_slave, new_slave); 2106 bond->force_primary = true; 2107 } 2108 } 2109 2110 switch (BOND_MODE(bond)) { 2111 case BOND_MODE_ACTIVEBACKUP: 2112 bond_set_slave_inactive_flags(new_slave, 2113 BOND_SLAVE_NOTIFY_NOW); 2114 break; 2115 case BOND_MODE_8023AD: 2116 /* in 802.3ad mode, the internal mechanism 2117 * will activate the slaves in the selected 2118 * aggregator 2119 */ 2120 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2121 /* if this is the first slave */ 2122 if (!prev_slave) { 2123 SLAVE_AD_INFO(new_slave)->id = 1; 2124 /* Initialize AD with the number of times that the AD timer is called in 1 second 2125 * can be called only after the mac address of the bond is set 2126 */ 2127 bond_3ad_initialize(bond); 2128 } else { 2129 SLAVE_AD_INFO(new_slave)->id = 2130 SLAVE_AD_INFO(prev_slave)->id + 1; 2131 } 2132 2133 bond_3ad_bind_slave(new_slave); 2134 break; 2135 case BOND_MODE_TLB: 2136 case BOND_MODE_ALB: 2137 bond_set_active_slave(new_slave); 2138 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2139 break; 2140 default: 2141 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2142 2143 /* always active in trunk mode */ 2144 bond_set_active_slave(new_slave); 2145 2146 /* In trunking mode there is little meaning to curr_active_slave 2147 * anyway (it holds no special properties of the bond device), 2148 * so we can change it without calling change_active_interface() 2149 */ 2150 if (!rcu_access_pointer(bond->curr_active_slave) && 2151 new_slave->link == BOND_LINK_UP) 2152 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2153 2154 break; 2155 } /* switch(bond_mode) */ 2156 2157 #ifdef CONFIG_NET_POLL_CONTROLLER 2158 if (bond->dev->npinfo) { 2159 if (slave_enable_netpoll(new_slave)) { 2160 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2161 res = -EBUSY; 2162 goto err_detach; 2163 } 2164 } 2165 #endif 2166 2167 if (!(bond_dev->features & NETIF_F_LRO)) 2168 dev_disable_lro(slave_dev); 2169 2170 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2171 new_slave); 2172 if (res) { 2173 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2174 goto err_detach; 2175 } 2176 2177 res = bond_master_upper_dev_link(bond, new_slave, extack); 2178 if (res) { 2179 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2180 goto err_unregister; 2181 } 2182 2183 bond_lower_state_changed(new_slave); 2184 2185 res = bond_sysfs_slave_add(new_slave); 2186 if (res) { 2187 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2188 goto err_upper_unlink; 2189 } 2190 2191 /* If the mode uses primary, then the following is handled by 2192 * bond_change_active_slave(). 2193 */ 2194 if (!bond_uses_primary(bond)) { 2195 /* set promiscuity level to new slave */ 2196 if (bond_dev->flags & IFF_PROMISC) { 2197 res = dev_set_promiscuity(slave_dev, 1); 2198 if (res) 2199 goto err_sysfs_del; 2200 } 2201 2202 /* set allmulti level to new slave */ 2203 if (bond_dev->flags & IFF_ALLMULTI) { 2204 res = dev_set_allmulti(slave_dev, 1); 2205 if (res) { 2206 if (bond_dev->flags & IFF_PROMISC) 2207 dev_set_promiscuity(slave_dev, -1); 2208 goto err_sysfs_del; 2209 } 2210 } 2211 2212 if (bond_dev->flags & IFF_UP) { 2213 netif_addr_lock_bh(bond_dev); 2214 dev_mc_sync_multiple(slave_dev, bond_dev); 2215 dev_uc_sync_multiple(slave_dev, bond_dev); 2216 netif_addr_unlock_bh(bond_dev); 2217 2218 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2219 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2220 } 2221 } 2222 2223 bond->slave_cnt++; 2224 netdev_compute_master_upper_features(bond->dev, true); 2225 bond_set_carrier(bond); 2226 2227 /* Needs to be called before bond_select_active_slave(), which will 2228 * remove the maddrs if the slave is selected as active slave. 2229 */ 2230 bond_slave_ns_maddrs_add(bond, new_slave); 2231 2232 if (bond_uses_primary(bond)) { 2233 block_netpoll_tx(); 2234 bond_select_active_slave(bond); 2235 unblock_netpoll_tx(); 2236 } 2237 2238 if (!slave_dev->netdev_ops->ndo_bpf || 2239 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2240 if (bond->xdp_prog) { 2241 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2242 "Slave does not support XDP"); 2243 res = -EOPNOTSUPP; 2244 goto err_sysfs_del; 2245 } 2246 } else if (bond->xdp_prog) { 2247 struct netdev_bpf xdp = { 2248 .command = XDP_SETUP_PROG, 2249 .flags = 0, 2250 .prog = bond->xdp_prog, 2251 .extack = extack, 2252 }; 2253 2254 if (dev_xdp_prog_count(slave_dev) > 0) { 2255 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2256 "Slave has XDP program loaded, please unload before enslaving"); 2257 res = -EOPNOTSUPP; 2258 goto err_sysfs_del; 2259 } 2260 2261 res = dev_xdp_propagate(slave_dev, &xdp); 2262 if (res < 0) { 2263 /* ndo_bpf() sets extack error message */ 2264 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2265 goto err_sysfs_del; 2266 } 2267 if (bond->xdp_prog) 2268 bpf_prog_inc(bond->xdp_prog); 2269 } 2270 2271 /* broadcast mode uses the all_slaves to loop through slaves. */ 2272 if (bond_mode_can_use_xmit_hash(bond) || 2273 BOND_MODE(bond) == BOND_MODE_BROADCAST) 2274 bond_update_slave_arr(bond, NULL); 2275 2276 bond_xdp_set_features(bond_dev); 2277 2278 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2279 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2280 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2281 2282 /* enslave is successful */ 2283 bond_queue_slave_event(new_slave); 2284 return 0; 2285 2286 /* Undo stages on error */ 2287 err_sysfs_del: 2288 bond_sysfs_slave_del(new_slave); 2289 2290 err_upper_unlink: 2291 bond_upper_dev_unlink(bond, new_slave); 2292 2293 err_unregister: 2294 netdev_rx_handler_unregister(slave_dev); 2295 2296 err_detach: 2297 vlan_vids_del_by_dev(slave_dev, bond_dev); 2298 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2299 RCU_INIT_POINTER(bond->primary_slave, NULL); 2300 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2301 block_netpoll_tx(); 2302 bond_change_active_slave(bond, NULL); 2303 bond_select_active_slave(bond); 2304 unblock_netpoll_tx(); 2305 } 2306 /* either primary_slave or curr_active_slave might've changed */ 2307 synchronize_rcu(); 2308 slave_disable_netpoll(new_slave); 2309 2310 err_close: 2311 if (!netif_is_bond_master(slave_dev)) 2312 slave_dev->priv_flags &= ~IFF_BONDING; 2313 dev_close(slave_dev); 2314 2315 err_restore_mac: 2316 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2317 if (!bond->params.fail_over_mac || 2318 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2319 /* XXX TODO - fom follow mode needs to change master's 2320 * MAC if this slave's MAC is in use by the bond, or at 2321 * least print a warning. 2322 */ 2323 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2324 new_slave->dev->addr_len); 2325 ss.ss_family = slave_dev->type; 2326 dev_set_mac_address(slave_dev, &ss, NULL); 2327 } 2328 2329 err_restore_mtu: 2330 dev_set_mtu(slave_dev, new_slave->original_mtu); 2331 2332 err_free: 2333 kobject_put(&new_slave->kobj); 2334 2335 err_undo_flags: 2336 /* Enslave of first slave has failed and we need to fix master's mac */ 2337 if (!bond_has_slaves(bond)) { 2338 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2339 slave_dev->dev_addr)) 2340 eth_hw_addr_random(bond_dev); 2341 if (bond_dev->type != ARPHRD_ETHER) { 2342 dev_close(bond_dev); 2343 bond_ether_setup(bond_dev); 2344 } 2345 } 2346 2347 return res; 2348 } 2349 2350 /* Try to release the slave device <slave> from the bond device <master> 2351 * It is legal to access curr_active_slave without a lock because all the function 2352 * is RTNL-locked. If "all" is true it means that the function is being called 2353 * while destroying a bond interface and all slaves are being released. 2354 * 2355 * The rules for slave state should be: 2356 * for Active/Backup: 2357 * Active stays on all backups go down 2358 * for Bonded connections: 2359 * The first up interface should be left on and all others downed. 2360 */ 2361 static int __bond_release_one(struct net_device *bond_dev, 2362 struct net_device *slave_dev, 2363 bool all, bool unregister) 2364 { 2365 struct bonding *bond = netdev_priv(bond_dev); 2366 struct slave *slave, *oldcurrent; 2367 struct sockaddr_storage ss; 2368 int old_flags = bond_dev->flags; 2369 netdev_features_t old_features = bond_dev->features; 2370 2371 /* slave is not a slave or master is not master of this slave */ 2372 if (!(slave_dev->flags & IFF_SLAVE) || 2373 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2374 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2375 return -EINVAL; 2376 } 2377 2378 block_netpoll_tx(); 2379 2380 slave = bond_get_slave_by_dev(bond, slave_dev); 2381 if (!slave) { 2382 /* not a slave of this bond */ 2383 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2384 unblock_netpoll_tx(); 2385 return -EINVAL; 2386 } 2387 2388 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2389 2390 bond_sysfs_slave_del(slave); 2391 2392 /* recompute stats just before removing the slave */ 2393 bond_get_stats(bond->dev, &bond->bond_stats); 2394 2395 if (bond->xdp_prog) { 2396 struct netdev_bpf xdp = { 2397 .command = XDP_SETUP_PROG, 2398 .flags = 0, 2399 .prog = NULL, 2400 .extack = NULL, 2401 }; 2402 if (dev_xdp_propagate(slave_dev, &xdp)) 2403 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2404 } 2405 2406 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2407 * for this slave anymore. 2408 */ 2409 netdev_rx_handler_unregister(slave_dev); 2410 2411 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2412 bond_3ad_unbind_slave(slave); 2413 2414 bond_upper_dev_unlink(bond, slave); 2415 2416 if (bond_mode_can_use_xmit_hash(bond) || 2417 BOND_MODE(bond) == BOND_MODE_BROADCAST) 2418 bond_update_slave_arr(bond, slave); 2419 2420 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2421 bond_is_active_slave(slave) ? "active" : "backup"); 2422 2423 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2424 2425 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2426 2427 if (!all && (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2428 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2429 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2430 bond_has_slaves(bond)) 2431 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", 2432 slave->perm_hwaddr); 2433 } 2434 2435 if (rtnl_dereference(bond->primary_slave) == slave) 2436 RCU_INIT_POINTER(bond->primary_slave, NULL); 2437 2438 if (oldcurrent == slave) 2439 bond_change_active_slave(bond, NULL); 2440 2441 /* Must be called after bond_change_active_slave () as the slave 2442 * might change from an active slave to a backup slave. Then it is 2443 * necessary to clear the maddrs on the backup slave. 2444 */ 2445 bond_slave_ns_maddrs_del(bond, slave); 2446 2447 if (bond_is_lb(bond)) { 2448 /* Must be called only after the slave has been 2449 * detached from the list and the curr_active_slave 2450 * has been cleared (if our_slave == old_current), 2451 * but before a new active slave is selected. 2452 */ 2453 bond_alb_deinit_slave(bond, slave); 2454 } 2455 2456 if (all) { 2457 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2458 } else if (oldcurrent == slave) { 2459 /* Note that we hold RTNL over this sequence, so there 2460 * is no concern that another slave add/remove event 2461 * will interfere. 2462 */ 2463 bond_select_active_slave(bond); 2464 } 2465 2466 bond_set_carrier(bond); 2467 if (!bond_has_slaves(bond)) 2468 eth_hw_addr_random(bond_dev); 2469 2470 unblock_netpoll_tx(); 2471 synchronize_rcu(); 2472 bond->slave_cnt--; 2473 2474 if (!bond_has_slaves(bond)) { 2475 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2476 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2477 } 2478 2479 netdev_compute_master_upper_features(bond->dev, true); 2480 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2481 (old_features & NETIF_F_VLAN_CHALLENGED)) 2482 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2483 2484 vlan_vids_del_by_dev(slave_dev, bond_dev); 2485 2486 /* If the mode uses primary, then this case was handled above by 2487 * bond_change_active_slave(..., NULL) 2488 */ 2489 if (!bond_uses_primary(bond)) { 2490 /* unset promiscuity level from slave 2491 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2492 * of the IFF_PROMISC flag in the bond_dev, but we need the 2493 * value of that flag before that change, as that was the value 2494 * when this slave was attached, so we cache at the start of the 2495 * function and use it here. Same goes for ALLMULTI below 2496 */ 2497 if (old_flags & IFF_PROMISC) 2498 dev_set_promiscuity(slave_dev, -1); 2499 2500 /* unset allmulti level from slave */ 2501 if (old_flags & IFF_ALLMULTI) 2502 dev_set_allmulti(slave_dev, -1); 2503 2504 if (old_flags & IFF_UP) 2505 bond_hw_addr_flush(bond_dev, slave_dev); 2506 } 2507 2508 slave_disable_netpoll(slave); 2509 2510 /* close slave before restoring its mac address */ 2511 dev_close(slave_dev); 2512 2513 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2514 2515 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2516 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2517 /* restore original ("permanent") mac address */ 2518 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2519 slave->dev->addr_len); 2520 ss.ss_family = slave_dev->type; 2521 dev_set_mac_address(slave_dev, &ss, NULL); 2522 } 2523 2524 if (unregister) { 2525 netdev_lock_ops(slave_dev); 2526 __netif_set_mtu(slave_dev, slave->original_mtu); 2527 netdev_unlock_ops(slave_dev); 2528 } else { 2529 dev_set_mtu(slave_dev, slave->original_mtu); 2530 } 2531 2532 if (!netif_is_bond_master(slave_dev)) 2533 slave_dev->priv_flags &= ~IFF_BONDING; 2534 2535 bond_xdp_set_features(bond_dev); 2536 kobject_put(&slave->kobj); 2537 2538 return 0; 2539 } 2540 2541 /* A wrapper used because of ndo_del_link */ 2542 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2543 { 2544 return __bond_release_one(bond_dev, slave_dev, false, false); 2545 } 2546 2547 /* First release a slave and then destroy the bond if no more slaves are left. 2548 * Must be under rtnl_lock when this function is called. 2549 */ 2550 static int bond_release_and_destroy(struct net_device *bond_dev, 2551 struct net_device *slave_dev) 2552 { 2553 struct bonding *bond = netdev_priv(bond_dev); 2554 int ret; 2555 2556 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2557 if (ret == 0 && !bond_has_slaves(bond) && 2558 bond_dev->reg_state != NETREG_UNREGISTERING) { 2559 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2560 netdev_info(bond_dev, "Destroying bond\n"); 2561 bond_remove_proc_entry(bond); 2562 unregister_netdevice(bond_dev); 2563 } 2564 return ret; 2565 } 2566 2567 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2568 { 2569 struct bonding *bond = netdev_priv(bond_dev); 2570 2571 bond_fill_ifbond(bond, info); 2572 } 2573 2574 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2575 { 2576 struct bonding *bond = netdev_priv(bond_dev); 2577 struct list_head *iter; 2578 int i = 0, res = -ENODEV; 2579 struct slave *slave; 2580 2581 bond_for_each_slave(bond, slave, iter) { 2582 if (i++ == (int)info->slave_id) { 2583 res = 0; 2584 bond_fill_ifslave(slave, info); 2585 break; 2586 } 2587 } 2588 2589 return res; 2590 } 2591 2592 /*-------------------------------- Monitoring -------------------------------*/ 2593 2594 /* called with rcu_read_lock() */ 2595 static int bond_miimon_inspect(struct bonding *bond) 2596 { 2597 bool ignore_updelay = false; 2598 int link_state, commit = 0; 2599 struct list_head *iter; 2600 struct slave *slave; 2601 2602 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2603 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2604 } else { 2605 struct bond_up_slave *usable_slaves; 2606 2607 usable_slaves = rcu_dereference(bond->usable_slaves); 2608 2609 if (usable_slaves && usable_slaves->count == 0) 2610 ignore_updelay = true; 2611 } 2612 2613 bond_for_each_slave_rcu(bond, slave, iter) { 2614 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2615 2616 link_state = netif_running(slave->dev) && 2617 netif_carrier_ok(slave->dev); 2618 2619 switch (slave->link) { 2620 case BOND_LINK_UP: 2621 if (link_state) 2622 continue; 2623 2624 bond_propose_link_state(slave, BOND_LINK_FAIL); 2625 commit++; 2626 slave->delay = bond->params.downdelay; 2627 if (slave->delay && net_ratelimit()) { 2628 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2629 (BOND_MODE(bond) == 2630 BOND_MODE_ACTIVEBACKUP) ? 2631 (bond_is_active_slave(slave) ? 