1 // SPDX-License-Identifier: GPL-1.0+ 2 /* 3 * originally based on the dummy device. 4 * 5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 6 * Based on dummy.c, and eql.c devices. 7 * 8 * bonding.c: an Ethernet Bonding driver 9 * 10 * This is useful to talk to a Cisco EtherChannel compatible equipment: 11 * Cisco 5500 12 * Sun Trunking (Solaris) 13 * Alteon AceDirector Trunks 14 * Linux Bonding 15 * and probably many L2 switches ... 16 * 17 * How it works: 18 * ifconfig bond0 ipaddress netmask up 19 * will setup a network device, with an ip address. No mac address 20 * will be assigned at this time. The hw mac address will come from 21 * the first slave bonded to the channel. All slaves will then use 22 * this hw mac address. 23 * 24 * ifconfig bond0 down 25 * will release all slaves, marking them as down. 26 * 27 * ifenslave bond0 eth0 28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 29 * a: be used as initial mac address 30 * b: if a hw mac address already is there, eth0's hw mac address 31 * will then be set from bond0. 32 * 33 */ 34 35 #include <linux/kernel.h> 36 #include <linux/module.h> 37 #include <linux/types.h> 38 #include <linux/fcntl.h> 39 #include <linux/filter.h> 40 #include <linux/interrupt.h> 41 #include <linux/ptrace.h> 42 #include <linux/ioport.h> 43 #include <linux/in.h> 44 #include <net/ip.h> 45 #include <linux/ip.h> 46 #include <linux/icmp.h> 47 #include <linux/icmpv6.h> 48 #include <linux/tcp.h> 49 #include <linux/udp.h> 50 #include <linux/slab.h> 51 #include <linux/string.h> 52 #include <linux/init.h> 53 #include <linux/timer.h> 54 #include <linux/socket.h> 55 #include <linux/ctype.h> 56 #include <linux/inet.h> 57 #include <linux/bitops.h> 58 #include <linux/io.h> 59 #include <asm/dma.h> 60 #include <linux/uaccess.h> 61 #include <linux/errno.h> 62 #include <linux/netdevice.h> 63 #include <linux/inetdevice.h> 64 #include <linux/igmp.h> 65 #include <linux/etherdevice.h> 66 #include <linux/skbuff.h> 67 #include <net/sock.h> 68 #include <linux/rtnetlink.h> 69 #include <linux/smp.h> 70 #include <linux/if_ether.h> 71 #include <net/arp.h> 72 #include <linux/mii.h> 73 #include <linux/ethtool.h> 74 #include <linux/if_vlan.h> 75 #include <linux/if_bonding.h> 76 #include <linux/phy.h> 77 #include <linux/jiffies.h> 78 #include <linux/preempt.h> 79 #include <net/route.h> 80 #include <net/net_namespace.h> 81 #include <net/netns/generic.h> 82 #include <net/pkt_sched.h> 83 #include <linux/rculist.h> 84 #include <net/flow_dissector.h> 85 #include <net/xfrm.h> 86 #include <net/bonding.h> 87 #include <net/bond_3ad.h> 88 #include <net/bond_alb.h> 89 #if IS_ENABLED(CONFIG_TLS_DEVICE) 90 #include <net/tls.h> 91 #endif 92 #include <net/ip6_route.h> 93 #include <net/netdev_lock.h> 94 #include <net/xdp.h> 95 96 #include "bonding_priv.h" 97 98 /*---------------------------- Module parameters ----------------------------*/ 99 100 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 101 102 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 103 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 104 static int num_peer_notif = 1; 105 static int miimon; 106 static int updelay; 107 static int downdelay; 108 static int use_carrier = 1; 109 static char *mode; 110 static char *primary; 111 static char *primary_reselect; 112 static char *lacp_rate; 113 static int min_links; 114 static char *ad_select; 115 static char *xmit_hash_policy; 116 static int arp_interval; 117 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 118 static char *arp_validate; 119 static char *arp_all_targets; 120 static char *fail_over_mac; 121 static int all_slaves_active; 122 static struct bond_params bonding_defaults; 123 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 124 static int packets_per_slave = 1; 125 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 126 127 module_param(max_bonds, int, 0); 128 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 129 module_param(tx_queues, int, 0); 130 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 131 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 132 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 133 "failover event (alias of num_unsol_na)"); 134 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 135 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 136 "failover event (alias of num_grat_arp)"); 137 module_param(miimon, int, 0); 138 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 139 module_param(updelay, int, 0); 140 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 141 module_param(downdelay, int, 0); 142 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 143 "in milliseconds"); 144 module_param(use_carrier, int, 0); 145 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 146 "0 for off, 1 for on (default)"); 147 module_param(mode, charp, 0); 148 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 149 "1 for active-backup, 2 for balance-xor, " 150 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 151 "6 for balance-alb"); 152 module_param(primary, charp, 0); 153 MODULE_PARM_DESC(primary, "Primary network device to use"); 154 module_param(primary_reselect, charp, 0); 155 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 156 "once it comes up; " 157 "0 for always (default), " 158 "1 for only if speed of primary is " 159 "better, " 160 "2 for only on active slave " 161 "failure"); 162 module_param(lacp_rate, charp, 0); 163 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 164 "0 for slow, 1 for fast"); 165 module_param(ad_select, charp, 0); 166 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 167 "0 for stable (default), 1 for bandwidth, " 168 "2 for count"); 169 module_param(min_links, int, 0); 170 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 171 172 module_param(xmit_hash_policy, charp, 0); 173 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 174 "0 for layer 2 (default), 1 for layer 3+4, " 175 "2 for layer 2+3, 3 for encap layer 2+3, " 176 "4 for encap layer 3+4, 5 for vlan+srcmac"); 177 module_param(arp_interval, int, 0); 178 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 179 module_param_array(arp_ip_target, charp, NULL, 0); 180 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 181 module_param(arp_validate, charp, 0); 182 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 183 "0 for none (default), 1 for active, " 184 "2 for backup, 3 for all"); 185 module_param(arp_all_targets, charp, 0); 186 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 187 module_param(fail_over_mac, charp, 0); 188 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 189 "the same MAC; 0 for none (default), " 190 "1 for active, 2 for follow"); 191 module_param(all_slaves_active, int, 0); 192 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 193 "by setting active flag for all slaves; " 194 "0 for never (default), 1 for always."); 195 module_param(resend_igmp, int, 0); 196 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 197 "link failure"); 198 module_param(packets_per_slave, int, 0); 199 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 200 "mode; 0 for a random slave, 1 packet per " 201 "slave (default), >1 packets per slave."); 202 module_param(lp_interval, uint, 0); 203 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 204 "the bonding driver sends learning packets to " 205 "each slaves peer switch. The default is 1."); 206 207 /*----------------------------- Global variables ----------------------------*/ 208 209 #ifdef CONFIG_NET_POLL_CONTROLLER 210 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 211 #endif 212 213 unsigned int bond_net_id __read_mostly; 214 215 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 216 { 217 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 218 .offset = offsetof(struct flow_keys, control), 219 }, 220 { 221 .key_id = FLOW_DISSECTOR_KEY_BASIC, 222 .offset = offsetof(struct flow_keys, basic), 223 }, 224 { 225 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 226 .offset = offsetof(struct flow_keys, addrs.v4addrs), 227 }, 228 { 229 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 230 .offset = offsetof(struct flow_keys, addrs.v6addrs), 231 }, 232 { 233 .key_id = FLOW_DISSECTOR_KEY_TIPC, 234 .offset = offsetof(struct flow_keys, addrs.tipckey), 235 }, 236 { 237 .key_id = FLOW_DISSECTOR_KEY_PORTS, 238 .offset = offsetof(struct flow_keys, ports), 239 }, 240 { 241 .key_id = FLOW_DISSECTOR_KEY_ICMP, 242 .offset = offsetof(struct flow_keys, icmp), 243 }, 244 { 245 .key_id = FLOW_DISSECTOR_KEY_VLAN, 246 .offset = offsetof(struct flow_keys, vlan), 247 }, 248 { 249 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 250 .offset = offsetof(struct flow_keys, tags), 251 }, 252 { 253 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 254 .offset = offsetof(struct flow_keys, keyid), 255 }, 256 }; 257 258 static struct flow_dissector flow_keys_bonding __read_mostly; 259 260 /*-------------------------- Forward declarations ---------------------------*/ 261 262 static int bond_init(struct net_device *bond_dev); 263 static void bond_uninit(struct net_device *bond_dev); 264 static void bond_get_stats(struct net_device *bond_dev, 265 struct rtnl_link_stats64 *stats); 266 static void bond_slave_arr_handler(struct work_struct *work); 267 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 268 int mod); 269 static void bond_netdev_notify_work(struct work_struct *work); 270 271 /*---------------------------- General routines -----------------------------*/ 272 273 const char *bond_mode_name(int mode) 274 { 275 static const char *names[] = { 276 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 277 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 278 [BOND_MODE_XOR] = "load balancing (xor)", 279 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 280 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 281 [BOND_MODE_TLB] = "transmit load balancing", 282 [BOND_MODE_ALB] = "adaptive load balancing", 283 }; 284 285 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 286 return "unknown"; 287 288 return names[mode]; 289 } 290 291 /** 292 * bond_dev_queue_xmit - Prepare skb for xmit. 293 * 294 * @bond: bond device that got this skb for tx. 295 * @skb: hw accel VLAN tagged skb to transmit 296 * @slave_dev: slave that is supposed to xmit this skbuff 297 */ 298 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 299 struct net_device *slave_dev) 300 { 301 skb->dev = slave_dev; 302 303 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 304 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 305 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 306 307 if (unlikely(netpoll_tx_running(bond->dev))) 308 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 309 310 return dev_queue_xmit(skb); 311 } 312 313 static bool bond_sk_check(struct bonding *bond) 314 { 315 switch (BOND_MODE(bond)) { 316 case BOND_MODE_8023AD: 317 case BOND_MODE_XOR: 318 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 319 return true; 320 fallthrough; 321 default: 322 return false; 323 } 324 } 325 326 static bool bond_xdp_check(struct bonding *bond) 327 { 328 switch (BOND_MODE(bond)) { 329 case BOND_MODE_ROUNDROBIN: 330 case BOND_MODE_ACTIVEBACKUP: 331 return true; 332 case BOND_MODE_8023AD: 333 case BOND_MODE_XOR: 334 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 335 * payload is not in the packet due to hardware offload. 