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/xdp.h> 94 95 #include "bonding_priv.h" 96 97 /*---------------------------- Module parameters ----------------------------*/ 98 99 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 100 101 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 102 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 103 static int num_peer_notif = 1; 104 static int miimon; 105 static int updelay; 106 static int downdelay; 107 static int use_carrier = 1; 108 static char *mode; 109 static char *primary; 110 static char *primary_reselect; 111 static char *lacp_rate; 112 static int min_links; 113 static char *ad_select; 114 static char *xmit_hash_policy; 115 static int arp_interval; 116 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 117 static char *arp_validate; 118 static char *arp_all_targets; 119 static char *fail_over_mac; 120 static int all_slaves_active; 121 static struct bond_params bonding_defaults; 122 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 123 static int packets_per_slave = 1; 124 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 125 126 module_param(max_bonds, int, 0); 127 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 128 module_param(tx_queues, int, 0); 129 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 130 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 131 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 132 "failover event (alias of num_unsol_na)"); 133 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 134 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 135 "failover event (alias of num_grat_arp)"); 136 module_param(miimon, int, 0); 137 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 138 module_param(updelay, int, 0); 139 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 140 module_param(downdelay, int, 0); 141 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 142 "in milliseconds"); 143 module_param(use_carrier, int, 0); 144 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 145 "0 for off, 1 for on (default)"); 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 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 215 { 216 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 217 .offset = offsetof(struct flow_keys, control), 218 }, 219 { 220 .key_id = FLOW_DISSECTOR_KEY_BASIC, 221 .offset = offsetof(struct flow_keys, basic), 222 }, 223 { 224 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 225 .offset = offsetof(struct flow_keys, addrs.v4addrs), 226 }, 227 { 228 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 229 .offset = offsetof(struct flow_keys, addrs.v6addrs), 230 }, 231 { 232 .key_id = FLOW_DISSECTOR_KEY_TIPC, 233 .offset = offsetof(struct flow_keys, addrs.tipckey), 234 }, 235 { 236 .key_id = FLOW_DISSECTOR_KEY_PORTS, 237 .offset = offsetof(struct flow_keys, ports), 238 }, 239 { 240 .key_id = FLOW_DISSECTOR_KEY_ICMP, 241 .offset = offsetof(struct flow_keys, icmp), 242 }, 243 { 244 .key_id = FLOW_DISSECTOR_KEY_VLAN, 245 .offset = offsetof(struct flow_keys, vlan), 246 }, 247 { 248 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 249 .offset = offsetof(struct flow_keys, tags), 250 }, 251 { 252 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 253 .offset = offsetof(struct flow_keys, keyid), 254 }, 255 }; 256 257 static struct flow_dissector flow_keys_bonding __read_mostly; 258 259 /*-------------------------- Forward declarations ---------------------------*/ 260 261 static int bond_init(struct net_device *bond_dev); 262 static void bond_uninit(struct net_device *bond_dev); 263 static void bond_get_stats(struct net_device *bond_dev, 264 struct rtnl_link_stats64 *stats); 265 static void bond_slave_arr_handler(struct work_struct *work); 266 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 267 int mod); 268 static void bond_netdev_notify_work(struct work_struct *work); 269 270 /*---------------------------- General routines -----------------------------*/ 271 272 const char *bond_mode_name(int mode) 273 { 274 static const char *names[] = { 275 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 276 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 277 [BOND_MODE_XOR] = "load balancing (xor)", 278 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 279 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 280 [BOND_MODE_TLB] = "transmit load balancing", 281 [BOND_MODE_ALB] = "adaptive load balancing", 282 }; 283 284 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 285 return "unknown"; 286 287 return names[mode]; 288 } 289 290 /** 291 * bond_dev_queue_xmit - Prepare skb for xmit. 292 * 293 * @bond: bond device that got this skb for tx. 294 * @skb: hw accel VLAN tagged skb to transmit 295 * @slave_dev: slave that is supposed to xmit this skbuff 296 */ 297 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 298 struct net_device *slave_dev) 299 { 300 skb->dev = slave_dev; 301 302 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 303 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 304 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 305 306 if (unlikely(netpoll_tx_running(bond->dev))) 307 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 308 309 return dev_queue_xmit(skb); 310 } 311 312 static bool bond_sk_check(struct bonding *bond) 313 { 314 switch (BOND_MODE(bond)) { 315 case BOND_MODE_8023AD: 316 case BOND_MODE_XOR: 317 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 318 return true; 319 fallthrough; 320 default: 321 return false; 322 } 323 } 324 325 static bool bond_xdp_check(struct bonding *bond) 326 { 327 switch (BOND_MODE(bond)) { 328 case BOND_MODE_ROUNDROBIN: 329 case BOND_MODE_ACTIVEBACKUP: 330 return true; 331 case BOND_MODE_8023AD: 332 case BOND_MODE_XOR: 333 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 334 * payload is not in the packet due to hardware offload. 335 */ 336 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 337 return true; 338 fallthrough; 339 default: 340 return false; 341 } 342 } 343 344 /*---------------------------------- VLAN -----------------------------------*/ 345 346 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 347 * We don't protect the slave list iteration with a lock because: 348 * a. This operation is performed in IOCTL context, 349 * b. The operation is protected by the RTNL semaphore in the 8021q code, 350 * c. Holding a lock with BH disabled while directly calling a base driver 351 * entry point is generally a BAD idea. 352 * 353 * The design of synchronization/protection for this operation in the 8021q 354 * module is good for one or more VLAN devices over a single physical device 355 * and cannot be extended for a teaming solution like bonding, so there is a 356 * potential race condition here where a net device from the vlan group might 357 * be referenced (either by a base driver or the 8021q code) while it is being 358 * removed from the system. However, it turns out we're not making matters 359 * worse, and if it works for regular VLAN usage it will work here too. 360 */ 361 362 /** 363 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 364 * @bond_dev: bonding net device that got called 365 * @proto: network protocol ID 366 * @vid: vlan id being added 367 */ 368 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 369 __be16 proto, u16 vid) 370 { 371 struct bonding *bond = netdev_priv(bond_dev); 372 struct slave *slave, *rollback_slave; 373 struct list_head *iter; 374 int res; 375 376 bond_for_each_slave(bond, slave, iter) { 377 res = vlan_vid_add(slave->dev, proto, vid); 378 if (res) 379 goto unwind; 380 } 381 382 return 0; 383 384 unwind: 385 /* unwind to the slave that failed */ 386 bond_for_each_slave(bond, rollback_slave, iter) { 387 if (rollback_slave == slave) 388 break; 389 390 vlan_vid_del(rollback_slave->dev, proto, vid); 391 } 392 393 return res; 394 } 395 396 /** 397 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 398 * @bond_dev: bonding net device that got called 399 * @proto: network protocol ID 400 * @vid: vlan id being removed 401 */ 402 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 403 __be16 proto, u16 vid) 404 { 405 struct bonding *bond = netdev_priv(bond_dev); 406 struct list_head *iter; 407 struct slave *slave; 408 409 bond_for_each_slave(bond, slave, iter) 410 vlan_vid_del(slave->dev, proto, vid); 411 412 if (bond_is_lb(bond)) 413 bond_alb_clear_vlan(bond, vid); 414 415 return 0; 416 } 417 418 /*---------------------------------- XFRM -----------------------------------*/ 419 420 #ifdef CONFIG_XFRM_OFFLOAD 421 /** 422 * bond_ipsec_dev - Get active device for IPsec offload 423 * @xs: pointer to transformer state struct 424 * 425 * Context: caller must hold rcu_read_lock. 426 * 427 * Return: the device for ipsec offload, or NULL if not exist. 428 **/ 429 static struct net_device *bond_ipsec_dev(struct xfrm_state *xs) 430 { 431 struct net_device *bond_dev = xs->xso.dev; 432 struct bonding *bond; 433 struct slave *slave; 434 435 if (!bond_dev) 436 return NULL; 437 438 bond = netdev_priv(bond_dev); 439 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 440 return NULL; 441 442 slave = rcu_dereference(bond->curr_active_slave); 443 if (!slave) 444 return NULL; 445 446 if (!xs->xso.real_dev) 447 return NULL; 448 449 if (xs->xso.real_dev != slave->dev) 450 pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n", 451 bond_dev->name, slave->dev->name, xs->xso.real_dev->name); 452 453 return slave->dev; 454 } 455 456 /** 457 * bond_ipsec_add_sa - program device with a security association 458 * @xs: pointer to transformer state struct 459 * @extack: extack point to fill failure reason 460 **/ 461 static int bond_ipsec_add_sa(struct xfrm_state *xs, 462 struct netlink_ext_ack *extack) 463 { 464 struct net_device *bond_dev = xs->xso.dev; 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 xs->xso.real_dev = real_dev; 501 err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack); 502 if (!err) { 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 ipsec->xs->xso.real_dev = real_dev; 545 if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) { 546 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__); 547 ipsec->xs->xso.real_dev = NULL; 548 } 549 } 550 out: 551 mutex_unlock(&bond->ipsec_lock); 552 } 553 554 /** 555 * bond_ipsec_del_sa - clear out this specific SA 556 * @xs: pointer to transformer state struct 557 **/ 558 static void bond_ipsec_del_sa(struct xfrm_state *xs) 559 { 560 struct net_device *bond_dev = xs->xso.dev; 561 struct net_device *real_dev; 562 netdevice_tracker tracker; 563 struct bond_ipsec *ipsec; 564 struct bonding *bond; 565 struct slave *slave; 566 567 if (!bond_dev) 568 return; 569 570 rcu_read_lock(); 571 bond = netdev_priv(bond_dev); 572 slave = rcu_dereference(bond->curr_active_slave); 573 real_dev = slave ? slave->dev : NULL; 574 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 575 rcu_read_unlock(); 576 577 if (!slave) 578 goto out; 579 580 if (!xs->xso.real_dev) 581 goto out; 582 583 WARN_ON(xs->xso.real_dev != real_dev); 584 585 if (!real_dev->xfrmdev_ops || 586 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 587 netif_is_bond_master(real_dev)) { 588 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__); 589 goto out; 590 } 591 592 real_dev->xfrmdev_ops->xdo_dev_state_delete(xs); 593 out: 594 netdev_put(real_dev, &tracker); 595 mutex_lock(&bond->ipsec_lock); 596 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 597 if (ipsec->xs == xs) { 598 list_del(&ipsec->list); 599 kfree(ipsec); 600 break; 601 } 602 } 603 mutex_unlock(&bond->ipsec_lock); 604 } 605 606 static void bond_ipsec_del_sa_all(struct bonding *bond) 607 { 608 struct net_device *bond_dev = bond->dev; 609 struct net_device *real_dev; 610 struct bond_ipsec *ipsec; 611 struct slave *slave; 612 613 slave = rtnl_dereference(bond->curr_active_slave); 614 real_dev = slave ? slave->dev : NULL; 615 if (!real_dev) 616 return; 617 618 mutex_lock(&bond->ipsec_lock); 619 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 620 if (!ipsec->xs->xso.real_dev) 621 continue; 622 623 if (!real_dev->xfrmdev_ops || 624 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 625 netif_is_bond_master(real_dev)) { 626 slave_warn(bond_dev, real_dev, 627 "%s: no slave xdo_dev_state_delete\n", 628 __func__); 629 } else { 630 real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 631 if (real_dev->xfrmdev_ops->xdo_dev_state_free) 632 real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs); 633 } 634 } 635 mutex_unlock(&bond->ipsec_lock); 636 } 637 638 static void bond_ipsec_free_sa(struct xfrm_state *xs) 639 { 640 struct net_device *bond_dev = xs->xso.dev; 641 struct net_device *real_dev; 642 netdevice_tracker tracker; 643 struct bonding *bond; 644 struct slave *slave; 645 646 if (!bond_dev) 647 return; 648 649 rcu_read_lock(); 650 bond = netdev_priv(bond_dev); 651 slave = rcu_dereference(bond->curr_active_slave); 652 real_dev = slave ? slave->dev : NULL; 653 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 654 rcu_read_unlock(); 655 656 if (!slave) 657 goto out; 658 659 if (!xs->xso.real_dev) 660 goto out; 661 662 WARN_ON(xs->xso.real_dev != real_dev); 663 664 if (real_dev && real_dev->xfrmdev_ops && 665 real_dev->xfrmdev_ops->xdo_dev_state_free) 666 real_dev->xfrmdev_ops->xdo_dev_state_free(xs); 667 out: 668 netdev_put(real_dev, &tracker); 669 } 670 671 /** 672 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 673 * @skb: current data packet 674 * @xs: pointer to transformer state struct 675 **/ 676 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 677 { 678 struct net_device *real_dev; 679 bool ok = false; 680 681 rcu_read_lock(); 682 real_dev = bond_ipsec_dev(xs); 683 if (!real_dev) 684 goto out; 685 686 if (!real_dev->xfrmdev_ops || 687 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 688 netif_is_bond_master(real_dev)) 689 goto out; 690 691 ok = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 692 out: 693 rcu_read_unlock(); 694 return ok; 695 } 696 697 /** 698 * bond_advance_esn_state - ESN support for IPSec HW offload 699 * @xs: pointer to transformer state struct 700 **/ 701 static void bond_advance_esn_state(struct xfrm_state *xs) 702 { 703 struct net_device *real_dev; 704 705 rcu_read_lock(); 706 real_dev = bond_ipsec_dev(xs); 707 if (!real_dev) 708 goto out; 709 710 if (!real_dev->xfrmdev_ops || 711 !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 712 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name); 713 goto out; 714 } 715 716 real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs); 717 out: 718 rcu_read_unlock(); 719 } 720 721 /** 722 * bond_xfrm_update_stats - Update xfrm state 723 * @xs: pointer to transformer state struct 724 **/ 725 static void bond_xfrm_update_stats(struct xfrm_state *xs) 726 { 727 struct net_device *real_dev; 728 729 rcu_read_lock(); 730 real_dev = bond_ipsec_dev(xs); 731 if (!real_dev) 732 goto out; 733 734 if (!real_dev->xfrmdev_ops || 735 !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) { 736 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name); 737 goto out; 738 } 739 740 real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs); 741 out: 742 rcu_read_unlock(); 743 } 744 745 static const struct xfrmdev_ops bond_xfrmdev_ops = { 746 .xdo_dev_state_add = bond_ipsec_add_sa, 747 .xdo_dev_state_delete = bond_ipsec_del_sa, 748 .xdo_dev_state_free = bond_ipsec_free_sa, 749 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 750 .xdo_dev_state_advance_esn = bond_advance_esn_state, 751 .xdo_dev_state_update_stats = bond_xfrm_update_stats, 752 }; 753 #endif /* CONFIG_XFRM_OFFLOAD */ 754 755 /*------------------------------- Link status -------------------------------*/ 756 757 /* Set the carrier state for the master according to the state of its 758 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 759 * do special 802.3ad magic. 760 * 761 * Returns zero if carrier state does not change, nonzero if it does. 762 */ 763 int bond_set_carrier(struct bonding *bond) 764 { 765 struct list_head *iter; 766 struct slave *slave; 767 768 if (!bond_has_slaves(bond)) 769 goto down; 770 771 if (BOND_MODE(bond) == BOND_MODE_8023AD) 772 return bond_3ad_set_carrier(bond); 773 774 bond_for_each_slave(bond, slave, iter) { 775 if (slave->link == BOND_LINK_UP) { 776 if (!netif_carrier_ok(bond->dev)) { 777 netif_carrier_on(bond->dev); 778 return 1; 779 } 780 return 0; 781 } 782 } 783 784 down: 785 if (netif_carrier_ok(bond->dev)) { 786 netif_carrier_off(bond->dev); 787 return 1; 788 } 789 return 0; 790 } 791 792 /* Get link speed and duplex from the slave's base driver 793 * using ethtool. If for some reason the call fails or the 794 * values are invalid, set speed and duplex to -1, 795 * and return. Return 1 if speed or duplex settings are 796 * UNKNOWN; 0 otherwise. 797 */ 798 static int bond_update_speed_duplex(struct slave *slave) 799 { 800 struct net_device *slave_dev = slave->dev; 801 struct ethtool_link_ksettings ecmd; 802 int res; 803 804 slave->speed = SPEED_UNKNOWN; 805 slave->duplex = DUPLEX_UNKNOWN; 806 807 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 808 if (res < 0) 809 return 1; 810 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 811 return 1; 812 switch (ecmd.base.duplex) { 813 case DUPLEX_FULL: 814 case DUPLEX_HALF: 815 break; 816 default: 817 return 1; 818 } 819 820 slave->speed = ecmd.base.speed; 821 slave->duplex = ecmd.base.duplex; 822 823 return 0; 824 } 825 826 const char *bond_slave_link_status(s8 link) 827 { 828 switch (link) { 829 case BOND_LINK_UP: 830 return "up"; 831 case BOND_LINK_FAIL: 832 return "going down"; 833 case BOND_LINK_DOWN: 834 return "down"; 835 case BOND_LINK_BACK: 836 return "going back"; 837 default: 838 return "unknown"; 839 } 840 } 841 842 /* if <dev> supports MII link status reporting, check its link status. 843 * 844 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 845 * depending upon the setting of the use_carrier parameter. 846 * 847 * Return either BMSR_LSTATUS, meaning that the link is up (or we 848 * can't tell and just pretend it is), or 0, meaning that the link is 849 * down. 850 * 851 * If reporting is non-zero, instead of faking link up, return -1 if 852 * both ETHTOOL and MII ioctls fail (meaning the device does not 853 * support them). If use_carrier is set, return whatever it says. 854 * It'd be nice if there was a good way to tell if a driver supports 855 * netif_carrier, but there really isn't. 856 */ 857 static int bond_check_dev_link(struct bonding *bond, 858 struct net_device *slave_dev, int reporting) 859 { 860 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 861 int (*ioctl)(struct net_device *, struct ifreq *, int); 862 struct ifreq ifr; 863 struct mii_ioctl_data *mii; 864 865 if (!reporting && !netif_running(slave_dev)) 866 return 0; 867 868 if (bond->params.use_carrier) 869 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 870 871 /* Try to get link status using Ethtool first. */ 872 if (slave_dev->ethtool_ops->get_link) 873 return slave_dev->ethtool_ops->get_link(slave_dev) ? 874 BMSR_LSTATUS : 0; 875 876 /* Ethtool can't be used, fallback to MII ioctls. */ 877 ioctl = slave_ops->ndo_eth_ioctl; 878 if (ioctl) { 879 /* TODO: set pointer to correct ioctl on a per team member 880 * bases to make this more efficient. that is, once 881 * we determine the correct ioctl, we will always 882 * call it and not the others for that team 883 * member. 884 */ 885 886 /* We cannot assume that SIOCGMIIPHY will also read a 887 * register; not all network drivers (e.g., e100) 888 * support that. 889 */ 890 891 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 892 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 893 mii = if_mii(&ifr); 894 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 895 mii->reg_num = MII_BMSR; 896 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 897 return mii->val_out & BMSR_LSTATUS; 898 } 899 } 900 901 /* If reporting, report that either there's no ndo_eth_ioctl, 902 * or both SIOCGMIIREG and get_link failed (meaning that we 903 * cannot report link status). If not reporting, pretend 904 * we're ok. 905 */ 906 return reporting ? -1 : BMSR_LSTATUS; 907 } 908 909 /*----------------------------- Multicast list ------------------------------*/ 910 911 /* Push the promiscuity flag down to appropriate slaves */ 912 static int bond_set_promiscuity(struct bonding *bond, int inc) 913 { 914 struct list_head *iter; 915 int err = 0; 916 917 if (bond_uses_primary(bond)) { 918 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 919 920 if (curr_active) 921 err = dev_set_promiscuity(curr_active->dev, inc); 922 } else { 923 struct slave *slave; 924 925 bond_for_each_slave(bond, slave, iter) { 926 err = dev_set_promiscuity(slave->dev, inc); 927 if (err) 928 return err; 929 } 930 } 931 return err; 932 } 933 934 /* Push the allmulti flag down to all slaves */ 935 static int bond_set_allmulti(struct bonding *bond, int inc) 936 { 937 struct list_head *iter; 938 int err = 0; 939 940 if (bond_uses_primary(bond)) { 941 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 942 943 if (curr_active) 944 err = dev_set_allmulti(curr_active->dev, inc); 945 } else { 946 struct slave *slave; 947 948 bond_for_each_slave(bond, slave, iter) { 949 err = dev_set_allmulti(slave->dev, inc); 950 if (err) 951 return err; 952 } 953 } 954 return err; 955 } 956 957 /* Retrieve the list of registered multicast addresses for the bonding 958 * device and retransmit an IGMP JOIN request to the current active 959 * slave. 960 */ 961 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 962 { 963 struct bonding *bond = container_of(work, struct bonding, 964 mcast_work.work); 965 966 if (!rtnl_trylock()) { 967 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 968 return; 969 } 970 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 971 972 if (bond->igmp_retrans > 1) { 973 bond->igmp_retrans--; 974 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 975 } 976 rtnl_unlock(); 977 } 978 979 /* Flush bond's hardware addresses from slave */ 980 static void bond_hw_addr_flush(struct net_device *bond_dev, 981 struct net_device *slave_dev) 982 { 983 struct bonding *bond = netdev_priv(bond_dev); 984 985 dev_uc_unsync(slave_dev, bond_dev); 986 dev_mc_unsync(slave_dev, bond_dev); 987 988 if (BOND_MODE(bond) == BOND_MODE_8023AD) 989 dev_mc_del(slave_dev, lacpdu_mcast_addr); 990 } 991 992 /*--------------------------- Active slave change ---------------------------*/ 993 994 /* Update the hardware address list and promisc/allmulti for the new and 995 * old active slaves (if any). Modes that are not using primary keep all 996 * slaves up date at all times; only the modes that use primary need to call 997 * this function to swap these settings during a failover. 