2632 "active " : "backup ") : "", 2633 bond->params.downdelay * bond->params.miimon); 2634 } 2635 fallthrough; 2636 case BOND_LINK_FAIL: 2637 if (link_state) { 2638 /* recovered before downdelay expired */ 2639 bond_propose_link_state(slave, BOND_LINK_UP); 2640 slave->last_link_up = jiffies; 2641 if (net_ratelimit()) 2642 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2643 (bond->params.downdelay - slave->delay) * 2644 bond->params.miimon); 2645 commit++; 2646 continue; 2647 } 2648 2649 if (slave->delay <= 0) { 2650 bond_propose_link_state(slave, BOND_LINK_DOWN); 2651 commit++; 2652 continue; 2653 } 2654 2655 slave->delay--; 2656 break; 2657 2658 case BOND_LINK_DOWN: 2659 if (!link_state) 2660 continue; 2661 2662 bond_propose_link_state(slave, BOND_LINK_BACK); 2663 commit++; 2664 slave->delay = bond->params.updelay; 2665 2666 if (slave->delay && net_ratelimit()) { 2667 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2668 ignore_updelay ? 0 : 2669 bond->params.updelay * 2670 bond->params.miimon); 2671 } 2672 fallthrough; 2673 case BOND_LINK_BACK: 2674 if (!link_state) { 2675 bond_propose_link_state(slave, BOND_LINK_DOWN); 2676 if (net_ratelimit()) 2677 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2678 (bond->params.updelay - slave->delay) * 2679 bond->params.miimon); 2680 commit++; 2681 continue; 2682 } 2683 2684 if (ignore_updelay) 2685 slave->delay = 0; 2686 2687 if (slave->delay <= 0) { 2688 bond_propose_link_state(slave, BOND_LINK_UP); 2689 commit++; 2690 ignore_updelay = false; 2691 continue; 2692 } 2693 2694 slave->delay--; 2695 break; 2696 } 2697 } 2698 2699 return commit; 2700 } 2701 2702 static void bond_miimon_link_change(struct bonding *bond, 2703 struct slave *slave, 2704 char link) 2705 { 2706 switch (BOND_MODE(bond)) { 2707 case BOND_MODE_8023AD: 2708 bond_3ad_handle_link_change(slave, link); 2709 break; 2710 case BOND_MODE_TLB: 2711 case BOND_MODE_ALB: 2712 bond_alb_handle_link_change(bond, slave, link); 2713 break; 2714 case BOND_MODE_XOR: 2715 bond_update_slave_arr(bond, NULL); 2716 break; 2717 } 2718 } 2719 2720 static void bond_miimon_commit(struct bonding *bond) 2721 { 2722 struct slave *slave, *primary, *active; 2723 bool do_failover = false; 2724 struct list_head *iter; 2725 2726 ASSERT_RTNL(); 2727 2728 bond_for_each_slave(bond, slave, iter) { 2729 switch (slave->link_new_state) { 2730 case BOND_LINK_NOCHANGE: 2731 /* For 802.3ad mode, check current slave speed and 2732 * duplex again in case its port was disabled after 2733 * invalid speed/duplex reporting but recovered before 2734 * link monitoring could make a decision on the actual 2735 * link status 2736 */ 2737 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2738 slave->link == BOND_LINK_UP) 2739 bond_3ad_adapter_speed_duplex_changed(slave); 2740 continue; 2741 2742 case BOND_LINK_UP: 2743 if (bond_update_speed_duplex(slave) && 2744 bond_needs_speed_duplex(bond)) { 2745 slave->link = BOND_LINK_DOWN; 2746 if (net_ratelimit()) 2747 slave_warn(bond->dev, slave->dev, 2748 "failed to get link speed/duplex\n"); 2749 continue; 2750 } 2751 bond_set_slave_link_state(slave, BOND_LINK_UP, 2752 BOND_SLAVE_NOTIFY_NOW); 2753 slave->last_link_up = jiffies; 2754 2755 primary = rtnl_dereference(bond->primary_slave); 2756 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2757 /* prevent it from being the active one */ 2758 bond_set_backup_slave(slave); 2759 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2760 /* make it immediately active */ 2761 bond_set_active_slave(slave); 2762 } 2763 2764 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2765 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2766 slave->duplex ? "full" : "half"); 2767 2768 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2769 2770 active = rtnl_dereference(bond->curr_active_slave); 2771 if (!active || slave == primary || slave->prio > active->prio) 2772 do_failover = true; 2773 2774 continue; 2775 2776 case BOND_LINK_DOWN: 2777 if (slave->link_failure_count < UINT_MAX) 2778 slave->link_failure_count++; 2779 2780 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2781 BOND_SLAVE_NOTIFY_NOW); 2782 2783 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2784 BOND_MODE(bond) == BOND_MODE_8023AD) 2785 bond_set_slave_inactive_flags(slave, 2786 BOND_SLAVE_NOTIFY_NOW); 2787 2788 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2789 2790 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2791 2792 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2793 do_failover = true; 2794 2795 continue; 2796 2797 default: 2798 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2799 slave->link_new_state); 2800 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2801 2802 continue; 2803 } 2804 } 2805 2806 if (do_failover) { 2807 block_netpoll_tx(); 2808 bond_select_active_slave(bond); 2809 unblock_netpoll_tx(); 2810 } 2811 2812 bond_set_carrier(bond); 2813 } 2814 2815 /* bond_mii_monitor 2816 * 2817 * Really a wrapper that splits the mii monitor into two phases: an 2818 * inspection, then (if inspection indicates something needs to be done) 2819 * an acquisition of appropriate locks followed by a commit phase to 2820 * implement whatever link state changes are indicated. 2821 */ 2822 static void bond_mii_monitor(struct work_struct *work) 2823 { 2824 struct bonding *bond = container_of(work, struct bonding, 2825 mii_work.work); 2826 struct list_head *iter; 2827 struct slave *slave; 2828 unsigned long delay; 2829 bool commit; 2830 2831 delay = msecs_to_jiffies(bond->params.miimon); 2832 2833 if (!bond_has_slaves(bond)) 2834 goto re_arm; 2835 2836 rcu_read_lock(); 2837 2838 commit = !!bond_miimon_inspect(bond); 2839 2840 rcu_read_unlock(); 2841 2842 if (commit || READ_ONCE(bond->send_peer_notif)) { 2843 /* Race avoidance with bond_close cancel of workqueue */ 2844 if (!rtnl_trylock()) { 2845 delay = 1; 2846 goto re_arm; 2847 } 2848 2849 if (commit) { 2850 bond_for_each_slave(bond, slave, iter) { 2851 bond_commit_link_state(slave, 2852 BOND_SLAVE_NOTIFY_LATER); 2853 } 2854 bond_miimon_commit(bond); 2855 } 2856 2857 if (bond->send_peer_notif) 2858 bond_peer_notify_may_events(bond, true); 2859 2860 rtnl_unlock(); /* might sleep, hold no other locks */ 2861 } 2862 2863 re_arm: 2864 if (bond->params.miimon) 2865 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2866 } 2867 2868 static int bond_upper_dev_walk(struct net_device *upper, 2869 struct netdev_nested_priv *priv) 2870 { 2871 __be32 ip = *(__be32 *)priv->data; 2872 2873 return ip == bond_confirm_addr(upper, 0, ip); 2874 } 2875 2876 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2877 { 2878 struct netdev_nested_priv priv = { 2879 .data = (void *)&ip, 2880 }; 2881 bool ret = false; 2882 2883 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2884 return true; 2885 2886 rcu_read_lock(); 2887 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2888 ret = true; 2889 rcu_read_unlock(); 2890 2891 return ret; 2892 } 2893 2894 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff) 2895 2896 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2897 struct sk_buff *skb) 2898 { 2899 struct net_device *bond_dev = slave->bond->dev; 2900 struct net_device *slave_dev = slave->dev; 2901 struct bond_vlan_tag *outer_tag = tags; 2902 2903 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE) 2904 return true; 2905 2906 tags++; 2907 2908 /* Go through all the tags backwards and add them to the packet */ 2909 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) { 2910 if (!tags->vlan_id) { 2911 tags++; 2912 continue; 2913 } 2914 2915 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2916 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2917 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2918 tags->vlan_id); 2919 if (!skb) { 2920 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2921 return false; 2922 } 2923 2924 tags++; 2925 } 2926 /* Set the outer tag */ 2927 if (outer_tag->vlan_id) { 2928 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2929 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2930 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2931 outer_tag->vlan_id); 2932 } 2933 2934 return true; 2935 } 2936 2937 /* We go to the (large) trouble of VLAN tagging ARP frames because 2938 * switches in VLAN mode (especially if ports are configured as 2939 * "native" to a VLAN) might not pass non-tagged frames. 2940 */ 2941 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2942 __be32 src_ip, struct bond_vlan_tag *tags) 2943 { 2944 struct net_device *bond_dev = slave->bond->dev; 2945 struct net_device *slave_dev = slave->dev; 2946 struct sk_buff *skb; 2947 2948 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2949 arp_op, &dest_ip, &src_ip); 2950 2951 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 2952 NULL, slave_dev->dev_addr, NULL); 2953 2954 if (!skb) { 2955 net_err_ratelimited("ARP packet allocation failed\n"); 2956 return; 2957 } 2958 2959 if (bond_handle_vlan(slave, tags, skb)) { 2960 slave_update_last_tx(slave); 2961 arp_xmit(skb); 2962 } 2963 2964 return; 2965 } 2966 2967 /* Validate the device path between the @start_dev and the @end_dev. 2968 * The path is valid if the @end_dev is reachable through device 2969 * stacking. 2970 * When the path is validated, collect any vlan information in the 2971 * path. 2972 */ 2973 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 2974 struct net_device *end_dev, 2975 int level) 2976 { 2977 struct bond_vlan_tag *tags; 2978 struct net_device *upper; 2979 struct list_head *iter; 2980 2981 if (start_dev == end_dev) { 2982 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 2983 if (!tags) 2984 return ERR_PTR(-ENOMEM); 2985 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE; 2986 return tags; 2987 } 2988 2989 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 2990 tags = bond_verify_device_path(upper, end_dev, level + 1); 2991 if (IS_ERR_OR_NULL(tags)) { 2992 if (IS_ERR(tags)) 2993 return tags; 2994 continue; 2995 } 2996 if (is_vlan_dev(upper)) { 2997 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 2998 tags[level].vlan_id = vlan_dev_vlan_id(upper); 2999 } 3000 3001 return tags; 3002 } 3003 3004 return NULL; 3005 } 3006 3007 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 3008 { 3009 struct rtable *rt; 3010 struct bond_vlan_tag *tags; 3011 __be32 *targets = bond->params.arp_targets, addr; 3012 int i; 3013 3014 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 3015 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 3016 __func__, &targets[i]); 3017 tags = NULL; 3018 3019 /* Find out through which dev should the packet go */ 3020 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0, 3021 RT_SCOPE_LINK); 3022 if (IS_ERR(rt)) { 3023 /* there's no route to target - try to send arp 3024 * probe to generate any traffic (arp_validate=0) 3025 */ 3026 if (bond->params.arp_validate) 3027 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 3028 bond->dev->name, 3029 &targets[i]); 3030 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 3031 0, tags); 3032 continue; 3033 } 3034 3035 /* bond device itself */ 3036 if (rt->dst.dev == bond->dev) 3037 goto found; 3038 3039 rcu_read_lock(); 3040 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 3041 rcu_read_unlock(); 3042 3043 if (!IS_ERR_OR_NULL(tags)) 3044 goto found; 3045 3046 /* Not our device - skip */ 3047 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 3048 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 3049 3050 ip_rt_put(rt); 3051 continue; 3052 3053 found: 3054 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 3055 ip_rt_put(rt); 3056 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 3057 kfree(tags); 3058 } 3059 } 3060 3061 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3062 { 3063 int i; 3064 3065 if (!sip || !bond_has_this_ip(bond, tip)) { 3066 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3067 __func__, &sip, &tip); 3068 return; 3069 } 3070 3071 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3072 if (i == -1) { 3073 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3074 __func__, &sip); 3075 return; 3076 } 3077 WRITE_ONCE(slave->last_rx, jiffies); 3078 WRITE_ONCE(slave->target_last_arp_rx[i], jiffies); 3079 } 3080 3081 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3082 struct slave *slave) 3083 { 3084 struct arphdr *arp = (struct arphdr *)skb->data; 3085 struct slave *curr_active_slave, *curr_arp_slave; 3086 unsigned char *arp_ptr; 3087 __be32 sip, tip; 3088 unsigned int alen; 3089 3090 alen = arp_hdr_len(bond->dev); 3091 3092 if (alen > skb_headlen(skb)) { 3093 arp = kmalloc(alen, GFP_ATOMIC); 3094 if (!arp) 3095 goto out_unlock; 3096 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3097 goto out_unlock; 3098 } 3099 3100 if (arp->ar_hln != bond->dev->addr_len || 3101 skb->pkt_type == PACKET_OTHERHOST || 3102 skb->pkt_type == PACKET_LOOPBACK || 3103 arp->ar_hrd != htons(ARPHRD_ETHER) || 3104 arp->ar_pro != htons(ETH_P_IP) || 3105 arp->ar_pln != 4) 3106 goto out_unlock; 3107 3108 arp_ptr = (unsigned char *)(arp + 1); 3109 arp_ptr += bond->dev->addr_len; 3110 memcpy(&sip, arp_ptr, 4); 3111 arp_ptr += 4 + bond->dev->addr_len; 3112 memcpy(&tip, arp_ptr, 4); 3113 3114 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3115 __func__, slave->dev->name, bond_slave_state(slave), 3116 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3117 &sip, &tip); 3118 3119 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3120 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3121 3122 /* We 'trust' the received ARP enough to validate it if: 3123 * 3124 * (a) the slave receiving the ARP is active (which includes the 3125 * current ARP slave, if any), or 3126 * 3127 * (b) the receiving slave isn't active, but there is a currently 3128 * active slave and it received valid arp reply(s) after it became 3129 * the currently active slave, or 3130 * 3131 * (c) there is an ARP slave that sent an ARP during the prior ARP 3132 * interval, and we receive an ARP reply on any slave. We accept 3133 * these because switch FDB update delays may deliver the ARP 3134 * reply to a slave other than the sender of the ARP request. 3135 * 3136 * Note: for (b), backup slaves are receiving the broadcast ARP 3137 * request, not a reply. This request passes from the sending 3138 * slave through the L2 switch(es) to the receiving slave. Since 3139 * this is checking the request, sip/tip are swapped for 3140 * validation. 3141 * 3142 * This is done to avoid endless looping when we can't reach the 3143 * arp_ip_target and fool ourselves with our own arp requests. 3144 */ 3145 if (bond_is_active_slave(slave)) 3146 bond_validate_arp(bond, slave, sip, tip); 3147 else if (curr_active_slave && 3148 time_after(slave_last_rx(bond, curr_active_slave), 3149 curr_active_slave->last_link_up)) 3150 bond_validate_arp(bond, slave, tip, sip); 3151 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3152 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3153 bond_validate_arp(bond, slave, sip, tip); 3154 3155 out_unlock: 3156 if (arp != (struct arphdr *)skb->data) 3157 kfree(arp); 3158 return RX_HANDLER_ANOTHER; 3159 } 3160 3161 #if IS_ENABLED(CONFIG_IPV6) 3162 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3163 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3164 { 3165 struct net_device *bond_dev = slave->bond->dev; 3166 struct net_device *slave_dev = slave->dev; 3167 struct in6_addr mcaddr; 3168 struct sk_buff *skb; 3169 3170 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3171 daddr, saddr); 3172 3173 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3174 if (!skb) { 3175 net_err_ratelimited("NS packet allocation failed\n"); 3176 return; 3177 } 3178 3179 addrconf_addr_solict_mult(daddr, &mcaddr); 3180 if (bond_handle_vlan(slave, tags, skb)) { 3181 slave_update_last_tx(slave); 3182 ndisc_send_skb(skb, &mcaddr, saddr); 3183 } 3184 } 3185 3186 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3187 { 3188 struct in6_addr *targets = bond->params.ns_targets; 3189 struct bond_vlan_tag *tags; 3190 struct dst_entry *dst; 3191 struct in6_addr saddr; 3192 struct flowi6 fl6; 3193 int i; 3194 3195 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3196 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3197 __func__, &targets[i]); 3198 tags = NULL; 3199 3200 /* Find out through which dev should the packet go */ 3201 memset(&fl6, 0, sizeof(struct flowi6)); 3202 fl6.daddr = targets[i]; 3203 3204 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3205 if (dst->error) { 3206 dst_release(dst); 3207 /* there's no route to target - try to send arp 3208 * probe to generate any traffic (arp_validate=0) 3209 */ 3210 if (bond->params.arp_validate) 3211 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3212 bond->dev->name, 3213 &targets[i]); 3214 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3215 continue; 3216 } 3217 3218 /* bond device itself */ 3219 if (dst->dev == bond->dev) 3220 goto found; 3221 3222 rcu_read_lock(); 3223 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3224 rcu_read_unlock(); 3225 3226 if (!IS_ERR_OR_NULL(tags)) 3227 goto found; 3228 3229 /* Not our device - skip */ 3230 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3231 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3232 3233 dst_release(dst); 3234 continue; 3235 3236 found: 3237 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3238 bond_ns_send(slave, &targets[i], &saddr, tags); 3239 else 3240 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3241 3242 dst_release(dst); 3243 kfree(tags); 3244 } 3245 } 3246 3247 static int bond_confirm_addr6(struct net_device *dev, 3248 struct netdev_nested_priv *priv) 3249 { 3250 struct in6_addr *addr = (struct in6_addr *)priv->data; 3251 3252 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3253 } 3254 3255 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3256 { 3257 struct netdev_nested_priv priv = { 3258 .data = addr, 3259 }; 3260 int ret = false; 3261 3262 if (bond_confirm_addr6(bond->dev, &priv)) 3263 return true; 3264 3265 rcu_read_lock(); 3266 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3267 ret = true; 3268 rcu_read_unlock(); 3269 3270 return ret; 3271 } 3272 3273 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3274 struct in6_addr *saddr, struct in6_addr *daddr) 3275 { 3276 int i; 3277 3278 /* Ignore NAs that: 3279 * 1. Source address is unspecified address. 