336 */ 337 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 338 return true; 339 fallthrough; 340 default: 341 return false; 342 } 343 } 344 345 /*---------------------------------- VLAN -----------------------------------*/ 346 347 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 348 * We don't protect the slave list iteration with a lock because: 349 * a. This operation is performed in IOCTL context, 350 * b. The operation is protected by the RTNL semaphore in the 8021q code, 351 * c. Holding a lock with BH disabled while directly calling a base driver 352 * entry point is generally a BAD idea. 353 * 354 * The design of synchronization/protection for this operation in the 8021q 355 * module is good for one or more VLAN devices over a single physical device 356 * and cannot be extended for a teaming solution like bonding, so there is a 357 * potential race condition here where a net device from the vlan group might 358 * be referenced (either by a base driver or the 8021q code) while it is being 359 * removed from the system. However, it turns out we're not making matters 360 * worse, and if it works for regular VLAN usage it will work here too. 361 */ 362 363 /** 364 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 365 * @bond_dev: bonding net device that got called 366 * @proto: network protocol ID 367 * @vid: vlan id being added 368 */ 369 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 370 __be16 proto, u16 vid) 371 { 372 struct bonding *bond = netdev_priv(bond_dev); 373 struct slave *slave, *rollback_slave; 374 struct list_head *iter; 375 int res; 376 377 bond_for_each_slave(bond, slave, iter) { 378 res = vlan_vid_add(slave->dev, proto, vid); 379 if (res) 380 goto unwind; 381 } 382 383 return 0; 384 385 unwind: 386 /* unwind to the slave that failed */ 387 bond_for_each_slave(bond, rollback_slave, iter) { 388 if (rollback_slave == slave) 389 break; 390 391 vlan_vid_del(rollback_slave->dev, proto, vid); 392 } 393 394 return res; 395 } 396 397 /** 398 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 399 * @bond_dev: bonding net device that got called 400 * @proto: network protocol ID 401 * @vid: vlan id being removed 402 */ 403 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 404 __be16 proto, u16 vid) 405 { 406 struct bonding *bond = netdev_priv(bond_dev); 407 struct list_head *iter; 408 struct slave *slave; 409 410 bond_for_each_slave(bond, slave, iter) 411 vlan_vid_del(slave->dev, proto, vid); 412 413 if (bond_is_lb(bond)) 414 bond_alb_clear_vlan(bond, vid); 415 416 return 0; 417 } 418 419 /*---------------------------------- XFRM -----------------------------------*/ 420 421 #ifdef CONFIG_XFRM_OFFLOAD 422 /** 423 * bond_ipsec_dev - Get active device for IPsec offload 424 * @xs: pointer to transformer state struct 425 * 426 * Context: caller must hold rcu_read_lock. 427 * 428 * Return: the device for ipsec offload, or NULL if not exist. 429 **/ 430 static struct net_device *bond_ipsec_dev(struct xfrm_state *xs) 431 { 432 struct net_device *bond_dev = xs->xso.dev; 433 struct bonding *bond; 434 struct slave *slave; 435 436 bond = netdev_priv(bond_dev); 437 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 438 return NULL; 439 440 slave = rcu_dereference(bond->curr_active_slave); 441 if (!slave) 442 return NULL; 443 444 if (!xs->xso.real_dev) 445 return NULL; 446 447 if (xs->xso.real_dev != slave->dev) 448 pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n", 449 bond_dev->name, slave->dev->name, xs->xso.real_dev->name); 450 451 return slave->dev; 452 } 453 454 /** 455 * bond_ipsec_add_sa - program device with a security association 456 * @xs: pointer to transformer state struct 457 * @extack: extack point to fill failure reason 458 **/ 459 static int bond_ipsec_add_sa(struct xfrm_state *xs, 460 struct netlink_ext_ack *extack) 461 { 462 struct net_device *bond_dev = xs->xso.dev; 463 struct net_device *real_dev; 464 netdevice_tracker tracker; 465 struct bond_ipsec *ipsec; 466 struct bonding *bond; 467 struct slave *slave; 468 int err; 469 470 if (!bond_dev) 471 return -EINVAL; 472 473 rcu_read_lock(); 474 bond = netdev_priv(bond_dev); 475 slave = rcu_dereference(bond->curr_active_slave); 476 real_dev = slave ? slave->dev : NULL; 477 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 478 rcu_read_unlock(); 479 if (!real_dev) { 480 err = -ENODEV; 481 goto out; 482 } 483 484 if (!real_dev->xfrmdev_ops || 485 !real_dev->xfrmdev_ops->xdo_dev_state_add || 486 netif_is_bond_master(real_dev)) { 487 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload"); 488 err = -EINVAL; 489 goto out; 490 } 491 492 ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL); 493 if (!ipsec) { 494 err = -ENOMEM; 495 goto out; 496 } 497 498 xs->xso.real_dev = real_dev; 499 err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack); 500 if (!err) { 501 ipsec->xs = xs; 502 INIT_LIST_HEAD(&ipsec->list); 503 mutex_lock(&bond->ipsec_lock); 504 list_add(&ipsec->list, &bond->ipsec_list); 505 mutex_unlock(&bond->ipsec_lock); 506 } else { 507 kfree(ipsec); 508 } 509 out: 510 netdev_put(real_dev, &tracker); 511 return err; 512 } 513 514 static void bond_ipsec_add_sa_all(struct bonding *bond) 515 { 516 struct net_device *bond_dev = bond->dev; 517 struct net_device *real_dev; 518 struct bond_ipsec *ipsec; 519 struct slave *slave; 520 521 slave = rtnl_dereference(bond->curr_active_slave); 522 real_dev = slave ? slave->dev : NULL; 523 if (!real_dev) 524 return; 525 526 mutex_lock(&bond->ipsec_lock); 527 if (!real_dev->xfrmdev_ops || 528 !real_dev->xfrmdev_ops->xdo_dev_state_add || 529 netif_is_bond_master(real_dev)) { 530 if (!list_empty(&bond->ipsec_list)) 531 slave_warn(bond_dev, real_dev, 532 "%s: no slave xdo_dev_state_add\n", 533 __func__); 534 goto out; 535 } 536 537 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 538 /* If new state is added before ipsec_lock acquired */ 539 if (ipsec->xs->xso.real_dev == real_dev) 540 continue; 541 542 ipsec->xs->xso.real_dev = real_dev; 543 if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) { 544 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__); 545 ipsec->xs->xso.real_dev = NULL; 546 } 547 } 548 out: 549 mutex_unlock(&bond->ipsec_lock); 550 } 551 552 /** 553 * bond_ipsec_del_sa - clear out this specific SA 554 * @xs: pointer to transformer state struct 555 **/ 556 static void bond_ipsec_del_sa(struct xfrm_state *xs) 557 { 558 struct net_device *bond_dev = xs->xso.dev; 559 struct net_device *real_dev; 560 netdevice_tracker tracker; 561 struct bond_ipsec *ipsec; 562 struct bonding *bond; 563 struct slave *slave; 564 565 if (!bond_dev) 566 return; 567 568 rcu_read_lock(); 569 bond = netdev_priv(bond_dev); 570 slave = rcu_dereference(bond->curr_active_slave); 571 real_dev = slave ? slave->dev : NULL; 572 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 573 rcu_read_unlock(); 574 575 if (!slave) 576 goto out; 577 578 if (!xs->xso.real_dev) 579 goto out; 580 581 WARN_ON(xs->xso.real_dev != real_dev); 582 583 if (!real_dev->xfrmdev_ops || 584 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 585 netif_is_bond_master(real_dev)) { 586 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__); 587 goto out; 588 } 589 590 real_dev->xfrmdev_ops->xdo_dev_state_delete(xs); 591 out: 592 netdev_put(real_dev, &tracker); 593 mutex_lock(&bond->ipsec_lock); 594 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 595 if (ipsec->xs == xs) { 596 list_del(&ipsec->list); 597 kfree(ipsec); 598 break; 599 } 600 } 601 mutex_unlock(&bond->ipsec_lock); 602 } 603 604 static void bond_ipsec_del_sa_all(struct bonding *bond) 605 { 606 struct net_device *bond_dev = bond->dev; 607 struct net_device *real_dev; 608 struct bond_ipsec *ipsec; 609 struct slave *slave; 610 611 slave = rtnl_dereference(bond->curr_active_slave); 612 real_dev = slave ? slave->dev : NULL; 613 if (!real_dev) 614 return; 615 616 mutex_lock(&bond->ipsec_lock); 617 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 618 if (!ipsec->xs->xso.real_dev) 619 continue; 620 621 if (!real_dev->xfrmdev_ops || 622 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 623 netif_is_bond_master(real_dev)) { 624 slave_warn(bond_dev, real_dev, 625 "%s: no slave xdo_dev_state_delete\n", 626 __func__); 627 } else { 628 real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 629 if (real_dev->xfrmdev_ops->xdo_dev_state_free) 630 real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs); 631 } 632 } 633 mutex_unlock(&bond->ipsec_lock); 634 } 635 636 static void bond_ipsec_free_sa(struct xfrm_state *xs) 637 { 638 struct net_device *bond_dev = xs->xso.dev; 639 struct net_device *real_dev; 640 netdevice_tracker tracker; 641 struct bonding *bond; 642 struct slave *slave; 643 644 if (!bond_dev) 645 return; 646 647 rcu_read_lock(); 648 bond = netdev_priv(bond_dev); 649 slave = rcu_dereference(bond->curr_active_slave); 650 real_dev = slave ? slave->dev : NULL; 651 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 652 rcu_read_unlock(); 653 654 if (!slave) 655 goto out; 656 657 if (!xs->xso.real_dev) 658 goto out; 659 660 WARN_ON(xs->xso.real_dev != real_dev); 661 662 if (real_dev && real_dev->xfrmdev_ops && 663 real_dev->xfrmdev_ops->xdo_dev_state_free) 664 real_dev->xfrmdev_ops->xdo_dev_state_free(xs); 665 out: 666 netdev_put(real_dev, &tracker); 667 } 668 669 /** 670 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 671 * @skb: current data packet 672 * @xs: pointer to transformer state struct 673 **/ 674 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 675 { 676 struct net_device *real_dev; 677 bool ok = false; 678 679 rcu_read_lock(); 680 real_dev = bond_ipsec_dev(xs); 681 if (!real_dev) 682 goto out; 683 684 if (!real_dev->xfrmdev_ops || 685 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 686 netif_is_bond_master(real_dev)) 687 goto out; 688 689 ok = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 690 out: 691 rcu_read_unlock(); 692 return ok; 693 } 694 695 /** 696 * bond_advance_esn_state - ESN support for IPSec HW offload 697 * @xs: pointer to transformer state struct 698 **/ 699 static void bond_advance_esn_state(struct xfrm_state *xs) 700 { 701 struct net_device *real_dev; 702 703 rcu_read_lock(); 704 real_dev = bond_ipsec_dev(xs); 705 if (!real_dev) 706 goto out; 707 708 if (!real_dev->xfrmdev_ops || 709 !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) { 710 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name); 711 goto out; 712 } 713 714 real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs); 715 out: 716 rcu_read_unlock(); 717 } 718 719 /** 720 * bond_xfrm_update_stats - Update xfrm state 721 * @xs: pointer to transformer state struct 722 **/ 723 static void bond_xfrm_update_stats(struct xfrm_state *xs) 724 { 725 struct net_device *real_dev; 726 727 rcu_read_lock(); 728 real_dev = bond_ipsec_dev(xs); 729 if (!real_dev) 730 goto out; 731 732 if (!real_dev->xfrmdev_ops || 733 !