998 */ 999 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 1000 struct slave *old_active) 1001 { 1002 if (old_active) { 1003 if (bond->dev->flags & IFF_PROMISC) 1004 dev_set_promiscuity(old_active->dev, -1); 1005 1006 if (bond->dev->flags & IFF_ALLMULTI) 1007 dev_set_allmulti(old_active->dev, -1); 1008 1009 if (bond->dev->flags & IFF_UP) 1010 bond_hw_addr_flush(bond->dev, old_active->dev); 1011 1012 bond_slave_ns_maddrs_add(bond, old_active); 1013 } 1014 1015 if (new_active) { 1016 /* FIXME: Signal errors upstream. */ 1017 if (bond->dev->flags & IFF_PROMISC) 1018 dev_set_promiscuity(new_active->dev, 1); 1019 1020 if (bond->dev->flags & IFF_ALLMULTI) 1021 dev_set_allmulti(new_active->dev, 1); 1022 1023 if (bond->dev->flags & IFF_UP) { 1024 netif_addr_lock_bh(bond->dev); 1025 dev_uc_sync(new_active->dev, bond->dev); 1026 dev_mc_sync(new_active->dev, bond->dev); 1027 netif_addr_unlock_bh(bond->dev); 1028 } 1029 1030 bond_slave_ns_maddrs_del(bond, new_active); 1031 } 1032 } 1033 1034 /** 1035 * bond_set_dev_addr - clone slave's address to bond 1036 * @bond_dev: bond net device 1037 * @slave_dev: slave net device 1038 * 1039 * Should be called with RTNL held. 1040 */ 1041 static int bond_set_dev_addr(struct net_device *bond_dev, 1042 struct net_device *slave_dev) 1043 { 1044 int err; 1045 1046 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 1047 bond_dev, slave_dev, slave_dev->addr_len); 1048 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 1049 if (err) 1050 return err; 1051 1052 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 1053 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 1054 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 1055 return 0; 1056 } 1057 1058 static struct slave *bond_get_old_active(struct bonding *bond, 1059 struct slave *new_active) 1060 { 1061 struct slave *slave; 1062 struct list_head *iter; 1063 1064 bond_for_each_slave(bond, slave, iter) { 1065 if (slave == new_active) 1066 continue; 1067 1068 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 1069 return slave; 1070 } 1071 1072 return NULL; 1073 } 1074 1075 /* bond_do_fail_over_mac 1076 * 1077 * Perform special MAC address swapping for fail_over_mac settings 1078 * 1079 * Called with RTNL 1080 */ 1081 static void bond_do_fail_over_mac(struct bonding *bond, 1082 struct slave *new_active, 1083 struct slave *old_active) 1084 { 1085 u8 tmp_mac[MAX_ADDR_LEN]; 1086 struct sockaddr_storage ss; 1087 int rv; 1088 1089 switch (bond->params.fail_over_mac) { 1090 case BOND_FOM_ACTIVE: 1091 if (new_active) { 1092 rv = bond_set_dev_addr(bond->dev, new_active->dev); 1093 if (rv) 1094 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 1095 -rv); 1096 } 1097 break; 1098 case BOND_FOM_FOLLOW: 1099 /* if new_active && old_active, swap them 1100 * if just old_active, do nothing (going to no active slave) 1101 * if just new_active, set new_active to bond's MAC 1102 */ 1103 if (!new_active) 1104 return; 1105 1106 if (!old_active) 1107 old_active = bond_get_old_active(bond, new_active); 1108 1109 if (old_active) { 1110 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 1111 new_active->dev->addr_len); 1112 bond_hw_addr_copy(ss.__data, 1113 old_active->dev->dev_addr, 1114 old_active->dev->addr_len); 1115 ss.ss_family = new_active->dev->type; 1116 } else { 1117 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 1118 bond->dev->addr_len); 1119 ss.ss_family = bond->dev->type; 1120 } 1121 1122 rv = dev_set_mac_address(new_active->dev, 1123 (struct sockaddr *)&ss, NULL); 1124 if (rv) { 1125 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1126 -rv); 1127 goto out; 1128 } 1129 1130 if (!old_active) 1131 goto out; 1132 1133 bond_hw_addr_copy(ss.__data, tmp_mac, 1134 new_active->dev->addr_len); 1135 ss.ss_family = old_active->dev->type; 1136 1137 rv = dev_set_mac_address(old_active->dev, 1138 (struct sockaddr *)&ss, NULL); 1139 if (rv) 1140 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1141 -rv); 1142 out: 1143 break; 1144 default: 1145 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1146 bond->params.fail_over_mac); 1147 break; 1148 } 1149 1150 } 1151 1152 /** 1153 * bond_choose_primary_or_current - select the primary or high priority slave 1154 * @bond: our bonding struct 1155 * 1156 * - Check if there is a primary link. If the primary link was set and is up, 1157 * go on and do link reselection. 1158 * 1159 * - If primary link is not set or down, find the highest priority link. 1160 * If the highest priority link is not current slave, set it as primary 1161 * link and do link reselection. 1162 */ 1163 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1164 { 1165 struct slave *prim = rtnl_dereference(bond->primary_slave); 1166 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1167 struct slave *slave, *hprio = NULL; 1168 struct list_head *iter; 1169 1170 if (!prim || prim->link != BOND_LINK_UP) { 1171 bond_for_each_slave(bond, slave, iter) { 1172 if (slave->link == BOND_LINK_UP) { 1173 hprio = hprio ?: slave; 1174 if (slave->prio > hprio->prio) 1175 hprio = slave; 1176 } 1177 } 1178 1179 if (hprio && hprio != curr) { 1180 prim = hprio; 1181 goto link_reselect; 1182 } 1183 1184 if (!curr || curr->link != BOND_LINK_UP) 1185 return NULL; 1186 return curr; 1187 } 1188 1189 if (bond->force_primary) { 1190 bond->force_primary = false; 1191 return prim; 1192 } 1193 1194 link_reselect: 1195 if (!curr || curr->link != BOND_LINK_UP) 1196 return prim; 1197 1198 /* At this point, prim and curr are both up */ 1199 switch (bond->params.primary_reselect) { 1200 case BOND_PRI_RESELECT_ALWAYS: 1201 return prim; 1202 case BOND_PRI_RESELECT_BETTER: 1203 if (prim->speed < curr->speed) 1204 return curr; 1205 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1206 return curr; 1207 return prim; 1208 case BOND_PRI_RESELECT_FAILURE: 1209 return curr; 1210 default: 1211 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1212 bond->params.primary_reselect); 1213 return curr; 1214 } 1215 } 1216 1217 /** 1218 * bond_find_best_slave - select the best available slave to be the active one 1219 * @bond: our bonding struct 1220 */ 1221 static struct slave *bond_find_best_slave(struct bonding *bond) 1222 { 1223 struct slave *slave, *bestslave = NULL; 1224 struct list_head *iter; 1225 int mintime = bond->params.updelay; 1226 1227 slave = bond_choose_primary_or_current(bond); 1228 if (slave) 1229 return slave; 1230 1231 bond_for_each_slave(bond, slave, iter) { 1232 if (slave->link == BOND_LINK_UP) 1233 return slave; 1234 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1235 slave->delay < mintime) { 1236 mintime = slave->delay; 1237 bestslave = slave; 1238 } 1239 } 1240 1241 return bestslave; 1242 } 1243 1244 /* must be called in RCU critical section or with RTNL held */ 1245 static bool bond_should_notify_peers(struct bonding *bond) 1246 { 1247 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave); 1248 1249 if (!slave || !bond->send_peer_notif || 1250 bond->send_peer_notif % 1251 max(1, bond->params.peer_notif_delay) != 0 || 1252 !netif_carrier_ok(bond->dev) || 1253 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1254 return false; 1255 1256 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1257 slave ? slave->dev->name : "NULL"); 1258 1259 return true; 1260 } 1261 1262 /** 1263 * bond_change_active_slave - change the active slave into the specified one 1264 * @bond: our bonding struct 1265 * @new_active: the new slave to make the active one 1266 * 1267 * Set the new slave to the bond's settings and unset them on the old 1268 * curr_active_slave. 1269 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1270 * 1271 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1272 * because it is apparently the best available slave we have, even though its 1273 * updelay hasn't timed out yet. 1274 * 1275 * Caller must hold RTNL. 1276 */ 1277 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1278 { 1279 struct slave *old_active; 1280 1281 ASSERT_RTNL(); 1282 1283 old_active = rtnl_dereference(bond->curr_active_slave); 1284 1285 if (old_active == new_active) 1286 return; 1287 1288 #ifdef CONFIG_XFRM_OFFLOAD 1289 bond_ipsec_del_sa_all(bond); 1290 #endif /* CONFIG_XFRM_OFFLOAD */ 1291 1292 if (new_active) { 1293 new_active->last_link_up = jiffies; 1294 1295 if (new_active->link == BOND_LINK_BACK) { 1296 if (bond_uses_primary(bond)) { 1297 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1298 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1299 } 1300 1301 new_active->delay = 0; 1302 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1303 BOND_SLAVE_NOTIFY_NOW); 1304 1305 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1306 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1307 1308 if (bond_is_lb(bond)) 1309 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1310 } else { 1311 if (bond_uses_primary(bond)) 1312 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1313 } 1314 } 1315 1316 if (bond_uses_primary(bond)) 1317 bond_hw_addr_swap(bond, new_active, old_active); 1318 1319 if (bond_is_lb(bond)) { 1320 bond_alb_handle_active_change(bond, new_active); 1321 if (old_active) 1322 bond_set_slave_inactive_flags(old_active, 1323 BOND_SLAVE_NOTIFY_NOW); 1324 if (new_active) 1325 bond_set_slave_active_flags(new_active, 1326 BOND_SLAVE_NOTIFY_NOW); 1327 } else { 1328 rcu_assign_pointer(bond->curr_active_slave, new_active); 1329 } 1330 1331 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1332 if (old_active) 1333 bond_set_slave_inactive_flags(old_active, 1334 BOND_SLAVE_NOTIFY_NOW); 1335 1336 if (new_active) { 1337 bool should_notify_peers = false; 1338 1339 bond_set_slave_active_flags(new_active, 1340 BOND_SLAVE_NOTIFY_NOW); 1341 1342 if (bond->params.fail_over_mac) 1343 bond_do_fail_over_mac(bond, new_active, 1344 old_active); 1345 1346 if (netif_running(bond->dev)) { 1347 bond->send_peer_notif = 1348 bond->params.num_peer_notif * 1349 max(1, bond->params.peer_notif_delay); 1350 should_notify_peers = 1351 bond_should_notify_peers(bond); 1352 } 1353 1354 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1355 if (should_notify_peers) { 1356 bond->send_peer_notif--; 1357 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1358 bond->dev); 1359 } 1360 } 1361 } 1362 1363 #ifdef CONFIG_XFRM_OFFLOAD 1364 bond_ipsec_add_sa_all(bond); 1365 #endif /* CONFIG_XFRM_OFFLOAD */ 1366 1367 /* resend IGMP joins since active slave has changed or 1368 * all were sent on curr_active_slave. 1369 * resend only if bond is brought up with the affected 1370 * bonding modes and the retransmission is enabled 1371 */ 1372 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1373 ((bond_uses_primary(bond) && new_active) || 1374 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1375 bond->igmp_retrans = bond->params.resend_igmp; 1376 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1377 } 1378 } 1379 1380 /** 1381 * bond_select_active_slave - select a new active slave, if needed 1382 * @bond: our bonding struct 1383 * 1384 * This functions should be called when one of the following occurs: 1385 * - The old curr_active_slave has been released or lost its link. 1386 * - The primary_slave has got its link back. 1387 * - A slave has got its link back and there's no old curr_active_slave. 1388 * 1389 * Caller must hold RTNL. 1390 */ 1391 void bond_select_active_slave(struct bonding *bond) 1392 { 1393 struct slave *best_slave; 1394 int rv; 1395 1396 ASSERT_RTNL(); 1397 1398 best_slave = bond_find_best_slave(bond); 1399 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1400 bond_change_active_slave(bond, best_slave); 1401 rv = bond_set_carrier(bond); 1402 if (!rv) 1403 return; 1404 1405 if (netif_carrier_ok(bond->dev)) 1406 netdev_info(bond->dev, "active interface up!\n"); 1407 else 1408 netdev_info(bond->dev, "now running without any active interface!\n"); 1409 } 1410 } 1411 1412 #ifdef CONFIG_NET_POLL_CONTROLLER 1413 static inline int slave_enable_netpoll(struct slave *slave) 1414 { 1415 struct netpoll *np; 1416 int err = 0; 1417 1418 np = kzalloc(sizeof(*np), GFP_KERNEL); 1419 err = -ENOMEM; 1420 if (!np) 1421 goto out; 1422 1423 err = __netpoll_setup(np, slave->dev); 1424 if (err) { 1425 kfree(np); 1426 goto out; 1427 } 1428 slave->np = np; 1429 out: 1430 return err; 1431 } 1432 static inline void slave_disable_netpoll(struct slave *slave) 1433 { 1434 struct netpoll *np = slave->np; 1435 1436 if (!np) 1437 return; 1438 1439 slave->np = NULL; 1440 1441 __netpoll_free(np); 1442 } 1443 1444 static void bond_poll_controller(struct net_device *bond_dev) 1445 { 1446 struct bonding *bond = netdev_priv(bond_dev); 1447 struct slave *slave = NULL; 1448 struct list_head *iter; 1449 struct ad_info ad_info; 1450 1451 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1452 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1453 return; 1454 1455 bond_for_each_slave_rcu(bond, slave, iter) { 1456 if (!bond_slave_is_up(slave)) 1457 continue; 1458 1459 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1460 struct aggregator *agg = 1461 SLAVE_AD_INFO(slave)->port.aggregator; 1462 1463 if (agg && 1464 agg->aggregator_identifier != ad_info.aggregator_id) 1465 continue; 1466 } 1467 1468 netpoll_poll_dev(slave->dev); 1469 } 1470 } 1471 1472 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1473 { 1474 struct bonding *bond = netdev_priv(bond_dev); 1475 struct list_head *iter; 1476 struct slave *slave; 1477 1478 bond_for_each_slave(bond, slave, iter) 1479 if (bond_slave_is_up(slave)) 1480 slave_disable_netpoll(slave); 1481 } 1482 1483 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1484 { 1485 struct bonding *bond = netdev_priv(dev); 1486 struct list_head *iter; 1487 struct slave *slave; 1488 int err = 0; 1489 1490 bond_for_each_slave(bond, slave, iter) { 1491 err = slave_enable_netpoll(slave); 1492 if (err) { 1493 bond_netpoll_cleanup(dev); 1494 break; 1495 } 1496 } 1497 return err; 1498 } 1499 #else 1500 static inline int slave_enable_netpoll(struct slave *slave) 1501 { 1502 return 0; 1503 } 1504 static inline void slave_disable_netpoll(struct slave *slave) 1505 { 1506 } 1507 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1508 { 1509 } 1510 #endif 1511 1512 /*---------------------------------- IOCTL ----------------------------------*/ 1513 1514 static netdev_features_t bond_fix_features(struct net_device *dev, 1515 netdev_features_t features) 1516 { 1517 struct bonding *bond = netdev_priv(dev); 1518 struct list_head *iter; 1519 netdev_features_t mask; 1520 struct slave *slave; 1521 1522 mask = features; 1523 1524 features &= ~NETIF_F_ONE_FOR_ALL; 1525 features |= NETIF_F_ALL_FOR_ALL; 1526 1527 bond_for_each_slave(bond, slave, iter) { 1528 features = netdev_increment_features(features, 1529 slave->dev->features, 1530 mask); 1531 } 1532 features = netdev_add_tso_features(features, mask); 1533 1534 return features; 1535 } 1536 1537 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1538 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1539 NETIF_F_HIGHDMA | NETIF_F_LRO) 1540 1541 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1542 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1543 1544 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1545 NETIF_F_GSO_SOFTWARE) 1546 1547 1548 static void bond_compute_features(struct bonding *bond) 1549 { 1550 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1551 IFF_XMIT_DST_RELEASE_PERM; 1552 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1553 netdev_features_t enc_features = BOND_ENC_FEATURES; 1554 #ifdef CONFIG_XFRM_OFFLOAD 1555 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1556 #endif /* CONFIG_XFRM_OFFLOAD */ 1557 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1558 struct net_device *bond_dev = bond->dev; 1559 struct list_head *iter; 1560 struct slave *slave; 1561 unsigned short max_hard_header_len = ETH_HLEN; 1562 unsigned int tso_max_size = TSO_MAX_SIZE; 1563 u16 tso_max_segs = TSO_MAX_SEGS; 1564 1565 if (!bond_has_slaves(bond)) 1566 goto done; 1567 vlan_features &= NETIF_F_ALL_FOR_ALL; 1568 mpls_features &= NETIF_F_ALL_FOR_ALL; 1569 1570 bond_for_each_slave(bond, slave, iter) { 1571 vlan_features = netdev_increment_features(vlan_features, 1572 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1573 1574 enc_features = netdev_increment_features(enc_features, 1575 slave->dev->hw_enc_features, 1576 BOND_ENC_FEATURES); 1577 1578 #ifdef CONFIG_XFRM_OFFLOAD 1579 xfrm_features = netdev_increment_features(xfrm_features, 1580 slave->dev->hw_enc_features, 1581 BOND_XFRM_FEATURES); 1582 #endif /* CONFIG_XFRM_OFFLOAD */ 1583 1584 mpls_features = netdev_increment_features(mpls_features, 1585 slave->dev->mpls_features, 1586 BOND_MPLS_FEATURES); 1587 1588 dst_release_flag &= slave->dev->priv_flags; 1589 if (slave->dev->hard_header_len > max_hard_header_len) 1590 max_hard_header_len = slave->dev->hard_header_len; 1591 1592 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1593 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1594 } 1595 bond_dev->hard_header_len = max_hard_header_len; 1596 1597 done: 1598 bond_dev->vlan_features = vlan_features; 1599 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1600 NETIF_F_HW_VLAN_CTAG_TX | 1601 NETIF_F_HW_VLAN_STAG_TX; 1602 #ifdef CONFIG_XFRM_OFFLOAD 1603 bond_dev->hw_enc_features |= xfrm_features; 1604 #endif /* CONFIG_XFRM_OFFLOAD */ 1605 bond_dev->mpls_features = mpls_features; 1606 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1607 netif_set_tso_max_size(bond_dev, tso_max_size); 1608 1609 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1610 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1611 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1612 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1613 1614 netdev_change_features(bond_dev); 1615 } 1616 1617 static void bond_setup_by_slave(struct net_device *bond_dev, 1618 struct net_device *slave_dev) 1619 { 1620 bool was_up = !!(bond_dev->flags & IFF_UP); 1621 1622 dev_close(bond_dev); 1623 1624 bond_dev->header_ops = slave_dev->header_ops; 1625 1626 bond_dev->type = slave_dev->type; 1627 bond_dev->hard_header_len = slave_dev->hard_header_len; 1628 bond_dev->needed_headroom = slave_dev->needed_headroom; 1629 bond_dev->addr_len = slave_dev->addr_len; 1630 1631 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1632 slave_dev->addr_len); 1633 1634 if (slave_dev->flags & IFF_POINTOPOINT) { 1635 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1636 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1637 } 1638 if (was_up) 1639 dev_open(bond_dev, NULL); 1640 } 1641 1642 /* On bonding slaves other than the currently active slave, suppress 1643 * duplicates except for alb non-mcast/bcast. 1644 */ 1645 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1646 struct slave *slave, 1647 struct bonding *bond) 1648 { 1649 if (bond_is_slave_inactive(slave)) { 1650 if (BOND_MODE(bond) == BOND_MODE_ALB && 1651 skb->pkt_type != PACKET_BROADCAST && 1652 skb->pkt_type != PACKET_MULTICAST) 1653 return false; 1654 return true; 1655 } 1656 return false; 1657 } 1658 1659 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1660 { 1661 struct sk_buff *skb = *pskb; 1662 struct slave *slave; 1663 struct bonding *bond; 1664 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1665 struct slave *); 1666 int ret = RX_HANDLER_ANOTHER; 1667 1668 skb = skb_share_check(skb, GFP_ATOMIC); 1669 if (unlikely(!skb)) 1670 return RX_HANDLER_CONSUMED; 1671 1672 *pskb = skb; 1673 1674 slave = bond_slave_get_rcu(skb->dev); 1675 bond = slave->bond; 1676 1677 recv_probe = READ_ONCE(bond->recv_probe); 1678 if (recv_probe) { 1679 ret = recv_probe(skb, bond, slave); 1680 if (ret == RX_HANDLER_CONSUMED) { 1681 consume_skb(skb); 1682 return ret; 1683 } 1684 } 1685 1686 /* 1687 * For packets determined by bond_should_deliver_exact_match() call to 1688 * be suppressed we want to make an exception for link-local packets. 1689 * This is necessary for e.g. LLDP daemons to be able to monitor 1690 * inactive slave links without being forced to bind to them 1691 * explicitly. 1692 * 1693 * At the same time, packets that are passed to the bonding master 1694 * (including link-local ones) can have their originating interface 1695 * determined via PACKET_ORIGDEV socket option. 1696 */ 1697 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1698 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1699 return RX_HANDLER_PASS; 1700 return RX_HANDLER_EXACT; 1701 } 1702 1703 skb->dev = bond->dev; 1704 1705 if (BOND_MODE(bond) == BOND_MODE_ALB && 1706 netif_is_bridge_port(bond->dev) && 1707 skb->pkt_type == PACKET_HOST) { 1708 1709 if (unlikely(skb_cow_head(skb, 1710 skb->data - skb_mac_header(skb)))) { 1711 kfree_skb(skb); 1712 return RX_HANDLER_CONSUMED; 1713 } 1714 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1715 bond->dev->addr_len); 1716 } 1717 1718 return ret; 1719 } 1720 1721 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1722 { 1723 switch (BOND_MODE(bond)) { 1724 case BOND_MODE_ROUNDROBIN: 1725 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1726 case BOND_MODE_ACTIVEBACKUP: 1727 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1728 case BOND_MODE_BROADCAST: 1729 return NETDEV_LAG_TX_TYPE_BROADCAST; 1730 case BOND_MODE_XOR: 1731 case BOND_MODE_8023AD: 1732 return NETDEV_LAG_TX_TYPE_HASH; 1733 default: 1734 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1735 } 1736 } 1737 1738 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1739 enum netdev_lag_tx_type type) 1740 { 1741 if (type != NETDEV_LAG_TX_TYPE_HASH) 1742 return NETDEV_LAG_HASH_NONE; 1743 1744 switch (bond->params.xmit_policy) { 1745 case BOND_XMIT_POLICY_LAYER2: 1746 return NETDEV_LAG_HASH_L2; 1747 case BOND_XMIT_POLICY_LAYER34: 1748 return NETDEV_LAG_HASH_L34; 1749 case BOND_XMIT_POLICY_LAYER23: 1750 return NETDEV_LAG_HASH_L23; 1751 case BOND_XMIT_POLICY_ENCAP23: 1752 return NETDEV_LAG_HASH_E23; 1753 case BOND_XMIT_POLICY_ENCAP34: 1754 return NETDEV_LAG_HASH_E34; 1755 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1756 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1757 default: 1758 return NETDEV_LAG_HASH_UNKNOWN; 1759 } 1760 } 1761 1762 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1763 struct netlink_ext_ack *extack) 1764 { 1765 struct netdev_lag_upper_info lag_upper_info; 1766 enum netdev_lag_tx_type type; 1767 int err; 1768 1769 type = bond_lag_tx_type(bond); 1770 lag_upper_info.tx_type = type; 1771 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1772 1773 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1774 &lag_upper_info, extack); 1775 if (err) 1776 return err; 1777 1778 slave->dev->flags |= IFF_SLAVE; 1779 return 0; 1780 } 1781 1782 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1783 { 1784 netdev_upper_dev_unlink(slave->dev, bond->dev); 1785 slave->dev->flags &= ~IFF_SLAVE; 1786 } 1787 1788 static void slave_kobj_release(struct kobject *kobj) 1789 { 1790 struct slave *slave = to_slave(kobj); 1791 struct bonding *bond = bond_get_bond_by_slave(slave); 1792 1793 cancel_delayed_work_sync(&slave->notify_work); 1794 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1795 kfree(SLAVE_AD_INFO(slave)); 1796 1797 kfree(slave); 1798 } 1799 1800 static struct kobj_type slave_ktype = { 1801 .release = slave_kobj_release, 1802 #ifdef CONFIG_SYSFS 1803 .sysfs_ops = &slave_sysfs_ops, 1804 #endif 1805 }; 1806 1807 static int bond_kobj_init(struct slave *slave) 1808 { 1809 int err; 1810 1811 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1812 &(slave->dev->dev.kobj), "bonding_slave"); 1813 if (err) 1814 kobject_put(&slave->kobj); 1815 1816 return err; 1817 } 1818 1819 static struct slave *bond_alloc_slave(struct bonding *bond, 1820 struct net_device *slave_dev) 1821 { 1822 struct slave *slave = NULL; 1823 1824 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1825 if (!slave) 1826 return NULL; 1827 1828 slave->bond = bond; 1829 slave->dev = slave_dev; 1830 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1831 1832 if (bond_kobj_init(slave)) 1833 return NULL; 1834 1835 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1836 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1837 GFP_KERNEL); 1838 if (!SLAVE_AD_INFO(slave)) { 1839 kobject_put(&slave->kobj); 1840 return NULL; 1841 } 1842 } 1843 1844 return slave; 1845 } 1846 1847 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1848 { 1849 info->bond_mode = BOND_MODE(bond); 1850 info->miimon = bond->params.miimon; 1851 info->num_slaves = bond->slave_cnt; 1852 } 1853 1854 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1855 { 1856 strcpy(info->slave_name, slave->dev->name); 1857 info->link = slave->link; 1858 info->state = bond_slave_state(slave); 1859 info->link_failure_count = slave->link_failure_count; 1860 } 1861 1862 static void