3280 * 2. Dest address is neither all-nodes multicast address nor 3281 * exist on bond interface. 3282 */ 3283 if (ipv6_addr_any(saddr) || 3284 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3285 !bond_has_this_ip6(bond, daddr))) { 3286 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3287 __func__, saddr, daddr); 3288 return; 3289 } 3290 3291 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3292 if (i == -1) { 3293 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3294 __func__, saddr); 3295 return; 3296 } 3297 WRITE_ONCE(slave->last_rx, jiffies); 3298 WRITE_ONCE(slave->target_last_arp_rx[i], jiffies); 3299 } 3300 3301 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3302 struct slave *slave) 3303 { 3304 struct slave *curr_active_slave, *curr_arp_slave; 3305 struct in6_addr *saddr, *daddr; 3306 struct { 3307 struct ipv6hdr ip6; 3308 struct icmp6hdr icmp6; 3309 } *combined, _combined; 3310 3311 if (skb->pkt_type == PACKET_OTHERHOST || 3312 skb->pkt_type == PACKET_LOOPBACK) 3313 goto out; 3314 3315 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3316 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3317 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION && 3318 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)) 3319 goto out; 3320 3321 saddr = &combined->ip6.saddr; 3322 daddr = &combined->ip6.daddr; 3323 3324 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3325 __func__, slave->dev->name, bond_slave_state(slave), 3326 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3327 saddr, daddr); 3328 3329 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3330 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3331 3332 /* We 'trust' the received ARP enough to validate it if: 3333 * see bond_arp_rcv(). 3334 */ 3335 if (bond_is_active_slave(slave)) 3336 bond_validate_na(bond, slave, saddr, daddr); 3337 else if (curr_active_slave && 3338 time_after(slave_last_rx(bond, curr_active_slave), 3339 curr_active_slave->last_link_up)) 3340 bond_validate_na(bond, slave, daddr, saddr); 3341 else if (curr_arp_slave && 3342 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3343 bond_validate_na(bond, slave, saddr, daddr); 3344 3345 out: 3346 return RX_HANDLER_ANOTHER; 3347 } 3348 #endif 3349 3350 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3351 struct slave *slave) 3352 { 3353 #if IS_ENABLED(CONFIG_IPV6) 3354 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3355 #endif 3356 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3357 3358 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3359 __func__, skb->dev->name); 3360 3361 /* Use arp validate logic for both ARP and NS */ 3362 if (!slave_do_arp_validate(bond, slave)) { 3363 if ((slave_do_arp_validate_only(bond) && is_arp) || 3364 #if IS_ENABLED(CONFIG_IPV6) 3365 (slave_do_arp_validate_only(bond) && is_ipv6) || 3366 #endif 3367 !slave_do_arp_validate_only(bond)) 3368 WRITE_ONCE(slave->last_rx, jiffies); 3369 return RX_HANDLER_ANOTHER; 3370 } else if (is_arp) { 3371 return bond_arp_rcv(skb, bond, slave); 3372 #if IS_ENABLED(CONFIG_IPV6) 3373 } else if (is_ipv6) { 3374 return bond_na_rcv(skb, bond, slave); 3375 #endif 3376 } else { 3377 return RX_HANDLER_ANOTHER; 3378 } 3379 } 3380 3381 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3382 { 3383 bond_arp_send_all(bond, slave); 3384 #if IS_ENABLED(CONFIG_IPV6) 3385 bond_ns_send_all(bond, slave); 3386 #endif 3387 } 3388 3389 /* function to verify if we're in the arp_interval timeslice, returns true if 3390 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3391 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3392 */ 3393 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3394 int mod) 3395 { 3396 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3397 3398 return time_in_range(jiffies, 3399 last_act - delta_in_ticks, 3400 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3401 } 3402 3403 /* This function is called regularly to monitor each slave's link 3404 * ensuring that traffic is being sent and received when arp monitoring 3405 * is used in load-balancing mode. if the adapter has been dormant, then an 3406 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3407 * arp monitoring in active backup mode. 3408 */ 3409 static void bond_loadbalance_arp_mon(struct bonding *bond) 3410 { 3411 struct slave *slave, *oldcurrent; 3412 struct list_head *iter; 3413 int do_failover = 0, slave_state_changed = 0; 3414 3415 if (!bond_has_slaves(bond)) 3416 goto re_arm; 3417 3418 rcu_read_lock(); 3419 3420 oldcurrent = rcu_dereference(bond->curr_active_slave); 3421 /* see if any of the previous devices are up now (i.e. they have 3422 * xmt and rcv traffic). the curr_active_slave does not come into 3423 * the picture unless it is null. also, slave->last_link_up is not 3424 * needed here because we send an arp on each slave and give a slave 3425 * as long as it needs to get the tx/rx within the delta. 3426 * TODO: what about up/down delay in arp mode? it wasn't here before 3427 * so it can wait 3428 */ 3429 bond_for_each_slave_rcu(bond, slave, iter) { 3430 unsigned long last_tx = slave_last_tx(slave); 3431 3432 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3433 3434 if (slave->link != BOND_LINK_UP) { 3435 if (bond_time_in_interval(bond, last_tx, 1) && 3436 bond_time_in_interval(bond, READ_ONCE(slave->last_rx), 1)) { 3437 3438 bond_propose_link_state(slave, BOND_LINK_UP); 3439 slave_state_changed = 1; 3440 3441 /* primary_slave has no meaning in round-robin 3442 * mode. the window of a slave being up and 3443 * curr_active_slave being null after enslaving 3444 * is closed. 3445 */ 3446 if (!oldcurrent) { 3447 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3448 do_failover = 1; 3449 } else { 3450 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3451 } 3452 } 3453 } else { 3454 /* slave->link == BOND_LINK_UP */ 3455 3456 /* not all switches will respond to an arp request 3457 * when the source ip is 0, so don't take the link down 3458 * if we don't know our ip yet 3459 */ 3460 if (!bond_time_in_interval(bond, last_tx, 3461 bond->params.missed_max) || 3462 !bond_time_in_interval(bond, READ_ONCE(slave->last_rx), 3463 bond->params.missed_max)) { 3464 3465 bond_propose_link_state(slave, BOND_LINK_DOWN); 3466 slave_state_changed = 1; 3467 3468 if (slave->link_failure_count < UINT_MAX) 3469 slave->link_failure_count++; 3470 3471 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3472 3473 if (slave == oldcurrent) 3474 do_failover = 1; 3475 } 3476 } 3477 3478 /* note: if switch is in round-robin mode, all links 3479 * must tx arp to ensure all links rx an arp - otherwise 3480 * links may oscillate or not come up at all; if switch is 3481 * in something like xor mode, there is nothing we can 3482 * do - all replies will be rx'ed on same link causing slaves 3483 * to be unstable during low/no traffic periods 3484 */ 3485 if (bond_slave_is_up(slave)) 3486 bond_send_validate(bond, slave); 3487 } 3488 3489 rcu_read_unlock(); 3490 3491 if (do_failover || slave_state_changed) { 3492 if (!rtnl_trylock()) 3493 goto re_arm; 3494 3495 bond_for_each_slave(bond, slave, iter) { 3496 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3497 slave->link = slave->link_new_state; 3498 } 3499 3500 if (slave_state_changed) { 3501 bond_slave_state_change(bond); 3502 if (BOND_MODE(bond) == BOND_MODE_XOR) 3503 bond_update_slave_arr(bond, NULL); 3504 } 3505 if (do_failover) { 3506 block_netpoll_tx(); 3507 bond_select_active_slave(bond); 3508 unblock_netpoll_tx(); 3509 } 3510 rtnl_unlock(); 3511 } 3512 3513 re_arm: 3514 if (bond->params.arp_interval) 3515 queue_delayed_work(bond->wq, &bond->arp_work, 3516 msecs_to_jiffies(bond->params.arp_interval)); 3517 } 3518 3519 /* Called to inspect slaves for active-backup mode ARP monitor link state 3520 * changes. Sets proposed link state in slaves to specify what action 3521 * should take place for the slave. Returns 0 if no changes are found, >0 3522 * if changes to link states must be committed. 3523 * 3524 * Called with rcu_read_lock held. 3525 */ 3526 static int bond_ab_arp_inspect(struct bonding *bond) 3527 { 3528 unsigned long last_tx, last_rx; 3529 struct list_head *iter; 3530 struct slave *slave; 3531 int commit = 0; 3532 3533 bond_for_each_slave_rcu(bond, slave, iter) { 3534 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3535 last_rx = slave_last_rx(bond, slave); 3536 3537 if (slave->link != BOND_LINK_UP) { 3538 if (bond_time_in_interval(bond, last_rx, 1)) { 3539 bond_propose_link_state(slave, BOND_LINK_UP); 3540 commit++; 3541 } else if (slave->link == BOND_LINK_BACK) { 3542 bond_propose_link_state(slave, BOND_LINK_FAIL); 3543 commit++; 3544 } 3545 continue; 3546 } 3547 3548 /* Give slaves 2*delta after being enslaved or made 3549 * active. This avoids bouncing, as the last receive 3550 * times need a full ARP monitor cycle to be updated. 3551 */ 3552 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3553 continue; 3554 3555 /* Backup slave is down if: 3556 * - No current_arp_slave AND 3557 * - more than (missed_max+1)*delta since last receive AND 3558 * - the bond has an IP address 3559 * 3560 * Note: a non-null current_arp_slave indicates 3561 * the curr_active_slave went down and we are 3562 * searching for a new one; under this condition 3563 * we only take the curr_active_slave down - this 3564 * gives each slave a chance to tx/rx traffic 3565 * before being taken out 3566 */ 3567 if (!bond_is_active_slave(slave) && 3568 !rcu_access_pointer(bond->current_arp_slave) && 3569 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3570 bond_propose_link_state(slave, BOND_LINK_DOWN); 3571 commit++; 3572 } 3573 3574 /* Active slave is down if: 3575 * - more than missed_max*delta since transmitting OR 3576 * - (more than missed_max*delta since receive AND 3577 * the bond has an IP address) 3578 */ 3579 last_tx = slave_last_tx(slave); 3580 if (bond_is_active_slave(slave) && 3581 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3582 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3583 bond_propose_link_state(slave, BOND_LINK_DOWN); 3584 commit++; 3585 } 3586 } 3587 3588 return commit; 3589 } 3590 3591 /* Called to commit link state changes noted by inspection step of 3592 * active-backup mode ARP monitor. 3593 * 3594 * Called with RTNL hold. 3595 */ 3596 static void bond_ab_arp_commit(struct bonding *bond) 3597 { 3598 bool do_failover = false; 3599 struct list_head *iter; 3600 unsigned long last_tx; 3601 struct slave *slave; 3602 3603 bond_for_each_slave(bond, slave, iter) { 3604 switch (slave->link_new_state) { 3605 case BOND_LINK_NOCHANGE: 3606 continue; 3607 3608 case BOND_LINK_UP: 3609 last_tx = slave_last_tx(slave); 3610 if (rtnl_dereference(bond->curr_active_slave) != slave || 3611 (!rtnl_dereference(bond->curr_active_slave) && 3612 bond_time_in_interval(bond, last_tx, 1))) { 3613 struct slave *current_arp_slave; 3614 3615 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3616 bond_set_slave_link_state(slave, BOND_LINK_UP, 3617 BOND_SLAVE_NOTIFY_NOW); 3618 if (current_arp_slave) { 3619 bond_set_slave_inactive_flags( 3620 current_arp_slave, 3621 BOND_SLAVE_NOTIFY_NOW); 3622 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3623 } 3624 3625 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3626 3627 if (!rtnl_dereference(bond->curr_active_slave) || 3628 slave == rtnl_dereference(bond->primary_slave) || 3629 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio) 3630 do_failover = true; 3631 3632 } 3633 3634 continue; 3635 3636 case BOND_LINK_DOWN: 3637 if (slave->link_failure_count < UINT_MAX) 3638 slave->link_failure_count++; 3639 3640 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3641 BOND_SLAVE_NOTIFY_NOW); 3642 bond_set_slave_inactive_flags(slave, 3643 BOND_SLAVE_NOTIFY_NOW); 3644 3645 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3646 3647 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3648 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3649 do_failover = true; 3650 } 3651 3652 continue; 3653 3654 case BOND_LINK_FAIL: 3655 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3656 BOND_SLAVE_NOTIFY_NOW); 3657 bond_set_slave_inactive_flags(slave, 3658 BOND_SLAVE_NOTIFY_NOW); 3659 3660 /* A slave has just been enslaved and has become 3661 * the current active slave. 3662 */ 3663 if (rtnl_dereference(bond->curr_active_slave)) 3664 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3665 continue; 3666 3667 default: 3668 slave_err(bond->dev, slave->dev, 3669 "impossible: link_new_state %d on slave\n", 3670 slave->link_new_state); 3671 continue; 3672 } 3673 } 3674 3675 if (do_failover) { 3676 block_netpoll_tx(); 3677 bond_select_active_slave(bond); 3678 unblock_netpoll_tx(); 3679 } 3680 3681 bond_set_carrier(bond); 3682 } 3683 3684 /* Send ARP probes for active-backup mode ARP monitor. 