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) { 734 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name); 735 goto out; 736 } 737 738 real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs); 739 out: 740 rcu_read_unlock(); 741 } 742 743 static const struct xfrmdev_ops bond_xfrmdev_ops = { 744 .xdo_dev_state_add = bond_ipsec_add_sa, 745 .xdo_dev_state_delete = bond_ipsec_del_sa, 746 .xdo_dev_state_free = bond_ipsec_free_sa, 747 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 748 .xdo_dev_state_advance_esn = bond_advance_esn_state, 749 .xdo_dev_state_update_stats = bond_xfrm_update_stats, 750 }; 751 #endif /* CONFIG_XFRM_OFFLOAD */ 752 753 /*------------------------------- Link status -------------------------------*/ 754 755 /* Set the carrier state for the master according to the state of its 756 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 757 * do special 802.3ad magic. 758 * 759 * Returns zero if carrier state does not change, nonzero if it does. 760 */ 761 int bond_set_carrier(struct bonding *bond) 762 { 763 struct list_head *iter; 764 struct slave *slave; 765 766 if (!bond_has_slaves(bond)) 767 goto down; 768 769 if (BOND_MODE(bond) == BOND_MODE_8023AD) 770 return bond_3ad_set_carrier(bond); 771 772 bond_for_each_slave(bond, slave, iter) { 773 if (slave->link == BOND_LINK_UP) { 774 if (!netif_carrier_ok(bond->dev)) { 775 netif_carrier_on(bond->dev); 776 return 1; 777 } 778 return 0; 779 } 780 } 781 782 down: 783 if (netif_carrier_ok(bond->dev)) { 784 netif_carrier_off(bond->dev); 785 return 1; 786 } 787 return 0; 788 } 789 790 /* Get link speed and duplex from the slave's base driver 791 * using ethtool. If for some reason the call fails or the 792 * values are invalid, set speed and duplex to -1, 793 * and return. Return 1 if speed or duplex settings are 794 * UNKNOWN; 0 otherwise. 795 */ 796 static int bond_update_speed_duplex(struct slave *slave) 797 { 798 struct net_device *slave_dev = slave->dev; 799 struct ethtool_link_ksettings ecmd; 800 int res; 801 802 slave->speed = SPEED_UNKNOWN; 803 slave->duplex = DUPLEX_UNKNOWN; 804 805 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 806 if (res < 0) 807 return 1; 808 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 809 return 1; 810 switch (ecmd.base.duplex) { 811 case DUPLEX_FULL: 812 case DUPLEX_HALF: 813 break; 814 default: 815 return 1; 816 } 817 818 slave->speed = ecmd.base.speed; 819 slave->duplex = ecmd.base.duplex; 820 821 return 0; 822 } 823 824 const char *bond_slave_link_status(s8 link) 825 { 826 switch (link) { 827 case BOND_LINK_UP: 828 return "up"; 829 case BOND_LINK_FAIL: 830 return "going down"; 831 case BOND_LINK_DOWN: 832 return "down"; 833 case BOND_LINK_BACK: 834 return "going back"; 835 default: 836 return "unknown"; 837 } 838 } 839 840 /* if <dev> supports MII link status reporting, check its link status. 841 * 842 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 843 * depending upon the setting of the use_carrier parameter. 844 * 845 * Return either BMSR_LSTATUS, meaning that the link is up (or we 846 * can't tell and just pretend it is), or 0, meaning that the link is 847 * down. 848 * 849 * If reporting is non-zero, instead of faking link up, return -1 if 850 * both ETHTOOL and MII ioctls fail (meaning the device does not 851 * support them). If use_carrier is set, return whatever it says. 852 * It'd be nice if there was a good way to tell if a driver supports 853 * netif_carrier, but there really isn't. 854 */ 855 static int bond_check_dev_link(struct bonding *bond, 856 struct net_device *slave_dev, int reporting) 857 { 858 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 859 struct ifreq ifr; 860 struct mii_ioctl_data *mii; 861 862 if (!reporting && !netif_running(slave_dev)) 863 return 0; 864 865 if (bond->params.use_carrier) 866 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 867 868 /* Try to get link status using Ethtool first. */ 869 if (slave_dev->ethtool_ops->get_link) 870 return slave_dev->ethtool_ops->get_link(slave_dev) ? 871 BMSR_LSTATUS : 0; 872 873 /* Ethtool can't be used, fallback to MII ioctls. */ 874 if (slave_ops->ndo_eth_ioctl) { 875 /* TODO: set pointer to correct ioctl on a per team member 876 * bases to make this more efficient. that is, once 877 * we determine the correct ioctl, we will always 878 * call it and not the others for that team 879 * member. 880 */ 881 882 /* We cannot assume that SIOCGMIIPHY will also read a 883 * register; not all network drivers (e.g., e100) 884 * support that. 885 */ 886 887 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 888 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 889 mii = if_mii(&ifr); 890 891 if (dev_eth_ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 892 mii->reg_num = MII_BMSR; 893 if (dev_eth_ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 894 return mii->val_out & BMSR_LSTATUS; 895 } 896 } 897 898 /* If reporting, report that either there's no ndo_eth_ioctl, 899 * or both SIOCGMIIREG and get_link failed (meaning that we 900 * cannot report link status). If not reporting, pretend 901 * we're ok. 902 */ 903 return reporting ? -1 : BMSR_LSTATUS; 904 } 905 906 /*----------------------------- Multicast list ------------------------------*/ 907 908 /* Push the promiscuity flag down to appropriate slaves */ 909 static int bond_set_promiscuity(struct bonding *bond, int inc) 910 { 911 struct list_head *iter; 912 int err = 0; 913 914 if (bond_uses_primary(bond)) { 915 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 916 917 if (curr_active) 918 err = dev_set_promiscuity(curr_active->dev, inc); 919 } else { 920 struct slave *slave; 921 922 bond_for_each_slave(bond, slave, iter) { 923 err = dev_set_promiscuity(slave->dev, inc); 924 if (err) 925 return err; 926 } 927 } 928 return err; 929 } 930 931 /* Push the allmulti flag down to all slaves */ 932 static int bond_set_allmulti(struct bonding *bond, int inc) 933 { 934 struct list_head *iter; 935 int err = 0; 936 937 if (bond_uses_primary(bond)) { 938 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 939 940 if (curr_active) 941 err = dev_set_allmulti(curr_active->dev, inc); 942 } else { 943 struct slave *slave; 944 945 bond_for_each_slave(bond, slave, iter) { 946 err = dev_set_allmulti(slave->dev, inc); 947 if (err) 948 return err; 949 } 950 } 951 return err; 952 } 953 954 /* Retrieve the list of registered multicast addresses for the bonding 955 * device and retransmit an IGMP JOIN request to the current active 956 * slave. 957 */ 958 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 959 { 960 struct bonding *bond = container_of(work, struct bonding, 961 mcast_work.work); 962 963 if (!rtnl_trylock()) { 964 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 965 return; 966 } 967 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 968 969 if (bond->igmp_retrans > 1) { 970 bond->igmp_retrans--; 971 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 972 } 973 rtnl_unlock(); 974 } 975 976 /* Flush bond's hardware addresses from slave */ 977 static void bond_hw_addr_flush(struct net_device *bond_dev, 978 struct net_device *slave_dev) 979 { 980 struct bonding *bond = netdev_priv(bond_dev); 981 982 dev_uc_unsync(slave_dev, bond_dev); 983 dev_mc_unsync(slave_dev, bond_dev); 984 985 if (BOND_MODE(bond) == BOND_MODE_8023AD) 986 dev_mc_del(slave_dev, lacpdu_mcast_addr); 987 } 988 989 /*--------------------------- Active slave change ---------------------------*/ 990 991 /* Update the hardware address list and promisc/allmulti for the new and 992 * old active slaves (if any). Modes that are not using primary keep all 993 * slaves up date at all times; only the modes that use primary need to call 994 * this function to swap these settings during a failover. 995 */ 996 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 997 struct slave *old_active) 998 { 999 if (old_active) { 1000 if (bond->dev->flags & IFF_PROMISC) 1001 dev_set_promiscuity(old_active->dev, -1); 1002 1003 if (bond->dev->flags & IFF_ALLMULTI) 1004 dev_set_allmulti(old_active->dev, -1); 1005 1006 if (bond->dev->flags & IFF_UP) 1007 bond_hw_addr_flush(bond->dev, old_active->dev); 1008 1009 bond_slave_ns_maddrs_add(bond, old_active); 1010 } 1011 1012 if (new_active) { 1013 /* FIXME: Signal errors upstream. */ 1014 if (bond->dev->flags & IFF_PROMISC) 1015 dev_set_promiscuity(new_active->dev, 1); 1016 1017 if (bond->dev->flags & IFF_ALLMULTI) 1018 dev_set_allmulti(new_active->dev, 1); 1019 1020 if (bond->dev->flags & IFF_UP) { 1021 netif_addr_lock_bh(bond->dev); 1022 dev_uc_sync(new_active->dev, bond->dev); 1023 dev_mc_sync(new_active->dev, bond->dev); 1024 netif_addr_unlock_bh(bond->dev); 1025 } 1026 1027 bond_slave_ns_maddrs_del(bond, new_active); 1028 } 1029 } 1030 1031 /** 1032 * bond_set_dev_addr - clone slave's address to bond 1033 * @bond_dev: bond net device 1034 * @slave_dev: slave net device 1035 * 1036 * Should be called with RTNL held. 1037 */ 1038 static int bond_set_dev_addr(struct net_device *bond_dev, 1039 struct net_device *slave_dev) 1040 { 1041 int err; 1042 1043 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 1044 bond_dev, slave_dev, slave_dev->addr_len); 1045 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 1046 if (err) 1047 return err; 1048 1049 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 1050 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 1051 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 1052 return 0; 1053 } 1054 1055 static struct slave *bond_get_old_active(struct bonding *bond, 1056 struct slave *new_active) 1057 { 1058 struct slave *slave; 1059 struct list_head *iter; 1060 1061 bond_for_each_slave(bond, slave, iter) { 1062 if (slave == new_active) 1063 continue; 1064 1065 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 1066 return slave; 1067 } 1068 1069 return NULL; 1070 } 1071 1072 /* bond_do_fail_over_mac 1073 * 1074 * Perform special MAC address swapping for fail_over_mac settings 1075 * 1076 * Called with RTNL 1077 */ 1078 static void bond_do_fail_over_mac(struct bonding *bond, 1079 struct slave *new_active, 1080 struct slave *old_active) 1081 { 1082 u8 tmp_mac[MAX_ADDR_LEN]; 1083 struct sockaddr_storage ss; 1084 int rv; 1085 1086 switch (bond->params.fail_over_mac) { 1087 case BOND_FOM_ACTIVE: 1088 if (new_active) { 1089 rv = bond_set_dev_addr(bond->dev, new_active->dev); 1090 if (rv) 1091 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 1092 -rv); 1093 } 1094 break; 1095 case BOND_FOM_FOLLOW: 1096 /* if new_active && old_active, swap them 1097 * if just old_active, do nothing (going to no active slave) 1098 * if just new_active, set new_active to bond's MAC 1099 */ 1100 if (!new_active) 1101 return; 1102 1103 if (!old_active) 1104 old_active = bond_get_old_active(bond, new_active); 1105 1106 if (old_active) { 1107 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 1108 new_active->dev->addr_len); 1109 bond_hw_addr_copy(ss.__data, 1110 old_active->dev->dev_addr, 1111 old_active->dev->addr_len); 1112 ss.ss_family = new_active->dev->type; 1113 } else { 1114 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 1115 bond->dev->addr_len); 1116 ss.ss_family = bond->dev->type; 1117 } 1118 1119 rv = dev_set_mac_address(new_active->dev, 1120 (struct sockaddr *)&ss, NULL); 1121 if (rv) { 1122 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1123 -rv); 1124 goto out; 1125 } 1126 1127 if (!old_active) 1128 goto out; 1129 1130 bond_hw_addr_copy(ss.