bond_netdev_notify_work(struct work_struct *_work) 1863 { 1864 struct slave *slave = container_of(_work, struct slave, 1865 notify_work.work); 1866 1867 if (rtnl_trylock()) { 1868 struct netdev_bonding_info binfo; 1869 1870 bond_fill_ifslave(slave, &binfo.slave); 1871 bond_fill_ifbond(slave->bond, &binfo.master); 1872 netdev_bonding_info_change(slave->dev, &binfo); 1873 rtnl_unlock(); 1874 } else { 1875 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1876 } 1877 } 1878 1879 void bond_queue_slave_event(struct slave *slave) 1880 { 1881 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1882 } 1883 1884 void bond_lower_state_changed(struct slave *slave) 1885 { 1886 struct netdev_lag_lower_state_info info; 1887 1888 info.link_up = slave->link == BOND_LINK_UP || 1889 slave->link == BOND_LINK_FAIL; 1890 info.tx_enabled = bond_is_active_slave(slave); 1891 netdev_lower_state_changed(slave->dev, &info); 1892 } 1893 1894 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1895 if (extack) \ 1896 NL_SET_ERR_MSG(extack, errmsg); \ 1897 else \ 1898 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1899 } while (0) 1900 1901 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1902 if (extack) \ 1903 NL_SET_ERR_MSG(extack, errmsg); \ 1904 else \ 1905 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1906 } while (0) 1907 1908 /* The bonding driver uses ether_setup() to convert a master bond device 1909 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always 1910 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP 1911 * if they were set 1912 */ 1913 static void bond_ether_setup(struct net_device *bond_dev) 1914 { 1915 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP); 1916 1917 ether_setup(bond_dev); 1918 bond_dev->flags |= IFF_MASTER | flags; 1919 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1920 } 1921 1922 void bond_xdp_set_features(struct net_device *bond_dev) 1923 { 1924 struct bonding *bond = netdev_priv(bond_dev); 1925 xdp_features_t val = NETDEV_XDP_ACT_MASK; 1926 struct list_head *iter; 1927 struct slave *slave; 1928 1929 ASSERT_RTNL(); 1930 1931 if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) { 1932 xdp_clear_features_flag(bond_dev); 1933 return; 1934 } 1935 1936 bond_for_each_slave(bond, slave, iter) 1937 val &= slave->dev->xdp_features; 1938 1939 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY; 1940 1941 xdp_set_features_flag(bond_dev, val); 1942 } 1943 1944 /* enslave device <slave> to bond device <master> */ 1945 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1946 struct netlink_ext_ack *extack) 1947 { 1948 struct bonding *bond = netdev_priv(bond_dev); 1949 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1950 struct slave *new_slave = NULL, *prev_slave; 1951 struct sockaddr_storage ss; 1952 int link_reporting; 1953 int res = 0, i; 1954 1955 if (slave_dev->flags & IFF_MASTER && 1956 !netif_is_bond_master(slave_dev)) { 1957 BOND_NL_ERR(bond_dev, extack, 1958 "Device type (master device) cannot be enslaved"); 1959 return -EPERM; 1960 } 1961 1962 if (!bond->params.use_carrier && 1963 slave_dev->ethtool_ops->get_link == NULL && 1964 slave_ops->ndo_eth_ioctl == NULL) { 1965 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1966 } 1967 1968 /* already in-use? */ 1969 if (netdev_is_rx_handler_busy(slave_dev)) { 1970 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1971 "Device is in use and cannot be enslaved"); 1972 return -EBUSY; 1973 } 1974 1975 if (bond_dev == slave_dev) { 1976 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1977 return -EPERM; 1978 } 1979 1980 /* vlan challenged mutual exclusion */ 1981 /* no need to lock since we're protected by rtnl_lock */ 1982 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1983 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1984 if (vlan_uses_dev(bond_dev)) { 1985 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1986 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1987 return -EPERM; 1988 } else { 1989 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1990 } 1991 } else { 1992 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1993 } 1994 1995 if (slave_dev->features & NETIF_F_HW_ESP) 1996 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1997 1998 /* Old ifenslave binaries are no longer supported. These can 1999 * be identified with moderate accuracy by the state of the slave: 2000 * the current ifenslave will set the interface down prior to 2001 * enslaving it; the old ifenslave will not. 2002 */ 2003 if (slave_dev->flags & IFF_UP) { 2004 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2005 "Device can not be enslaved while up"); 2006 return -EPERM; 2007 } 2008 2009 /* set bonding device ether type by slave - bonding netdevices are 2010 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 2011 * there is a need to override some of the type dependent attribs/funcs. 2012 * 2013 * bond ether type mutual exclusion - don't allow slaves of dissimilar 2014 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 2015 */ 2016 if (!bond_has_slaves(bond)) { 2017 if (bond_dev->type != slave_dev->type) { 2018 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 2019 bond_dev->type, slave_dev->type); 2020 2021 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 2022 bond_dev); 2023 res = notifier_to_errno(res); 2024 if (res) { 2025 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 2026 return -EBUSY; 2027 } 2028 2029 /* Flush unicast and multicast addresses */ 2030 dev_uc_flush(bond_dev); 2031 dev_mc_flush(bond_dev); 2032 2033 if (slave_dev->type != ARPHRD_ETHER) 2034 bond_setup_by_slave(bond_dev, slave_dev); 2035 else 2036 bond_ether_setup(bond_dev); 2037 2038 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 2039 bond_dev); 2040 } 2041 } else if (bond_dev->type != slave_dev->type) { 2042 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2043 "Device type is different from other slaves"); 2044 return -EINVAL; 2045 } 2046 2047 if (slave_dev->type == ARPHRD_INFINIBAND && 2048 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2049 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2050 "Only active-backup mode is supported for infiniband slaves"); 2051 res = -EOPNOTSUPP; 2052 goto err_undo_flags; 2053 } 2054 2055 if (!slave_ops->ndo_set_mac_address || 2056 slave_dev->type == ARPHRD_INFINIBAND) { 2057 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 2058 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 2059 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 2060 if (!bond_has_slaves(bond)) { 2061 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 2062 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 2063 } else { 2064 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2065 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 2066 res = -EOPNOTSUPP; 2067 goto err_undo_flags; 2068 } 2069 } 2070 } 2071 2072 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 2073 2074 /* If this is the first slave, then we need to set the master's hardware 2075 * address to be the same as the slave's. 2076 */ 2077 if (!bond_has_slaves(bond) && 2078 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 2079 res = bond_set_dev_addr(bond->dev, slave_dev); 2080 if (res) 2081 goto err_undo_flags; 2082 } 2083 2084 new_slave = bond_alloc_slave(bond, slave_dev); 2085 if (!new_slave) { 2086 res = -ENOMEM; 2087 goto err_undo_flags; 2088 } 2089 2090 /* Set the new_slave's queue_id to be zero. Queue ID mapping 2091 * is set via sysfs or module option if desired. 2092 */ 2093 new_slave->queue_id = 0; 2094 2095 /* Save slave's original mtu and then set it to match the bond */ 2096 new_slave->original_mtu = slave_dev->mtu; 2097 res = dev_set_mtu(slave_dev, bond->dev->mtu); 2098 if (res) { 2099 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 2100 goto err_free; 2101 } 2102 2103 /* Save slave's original ("permanent") mac address for modes 2104 * that need it, and for restoring it upon release, and then 2105 * set it to the master's address 2106 */ 2107 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 2108 slave_dev->addr_len); 2109 2110 if (!bond->params.fail_over_mac || 2111 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2112 /* Set slave to master's mac address. The application already 2113 * set the master's mac address to that of the first slave 2114 */ 2115 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 2116 ss.ss_family = slave_dev->type; 2117 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 2118 extack); 2119 if (res) { 2120 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 2121 goto err_restore_mtu; 2122 } 2123 } 2124 2125 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 2126 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 2127 2128 /* open the slave since the application closed it */ 2129 res = dev_open(slave_dev, extack); 2130 if (res) { 2131 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 2132 goto err_restore_mac; 2133 } 2134 2135 slave_dev->priv_flags |= IFF_BONDING; 2136 /* initialize slave stats */ 2137 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 2138 2139 if (bond_is_lb(bond)) { 2140 /* bond_alb_init_slave() must be called before all other stages since 2141 * it might fail and we do not want to have to undo everything 2142 */ 2143 res = bond_alb_init_slave(bond, new_slave); 2144 if (res) 2145 goto err_close; 2146 } 2147 2148 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 2149 if (res) { 2150 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 2151 goto err_close; 2152 } 2153 2154 prev_slave = bond_last_slave(bond); 2155 2156 new_slave->delay = 0; 2157 new_slave->link_failure_count = 0; 2158 2159 if (bond_update_speed_duplex(new_slave) && 2160 bond_needs_speed_duplex(bond)) 2161 new_slave->link = BOND_LINK_DOWN; 2162 2163 new_slave->last_rx = jiffies - 2164 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2165 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2166 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2167 2168 new_slave->last_tx = new_slave->last_rx; 2169 2170 if (bond->params.miimon && !bond->params.use_carrier) { 2171 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2172 2173 if ((link_reporting == -1) && !bond->params.arp_interval) { 2174 /* miimon is set but a bonded network driver 2175 * does not support ETHTOOL/MII and 2176 * arp_interval is not set. Note: if 2177 * use_carrier is enabled, we will never go 2178 * here (because netif_carrier is always 2179 * supported); thus, we don't need to change 2180 * the messages for netif_carrier. 2181 */ 2182 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n"); 2183 } else if (link_reporting == -1) { 2184 /* unable get link status using mii/ethtool */ 2185 slave_warn(bond_dev, slave_dev, "can't get link status from slave; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n"); 2186 } 2187 } 2188 2189 /* check for initial state */ 2190 new_slave->link = BOND_LINK_NOCHANGE; 2191 if (bond->params.miimon) { 2192 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2193 if (bond->params.updelay) { 2194 bond_set_slave_link_state(new_slave, 2195 BOND_LINK_BACK, 2196 BOND_SLAVE_NOTIFY_NOW); 2197 new_slave->delay = bond->params.updelay; 2198 } else { 2199 bond_set_slave_link_state(new_slave, 2200 BOND_LINK_UP, 2201 BOND_SLAVE_NOTIFY_NOW); 2202 } 2203 } else { 2204 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2205 BOND_SLAVE_NOTIFY_NOW); 2206 } 2207 } else if (bond->params.arp_interval) { 2208 bond_set_slave_link_state(new_slave, 2209 (netif_carrier_ok(slave_dev) ? 2210 BOND_LINK_UP : BOND_LINK_DOWN), 2211 BOND_SLAVE_NOTIFY_NOW); 2212 } else { 2213 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2214 BOND_SLAVE_NOTIFY_NOW); 2215 } 2216 2217 if (new_slave->link != BOND_LINK_DOWN) 2218 new_slave->last_link_up = jiffies; 2219 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2220 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2221 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2222 2223 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2224 /* if there is a primary slave, remember it */ 2225 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2226 rcu_assign_pointer(bond->primary_slave, new_slave); 2227 bond->force_primary = true; 2228 } 2229 } 2230 2231 switch (BOND_MODE(bond)) { 2232 case BOND_MODE_ACTIVEBACKUP: 2233 bond_set_slave_inactive_flags(new_slave, 2234 BOND_SLAVE_NOTIFY_NOW); 2235 break; 2236 case BOND_MODE_8023AD: 2237 /* in 802.3ad mode, the internal mechanism 2238 * will activate the slaves in the selected 2239 * aggregator 2240 */ 2241 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2242 /* if this is the first slave */ 2243 if (!prev_slave) { 2244 SLAVE_AD_INFO(new_slave)->id = 1; 2245 /* Initialize AD with the number of times that the AD timer is called in 1 second 2246 * can be called only after the mac address of the bond is set 2247 */ 2248 bond_3ad_initialize(bond); 2249 } else { 2250 SLAVE_AD_INFO(new_slave)->id = 2251 SLAVE_AD_INFO(prev_slave)->id + 1; 2252 } 2253 2254 bond_3ad_bind_slave(new_slave); 2255 break; 2256 case BOND_MODE_TLB: 2257 case BOND_MODE_ALB: 2258 bond_set_active_slave(new_slave); 2259 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2260 break; 2261 default: 2262 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2263 2264 /* always active in trunk mode */ 2265 bond_set_active_slave(new_slave); 2266 2267 /* In trunking mode there is little meaning to curr_active_slave 2268 * anyway (it holds no special properties of the bond device), 2269 * so we can change it without calling change_active_interface() 2270 */ 2271 if (!rcu_access_pointer(bond->curr_active_slave) && 2272 new_slave->link == BOND_LINK_UP) 2273 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2274 2275 break; 2276 } /* switch(bond_mode) */ 2277 2278 #ifdef CONFIG_NET_POLL_CONTROLLER 2279 if (bond->dev->npinfo) { 2280 if (slave_enable_netpoll(new_slave)) { 2281 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2282 res = -EBUSY; 2283 goto err_detach; 2284 } 2285 } 2286 #endif 2287 2288 if (!(bond_dev->features & NETIF_F_LRO)) 2289 dev_disable_lro(slave_dev); 2290 2291 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2292 new_slave); 2293 if (res) { 2294 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2295 goto err_detach; 2296 } 2297 2298 res = bond_master_upper_dev_link(bond, new_slave, extack); 2299 if (res) { 2300 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2301 goto err_unregister; 2302 } 2303 2304 bond_lower_state_changed(new_slave); 2305 2306 res = bond_sysfs_slave_add(new_slave); 2307 if (res) { 2308 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2309 goto err_upper_unlink; 2310 } 2311 2312 /* If the mode uses primary, then the following is handled by 2313 * bond_change_active_slave(). 2314 */ 2315 if (!bond_uses_primary(bond)) { 2316 /* set promiscuity level to new slave */ 2317 if (bond_dev->flags & IFF_PROMISC) { 2318 res = dev_set_promiscuity(slave_dev, 1); 2319 if (res) 2320 goto err_sysfs_del; 2321 } 2322 2323 /* set allmulti level to new slave */ 2324 if (bond_dev->flags & IFF_ALLMULTI) { 2325 res = dev_set_allmulti(slave_dev, 1); 2326 if (res) { 2327 if (bond_dev->flags & IFF_PROMISC) 2328 dev_set_promiscuity(slave_dev, -1); 2329 goto err_sysfs_del; 2330 } 2331 } 2332 2333 if (bond_dev->flags & IFF_UP) { 2334 netif_addr_lock_bh(bond_dev); 2335 dev_mc_sync_multiple(slave_dev, bond_dev); 2336 dev_uc_sync_multiple(slave_dev, bond_dev); 2337 netif_addr_unlock_bh(bond_dev); 2338 2339 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2340 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2341 } 2342 } 2343 2344 bond->slave_cnt++; 2345 bond_compute_features(bond); 2346 bond_set_carrier(bond); 2347 2348 /* Needs to be called before bond_select_active_slave(), which will 2349 * remove the maddrs if the slave is selected as active slave. 2350 */ 2351 bond_slave_ns_maddrs_add(bond, new_slave); 2352 2353 if (bond_uses_primary(bond)) { 2354 block_netpoll_tx(); 2355 bond_select_active_slave(bond); 2356 unblock_netpoll_tx(); 2357 } 2358 2359 if (bond_mode_can_use_xmit_hash(bond)) 2360 bond_update_slave_arr(bond, NULL); 2361 2362 if (!slave_dev->netdev_ops->ndo_bpf || 2363 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2364 if (bond->xdp_prog) { 2365 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2366 "Slave does not support XDP"); 2367 res = -EOPNOTSUPP; 2368 goto err_sysfs_del; 2369 } 2370 } else if (bond->xdp_prog) { 2371 struct netdev_bpf xdp = { 2372 .command = XDP_SETUP_PROG, 2373 .flags = 0, 2374 .prog = bond->xdp_prog, 2375 .extack = extack, 2376 }; 2377 2378 if (dev_xdp_prog_count(slave_dev) > 0) { 2379 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2380 "Slave has XDP program loaded, please unload before enslaving"); 2381 res = -EOPNOTSUPP; 2382 goto err_sysfs_del; 2383 } 2384 2385 res = dev_xdp_propagate(slave_dev, &xdp); 2386 if (res < 0) { 2387 /* ndo_bpf() sets extack error message */ 2388 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2389 goto err_sysfs_del; 2390 } 2391 if (bond->xdp_prog) 2392 bpf_prog_inc(bond->xdp_prog); 2393 } 2394 2395 bond_xdp_set_features(bond_dev); 2396 2397 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2398 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2399 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2400 2401 /* enslave is successful */ 2402 bond_queue_slave_event(new_slave); 2403 return 0; 2404 2405 /* Undo stages on error */ 2406 err_sysfs_del: 2407 bond_sysfs_slave_del(new_slave); 2408 2409 err_upper_unlink: 2410 bond_upper_dev_unlink(bond, new_slave); 2411 2412 err_unregister: 2413 netdev_rx_handler_unregister(slave_dev); 2414 2415 err_detach: 2416 vlan_vids_del_by_dev(slave_dev, bond_dev); 2417 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2418 RCU_INIT_POINTER(bond->primary_slave, NULL); 2419 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2420 block_netpoll_tx(); 2421 bond_change_active_slave(bond, NULL); 2422 bond_select_active_slave(bond); 2423 unblock_netpoll_tx(); 2424 } 2425 /* either primary_slave or curr_active_slave might've changed */ 2426 synchronize_rcu(); 2427 slave_disable_netpoll(new_slave); 2428 2429 err_close: 2430 if (!netif_is_bond_master(slave_dev)) 2431 slave_dev->priv_flags &= ~IFF_BONDING; 2432 dev_close(slave_dev); 2433 2434 err_restore_mac: 2435 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2436 if (!bond->params.fail_over_mac || 2437 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2438 /* XXX TODO - fom follow mode needs to change master's 2439 * MAC if this slave's MAC is in use by the bond, or at 2440 * least print a warning. 2441 */ 2442 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2443 new_slave->dev->addr_len); 2444 ss.ss_family = slave_dev->type; 2445 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2446 } 2447 2448 err_restore_mtu: 2449 dev_set_mtu(slave_dev, new_slave->original_mtu); 2450 2451 err_free: 2452 kobject_put(&new_slave->kobj); 2453 2454 err_undo_flags: 2455 /* Enslave of first slave has failed and we need to fix master's mac */ 2456 if (!bond_has_slaves(bond)) { 2457 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2458 slave_dev->dev_addr)) 2459 eth_hw_addr_random(bond_dev); 2460 if (bond_dev->type != ARPHRD_ETHER) { 2461 dev_close(bond_dev); 2462 bond_ether_setup(bond_dev); 2463 } 2464 } 2465 2466 return res; 2467 } 2468 2469 /* Try to release the slave device <slave> from the bond device <master> 2470 * It is legal to access curr_active_slave without a lock because all the function 2471 * is RTNL-locked. If "all" is true it means that the function is being called 2472 * while destroying a bond interface and all slaves are being released. 2473 * 2474 * The rules for slave state should be: 2475 * for Active/Backup: 2476 * Active stays on all backups go down 2477 * for Bonded connections: 2478 * The first up interface should be left on and all others downed. 2479 */ 2480 static int __bond_release_one(struct net_device *bond_dev, 2481 struct net_device *slave_dev, 2482 bool all, bool unregister) 2483 { 2484 struct bonding *bond = netdev_priv(bond_dev); 2485 struct slave *slave, *oldcurrent; 2486 struct sockaddr_storage ss; 2487 int old_flags = bond_dev->flags; 2488 netdev_features_t old_features = bond_dev->features; 2489 2490 /* slave is not a slave or master is not master of this slave */ 2491 if (!(slave_dev->flags & IFF_SLAVE) || 2492 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2493 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2494 return -EINVAL; 2495 } 2496 2497 block_netpoll_tx(); 2498 2499 slave = bond_get_slave_by_dev(bond, slave_dev); 2500 if (!slave) { 2501 /* not a slave of this bond */ 2502 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2503 unblock_netpoll_tx(); 2504 return -EINVAL; 2505 } 2506 2507 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2508 2509 bond_sysfs_slave_del(slave); 2510 2511 /* recompute stats just before removing the slave */ 2512 bond_get_stats(bond->dev, &bond->bond_stats); 2513 2514 if (bond->xdp_prog) { 2515 struct netdev_bpf xdp = { 2516 .command = XDP_SETUP_PROG, 2517 .flags = 0, 2518 .prog = NULL, 2519 .extack = NULL, 2520 }; 2521 if (dev_xdp_propagate(slave_dev, &xdp)) 2522 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2523 } 2524 2525 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2526 * for this slave anymore. 2527 */ 2528 netdev_rx_handler_unregister(slave_dev); 2529 2530 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2531 bond_3ad_unbind_slave(slave); 2532 2533 bond_upper_dev_unlink(bond, slave); 2534 2535 if (bond_mode_can_use_xmit_hash(bond)) 2536 bond_update_slave_arr(bond, slave); 2537 2538 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2539 bond_is_active_slave(slave) ? "active" : "backup"); 2540 2541 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2542 2543 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2544 2545 if (!all && (!bond->params.fail_over_mac || 2546 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2547 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2548 bond_has_slaves(bond)) 2549 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", 2550 slave->perm_hwaddr); 2551 } 2552 2553 if (rtnl_dereference(bond->primary_slave) == slave) 2554 RCU_INIT_POINTER(bond->primary_slave, NULL); 2555 2556 if (oldcurrent == slave) 2557 bond_change_active_slave(bond, NULL); 2558 2559 /* Must be called after bond_change_active_slave () as the slave 2560 * might change from an active slave to a backup slave. Then it is 2561 * necessary to clear the maddrs on the backup slave. 2562 */ 2563 bond_slave_ns_maddrs_del(bond, slave); 2564 2565 if (bond_is_lb(bond)) { 2566 /* Must be called only after the slave has been 2567 * detached from the list and the curr_active_slave 2568 * has been cleared (if our_slave == old_current), 2569 * but before a new active slave is selected. 2570 */ 2571 bond_alb_deinit_slave(bond, slave); 2572 } 2573 2574 if (all) { 2575 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2576 } else if (oldcurrent == slave) { 2577 /* Note that we hold RTNL over this sequence, so there 2578 * is no concern that another slave add/remove event 2579 * will interfere. 