3685 * 3686 * Called with rcu_read_lock held. 3687 */ 3688 static bool bond_ab_arp_probe(struct bonding *bond) 3689 { 3690 struct slave *slave, *before = NULL, *new_slave = NULL, 3691 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3692 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3693 struct list_head *iter; 3694 bool found = false; 3695 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3696 3697 if (curr_arp_slave && curr_active_slave) 3698 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3699 curr_arp_slave->dev->name, 3700 curr_active_slave->dev->name); 3701 3702 if (curr_active_slave) { 3703 bond_send_validate(bond, curr_active_slave); 3704 return should_notify_rtnl; 3705 } 3706 3707 /* if we don't have a curr_active_slave, search for the next available 3708 * backup slave from the current_arp_slave and make it the candidate 3709 * for becoming the curr_active_slave 3710 */ 3711 3712 if (!curr_arp_slave) { 3713 curr_arp_slave = bond_first_slave_rcu(bond); 3714 if (!curr_arp_slave) 3715 return should_notify_rtnl; 3716 } 3717 3718 bond_for_each_slave_rcu(bond, slave, iter) { 3719 if (!found && !before && bond_slave_is_up(slave)) 3720 before = slave; 3721 3722 if (found && !new_slave && bond_slave_is_up(slave)) 3723 new_slave = slave; 3724 /* if the link state is up at this point, we 3725 * mark it down - this can happen if we have 3726 * simultaneous link failures and 3727 * reselect_active_interface doesn't make this 3728 * one the current slave so it is still marked 3729 * up when it is actually down 3730 */ 3731 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3732 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3733 BOND_SLAVE_NOTIFY_LATER); 3734 if (slave->link_failure_count < UINT_MAX) 3735 slave->link_failure_count++; 3736 3737 bond_set_slave_inactive_flags(slave, 3738 BOND_SLAVE_NOTIFY_LATER); 3739 3740 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3741 } 3742 if (slave == curr_arp_slave) 3743 found = true; 3744 } 3745 3746 if (!new_slave && before) 3747 new_slave = before; 3748 3749 if (!new_slave) 3750 goto check_state; 3751 3752 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3753 BOND_SLAVE_NOTIFY_LATER); 3754 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3755 bond_send_validate(bond, new_slave); 3756 new_slave->last_link_up = jiffies; 3757 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3758 3759 check_state: 3760 bond_for_each_slave_rcu(bond, slave, iter) { 3761 if (slave->should_notify || slave->should_notify_link) { 3762 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3763 break; 3764 } 3765 } 3766 return should_notify_rtnl; 3767 } 3768 3769 static void bond_activebackup_arp_mon(struct bonding *bond) 3770 { 3771 bool should_notify_rtnl; 3772 int delta_in_ticks; 3773 3774 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3775 3776 if (!bond_has_slaves(bond)) 3777 goto re_arm; 3778 3779 rcu_read_lock(); 3780 3781 if (bond_ab_arp_inspect(bond)) { 3782 rcu_read_unlock(); 3783 3784 /* Race avoidance with bond_close flush of workqueue */ 3785 if (!rtnl_trylock()) { 3786 delta_in_ticks = 1; 3787 goto re_arm; 3788 } 3789 3790 bond_ab_arp_commit(bond); 3791 3792 rtnl_unlock(); 3793 rcu_read_lock(); 3794 } 3795 3796 should_notify_rtnl = bond_ab_arp_probe(bond); 3797 rcu_read_unlock(); 3798 3799 if (READ_ONCE(bond->send_peer_notif) || should_notify_rtnl) { 3800 if (!rtnl_trylock()) { 3801 delta_in_ticks = 1; 3802 goto re_arm; 3803 } 3804 3805 if (bond->send_peer_notif) 3806 bond_peer_notify_may_events(bond, true); 3807 3808 if (should_notify_rtnl) { 3809 bond_slave_state_notify(bond); 3810 bond_slave_link_notify(bond); 3811 } 3812 3813 rtnl_unlock(); 3814 } 3815 3816 re_arm: 3817 if (bond->params.arp_interval) 3818 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3819 } 3820 3821 static void bond_arp_monitor(struct work_struct *work) 3822 { 3823 struct bonding *bond = container_of(work, struct bonding, 3824 arp_work.work); 3825 3826 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3827 bond_activebackup_arp_mon(bond); 3828 else 3829 bond_loadbalance_arp_mon(bond); 3830 } 3831 3832 /*-------------------------- netdev event handling --------------------------*/ 3833 3834 /* Change device name */ 3835 static int bond_event_changename(struct bonding *bond) 3836 { 3837 bond_remove_proc_entry(bond); 3838 bond_create_proc_entry(bond); 3839 3840 bond_debug_reregister(bond); 3841 3842 return NOTIFY_DONE; 3843 } 3844 3845 static int bond_master_netdev_event(unsigned long event, 3846 struct net_device *bond_dev) 3847 { 3848 struct bonding *event_bond = netdev_priv(bond_dev); 3849 3850 netdev_dbg(bond_dev, "%s called\n", __func__); 3851 3852 switch (event) { 3853 case NETDEV_CHANGENAME: 3854 return bond_event_changename(event_bond); 3855 case NETDEV_UNREGISTER: 3856 bond_remove_proc_entry(event_bond); 3857 #ifdef CONFIG_XFRM_OFFLOAD 3858 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3859 #endif /* CONFIG_XFRM_OFFLOAD */ 3860 break; 3861 case NETDEV_REGISTER: 3862 bond_create_proc_entry(event_bond); 3863 break; 3864 default: 3865 break; 3866 } 3867 3868 return NOTIFY_DONE; 3869 } 3870 3871 static int bond_slave_netdev_event(unsigned long event, 3872 struct net_device *slave_dev) 3873 { 3874 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3875 struct bonding *bond; 3876 struct net_device *bond_dev; 3877 3878 /* A netdev event can be generated while enslaving a device 3879 * before netdev_rx_handler_register is called in which case 3880 * slave will be NULL 3881 */ 3882 if (!slave) { 3883 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3884 return NOTIFY_DONE; 3885 } 3886 3887 bond_dev = slave->bond->dev; 3888 bond = slave->bond; 3889 primary = rtnl_dereference(bond->primary_slave); 3890 3891 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3892 3893 switch (event) { 3894 case NETDEV_UNREGISTER: 3895 if (bond_dev->type != ARPHRD_ETHER) 3896 bond_release_and_destroy(bond_dev, slave_dev); 3897 else 3898 __bond_release_one(bond_dev, slave_dev, false, true); 3899 break; 3900 case NETDEV_UP: 3901 case NETDEV_CHANGE: 3902 /* For 802.3ad mode only: 3903 * Getting invalid Speed/Duplex values here will put slave 3904 * in weird state. Mark it as link-fail if the link was 3905 * previously up or link-down if it hasn't yet come up, and 3906 * let link-monitoring (miimon) set it right when correct 3907 * speeds/duplex are available. 3908 */ 3909 if (bond_update_speed_duplex(slave) && 3910 BOND_MODE(bond) == BOND_MODE_8023AD) { 3911 if (slave->last_link_up) 3912 slave->link = BOND_LINK_FAIL; 3913 else 3914 slave->link = BOND_LINK_DOWN; 3915 } 3916 3917 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3918 bond_3ad_adapter_speed_duplex_changed(slave); 3919 fallthrough; 3920 case NETDEV_DOWN: 3921 /* Refresh slave-array if applicable! 3922 * If the setup does not use miimon or arpmon (mode-specific!), 3923 * then these events will not cause the slave-array to be 3924 * refreshed. This will cause xmit to use a slave that is not 3925 * usable. Avoid such situation by refeshing the array at these 3926 * events. If these (miimon/arpmon) parameters are configured 3927 * then array gets refreshed twice and that should be fine! 3928 */ 3929 if (bond_mode_can_use_xmit_hash(bond)) 3930 bond_update_slave_arr(bond, NULL); 3931 break; 3932 case NETDEV_CHANGEMTU: 3933 /* TODO: Should slaves be allowed to 3934 * independently alter their MTU? For 3935 * an active-backup bond, slaves need 3936 * not be the same type of device, so 3937 * MTUs may vary. For other modes, 3938 * slaves arguably should have the 3939 * same MTUs. To do this, we'd need to 3940 * take over the slave's change_mtu 3941 * function for the duration of their 3942 * servitude. 3943 */ 3944 break; 3945 case NETDEV_CHANGENAME: 3946 /* we don't care if we don't have primary set */ 3947 if (!bond_uses_primary(bond) || 3948 !bond->params.primary[0]) 3949 break; 3950 3951 if (slave == primary) { 3952 /* slave's name changed - he's no longer primary */ 3953 RCU_INIT_POINTER(bond->primary_slave, NULL); 3954 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 3955 /* we have a new primary slave */ 3956 rcu_assign_pointer(bond->primary_slave, slave); 3957 } else { /* we didn't change primary - exit */ 3958 break; 3959 } 3960 3961 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 3962 primary ? slave_dev->name : "none"); 3963 3964 block_netpoll_tx(); 3965 bond_select_active_slave(bond); 3966 unblock_netpoll_tx(); 3967 break; 3968 case NETDEV_FEAT_CHANGE: 3969 if (!bond->notifier_ctx) { 3970 bond->notifier_ctx = true; 3971 netdev_compute_master_upper_features(bond->dev, true); 3972 bond->notifier_ctx = false; 3973 } 3974 break; 3975 case NETDEV_RESEND_IGMP: 3976 /* Propagate to master device */ 3977 call_netdevice_notifiers(event, slave->bond->dev); 3978 break; 3979 case NETDEV_XDP_FEAT_CHANGE: 3980 bond_xdp_set_features(bond_dev); 3981 break; 3982 default: 3983 break; 3984 } 3985 3986 return NOTIFY_DONE; 3987 } 3988 3989 /* bond_netdev_event: handle netdev notifier chain events. 3990 * 3991 * This function receives events for the netdev chain. The caller (an 3992 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 3993 * locks for us to safely manipulate the slave devices (RTNL lock, 3994 * dev_probe_lock). 3995 */ 3996 static int bond_netdev_event(struct notifier_block *this, 3997 unsigned long event, void *ptr) 3998 { 3999 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 4000 4001 netdev_dbg(event_dev, "%s received %s\n", 4002 __func__, netdev_cmd_to_name(event)); 4003 4004 if (!(event_dev->priv_flags & IFF_BONDING)) 4005 return NOTIFY_DONE; 4006 4007 if (event_dev->flags & IFF_MASTER) { 4008 int ret; 4009 4010 ret = bond_master_netdev_event(event, event_dev); 4011 if (ret != NOTIFY_DONE) 4012 return ret; 4013 } 4014 4015 if (event_dev->flags & IFF_SLAVE) 4016 return bond_slave_netdev_event(event, event_dev); 4017 4018 return NOTIFY_DONE; 4019 } 4020 4021 static struct notifier_block bond_netdev_notifier = { 4022 .notifier_call = bond_netdev_event, 4023 }; 4024 4025 /*---------------------------- Hashing Policies -----------------------------*/ 4026 4027 /* Helper to access data in a packet, with or without a backing skb. 4028 * If skb is given the data is linearized if necessary via pskb_may_pull. 4029 */ 4030 static inline const void *bond_pull_data(struct sk_buff *skb, 4031 const void *data, int hlen, int n) 4032 { 4033 if (likely(n <= hlen)) 4034 return data; 4035 else if (skb && likely(pskb_may_pull(skb, n))) 4036 return skb->data; 4037 4038 return NULL; 4039 } 4040 4041 /* L2 hash helper */ 4042 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4043 { 4044 struct ethhdr *ep; 4045 4046 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4047 if (!data) 4048 return 0; 4049 4050 ep = (struct ethhdr *)(data + mhoff); 4051 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 4052 } 4053 4054 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 4055 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 4056 { 4057 const struct ipv6hdr *iph6; 4058 const struct iphdr *iph; 4059 4060 if (l2_proto == htons(ETH_P_IP)) { 4061 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 4062 if (!data) 4063 return false; 4064 4065 iph = (const struct iphdr *)(data + *nhoff); 4066 iph_to_flow_copy_v4addrs(fk, iph); 4067 *nhoff += iph->ihl << 2; 4068 if (!ip_is_fragment(iph)) 4069 *ip_proto = iph->protocol; 4070 } else if (l2_proto == htons(ETH_P_IPV6)) { 4071 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4072 if (!data) 4073 return false; 4074 4075 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4076 iph_to_flow_copy_v6addrs(fk, iph6); 4077 *nhoff += sizeof(*iph6); 4078 *ip_proto = iph6->nexthdr; 4079 } else { 4080 return false; 4081 } 4082 4083 if (l34 && *ip_proto >= 0) 4084 fk->ports.ports = skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4085 4086 return true; 4087 } 4088 4089 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4090 { 4091 u32 srcmac_vendor = 0, srcmac_dev = 0; 4092 struct ethhdr *mac_hdr; 4093 u16 vlan = 0; 4094 int i; 4095 4096 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4097 if (!data) 4098 return 0; 4099 mac_hdr = (struct ethhdr *)(data + mhoff); 4100 4101 for (i = 0; i < 3; i++) 4102 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4103 4104 for (i = 3; i < ETH_ALEN; i++) 4105 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4106 4107 if (skb && skb_vlan_tag_present(skb)) 4108 vlan = skb_vlan_tag_get(skb); 4109 4110 return vlan ^ srcmac_vendor ^ srcmac_dev; 4111 } 4112 4113 /* Extract the appropriate headers based on bond's xmit policy */ 4114 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4115 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4116 { 4117 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4118 int ip_proto = -1; 4119 4120 switch (bond->params.xmit_policy) { 4121 case BOND_XMIT_POLICY_ENCAP23: 4122 case BOND_XMIT_POLICY_ENCAP34: 4123 memset(fk, 0, sizeof(*fk)); 4124 return __skb_flow_dissect(dev_net(bond->dev), skb, 4125 &flow_keys_bonding, fk, data, 4126 l2_proto, nhoff, hlen, 0); 4127 default: 4128 break; 4129 } 4130 4131 fk->ports.ports = 0; 4132 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4133 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4134 return false; 4135 4136 /* ICMP error packets contains at least 8 bytes of the header 4137 * of the packet which generated the error. Use this information 4138 * to correlate ICMP error packets within the same flow which 4139 * generated the error. 4140 */ 4141 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4142 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4143 if (ip_proto == IPPROTO_ICMP) { 4144 if (!icmp_is_err(fk->icmp.type)) 4145 return true; 4146 4147 nhoff += sizeof(struct icmphdr); 4148 } else if (ip_proto == IPPROTO_ICMPV6) { 4149 if (!icmpv6_is_err(fk->icmp.type)) 4150 return true; 4151 4152 nhoff += sizeof(struct icmp6hdr); 4153 } 4154 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4155 } 4156 4157 return true; 4158 } 4159 4160 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4161 { 4162 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4163 (__force u32)flow_get_u32_src(flow); 4164 hash ^= (hash >> 16); 4165 hash ^= (hash >> 8); 4166 4167 /* discard lowest hash bit to deal with the common even ports pattern */ 4168 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4169 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4170 return hash >> 1; 4171 4172 return hash; 4173 } 4174 4175 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4176 * the data as required, but this function can be used without it if the data is 4177 * known to be linear (e.g. with xdp_buff). 4178 */ 4179 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4180 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4181 { 4182 struct flow_keys flow; 4183 u32 hash; 4184 4185 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4186 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4187 4188 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4189 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4190 return bond_eth_hash(skb, data, mhoff, hlen); 4191 4192 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4193 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4194 hash = bond_eth_hash(skb, data, mhoff, hlen); 4195 } else { 4196 if (flow.icmp.id) 4197 memcpy(&hash, &flow.icmp, sizeof(hash)); 4198 else 4199 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4200 } 4201 4202 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4203 } 4204 4205 /** 4206 * bond_xmit_hash - generate a hash value based on the xmit policy 4207 * @bond: bonding device 4208 * @skb: buffer to use for headers 4209 * 4210 * This function will extract the necessary headers from the skb buffer and use 4211 * them to generate a hash based on the xmit_policy set in the bonding device 4212 */ 4213 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4214 { 4215 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4216 skb->l4_hash) 4217 return skb->hash; 4218 4219 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4220 0, skb_network_offset(skb), 4221 skb_headlen(skb)); 4222 } 4223 4224 /** 4225 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4226 * @bond: bonding device 4227 * @xdp: buffer to use for headers 4228 * 4229 * The XDP variant of bond_xmit_hash. 4230 */ 4231 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4232 { 4233 struct ethhdr *eth; 4234 4235 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4236 return 0; 4237 4238 eth = (struct ethhdr *)xdp->data; 4239 4240 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4241 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4242 } 4243 4244 /*-------------------------- Device entry points ----------------------------*/ 4245 4246 void bond_work_init_all(struct bonding *bond) 4247 { 4248 /* ndo_stop, bond_close() will try to flush the work under 4249 * the rtnl lock. The workqueue must not block on rtnl lock 4250 * to avoid deadlock. 4251 */ 4252 INIT_DELAYED_WORK(&bond->mcast_work, 4253 bond_resend_igmp_join_requests_delayed); 4254 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4255 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4256 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4257 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4258 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4259 INIT_DELAYED_WORK(&bond->peer_notify_work, bond_peer_notify_handler); 4260 } 4261 4262 void bond_work_cancel_all(struct bonding *bond) 4263 { 4264 cancel_delayed_work_sync(&bond->mii_work); 4265 cancel_delayed_work_sync(&bond->arp_work); 4266 cancel_delayed_work_sync(&bond->alb_work); 4267 cancel_delayed_work_sync(&bond->ad_work); 4268 cancel_delayed_work_sync(&bond->mcast_work); 4269 cancel_delayed_work_sync(&bond->slave_arr_work); 4270 cancel_delayed_work_sync(&bond->peer_notify_work); 4271 } 4272 4273 static int bond_open(struct net_device *bond_dev) 4274 { 4275 struct bonding *bond = netdev_priv(bond_dev); 4276 struct list_head *iter; 4277 struct slave *slave; 4278 4279 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4280 bond->rr_tx_counter = alloc_percpu(u32); 4281 if (!bond->rr_tx_counter) 4282 return -ENOMEM; 4283 } 4284 4285 /* reset slave->backup and slave->inactive */ 4286 if (bond_has_slaves(bond)) { 4287 bond_for_each_slave(bond, slave, iter) { 4288 if (bond_uses_primary(bond) && 4289 slave != rcu_access_pointer(bond->curr_active_slave)) { 4290 bond_set_slave_inactive_flags(slave, 4291 BOND_SLAVE_NOTIFY_NOW); 4292 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4293 bond_set_slave_active_flags(slave, 4294 BOND_SLAVE_NOTIFY_NOW); 4295 } 4296 } 4297 } 4298 4299 if (bond_is_lb(bond)) { 4300 /* bond_alb_initialize must be called before the timer 4301 * is started. 