__data, tmp_mac, 1131 new_active->dev->addr_len); 1132 ss.ss_family = old_active->dev->type; 1133 1134 rv = dev_set_mac_address(old_active->dev, 1135 (struct sockaddr *)&ss, NULL); 1136 if (rv) 1137 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1138 -rv); 1139 out: 1140 break; 1141 default: 1142 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1143 bond->params.fail_over_mac); 1144 break; 1145 } 1146 1147 } 1148 1149 /** 1150 * bond_choose_primary_or_current - select the primary or high priority slave 1151 * @bond: our bonding struct 1152 * 1153 * - Check if there is a primary link. If the primary link was set and is up, 1154 * go on and do link reselection. 1155 * 1156 * - If primary link is not set or down, find the highest priority link. 1157 * If the highest priority link is not current slave, set it as primary 1158 * link and do link reselection. 1159 */ 1160 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1161 { 1162 struct slave *prim = rtnl_dereference(bond->primary_slave); 1163 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1164 struct slave *slave, *hprio = NULL; 1165 struct list_head *iter; 1166 1167 if (!prim || prim->link != BOND_LINK_UP) { 1168 bond_for_each_slave(bond, slave, iter) { 1169 if (slave->link == BOND_LINK_UP) { 1170 hprio = hprio ?: slave; 1171 if (slave->prio > hprio->prio) 1172 hprio = slave; 1173 } 1174 } 1175 1176 if (hprio && hprio != curr) { 1177 prim = hprio; 1178 goto link_reselect; 1179 } 1180 1181 if (!curr || curr->link != BOND_LINK_UP) 1182 return NULL; 1183 return curr; 1184 } 1185 1186 if (bond->force_primary) { 1187 bond->force_primary = false; 1188 return prim; 1189 } 1190 1191 link_reselect: 1192 if (!curr || curr->link != BOND_LINK_UP) 1193 return prim; 1194 1195 /* At this point, prim and curr are both up */ 1196 switch (bond->params.primary_reselect) { 1197 case BOND_PRI_RESELECT_ALWAYS: 1198 return prim; 1199 case BOND_PRI_RESELECT_BETTER: 1200 if (prim->speed < curr->speed) 1201 return curr; 1202 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1203 return curr; 1204 return prim; 1205 case BOND_PRI_RESELECT_FAILURE: 1206 return curr; 1207 default: 1208 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1209 bond->params.primary_reselect); 1210 return curr; 1211 } 1212 } 1213 1214 /** 1215 * bond_find_best_slave - select the best available slave to be the active one 1216 * @bond: our bonding struct 1217 */ 1218 static struct slave *bond_find_best_slave(struct bonding *bond) 1219 { 1220 struct slave *slave, *bestslave = NULL; 1221 struct list_head *iter; 1222 int mintime = bond->params.updelay; 1223 1224 slave = bond_choose_primary_or_current(bond); 1225 if (slave) 1226 return slave; 1227 1228 bond_for_each_slave(bond, slave, iter) { 1229 if (slave->link == BOND_LINK_UP) 1230 return slave; 1231 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1232 slave->delay < mintime) { 1233 mintime = slave->delay; 1234 bestslave = slave; 1235 } 1236 } 1237 1238 return bestslave; 1239 } 1240 1241 /* must be called in RCU critical section or with RTNL held */ 1242 static bool bond_should_notify_peers(struct bonding *bond) 1243 { 1244 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave); 1245 1246 if (!slave || !bond->send_peer_notif || 1247 bond->send_peer_notif % 1248 max(1, bond->params.peer_notif_delay) != 0 || 1249 !netif_carrier_ok(bond->dev) || 1250 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1251 return false; 1252 1253 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1254 slave ? slave->dev->name : "NULL"); 1255 1256 return true; 1257 } 1258 1259 /** 1260 * bond_change_active_slave - change the active slave into the specified one 1261 * @bond: our bonding struct 1262 * @new_active: the new slave to make the active one 1263 * 1264 * Set the new slave to the bond's settings and unset them on the old 1265 * curr_active_slave. 1266 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1267 * 1268 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1269 * because it is apparently the best available slave we have, even though its 1270 * updelay hasn't timed out yet. 1271 * 1272 * Caller must hold RTNL. 1273 */ 1274 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1275 { 1276 struct slave *old_active; 1277 1278 ASSERT_RTNL(); 1279 1280 old_active = rtnl_dereference(bond->curr_active_slave); 1281 1282 if (old_active == new_active) 1283 return; 1284 1285 #ifdef CONFIG_XFRM_OFFLOAD 1286 bond_ipsec_del_sa_all(bond); 1287 #endif /* CONFIG_XFRM_OFFLOAD */ 1288 1289 if (new_active) { 1290 new_active->last_link_up = jiffies; 1291 1292 if (new_active->link == BOND_LINK_BACK) { 1293 if (bond_uses_primary(bond)) { 1294 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1295 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1296 } 1297 1298 new_active->delay = 0; 1299 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1300 BOND_SLAVE_NOTIFY_NOW); 1301 1302 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1303 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1304 1305 if (bond_is_lb(bond)) 1306 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1307 } else { 1308 if (bond_uses_primary(bond)) 1309 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1310 } 1311 } 1312 1313 if (bond_uses_primary(bond)) 1314 bond_hw_addr_swap(bond, new_active, old_active); 1315 1316 if (bond_is_lb(bond)) { 1317 bond_alb_handle_active_change(bond, new_active); 1318 if (old_active) 1319 bond_set_slave_inactive_flags(old_active, 1320 BOND_SLAVE_NOTIFY_NOW); 1321 if (new_active) 1322 bond_set_slave_active_flags(new_active, 1323 BOND_SLAVE_NOTIFY_NOW); 1324 } else { 1325 rcu_assign_pointer(bond->curr_active_slave, new_active); 1326 } 1327 1328 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1329 if (old_active) 1330 bond_set_slave_inactive_flags(old_active, 1331 BOND_SLAVE_NOTIFY_NOW); 1332 1333 if (new_active) { 1334 bool should_notify_peers = false; 1335 1336 bond_set_slave_active_flags(new_active, 1337 BOND_SLAVE_NOTIFY_NOW); 1338 1339 if (bond->params.fail_over_mac) 1340 bond_do_fail_over_mac(bond, new_active, 1341 old_active); 1342 1343 if (netif_running(bond->dev)) { 1344 bond->send_peer_notif = 1345 bond->params.num_peer_notif * 1346 max(1, bond->params.peer_notif_delay); 1347 should_notify_peers = 1348 bond_should_notify_peers(bond); 1349 } 1350 1351 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1352 if (should_notify_peers) { 1353 bond->send_peer_notif--; 1354 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1355 bond->dev); 1356 } 1357 } 1358 } 1359 1360 #ifdef CONFIG_XFRM_OFFLOAD 1361 bond_ipsec_add_sa_all(bond); 1362 #endif /* CONFIG_XFRM_OFFLOAD */ 1363 1364 /* resend IGMP joins since active slave has changed or 1365 * all were sent on curr_active_slave. 1366 * resend only if bond is brought up with the affected 1367 * bonding modes and the retransmission is enabled 1368 */ 1369 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1370 ((bond_uses_primary(bond) && new_active) || 1371 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1372 bond->igmp_retrans = bond->params.resend_igmp; 1373 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1374 } 1375 } 1376 1377 /** 1378 * bond_select_active_slave - select a new active slave, if needed 1379 * @bond: our bonding struct 1380 * 1381 * This functions should be called when one of the following occurs: 1382 * - The old curr_active_slave has been released or lost its link. 1383 * - The primary_slave has got its link back. 1384 * - A slave has got its link back and there's no old curr_active_slave. 1385 * 1386 * Caller must hold RTNL. 1387 */ 1388 void bond_select_active_slave(struct bonding *bond) 1389 { 1390 struct slave *best_slave; 1391 int rv; 1392 1393 ASSERT_RTNL(); 1394 1395 best_slave = bond_find_best_slave(bond); 1396 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1397 bond_change_active_slave(bond, best_slave); 1398 rv = bond_set_carrier(bond); 1399 if (!rv) 1400 return; 1401 1402 if (netif_carrier_ok(bond->dev)) 1403 netdev_info(bond->dev, "active interface up!\n"); 1404 else 1405 netdev_info(bond->dev, "now running without any active interface!\n"); 1406 } 1407 } 1408 1409 #ifdef CONFIG_NET_POLL_CONTROLLER 1410 static inline int slave_enable_netpoll(struct slave *slave) 1411 { 1412 struct netpoll *np; 1413 int err = 0; 1414 1415 np = kzalloc(sizeof(*np), GFP_KERNEL); 1416 err = -ENOMEM; 1417 if (!np) 1418 goto out; 1419 1420 err = __netpoll_setup(np, slave->dev); 1421 if (err) { 1422 kfree(np); 1423 goto out; 1424 } 1425 slave->np = np; 1426 out: 1427 return err; 1428 } 1429 static inline void slave_disable_netpoll(struct slave *slave) 1430 { 1431 struct netpoll *np = slave->np; 1432 1433 if (!np) 1434 return; 1435 1436 slave->np = NULL; 1437 1438 __netpoll_free(np); 1439 } 1440 1441 static void bond_poll_controller(struct net_device *bond_dev) 1442 { 1443 struct bonding *bond = netdev_priv(bond_dev); 1444 struct slave *slave = NULL; 1445 struct list_head *iter; 1446 struct ad_info ad_info; 1447 1448 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1449 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1450 return; 1451 1452 bond_for_each_slave_rcu(bond, slave, iter) { 1453 if (!bond_slave_is_up(slave)) 1454 continue; 1455 1456 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1457 struct aggregator *agg = 1458 SLAVE_AD_INFO(slave)->port.aggregator; 1459 1460 if (agg && 1461 agg->aggregator_identifier != ad_info.aggregator_id) 1462 continue; 1463 } 1464 1465 netpoll_poll_dev(slave->dev); 1466 } 1467 } 1468 1469 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1470 { 1471 struct bonding *bond = netdev_priv(bond_dev); 1472 struct list_head *iter; 1473 struct slave *slave; 1474 1475 bond_for_each_slave(bond, slave, iter) 1476 if (bond_slave_is_up(slave)) 1477 slave_disable_netpoll(slave); 1478 } 1479 1480 static int bond_netpoll_setup(struct net_device *dev) 1481 { 1482 struct bonding *bond = netdev_priv(dev); 1483 struct list_head *iter; 1484 struct slave *slave; 1485 int err = 0; 1486 1487 bond_for_each_slave(bond, slave, iter) { 1488 err = slave_enable_netpoll(slave); 1489 if (err) { 1490 bond_netpoll_cleanup(dev); 1491 break; 1492 } 1493 } 1494 return err; 1495 } 1496 #else 1497 static inline int slave_enable_netpoll(struct slave *slave) 1498 { 1499 return 0; 1500 } 1501 static inline void slave_disable_netpoll(struct slave *slave) 1502 { 1503 } 1504 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1505 { 1506 } 1507 #endif 1508 1509 /*---------------------------------- IOCTL ----------------------------------*/ 1510 1511 static netdev_features_t bond_fix_features(struct net_device *dev, 1512 netdev_features_t features) 1513 { 1514 struct bonding *bond = netdev_priv(dev); 1515 struct list_head *iter; 1516 netdev_features_t mask; 1517 struct slave *slave; 1518 1519 mask = features; 1520 features = netdev_base_features(features); 1521 1522 bond_for_each_slave(bond, slave, iter) { 1523 features = netdev_increment_features(features, 1524 slave->dev->features, 1525 mask); 1526 } 1527 features = netdev_add_tso_features(features, mask); 1528 1529 return features; 1530 } 1531 1532 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1533 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1534 NETIF_F_GSO_ENCAP_ALL | \ 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 NETIF_F_GSO_PARTIAL) 1540 1541 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1542 NETIF_F_GSO_SOFTWARE) 1543 1544 #define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP) 1545 1546 1547 static void bond_compute_features(struct bonding *bond) 1548 { 1549 netdev_features_t gso_partial_features = BOND_GSO_PARTIAL_FEATURES; 1550 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1551 IFF_XMIT_DST_RELEASE_PERM; 1552 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1553 netdev_features_t enc_features = BOND_ENC_FEATURES; 1554 #ifdef CONFIG_XFRM_OFFLOAD 1555 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1556 #endif /* CONFIG_XFRM_OFFLOAD */ 1557 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1558 struct net_device *bond_dev = bond->dev; 1559 struct list_head *iter; 1560 struct slave *slave; 1561 unsigned short max_hard_header_len = ETH_HLEN; 1562 unsigned int tso_max_size = TSO_MAX_SIZE; 1563 u16 tso_max_segs = TSO_MAX_SEGS; 1564 1565 if (!bond_has_slaves(bond)) 1566 goto done; 1567 1568 vlan_features = netdev_base_features(vlan_features); 1569 mpls_features = netdev_base_features(mpls_features); 1570 1571 bond_for_each_slave(bond, slave, iter) { 1572 vlan_features = netdev_increment_features(vlan_features, 1573 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1574 1575 enc_features = netdev_increment_features(enc_features, 1576 slave->dev->hw_enc_features, 1577 BOND_ENC_FEATURES); 1578 1579 #ifdef CONFIG_XFRM_OFFLOAD 1580 xfrm_features = netdev_increment_features(xfrm_features, 1581 slave->dev->hw_enc_features, 1582 BOND_XFRM_FEATURES); 1583 #endif /* CONFIG_XFRM_OFFLOAD */ 1584 1585 gso_partial_features = netdev_increment_features(gso_partial_features, 1586 slave->dev->gso_partial_features, 1587 BOND_GSO_PARTIAL_FEATURES); 1588 1589 mpls_features = netdev_increment_features(mpls_features, 1590 slave->dev->mpls_features, 1591 BOND_MPLS_FEATURES); 1592 1593 dst_release_flag &= slave->dev->priv_flags; 1594 if (slave->dev->hard_header_len > max_hard_header_len) 1595 max_hard_header_len = slave->dev->hard_header_len; 1596 1597 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1598 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1599 } 1600 bond_dev->hard_header_len = max_hard_header_len; 1601 1602 done: 1603 bond_dev->gso_partial_features = gso_partial_features; 1604 bond_dev->vlan_features = vlan_features; 1605 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1606 NETIF_F_HW_VLAN_CTAG_TX | 1607 NETIF_F_HW_VLAN_STAG_TX; 1608 #ifdef CONFIG_XFRM_OFFLOAD 1609 bond_dev->hw_enc_features |= xfrm_features; 1610 #endif /* CONFIG_XFRM_OFFLOAD */ 1611 bond_dev->mpls_features = mpls_features; 1612 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1613 netif_set_tso_max_size(bond_dev, tso_max_size); 1614 1615 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1616 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1617 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1618 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1619 1620 netdev_change_features(bond_dev); 1621 } 1622 1623 static void bond_setup_by_slave(struct net_device *bond_dev, 1624 struct net_device *slave_dev) 1625 { 1626 bool was_up = !!(bond_dev->flags & IFF_UP); 1627 1628 dev_close(bond_dev); 1629 1630 bond_dev->header_ops = slave_dev->header_ops; 1631 1632 bond_dev->type = slave_dev->type; 1633 bond_dev->hard_header_len = slave_dev->hard_header_len; 1634 bond_dev->needed_headroom = slave_dev->needed_headroom; 1635 bond_dev->addr_len = slave_dev->addr_len; 1636 1637 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1638 slave_dev->addr_len); 1639 1640 if (slave_dev->flags & IFF_POINTOPOINT) { 1641 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1642 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1643 } 1644 if (was_up) 1645 dev_open(bond_dev, NULL); 1646 } 1647 1648 /* On bonding slaves other than the currently active slave, suppress 1649 * duplicates except for alb non-mcast/bcast. 1650 */ 1651 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1652 struct slave *slave, 1653 struct bonding *bond) 1654 { 1655 if (bond_is_slave_inactive(slave)) { 1656 if (BOND_MODE(bond) == BOND_MODE_ALB && 1657 skb->pkt_type != PACKET_BROADCAST && 1658 skb->pkt_type != PACKET_MULTICAST) 1659 return false; 1660 return true; 1661 } 1662 return false; 1663 } 1664 1665 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1666 { 1667 struct sk_buff *skb = *pskb; 1668 struct slave *slave; 1669 struct bonding *bond; 1670 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1671 struct slave *); 1672 int ret = RX_HANDLER_ANOTHER; 1673 1674 skb = skb_share_check(skb, GFP_ATOMIC); 1675 if (unlikely(!skb)) 1676 return RX_HANDLER_CONSUMED; 1677 1678 *pskb = skb; 1679 1680 slave = bond_slave_get_rcu(skb->dev); 1681 bond = slave->bond; 1682 1683 recv_probe = READ_ONCE(bond->recv_probe); 1684 if (recv_probe) { 1685 ret = recv_probe(skb, bond, slave); 1686 if (ret == RX_HANDLER_CONSUMED) { 1687 consume_skb(skb); 1688 return ret; 1689 } 1690 } 1691 1692 /* 1693 * For packets determined by bond_should_deliver_exact_match() call to 1694 * be suppressed we want to make an exception for link-local packets. 1695 * This is necessary for e.g. LLDP daemons to be able to monitor 1696 * inactive slave links without being forced to bind to them 1697 * explicitly. 1698 * 1699 * At the same time, packets that are passed to the bonding master 1700 * (including link-local ones) can have their originating interface 1701 * determined via PACKET_ORIGDEV socket option. 1702 */ 1703 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1704 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1705 return RX_HANDLER_PASS; 1706 return RX_HANDLER_EXACT; 1707 } 1708 1709 skb->dev = bond->dev; 1710 1711 if (BOND_MODE(bond) == BOND_MODE_ALB && 1712 netif_is_bridge_port(bond->dev) && 1713 skb->pkt_type == PACKET_HOST) { 1714 1715 if (unlikely(skb_cow_head(skb, 1716 skb->data - skb_mac_header(skb)))) { 1717 kfree_skb(skb); 1718 return RX_HANDLER_CONSUMED; 1719 } 1720 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1721 bond->dev->addr_len); 1722 } 1723 1724 return ret; 1725 } 1726 1727 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1728 { 1729 switch (BOND_MODE(bond)) { 1730 case BOND_MODE_ROUNDROBIN: 1731 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1732 case BOND_MODE_ACTIVEBACKUP: 1733 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1734 case BOND_MODE_BROADCAST: 1735 return NETDEV_LAG_TX_TYPE_BROADCAST; 1736 case BOND_MODE_XOR: 1737 case BOND_MODE_8023AD: 1738 return NETDEV_LAG_TX_TYPE_HASH; 1739 default: 1740 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1741 } 1742 } 1743 1744 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1745 enum netdev_lag_tx_type type) 1746 { 1747 if (type != NETDEV_LAG_TX_TYPE_HASH) 1748 return NETDEV_LAG_HASH_NONE; 1749 1750 switch (bond->params.xmit_policy) { 1751 case BOND_XMIT_POLICY_LAYER2: 1752 return NETDEV_LAG_HASH_L2; 1753 case BOND_XMIT_POLICY_LAYER34: 1754 return NETDEV_LAG_HASH_L34; 1755 case BOND_XMIT_POLICY_LAYER23: 1756 return NETDEV_LAG_HASH_L23; 1757 case BOND_XMIT_POLICY_ENCAP23: 1758 return NETDEV_LAG_HASH_E23; 1759 case BOND_XMIT_POLICY_ENCAP34: 1760 return NETDEV_LAG_HASH_E34; 1761 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1762 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1763 default: 1764 return NETDEV_LAG_HASH_UNKNOWN; 1765 } 1766 } 1767 1768 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1769 struct netlink_ext_ack *extack) 1770 { 1771 struct netdev_lag_upper_info lag_upper_info; 1772 enum netdev_lag_tx_type type; 1773 int err; 1774 1775 type = bond_lag_tx_type(bond); 1776 lag_upper_info.tx_type = type; 1777 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1778 1779 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1780 &lag_upper_info, extack); 1781 if (err) 1782 return err; 1783 1784 slave->dev->flags |= IFF_SLAVE; 1785 return 0; 1786 } 1787 1788 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1789 { 1790 netdev_upper_dev_unlink(slave->dev, bond->dev); 1791 slave->dev->flags &= ~IFF_SLAVE; 1792 } 1793 1794 static void slave_kobj_release(struct kobject *kobj) 1795 { 1796 struct slave *slave = to_slave(kobj); 1797 struct bonding *bond = bond_get_bond_by_slave(slave); 1798 1799 cancel_delayed_work_sync(&slave->notify_work); 1800 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1801 kfree(SLAVE_AD_INFO(slave)); 1802 1803 kfree(slave); 1804 } 1805 1806 static struct kobj_type slave_ktype = { 1807 .release = slave_kobj_release, 1808 #ifdef CONFIG_SYSFS 1809 .sysfs_ops = &slave_sysfs_ops, 1810 #endif 1811 }; 1812 1813 static int bond_kobj_init(struct slave *slave) 1814 { 1815 int err; 1816 1817 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1818 &(slave->dev->dev.kobj), "bonding_slave"); 1819 if (err) 1820 kobject_put(&slave->kobj); 1821 1822 return err; 1823 } 1824 1825 static struct slave *bond_alloc_slave(struct bonding *bond, 1826 struct net_device *slave_dev) 1827 { 1828 struct slave *slave = NULL; 1829 1830 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1831 if (!slave) 1832 return NULL; 1833 1834 slave->bond = bond; 1835 slave->dev = slave_dev; 1836 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1837 1838 if (bond_kobj_init(slave)) 1839 return NULL; 1840 1841 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1842 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1843 GFP_KERNEL); 1844 if (!SLAVE_AD_INFO(slave)) { 1845 kobject_put(&slave->kobj); 1846 return NULL; 1847 } 1848 } 1849 1850 return slave; 1851 } 1852 1853 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1854 { 1855 info->bond_mode = BOND_MODE(bond); 1856 info->miimon = bond->params.miimon; 1857 info->num_slaves = bond->slave_cnt; 1858 } 1859 1860 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1861 { 1862 strcpy(info->slave_name, slave->dev->name); 1863 info->link = slave->link; 1864 info->state = bond_slave_state(slave); 1865 info->link_failure_count = slave->link_failure_count; 1866 } 1867 1868 static void bond_netdev_notify_work(struct work_struct *_work) 1869 { 1870 struct slave *slave = container_of(_work, struct slave, 1871 notify_work.work); 1872 1873 if (rtnl_trylock()) { 1874 struct netdev_bonding_info binfo; 1875 1876 bond_fill_ifslave(slave, &binfo.slave); 1877 bond_fill_ifbond(slave->bond, &binfo.master); 1878 netdev_bonding_info_change(slave->dev, &binfo); 1879 rtnl_unlock(); 1880 } else { 1881 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1882 } 1883 } 1884 1885 void bond_queue_slave_event(struct slave *slave) 1886 { 1887 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1888 } 1889 1890 void bond_lower_state_changed(struct slave *slave) 1891 { 1892 struct netdev_lag_lower_state_info info; 1893 1894 info.link_up = slave->link == BOND_LINK_UP || 1895 slave->link == BOND_LINK_FAIL; 1896 info.tx_enabled = bond_is_active_slave(slave); 1897 netdev_lower_state_changed(slave->dev, &info); 1898 } 1899 1900 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1901 if (extack) \ 1902 NL_SET_ERR_MSG(extack, errmsg); \ 1903 else \ 1904 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1905 } while (0) 1906 1907 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1908 if (extack) \ 1909 NL_SET_ERR_MSG(extack, errmsg); \ 1910 else \ 1911 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1912 } while (0) 1913 1914 /* The bonding driver uses ether_setup() to convert a master bond device 1915 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always 1916 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP 1917 * if they were set 1918 */ 1919 static void bond_ether_setup(struct net_device *bond_dev) 1920 { 1921 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP); 1922 1923 ether_setup(bond_dev); 1924 bond_dev->flags |= IFF_MASTER | flags; 1925 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1926 } 1927 1928 void bond_xdp_set_features(struct net_device *bond_dev) 1929 { 1930 struct bonding *bond = netdev_priv(bond_dev); 1931 xdp_features_t val = NETDEV_XDP_ACT_MASK; 1932 struct list_head *iter; 1933 struct slave *slave; 1934 1935 ASSERT_RTNL(); 1936 1937 if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) { 1938 xdp_clear_features_flag(bond_dev); 1939 return; 1940 } 1941 1942 bond_for_each_slave(bond, slave, iter) 1943 val &= slave->dev->xdp_features; 1944 1945 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY; 1946 1947 xdp_set_features_flag(bond_dev, val); 1948 } 1949 1950 /* enslave device <slave> to bond device <master> */ 1951 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1952 struct netlink_ext_ack *extack) 1953 { 1954 struct bonding *bond = netdev_priv(bond_dev); 1955 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1956 struct slave *new_slave = NULL, *prev_slave; 1957 struct sockaddr_storage ss; 1958 int link_reporting; 1959 int res = 0, i; 1960 1961 if (slave_dev->flags & IFF_MASTER && 1962 !netif_is_bond_master(slave_dev)) { 1963 BOND_NL_ERR(bond_dev, extack, 1964 "Device type (master device) cannot be enslaved"); 1965 return -EPERM; 1966 } 1967 1968 if (!bond->params.use_carrier && 1969 slave_dev->ethtool_ops->get_link == NULL && 1970 slave_ops->ndo_eth_ioctl == NULL) { 1971 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1972 } 1973 1974 /* already in-use? */ 1975 if (netdev_is_rx_handler_busy(slave_dev)) { 1976 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1977 "Device is in use and cannot be enslaved"); 1978 return -EBUSY; 1979 } 1980 1981 if (bond_dev == slave_dev) { 1982 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1983 return -EPERM; 1984 } 1985 1986 /* vlan challenged mutual exclusion */ 1987 /* no need to lock since we're protected by rtnl_lock */ 1988 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1989 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1990 if (vlan_uses_dev(bond_dev)) { 1991 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1992 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1993 return -EPERM; 1994 } else { 1995 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1996 } 1997 } else { 1998 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1999 } 2000 2001 if (slave_dev->features & NETIF_F_HW_ESP) 2002 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 2003 2004 /* Old ifenslave binaries are no longer supported. These can 2005 * be identified with moderate accuracy by the state of the slave: 2006 * the current ifenslave will set the interface down prior to 2007 * enslaving it; the old ifenslave will not. 2008 */ 2009 if (slave_dev->flags & IFF_UP) { 2010 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2011 "Device can not be enslaved while up"); 2012 return -EPERM; 2013 } 2014 2015 /* set bonding device ether type by slave - bonding netdevices are 2016 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 2017 * there is a need to override some of the type dependent attribs/funcs. 2018 * 2019 * bond ether type mutual exclusion - don't allow slaves of dissimilar 2020 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 2021 */ 2022 if (!bond_has_slaves(bond)) { 2023 if (bond_dev->type != slave_dev->type) { 2024 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 2025 bond_dev->type, slave_dev->type); 2026 2027 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 2028 bond_dev); 2029 res = notifier_to_errno(res); 2030 if (res) { 2031 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 2032 return -EBUSY; 2033 } 2034 2035 /* Flush unicast and multicast addresses */ 2036 dev_uc_flush(bond_dev); 2037 dev_mc_flush(bond_dev); 2038 2039 if (slave_dev->type != ARPHRD_ETHER) 2040 bond_setup_by_slave(bond_dev, slave_dev); 2041 else 2042 bond_ether_setup(bond_dev); 2043 2044 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 2045 bond_dev); 2046 } 2047 } else if (bond_dev->type != slave_dev->type) { 2048 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2049 "Device type is different from other slaves"); 2050 return -EINVAL; 2051 } 2052 2053 if (slave_dev->type == ARPHRD_INFINIBAND && 2054 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2055 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2056 "Only active-backup mode is supported for infiniband slaves"); 2057 res = -EOPNOTSUPP; 2058 goto err_undo_flags; 2059 } 2060 2061 if (!slave_ops->ndo_set_mac_address || 2062 slave_dev->type == ARPHRD_INFINIBAND) { 2063 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 2064 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 2065 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 2066 if (!bond_has_slaves(bond)) { 2067 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 2068 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 2069 } else { 2070 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2071 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 2072 res = -EOPNOTSUPP; 2073 goto err_undo_flags; 2074 } 2075 } 2076 } 2077 2078 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 2079 2080 /* If this is the first slave, then we need to set the master's hardware 2081 * address to be the same as the slave's. 2082 */ 2083 if (!bond_has_slaves(bond) && 2084 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 2085 res = bond_set_dev_addr(bond->dev, slave_dev); 2086 if (res) 2087 goto err_undo_flags; 2088 } 2089 2090 new_slave = bond_alloc_slave(bond, slave_dev); 2091 if (!new_slave) { 2092 res = -ENOMEM; 2093 goto err_undo_flags; 2094 } 2095 2096 /* Set the new_slave's queue_id to be zero. Queue ID mapping 2097 * is set via sysfs or module option if desired. 2098 */ 2099 new_slave->queue_id = 0; 2100 2101 /* Save slave's original mtu and then set it to match the bond */ 2102 new_slave->original_mtu = slave_dev->mtu; 2103 res = dev_set_mtu(slave_dev, bond->dev->mtu); 2104 if (res) { 2105 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 2106 goto err_free; 2107 } 2108 2109 /* Save slave's original ("permanent") mac address for modes 2110 * that need it, and for restoring it upon release, and then 2111 * set it to the master's address 2112 */ 2113 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 2114 slave_dev->addr_len); 2115 2116 if (!bond->params.fail_over_mac || 2117 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2118 /* Set slave to master's mac address. The application already 2119 * set the master's mac address to that of the first slave 2120 */ 2121 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 2122 ss.ss_family = slave_dev->type; 2123 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 2124 extack); 2125 if (res) { 2126 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 2127 goto err_restore_mtu; 2128 } 2129 } 2130 2131 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 2132 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 2133 2134 /* open the slave since the application closed it */ 2135 res = dev_open(slave_dev, extack); 2136 if (res) { 2137 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 2138 goto err_restore_mac; 2139 } 2140 2141 slave_dev->priv_flags |= IFF_BONDING; 2142 /* initialize slave stats */ 2143 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 2144 2145 if (bond_is_lb(bond)) { 2146 /* bond_alb_init_slave() must be called before all other stages since 2147 * it might fail and we do not want to have to undo everything 2148 */ 2149 res = bond_alb_init_slave(bond, new_slave); 2150 if (res) 2151 goto err_close; 2152 } 2153 2154 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 2155 if (res) { 2156 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 2157 goto err_close; 2158 } 2159 2160 prev_slave = bond_last_slave(bond); 2161 2162 new_slave->delay = 0; 2163 new_slave->link_failure_count = 0; 2164 2165 if (bond_update_speed_duplex(new_slave) && 2166 bond_needs_speed_duplex(bond)) 2167 new_slave->link = BOND_LINK_DOWN; 2168 2169 new_slave->last_rx = jiffies - 2170 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2171 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2172 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2173 2174 new_slave->last_tx = new_slave->last_rx; 2175 2176 if (bond->params.miimon && !bond->params.use_carrier) { 2177 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2178 2179 if ((link_reporting == -1) && !bond->params.arp_interval) { 2180 /* miimon is set but a bonded network driver 2181 * does not support ETHTOOL/MII and 2182 * arp_interval is not set. Note: if 2183 * use_carrier is enabled, we will never go 2184 * here (because netif_carrier is always 2185 * supported); thus, we don't need to change 2186 * the messages for netif_carrier. 2187 */ 2188 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"); 2189 } else if (link_reporting == -1) { 2190 /* unable get link status using mii/ethtool */ 2191 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"); 2192 } 2193 } 2194 2195 /* check for initial state */ 2196 new_slave->link = BOND_LINK_NOCHANGE; 2197 if (bond->params.miimon) { 2198 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2199 if (bond->params.updelay) { 2200 bond_set_slave_link_state(new_slave, 2201 BOND_LINK_BACK, 2202 BOND_SLAVE_NOTIFY_NOW); 2203 new_slave->delay = bond->params.updelay; 2204 } else { 2205 bond_set_slave_link_state(new_slave, 2206 BOND_LINK_UP, 2207 BOND_SLAVE_NOTIFY_NOW); 2208 } 2209 } else { 2210 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2211 BOND_SLAVE_NOTIFY_NOW); 2212 } 2213 } else if (bond->params.arp_interval) { 2214 bond_set_slave_link_state(new_slave, 2215 (netif_carrier_ok(slave_dev) ? 