2580 */ 2581 bond_select_active_slave(bond); 2582 } 2583 2584 bond_set_carrier(bond); 2585 if (!bond_has_slaves(bond)) 2586 eth_hw_addr_random(bond_dev); 2587 2588 unblock_netpoll_tx(); 2589 synchronize_rcu(); 2590 bond->slave_cnt--; 2591 2592 if (!bond_has_slaves(bond)) { 2593 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2594 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2595 } 2596 2597 bond_compute_features(bond); 2598 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2599 (old_features & NETIF_F_VLAN_CHALLENGED)) 2600 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2601 2602 vlan_vids_del_by_dev(slave_dev, bond_dev); 2603 2604 /* If the mode uses primary, then this case was handled above by 2605 * bond_change_active_slave(..., NULL) 2606 */ 2607 if (!bond_uses_primary(bond)) { 2608 /* unset promiscuity level from slave 2609 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2610 * of the IFF_PROMISC flag in the bond_dev, but we need the 2611 * value of that flag before that change, as that was the value 2612 * when this slave was attached, so we cache at the start of the 2613 * function and use it here. Same goes for ALLMULTI below 2614 */ 2615 if (old_flags & IFF_PROMISC) 2616 dev_set_promiscuity(slave_dev, -1); 2617 2618 /* unset allmulti level from slave */ 2619 if (old_flags & IFF_ALLMULTI) 2620 dev_set_allmulti(slave_dev, -1); 2621 2622 if (old_flags & IFF_UP) 2623 bond_hw_addr_flush(bond_dev, slave_dev); 2624 } 2625 2626 slave_disable_netpoll(slave); 2627 2628 /* close slave before restoring its mac address */ 2629 dev_close(slave_dev); 2630 2631 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2632 2633 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2634 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2635 /* restore original ("permanent") mac address */ 2636 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2637 slave->dev->addr_len); 2638 ss.ss_family = slave_dev->type; 2639 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2640 } 2641 2642 if (unregister) 2643 __dev_set_mtu(slave_dev, slave->original_mtu); 2644 else 2645 dev_set_mtu(slave_dev, slave->original_mtu); 2646 2647 if (!netif_is_bond_master(slave_dev)) 2648 slave_dev->priv_flags &= ~IFF_BONDING; 2649 2650 bond_xdp_set_features(bond_dev); 2651 kobject_put(&slave->kobj); 2652 2653 return 0; 2654 } 2655 2656 /* A wrapper used because of ndo_del_link */ 2657 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2658 { 2659 return __bond_release_one(bond_dev, slave_dev, false, false); 2660 } 2661 2662 /* First release a slave and then destroy the bond if no more slaves are left. 2663 * Must be under rtnl_lock when this function is called. 2664 */ 2665 static int bond_release_and_destroy(struct net_device *bond_dev, 2666 struct net_device *slave_dev) 2667 { 2668 struct bonding *bond = netdev_priv(bond_dev); 2669 int ret; 2670 2671 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2672 if (ret == 0 && !bond_has_slaves(bond) && 2673 bond_dev->reg_state != NETREG_UNREGISTERING) { 2674 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2675 netdev_info(bond_dev, "Destroying bond\n"); 2676 bond_remove_proc_entry(bond); 2677 unregister_netdevice(bond_dev); 2678 } 2679 return ret; 2680 } 2681 2682 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2683 { 2684 struct bonding *bond = netdev_priv(bond_dev); 2685 2686 bond_fill_ifbond(bond, info); 2687 } 2688 2689 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2690 { 2691 struct bonding *bond = netdev_priv(bond_dev); 2692 struct list_head *iter; 2693 int i = 0, res = -ENODEV; 2694 struct slave *slave; 2695 2696 bond_for_each_slave(bond, slave, iter) { 2697 if (i++ == (int)info->slave_id) { 2698 res = 0; 2699 bond_fill_ifslave(slave, info); 2700 break; 2701 } 2702 } 2703 2704 return res; 2705 } 2706 2707 /*-------------------------------- Monitoring -------------------------------*/ 2708 2709 /* called with rcu_read_lock() */ 2710 static int bond_miimon_inspect(struct bonding *bond) 2711 { 2712 bool ignore_updelay = false; 2713 int link_state, commit = 0; 2714 struct list_head *iter; 2715 struct slave *slave; 2716 2717 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2718 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2719 } else { 2720 struct bond_up_slave *usable_slaves; 2721 2722 usable_slaves = rcu_dereference(bond->usable_slaves); 2723 2724 if (usable_slaves && usable_slaves->count == 0) 2725 ignore_updelay = true; 2726 } 2727 2728 bond_for_each_slave_rcu(bond, slave, iter) { 2729 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2730 2731 link_state = bond_check_dev_link(bond, slave->dev, 0); 2732 2733 switch (slave->link) { 2734 case BOND_LINK_UP: 2735 if (link_state) 2736 continue; 2737 2738 bond_propose_link_state(slave, BOND_LINK_FAIL); 2739 commit++; 2740 slave->delay = bond->params.downdelay; 2741 if (slave->delay && net_ratelimit()) { 2742 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2743 (BOND_MODE(bond) == 2744 BOND_MODE_ACTIVEBACKUP) ? 2745 (bond_is_active_slave(slave) ? 2746 "active " : "backup ") : "", 2747 bond->params.downdelay * bond->params.miimon); 2748 } 2749 fallthrough; 2750 case BOND_LINK_FAIL: 2751 if (link_state) { 2752 /* recovered before downdelay expired */ 2753 bond_propose_link_state(slave, BOND_LINK_UP); 2754 slave->last_link_up = jiffies; 2755 if (net_ratelimit()) 2756 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2757 (bond->params.downdelay - slave->delay) * 2758 bond->params.miimon); 2759 commit++; 2760 continue; 2761 } 2762 2763 if (slave->delay <= 0) { 2764 bond_propose_link_state(slave, BOND_LINK_DOWN); 2765 commit++; 2766 continue; 2767 } 2768 2769 slave->delay--; 2770 break; 2771 2772 case BOND_LINK_DOWN: 2773 if (!link_state) 2774 continue; 2775 2776 bond_propose_link_state(slave, BOND_LINK_BACK); 2777 commit++; 2778 slave->delay = bond->params.updelay; 2779 2780 if (slave->delay && net_ratelimit()) { 2781 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2782 ignore_updelay ? 0 : 2783 bond->params.updelay * 2784 bond->params.miimon); 2785 } 2786 fallthrough; 2787 case BOND_LINK_BACK: 2788 if (!link_state) { 2789 bond_propose_link_state(slave, BOND_LINK_DOWN); 2790 if (net_ratelimit()) 2791 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2792 (bond->params.updelay - slave->delay) * 2793 bond->params.miimon); 2794 commit++; 2795 continue; 2796 } 2797 2798 if (ignore_updelay) 2799 slave->delay = 0; 2800 2801 if (slave->delay <= 0) { 2802 bond_propose_link_state(slave, BOND_LINK_UP); 2803 commit++; 2804 ignore_updelay = false; 2805 continue; 2806 } 2807 2808 slave->delay--; 2809 break; 2810 } 2811 } 2812 2813 return commit; 2814 } 2815 2816 static void bond_miimon_link_change(struct bonding *bond, 2817 struct slave *slave, 2818 char link) 2819 { 2820 switch (BOND_MODE(bond)) { 2821 case BOND_MODE_8023AD: 2822 bond_3ad_handle_link_change(slave, link); 2823 break; 2824 case BOND_MODE_TLB: 2825 case BOND_MODE_ALB: 2826 bond_alb_handle_link_change(bond, slave, link); 2827 break; 2828 case BOND_MODE_XOR: 2829 bond_update_slave_arr(bond, NULL); 2830 break; 2831 } 2832 } 2833 2834 static void bond_miimon_commit(struct bonding *bond) 2835 { 2836 struct slave *slave, *primary, *active; 2837 bool do_failover = false; 2838 struct list_head *iter; 2839 2840 ASSERT_RTNL(); 2841 2842 bond_for_each_slave(bond, slave, iter) { 2843 switch (slave->link_new_state) { 2844 case BOND_LINK_NOCHANGE: 2845 /* For 802.3ad mode, check current slave speed and 2846 * duplex again in case its port was disabled after 2847 * invalid speed/duplex reporting but recovered before 2848 * link monitoring could make a decision on the actual 2849 * link status 2850 */ 2851 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2852 slave->link == BOND_LINK_UP) 2853 bond_3ad_adapter_speed_duplex_changed(slave); 2854 continue; 2855 2856 case BOND_LINK_UP: 2857 if (bond_update_speed_duplex(slave) && 2858 bond_needs_speed_duplex(bond)) { 2859 slave->link = BOND_LINK_DOWN; 2860 if (net_ratelimit()) 2861 slave_warn(bond->dev, slave->dev, 2862 "failed to get link speed/duplex\n"); 2863 continue; 2864 } 2865 bond_set_slave_link_state(slave, BOND_LINK_UP, 2866 BOND_SLAVE_NOTIFY_NOW); 2867 slave->last_link_up = jiffies; 2868 2869 primary = rtnl_dereference(bond->primary_slave); 2870 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2871 /* prevent it from being the active one */ 2872 bond_set_backup_slave(slave); 2873 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2874 /* make it immediately active */ 2875 bond_set_active_slave(slave); 2876 } 2877 2878 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2879 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2880 slave->duplex ? "full" : "half"); 2881 2882 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2883 2884 active = rtnl_dereference(bond->curr_active_slave); 2885 if (!active || slave == primary || slave->prio > active->prio) 2886 do_failover = true; 2887 2888 continue; 2889 2890 case BOND_LINK_DOWN: 2891 if (slave->link_failure_count < UINT_MAX) 2892 slave->link_failure_count++; 2893 2894 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2895 BOND_SLAVE_NOTIFY_NOW); 2896 2897 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2898 BOND_MODE(bond) == BOND_MODE_8023AD) 2899 bond_set_slave_inactive_flags(slave, 2900 BOND_SLAVE_NOTIFY_NOW); 2901 2902 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2903 2904 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2905 2906 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2907 do_failover = true; 2908 2909 continue; 2910 2911 default: 2912 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2913 slave->link_new_state); 2914 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2915 2916 continue; 2917 } 2918 } 2919 2920 if (do_failover) { 2921 block_netpoll_tx(); 2922 bond_select_active_slave(bond); 2923 unblock_netpoll_tx(); 2924 } 2925 2926 bond_set_carrier(bond); 2927 } 2928 2929 /* bond_mii_monitor 2930 * 2931 * Really a wrapper that splits the mii monitor into two phases: an 2932 * inspection, then (if inspection indicates something needs to be done) 2933 * an acquisition of appropriate locks followed by a commit phase to 2934 * implement whatever link state changes are indicated. 2935 */ 2936 static void bond_mii_monitor(struct work_struct *work) 2937 { 2938 struct bonding *bond = container_of(work, struct bonding, 2939 mii_work.work); 2940 bool should_notify_peers = false; 2941 bool commit; 2942 unsigned long delay; 2943 struct slave *slave; 2944 struct list_head *iter; 2945 2946 delay = msecs_to_jiffies(bond->params.miimon); 2947 2948 if (!bond_has_slaves(bond)) 2949 goto re_arm; 2950 2951 rcu_read_lock(); 2952 should_notify_peers = bond_should_notify_peers(bond); 2953 commit = !!bond_miimon_inspect(bond); 2954 if (bond->send_peer_notif) { 2955 rcu_read_unlock(); 2956 if (rtnl_trylock()) { 2957 bond->send_peer_notif--; 2958 rtnl_unlock(); 2959 } 2960 } else { 2961 rcu_read_unlock(); 2962 } 2963 2964 if (commit) { 2965 /* Race avoidance with bond_close cancel of workqueue */ 2966 if (!rtnl_trylock()) { 2967 delay = 1; 2968 should_notify_peers = false; 2969 goto re_arm; 2970 } 2971 2972 bond_for_each_slave(bond, slave, iter) { 2973 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2974 } 2975 bond_miimon_commit(bond); 2976 2977 rtnl_unlock(); /* might sleep, hold no other locks */ 2978 } 2979 2980 re_arm: 2981 if (bond->params.miimon) 2982 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2983 2984 if (should_notify_peers) { 2985 if (!rtnl_trylock()) 2986 return; 2987 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2988 rtnl_unlock(); 2989 } 2990 } 2991 2992 static int bond_upper_dev_walk(struct net_device *upper, 2993 struct netdev_nested_priv *priv) 2994 { 2995 __be32 ip = *(__be32 *)priv->data; 2996 2997 return ip == bond_confirm_addr(upper, 0, ip); 2998 } 2999 3000 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 3001 { 3002 struct netdev_nested_priv priv = { 3003 .data = (void *)&ip, 3004 }; 3005 bool ret = false; 3006 3007 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 3008 return true; 3009 3010 rcu_read_lock(); 3011 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 3012 ret = true; 3013 rcu_read_unlock(); 3014 3015 return ret; 3016 } 3017 3018 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff) 3019 3020 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 3021 struct sk_buff *skb) 3022 { 3023 struct net_device *bond_dev = slave->bond->dev; 3024 struct net_device *slave_dev = slave->dev; 3025 struct bond_vlan_tag *outer_tag = tags; 3026 3027 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE) 3028 return true; 3029 3030 tags++; 3031 3032 /* Go through all the tags backwards and add them to the packet */ 3033 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) { 3034 if (!tags->vlan_id) { 3035 tags++; 3036 continue; 3037 } 3038 3039 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 3040 ntohs(outer_tag->vlan_proto), tags->vlan_id); 3041 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 3042 tags->vlan_id); 3043 if (!skb) { 3044 net_err_ratelimited("failed to insert inner VLAN tag\n"); 3045 return false; 3046 } 3047 3048 tags++; 3049 } 3050 /* Set the outer tag */ 3051 if (outer_tag->vlan_id) { 3052 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 3053 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 3054 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 3055 outer_tag->vlan_id); 3056 } 3057 3058 return true; 3059 } 3060 3061 /* We go to the (large) trouble of VLAN tagging ARP frames because 3062 * switches in VLAN mode (especially if ports are configured as 3063 * "native" to a VLAN) might not pass non-tagged frames. 3064 */ 3065 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 3066 __be32 src_ip, struct bond_vlan_tag *tags) 3067 { 3068 struct net_device *bond_dev = slave->bond->dev; 3069 struct net_device *slave_dev = slave->dev; 3070 struct sk_buff *skb; 3071 3072 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 3073 arp_op, &dest_ip, &src_ip); 3074 3075 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 3076 NULL, slave_dev->dev_addr, NULL); 3077 3078 if (!skb) { 3079 net_err_ratelimited("ARP packet allocation failed\n"); 3080 return; 3081 } 3082 3083 if (bond_handle_vlan(slave, tags, skb)) { 3084 slave_update_last_tx(slave); 3085 arp_xmit(skb); 3086 } 3087 3088 return; 3089 } 3090 3091 /* Validate the device path between the @start_dev and the @end_dev. 3092 * The path is valid if the @end_dev is reachable through device 3093 * stacking. 3094 * When the path is validated, collect any vlan information in the 3095 * path. 3096 */ 3097 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 3098 struct net_device *end_dev, 3099 int level) 3100 { 3101 struct bond_vlan_tag *tags; 3102 struct net_device *upper; 3103 struct list_head *iter; 3104 3105 if (start_dev == end_dev) { 3106 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 3107 if (!tags) 3108 return ERR_PTR(-ENOMEM); 3109 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE; 3110 return tags; 3111 } 3112 3113 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 3114 tags = bond_verify_device_path(upper, end_dev, level + 1); 3115 if (IS_ERR_OR_NULL(tags)) { 3116 if (IS_ERR(tags)) 3117 return tags; 3118 continue; 3119 } 3120 if (is_vlan_dev(upper)) { 3121 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 3122 tags[level].vlan_id = vlan_dev_vlan_id(upper); 3123 } 3124 3125 return tags; 3126 } 3127 3128 return NULL; 3129 } 3130 3131 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 3132 { 3133 struct rtable *rt; 3134 struct bond_vlan_tag *tags; 3135 __be32 *targets = bond->params.arp_targets, addr; 3136 int i; 3137 3138 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 3139 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 3140 __func__, &targets[i]); 3141 tags = NULL; 3142 3143 /* Find out through which dev should the packet go */ 3144 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0, 3145 RT_SCOPE_LINK); 3146 if (IS_ERR(rt)) { 3147 /* there's no route to target - try to send arp 3148 * probe to generate any traffic (arp_validate=0) 3149 */ 3150 if (bond->params.arp_validate) 3151 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 3152 bond->dev->name, 3153 &targets[i]); 3154 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 3155 0, tags); 3156 continue; 3157 } 3158 3159 /* bond device itself */ 3160 if (rt->dst.dev == bond->dev) 3161 goto found; 3162 3163 rcu_read_lock(); 3164 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 3165 rcu_read_unlock(); 3166 3167 if (!IS_ERR_OR_NULL(tags)) 3168 goto found; 3169 3170 /* Not our device - skip */ 3171 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 3172 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 3173 3174 ip_rt_put(rt); 3175 continue; 3176 3177 found: 3178 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 3179 ip_rt_put(rt); 3180 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 3181 kfree(tags); 3182 } 3183 } 3184 3185 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3186 { 3187 int i; 3188 3189 if (!sip || !bond_has_this_ip(bond, tip)) { 3190 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3191 __func__, &sip, &tip); 3192 return; 3193 } 3194 3195 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3196 if (i == -1) { 3197 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3198 __func__, &sip); 3199 return; 3200 } 3201 slave->last_rx = jiffies; 3202 slave->target_last_arp_rx[i] = jiffies; 3203 } 3204 3205 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3206 struct slave *slave) 3207 { 3208 struct arphdr *arp = (struct arphdr *)skb->data; 3209 struct slave *curr_active_slave, *curr_arp_slave; 3210 unsigned char *arp_ptr; 3211 __be32 sip, tip; 3212 unsigned int alen; 3213 3214 alen = arp_hdr_len(bond->dev); 3215 3216 if (alen > skb_headlen(skb)) { 3217 arp = kmalloc(alen, GFP_ATOMIC); 3218 if (!arp) 3219 goto out_unlock; 3220 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3221 goto out_unlock; 3222 } 3223 3224 if (arp->ar_hln != bond->dev->addr_len || 3225 skb->pkt_type == PACKET_OTHERHOST || 3226 skb->pkt_type == PACKET_LOOPBACK || 3227 arp->ar_hrd != htons(ARPHRD_ETHER) || 3228 arp->ar_pro != htons(ETH_P_IP) || 3229 arp->ar_pln != 4) 3230 goto out_unlock; 3231 3232 arp_ptr = (unsigned char *)(arp + 1); 3233 arp_ptr += bond->dev->addr_len; 3234 memcpy(&sip, arp_ptr, 4); 3235 arp_ptr += 4 + bond->dev->addr_len; 3236 memcpy(&tip, arp_ptr, 4); 3237 3238 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3239 __func__, slave->dev->name, bond_slave_state(slave), 3240 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3241 &sip, &tip); 3242 3243 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3244 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3245 3246 /* We 'trust' the received ARP enough to validate it if: 3247 * 3248 * (a) the slave receiving the ARP is active (which includes the 3249 * current ARP slave, if any), or 3250 * 3251 * (b) the receiving slave isn't active, but there is a currently 3252 * active slave and it received valid arp reply(s) after it became 3253 * the currently active slave, or 3254 * 3255 * (c) there is an ARP slave that sent an ARP during the prior ARP 3256 * interval, and we receive an ARP reply on any slave. We accept 3257 * these because switch FDB update delays may deliver the ARP 3258 * reply to a slave other than the sender of the ARP request. 3259 * 3260 * Note: for (b), backup slaves are receiving the broadcast ARP 3261 * request, not a reply. This request passes from the sending 3262 * slave through the L2 switch(es) to the receiving slave. Since 3263 * this is checking the request, sip/tip are swapped for 3264 * validation. 3265 * 3266 * This is done to avoid endless looping when we can't reach the 3267 * arp_ip_target and fool ourselves with our own arp requests. 3268 */ 3269 if (bond_is_active_slave(slave)) 3270 bond_validate_arp(bond, slave, sip, tip); 3271 else if (curr_active_slave && 3272 time_after(slave_last_rx(bond, curr_active_slave), 3273 curr_active_slave->last_link_up)) 3274 bond_validate_arp(bond, slave, tip, sip); 3275 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3276 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3277 bond_validate_arp(bond, slave, sip, tip); 3278 3279 out_unlock: 3280 if (arp != (struct arphdr *)skb->data) 3281 kfree(arp); 3282 return RX_HANDLER_ANOTHER; 3283 } 3284 3285 #if IS_ENABLED(CONFIG_IPV6) 3286 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3287 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3288 { 3289 struct net_device *bond_dev = slave->bond->dev; 3290 struct net_device *slave_dev = slave->dev; 3291 struct in6_addr mcaddr; 3292 struct sk_buff *skb; 3293 3294 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3295 daddr, saddr); 3296 3297 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3298 if (!skb) { 3299 net_err_ratelimited("NS packet allocation failed\n"); 3300 return; 3301 } 3302 3303 addrconf_addr_solict_mult(daddr, &mcaddr); 3304 if (bond_handle_vlan(slave, tags, skb)) { 3305 slave_update_last_tx(slave); 3306 ndisc_send_skb(skb, &mcaddr, saddr); 3307 } 3308 } 3309 3310 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3311 { 3312 struct in6_addr *targets = bond->params.ns_targets; 3313 struct bond_vlan_tag *tags; 3314 struct dst_entry *dst; 3315 struct in6_addr saddr; 3316 struct flowi6 fl6; 3317 int i; 3318 3319 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3320 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3321 __func__, &targets[i]); 3322 tags = NULL; 3323 3324 /* Find out through which dev should the packet go */ 3325 memset(&fl6, 0, sizeof(struct flowi6)); 3326 fl6.daddr = targets[i]; 3327 fl6.flowi6_oif = bond->dev->ifindex; 3328 3329 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3330 if (dst->error) { 3331 dst_release(dst); 3332 /* there's no route to target - try to send arp 3333 * probe to generate any traffic (arp_validate=0) 3334 */ 3335 if (bond->params.arp_validate) 3336 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3337 bond->dev->name, 3338 &targets[i]); 3339 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3340 continue; 3341 } 3342 3343 /* bond device itself */ 3344 if (dst->dev == bond->dev) 3345 goto found; 3346 3347 rcu_read_lock(); 3348 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3349 rcu_read_unlock(); 3350 3351 if (!IS_ERR_OR_NULL(tags)) 3352 goto found; 3353 3354 /* Not our device - skip */ 3355 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3356 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3357 3358 dst_release(dst); 3359 continue; 3360 3361 found: 3362 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3363 bond_ns_send(slave, &targets[i], &saddr, tags); 3364 else 3365 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3366 3367 dst_release(dst); 3368 kfree(tags); 3369 } 3370 } 3371 3372 static int bond_confirm_addr6(struct net_device *dev, 3373 struct netdev_nested_priv *priv) 3374 { 3375 struct in6_addr *addr = (struct in6_addr *)priv->data; 3376 3377 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3378 } 3379 3380 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3381 { 3382 struct netdev_nested_priv priv = { 3383 .data = addr, 3384 }; 3385 int ret = false; 3386 3387 if (bond_confirm_addr6(bond->dev, &priv)) 3388 return true; 3389 3390 rcu_read_lock(); 3391 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3392 ret = true; 3393 rcu_read_unlock(); 3394 3395 return ret; 3396 } 3397 3398 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3399 struct in6_addr *saddr, struct in6_addr *daddr) 3400 { 3401 int i; 3402 3403 /* Ignore NAs that: 3404 * 1. Source address is unspecified address. 3405 * 2. Dest address is neither all-nodes multicast address nor 3406 * exist on bond interface. 