4302 */ 4303 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4304 return -ENOMEM; 4305 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4306 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4307 } 4308 4309 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4310 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4311 4312 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4313 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4314 bond->recv_probe = bond_rcv_validate; 4315 } 4316 4317 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4318 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4319 /* register to receive LACPDUs */ 4320 bond->recv_probe = bond_3ad_lacpdu_recv; 4321 bond_3ad_initiate_agg_selection(bond, 1); 4322 4323 bond_for_each_slave(bond, slave, iter) 4324 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4325 4326 if (bond->params.broadcast_neighbor) 4327 static_branch_inc(&bond_bcast_neigh_enabled); 4328 } 4329 4330 if (bond_mode_can_use_xmit_hash(bond)) 4331 bond_update_slave_arr(bond, NULL); 4332 4333 return 0; 4334 } 4335 4336 static int bond_close(struct net_device *bond_dev) 4337 { 4338 struct bonding *bond = netdev_priv(bond_dev); 4339 struct slave *slave; 4340 4341 bond_work_cancel_all(bond); 4342 bond->send_peer_notif = 0; 4343 if (bond_is_lb(bond)) 4344 bond_alb_deinitialize(bond); 4345 bond->recv_probe = NULL; 4346 4347 if (BOND_MODE(bond) == BOND_MODE_8023AD && 4348 bond->params.broadcast_neighbor) 4349 static_branch_dec(&bond_bcast_neigh_enabled); 4350 4351 if (bond_uses_primary(bond)) { 4352 rcu_read_lock(); 4353 slave = rcu_dereference(bond->curr_active_slave); 4354 if (slave) 4355 bond_hw_addr_flush(bond_dev, slave->dev); 4356 rcu_read_unlock(); 4357 } else { 4358 struct list_head *iter; 4359 4360 bond_for_each_slave(bond, slave, iter) 4361 bond_hw_addr_flush(bond_dev, slave->dev); 4362 } 4363 4364 return 0; 4365 } 4366 4367 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4368 * that some drivers can provide 32bit values only. 4369 */ 4370 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4371 const struct rtnl_link_stats64 *_new, 4372 const struct rtnl_link_stats64 *_old) 4373 { 4374 const u64 *new = (const u64 *)_new; 4375 const u64 *old = (const u64 *)_old; 4376 u64 *res = (u64 *)_res; 4377 int i; 4378 4379 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4380 u64 nv = new[i]; 4381 u64 ov = old[i]; 4382 s64 delta = nv - ov; 4383 4384 /* detects if this particular field is 32bit only */ 4385 if (((nv | ov) >> 32) == 0) 4386 delta = (s64)(s32)((u32)nv - (u32)ov); 4387 4388 /* filter anomalies, some drivers reset their stats 4389 * at down/up events. 4390 */ 4391 if (delta > 0) 4392 res[i] += delta; 4393 } 4394 } 4395 4396 #ifdef CONFIG_LOCKDEP 4397 static int bond_get_lowest_level_rcu(struct net_device *dev) 4398 { 4399 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4400 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4401 int cur = 0, max = 0; 4402 4403 now = dev; 4404 iter = &dev->adj_list.lower; 4405 4406 while (1) { 4407 next = NULL; 4408 while (1) { 4409 ldev = netdev_next_lower_dev_rcu(now, &iter); 4410 if (!ldev) 4411 break; 4412 4413 next = ldev; 4414 niter = &ldev->adj_list.lower; 4415 dev_stack[cur] = now; 4416 iter_stack[cur++] = iter; 4417 if (max <= cur) 4418 max = cur; 4419 break; 4420 } 4421 4422 if (!next) { 4423 if (!cur) 4424 return max; 4425 next = dev_stack[--cur]; 4426 niter = iter_stack[cur]; 4427 } 4428 4429 now = next; 4430 iter = niter; 4431 } 4432 4433 return max; 4434 } 4435 #endif 4436 4437 static void bond_get_stats(struct net_device *bond_dev, 4438 struct rtnl_link_stats64 *stats) 4439 { 4440 struct bonding *bond = netdev_priv(bond_dev); 4441 struct rtnl_link_stats64 temp; 4442 struct list_head *iter; 4443 struct slave *slave; 4444 int nest_level = 0; 4445 4446 4447 rcu_read_lock(); 4448 #ifdef CONFIG_LOCKDEP 4449 nest_level = bond_get_lowest_level_rcu(bond_dev); 4450 #endif 4451 4452 spin_lock_nested(&bond->stats_lock, nest_level); 4453 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4454 4455 bond_for_each_slave_rcu(bond, slave, iter) { 4456 const struct rtnl_link_stats64 *new = 4457 dev_get_stats(slave->dev, &temp); 4458 4459 bond_fold_stats(stats, new, &slave->slave_stats); 4460 4461 /* save off the slave stats for the next run */ 4462 memcpy(&slave->slave_stats, new, sizeof(*new)); 4463 } 4464 4465 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4466 spin_unlock(&bond->stats_lock); 4467 rcu_read_unlock(); 4468 } 4469 4470 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4471 { 4472 struct bonding *bond = netdev_priv(bond_dev); 4473 struct mii_ioctl_data *mii = NULL; 4474 4475 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4476 4477 switch (cmd) { 4478 case SIOCGMIIPHY: 4479 mii = if_mii(ifr); 4480 if (!mii) 4481 return -EINVAL; 4482 4483 mii->phy_id = 0; 4484 fallthrough; 4485 case SIOCGMIIREG: 4486 /* We do this again just in case we were called by SIOCGMIIREG 4487 * instead of SIOCGMIIPHY. 4488 */ 4489 mii = if_mii(ifr); 4490 if (!mii) 4491 return -EINVAL; 4492 4493 if (mii->reg_num == 1) { 4494 mii->val_out = 0; 4495 if (netif_carrier_ok(bond->dev)) 4496 mii->val_out = BMSR_LSTATUS; 4497 } 4498 4499 break; 4500 default: 4501 return -EOPNOTSUPP; 4502 } 4503 4504 return 0; 4505 } 4506 4507 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4508 { 4509 struct bonding *bond = netdev_priv(bond_dev); 4510 struct net_device *slave_dev = NULL; 4511 struct ifbond k_binfo; 4512 struct ifbond __user *u_binfo = NULL; 4513 struct ifslave k_sinfo; 4514 struct ifslave __user *u_sinfo = NULL; 4515 struct bond_opt_value newval; 4516 struct net *net; 4517 int res = 0; 4518 4519 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4520 4521 switch (cmd) { 4522 case SIOCBONDINFOQUERY: 4523 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4524 4525 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4526 return -EFAULT; 4527 4528 bond_info_query(bond_dev, &k_binfo); 4529 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4530 return -EFAULT; 4531 4532 return 0; 4533 case SIOCBONDSLAVEINFOQUERY: 4534 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4535 4536 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4537 return -EFAULT; 4538 4539 res = bond_slave_info_query(bond_dev, &k_sinfo); 4540 if (res == 0 && 4541 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4542 return -EFAULT; 4543 4544 return res; 4545 default: 4546 break; 4547 } 4548 4549 net = dev_net(bond_dev); 4550 4551 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4552 return -EPERM; 4553 4554 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4555 4556 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4557 4558 if (!slave_dev) 4559 return -ENODEV; 4560 4561 switch (cmd) { 4562 case SIOCBONDENSLAVE: 4563 res = bond_enslave(bond_dev, slave_dev, NULL); 4564 break; 4565 case SIOCBONDRELEASE: 4566 res = bond_release(bond_dev, slave_dev); 4567 break; 4568 case SIOCBONDSETHWADDR: 4569 res = bond_set_dev_addr(bond_dev, slave_dev); 4570 break; 4571 case SIOCBONDCHANGEACTIVE: 4572 bond_opt_initstr(&newval, slave_dev->name); 4573 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4574 &newval); 4575 break; 4576 default: 4577 res = -EOPNOTSUPP; 4578 } 4579 4580 return res; 4581 } 4582 4583 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4584 void __user *data, int cmd) 4585 { 4586 struct ifreq ifrdata = { .ifr_data = data }; 4587 4588 switch (cmd) { 4589 case BOND_INFO_QUERY_OLD: 4590 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4591 case BOND_SLAVE_INFO_QUERY_OLD: 4592 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4593 case BOND_ENSLAVE_OLD: 4594 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4595 case BOND_RELEASE_OLD: 4596 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4597 case BOND_SETHWADDR_OLD: 4598 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4599 case BOND_CHANGE_ACTIVE_OLD: 4600 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4601 } 4602 4603 return -EOPNOTSUPP; 4604 } 4605 4606 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4607 { 4608 struct bonding *bond = netdev_priv(bond_dev); 4609 4610 if (change & IFF_PROMISC) 4611 bond_set_promiscuity(bond, 4612 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4613 4614 if (change & IFF_ALLMULTI) 4615 bond_set_allmulti(bond, 4616 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4617 } 4618 4619 static void bond_set_rx_mode(struct net_device *bond_dev) 4620 { 4621 struct bonding *bond = netdev_priv(bond_dev); 4622 struct list_head *iter; 4623 struct slave *slave; 4624 4625 rcu_read_lock(); 4626 if (bond_uses_primary(bond)) { 4627 slave = rcu_dereference(bond->curr_active_slave); 4628 if (slave) { 4629 dev_uc_sync(slave->dev, bond_dev); 4630 dev_mc_sync(slave->dev, bond_dev); 4631 } 4632 } else { 4633 bond_for_each_slave_rcu(bond, slave, iter) { 4634 dev_uc_sync_multiple(slave->dev, bond_dev); 4635 dev_mc_sync_multiple(slave->dev, bond_dev); 4636 } 4637 } 4638 rcu_read_unlock(); 4639 } 4640 4641 static int bond_neigh_init(struct neighbour *n) 4642 { 4643 struct bonding *bond = netdev_priv(n->dev); 4644 const struct net_device_ops *slave_ops; 4645 struct neigh_parms parms; 4646 struct slave *slave; 4647 int ret = 0; 4648 4649 rcu_read_lock(); 4650 slave = bond_first_slave_rcu(bond); 4651 if (!slave) 4652 goto out; 4653 slave_ops = slave->dev->netdev_ops; 4654 if (!slave_ops->ndo_neigh_setup) 4655 goto out; 4656 4657 /* TODO: find another way [1] to implement this. 4658 * Passing a zeroed structure is fragile, 4659 * but at least we do not pass garbage. 4660 * 4661 * [1] One way would be that ndo_neigh_setup() never touch 4662 * struct neigh_parms, but propagate the new neigh_setup() 4663 * back to ___neigh_create() / neigh_parms_alloc() 4664 */ 4665 memset(&parms, 0, sizeof(parms)); 4666 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4667 4668 if (ret) 4669 goto out; 4670 4671 if (parms.neigh_setup) 4672 ret = parms.neigh_setup(n); 4673 out: 4674 rcu_read_unlock(); 4675 return ret; 4676 } 4677 4678 /* The bonding ndo_neigh_setup is called at init time beofre any 4679 * slave exists. So we must declare proxy setup function which will 4680 * be used at run time to resolve the actual slave neigh param setup. 4681 * 4682 * It's also called by master devices (such as vlans) to setup their 4683 * underlying devices. In that case - do nothing, we're already set up from 4684 * our init. 4685 */ 4686 static int bond_neigh_setup(struct net_device *dev, 4687 struct neigh_parms *parms) 4688 { 4689 /* modify only our neigh_parms */ 4690 if (parms->dev == dev) 4691 parms->neigh_setup = bond_neigh_init; 4692 4693 return 0; 4694 } 4695 4696 /* Change the MTU of all of a master's slaves to match the master */ 4697 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4698 { 4699 struct bonding *bond = netdev_priv(bond_dev); 4700 struct slave *slave, *rollback_slave; 4701 struct list_head *iter; 4702 int res = 0; 4703 4704 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4705 4706 bond_for_each_slave(bond, slave, iter) { 4707 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4708 slave, slave->dev->netdev_ops->ndo_change_mtu); 4709 4710 res = dev_set_mtu(slave->dev, new_mtu); 4711 4712 if (res) { 4713 /* If we failed to set the slave's mtu to the new value 4714 * we must abort the operation even in ACTIVE_BACKUP 4715 * mode, because if we allow the backup slaves to have 4716 * different mtu values than the active slave we'll 4717 * need to change their mtu when doing a failover. That 4718 * means changing their mtu from timer context, which 4719 * is probably not a good idea. 4720 */ 4721 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4722 res, new_mtu); 4723 goto unwind; 4724 } 4725 } 4726 4727 WRITE_ONCE(bond_dev->mtu, new_mtu); 4728 4729 return 0; 4730 4731 unwind: 4732 /* unwind from head to the slave that failed */ 4733 bond_for_each_slave(bond, rollback_slave, iter) { 4734 int tmp_res; 4735 4736 if (rollback_slave == slave) 4737 break; 4738 4739 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4740 if (tmp_res) 4741 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4742 tmp_res); 4743 } 4744 4745 return res; 4746 } 4747 4748 /* Change HW address 4749 * 4750 * Note that many devices must be down to change the HW address, and 4751 * downing the master releases all slaves. We can make bonds full of 4752 * bonding devices to test this, however. 4753 */ 4754 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4755 { 4756 struct bonding *bond = netdev_priv(bond_dev); 4757 struct slave *slave, *rollback_slave; 4758 struct sockaddr_storage *ss = addr, tmp_ss; 4759 struct list_head *iter; 4760 int res = 0; 4761 4762 if (BOND_MODE(bond) == BOND_MODE_ALB) 4763 return bond_alb_set_mac_address(bond_dev, addr); 4764 4765 4766 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4767 4768 /* If fail_over_mac is enabled, do nothing and return success. 4769 * Returning an error causes ifenslave to fail. 4770 */ 4771 if (bond->params.fail_over_mac && 4772 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4773 return 0; 4774 4775 if (!is_valid_ether_addr(ss->__data)) 4776 return -EADDRNOTAVAIL; 4777 4778 bond_for_each_slave(bond, slave, iter) { 4779 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4780 __func__, slave); 4781 res = dev_set_mac_address(slave->dev, addr, NULL); 4782 if (res) { 4783 /* TODO: consider downing the slave 4784 * and retry ? 4785 * User should expect communications 4786 * breakage anyway until ARP finish 4787 * updating, so... 4788 */ 4789 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4790 __func__, res); 4791 goto unwind; 4792 } 4793 } 4794 4795 /* success */ 4796 dev_addr_set(bond_dev, ss->__data); 4797 return 0; 4798 4799 unwind: 4800 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4801 tmp_ss.ss_family = bond_dev->type; 4802 4803 /* unwind from head to the slave that failed */ 4804 bond_for_each_slave(bond, rollback_slave, iter) { 4805 int tmp_res; 4806 4807 if (rollback_slave == slave) 4808 break; 4809 4810 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_ss, NULL); 4811 if (tmp_res) { 4812 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4813 __func__, tmp_res); 4814 } 4815 } 4816 4817 return res; 4818 } 4819 4820 /** 4821 * bond_get_slave_by_id - get xmit slave with slave_id 4822 * @bond: bonding device that is transmitting 4823 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4824 * 4825 * This function tries to get slave with slave_id but in case 4826 * it fails, it tries to find the first available slave for transmission. 4827 */ 4828 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4829 int slave_id) 4830 { 4831 struct list_head *iter; 4832 struct slave *slave; 4833 int i = slave_id; 4834 4835 /* Here we start from the slave with slave_id */ 4836 bond_for_each_slave_rcu(bond, slave, iter) { 4837 if (--i < 0) { 4838 if (bond_slave_can_tx(slave)) 4839 return slave; 4840 } 4841 } 4842 4843 /* Here we start from the first slave up to slave_id */ 4844 i = slave_id; 4845 bond_for_each_slave_rcu(bond, slave, iter) { 4846 if (--i < 0) 4847 break; 4848 if (bond_slave_can_tx(slave)) 4849 return slave; 4850 } 4851 /* no slave that can tx has been found */ 4852 return NULL; 4853 } 4854 4855 /** 4856 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4857 * @bond: bonding device to use 4858 * 4859 * Based on the value of the bonding device's packets_per_slave parameter 4860 * this function generates a slave id, which is usually used as the next 4861 * slave to transmit through. 4862 */ 4863 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4864 { 4865 u32 slave_id; 4866 struct reciprocal_value reciprocal_packets_per_slave; 4867 int packets_per_slave = bond->params.packets_per_slave; 4868 4869 switch (packets_per_slave) { 4870 case 0: 4871 slave_id = get_random_u32(); 4872 break; 4873 case 1: 4874 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4875 break; 4876 default: 4877 reciprocal_packets_per_slave = 4878 bond->params.reciprocal_packets_per_slave; 4879 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4880 slave_id = reciprocal_divide(slave_id, 4881 reciprocal_packets_per_slave); 4882 break; 4883 } 4884 4885 return slave_id; 4886 } 4887 4888 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4889 struct sk_buff *skb) 4890 { 4891 struct slave *slave; 4892 int slave_cnt; 4893 u32 slave_id; 4894 4895 /* Start with the curr_active_slave that joined the bond as the 4896 * default for sending IGMP traffic. For failover purposes one 4897 * needs to maintain some consistency for the interface that will 4898 * send the join/membership reports. The curr_active_slave found 4899 * will send all of this type of traffic. 