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