3407 */ 3408 if (ipv6_addr_any(saddr) || 3409 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3410 !bond_has_this_ip6(bond, daddr))) { 3411 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3412 __func__, saddr, daddr); 3413 return; 3414 } 3415 3416 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3417 if (i == -1) { 3418 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3419 __func__, saddr); 3420 return; 3421 } 3422 slave->last_rx = jiffies; 3423 slave->target_last_arp_rx[i] = jiffies; 3424 } 3425 3426 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3427 struct slave *slave) 3428 { 3429 struct slave *curr_active_slave, *curr_arp_slave; 3430 struct in6_addr *saddr, *daddr; 3431 struct { 3432 struct ipv6hdr ip6; 3433 struct icmp6hdr icmp6; 3434 } *combined, _combined; 3435 3436 if (skb->pkt_type == PACKET_OTHERHOST || 3437 skb->pkt_type == PACKET_LOOPBACK) 3438 goto out; 3439 3440 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3441 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3442 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION && 3443 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)) 3444 goto out; 3445 3446 saddr = &combined->ip6.saddr; 3447 daddr = &combined->ip6.daddr; 3448 3449 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3450 __func__, slave->dev->name, bond_slave_state(slave), 3451 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3452 saddr, daddr); 3453 3454 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3455 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3456 3457 /* We 'trust' the received ARP enough to validate it if: 3458 * see bond_arp_rcv(). 3459 */ 3460 if (bond_is_active_slave(slave)) 3461 bond_validate_na(bond, slave, saddr, daddr); 3462 else if (curr_active_slave && 3463 time_after(slave_last_rx(bond, curr_active_slave), 3464 curr_active_slave->last_link_up)) 3465 bond_validate_na(bond, slave, daddr, saddr); 3466 else if (curr_arp_slave && 3467 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3468 bond_validate_na(bond, slave, saddr, daddr); 3469 3470 out: 3471 return RX_HANDLER_ANOTHER; 3472 } 3473 #endif 3474 3475 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3476 struct slave *slave) 3477 { 3478 #if IS_ENABLED(CONFIG_IPV6) 3479 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3480 #endif 3481 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3482 3483 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3484 __func__, skb->dev->name); 3485 3486 /* Use arp validate logic for both ARP and NS */ 3487 if (!slave_do_arp_validate(bond, slave)) { 3488 if ((slave_do_arp_validate_only(bond) && is_arp) || 3489 #if IS_ENABLED(CONFIG_IPV6) 3490 (slave_do_arp_validate_only(bond) && is_ipv6) || 3491 #endif 3492 !slave_do_arp_validate_only(bond)) 3493 slave->last_rx = jiffies; 3494 return RX_HANDLER_ANOTHER; 3495 } else if (is_arp) { 3496 return bond_arp_rcv(skb, bond, slave); 3497 #if IS_ENABLED(CONFIG_IPV6) 3498 } else if (is_ipv6) { 3499 return bond_na_rcv(skb, bond, slave); 3500 #endif 3501 } else { 3502 return RX_HANDLER_ANOTHER; 3503 } 3504 } 3505 3506 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3507 { 3508 bond_arp_send_all(bond, slave); 3509 #if IS_ENABLED(CONFIG_IPV6) 3510 bond_ns_send_all(bond, slave); 3511 #endif 3512 } 3513 3514 /* function to verify if we're in the arp_interval timeslice, returns true if 3515 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3516 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3517 */ 3518 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3519 int mod) 3520 { 3521 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3522 3523 return time_in_range(jiffies, 3524 last_act - delta_in_ticks, 3525 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3526 } 3527 3528 /* This function is called regularly to monitor each slave's link 3529 * ensuring that traffic is being sent and received when arp monitoring 3530 * is used in load-balancing mode. if the adapter has been dormant, then an 3531 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3532 * arp monitoring in active backup mode. 3533 */ 3534 static void bond_loadbalance_arp_mon(struct bonding *bond) 3535 { 3536 struct slave *slave, *oldcurrent; 3537 struct list_head *iter; 3538 int do_failover = 0, slave_state_changed = 0; 3539 3540 if (!bond_has_slaves(bond)) 3541 goto re_arm; 3542 3543 rcu_read_lock(); 3544 3545 oldcurrent = rcu_dereference(bond->curr_active_slave); 3546 /* see if any of the previous devices are up now (i.e. they have 3547 * xmt and rcv traffic). the curr_active_slave does not come into 3548 * the picture unless it is null. also, slave->last_link_up is not 3549 * needed here because we send an arp on each slave and give a slave 3550 * as long as it needs to get the tx/rx within the delta. 3551 * TODO: what about up/down delay in arp mode? it wasn't here before 3552 * so it can wait 3553 */ 3554 bond_for_each_slave_rcu(bond, slave, iter) { 3555 unsigned long last_tx = slave_last_tx(slave); 3556 3557 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3558 3559 if (slave->link != BOND_LINK_UP) { 3560 if (bond_time_in_interval(bond, last_tx, 1) && 3561 bond_time_in_interval(bond, slave->last_rx, 1)) { 3562 3563 bond_propose_link_state(slave, BOND_LINK_UP); 3564 slave_state_changed = 1; 3565 3566 /* primary_slave has no meaning in round-robin 3567 * mode. the window of a slave being up and 3568 * curr_active_slave being null after enslaving 3569 * is closed. 3570 */ 3571 if (!oldcurrent) { 3572 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3573 do_failover = 1; 3574 } else { 3575 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3576 } 3577 } 3578 } else { 3579 /* slave->link == BOND_LINK_UP */ 3580 3581 /* not all switches will respond to an arp request 3582 * when the source ip is 0, so don't take the link down 3583 * if we don't know our ip yet 3584 */ 3585 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3586 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3587 3588 bond_propose_link_state(slave, BOND_LINK_DOWN); 3589 slave_state_changed = 1; 3590 3591 if (slave->link_failure_count < UINT_MAX) 3592 slave->link_failure_count++; 3593 3594 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3595 3596 if (slave == oldcurrent) 3597 do_failover = 1; 3598 } 3599 } 3600 3601 /* note: if switch is in round-robin mode, all links 3602 * must tx arp to ensure all links rx an arp - otherwise 3603 * links may oscillate or not come up at all; if switch is 3604 * in something like xor mode, there is nothing we can 3605 * do - all replies will be rx'ed on same link causing slaves 3606 * to be unstable during low/no traffic periods 3607 */ 3608 if (bond_slave_is_up(slave)) 3609 bond_send_validate(bond, slave); 3610 } 3611 3612 rcu_read_unlock(); 3613 3614 if (do_failover || slave_state_changed) { 3615 if (!rtnl_trylock()) 3616 goto re_arm; 3617 3618 bond_for_each_slave(bond, slave, iter) { 3619 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3620 slave->link = slave->link_new_state; 3621 } 3622 3623 if (slave_state_changed) { 3624 bond_slave_state_change(bond); 3625 if (BOND_MODE(bond) == BOND_MODE_XOR) 3626 bond_update_slave_arr(bond, NULL); 3627 } 3628 if (do_failover) { 3629 block_netpoll_tx(); 3630 bond_select_active_slave(bond); 3631 unblock_netpoll_tx(); 3632 } 3633 rtnl_unlock(); 3634 } 3635 3636 re_arm: 3637 if (bond->params.arp_interval) 3638 queue_delayed_work(bond->wq, &bond->arp_work, 3639 msecs_to_jiffies(bond->params.arp_interval)); 3640 } 3641 3642 /* Called to inspect slaves for active-backup mode ARP monitor link state 3643 * changes. Sets proposed link state in slaves to specify what action 3644 * should take place for the slave. Returns 0 if no changes are found, >0 3645 * if changes to link states must be committed. 3646 * 3647 * Called with rcu_read_lock held. 3648 */ 3649 static int bond_ab_arp_inspect(struct bonding *bond) 3650 { 3651 unsigned long last_tx, last_rx; 3652 struct list_head *iter; 3653 struct slave *slave; 3654 int commit = 0; 3655 3656 bond_for_each_slave_rcu(bond, slave, iter) { 3657 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3658 last_rx = slave_last_rx(bond, slave); 3659 3660 if (slave->link != BOND_LINK_UP) { 3661 if (bond_time_in_interval(bond, last_rx, 1)) { 3662 bond_propose_link_state(slave, BOND_LINK_UP); 3663 commit++; 3664 } else if (slave->link == BOND_LINK_BACK) { 3665 bond_propose_link_state(slave, BOND_LINK_FAIL); 3666 commit++; 3667 } 3668 continue; 3669 } 3670 3671 /* Give slaves 2*delta after being enslaved or made 3672 * active. This avoids bouncing, as the last receive 3673 * times need a full ARP monitor cycle to be updated. 3674 */ 3675 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3676 continue; 3677 3678 /* Backup slave is down if: 3679 * - No current_arp_slave AND 3680 * - more than (missed_max+1)*delta since last receive AND 3681 * - the bond has an IP address 3682 * 3683 * Note: a non-null current_arp_slave indicates 3684 * the curr_active_slave went down and we are 3685 * searching for a new one; under this condition 3686 * we only take the curr_active_slave down - this 3687 * gives each slave a chance to tx/rx traffic 3688 * before being taken out 3689 */ 3690 if (!bond_is_active_slave(slave) && 3691 !rcu_access_pointer(bond->current_arp_slave) && 3692 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3693 bond_propose_link_state(slave, BOND_LINK_DOWN); 3694 commit++; 3695 } 3696 3697 /* Active slave is down if: 3698 * - more than missed_max*delta since transmitting OR 3699 * - (more than missed_max*delta since receive AND 3700 * the bond has an IP address) 3701 */ 3702 last_tx = slave_last_tx(slave); 3703 if (bond_is_active_slave(slave) && 3704 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3705 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3706 bond_propose_link_state(slave, BOND_LINK_DOWN); 3707 commit++; 3708 } 3709 } 3710 3711 return commit; 3712 } 3713 3714 /* Called to commit link state changes noted by inspection step of 3715 * active-backup mode ARP monitor. 3716 * 3717 * Called with RTNL hold. 3718 */ 3719 static void bond_ab_arp_commit(struct bonding *bond) 3720 { 3721 bool do_failover = false; 3722 struct list_head *iter; 3723 unsigned long last_tx; 3724 struct slave *slave; 3725 3726 bond_for_each_slave(bond, slave, iter) { 3727 switch (slave->link_new_state) { 3728 case BOND_LINK_NOCHANGE: 3729 continue; 3730 3731 case BOND_LINK_UP: 3732 last_tx = slave_last_tx(slave); 3733 if (rtnl_dereference(bond->curr_active_slave) != slave || 3734 (!rtnl_dereference(bond->curr_active_slave) && 3735 bond_time_in_interval(bond, last_tx, 1))) { 3736 struct slave *current_arp_slave; 3737 3738 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3739 bond_set_slave_link_state(slave, BOND_LINK_UP, 3740 BOND_SLAVE_NOTIFY_NOW); 3741 if (current_arp_slave) { 3742 bond_set_slave_inactive_flags( 3743 current_arp_slave, 3744 BOND_SLAVE_NOTIFY_NOW); 3745 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3746 } 3747 3748 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3749 3750 if (!rtnl_dereference(bond->curr_active_slave) || 3751 slave == rtnl_dereference(bond->primary_slave) || 3752 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio) 3753 do_failover = true; 3754 3755 } 3756 3757 continue; 3758 3759 case BOND_LINK_DOWN: 3760 if (slave->link_failure_count < UINT_MAX) 3761 slave->link_failure_count++; 3762 3763 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3764 BOND_SLAVE_NOTIFY_NOW); 3765 bond_set_slave_inactive_flags(slave, 3766 BOND_SLAVE_NOTIFY_NOW); 3767 3768 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3769 3770 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3771 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3772 do_failover = true; 3773 } 3774 3775 continue; 3776 3777 case BOND_LINK_FAIL: 3778 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3779 BOND_SLAVE_NOTIFY_NOW); 3780 bond_set_slave_inactive_flags(slave, 3781 BOND_SLAVE_NOTIFY_NOW); 3782 3783 /* A slave has just been enslaved and has become 3784 * the current active slave. 3785 */ 3786 if (rtnl_dereference(bond->curr_active_slave)) 3787 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3788 continue; 3789 3790 default: 3791 slave_err(bond->dev, slave->dev, 3792 "impossible: link_new_state %d on slave\n", 3793 slave->link_new_state); 3794 continue; 3795 } 3796 } 3797 3798 if (do_failover) { 3799 block_netpoll_tx(); 3800 bond_select_active_slave(bond); 3801 unblock_netpoll_tx(); 3802 } 3803 3804 bond_set_carrier(bond); 3805 } 3806 3807 /* Send ARP probes for active-backup mode ARP monitor. 3808 * 3809 * Called with rcu_read_lock held. 3810 */ 3811 static bool bond_ab_arp_probe(struct bonding *bond) 3812 { 3813 struct slave *slave, *before = NULL, *new_slave = NULL, 3814 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3815 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3816 struct list_head *iter; 3817 bool found = false; 3818 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3819 3820 if (curr_arp_slave && curr_active_slave) 3821 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3822 curr_arp_slave->dev->name, 3823 curr_active_slave->dev->name); 3824 3825 if (curr_active_slave) { 3826 bond_send_validate(bond, curr_active_slave); 3827 return should_notify_rtnl; 3828 } 3829 3830 /* if we don't have a curr_active_slave, search for the next available 3831 * backup slave from the current_arp_slave and make it the candidate 3832 * for becoming the curr_active_slave 3833 */ 3834 3835 if (!curr_arp_slave) { 3836 curr_arp_slave = bond_first_slave_rcu(bond); 3837 if (!curr_arp_slave) 3838 return should_notify_rtnl; 3839 } 3840 3841 bond_for_each_slave_rcu(bond, slave, iter) { 3842 if (!found && !before && bond_slave_is_up(slave)) 3843 before = slave; 3844 3845 if (found && !new_slave && bond_slave_is_up(slave)) 3846 new_slave = slave; 3847 /* if the link state is up at this point, we 3848 * mark it down - this can happen if we have 3849 * simultaneous link failures and 3850 * reselect_active_interface doesn't make this 3851 * one the current slave so it is still marked 3852 * up when it is actually down 3853 */ 3854 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3855 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3856 BOND_SLAVE_NOTIFY_LATER); 3857 if (slave->link_failure_count < UINT_MAX) 3858 slave->link_failure_count++; 3859 3860 bond_set_slave_inactive_flags(slave, 3861 BOND_SLAVE_NOTIFY_LATER); 3862 3863 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3864 } 3865 if (slave == curr_arp_slave) 3866 found = true; 3867 } 3868 3869 if (!new_slave && before) 3870 new_slave = before; 3871 3872 if (!new_slave) 3873 goto check_state; 3874 3875 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3876 BOND_SLAVE_NOTIFY_LATER); 3877 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3878 bond_send_validate(bond, new_slave); 3879 new_slave->last_link_up = jiffies; 3880 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3881 3882 check_state: 3883 bond_for_each_slave_rcu(bond, slave, iter) { 3884 if (slave->should_notify || slave->should_notify_link) { 3885 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3886 break; 3887 } 3888 } 3889 return should_notify_rtnl; 3890 } 3891 3892 static void bond_activebackup_arp_mon(struct bonding *bond) 3893 { 3894 bool should_notify_peers = false; 3895 bool should_notify_rtnl = false; 3896 int delta_in_ticks; 3897 3898 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3899 3900 if (!bond_has_slaves(bond)) 3901 goto re_arm; 3902 3903 rcu_read_lock(); 3904 3905 should_notify_peers = bond_should_notify_peers(bond); 3906 3907 if (bond_ab_arp_inspect(bond)) { 3908 rcu_read_unlock(); 3909 3910 /* Race avoidance with bond_close flush of workqueue */ 3911 if (!rtnl_trylock()) { 3912 delta_in_ticks = 1; 3913 should_notify_peers = false; 3914 goto re_arm; 3915 } 3916 3917 bond_ab_arp_commit(bond); 3918 3919 rtnl_unlock(); 3920 rcu_read_lock(); 3921 } 3922 3923 should_notify_rtnl = bond_ab_arp_probe(bond); 3924 rcu_read_unlock(); 3925 3926 re_arm: 3927 if (bond->params.arp_interval) 3928 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3929 3930 if (should_notify_peers || should_notify_rtnl) { 3931 if (!rtnl_trylock()) 3932 return; 3933 3934 if (should_notify_peers) { 3935 bond->send_peer_notif--; 3936 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3937 bond->dev); 3938 } 3939 if (should_notify_rtnl) { 3940 bond_slave_state_notify(bond); 3941 bond_slave_link_notify(bond); 3942 } 3943 3944 rtnl_unlock(); 3945 } 3946 } 3947 3948 static void bond_arp_monitor(struct work_struct *work) 3949 { 3950 struct bonding *bond = container_of(work, struct bonding, 3951 arp_work.work); 3952 3953 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3954 bond_activebackup_arp_mon(bond); 3955 else 3956 bond_loadbalance_arp_mon(bond); 3957 } 3958 3959 /*-------------------------- netdev event handling --------------------------*/ 3960 3961 /* Change device name */ 3962 static int bond_event_changename(struct bonding *bond) 3963 { 3964 bond_remove_proc_entry(bond); 3965 bond_create_proc_entry(bond); 3966 3967 bond_debug_reregister(bond); 3968 3969 return NOTIFY_DONE; 3970 } 3971 3972 static int bond_master_netdev_event(unsigned long event, 3973 struct net_device *bond_dev) 3974 { 3975 struct bonding *event_bond = netdev_priv(bond_dev); 3976 3977 netdev_dbg(bond_dev, "%s called\n", __func__); 3978 3979 switch (event) { 3980 case NETDEV_CHANGENAME: 3981 return bond_event_changename(event_bond); 3982 case NETDEV_UNREGISTER: 3983 bond_remove_proc_entry(event_bond); 3984 #ifdef CONFIG_XFRM_OFFLOAD 3985 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3986 #endif /* CONFIG_XFRM_OFFLOAD */ 3987 break; 3988 case NETDEV_REGISTER: 3989 bond_create_proc_entry(event_bond); 3990 break; 3991 default: 3992 break; 3993 } 3994 3995 return NOTIFY_DONE; 3996 } 3997 3998 static int bond_slave_netdev_event(unsigned long event, 3999 struct net_device *slave_dev) 4000 { 4001 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 4002 struct bonding *bond; 4003 struct net_device *bond_dev; 4004 4005 /* A netdev event can be generated while enslaving a device 4006 * before netdev_rx_handler_register is called in which case 4007 * slave will be NULL 4008 */ 4009 if (!slave) { 4010 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 4011 return NOTIFY_DONE; 4012 } 4013 4014 bond_dev = slave->bond->dev; 4015 bond = slave->bond; 4016 primary = rtnl_dereference(bond->primary_slave); 4017 4018 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 4019 4020 switch (event) { 4021 case NETDEV_UNREGISTER: 4022 if (bond_dev->type != ARPHRD_ETHER) 4023 bond_release_and_destroy(bond_dev, slave_dev); 4024 else 4025 __bond_release_one(bond_dev, slave_dev, false, true); 4026 break; 4027 case NETDEV_UP: 4028 case NETDEV_CHANGE: 4029 /* For 802.3ad mode only: 4030 * Getting invalid Speed/Duplex values here will put slave 4031 * in weird state. Mark it as link-fail if the link was 4032 * previously up or link-down if it hasn't yet come up, and 4033 * let link-monitoring (miimon) set it right when correct 4034 * speeds/duplex are available. 4035 */ 4036 if (bond_update_speed_duplex(slave) && 4037 BOND_MODE(bond) == BOND_MODE_8023AD) { 4038 if (slave->last_link_up) 4039 slave->link = BOND_LINK_FAIL; 4040 else 4041 slave->link = BOND_LINK_DOWN; 4042 } 4043 4044 if (BOND_MODE(bond) == BOND_MODE_8023AD) 4045 bond_3ad_adapter_speed_duplex_changed(slave); 4046 fallthrough; 4047 case NETDEV_DOWN: 4048 /* Refresh slave-array if applicable! 4049 * If the setup does not use miimon or arpmon (mode-specific!), 4050 * then these events will not cause the slave-array to be 4051 * refreshed. This will cause xmit to use a slave that is not 4052 * usable. Avoid such situation by refeshing the array at these 4053 * events. If these (miimon/arpmon) parameters are configured 4054 * then array gets refreshed twice and that should be fine! 4055 */ 4056 if (bond_mode_can_use_xmit_hash(bond)) 4057 bond_update_slave_arr(bond, NULL); 4058 break; 4059 case NETDEV_CHANGEMTU: 4060 /* TODO: Should slaves be allowed to 4061 * independently alter their MTU? For 4062 * an active-backup bond, slaves need 4063 * not be the same type of device, so 4064 * MTUs may vary. For other modes, 4065 * slaves arguably should have the 4066 * same MTUs. To do this, we'd need to 4067 * take over the slave's change_mtu 4068 * function for the duration of their 4069 * servitude. 4070 */ 4071 break; 4072 case NETDEV_CHANGENAME: 4073 /* we don't care if we don't have primary set */ 4074 if (!bond_uses_primary(bond) || 4075 !bond->params.primary[0]) 4076 break; 4077 4078 if (slave == primary) { 4079 /* slave's name changed - he's no longer primary */ 4080 RCU_INIT_POINTER(bond->primary_slave, NULL); 4081 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 4082 /* we have a new primary slave */ 4083 rcu_assign_pointer(bond->primary_slave, slave); 4084 } else { /* we didn't change primary - exit */ 4085 break; 4086 } 4087 4088 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 4089 primary ? slave_dev->name : "none"); 4090 4091 block_netpoll_tx(); 4092 bond_select_active_slave(bond); 4093 unblock_netpoll_tx(); 4094 break; 4095 case NETDEV_FEAT_CHANGE: 4096 if (!bond->notifier_ctx) { 4097 bond->notifier_ctx = true; 4098 bond_compute_features(bond); 4099 bond->notifier_ctx = false; 4100 } 4101 break; 4102 case NETDEV_RESEND_IGMP: 4103 /* Propagate to master device */ 4104 call_netdevice_notifiers(event, slave->bond->dev); 4105 break; 4106 case NETDEV_XDP_FEAT_CHANGE: 4107 bond_xdp_set_features(bond_dev); 4108 break; 4109 default: 4110 break; 4111 } 4112 4113 return NOTIFY_DONE; 4114 } 4115 4116 /* bond_netdev_event: handle netdev notifier chain events. 4117 * 4118 * This function receives events for the netdev chain. The caller (an 4119 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 4120 * locks for us to safely manipulate the slave devices (RTNL lock, 4121 * dev_probe_lock). 4122 */ 4123 static int bond_netdev_event(struct notifier_block *this, 4124 unsigned long event, void *ptr) 4125 { 4126 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 4127 4128 netdev_dbg(event_dev, "%s received %s\n", 4129 __func__, netdev_cmd_to_name(event)); 4130 4131 if (!