4900 */ 4901 if (skb->protocol == htons(ETH_P_IP)) { 4902 int noff = skb_network_offset(skb); 4903 struct iphdr *iph; 4904 4905 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4906 goto non_igmp; 4907 4908 iph = ip_hdr(skb); 4909 if (iph->protocol == IPPROTO_IGMP) { 4910 slave = rcu_dereference(bond->curr_active_slave); 4911 if (slave) 4912 return slave; 4913 return bond_get_slave_by_id(bond, 0); 4914 } 4915 } 4916 4917 non_igmp: 4918 slave_cnt = READ_ONCE(bond->slave_cnt); 4919 if (likely(slave_cnt)) { 4920 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4921 return bond_get_slave_by_id(bond, slave_id); 4922 } 4923 return NULL; 4924 } 4925 4926 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4927 struct xdp_buff *xdp) 4928 { 4929 struct slave *slave; 4930 int slave_cnt; 4931 u32 slave_id; 4932 const struct ethhdr *eth; 4933 void *data = xdp->data; 4934 4935 if (data + sizeof(struct ethhdr) > xdp->data_end) 4936 goto non_igmp; 4937 4938 eth = (struct ethhdr *)data; 4939 data += sizeof(struct ethhdr); 4940 4941 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4942 if (eth->h_proto == htons(ETH_P_IP)) { 4943 const struct iphdr *iph; 4944 4945 if (data + sizeof(struct iphdr) > xdp->data_end) 4946 goto non_igmp; 4947 4948 iph = (struct iphdr *)data; 4949 4950 if (iph->protocol == IPPROTO_IGMP) { 4951 slave = rcu_dereference(bond->curr_active_slave); 4952 if (slave) 4953 return slave; 4954 return bond_get_slave_by_id(bond, 0); 4955 } 4956 } 4957 4958 non_igmp: 4959 slave_cnt = READ_ONCE(bond->slave_cnt); 4960 if (likely(slave_cnt)) { 4961 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4962 return bond_get_slave_by_id(bond, slave_id); 4963 } 4964 return NULL; 4965 } 4966 4967 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 4968 struct net_device *bond_dev) 4969 { 4970 struct bonding *bond = netdev_priv(bond_dev); 4971 struct slave *slave; 4972 4973 slave = bond_xmit_roundrobin_slave_get(bond, skb); 4974 if (likely(slave)) 4975 return bond_dev_queue_xmit(bond, skb, slave->dev); 4976 4977 return bond_tx_drop(bond_dev, skb); 4978 } 4979 4980 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 4981 { 4982 return rcu_dereference(bond->curr_active_slave); 4983 } 4984 4985 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 4986 * the bond has a usable interface. 4987 */ 4988 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 4989 struct net_device *bond_dev) 4990 { 4991 struct bonding *bond = netdev_priv(bond_dev); 4992 struct slave *slave; 4993 4994 slave = bond_xmit_activebackup_slave_get(bond); 4995 if (slave) 4996 return bond_dev_queue_xmit(bond, skb, slave->dev); 4997 4998 return bond_tx_drop(bond_dev, skb); 4999 } 5000 5001 /* Use this to update slave_array when (a) it's not appropriate to update 5002 * slave_array right away (note that update_slave_array() may sleep) 5003 * and / or (b) RTNL is not held. 5004 */ 5005 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 5006 { 5007 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 5008 } 5009 5010 /* Slave array work handler. Holds only RTNL */ 5011 static void bond_slave_arr_handler(struct work_struct *work) 5012 { 5013 struct bonding *bond = container_of(work, struct bonding, 5014 slave_arr_work.work); 5015 int ret; 5016 5017 if (!rtnl_trylock()) 5018 goto err; 5019 5020 ret = bond_update_slave_arr(bond, NULL); 5021 rtnl_unlock(); 5022 if (ret) { 5023 pr_warn_ratelimited("Failed to update slave array from WT\n"); 5024 goto err; 5025 } 5026 return; 5027 5028 err: 5029 bond_slave_arr_work_rearm(bond, 1); 5030 } 5031 5032 static void bond_skip_slave(struct bond_up_slave *slaves, 5033 struct slave *skipslave) 5034 { 5035 int idx; 5036 5037 /* Rare situation where caller has asked to skip a specific 5038 * slave but allocation failed (most likely!). BTW this is 5039 * only possible when the call is initiated from 5040 * __bond_release_one(). In this situation; overwrite the 5041 * skipslave entry in the array with the last entry from the 5042 * array to avoid a situation where the xmit path may choose 5043 * this to-be-skipped slave to send a packet out. 5044 */ 5045 for (idx = 0; slaves && idx < slaves->count; idx++) { 5046 if (skipslave == slaves->arr[idx]) { 5047 slaves->arr[idx] = 5048 slaves->arr[slaves->count - 1]; 5049 slaves->count--; 5050 break; 5051 } 5052 } 5053 } 5054 5055 static void bond_set_slave_arr(struct bonding *bond, 5056 struct bond_up_slave *usable_slaves, 5057 struct bond_up_slave *all_slaves) 5058 { 5059 struct bond_up_slave *usable, *all; 5060 5061 usable = rtnl_dereference(bond->usable_slaves); 5062 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5063 kfree_rcu(usable, rcu); 5064 5065 all = rtnl_dereference(bond->all_slaves); 5066 rcu_assign_pointer(bond->all_slaves, all_slaves); 5067 kfree_rcu(all, rcu); 5068 } 5069 5070 static void bond_reset_slave_arr(struct bonding *bond) 5071 { 5072 bond_set_slave_arr(bond, NULL, NULL); 5073 } 5074 5075 /* Build the usable slaves array in control path for modes that use xmit-hash 5076 * to determine the slave interface - 5077 * (a) BOND_MODE_8023AD 5078 * (b) BOND_MODE_XOR 5079 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5080 * 5081 * The caller is expected to hold RTNL only and NO other lock! 5082 */ 5083 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5084 { 5085 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5086 struct slave *slave; 5087 struct list_head *iter; 5088 int agg_id = 0; 5089 int ret = 0; 5090 5091 might_sleep(); 5092 5093 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5094 bond->slave_cnt), GFP_KERNEL); 5095 all_slaves = kzalloc(struct_size(all_slaves, arr, 5096 bond->slave_cnt), GFP_KERNEL); 5097 if (!usable_slaves || !all_slaves) { 5098 ret = -ENOMEM; 5099 goto out; 5100 } 5101 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5102 struct ad_info ad_info; 5103 5104 spin_lock_bh(&bond->mode_lock); 5105 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5106 spin_unlock_bh(&bond->mode_lock); 5107 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5108 /* No active aggragator means it's not safe to use 5109 * the previous array. 5110 */ 5111 bond_reset_slave_arr(bond); 5112 goto out; 5113 } 5114 spin_unlock_bh(&bond->mode_lock); 5115 agg_id = ad_info.aggregator_id; 5116 } 5117 bond_for_each_slave(bond, slave, iter) { 5118 if (skipslave == slave) 5119 continue; 5120 5121 all_slaves->arr[all_slaves->count++] = slave; 5122 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5123 struct aggregator *agg; 5124 5125 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5126 if (!agg || agg->aggregator_identifier != agg_id) 5127 continue; 5128 } 5129 if (!bond_slave_can_tx(slave)) 5130 continue; 5131 5132 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5133 usable_slaves->count); 5134 5135 usable_slaves->arr[usable_slaves->count++] = slave; 5136 } 5137 5138 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5139 return ret; 5140 out: 5141 if (ret != 0 && skipslave) { 5142 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5143 skipslave); 5144 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5145 skipslave); 5146 } 5147 kfree_rcu(all_slaves, rcu); 5148 kfree_rcu(usable_slaves, rcu); 5149 5150 return ret; 5151 } 5152 5153 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5154 struct sk_buff *skb, 5155 struct bond_up_slave *slaves) 5156 { 5157 struct slave *slave; 5158 unsigned int count; 5159 u32 hash; 5160 5161 hash = bond_xmit_hash(bond, skb); 5162 count = slaves ? READ_ONCE(slaves->count) : 0; 5163 if (unlikely(!count)) 5164 return NULL; 5165 5166 slave = slaves->arr[hash % count]; 5167 return slave; 5168 } 5169 5170 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5171 struct xdp_buff *xdp) 5172 { 5173 struct bond_up_slave *slaves; 5174 unsigned int count; 5175 u32 hash; 5176 5177 hash = bond_xmit_hash_xdp(bond, xdp); 5178 slaves = rcu_dereference(bond->usable_slaves); 5179 count = slaves ? READ_ONCE(slaves->count) : 0; 5180 if (unlikely(!count)) 5181 return NULL; 5182 5183 return slaves->arr[hash % count]; 5184 } 5185 5186 static bool bond_should_broadcast_neighbor(struct sk_buff *skb, 5187 struct net_device *dev) 5188 { 5189 struct bonding *bond = netdev_priv(dev); 5190 struct { 5191 struct ipv6hdr ip6; 5192 struct icmp6hdr icmp6; 5193 } *combined, _combined; 5194 5195 if (!static_branch_unlikely(&bond_bcast_neigh_enabled)) 5196 return false; 5197 5198 if (!bond->params.broadcast_neighbor) 5199 return false; 5200 5201 if (skb->protocol == htons(ETH_P_ARP)) 5202 return true; 5203 5204 if (skb->protocol == htons(ETH_P_IPV6)) { 5205 combined = skb_header_pointer(skb, skb_mac_header_len(skb), 5206 sizeof(_combined), 5207 &_combined); 5208 if (combined && combined->ip6.nexthdr == NEXTHDR_ICMP && 5209 (combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION || 5210 combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT)) 5211 return true; 5212 } 5213 5214 return false; 5215 } 5216 5217 /* Use this Xmit function for 3AD as well as XOR modes. The current 5218 * usable slave array is formed in the control path. The xmit function 5219 * just calculates hash and sends the packet out. 5220 */ 5221 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5222 struct net_device *dev) 5223 { 5224 struct bonding *bond = netdev_priv(dev); 5225 struct bond_up_slave *slaves; 5226 struct slave *slave; 5227 5228 slaves = rcu_dereference(bond->usable_slaves); 5229 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5230 if (likely(slave)) 5231 return bond_dev_queue_xmit(bond, skb, slave->dev); 5232 5233 return bond_tx_drop(dev, skb); 5234 } 5235 5236 /* in broadcast mode, we send everything to all or usable slave interfaces. 5237 * under rcu_read_lock when this function is called. 5238 */ 5239 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5240 struct net_device *bond_dev, 5241 bool all_slaves) 5242 { 5243 struct bonding *bond = netdev_priv(bond_dev); 5244 struct bond_up_slave *slaves; 5245 bool xmit_suc = false; 5246 bool skb_used = false; 5247 int slaves_count, i; 5248 5249 if (all_slaves) 5250 slaves = rcu_dereference(bond->all_slaves); 5251 else 5252 slaves = rcu_dereference(bond->usable_slaves); 5253 5254 slaves_count = slaves ? READ_ONCE(slaves->count) : 0; 5255 for (i = 0; i < slaves_count; i++) { 5256 struct slave *slave = slaves->arr[i]; 5257 struct sk_buff *skb2; 5258 5259 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5260 continue; 5261 5262 if (bond_is_last_slave(bond, slave)) { 5263 skb2 = skb; 5264 skb_used = true; 5265 } else { 5266 skb2 = skb_clone(skb, GFP_ATOMIC); 5267 if (!skb2) { 5268 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5269 bond_dev->name, __func__); 5270 continue; 5271 } 5272 } 5273 5274 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5275 xmit_suc = true; 5276 } 5277 5278 if (!skb_used) 5279 dev_kfree_skb_any(skb); 5280 5281 if (xmit_suc) 5282 return NETDEV_TX_OK; 5283 5284 dev_core_stats_tx_dropped_inc(bond_dev); 5285 return NET_XMIT_DROP; 5286 } 5287 5288 /*------------------------- Device initialization ---------------------------*/ 5289 5290 /* Lookup the slave that corresponds to a qid */ 5291 static inline int bond_slave_override(struct bonding *bond, 5292 struct sk_buff *skb) 5293 { 5294 struct slave *slave = NULL; 5295 struct list_head *iter; 5296 5297 if (!skb_rx_queue_recorded(skb)) 5298 return 1; 5299 5300 /* Find out if any slaves have the same mapping as this skb. */ 5301 bond_for_each_slave_rcu(bond, slave, iter) { 5302 if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) { 5303 if (bond_slave_is_up(slave) && 5304 slave->link == BOND_LINK_UP) { 5305 bond_dev_queue_xmit(bond, skb, slave->dev); 5306 return 0; 5307 } 5308 /* If the slave isn't UP, use default transmit policy. */ 5309 break; 5310 } 5311 } 5312 5313 return 1; 5314 } 5315 5316 5317 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5318 struct net_device *sb_dev) 5319 { 5320 /* This helper function exists to help dev_pick_tx get the correct 5321 * destination queue. Using a helper function skips a call to 5322 * skb_tx_hash and will put the skbs in the queue we expect on their 5323 * way down to the bonding driver. 5324 */ 5325 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5326 5327 /* Save the original txq to restore before passing to the driver */ 5328 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5329 5330 if (unlikely(txq >= dev->real_num_tx_queues)) { 5331 do { 5332 txq -= dev->real_num_tx_queues; 5333 } while (txq >= dev->real_num_tx_queues); 5334 } 5335 return txq; 5336 } 5337 5338 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5339 struct sk_buff *skb, 5340 bool all_slaves) 5341 { 5342 struct bonding *bond = netdev_priv(master_dev); 5343 struct bond_up_slave *slaves; 5344 struct slave *slave = NULL; 5345 5346 switch (BOND_MODE(bond)) { 5347 case BOND_MODE_ROUNDROBIN: 5348 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5349 break; 5350 case BOND_MODE_ACTIVEBACKUP: 5351 slave = bond_xmit_activebackup_slave_get(bond); 5352 break; 5353 case BOND_MODE_8023AD: 5354 case BOND_MODE_XOR: 5355 if (all_slaves) 5356 slaves = rcu_dereference(bond->all_slaves); 5357 else 5358 slaves = rcu_dereference(bond->usable_slaves); 5359 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5360 break; 5361 case BOND_MODE_BROADCAST: 5362 break; 5363 case BOND_MODE_ALB: 5364 slave = bond_xmit_alb_slave_get(bond, skb); 5365 break; 5366 case BOND_MODE_TLB: 5367 slave = bond_xmit_tlb_slave_get(bond, skb); 5368 break; 5369 default: 5370 /* Should never happen, mode already checked */ 5371 WARN_ONCE(true, "Unknown bonding mode"); 5372 break; 5373 } 5374 5375 if (slave) 5376 return slave->dev; 5377 return NULL; 5378 } 5379 5380 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5381 { 5382 switch (sk->sk_family) { 5383 #if IS_ENABLED(CONFIG_IPV6) 5384 case AF_INET6: 5385 if (ipv6_only_sock(sk) || 5386 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5387 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5388 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5389 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5390 break; 5391 } 5392 fallthrough; 5393 #endif 5394 default: /* AF_INET */ 5395 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5396 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5397 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5398 break; 5399 } 5400 5401 flow->ports.src = inet_sk(sk)->inet_sport; 5402 flow->ports.dst = inet_sk(sk)->inet_dport; 5403 } 5404 5405 /** 5406 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5407 * @sk: socket to use for headers 5408 * 5409 * This function will extract the necessary field from the socket and use 5410 * them to generate a hash based on the LAYER34 xmit_policy. 5411 * Assumes that sk is a TCP or UDP socket. 5412 */ 5413 static u32 bond_sk_hash_l34(struct sock *sk) 5414 { 5415 struct flow_keys flow; 5416 u32 hash; 5417 5418 bond_sk_to_flow(sk, &flow); 5419 5420 /* L4 */ 5421 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5422 /* L3 */ 5423 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5424 } 5425 5426 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5427 struct sock *sk) 5428 { 5429 struct bond_up_slave *slaves; 5430 struct slave *slave; 5431 unsigned int count; 5432 u32 hash; 5433 5434 slaves = rcu_dereference(bond->usable_slaves); 5435 count = slaves ? READ_ONCE(slaves->count) : 0; 5436 if (unlikely(!count)) 5437 return NULL; 5438 5439 hash = bond_sk_hash_l34(sk); 5440 slave = slaves->arr[hash % count]; 5441 5442 return slave->dev; 5443 } 5444 5445 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5446 struct sock *sk) 5447 { 5448 struct bonding *bond = netdev_priv(dev); 5449 struct net_device *lower = NULL; 5450 5451 rcu_read_lock(); 5452 if (bond_sk_check(bond)) 5453 lower = __bond_sk_get_lower_dev(bond, sk); 5454 rcu_read_unlock(); 5455 5456 return lower; 5457 } 5458 5459 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5460 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5461 struct net_device *dev) 5462 { 5463 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5464 5465 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded 5466 * was true, if tls_device_down is running in parallel, but it's OK, 5467 * because bond_get_slave_by_dev has a NULL check. 