2216 BOND_LINK_UP : BOND_LINK_DOWN), 2217 BOND_SLAVE_NOTIFY_NOW); 2218 } else { 2219 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2220 BOND_SLAVE_NOTIFY_NOW); 2221 } 2222 2223 if (new_slave->link != BOND_LINK_DOWN) 2224 new_slave->last_link_up = jiffies; 2225 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2226 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2227 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2228 2229 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2230 /* if there is a primary slave, remember it */ 2231 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2232 rcu_assign_pointer(bond->primary_slave, new_slave); 2233 bond->force_primary = true; 2234 } 2235 } 2236 2237 switch (BOND_MODE(bond)) { 2238 case BOND_MODE_ACTIVEBACKUP: 2239 bond_set_slave_inactive_flags(new_slave, 2240 BOND_SLAVE_NOTIFY_NOW); 2241 break; 2242 case BOND_MODE_8023AD: 2243 /* in 802.3ad mode, the internal mechanism 2244 * will activate the slaves in the selected 2245 * aggregator 2246 */ 2247 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2248 /* if this is the first slave */ 2249 if (!prev_slave) { 2250 SLAVE_AD_INFO(new_slave)->id = 1; 2251 /* Initialize AD with the number of times that the AD timer is called in 1 second 2252 * can be called only after the mac address of the bond is set 2253 */ 2254 bond_3ad_initialize(bond); 2255 } else { 2256 SLAVE_AD_INFO(new_slave)->id = 2257 SLAVE_AD_INFO(prev_slave)->id + 1; 2258 } 2259 2260 bond_3ad_bind_slave(new_slave); 2261 break; 2262 case BOND_MODE_TLB: 2263 case BOND_MODE_ALB: 2264 bond_set_active_slave(new_slave); 2265 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2266 break; 2267 default: 2268 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2269 2270 /* always active in trunk mode */ 2271 bond_set_active_slave(new_slave); 2272 2273 /* In trunking mode there is little meaning to curr_active_slave 2274 * anyway (it holds no special properties of the bond device), 2275 * so we can change it without calling change_active_interface() 2276 */ 2277 if (!rcu_access_pointer(bond->curr_active_slave) && 2278 new_slave->link == BOND_LINK_UP) 2279 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2280 2281 break; 2282 } /* switch(bond_mode) */ 2283 2284 #ifdef CONFIG_NET_POLL_CONTROLLER 2285 if (bond->dev->npinfo) { 2286 if (slave_enable_netpoll(new_slave)) { 2287 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2288 res = -EBUSY; 2289 goto err_detach; 2290 } 2291 } 2292 #endif 2293 2294 if (!(bond_dev->features & NETIF_F_LRO)) 2295 dev_disable_lro(slave_dev); 2296 2297 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2298 new_slave); 2299 if (res) { 2300 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2301 goto err_detach; 2302 } 2303 2304 res = bond_master_upper_dev_link(bond, new_slave, extack); 2305 if (res) { 2306 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2307 goto err_unregister; 2308 } 2309 2310 bond_lower_state_changed(new_slave); 2311 2312 res = bond_sysfs_slave_add(new_slave); 2313 if (res) { 2314 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2315 goto err_upper_unlink; 2316 } 2317 2318 /* If the mode uses primary, then the following is handled by 2319 * bond_change_active_slave(). 2320 */ 2321 if (!bond_uses_primary(bond)) { 2322 /* set promiscuity level to new slave */ 2323 if (bond_dev->flags & IFF_PROMISC) { 2324 res = dev_set_promiscuity(slave_dev, 1); 2325 if (res) 2326 goto err_sysfs_del; 2327 } 2328 2329 /* set allmulti level to new slave */ 2330 if (bond_dev->flags & IFF_ALLMULTI) { 2331 res = dev_set_allmulti(slave_dev, 1); 2332 if (res) { 2333 if (bond_dev->flags & IFF_PROMISC) 2334 dev_set_promiscuity(slave_dev, -1); 2335 goto err_sysfs_del; 2336 } 2337 } 2338 2339 if (bond_dev->flags & IFF_UP) { 2340 netif_addr_lock_bh(bond_dev); 2341 dev_mc_sync_multiple(slave_dev, bond_dev); 2342 dev_uc_sync_multiple(slave_dev, bond_dev); 2343 netif_addr_unlock_bh(bond_dev); 2344 2345 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2346 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2347 } 2348 } 2349 2350 bond->slave_cnt++; 2351 bond_compute_features(bond); 2352 bond_set_carrier(bond); 2353 2354 /* Needs to be called before bond_select_active_slave(), which will 2355 * remove the maddrs if the slave is selected as active slave. 2356 */ 2357 bond_slave_ns_maddrs_add(bond, new_slave); 2358 2359 if (bond_uses_primary(bond)) { 2360 block_netpoll_tx(); 2361 bond_select_active_slave(bond); 2362 unblock_netpoll_tx(); 2363 } 2364 2365 if (bond_mode_can_use_xmit_hash(bond)) 2366 bond_update_slave_arr(bond, NULL); 2367 2368 if (!slave_dev->netdev_ops->ndo_bpf || 2369 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2370 if (bond->xdp_prog) { 2371 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2372 "Slave does not support XDP"); 2373 res = -EOPNOTSUPP; 2374 goto err_sysfs_del; 2375 } 2376 } else if (bond->xdp_prog) { 2377 struct netdev_bpf xdp = { 2378 .command = XDP_SETUP_PROG, 2379 .flags = 0, 2380 .prog = bond->xdp_prog, 2381 .extack = extack, 2382 }; 2383 2384 if (dev_xdp_prog_count(slave_dev) > 0) { 2385 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2386 "Slave has XDP program loaded, please unload before enslaving"); 2387 res = -EOPNOTSUPP; 2388 goto err_sysfs_del; 2389 } 2390 2391 res = dev_xdp_propagate(slave_dev, &xdp); 2392 if (res < 0) { 2393 /* ndo_bpf() sets extack error message */ 2394 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2395 goto err_sysfs_del; 2396 } 2397 if (bond->xdp_prog) 2398 bpf_prog_inc(bond->xdp_prog); 2399 } 2400 2401 bond_xdp_set_features(bond_dev); 2402 2403 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2404 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2405 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2406 2407 /* enslave is successful */ 2408 bond_queue_slave_event(new_slave); 2409 return 0; 2410 2411 /* Undo stages on error */ 2412 err_sysfs_del: 2413 bond_sysfs_slave_del(new_slave); 2414 2415 err_upper_unlink: 2416 bond_upper_dev_unlink(bond, new_slave); 2417 2418 err_unregister: 2419 netdev_rx_handler_unregister(slave_dev); 2420 2421 err_detach: 2422 vlan_vids_del_by_dev(slave_dev, bond_dev); 2423 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2424 RCU_INIT_POINTER(bond->primary_slave, NULL); 2425 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2426 block_netpoll_tx(); 2427 bond_change_active_slave(bond, NULL); 2428 bond_select_active_slave(bond); 2429 unblock_netpoll_tx(); 2430 } 2431 /* either primary_slave or curr_active_slave might've changed */ 2432 synchronize_rcu(); 2433 slave_disable_netpoll(new_slave); 2434 2435 err_close: 2436 if (!netif_is_bond_master(slave_dev)) 2437 slave_dev->priv_flags &= ~IFF_BONDING; 2438 dev_close(slave_dev); 2439 2440 err_restore_mac: 2441 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2442 if (!bond->params.fail_over_mac || 2443 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2444 /* XXX TODO - fom follow mode needs to change master's 2445 * MAC if this slave's MAC is in use by the bond, or at 2446 * least print a warning. 2447 */ 2448 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2449 new_slave->dev->addr_len); 2450 ss.ss_family = slave_dev->type; 2451 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2452 } 2453 2454 err_restore_mtu: 2455 dev_set_mtu(slave_dev, new_slave->original_mtu); 2456 2457 err_free: 2458 kobject_put(&new_slave->kobj); 2459 2460 err_undo_flags: 2461 /* Enslave of first slave has failed and we need to fix master's mac */ 2462 if (!bond_has_slaves(bond)) { 2463 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2464 slave_dev->dev_addr)) 2465 eth_hw_addr_random(bond_dev); 2466 if (bond_dev->type != ARPHRD_ETHER) { 2467 dev_close(bond_dev); 2468 bond_ether_setup(bond_dev); 2469 } 2470 } 2471 2472 return res; 2473 } 2474 2475 /* Try to release the slave device <slave> from the bond device <master> 2476 * It is legal to access curr_active_slave without a lock because all the function 2477 * is RTNL-locked. If "all" is true it means that the function is being called 2478 * while destroying a bond interface and all slaves are being released. 2479 * 2480 * The rules for slave state should be: 2481 * for Active/Backup: 2482 * Active stays on all backups go down 2483 * for Bonded connections: 2484 * The first up interface should be left on and all others downed. 2485 */ 2486 static int __bond_release_one(struct net_device *bond_dev, 2487 struct net_device *slave_dev, 2488 bool all, bool unregister) 2489 { 2490 struct bonding *bond = netdev_priv(bond_dev); 2491 struct slave *slave, *oldcurrent; 2492 struct sockaddr_storage ss; 2493 int old_flags = bond_dev->flags; 2494 netdev_features_t old_features = bond_dev->features; 2495 2496 /* slave is not a slave or master is not master of this slave */ 2497 if (!(slave_dev->flags & IFF_SLAVE) || 2498 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2499 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2500 return -EINVAL; 2501 } 2502 2503 block_netpoll_tx(); 2504 2505 slave = bond_get_slave_by_dev(bond, slave_dev); 2506 if (!slave) { 2507 /* not a slave of this bond */ 2508 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2509 unblock_netpoll_tx(); 2510 return -EINVAL; 2511 } 2512 2513 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2514 2515 bond_sysfs_slave_del(slave); 2516 2517 /* recompute stats just before removing the slave */ 2518 bond_get_stats(bond->dev, &bond->bond_stats); 2519 2520 if (bond->xdp_prog) { 2521 struct netdev_bpf xdp = { 2522 .command = XDP_SETUP_PROG, 2523 .flags = 0, 2524 .prog = NULL, 2525 .extack = NULL, 2526 }; 2527 if (dev_xdp_propagate(slave_dev, &xdp)) 2528 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2529 } 2530 2531 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2532 * for this slave anymore. 2533 */ 2534 netdev_rx_handler_unregister(slave_dev); 2535 2536 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2537 bond_3ad_unbind_slave(slave); 2538 2539 bond_upper_dev_unlink(bond, slave); 2540 2541 if (bond_mode_can_use_xmit_hash(bond)) 2542 bond_update_slave_arr(bond, slave); 2543 2544 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2545 bond_is_active_slave(slave) ? "active" : "backup"); 2546 2547 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2548 2549 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2550 2551 if (!all && (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2552 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2553 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2554 bond_has_slaves(bond)) 2555 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", 2556 slave->perm_hwaddr); 2557 } 2558 2559 if (rtnl_dereference(bond->primary_slave) == slave) 2560 RCU_INIT_POINTER(bond->primary_slave, NULL); 2561 2562 if (oldcurrent == slave) 2563 bond_change_active_slave(bond, NULL); 2564 2565 /* Must be called after bond_change_active_slave () as the slave 2566 * might change from an active slave to a backup slave. Then it is 2567 * necessary to clear the maddrs on the backup slave. 2568 */ 2569 bond_slave_ns_maddrs_del(bond, slave); 2570 2571 if (bond_is_lb(bond)) { 2572 /* Must be called only after the slave has been 2573 * detached from the list and the curr_active_slave 2574 * has been cleared (if our_slave == old_current), 2575 * but before a new active slave is selected. 2576 */ 2577 bond_alb_deinit_slave(bond, slave); 2578 } 2579 2580 if (all) { 2581 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2582 } else if (oldcurrent == slave) { 2583 /* Note that we hold RTNL over this sequence, so there 2584 * is no concern that another slave add/remove event 2585 * will interfere. 