(event_dev->priv_flags & IFF_BONDING)) 4132 return NOTIFY_DONE; 4133 4134 if (event_dev->flags & IFF_MASTER) { 4135 int ret; 4136 4137 ret = bond_master_netdev_event(event, event_dev); 4138 if (ret != NOTIFY_DONE) 4139 return ret; 4140 } 4141 4142 if (event_dev->flags & IFF_SLAVE) 4143 return bond_slave_netdev_event(event, event_dev); 4144 4145 return NOTIFY_DONE; 4146 } 4147 4148 static struct notifier_block bond_netdev_notifier = { 4149 .notifier_call = bond_netdev_event, 4150 }; 4151 4152 /*---------------------------- Hashing Policies -----------------------------*/ 4153 4154 /* Helper to access data in a packet, with or without a backing skb. 4155 * If skb is given the data is linearized if necessary via pskb_may_pull. 4156 */ 4157 static inline const void *bond_pull_data(struct sk_buff *skb, 4158 const void *data, int hlen, int n) 4159 { 4160 if (likely(n <= hlen)) 4161 return data; 4162 else if (skb && likely(pskb_may_pull(skb, n))) 4163 return skb->data; 4164 4165 return NULL; 4166 } 4167 4168 /* L2 hash helper */ 4169 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4170 { 4171 struct ethhdr *ep; 4172 4173 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4174 if (!data) 4175 return 0; 4176 4177 ep = (struct ethhdr *)(data + mhoff); 4178 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 4179 } 4180 4181 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 4182 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 4183 { 4184 const struct ipv6hdr *iph6; 4185 const struct iphdr *iph; 4186 4187 if (l2_proto == htons(ETH_P_IP)) { 4188 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 4189 if (!data) 4190 return false; 4191 4192 iph = (const struct iphdr *)(data + *nhoff); 4193 iph_to_flow_copy_v4addrs(fk, iph); 4194 *nhoff += iph->ihl << 2; 4195 if (!ip_is_fragment(iph)) 4196 *ip_proto = iph->protocol; 4197 } else if (l2_proto == htons(ETH_P_IPV6)) { 4198 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4199 if (!data) 4200 return false; 4201 4202 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4203 iph_to_flow_copy_v6addrs(fk, iph6); 4204 *nhoff += sizeof(*iph6); 4205 *ip_proto = iph6->nexthdr; 4206 } else { 4207 return false; 4208 } 4209 4210 if (l34 && *ip_proto >= 0) 4211 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4212 4213 return true; 4214 } 4215 4216 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4217 { 4218 u32 srcmac_vendor = 0, srcmac_dev = 0; 4219 struct ethhdr *mac_hdr; 4220 u16 vlan = 0; 4221 int i; 4222 4223 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4224 if (!data) 4225 return 0; 4226 mac_hdr = (struct ethhdr *)(data + mhoff); 4227 4228 for (i = 0; i < 3; i++) 4229 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4230 4231 for (i = 3; i < ETH_ALEN; i++) 4232 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4233 4234 if (skb && skb_vlan_tag_present(skb)) 4235 vlan = skb_vlan_tag_get(skb); 4236 4237 return vlan ^ srcmac_vendor ^ srcmac_dev; 4238 } 4239 4240 /* Extract the appropriate headers based on bond's xmit policy */ 4241 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4242 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4243 { 4244 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4245 int ip_proto = -1; 4246 4247 switch (bond->params.xmit_policy) { 4248 case BOND_XMIT_POLICY_ENCAP23: 4249 case BOND_XMIT_POLICY_ENCAP34: 4250 memset(fk, 0, sizeof(*fk)); 4251 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4252 fk, data, l2_proto, nhoff, hlen, 0); 4253 default: 4254 break; 4255 } 4256 4257 fk->ports.ports = 0; 4258 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4259 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4260 return false; 4261 4262 /* ICMP error packets contains at least 8 bytes of the header 4263 * of the packet which generated the error. Use this information 4264 * to correlate ICMP error packets within the same flow which 4265 * generated the error. 4266 */ 4267 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4268 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4269 if (ip_proto == IPPROTO_ICMP) { 4270 if (!icmp_is_err(fk->icmp.type)) 4271 return true; 4272 4273 nhoff += sizeof(struct icmphdr); 4274 } else if (ip_proto == IPPROTO_ICMPV6) { 4275 if (!icmpv6_is_err(fk->icmp.type)) 4276 return true; 4277 4278 nhoff += sizeof(struct icmp6hdr); 4279 } 4280 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4281 } 4282 4283 return true; 4284 } 4285 4286 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4287 { 4288 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4289 (__force u32)flow_get_u32_src(flow); 4290 hash ^= (hash >> 16); 4291 hash ^= (hash >> 8); 4292 4293 /* discard lowest hash bit to deal with the common even ports pattern */ 4294 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4295 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4296 return hash >> 1; 4297 4298 return hash; 4299 } 4300 4301 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4302 * the data as required, but this function can be used without it if the data is 4303 * known to be linear (e.g. with xdp_buff). 4304 */ 4305 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4306 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4307 { 4308 struct flow_keys flow; 4309 u32 hash; 4310 4311 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4312 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4313 4314 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4315 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4316 return bond_eth_hash(skb, data, mhoff, hlen); 4317 4318 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4319 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4320 hash = bond_eth_hash(skb, data, mhoff, hlen); 4321 } else { 4322 if (flow.icmp.id) 4323 memcpy(&hash, &flow.icmp, sizeof(hash)); 4324 else 4325 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4326 } 4327 4328 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4329 } 4330 4331 /** 4332 * bond_xmit_hash - generate a hash value based on the xmit policy 4333 * @bond: bonding device 4334 * @skb: buffer to use for headers 4335 * 4336 * This function will extract the necessary headers from the skb buffer and use 4337 * them to generate a hash based on the xmit_policy set in the bonding device 4338 */ 4339 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4340 { 4341 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4342 skb->l4_hash) 4343 return skb->hash; 4344 4345 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4346 0, skb_network_offset(skb), 4347 skb_headlen(skb)); 4348 } 4349 4350 /** 4351 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4352 * @bond: bonding device 4353 * @xdp: buffer to use for headers 4354 * 4355 * The XDP variant of bond_xmit_hash. 4356 */ 4357 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4358 { 4359 struct ethhdr *eth; 4360 4361 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4362 return 0; 4363 4364 eth = (struct ethhdr *)xdp->data; 4365 4366 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4367 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4368 } 4369 4370 /*-------------------------- Device entry points ----------------------------*/ 4371 4372 void bond_work_init_all(struct bonding *bond) 4373 { 4374 INIT_DELAYED_WORK(&bond->mcast_work, 4375 bond_resend_igmp_join_requests_delayed); 4376 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4377 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4378 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4379 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4380 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4381 } 4382 4383 static void bond_work_cancel_all(struct bonding *bond) 4384 { 4385 cancel_delayed_work_sync(&bond->mii_work); 4386 cancel_delayed_work_sync(&bond->arp_work); 4387 cancel_delayed_work_sync(&bond->alb_work); 4388 cancel_delayed_work_sync(&bond->ad_work); 4389 cancel_delayed_work_sync(&bond->mcast_work); 4390 cancel_delayed_work_sync(&bond->slave_arr_work); 4391 } 4392 4393 static int bond_open(struct net_device *bond_dev) 4394 { 4395 struct bonding *bond = netdev_priv(bond_dev); 4396 struct list_head *iter; 4397 struct slave *slave; 4398 4399 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4400 bond->rr_tx_counter = alloc_percpu(u32); 4401 if (!bond->rr_tx_counter) 4402 return -ENOMEM; 4403 } 4404 4405 /* reset slave->backup and slave->inactive */ 4406 if (bond_has_slaves(bond)) { 4407 bond_for_each_slave(bond, slave, iter) { 4408 if (bond_uses_primary(bond) && 4409 slave != rcu_access_pointer(bond->curr_active_slave)) { 4410 bond_set_slave_inactive_flags(slave, 4411 BOND_SLAVE_NOTIFY_NOW); 4412 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4413 bond_set_slave_active_flags(slave, 4414 BOND_SLAVE_NOTIFY_NOW); 4415 } 4416 } 4417 } 4418 4419 if (bond_is_lb(bond)) { 4420 /* bond_alb_initialize must be called before the timer 4421 * is started. 4422 */ 4423 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4424 return -ENOMEM; 4425 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4426 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4427 } 4428 4429 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4430 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4431 4432 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4433 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4434 bond->recv_probe = bond_rcv_validate; 4435 } 4436 4437 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4438 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4439 /* register to receive LACPDUs */ 4440 bond->recv_probe = bond_3ad_lacpdu_recv; 4441 bond_3ad_initiate_agg_selection(bond, 1); 4442 4443 bond_for_each_slave(bond, slave, iter) 4444 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4445 } 4446 4447 if (bond_mode_can_use_xmit_hash(bond)) 4448 bond_update_slave_arr(bond, NULL); 4449 4450 return 0; 4451 } 4452 4453 static int bond_close(struct net_device *bond_dev) 4454 { 4455 struct bonding *bond = netdev_priv(bond_dev); 4456 struct slave *slave; 4457 4458 bond_work_cancel_all(bond); 4459 bond->send_peer_notif = 0; 4460 if (bond_is_lb(bond)) 4461 bond_alb_deinitialize(bond); 4462 bond->recv_probe = NULL; 4463 4464 if (bond_uses_primary(bond)) { 4465 rcu_read_lock(); 4466 slave = rcu_dereference(bond->curr_active_slave); 4467 if (slave) 4468 bond_hw_addr_flush(bond_dev, slave->dev); 4469 rcu_read_unlock(); 4470 } else { 4471 struct list_head *iter; 4472 4473 bond_for_each_slave(bond, slave, iter) 4474 bond_hw_addr_flush(bond_dev, slave->dev); 4475 } 4476 4477 return 0; 4478 } 4479 4480 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4481 * that some drivers can provide 32bit values only. 4482 */ 4483 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4484 const struct rtnl_link_stats64 *_new, 4485 const struct rtnl_link_stats64 *_old) 4486 { 4487 const u64 *new = (const u64 *)_new; 4488 const u64 *old = (const u64 *)_old; 4489 u64 *res = (u64 *)_res; 4490 int i; 4491 4492 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4493 u64 nv = new[i]; 4494 u64 ov = old[i]; 4495 s64 delta = nv - ov; 4496 4497 /* detects if this particular field is 32bit only */ 4498 if (((nv | ov) >> 32) == 0) 4499 delta = (s64)(s32)((u32)nv - (u32)ov); 4500 4501 /* filter anomalies, some drivers reset their stats 4502 * at down/up events. 4503 */ 4504 if (delta > 0) 4505 res[i] += delta; 4506 } 4507 } 4508 4509 #ifdef CONFIG_LOCKDEP 4510 static int bond_get_lowest_level_rcu(struct net_device *dev) 4511 { 4512 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4513 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4514 int cur = 0, max = 0; 4515 4516 now = dev; 4517 iter = &dev->adj_list.lower; 4518 4519 while (1) { 4520 next = NULL; 4521 while (1) { 4522 ldev = netdev_next_lower_dev_rcu(now, &iter); 4523 if (!ldev) 4524 break; 4525 4526 next = ldev; 4527 niter = &ldev->adj_list.lower; 4528 dev_stack[cur] = now; 4529 iter_stack[cur++] = iter; 4530 if (max <= cur) 4531 max = cur; 4532 break; 4533 } 4534 4535 if (!next) { 4536 if (!cur) 4537 return max; 4538 next = dev_stack[--cur]; 4539 niter = iter_stack[cur]; 4540 } 4541 4542 now = next; 4543 iter = niter; 4544 } 4545 4546 return max; 4547 } 4548 #endif 4549 4550 static void bond_get_stats(struct net_device *bond_dev, 4551 struct rtnl_link_stats64 *stats) 4552 { 4553 struct bonding *bond = netdev_priv(bond_dev); 4554 struct rtnl_link_stats64 temp; 4555 struct list_head *iter; 4556 struct slave *slave; 4557 int nest_level = 0; 4558 4559 4560 rcu_read_lock(); 4561 #ifdef CONFIG_LOCKDEP 4562 nest_level = bond_get_lowest_level_rcu(bond_dev); 4563 #endif 4564 4565 spin_lock_nested(&bond->stats_lock, nest_level); 4566 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4567 4568 bond_for_each_slave_rcu(bond, slave, iter) { 4569 const struct rtnl_link_stats64 *new = 4570 dev_get_stats(slave->dev, &temp); 4571 4572 bond_fold_stats(stats, new, &slave->slave_stats); 4573 4574 /* save off the slave stats for the next run */ 4575 memcpy(&slave->slave_stats, new, sizeof(*new)); 4576 } 4577 4578 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4579 spin_unlock(&bond->stats_lock); 4580 rcu_read_unlock(); 4581 } 4582 4583 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4584 { 4585 struct bonding *bond = netdev_priv(bond_dev); 4586 struct mii_ioctl_data *mii = NULL; 4587 4588 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4589 4590 switch (cmd) { 4591 case SIOCGMIIPHY: 4592 mii = if_mii(ifr); 4593 if (!mii) 4594 return -EINVAL; 4595 4596 mii->phy_id = 0; 4597 fallthrough; 4598 case SIOCGMIIREG: 4599 /* We do this again just in case we were called by SIOCGMIIREG 4600 * instead of SIOCGMIIPHY. 4601 */ 4602 mii = if_mii(ifr); 4603 if (!mii) 4604 return -EINVAL; 4605 4606 if (mii->reg_num == 1) { 4607 mii->val_out = 0; 4608 if (netif_carrier_ok(bond->dev)) 4609 mii->val_out = BMSR_LSTATUS; 4610 } 4611 4612 break; 4613 default: 4614 return -EOPNOTSUPP; 4615 } 4616 4617 return 0; 4618 } 4619 4620 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4621 { 4622 struct bonding *bond = netdev_priv(bond_dev); 4623 struct net_device *slave_dev = NULL; 4624 struct ifbond k_binfo; 4625 struct ifbond __user *u_binfo = NULL; 4626 struct ifslave k_sinfo; 4627 struct ifslave __user *u_sinfo = NULL; 4628 struct bond_opt_value newval; 4629 struct net *net; 4630 int res = 0; 4631 4632 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4633 4634 switch (cmd) { 4635 case SIOCBONDINFOQUERY: 4636 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4637 4638 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4639 return -EFAULT; 4640 4641 bond_info_query(bond_dev, &k_binfo); 4642 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4643 return -EFAULT; 4644 4645 return 0; 4646 case SIOCBONDSLAVEINFOQUERY: 4647 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4648 4649 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4650 return -EFAULT; 4651 4652 res = bond_slave_info_query(bond_dev, &k_sinfo); 4653 if (res == 0 && 4654 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4655 return -EFAULT; 4656 4657 return res; 4658 default: 4659 break; 4660 } 4661 4662 net = dev_net(bond_dev); 4663 4664 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4665 return -EPERM; 4666 4667 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4668 4669 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4670 4671 if (!slave_dev) 4672 return -ENODEV; 4673 4674 switch (cmd) { 4675 case SIOCBONDENSLAVE: 4676 res = bond_enslave(bond_dev, slave_dev, NULL); 4677 break; 4678 case SIOCBONDRELEASE: 4679 res = bond_release(bond_dev, slave_dev); 4680 break; 4681 case SIOCBONDSETHWADDR: 4682 res = bond_set_dev_addr(bond_dev, slave_dev); 4683 break; 4684 case SIOCBONDCHANGEACTIVE: 4685 bond_opt_initstr(&newval, slave_dev->name); 4686 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4687 &newval); 4688 break; 4689 default: 4690 res = -EOPNOTSUPP; 4691 } 4692 4693 return res; 4694 } 4695 4696 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4697 void __user *data, int cmd) 4698 { 4699 struct ifreq ifrdata = { .ifr_data = data }; 4700 4701 switch (cmd) { 4702 case BOND_INFO_QUERY_OLD: 4703 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4704 case BOND_SLAVE_INFO_QUERY_OLD: 4705 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4706 case BOND_ENSLAVE_OLD: 4707 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4708 case BOND_RELEASE_OLD: 4709 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4710 case BOND_SETHWADDR_OLD: 4711 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4712 case BOND_CHANGE_ACTIVE_OLD: 4713 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4714 } 4715 4716 return -EOPNOTSUPP; 4717 } 4718 4719 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4720 { 4721 struct bonding *bond = netdev_priv(bond_dev); 4722 4723 if (change & IFF_PROMISC) 4724 bond_set_promiscuity(bond, 4725 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4726 4727 if (change & IFF_ALLMULTI) 4728 bond_set_allmulti(bond, 4729 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4730 } 4731 4732 static void bond_set_rx_mode(struct net_device *bond_dev) 4733 { 4734 struct bonding *bond = netdev_priv(bond_dev); 4735 struct list_head *iter; 4736 struct slave *slave; 4737 4738 rcu_read_lock(); 4739 if (bond_uses_primary(bond)) { 4740 slave = rcu_dereference(bond->curr_active_slave); 4741 if (slave) { 4742 dev_uc_sync(slave->dev, bond_dev); 4743 dev_mc_sync(slave->dev, bond_dev); 4744 } 4745 } else { 4746 bond_for_each_slave_rcu(bond, slave, iter) { 4747 dev_uc_sync_multiple(slave->dev, bond_dev); 4748 dev_mc_sync_multiple(slave->dev, bond_dev); 4749 } 4750 } 4751 rcu_read_unlock(); 4752 } 4753 4754 static int bond_neigh_init(struct neighbour *n) 4755 { 4756 struct bonding *bond = netdev_priv(n->dev); 4757 const struct net_device_ops *slave_ops; 4758 struct neigh_parms parms; 4759 struct slave *slave; 4760 int ret = 0; 4761 4762 rcu_read_lock(); 4763 slave = bond_first_slave_rcu(bond); 4764 if (!slave) 4765 goto out; 4766 slave_ops = slave->dev->netdev_ops; 4767 if (!slave_ops->ndo_neigh_setup) 4768 goto out; 4769 4770 /* TODO: find another way [1] to implement this. 4771 * Passing a zeroed structure is fragile, 4772 * but at least we do not pass garbage. 4773 * 4774 * [1] One way would be that ndo_neigh_setup() never touch 4775 * struct neigh_parms, but propagate the new neigh_setup() 4776 * back to ___neigh_create() / neigh_parms_alloc() 4777 */ 4778 memset(&parms, 0, sizeof(parms)); 4779 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4780 4781 if (ret) 4782 goto out; 4783 4784 if (parms.neigh_setup) 4785 ret = parms.neigh_setup(n); 4786 out: 4787 rcu_read_unlock(); 4788 return ret; 4789 } 4790 4791 /* The bonding ndo_neigh_setup is called at init time beofre any 4792 * slave exists. So we must declare proxy setup function which will 4793 * be used at run time to resolve the actual slave neigh param setup. 4794 * 4795 * It's also called by master devices (such as vlans) to setup their 4796 * underlying devices. In that case - do nothing, we're already set up from 4797 * our init. 4798 */ 4799 static int bond_neigh_setup(struct net_device *dev, 4800 struct neigh_parms *parms) 4801 { 4802 /* modify only our neigh_parms */ 4803 if (parms->dev == dev) 4804 parms->neigh_setup = bond_neigh_init; 4805 4806 return 0; 4807 } 4808 4809 /* Change the MTU of all of a master's slaves to match the master */ 4810 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4811 { 4812 struct bonding *bond = netdev_priv(bond_dev); 4813 struct slave *slave, *rollback_slave; 4814 struct list_head *iter; 4815 int res = 0; 4816 4817 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4818 4819 bond_for_each_slave(bond, slave, iter) { 4820 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4821 slave, slave->dev->netdev_ops->ndo_change_mtu); 4822 4823 res = dev_set_mtu(slave->dev, new_mtu); 4824 4825 if (res) { 4826 /* If we failed to set the slave's mtu to the new value 4827 * we must abort the operation even in ACTIVE_BACKUP 4828 * mode, because if we allow the backup slaves to have 4829 * different mtu values than the active slave we'll 4830 * need to change their mtu when doing a failover. That 4831 * means changing their mtu from timer context, which 4832 * is probably not a good idea. 4833 */ 4834 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4835 res, new_mtu); 4836 goto unwind; 4837 } 4838 } 4839 4840 WRITE_ONCE(bond_dev->mtu, new_mtu); 4841 4842 return 0; 4843 4844 unwind: 4845 /* unwind from head to the slave that failed */ 4846 bond_for_each_slave(bond, rollback_slave, iter) { 4847 int tmp_res; 4848 4849 if (rollback_slave == slave) 4850 break; 4851 4852 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4853 if (tmp_res) 4854 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4855 tmp_res); 4856 } 4857 4858 return res; 4859 } 4860 4861 /* Change HW address 4862 * 4863 * Note that many devices must be down to change the HW address, and 4864 * downing the master releases all slaves. We can make bonds full of 4865 * bonding devices to test this, however. 4866 */ 4867 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4868 { 4869 struct bonding *bond = netdev_priv(bond_dev); 4870 struct slave *slave, *rollback_slave; 4871 struct sockaddr_storage *ss = addr, tmp_ss; 4872 struct list_head *iter; 4873 int res = 0; 4874 4875 if (BOND_MODE(bond) == BOND_MODE_ALB) 4876 return bond_alb_set_mac_address(bond_dev, addr); 4877 4878 4879 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4880 4881 /* If fail_over_mac is enabled, do nothing and return success. 4882 * Returning an error causes ifenslave to fail. 4883 */ 4884 if (bond->params.fail_over_mac && 4885 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4886 return 0; 4887 4888 if (!is_valid_ether_addr(ss->__data)) 4889 return -EADDRNOTAVAIL; 4890 4891 bond_for_each_slave(bond, slave, iter) { 4892 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4893 __func__, slave); 4894 res = dev_set_mac_address(slave->dev, addr, NULL); 4895 if (res) { 4896 /* TODO: consider downing the slave 4897 * and retry ? 4898 * User should expect communications 4899 * breakage anyway until ARP finish 4900 * updating, so... 4901 */ 4902 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4903 __func__, res); 4904 goto unwind; 4905 } 4906 } 4907 4908 /* success */ 4909 dev_addr_set(bond_dev, ss->__data); 4910 return 0; 4911 4912 unwind: 4913 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4914 tmp_ss.ss_family = bond_dev->type; 4915 4916 /* unwind from head to the slave that failed */ 4917 bond_for_each_slave(bond, rollback_slave, iter) { 4918 int tmp_res; 4919 4920 if (rollback_slave == slave) 4921 break; 4922 4923 tmp_res = dev_set_mac_address(rollback_slave->dev, 4924 (struct sockaddr *)&tmp_ss, NULL); 4925 if (tmp_res) { 4926 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4927 __func__, tmp_res); 4928 } 4929 } 4930 4931 return res; 4932 } 4933 4934 /** 4935 * bond_get_slave_by_id - get xmit slave with slave_id 4936 * @bond: bonding device that is transmitting 4937 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4938 * 4939 * This function tries to get slave with slave_id but in case 4940 * it fails, it tries to find the first available slave for transmission. 4941 */ 4942 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4943 int slave_id) 4944 { 4945 struct list_head *iter; 4946 struct slave *slave; 4947 int i = slave_id; 4948 4949 /* Here we start from the slave with slave_id */ 4950 bond_for_each_slave_rcu(bond, slave, iter) { 4951 if (--i < 0) { 4952 if (bond_slave_can_tx(slave)) 4953 return slave; 4954 } 4955 } 4956 4957 /* Here we start from the first slave up to slave_id */ 4958 i = slave_id; 4959 bond_for_each_slave_rcu(bond, slave, iter) { 4960 if (--i < 0) 4961 break; 4962 if (bond_slave_can_tx(slave)) 4963 return slave; 4964 } 4965 /* no slave that can tx has been found */ 4966 return NULL; 4967 } 4968 4969 /** 4970 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4971 * @bond: bonding device to use 4972 * 4973 * Based on the value of the bonding device's packets_per_slave parameter 4974 * this function generates a slave id, which is usually used as the next 4975 * slave to transmit through. 4976 */ 4977 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4978 { 4979 u32 slave_id; 4980 struct reciprocal_value reciprocal_packets_per_slave; 4981 int packets_per_slave = bond->params.packets_per_slave; 4982 4983 switch (packets_per_slave) { 4984 case 0: 4985 slave_id = get_random_u32(); 4986 break; 4987 case 1: 4988 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4989 break; 4990 default: 4991 reciprocal_packets_per_slave = 4992 bond->params.reciprocal_packets_per_slave; 4993 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4994 slave_id = reciprocal_divide(slave_id, 4995 reciprocal_packets_per_slave); 4996 break; 4997 } 4998 4999 return slave_id; 5000 } 5001 5002 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 5003 struct sk_buff *skb) 5004 { 5005 struct slave *slave; 5006 int slave_cnt; 5007 u32 slave_id; 5008 5009 /* Start with the curr_active_slave that joined the bond as the 5010 * default for sending IGMP traffic. For failover purposes one 5011 * needs to maintain some consistency for the interface that will 5012 * send the join/membership reports. The curr_active_slave found 5013 * will send all of this type of traffic. 