5468 */ 5469 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5470 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5471 return bond_tx_drop(dev, skb); 5472 } 5473 #endif 5474 5475 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5476 { 5477 struct bonding *bond = netdev_priv(dev); 5478 5479 if (bond_should_override_tx_queue(bond) && 5480 !bond_slave_override(bond, skb)) 5481 return NETDEV_TX_OK; 5482 5483 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5484 if (tls_is_skb_tx_device_offloaded(skb)) 5485 return bond_tls_device_xmit(bond, skb, dev); 5486 #endif 5487 5488 switch (BOND_MODE(bond)) { 5489 case BOND_MODE_ROUNDROBIN: 5490 return bond_xmit_roundrobin(skb, dev); 5491 case BOND_MODE_ACTIVEBACKUP: 5492 return bond_xmit_activebackup(skb, dev); 5493 case BOND_MODE_8023AD: 5494 if (bond_should_broadcast_neighbor(skb, dev)) 5495 return bond_xmit_broadcast(skb, dev, false); 5496 fallthrough; 5497 case BOND_MODE_XOR: 5498 return bond_3ad_xor_xmit(skb, dev); 5499 case BOND_MODE_BROADCAST: 5500 return bond_xmit_broadcast(skb, dev, true); 5501 case BOND_MODE_ALB: 5502 return bond_alb_xmit(skb, dev); 5503 case BOND_MODE_TLB: 5504 return bond_tlb_xmit(skb, dev); 5505 default: 5506 /* Should never happen, mode already checked */ 5507 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5508 WARN_ON_ONCE(1); 5509 return bond_tx_drop(dev, skb); 5510 } 5511 } 5512 5513 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5514 { 5515 struct bonding *bond = netdev_priv(dev); 5516 netdev_tx_t ret = NETDEV_TX_OK; 5517 5518 /* If we risk deadlock from transmitting this in the 5519 * netpoll path, tell netpoll to queue the frame for later tx 5520 */ 5521 if (unlikely(is_netpoll_tx_blocked(dev))) 5522 return NETDEV_TX_BUSY; 5523 5524 rcu_read_lock(); 5525 if (bond_has_slaves(bond)) 5526 ret = __bond_start_xmit(skb, dev); 5527 else 5528 ret = bond_tx_drop(dev, skb); 5529 rcu_read_unlock(); 5530 5531 return ret; 5532 } 5533 5534 static struct net_device * 5535 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5536 { 5537 struct bonding *bond = netdev_priv(bond_dev); 5538 struct slave *slave; 5539 5540 /* Caller needs to hold rcu_read_lock() */ 5541 5542 switch (BOND_MODE(bond)) { 5543 case BOND_MODE_ROUNDROBIN: 5544 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5545 break; 5546 5547 case BOND_MODE_ACTIVEBACKUP: 5548 slave = bond_xmit_activebackup_slave_get(bond); 5549 break; 5550 5551 case BOND_MODE_8023AD: 5552 case BOND_MODE_XOR: 5553 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5554 break; 5555 5556 default: 5557 if (net_ratelimit()) 5558 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", 5559 BOND_MODE(bond)); 5560 return NULL; 5561 } 5562 5563 if (slave) 5564 return slave->dev; 5565 5566 return NULL; 5567 } 5568 5569 static int bond_xdp_xmit(struct net_device *bond_dev, 5570 int n, struct xdp_frame **frames, u32 flags) 5571 { 5572 int nxmit, err = -ENXIO; 5573 5574 rcu_read_lock(); 5575 5576 for (nxmit = 0; nxmit < n; nxmit++) { 5577 struct xdp_frame *frame = frames[nxmit]; 5578 struct xdp_frame *frames1[] = {frame}; 5579 struct net_device *slave_dev; 5580 struct xdp_buff xdp; 5581 5582 xdp_convert_frame_to_buff(frame, &xdp); 5583 5584 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5585 if (!slave_dev) { 5586 err = -ENXIO; 5587 break; 5588 } 5589 5590 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5591 if (err < 1) 5592 break; 5593 } 5594 5595 rcu_read_unlock(); 5596 5597 /* If error happened on the first frame then we can pass the error up, otherwise 5598 * report the number of frames that were xmitted. 5599 */ 5600 if (err < 0) 5601 return (nxmit == 0 ? err : nxmit); 5602 5603 return nxmit; 5604 } 5605 5606 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5607 struct netlink_ext_ack *extack) 5608 { 5609 struct bonding *bond = netdev_priv(dev); 5610 struct list_head *iter; 5611 struct slave *slave, *rollback_slave; 5612 struct bpf_prog *old_prog; 5613 struct netdev_bpf xdp = { 5614 .command = XDP_SETUP_PROG, 5615 .flags = 0, 5616 .prog = prog, 5617 .extack = extack, 5618 }; 5619 int err; 5620 5621 ASSERT_RTNL(); 5622 5623 if (!bond_xdp_check(bond, BOND_MODE(bond))) { 5624 BOND_NL_ERR(dev, extack, 5625 "No native XDP support for the current bonding mode"); 5626 return -EOPNOTSUPP; 5627 } 5628 5629 old_prog = bond->xdp_prog; 5630 bond->xdp_prog = prog; 5631 5632 bond_for_each_slave(bond, slave, iter) { 5633 struct net_device *slave_dev = slave->dev; 5634 5635 if (!slave_dev->netdev_ops->ndo_bpf || 5636 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5637 SLAVE_NL_ERR(dev, slave_dev, extack, 5638 "Slave device does not support XDP"); 5639 err = -EOPNOTSUPP; 5640 goto err; 5641 } 5642 5643 if (dev_xdp_prog_count(slave_dev) > 0) { 5644 SLAVE_NL_ERR(dev, slave_dev, extack, 5645 "Slave has XDP program loaded, please unload before enslaving"); 5646 err = -EOPNOTSUPP; 5647 goto err; 5648 } 5649 5650 err = dev_xdp_propagate(slave_dev, &xdp); 5651 if (err < 0) { 5652 /* ndo_bpf() sets extack error message */ 5653 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5654 goto err; 5655 } 5656 if (prog) 5657 bpf_prog_inc(prog); 5658 } 5659 5660 if (prog) { 5661 static_branch_inc(&bpf_master_redirect_enabled_key); 5662 } else if (old_prog) { 5663 bpf_prog_put(old_prog); 5664 static_branch_dec(&bpf_master_redirect_enabled_key); 5665 } 5666 5667 return 0; 5668 5669 err: 5670 /* unwind the program changes */ 5671 bond->xdp_prog = old_prog; 5672 xdp.prog = old_prog; 5673 xdp.extack = NULL; /* do not overwrite original error */ 5674 5675 bond_for_each_slave(bond, rollback_slave, iter) { 5676 struct net_device *slave_dev = rollback_slave->dev; 5677 int err_unwind; 5678 5679 if (slave == rollback_slave) 5680 break; 5681 5682 err_unwind = dev_xdp_propagate(slave_dev, &xdp); 5683 if (err_unwind < 0) 5684 slave_err(dev, slave_dev, 5685 "Error %d when unwinding XDP program change\n", err_unwind); 5686 else if (xdp.prog) 5687 bpf_prog_inc(xdp.prog); 5688 } 5689 return err; 5690 } 5691 5692 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5693 { 5694 switch (xdp->command) { 5695 case XDP_SETUP_PROG: 5696 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5697 default: 5698 return -EINVAL; 5699 } 5700 } 5701 5702 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5703 { 5704 if (speed == 0 || speed == SPEED_UNKNOWN) 5705 speed = slave->speed; 5706 else 5707 speed = min(speed, slave->speed); 5708 5709 return speed; 5710 } 5711 5712 /* Set the BOND_PHC_INDEX flag to notify user space */ 5713 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg) 5714 { 5715 struct ifreq *ifr = kernel_cfg->ifr; 5716 struct hwtstamp_config cfg; 5717 5718 if (kernel_cfg->copied_to_user) { 5719 /* Lower device has a legacy implementation */ 5720 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 5721 return -EFAULT; 5722 5723 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5724 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg))) 5725 return -EFAULT; 5726 } else { 5727 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5728 } 5729 5730 return 0; 5731 } 5732 5733 static int bond_hwtstamp_get(struct net_device *dev, 5734 struct kernel_hwtstamp_config *cfg) 5735 { 5736 struct bonding *bond = netdev_priv(dev); 5737 struct net_device *real_dev; 5738 int err; 5739 5740 real_dev = bond_option_active_slave_get_rcu(bond); 5741 if (!real_dev) 5742 return -EOPNOTSUPP; 5743 5744 err = generic_hwtstamp_get_lower(real_dev, cfg); 5745 if (err) 5746 return err; 5747 5748 return bond_set_phc_index_flag(cfg); 5749 } 5750 5751 static int bond_hwtstamp_set(struct net_device *dev, 5752 struct kernel_hwtstamp_config *cfg, 5753 struct netlink_ext_ack *extack) 5754 { 5755 struct bonding *bond = netdev_priv(dev); 5756 struct net_device *real_dev; 5757 int err; 5758 5759 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 5760 return -EOPNOTSUPP; 5761 5762 real_dev = bond_option_active_slave_get_rcu(bond); 5763 if (!real_dev) 5764 return -EOPNOTSUPP; 5765 5766 err = generic_hwtstamp_set_lower(real_dev, cfg, extack); 5767 if (err) 5768 return err; 5769 5770 return bond_set_phc_index_flag(cfg); 5771 } 5772 5773 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5774 struct ethtool_link_ksettings *cmd) 5775 { 5776 struct bonding *bond = netdev_priv(bond_dev); 5777 struct list_head *iter; 5778 struct slave *slave; 5779 u32 speed = 0; 5780 5781 cmd->base.duplex = DUPLEX_UNKNOWN; 5782 cmd->base.port = PORT_OTHER; 5783 5784 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5785 * do not need to check mode. Though link speed might not represent 5786 * the true receive or transmit bandwidth (not all modes are symmetric) 5787 * this is an accurate maximum. 5788 */ 5789 bond_for_each_slave(bond, slave, iter) { 5790 if (bond_slave_can_tx(slave)) { 5791 bond_update_speed_duplex(slave); 5792 if (slave->speed != SPEED_UNKNOWN) { 5793 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5794 speed = bond_mode_bcast_speed(slave, 5795 speed); 5796 else 5797 speed += slave->speed; 5798 } 5799 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5800 slave->duplex != DUPLEX_UNKNOWN) 5801 cmd->base.duplex = slave->duplex; 5802 } 5803 } 5804 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5805 5806 return 0; 5807 } 5808 5809 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5810 struct ethtool_drvinfo *drvinfo) 5811 { 5812 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5813 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5814 BOND_ABI_VERSION); 5815 } 5816 5817 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5818 struct kernel_ethtool_ts_info *info) 5819 { 5820 struct bonding *bond = netdev_priv(bond_dev); 5821 struct kernel_ethtool_ts_info ts_info; 5822 struct net_device *real_dev; 5823 bool sw_tx_support = false; 5824 struct list_head *iter; 5825 struct slave *slave; 5826 int ret = 0; 5827 5828 rcu_read_lock(); 5829 real_dev = bond_option_active_slave_get_rcu(bond); 5830 dev_hold(real_dev); 5831 rcu_read_unlock(); 5832 5833 if (real_dev) { 5834 ret = ethtool_get_ts_info_by_layer(real_dev, info); 5835 } else { 5836 /* Check if all slaves support software tx timestamping */ 5837 rcu_read_lock(); 5838 bond_for_each_slave_rcu(bond, slave, iter) { 5839 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info); 5840 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) { 5841 sw_tx_support = true; 5842 continue; 5843 } 5844 5845 sw_tx_support = false; 5846 break; 5847 } 5848 rcu_read_unlock(); 5849 } 5850 5851 if (sw_tx_support) 5852 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE; 5853 5854 dev_put(real_dev); 5855 return ret; 5856 } 5857 5858 static const struct ethtool_ops bond_ethtool_ops = { 5859 .get_drvinfo = bond_ethtool_get_drvinfo, 5860 .get_link = ethtool_op_get_link, 5861 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5862 .get_ts_info = bond_ethtool_get_ts_info, 5863 }; 5864 5865 static const struct net_device_ops bond_netdev_ops = { 5866 .ndo_init = bond_init, 5867 .ndo_uninit = bond_uninit, 5868 .ndo_open = bond_open, 5869 .ndo_stop = bond_close, 5870 .ndo_start_xmit = bond_start_xmit, 5871 .ndo_select_queue = bond_select_queue, 5872 .ndo_get_stats64 = bond_get_stats, 5873 .ndo_eth_ioctl = bond_eth_ioctl, 5874 .ndo_siocbond = bond_do_ioctl, 5875 .ndo_siocdevprivate = bond_siocdevprivate, 5876 .ndo_change_rx_flags = bond_change_rx_flags, 5877 .ndo_set_rx_mode = bond_set_rx_mode, 5878 .ndo_change_mtu = bond_change_mtu, 5879 .ndo_set_mac_address = bond_set_mac_address, 5880 .ndo_neigh_setup = bond_neigh_setup, 5881 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5882 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5883 #ifdef CONFIG_NET_POLL_CONTROLLER 5884 .ndo_netpoll_setup = bond_netpoll_setup, 5885 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5886 .ndo_poll_controller = bond_poll_controller, 5887 #endif 5888 .ndo_add_slave = bond_enslave, 5889 .ndo_del_slave = bond_release, 5890 .ndo_fix_features = bond_fix_features, 5891 .ndo_features_check = passthru_features_check, 5892 .ndo_get_xmit_slave = bond_xmit_get_slave, 5893 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5894 .ndo_bpf = bond_xdp, 5895 .ndo_xdp_xmit = bond_xdp_xmit, 5896 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5897 .ndo_hwtstamp_get = bond_hwtstamp_get, 5898 .ndo_hwtstamp_set = bond_hwtstamp_set, 5899 }; 5900 5901 static const struct device_type bond_type = { 5902 .name = "bond", 5903 }; 5904 5905 static void bond_destructor(struct net_device *bond_dev) 5906 { 5907 struct bonding *bond = netdev_priv(bond_dev); 5908 5909 if (bond->wq) 5910 destroy_workqueue(bond->wq); 5911 5912 free_percpu(bond->rr_tx_counter); 5913 } 5914 5915 void bond_setup(struct net_device *bond_dev) 5916 { 5917 struct bonding *bond = netdev_priv(bond_dev); 5918 5919 spin_lock_init(&bond->mode_lock); 5920 bond->params = bonding_defaults; 5921 5922 /* Initialize pointers */ 5923 bond->dev = bond_dev; 5924 5925 /* Initialize the device entry points */ 5926 ether_setup(bond_dev); 5927 bond_dev->max_mtu = ETH_MAX_MTU; 5928 bond_dev->netdev_ops = &bond_netdev_ops; 5929 bond_dev->ethtool_ops = &bond_ethtool_ops; 5930 5931 bond_dev->needs_free_netdev = true; 5932 bond_dev->priv_destructor = bond_destructor; 5933 5934 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5935 5936 /* Initialize the device options */ 5937 bond_dev->flags |= IFF_MASTER; 5938 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5939 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5940 5941 #ifdef CONFIG_XFRM_OFFLOAD 5942 /* set up xfrm device ops (only supported in active-backup right now) */ 5943 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5944 INIT_LIST_HEAD(&bond->ipsec_list); 5945 mutex_init(&bond->ipsec_lock); 5946 #endif /* CONFIG_XFRM_OFFLOAD */ 5947 5948 /* don't acquire bond device's netif_tx_lock when transmitting */ 5949 bond_dev->lltx = true; 5950 5951 /* Don't allow bond devices to change network namespaces. */ 5952 bond_dev->netns_immutable = true; 5953 5954 /* By default, we declare the bond to be fully 5955 * VLAN hardware accelerated capable. Special 5956 * care is taken in the various xmit functions 5957 * when there are slaves that are not hw accel 5958 * capable 5959 */ 5960 5961 bond_dev->hw_features = MASTER_UPPER_DEV_VLAN_FEATURES | 5962 NETIF_F_HW_VLAN_CTAG_RX | 5963 NETIF_F_HW_VLAN_CTAG_FILTER | 5964 NETIF_F_HW_VLAN_STAG_RX | 5965 NETIF_F_HW_VLAN_STAG_FILTER; 5966 5967 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5968 bond_dev->features |= bond_dev->hw_features; 5969 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5970 bond_dev->features |= NETIF_F_GSO_PARTIAL; 5971 #ifdef CONFIG_XFRM_OFFLOAD 5972 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5973 /* Only enable XFRM features if this is an active-backup config */ 5974 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5975 bond_dev->features |= BOND_XFRM_FEATURES; 5976 #endif /* CONFIG_XFRM_OFFLOAD */ 5977 } 5978 5979 /* Destroy a bonding device. 5980 * Must be under rtnl_lock when this function is called. 