5014 */ 5015 if (skb->protocol == htons(ETH_P_IP)) { 5016 int noff = skb_network_offset(skb); 5017 struct iphdr *iph; 5018 5019 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 5020 goto non_igmp; 5021 5022 iph = ip_hdr(skb); 5023 if (iph->protocol == IPPROTO_IGMP) { 5024 slave = rcu_dereference(bond->curr_active_slave); 5025 if (slave) 5026 return slave; 5027 return bond_get_slave_by_id(bond, 0); 5028 } 5029 } 5030 5031 non_igmp: 5032 slave_cnt = READ_ONCE(bond->slave_cnt); 5033 if (likely(slave_cnt)) { 5034 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 5035 return bond_get_slave_by_id(bond, slave_id); 5036 } 5037 return NULL; 5038 } 5039 5040 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 5041 struct xdp_buff *xdp) 5042 { 5043 struct slave *slave; 5044 int slave_cnt; 5045 u32 slave_id; 5046 const struct ethhdr *eth; 5047 void *data = xdp->data; 5048 5049 if (data + sizeof(struct ethhdr) > xdp->data_end) 5050 goto non_igmp; 5051 5052 eth = (struct ethhdr *)data; 5053 data += sizeof(struct ethhdr); 5054 5055 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 5056 if (eth->h_proto == htons(ETH_P_IP)) { 5057 const struct iphdr *iph; 5058 5059 if (data + sizeof(struct iphdr) > xdp->data_end) 5060 goto non_igmp; 5061 5062 iph = (struct iphdr *)data; 5063 5064 if (iph->protocol == IPPROTO_IGMP) { 5065 slave = rcu_dereference(bond->curr_active_slave); 5066 if (slave) 5067 return slave; 5068 return bond_get_slave_by_id(bond, 0); 5069 } 5070 } 5071 5072 non_igmp: 5073 slave_cnt = READ_ONCE(bond->slave_cnt); 5074 if (likely(slave_cnt)) { 5075 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 5076 return bond_get_slave_by_id(bond, slave_id); 5077 } 5078 return NULL; 5079 } 5080 5081 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 5082 struct net_device *bond_dev) 5083 { 5084 struct bonding *bond = netdev_priv(bond_dev); 5085 struct slave *slave; 5086 5087 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5088 if (likely(slave)) 5089 return bond_dev_queue_xmit(bond, skb, slave->dev); 5090 5091 return bond_tx_drop(bond_dev, skb); 5092 } 5093 5094 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 5095 { 5096 return rcu_dereference(bond->curr_active_slave); 5097 } 5098 5099 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 5100 * the bond has a usable interface. 5101 */ 5102 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 5103 struct net_device *bond_dev) 5104 { 5105 struct bonding *bond = netdev_priv(bond_dev); 5106 struct slave *slave; 5107 5108 slave = bond_xmit_activebackup_slave_get(bond); 5109 if (slave) 5110 return bond_dev_queue_xmit(bond, skb, slave->dev); 5111 5112 return bond_tx_drop(bond_dev, skb); 5113 } 5114 5115 /* Use this to update slave_array when (a) it's not appropriate to update 5116 * slave_array right away (note that update_slave_array() may sleep) 5117 * and / or (b) RTNL is not held. 5118 */ 5119 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 5120 { 5121 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 5122 } 5123 5124 /* Slave array work handler. Holds only RTNL */ 5125 static void bond_slave_arr_handler(struct work_struct *work) 5126 { 5127 struct bonding *bond = container_of(work, struct bonding, 5128 slave_arr_work.work); 5129 int ret; 5130 5131 if (!rtnl_trylock()) 5132 goto err; 5133 5134 ret = bond_update_slave_arr(bond, NULL); 5135 rtnl_unlock(); 5136 if (ret) { 5137 pr_warn_ratelimited("Failed to update slave array from WT\n"); 5138 goto err; 5139 } 5140 return; 5141 5142 err: 5143 bond_slave_arr_work_rearm(bond, 1); 5144 } 5145 5146 static void bond_skip_slave(struct bond_up_slave *slaves, 5147 struct slave *skipslave) 5148 { 5149 int idx; 5150 5151 /* Rare situation where caller has asked to skip a specific 5152 * slave but allocation failed (most likely!). BTW this is 5153 * only possible when the call is initiated from 5154 * __bond_release_one(). In this situation; overwrite the 5155 * skipslave entry in the array with the last entry from the 5156 * array to avoid a situation where the xmit path may choose 5157 * this to-be-skipped slave to send a packet out. 5158 */ 5159 for (idx = 0; slaves && idx < slaves->count; idx++) { 5160 if (skipslave == slaves->arr[idx]) { 5161 slaves->arr[idx] = 5162 slaves->arr[slaves->count - 1]; 5163 slaves->count--; 5164 break; 5165 } 5166 } 5167 } 5168 5169 static void bond_set_slave_arr(struct bonding *bond, 5170 struct bond_up_slave *usable_slaves, 5171 struct bond_up_slave *all_slaves) 5172 { 5173 struct bond_up_slave *usable, *all; 5174 5175 usable = rtnl_dereference(bond->usable_slaves); 5176 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5177 kfree_rcu(usable, rcu); 5178 5179 all = rtnl_dereference(bond->all_slaves); 5180 rcu_assign_pointer(bond->all_slaves, all_slaves); 5181 kfree_rcu(all, rcu); 5182 } 5183 5184 static void bond_reset_slave_arr(struct bonding *bond) 5185 { 5186 bond_set_slave_arr(bond, NULL, NULL); 5187 } 5188 5189 /* Build the usable slaves array in control path for modes that use xmit-hash 5190 * to determine the slave interface - 5191 * (a) BOND_MODE_8023AD 5192 * (b) BOND_MODE_XOR 5193 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5194 * 5195 * The caller is expected to hold RTNL only and NO other lock! 5196 */ 5197 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5198 { 5199 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5200 struct slave *slave; 5201 struct list_head *iter; 5202 int agg_id = 0; 5203 int ret = 0; 5204 5205 might_sleep(); 5206 5207 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5208 bond->slave_cnt), GFP_KERNEL); 5209 all_slaves = kzalloc(struct_size(all_slaves, arr, 5210 bond->slave_cnt), GFP_KERNEL); 5211 if (!usable_slaves || !all_slaves) { 5212 ret = -ENOMEM; 5213 goto out; 5214 } 5215 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5216 struct ad_info ad_info; 5217 5218 spin_lock_bh(&bond->mode_lock); 5219 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5220 spin_unlock_bh(&bond->mode_lock); 5221 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5222 /* No active aggragator means it's not safe to use 5223 * the previous array. 5224 */ 5225 bond_reset_slave_arr(bond); 5226 goto out; 5227 } 5228 spin_unlock_bh(&bond->mode_lock); 5229 agg_id = ad_info.aggregator_id; 5230 } 5231 bond_for_each_slave(bond, slave, iter) { 5232 if (skipslave == slave) 5233 continue; 5234 5235 all_slaves->arr[all_slaves->count++] = slave; 5236 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5237 struct aggregator *agg; 5238 5239 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5240 if (!agg || agg->aggregator_identifier != agg_id) 5241 continue; 5242 } 5243 if (!bond_slave_can_tx(slave)) 5244 continue; 5245 5246 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5247 usable_slaves->count); 5248 5249 usable_slaves->arr[usable_slaves->count++] = slave; 5250 } 5251 5252 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5253 return ret; 5254 out: 5255 if (ret != 0 && skipslave) { 5256 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5257 skipslave); 5258 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5259 skipslave); 5260 } 5261 kfree_rcu(all_slaves, rcu); 5262 kfree_rcu(usable_slaves, rcu); 5263 5264 return ret; 5265 } 5266 5267 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5268 struct sk_buff *skb, 5269 struct bond_up_slave *slaves) 5270 { 5271 struct slave *slave; 5272 unsigned int count; 5273 u32 hash; 5274 5275 hash = bond_xmit_hash(bond, skb); 5276 count = slaves ? READ_ONCE(slaves->count) : 0; 5277 if (unlikely(!count)) 5278 return NULL; 5279 5280 slave = slaves->arr[hash % count]; 5281 return slave; 5282 } 5283 5284 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5285 struct xdp_buff *xdp) 5286 { 5287 struct bond_up_slave *slaves; 5288 unsigned int count; 5289 u32 hash; 5290 5291 hash = bond_xmit_hash_xdp(bond, xdp); 5292 slaves = rcu_dereference(bond->usable_slaves); 5293 count = slaves ? READ_ONCE(slaves->count) : 0; 5294 if (unlikely(!count)) 5295 return NULL; 5296 5297 return slaves->arr[hash % count]; 5298 } 5299 5300 /* Use this Xmit function for 3AD as well as XOR modes. The current 5301 * usable slave array is formed in the control path. The xmit function 5302 * just calculates hash and sends the packet out. 5303 */ 5304 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5305 struct net_device *dev) 5306 { 5307 struct bonding *bond = netdev_priv(dev); 5308 struct bond_up_slave *slaves; 5309 struct slave *slave; 5310 5311 slaves = rcu_dereference(bond->usable_slaves); 5312 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5313 if (likely(slave)) 5314 return bond_dev_queue_xmit(bond, skb, slave->dev); 5315 5316 return bond_tx_drop(dev, skb); 5317 } 5318 5319 /* in broadcast mode, we send everything to all usable interfaces. */ 5320 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5321 struct net_device *bond_dev) 5322 { 5323 struct bonding *bond = netdev_priv(bond_dev); 5324 struct slave *slave = NULL; 5325 struct list_head *iter; 5326 bool xmit_suc = false; 5327 bool skb_used = false; 5328 5329 bond_for_each_slave_rcu(bond, slave, iter) { 5330 struct sk_buff *skb2; 5331 5332 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5333 continue; 5334 5335 if (bond_is_last_slave(bond, slave)) { 5336 skb2 = skb; 5337 skb_used = true; 5338 } else { 5339 skb2 = skb_clone(skb, GFP_ATOMIC); 5340 if (!skb2) { 5341 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5342 bond_dev->name, __func__); 5343 continue; 5344 } 5345 } 5346 5347 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5348 xmit_suc = true; 5349 } 5350 5351 if (!skb_used) 5352 dev_kfree_skb_any(skb); 5353 5354 if (xmit_suc) 5355 return NETDEV_TX_OK; 5356 5357 dev_core_stats_tx_dropped_inc(bond_dev); 5358 return NET_XMIT_DROP; 5359 } 5360 5361 /*------------------------- Device initialization ---------------------------*/ 5362 5363 /* Lookup the slave that corresponds to a qid */ 5364 static inline int bond_slave_override(struct bonding *bond, 5365 struct sk_buff *skb) 5366 { 5367 struct slave *slave = NULL; 5368 struct list_head *iter; 5369 5370 if (!skb_rx_queue_recorded(skb)) 5371 return 1; 5372 5373 /* Find out if any slaves have the same mapping as this skb. */ 5374 bond_for_each_slave_rcu(bond, slave, iter) { 5375 if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) { 5376 if (bond_slave_is_up(slave) && 5377 slave->link == BOND_LINK_UP) { 5378 bond_dev_queue_xmit(bond, skb, slave->dev); 5379 return 0; 5380 } 5381 /* If the slave isn't UP, use default transmit policy. */ 5382 break; 5383 } 5384 } 5385 5386 return 1; 5387 } 5388 5389 5390 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5391 struct net_device *sb_dev) 5392 { 5393 /* This helper function exists to help dev_pick_tx get the correct 5394 * destination queue. Using a helper function skips a call to 5395 * skb_tx_hash and will put the skbs in the queue we expect on their 5396 * way down to the bonding driver. 5397 */ 5398 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5399 5400 /* Save the original txq to restore before passing to the driver */ 5401 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5402 5403 if (unlikely(txq >= dev->real_num_tx_queues)) { 5404 do { 5405 txq -= dev->real_num_tx_queues; 5406 } while (txq >= dev->real_num_tx_queues); 5407 } 5408 return txq; 5409 } 5410 5411 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5412 struct sk_buff *skb, 5413 bool all_slaves) 5414 { 5415 struct bonding *bond = netdev_priv(master_dev); 5416 struct bond_up_slave *slaves; 5417 struct slave *slave = NULL; 5418 5419 switch (BOND_MODE(bond)) { 5420 case BOND_MODE_ROUNDROBIN: 5421 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5422 break; 5423 case BOND_MODE_ACTIVEBACKUP: 5424 slave = bond_xmit_activebackup_slave_get(bond); 5425 break; 5426 case BOND_MODE_8023AD: 5427 case BOND_MODE_XOR: 5428 if (all_slaves) 5429 slaves = rcu_dereference(bond->all_slaves); 5430 else 5431 slaves = rcu_dereference(bond->usable_slaves); 5432 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5433 break; 5434 case BOND_MODE_BROADCAST: 5435 break; 5436 case BOND_MODE_ALB: 5437 slave = bond_xmit_alb_slave_get(bond, skb); 5438 break; 5439 case BOND_MODE_TLB: 5440 slave = bond_xmit_tlb_slave_get(bond, skb); 5441 break; 5442 default: 5443 /* Should never happen, mode already checked */ 5444 WARN_ONCE(true, "Unknown bonding mode"); 5445 break; 5446 } 5447 5448 if (slave) 5449 return slave->dev; 5450 return NULL; 5451 } 5452 5453 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5454 { 5455 switch (sk->sk_family) { 5456 #if IS_ENABLED(CONFIG_IPV6) 5457 case AF_INET6: 5458 if (ipv6_only_sock(sk) || 5459 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5460 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5461 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5462 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5463 break; 5464 } 5465 fallthrough; 5466 #endif 5467 default: /* AF_INET */ 5468 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5469 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5470 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5471 break; 5472 } 5473 5474 flow->ports.src = inet_sk(sk)->inet_sport; 5475 flow->ports.dst = inet_sk(sk)->inet_dport; 5476 } 5477 5478 /** 5479 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5480 * @sk: socket to use for headers 5481 * 5482 * This function will extract the necessary field from the socket and use 5483 * them to generate a hash based on the LAYER34 xmit_policy. 5484 * Assumes that sk is a TCP or UDP socket. 5485 */ 5486 static u32 bond_sk_hash_l34(struct sock *sk) 5487 { 5488 struct flow_keys flow; 5489 u32 hash; 5490 5491 bond_sk_to_flow(sk, &flow); 5492 5493 /* L4 */ 5494 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5495 /* L3 */ 5496 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5497 } 5498 5499 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5500 struct sock *sk) 5501 { 5502 struct bond_up_slave *slaves; 5503 struct slave *slave; 5504 unsigned int count; 5505 u32 hash; 5506 5507 slaves = rcu_dereference(bond->usable_slaves); 5508 count = slaves ? READ_ONCE(slaves->count) : 0; 5509 if (unlikely(!count)) 5510 return NULL; 5511 5512 hash = bond_sk_hash_l34(sk); 5513 slave = slaves->arr[hash % count]; 5514 5515 return slave->dev; 5516 } 5517 5518 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5519 struct sock *sk) 5520 { 5521 struct bonding *bond = netdev_priv(dev); 5522 struct net_device *lower = NULL; 5523 5524 rcu_read_lock(); 5525 if (bond_sk_check(bond)) 5526 lower = __bond_sk_get_lower_dev(bond, sk); 5527 rcu_read_unlock(); 5528 5529 return lower; 5530 } 5531 5532 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5533 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5534 struct net_device *dev) 5535 { 5536 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5537 5538 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded 5539 * was true, if tls_device_down is running in parallel, but it's OK, 5540 * because bond_get_slave_by_dev has a NULL check. 5541 */ 5542 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5543 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5544 return bond_tx_drop(dev, skb); 5545 } 5546 #endif 5547 5548 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5549 { 5550 struct bonding *bond = netdev_priv(dev); 5551 5552 if (bond_should_override_tx_queue(bond) && 5553 !bond_slave_override(bond, skb)) 5554 return NETDEV_TX_OK; 5555 5556 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5557 if (tls_is_skb_tx_device_offloaded(skb)) 5558 return bond_tls_device_xmit(bond, skb, dev); 5559 #endif 5560 5561 switch (BOND_MODE(bond)) { 5562 case BOND_MODE_ROUNDROBIN: 5563 return bond_xmit_roundrobin(skb, dev); 5564 case BOND_MODE_ACTIVEBACKUP: 5565 return bond_xmit_activebackup(skb, dev); 5566 case BOND_MODE_8023AD: 5567 case BOND_MODE_XOR: 5568 return bond_3ad_xor_xmit(skb, dev); 5569 case BOND_MODE_BROADCAST: 5570 return bond_xmit_broadcast(skb, dev); 5571 case BOND_MODE_ALB: 5572 return bond_alb_xmit(skb, dev); 5573 case BOND_MODE_TLB: 5574 return bond_tlb_xmit(skb, dev); 5575 default: 5576 /* Should never happen, mode already checked */ 5577 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5578 WARN_ON_ONCE(1); 5579 return bond_tx_drop(dev, skb); 5580 } 5581 } 5582 5583 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5584 { 5585 struct bonding *bond = netdev_priv(dev); 5586 netdev_tx_t ret = NETDEV_TX_OK; 5587 5588 /* If we risk deadlock from transmitting this in the 5589 * netpoll path, tell netpoll to queue the frame for later tx 5590 */ 5591 if (unlikely(is_netpoll_tx_blocked(dev))) 5592 return NETDEV_TX_BUSY; 5593 5594 rcu_read_lock(); 5595 if (bond_has_slaves(bond)) 5596 ret = __bond_start_xmit(skb, dev); 5597 else 5598 ret = bond_tx_drop(dev, skb); 5599 rcu_read_unlock(); 5600 5601 return ret; 5602 } 5603 5604 static struct net_device * 5605 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5606 { 5607 struct bonding *bond = netdev_priv(bond_dev); 5608 struct slave *slave; 5609 5610 /* Caller needs to hold rcu_read_lock() */ 5611 5612 switch (BOND_MODE(bond)) { 5613 case BOND_MODE_ROUNDROBIN: 5614 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5615 break; 5616 5617 case BOND_MODE_ACTIVEBACKUP: 5618 slave = bond_xmit_activebackup_slave_get(bond); 5619 break; 5620 5621 case BOND_MODE_8023AD: 5622 case BOND_MODE_XOR: 5623 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5624 break; 5625 5626 default: 5627 if (net_ratelimit()) 5628 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", 5629 BOND_MODE(bond)); 5630 return NULL; 5631 } 5632 5633 if (slave) 5634 return slave->dev; 5635 5636 return NULL; 5637 } 5638 5639 static int bond_xdp_xmit(struct net_device *bond_dev, 5640 int n, struct xdp_frame **frames, u32 flags) 5641 { 5642 int nxmit, err = -ENXIO; 5643 5644 rcu_read_lock(); 5645 5646 for (nxmit = 0; nxmit < n; nxmit++) { 5647 struct xdp_frame *frame = frames[nxmit]; 5648 struct xdp_frame *frames1[] = {frame}; 5649 struct net_device *slave_dev; 5650 struct xdp_buff xdp; 5651 5652 xdp_convert_frame_to_buff(frame, &xdp); 5653 5654 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5655 if (!slave_dev) { 5656 err = -ENXIO; 5657 break; 5658 } 5659 5660 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5661 if (err < 1) 5662 break; 5663 } 5664 5665 rcu_read_unlock(); 5666 5667 /* If error happened on the first frame then we can pass the error up, otherwise 5668 * report the number of frames that were xmitted. 5669 */ 5670 if (err < 0) 5671 return (nxmit == 0 ? err : nxmit); 5672 5673 return nxmit; 5674 } 5675 5676 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5677 struct netlink_ext_ack *extack) 5678 { 5679 struct bonding *bond = netdev_priv(dev); 5680 struct list_head *iter; 5681 struct slave *slave, *rollback_slave; 5682 struct bpf_prog *old_prog; 5683 struct netdev_bpf xdp = { 5684 .command = XDP_SETUP_PROG, 5685 .flags = 0, 5686 .prog = prog, 5687 .extack = extack, 5688 }; 5689 int err; 5690 5691 ASSERT_RTNL(); 5692 5693 if (!bond_xdp_check(bond)) 5694 return -EOPNOTSUPP; 5695 5696 old_prog = bond->xdp_prog; 5697 bond->xdp_prog = prog; 5698 5699 bond_for_each_slave(bond, slave, iter) { 5700 struct net_device *slave_dev = slave->dev; 5701 5702 if (!slave_dev->netdev_ops->ndo_bpf || 5703 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5704 SLAVE_NL_ERR(dev, slave_dev, extack, 5705 "Slave device does not support XDP"); 5706 err = -EOPNOTSUPP; 5707 goto err; 5708 } 5709 5710 if (dev_xdp_prog_count(slave_dev) > 0) { 5711 SLAVE_NL_ERR(dev, slave_dev, extack, 5712 "Slave has XDP program loaded, please unload before enslaving"); 5713 err = -EOPNOTSUPP; 5714 goto err; 5715 } 5716 5717 err = dev_xdp_propagate(slave_dev, &xdp); 5718 if (err < 0) { 5719 /* ndo_bpf() sets extack error message */ 5720 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5721 goto err; 5722 } 5723 if (prog) 5724 bpf_prog_inc(prog); 5725 } 5726 5727 if (prog) { 5728 static_branch_inc(&bpf_master_redirect_enabled_key); 5729 } else if (old_prog) { 5730 bpf_prog_put(old_prog); 5731 static_branch_dec(&bpf_master_redirect_enabled_key); 5732 } 5733 5734 return 0; 5735 5736 err: 5737 /* unwind the program changes */ 5738 bond->xdp_prog = old_prog; 5739 xdp.prog = old_prog; 5740 xdp.extack = NULL; /* do not overwrite original error */ 5741 5742 bond_for_each_slave(bond, rollback_slave, iter) { 5743 struct net_device *slave_dev = rollback_slave->dev; 5744 int err_unwind; 5745 5746 if (slave == rollback_slave) 5747 break; 5748 5749 err_unwind = dev_xdp_propagate(slave_dev, &xdp); 5750 if (err_unwind < 0) 5751 slave_err(dev, slave_dev, 5752 "Error %d when unwinding XDP program change\n", err_unwind); 5753 else if (xdp.prog) 5754 bpf_prog_inc(xdp.prog); 5755 } 5756 return err; 5757 } 5758 5759 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5760 { 5761 switch (xdp->command) { 5762 case XDP_SETUP_PROG: 5763 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5764 default: 5765 return -EINVAL; 5766 } 5767 } 5768 5769 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5770 { 5771 if (speed == 0 || speed == SPEED_UNKNOWN) 5772 speed = slave->speed; 5773 else 5774 speed = min(speed, slave->speed); 5775 5776 return speed; 5777 } 5778 5779 /* Set the BOND_PHC_INDEX flag to notify user space */ 5780 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg) 5781 { 5782 struct ifreq *ifr = kernel_cfg->ifr; 5783 struct hwtstamp_config cfg; 5784 5785 if (kernel_cfg->copied_to_user) { 5786 /* Lower device has a legacy implementation */ 5787 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 5788 return -EFAULT; 5789 5790 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5791 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg))) 5792 return -EFAULT; 5793 } else { 5794 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5795 } 5796 5797 return 0; 5798 } 5799 5800 static int bond_hwtstamp_get(struct net_device *dev, 5801 struct kernel_hwtstamp_config *cfg) 5802 { 5803 struct bonding *bond = netdev_priv(dev); 5804 struct net_device *real_dev; 5805 int err; 5806 5807 real_dev = bond_option_active_slave_get_rcu(bond); 5808 if (!real_dev) 5809 return -EOPNOTSUPP; 5810 5811 err = generic_hwtstamp_get_lower(real_dev, cfg); 5812 if (err) 5813 return err; 5814 5815 return bond_set_phc_index_flag(cfg); 5816 } 5817 5818 static int bond_hwtstamp_set(struct net_device *dev, 5819 struct kernel_hwtstamp_config *cfg, 5820 struct netlink_ext_ack *extack) 5821 { 5822 struct bonding *bond = netdev_priv(dev); 5823 struct net_device *real_dev; 5824 int err; 5825 5826 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 5827 return -EOPNOTSUPP; 5828 5829 real_dev = bond_option_active_slave_get_rcu(bond); 5830 if (!real_dev) 5831 return -EOPNOTSUPP; 5832 5833 err = generic_hwtstamp_set_lower(real_dev, cfg, extack); 5834 if (err) 5835 return err; 5836 5837 return bond_set_phc_index_flag(cfg); 5838 } 5839 5840 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5841 struct ethtool_link_ksettings *cmd) 5842 { 5843 struct bonding *bond = netdev_priv(bond_dev); 5844 struct list_head *iter; 5845 struct slave *slave; 5846 u32 speed = 0; 5847 5848 cmd->base.duplex = DUPLEX_UNKNOWN; 5849 cmd->base.port = PORT_OTHER; 5850 5851 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5852 * do not need to check mode. Though link speed might not represent 5853 * the true receive or transmit bandwidth (not all modes are symmetric) 5854 * this is an accurate maximum. 5855 */ 5856 bond_for_each_slave(bond, slave, iter) { 5857 if (bond_slave_can_tx(slave)) { 5858 bond_update_speed_duplex(slave); 5859 if (slave->speed != SPEED_UNKNOWN) { 5860 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5861 speed = bond_mode_bcast_speed(slave, 5862 speed); 5863 else 5864 speed += slave->speed; 5865 } 5866 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5867 slave->duplex != DUPLEX_UNKNOWN) 5868 cmd->base.duplex = slave->duplex; 5869 } 5870 } 5871 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5872 5873 return 0; 5874 } 5875 5876 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5877 struct ethtool_drvinfo *drvinfo) 5878 { 5879 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5880 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5881 BOND_ABI_VERSION); 5882 } 5883 5884 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5885 struct kernel_ethtool_ts_info *info) 5886 { 5887 struct bonding *bond = netdev_priv(bond_dev); 5888 struct kernel_ethtool_ts_info ts_info; 5889 struct net_device *real_dev; 5890 bool sw_tx_support = false; 5891 struct list_head *iter; 5892 struct slave *slave; 5893 int ret = 0; 5894 5895 rcu_read_lock(); 5896 real_dev = bond_option_active_slave_get_rcu(bond); 5897 dev_hold(real_dev); 5898 rcu_read_unlock(); 5899 5900 if (real_dev) { 5901 ret = ethtool_get_ts_info_by_layer(real_dev, info); 5902 } else { 5903 /* Check if all slaves support software tx timestamping */ 5904 rcu_read_lock(); 5905 bond_for_each_slave_rcu(bond, slave, iter) { 5906 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info); 5907 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) { 5908 sw_tx_support = true; 5909 continue; 5910 } 5911 5912 sw_tx_support = false; 5913 break; 5914 } 5915 rcu_read_unlock(); 5916 } 5917 5918 if (sw_tx_support) 5919 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE; 5920 5921 dev_put(real_dev); 5922 return ret; 5923 } 5924 5925 static const struct ethtool_ops bond_ethtool_ops = { 5926 .get_drvinfo = bond_ethtool_get_drvinfo, 5927 .get_link = ethtool_op_get_link, 5928 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5929 .get_ts_info = bond_ethtool_get_ts_info, 5930 }; 5931 5932 static const struct net_device_ops bond_netdev_ops = { 5933 .ndo_init = bond_init, 5934 .ndo_uninit = bond_uninit, 5935 .ndo_open = bond_open, 5936 .ndo_stop = bond_close, 5937 .ndo_start_xmit = bond_start_xmit, 5938 .ndo_select_queue = bond_select_queue, 5939 .ndo_get_stats64 = bond_get_stats, 5940 .ndo_eth_ioctl = bond_eth_ioctl, 5941 .ndo_siocbond = bond_do_ioctl, 5942 .ndo_siocdevprivate = bond_siocdevprivate, 5943 .ndo_change_rx_flags = bond_change_rx_flags, 5944 .ndo_set_rx_mode = bond_set_rx_mode, 5945 .ndo_change_mtu = bond_change_mtu, 5946 .ndo_set_mac_address = bond_set_mac_address, 5947 .ndo_neigh_setup = bond_neigh_setup, 5948 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5949 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5950 #ifdef CONFIG_NET_POLL_CONTROLLER 5951 .ndo_netpoll_setup = bond_netpoll_setup, 5952 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5953 .ndo_poll_controller = bond_poll_controller, 5954 #endif 5955 .ndo_add_slave = bond_enslave, 5956 .ndo_del_slave = bond_release, 5957 .ndo_fix_features = bond_fix_features, 5958 .ndo_features_check = passthru_features_check, 5959 .ndo_get_xmit_slave = bond_xmit_get_slave, 5960 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5961 .ndo_bpf = bond_xdp, 5962 .ndo_xdp_xmit = bond_xdp_xmit, 5963 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5964 .ndo_hwtstamp_get = bond_hwtstamp_get, 5965 .ndo_hwtstamp_set = bond_hwtstamp_set, 5966 }; 5967 5968 static const struct device_type bond_type = { 5969 .name = "bond", 5970 }; 5971 5972 static void bond_destructor(struct net_device *bond_dev) 5973 { 5974 struct bonding *bond = netdev_priv(bond_dev); 5975 5976 if (bond->wq) 5977 destroy_workqueue(bond->wq); 5978 5979 free_percpu(bond->rr_tx_counter); 5980 } 5981 5982 void bond_setup(struct net_device *bond_dev) 5983 { 5984 struct bonding *bond = netdev_priv(bond_dev); 5985 5986 spin_lock_init(&bond->mode_lock); 5987 bond->params = bonding_defaults; 5988 5989 /* Initialize pointers */ 5990 bond->dev = bond_dev; 5991 5992 /* Initialize the device entry points */ 5993 ether_setup(bond_dev); 5994 bond_dev->max_mtu = ETH_MAX_MTU; 5995 bond_dev->netdev_ops = &bond_netdev_ops; 5996 bond_dev->ethtool_ops = &bond_ethtool_ops; 5997 5998 bond_dev->needs_free_netdev = true; 5999 bond_dev->priv_destructor = bond_destructor; 6000 6001 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 6002 6003 /* Initialize the device options */ 6004 bond_dev->flags |= IFF_MASTER; 6005 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 6006 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 6007 6008 #ifdef CONFIG_XFRM_OFFLOAD 6009 /* set up xfrm device ops (only supported in active-backup right now) */ 6010 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 6011 INIT_LIST_HEAD(&bond->ipsec_list); 6012 mutex_init(&bond->ipsec_lock); 6013 #endif /* CONFIG_XFRM_OFFLOAD */ 6014 6015 /* don't acquire bond device's netif_tx_lock when transmitting */ 6016 bond_dev->lltx = true; 6017 6018 /* Don't allow bond devices to change network namespaces. */ 6019 bond_dev->netns_local = true; 6020 6021 /* By default, we declare the bond to be fully 6022 * VLAN hardware accelerated capable. Special 6023 * care is taken in the various xmit functions 6024 * when there are slaves that are not hw accel 6025 * capable 6026 */ 6027 6028 bond_dev->hw_features = BOND_VLAN_FEATURES | 6029 NETIF_F_HW_VLAN_CTAG_RX | 6030 NETIF_F_HW_VLAN_CTAG_FILTER | 6031 NETIF_F_HW_VLAN_STAG_RX | 6032 NETIF_F_HW_VLAN_STAG_FILTER; 6033 6034 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 6035 bond_dev->features |= bond_dev->hw_features; 6036 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 6037 #ifdef CONFIG_XFRM_OFFLOAD 6038 bond_dev->hw_features |= BOND_XFRM_FEATURES; 6039 /* Only enable XFRM features if this is an active-backup config */ 6040 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 6041 bond_dev->features |= BOND_XFRM_FEATURES; 6042 #endif /* CONFIG_XFRM_OFFLOAD */ 6043 } 6044 6045 /* Destroy a bonding device. 6046 * Must be under rtnl_lock when this function is called. 6047 */ 6048 static void bond_uninit(struct net_device *bond_dev) 6049 { 6050 struct bonding *bond = netdev_priv(bond_dev); 6051 struct list_head *iter; 6052 struct slave *slave; 6053 6054 bond_netpoll_cleanup(bond_dev); 6055 6056 /* Release the bonded slaves */ 6057 bond_for_each_slave(bond, slave, iter) 6058 __bond_release_one(bond_dev, slave->dev, true, true); 6059 netdev_info(bond_dev, "Released all slaves\n"); 6060 6061 #ifdef CONFIG_XFRM_OFFLOAD 6062 mutex_destroy(&bond->ipsec_lock); 6063 #endif /* CONFIG_XFRM_OFFLOAD */ 6064 6065 bond_set_slave_arr(bond, NULL, NULL); 6066 6067 list_del_rcu(&bond->bond_list); 6068 6069 bond_debug_unregister(bond); 6070 } 6071 6072 /*------------------------- Module initialization ---------------------------*/ 6073 6074 static int __init bond_check_params(struct bond_params *params) 6075 { 6076 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 6077 struct bond_opt_value newval; 6078 const struct bond_opt_value *valptr; 6079 int arp_all_targets_value = 0; 6080 u16 ad_actor_sys_prio = 0; 6081 u16 ad_user_port_key = 0; 6082 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 6083 int arp_ip_count; 6084 int bond_mode = BOND_MODE_ROUNDROBIN; 6085 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 6086 int lacp_fast = 0; 6087 int tlb_dynamic_lb; 6088 6089 /* Convert string parameters. */ 6090 if (mode) { 6091 bond_opt_initstr(&newval, mode); 6092 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 6093 if (!valptr) { 6094 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 6095 return -EINVAL; 6096 } 6097 bond_mode = valptr->value; 6098 } 6099 6100 if (xmit_hash_policy) { 6101 if (bond_mode == BOND_MODE_ROUNDROBIN || 6102 bond_mode == BOND_MODE_ACTIVEBACKUP || 6103 bond_mode == BOND_MODE_BROADCAST) { 6104 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 6105 bond_mode_name(bond_mode)); 6106 } else { 6107 bond_opt_initstr(&newval, xmit_hash_policy); 6108 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 6109 &newval); 6110 if (!valptr) { 6111 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 6112 xmit_hash_policy); 6113 return -EINVAL; 6114 } 6115 xmit_hashtype = valptr->value; 6116 } 6117 } 6118 6119 if (lacp_rate) { 6120 if (bond_mode != BOND_MODE_8023AD) { 6121 pr_info("lacp_rate param is irrelevant in mode %s\n", 6122 bond_mode_name(bond_mode)); 6123 } else { 6124 bond_opt_initstr(&newval, lacp_rate); 6125 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 6126 &newval); 6127 if (!valptr) { 6128 pr_err("Error: Invalid lacp rate \"%s\"\n", 6129 lacp_rate); 6130 return -EINVAL; 6131 } 6132 lacp_fast = valptr->value; 6133 } 6134 } 6135 6136 if (ad_select) { 6137 bond_opt_initstr(&newval, ad_select); 6138 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 6139 &newval); 6140 if (!valptr) { 6141 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 6142 return -EINVAL; 6143 } 6144 params->ad_select = valptr->value; 6145 if (bond_mode != BOND_MODE_8023AD) 6146 pr_warn("ad_select param only affects 802.3ad mode\n"); 6147 } else { 6148 params->ad_select = BOND_AD_STABLE; 6149 } 6150 6151 if (max_bonds < 0) { 6152 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 6153 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 6154 max_bonds = BOND_DEFAULT_MAX_BONDS; 6155 } 6156 6157 if (miimon < 0) { 6158 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6159 miimon, INT_MAX); 6160 miimon = 0; 6161 } 6162 6163 if (updelay < 0) { 6164 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6165 updelay, INT_MAX); 6166 updelay = 0; 6167 } 6168 6169 if (downdelay < 0) { 6170 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6171 downdelay, INT_MAX); 6172 downdelay = 0; 6173 } 6174 6175 if ((use_carrier != 0) && (use_carrier != 1)) { 6176 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 6177 use_carrier); 6178 use_carrier = 1; 6179 } 6180 6181 if (num_peer_notif < 0 || num_peer_notif > 255) { 6182 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 6183 num_peer_notif); 6184 num_peer_notif = 1; 6185 } 6186 6187 /* reset values for 802.3ad/TLB/ALB */ 6188 if (!bond_mode_uses_arp(bond_mode)) { 6189 if (!miimon) { 6190 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"); 6191 pr_warn("Forcing miimon to 100msec\n"); 6192 miimon = BOND_DEFAULT_MIIMON; 6193 } 6194 } 6195 6196 if (tx_queues < 1 || tx_queues > 255) { 6197 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 6198 tx_queues, BOND_DEFAULT_TX_QUEUES); 6199 tx_queues = BOND_DEFAULT_TX_QUEUES; 6200 } 6201 6202 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 6203 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 6204 all_slaves_active); 6205 all_slaves_active = 0; 6206 } 6207 6208 if (resend_igmp < 0 || resend_igmp > 255) { 6209 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6210 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6211 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6212 } 6213 6214 bond_opt_initval(&newval, packets_per_slave); 6215 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6216 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6217 packets_per_slave, USHRT_MAX); 6218 packets_per_slave = 1; 6219 } 6220 6221 if (bond_mode == BOND_MODE_ALB) { 6222 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", 6223 updelay); 6224 } 6225 6226 if (!miimon) { 6227 if (updelay || downdelay) { 6228 /* just warn the user the up/down delay will have 6229 * no effect since miimon is zero... 6230 */ 6231 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", 6232 updelay, downdelay); 6233 } 6234 } else { 6235 /* don't allow arp monitoring */ 6236 if (arp_interval) { 6237 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6238 miimon, arp_interval); 6239 arp_interval = 0; 6240 } 6241 6242 if ((updelay % miimon) != 0) { 6243 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6244 updelay, miimon, (updelay / miimon) * miimon); 6245 } 6246 6247 updelay /= miimon; 6248 6249 if ((downdelay % miimon) != 0) { 6250 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6251 downdelay, miimon, 6252 (downdelay / miimon) * miimon); 6253 } 6254 6255 downdelay /= miimon; 6256 } 6257 6258 if (arp_interval < 0) { 6259 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6260 arp_interval, INT_MAX); 6261 arp_interval = 0; 6262 } 6263 6264 for (arp_ip_count = 0, i = 0; 6265 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6266 __be32 ip; 6267 6268 /* not a complete check, but good enough to catch mistakes */ 6269 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6270 !bond_is_ip_target_ok(ip)) { 6271 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6272 arp_ip_target[i]); 6273 arp_interval = 0; 6274 } else { 6275 if (bond_get_targets_ip(arp_target, ip) == -1) 6276 arp_target[arp_ip_count++] = ip; 6277 else 6278 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6279 &ip); 6280 } 6281 } 6282 6283 if (arp_interval && !arp_ip_count) { 6284 /* don't allow arping if no arp_ip_target given... */ 6285 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6286 arp_interval); 6287 arp_interval = 0; 6288 } 6289 6290 if (arp_validate) { 6291 if (!arp_interval) { 6292 pr_err("arp_validate requires arp_interval\n"); 6293 return -EINVAL; 6294 } 6295 6296 bond_opt_initstr(&newval, arp_validate); 6297 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6298 &newval); 6299 if (!valptr) { 6300 pr_err("Error: invalid arp_validate \"%s\"\n", 6301 arp_validate); 6302 return -EINVAL; 6303 } 6304 arp_validate_value = valptr->value; 6305 } else { 6306 arp_validate_value = 0; 6307 } 6308 6309 if (arp_all_targets) { 6310 bond_opt_initstr(&newval, arp_all_targets); 6311 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6312 &newval); 6313 if (!valptr) { 6314 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6315 arp_all_targets); 6316 arp_all_targets_value = 0; 6317 } else { 6318 arp_all_targets_value = valptr->value; 6319 } 6320 } 6321 6322 if (miimon) { 6323 pr_info("MII link monitoring set to %d ms\n", miimon); 6324 } else if (arp_interval) { 6325 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6326 arp_validate_value); 6327 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6328 arp_interval, valptr->string, arp_ip_count); 6329 6330 for (i = 0; i < arp_ip_count; i++) 6331 pr_cont(" %s", arp_ip_target[i]); 6332 6333 pr_cont("\n"); 6334 6335 } else if (max_bonds) { 6336 /* miimon and arp_interval not set, we need one so things 6337 * work as expected, see bonding.txt for details 6338 */ 6339 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"); 6340 } 6341 6342 if (primary && !bond_mode_uses_primary(bond_mode)) { 6343 /* currently, using a primary only makes sense 6344 * in active backup, TLB or ALB modes 6345 */ 6346 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6347 primary, bond_mode_name(bond_mode)); 6348 primary = NULL; 6349 } 6350 6351 if (primary && primary_reselect) { 6352 bond_opt_initstr(&newval, primary_reselect); 6353 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6354 &newval); 6355 if (!valptr) { 6356 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6357 primary_reselect); 6358 return -EINVAL; 6359 } 6360 primary_reselect_value = valptr->value; 6361 } else { 6362 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6363 } 6364 6365 if (fail_over_mac) { 6366 bond_opt_initstr(&newval, fail_over_mac); 6367 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6368 &newval); 6369 if (!valptr) { 6370 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6371 fail_over_mac); 6372 return -EINVAL; 6373 } 6374 fail_over_mac_value = valptr->value; 6375 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6376 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6377 } else { 6378 fail_over_mac_value = BOND_FOM_NONE; 6379 } 6380 6381 bond_opt_initstr(&newval, "default"); 6382 valptr = bond_opt_parse( 6383 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6384 &newval); 6385 if (!valptr) { 6386 pr_err("Error: No ad_actor_sys_prio default value"); 6387 return -EINVAL; 6388 } 6389 ad_actor_sys_prio = valptr->value; 6390 6391 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6392 &newval); 6393 if (!valptr) { 6394 pr_err("Error: No ad_user_port_key default value"); 6395 return -EINVAL; 6396 } 6397 ad_user_port_key = valptr->value; 6398 6399 bond_opt_initstr(&newval, "default"); 6400 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6401 if (!valptr) { 6402 pr_err("Error: No tlb_dynamic_lb default value"); 6403 return -EINVAL; 6404 } 6405 tlb_dynamic_lb = valptr->value; 6406 6407 if (lp_interval == 0) { 6408 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6409 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6410 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6411 } 6412 6413 /* fill params struct with the proper values */ 6414 params->mode = bond_mode; 6415 params->xmit_policy = xmit_hashtype; 6416 params->miimon = miimon; 6417 params->num_peer_notif = num_peer_notif; 6418 params->arp_interval = arp_interval; 6419 params->arp_validate = arp_validate_value; 6420 params->arp_all_targets = arp_all_targets_value; 6421 params->missed_max = 2; 6422 params->updelay = updelay; 6423 params->downdelay = downdelay; 6424 params->peer_notif_delay = 0; 6425 params->use_carrier = use_carrier; 6426 params->lacp_active = 1; 6427 params->lacp_fast = lacp_fast; 6428 params->primary[0] = 0; 6429 params->primary_reselect = primary_reselect_value; 6430 params->fail_over_mac = fail_over_mac_value; 6431 params->tx_queues = tx_queues; 6432 params->all_slaves_active = all_slaves_active; 6433 params->resend_igmp = resend_igmp; 6434 params->min_links = min_links; 6435 params->lp_interval = lp_interval; 6436 params->packets_per_slave = packets_per_slave; 6437 params->tlb_dynamic_lb = tlb_dynamic_lb; 6438 params->ad_actor_sys_prio = ad_actor_sys_prio; 6439 eth_zero_addr(params->ad_actor_system); 6440 params->ad_user_port_key = ad_user_port_key; 6441 params->coupled_control = 1; 6442 if (packets_per_slave > 0) { 6443 params->reciprocal_packets_per_slave = 6444 reciprocal_value(packets_per_slave); 6445 } else { 6446 /* reciprocal_packets_per_slave is unused if 6447 * packets_per_slave is 0 or 1, just initialize it 6448 */ 6449 params->reciprocal_packets_per_slave = 6450 (struct reciprocal_value) { 0 }; 6451 } 6452 6453 if (primary) 6454 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6455 6456 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6457 #if IS_ENABLED(CONFIG_IPV6) 6458 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6459 #endif 6460 6461 return 0; 6462 } 6463 6464 /* Called from registration process */ 6465 static int bond_init(struct net_device *bond_dev) 6466 { 6467 struct bonding *bond = netdev_priv(bond_dev); 6468 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6469 6470 netdev_dbg(bond_dev, "Begin bond_init\n"); 6471 6472 bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM, 6473 bond_dev->name); 6474 if (!bond->wq) 6475 return -ENOMEM; 6476 6477 bond->notifier_ctx = false; 6478 6479 spin_lock_init(&bond->stats_lock); 6480 netdev_lockdep_set_classes(bond_dev); 6481 6482 list_add_tail_rcu(&bond->bond_list, &bn->dev_list); 6483 6484 bond_prepare_sysfs_group(bond); 6485 6486 bond_debug_register(bond); 6487 6488 /* Ensure valid dev_addr */ 6489 if (is_zero_ether_addr(bond_dev->dev_addr) && 6490 bond_dev->addr_assign_type == NET_ADDR_PERM) 6491 eth_hw_addr_random(bond_dev); 6492 6493 return 0; 6494 } 6495 6496 unsigned int bond_get_num_tx_queues(void) 6497 { 6498 return tx_queues; 6499 } 6500 6501 /* Create a new bond based on the specified name and bonding parameters. 6502 * If name is NULL, obtain a suitable "bond%d" name for us. 6503 * Caller must NOT hold rtnl_lock; we need to release it here before we 6504 * set up our sysfs entries. 6505 */ 6506 int bond_create(struct net *net, const char *name) 6507 { 6508 struct net_device *bond_dev; 6509 struct bonding *bond; 6510 int res = -ENOMEM; 6511 6512 rtnl_lock(); 6513 6514 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6515 name ? name : "bond%d", NET_NAME_UNKNOWN, 6516 bond_setup, tx_queues); 6517 if (!bond_dev) 6518 goto out; 6519 6520 bond = netdev_priv(bond_dev); 6521 dev_net_set(bond_dev, net); 6522 bond_dev->rtnl_link_ops = &bond_link_ops; 6523 6524 res = register_netdevice(bond_dev); 6525 if (res < 0) { 6526 free_netdev(bond_dev); 6527 goto out; 6528 } 6529 6530 netif_carrier_off(bond_dev); 6531 6532 bond_work_init_all(bond); 6533 6534 out: 6535 rtnl_unlock(); 6536 return res; 6537 } 6538 6539 static int __net_init bond_net_init(struct net *net) 6540 { 6541 struct bond_net *bn = net_generic(net, bond_net_id); 6542 6543 bn->net = net; 6544 INIT_LIST_HEAD(&bn->dev_list); 6545 6546 bond_create_proc_dir(bn); 6547 bond_create_sysfs(bn); 6548 6549 return 0; 6550 } 6551 6552 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters 6553 * race condition in bond unloading") we need to remove sysfs files 6554 * before we remove our devices (done later in bond_net_exit_batch_rtnl()) 6555 */ 6556 static void __net_exit bond_net_pre_exit(struct net *net) 6557 { 6558 struct bond_net *bn = net_generic(net, bond_net_id); 6559 6560 bond_destroy_sysfs(bn); 6561 } 6562 6563 static void __net_exit bond_net_exit_batch_rtnl(struct list_head *net_list, 6564 struct list_head *dev_kill_list) 6565 { 6566 struct bond_net *bn; 6567 struct net *net; 6568 6569 /* Kill off any bonds created after unregistering bond rtnl ops */ 6570 list_for_each_entry(net, net_list, exit_list) { 6571 struct bonding *bond, *tmp_bond; 6572 6573 bn = net_generic(net, bond_net_id); 6574 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6575 unregister_netdevice_queue(bond->dev, dev_kill_list); 6576 } 6577 } 6578 6579 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory 6580 * only after all bonds are gone") bond_destroy_proc_dir() is called 6581 * after bond_net_exit_batch_rtnl() has completed. 6582 */ 6583 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6584 { 6585 struct bond_net *bn; 6586 struct net *net; 6587 6588 list_for_each_entry(net, net_list, exit_list) { 6589 bn = net_generic(net, bond_net_id); 6590 bond_destroy_proc_dir(bn); 6591 } 6592 } 6593 6594 static struct pernet_operations bond_net_ops = { 6595 .init = bond_net_init, 6596 .pre_exit = bond_net_pre_exit, 6597 .exit_batch_rtnl = bond_net_exit_batch_rtnl, 6598 .exit_batch = bond_net_exit_batch, 6599 .id = &bond_net_id, 6600 .size = sizeof(struct bond_net), 6601 }; 6602 6603 static int __init bonding_init(void) 6604 { 6605 int i; 6606 int res; 6607 6608 res = bond_check_params(&bonding_defaults); 6609 if (res) 6610 goto out; 6611 6612 bond_create_debugfs(); 6613 6614 res = register_pernet_subsys(&bond_net_ops); 6615 if (res) 6616 goto err_net_ops; 6617 6618 res = bond_netlink_init(); 6619 if (res) 6620 goto err_link; 6621 6622 for (i = 0; i < max_bonds; i++) { 6623 res = bond_create(&init_net, NULL); 6624 if (res) 6625 goto err; 6626 } 6627 6628 skb_flow_dissector_init(&flow_keys_bonding, 6629 flow_keys_bonding_keys, 6630 ARRAY_SIZE(flow_keys_bonding_keys)); 6631 6632 register_netdevice_notifier(&bond_netdev_notifier); 6633 out: 6634 return res; 6635 err: 6636 bond_netlink_fini(); 6637 err_link: 6638 unregister_pernet_subsys(&bond_net_ops); 6639 err_net_ops: 6640 bond_destroy_debugfs(); 6641 goto out; 6642 6643 } 6644 6645 static void __exit bonding_exit(void) 6646 { 6647 unregister_netdevice_notifier(&bond_netdev_notifier); 6648 6649 bond_netlink_fini(); 6650 unregister_pernet_subsys(&bond_net_ops); 6651 6652 bond_destroy_debugfs(); 6653 6654 #ifdef CONFIG_NET_POLL_CONTROLLER 6655 /* Make sure we don't have an imbalance on our netpoll blocking */ 6656 WARN_ON(atomic_read(&netpoll_block_tx)); 6657 #endif 6658 } 6659 6660 module_init(bonding_init); 6661 module_exit(bonding_exit); 6662 MODULE_LICENSE("GPL"); 6663 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6664 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6665