5981 */ 5982 static void bond_uninit(struct net_device *bond_dev) 5983 { 5984 struct bonding *bond = netdev_priv(bond_dev); 5985 struct list_head *iter; 5986 struct slave *slave; 5987 5988 bond_netpoll_cleanup(bond_dev); 5989 5990 /* Release the bonded slaves */ 5991 bond_for_each_slave(bond, slave, iter) 5992 __bond_release_one(bond_dev, slave->dev, true, true); 5993 netdev_info(bond_dev, "Released all slaves\n"); 5994 5995 #ifdef CONFIG_XFRM_OFFLOAD 5996 mutex_destroy(&bond->ipsec_lock); 5997 #endif /* CONFIG_XFRM_OFFLOAD */ 5998 5999 bond_set_slave_arr(bond, NULL, NULL); 6000 6001 list_del_rcu(&bond->bond_list); 6002 6003 bond_debug_unregister(bond); 6004 } 6005 6006 /*------------------------- Module initialization ---------------------------*/ 6007 6008 static int __init bond_check_params(struct bond_params *params) 6009 { 6010 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 6011 struct bond_opt_value newval; 6012 const struct bond_opt_value *valptr; 6013 int arp_all_targets_value = 0; 6014 u16 ad_actor_sys_prio = 0; 6015 u16 ad_user_port_key = 0; 6016 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 6017 int arp_ip_count; 6018 int bond_mode = BOND_MODE_ROUNDROBIN; 6019 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 6020 int lacp_fast = 0; 6021 int tlb_dynamic_lb; 6022 6023 /* Convert string parameters. */ 6024 if (mode) { 6025 bond_opt_initstr(&newval, mode); 6026 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 6027 if (!valptr) { 6028 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 6029 return -EINVAL; 6030 } 6031 bond_mode = valptr->value; 6032 } 6033 6034 if (xmit_hash_policy) { 6035 if (bond_mode == BOND_MODE_ROUNDROBIN || 6036 bond_mode == BOND_MODE_ACTIVEBACKUP || 6037 bond_mode == BOND_MODE_BROADCAST) { 6038 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 6039 bond_mode_name(bond_mode)); 6040 } else { 6041 bond_opt_initstr(&newval, xmit_hash_policy); 6042 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 6043 &newval); 6044 if (!valptr) { 6045 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 6046 xmit_hash_policy); 6047 return -EINVAL; 6048 } 6049 xmit_hashtype = valptr->value; 6050 } 6051 } 6052 6053 if (lacp_rate) { 6054 if (bond_mode != BOND_MODE_8023AD) { 6055 pr_info("lacp_rate param is irrelevant in mode %s\n", 6056 bond_mode_name(bond_mode)); 6057 } else { 6058 bond_opt_initstr(&newval, lacp_rate); 6059 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 6060 &newval); 6061 if (!valptr) { 6062 pr_err("Error: Invalid lacp rate \"%s\"\n", 6063 lacp_rate); 6064 return -EINVAL; 6065 } 6066 lacp_fast = valptr->value; 6067 } 6068 } 6069 6070 if (ad_select) { 6071 bond_opt_initstr(&newval, ad_select); 6072 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 6073 &newval); 6074 if (!valptr) { 6075 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 6076 return -EINVAL; 6077 } 6078 params->ad_select = valptr->value; 6079 if (bond_mode != BOND_MODE_8023AD) 6080 pr_warn("ad_select param only affects 802.3ad mode\n"); 6081 } else { 6082 params->ad_select = BOND_AD_STABLE; 6083 } 6084 6085 if (max_bonds < 0) { 6086 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 6087 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 6088 max_bonds = BOND_DEFAULT_MAX_BONDS; 6089 } 6090 6091 if (miimon < 0) { 6092 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6093 miimon, INT_MAX); 6094 miimon = 0; 6095 } 6096 6097 if (updelay < 0) { 6098 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6099 updelay, INT_MAX); 6100 updelay = 0; 6101 } 6102 6103 if (downdelay < 0) { 6104 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6105 downdelay, INT_MAX); 6106 downdelay = 0; 6107 } 6108 6109 if (use_carrier != 1) { 6110 pr_err("Error: invalid use_carrier parameter (%d)\n", 6111 use_carrier); 6112 return -EINVAL; 6113 } 6114 6115 if (num_peer_notif < 0 || num_peer_notif > 255) { 6116 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 6117 num_peer_notif); 6118 num_peer_notif = 1; 6119 } 6120 6121 /* reset values for 802.3ad/TLB/ALB */ 6122 if (!bond_mode_uses_arp(bond_mode)) { 6123 if (!miimon) { 6124 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"); 6125 pr_warn("Forcing miimon to 100msec\n"); 6126 miimon = BOND_DEFAULT_MIIMON; 6127 } 6128 } 6129 6130 if (tx_queues < 1 || tx_queues > 255) { 6131 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 6132 tx_queues, BOND_DEFAULT_TX_QUEUES); 6133 tx_queues = BOND_DEFAULT_TX_QUEUES; 6134 } 6135 6136 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 6137 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 6138 all_slaves_active); 6139 all_slaves_active = 0; 6140 } 6141 6142 if (resend_igmp < 0 || resend_igmp > 255) { 6143 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6144 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6145 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6146 } 6147 6148 bond_opt_initval(&newval, packets_per_slave); 6149 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6150 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6151 packets_per_slave, USHRT_MAX); 6152 packets_per_slave = 1; 6153 } 6154 6155 if (bond_mode == BOND_MODE_ALB) { 6156 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", 6157 updelay); 6158 } 6159 6160 if (!miimon) { 6161 if (updelay || downdelay) { 6162 /* just warn the user the up/down delay will have 6163 * no effect since miimon is zero... 6164 */ 6165 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", 6166 updelay, downdelay); 6167 } 6168 } else { 6169 /* don't allow arp monitoring */ 6170 if (arp_interval) { 6171 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6172 miimon, arp_interval); 6173 arp_interval = 0; 6174 } 6175 6176 if ((updelay % miimon) != 0) { 6177 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6178 updelay, miimon, (updelay / miimon) * miimon); 6179 } 6180 6181 updelay /= miimon; 6182 6183 if ((downdelay % miimon) != 0) { 6184 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6185 downdelay, miimon, 6186 (downdelay / miimon) * miimon); 6187 } 6188 6189 downdelay /= miimon; 6190 } 6191 6192 if (arp_interval < 0) { 6193 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6194 arp_interval, INT_MAX); 6195 arp_interval = 0; 6196 } 6197 6198 for (arp_ip_count = 0, i = 0; 6199 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6200 __be32 ip; 6201 6202 /* not a complete check, but good enough to catch mistakes */ 6203 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6204 !bond_is_ip_target_ok(ip)) { 6205 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6206 arp_ip_target[i]); 6207 arp_interval = 0; 6208 } else { 6209 if (bond_get_targets_ip(arp_target, ip) == -1) 6210 arp_target[arp_ip_count++] = ip; 6211 else 6212 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6213 &ip); 6214 } 6215 } 6216 6217 if (arp_interval && !arp_ip_count) { 6218 /* don't allow arping if no arp_ip_target given... */ 6219 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6220 arp_interval); 6221 arp_interval = 0; 6222 } 6223 6224 if (arp_validate) { 6225 if (!arp_interval) { 6226 pr_err("arp_validate requires arp_interval\n"); 6227 return -EINVAL; 6228 } 6229 6230 bond_opt_initstr(&newval, arp_validate); 6231 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6232 &newval); 6233 if (!valptr) { 6234 pr_err("Error: invalid arp_validate \"%s\"\n", 6235 arp_validate); 6236 return -EINVAL; 6237 } 6238 arp_validate_value = valptr->value; 6239 } else { 6240 arp_validate_value = 0; 6241 } 6242 6243 if (arp_all_targets) { 6244 bond_opt_initstr(&newval, arp_all_targets); 6245 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6246 &newval); 6247 if (!valptr) { 6248 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6249 arp_all_targets); 6250 arp_all_targets_value = 0; 6251 } else { 6252 arp_all_targets_value = valptr->value; 6253 } 6254 } 6255 6256 if (miimon) { 6257 pr_info("MII link monitoring set to %d ms\n", miimon); 6258 } else if (arp_interval) { 6259 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6260 arp_validate_value); 6261 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6262 arp_interval, valptr->string, arp_ip_count); 6263 6264 for (i = 0; i < arp_ip_count; i++) 6265 pr_cont(" %s", arp_ip_target[i]); 6266 6267 pr_cont("\n"); 6268 6269 } else if (max_bonds) { 6270 /* miimon and arp_interval not set, we need one so things 6271 * work as expected, see bonding.txt for details 6272 */ 6273 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"); 6274 } 6275 6276 if (primary && !bond_mode_uses_primary(bond_mode)) { 6277 /* currently, using a primary only makes sense 6278 * in active backup, TLB or ALB modes 6279 */ 6280 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6281 primary, bond_mode_name(bond_mode)); 6282 primary = NULL; 6283 } 6284 6285 if (primary && primary_reselect) { 6286 bond_opt_initstr(&newval, primary_reselect); 6287 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6288 &newval); 6289 if (!valptr) { 6290 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6291 primary_reselect); 6292 return -EINVAL; 6293 } 6294 primary_reselect_value = valptr->value; 6295 } else { 6296 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6297 } 6298 6299 if (fail_over_mac) { 6300 bond_opt_initstr(&newval, fail_over_mac); 6301 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6302 &newval); 6303 if (!valptr) { 6304 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6305 fail_over_mac); 6306 return -EINVAL; 6307 } 6308 fail_over_mac_value = valptr->value; 6309 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6310 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6311 } else { 6312 fail_over_mac_value = BOND_FOM_NONE; 6313 } 6314 6315 bond_opt_initstr(&newval, "default"); 6316 valptr = bond_opt_parse( 6317 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6318 &newval); 6319 if (!valptr) { 6320 pr_err("Error: No ad_actor_sys_prio default value"); 6321 return -EINVAL; 6322 } 6323 ad_actor_sys_prio = valptr->value; 6324 6325 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6326 &newval); 6327 if (!valptr) { 6328 pr_err("Error: No ad_user_port_key default value"); 6329 return -EINVAL; 6330 } 6331 ad_user_port_key = valptr->value; 6332 6333 bond_opt_initstr(&newval, "default"); 6334 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6335 if (!valptr) { 6336 pr_err("Error: No tlb_dynamic_lb default value"); 6337 return -EINVAL; 6338 } 6339 tlb_dynamic_lb = valptr->value; 6340 6341 if (lp_interval == 0) { 6342 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6343 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6344 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6345 } 6346 6347 /* fill params struct with the proper values */ 6348 params->mode = bond_mode; 6349 params->xmit_policy = xmit_hashtype; 6350 params->miimon = miimon; 6351 params->num_peer_notif = num_peer_notif; 6352 params->arp_interval = arp_interval; 6353 params->arp_validate = arp_validate_value; 6354 params->arp_all_targets = arp_all_targets_value; 6355 params->missed_max = 2; 6356 params->updelay = updelay; 6357 params->downdelay = downdelay; 6358 params->peer_notif_delay = 0; 6359 params->lacp_active = 1; 6360 params->lacp_fast = lacp_fast; 6361 params->primary[0] = 0; 6362 params->primary_reselect = primary_reselect_value; 6363 params->fail_over_mac = fail_over_mac_value; 6364 params->tx_queues = tx_queues; 6365 params->all_slaves_active = all_slaves_active; 6366 params->resend_igmp = resend_igmp; 6367 params->min_links = min_links; 6368 params->lp_interval = lp_interval; 6369 params->packets_per_slave = packets_per_slave; 6370 params->tlb_dynamic_lb = tlb_dynamic_lb; 6371 params->ad_actor_sys_prio = ad_actor_sys_prio; 6372 eth_zero_addr(params->ad_actor_system); 6373 params->ad_user_port_key = ad_user_port_key; 6374 params->coupled_control = 1; 6375 params->broadcast_neighbor = 0; 6376 if (packets_per_slave > 0) { 6377 params->reciprocal_packets_per_slave = 6378 reciprocal_value(packets_per_slave); 6379 } else { 6380 /* reciprocal_packets_per_slave is unused if 6381 * packets_per_slave is 0 or 1, just initialize it 6382 */ 6383 params->reciprocal_packets_per_slave = 6384 (struct reciprocal_value) { 0 }; 6385 } 6386 6387 if (primary) 6388 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6389 6390 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6391 #if IS_ENABLED(CONFIG_IPV6) 6392 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6393 #endif 6394 6395 return 0; 6396 } 6397 6398 /* Called from registration process */ 6399 static int bond_init(struct net_device *bond_dev) 6400 { 6401 struct bonding *bond = netdev_priv(bond_dev); 6402 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6403 6404 netdev_dbg(bond_dev, "Begin bond_init\n"); 6405 6406 bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM, 6407 bond_dev->name); 6408 if (!bond->wq) 6409 return -ENOMEM; 6410 6411 bond->notifier_ctx = false; 6412 6413 spin_lock_init(&bond->stats_lock); 6414 netdev_lockdep_set_classes(bond_dev); 6415 6416 list_add_tail_rcu(&bond->bond_list, &bn->dev_list); 6417 6418 bond_prepare_sysfs_group(bond); 6419 6420 bond_debug_register(bond); 6421 6422 /* Ensure valid dev_addr */ 6423 if (is_zero_ether_addr(bond_dev->dev_addr) && 6424 bond_dev->addr_assign_type == NET_ADDR_PERM) 6425 eth_hw_addr_random(bond_dev); 6426 6427 return 0; 6428 } 6429 6430 unsigned int bond_get_num_tx_queues(void) 6431 { 6432 return tx_queues; 6433 } 6434 6435 /* Create a new bond based on the specified name and bonding parameters. 6436 * If name is NULL, obtain a suitable "bond%d" name for us. 6437 * Caller must NOT hold rtnl_lock; we need to release it here before we 6438 * set up our sysfs entries. 6439 */ 6440 int bond_create(struct net *net, const char *name) 6441 { 6442 struct net_device *bond_dev; 6443 struct bonding *bond; 6444 int res = -ENOMEM; 6445 6446 rtnl_lock(); 6447 6448 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6449 name ? name : "bond%d", NET_NAME_UNKNOWN, 6450 bond_setup, tx_queues); 6451 if (!bond_dev) 6452 goto out; 6453 6454 bond = netdev_priv(bond_dev); 6455 dev_net_set(bond_dev, net); 6456 bond_dev->rtnl_link_ops = &bond_link_ops; 6457 6458 res = register_netdevice(bond_dev); 6459 if (res < 0) { 6460 free_netdev(bond_dev); 6461 goto out; 6462 } 6463 6464 netif_carrier_off(bond_dev); 6465 6466 bond_work_init_all(bond); 6467 6468 out: 6469 rtnl_unlock(); 6470 return res; 6471 } 6472 6473 static int __net_init bond_net_init(struct net *net) 6474 { 6475 struct bond_net *bn = net_generic(net, bond_net_id); 6476 6477 bn->net = net; 6478 INIT_LIST_HEAD(&bn->dev_list); 6479 6480 bond_create_proc_dir(bn); 6481 bond_create_sysfs(bn); 6482 6483 return 0; 6484 } 6485 6486 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters 6487 * race condition in bond unloading") we need to remove sysfs files 6488 * before we remove our devices (done later in bond_net_exit_rtnl()) 6489 */ 6490 static void __net_exit bond_net_pre_exit(struct net *net) 6491 { 6492 struct bond_net *bn = net_generic(net, bond_net_id); 6493 6494 bond_destroy_sysfs(bn); 6495 } 6496 6497 static void __net_exit bond_net_exit_rtnl(struct net *net, 6498 struct list_head *dev_kill_list) 6499 { 6500 struct bond_net *bn = net_generic(net, bond_net_id); 6501 struct bonding *bond, *tmp_bond; 6502 6503 /* Kill off any bonds created after unregistering bond rtnl ops */ 6504 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6505 unregister_netdevice_queue(bond->dev, dev_kill_list); 6506 } 6507 6508 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory 6509 * only after all bonds are gone") bond_destroy_proc_dir() is called 6510 * after bond_net_exit_rtnl() has completed. 6511 */ 6512 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6513 { 6514 struct bond_net *bn; 6515 struct net *net; 6516 6517 list_for_each_entry(net, net_list, exit_list) { 6518 bn = net_generic(net, bond_net_id); 6519 bond_destroy_proc_dir(bn); 6520 } 6521 } 6522 6523 static struct pernet_operations bond_net_ops = { 6524 .init = bond_net_init, 6525 .pre_exit = bond_net_pre_exit, 6526 .exit_rtnl = bond_net_exit_rtnl, 6527 .exit_batch = bond_net_exit_batch, 6528 .id = &bond_net_id, 6529 .size = sizeof(struct bond_net), 6530 }; 6531 6532 static int __init bonding_init(void) 6533 { 6534 int i; 6535 int res; 6536 6537 res = bond_check_params(&bonding_defaults); 6538 if (res) 6539 goto out; 6540 6541 bond_create_debugfs(); 6542 6543 res = register_pernet_subsys(&bond_net_ops); 6544 if (res) 6545 goto err_net_ops; 6546 6547 res = bond_netlink_init(); 6548 if (res) 6549 goto err_link; 6550 6551 for (i = 0; i < max_bonds; i++) { 6552 res = bond_create(&init_net, NULL); 6553 if (res) 6554 goto err; 6555 } 6556 6557 skb_flow_dissector_init(&flow_keys_bonding, 6558 flow_keys_bonding_keys, 6559 ARRAY_SIZE(flow_keys_bonding_keys)); 6560 6561 register_netdevice_notifier(&bond_netdev_notifier); 6562 out: 6563 return res; 6564 err: 6565 bond_netlink_fini(); 6566 err_link: 6567 unregister_pernet_subsys(&bond_net_ops); 6568 err_net_ops: 6569 bond_destroy_debugfs(); 6570 goto out; 6571 6572 } 6573 6574 static void __exit bonding_exit(void) 6575 { 6576 unregister_netdevice_notifier(&bond_netdev_notifier); 6577 6578 bond_netlink_fini(); 6579 unregister_pernet_subsys(&bond_net_ops); 6580 6581 bond_destroy_debugfs(); 6582 6583 #ifdef CONFIG_NET_POLL_CONTROLLER 6584 /* Make sure we don't have an imbalance on our netpoll blocking */ 6585 WARN_ON(atomic_read(&netpoll_block_tx)); 6586 #endif 6587 } 6588 6589 module_init(bonding_init); 6590 module_exit(bonding_exit); 6591 MODULE_LICENSE("GPL"); 6592 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6593 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6594 MODULE_IMPORT_NS("NETDEV_INTERNAL"); 6595