2586 */ 2587 bond_select_active_slave(bond); 2588 } 2589 2590 bond_set_carrier(bond); 2591 if (!bond_has_slaves(bond)) 2592 eth_hw_addr_random(bond_dev); 2593 2594 unblock_netpoll_tx(); 2595 synchronize_rcu(); 2596 bond->slave_cnt--; 2597 2598 if (!bond_has_slaves(bond)) { 2599 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2600 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2601 } 2602 2603 bond_compute_features(bond); 2604 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2605 (old_features & NETIF_F_VLAN_CHALLENGED)) 2606 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2607 2608 vlan_vids_del_by_dev(slave_dev, bond_dev); 2609 2610 /* If the mode uses primary, then this case was handled above by 2611 * bond_change_active_slave(..., NULL) 2612 */ 2613 if (!bond_uses_primary(bond)) { 2614 /* unset promiscuity level from slave 2615 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2616 * of the IFF_PROMISC flag in the bond_dev, but we need the 2617 * value of that flag before that change, as that was the value 2618 * when this slave was attached, so we cache at the start of the 2619 * function and use it here. Same goes for ALLMULTI below 2620 */ 2621 if (old_flags & IFF_PROMISC) 2622 dev_set_promiscuity(slave_dev, -1); 2623 2624 /* unset allmulti level from slave */ 2625 if (old_flags & IFF_ALLMULTI) 2626 dev_set_allmulti(slave_dev, -1); 2627 2628 if (old_flags & IFF_UP) 2629 bond_hw_addr_flush(bond_dev, slave_dev); 2630 } 2631 2632 slave_disable_netpoll(slave); 2633 2634 /* close slave before restoring its mac address */ 2635 dev_close(slave_dev); 2636 2637 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2638 2639 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2640 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2641 /* restore original ("permanent") mac address */ 2642 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2643 slave->dev->addr_len); 2644 ss.ss_family = slave_dev->type; 2645 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2646 } 2647 2648 if (unregister) { 2649 netdev_lock_ops(slave_dev); 2650 __dev_set_mtu(slave_dev, slave->original_mtu); 2651 netdev_unlock_ops(slave_dev); 2652 } else { 2653 dev_set_mtu(slave_dev, slave->original_mtu); 2654 } 2655 2656 if (!netif_is_bond_master(slave_dev)) 2657 slave_dev->priv_flags &= ~IFF_BONDING; 2658 2659 bond_xdp_set_features(bond_dev); 2660 kobject_put(&slave->kobj); 2661 2662 return 0; 2663 } 2664 2665 /* A wrapper used because of ndo_del_link */ 2666 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2667 { 2668 return __bond_release_one(bond_dev, slave_dev, false, false); 2669 } 2670 2671 /* First release a slave and then destroy the bond if no more slaves are left. 2672 * Must be under rtnl_lock when this function is called. 2673 */ 2674 static int bond_release_and_destroy(struct net_device *bond_dev, 2675 struct net_device *slave_dev) 2676 { 2677 struct bonding *bond = netdev_priv(bond_dev); 2678 int ret; 2679 2680 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2681 if (ret == 0 && !bond_has_slaves(bond) && 2682 bond_dev->reg_state != NETREG_UNREGISTERING) { 2683 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2684 netdev_info(bond_dev, "Destroying bond\n"); 2685 bond_remove_proc_entry(bond); 2686 unregister_netdevice(bond_dev); 2687 } 2688 return ret; 2689 } 2690 2691 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2692 { 2693 struct bonding *bond = netdev_priv(bond_dev); 2694 2695 bond_fill_ifbond(bond, info); 2696 } 2697 2698 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2699 { 2700 struct bonding *bond = netdev_priv(bond_dev); 2701 struct list_head *iter; 2702 int i = 0, res = -ENODEV; 2703 struct slave *slave; 2704 2705 bond_for_each_slave(bond, slave, iter) { 2706 if (i++ == (int)info->slave_id) { 2707 res = 0; 2708 bond_fill_ifslave(slave, info); 2709 break; 2710 } 2711 } 2712 2713 return res; 2714 } 2715 2716 /*-------------------------------- Monitoring -------------------------------*/ 2717 2718 /* called with rcu_read_lock() */ 2719 static int bond_miimon_inspect(struct bonding *bond) 2720 { 2721 bool ignore_updelay = false; 2722 int link_state, commit = 0; 2723 struct list_head *iter; 2724 struct slave *slave; 2725 2726 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2727 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2728 } else { 2729 struct bond_up_slave *usable_slaves; 2730 2731 usable_slaves = rcu_dereference(bond->usable_slaves); 2732 2733 if (usable_slaves && usable_slaves->count == 0) 2734 ignore_updelay = true; 2735 } 2736 2737 bond_for_each_slave_rcu(bond, slave, iter) { 2738 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2739 2740 link_state = bond_check_dev_link(bond, slave->dev, 0); 2741 2742 switch (slave->link) { 2743 case BOND_LINK_UP: 2744 if (link_state) 2745 continue; 2746 2747 bond_propose_link_state(slave, BOND_LINK_FAIL); 2748 commit++; 2749 slave->delay = bond->params.downdelay; 2750 if (slave->delay && net_ratelimit()) { 2751 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2752 (BOND_MODE(bond) == 2753 BOND_MODE_ACTIVEBACKUP) ? 2754 (bond_is_active_slave(slave) ? 2755 "active " : "backup ") : "", 2756 bond->params.downdelay * bond->params.miimon); 2757 } 2758 fallthrough; 2759 case BOND_LINK_FAIL: 2760 if (link_state) { 2761 /* recovered before downdelay expired */ 2762 bond_propose_link_state(slave, BOND_LINK_UP); 2763 slave->last_link_up = jiffies; 2764 if (net_ratelimit()) 2765 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2766 (bond->params.downdelay - slave->delay) * 2767 bond->params.miimon); 2768 commit++; 2769 continue; 2770 } 2771 2772 if (slave->delay <= 0) { 2773 bond_propose_link_state(slave, BOND_LINK_DOWN); 2774 commit++; 2775 continue; 2776 } 2777 2778 slave->delay--; 2779 break; 2780 2781 case BOND_LINK_DOWN: 2782 if (!link_state) 2783 continue; 2784 2785 bond_propose_link_state(slave, BOND_LINK_BACK); 2786 commit++; 2787 slave->delay = bond->params.updelay; 2788 2789 if (slave->delay && net_ratelimit()) { 2790 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2791 ignore_updelay ? 0 : 2792 bond->params.updelay * 2793 bond->params.miimon); 2794 } 2795 fallthrough; 2796 case BOND_LINK_BACK: 2797 if (!link_state) { 2798 bond_propose_link_state(slave, BOND_LINK_DOWN); 2799 if (net_ratelimit()) 2800 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2801 (bond->params.updelay - slave->delay) * 2802 bond->params.miimon); 2803 commit++; 2804 continue; 2805 } 2806 2807 if (ignore_updelay) 2808 slave->delay = 0; 2809 2810 if (slave->delay <= 0) { 2811 bond_propose_link_state(slave, BOND_LINK_UP); 2812 commit++; 2813 ignore_updelay = false; 2814 continue; 2815 } 2816 2817 slave->delay--; 2818 break; 2819 } 2820 } 2821 2822 return commit; 2823 } 2824 2825 static void bond_miimon_link_change(struct bonding *bond, 2826 struct slave *slave, 2827 char link) 2828 { 2829 switch (BOND_MODE(bond)) { 2830 case BOND_MODE_8023AD: 2831 bond_3ad_handle_link_change(slave, link); 2832 break; 2833 case BOND_MODE_TLB: 2834 case BOND_MODE_ALB: 2835 bond_alb_handle_link_change(bond, slave, link); 2836 break; 2837 case BOND_MODE_XOR: 2838 bond_update_slave_arr(bond, NULL); 2839 break; 2840 } 2841 } 2842 2843 static void bond_miimon_commit(struct bonding *bond) 2844 { 2845 struct slave *slave, *primary, *active; 2846 bool do_failover = false; 2847 struct list_head *iter; 2848 2849 ASSERT_RTNL(); 2850 2851 bond_for_each_slave(bond, slave, iter) { 2852 switch (slave->link_new_state) { 2853 case BOND_LINK_NOCHANGE: 2854 /* For 802.3ad mode, check current slave speed and 2855 * duplex again in case its port was disabled after 2856 * invalid speed/duplex reporting but recovered before 2857 * link monitoring could make a decision on the actual 2858 * link status 2859 */ 2860 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2861 slave->link == BOND_LINK_UP) 2862 bond_3ad_adapter_speed_duplex_changed(slave); 2863 continue; 2864 2865 case BOND_LINK_UP: 2866 if (bond_update_speed_duplex(slave) && 2867 bond_needs_speed_duplex(bond)) { 2868 slave->link = BOND_LINK_DOWN; 2869 if (net_ratelimit()) 2870 slave_warn(bond->dev, slave->dev, 2871 "failed to get link speed/duplex\n"); 2872 continue; 2873 } 2874 bond_set_slave_link_state(slave, BOND_LINK_UP, 2875 BOND_SLAVE_NOTIFY_NOW); 2876 slave->last_link_up = jiffies; 2877 2878 primary = rtnl_dereference(bond->primary_slave); 2879 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2880 /* prevent it from being the active one */ 2881 bond_set_backup_slave(slave); 2882 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2883 /* make it immediately active */ 2884 bond_set_active_slave(slave); 2885 } 2886 2887 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2888 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2889 slave->duplex ? "full" : "half"); 2890 2891 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2892 2893 active = rtnl_dereference(bond->curr_active_slave); 2894 if (!active || slave == primary || slave->prio > active->prio) 2895 do_failover = true; 2896 2897 continue; 2898 2899 case BOND_LINK_DOWN: 2900 if (slave->link_failure_count < UINT_MAX) 2901 slave->link_failure_count++; 2902 2903 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2904 BOND_SLAVE_NOTIFY_NOW); 2905 2906 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2907 BOND_MODE(bond) == BOND_MODE_8023AD) 2908 bond_set_slave_inactive_flags(slave, 2909 BOND_SLAVE_NOTIFY_NOW); 2910 2911 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2912 2913 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2914 2915 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2916 do_failover = true; 2917 2918 continue; 2919 2920 default: 2921 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2922 slave->link_new_state); 2923 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2924 2925 continue; 2926 } 2927 } 2928 2929 if (do_failover) { 2930 block_netpoll_tx(); 2931 bond_select_active_slave(bond); 2932 unblock_netpoll_tx(); 2933 } 2934 2935 bond_set_carrier(bond); 2936 } 2937 2938 /* bond_mii_monitor 2939 * 2940 * Really a wrapper that splits the mii monitor into two phases: an 2941 * inspection, then (if inspection indicates something needs to be done) 2942 * an acquisition of appropriate locks followed by a commit phase to 2943 * implement whatever link state changes are indicated. 2944 */ 2945 static void bond_mii_monitor(struct work_struct *work) 2946 { 2947 struct bonding *bond = container_of(work, struct bonding, 2948 mii_work.work); 2949 bool should_notify_peers = false; 2950 bool commit; 2951 unsigned long delay; 2952 struct slave *slave; 2953 struct list_head *iter; 2954 2955 delay = msecs_to_jiffies(bond->params.miimon); 2956 2957 if (!bond_has_slaves(bond)) 2958 goto re_arm; 2959 2960 rcu_read_lock(); 2961 should_notify_peers = bond_should_notify_peers(bond); 2962 commit = !!bond_miimon_inspect(bond); 2963 if (bond->send_peer_notif) { 2964 rcu_read_unlock(); 2965 if (rtnl_trylock()) { 2966 bond->send_peer_notif--; 2967 rtnl_unlock(); 2968 } 2969 } else { 2970 rcu_read_unlock(); 2971 } 2972 2973 if (commit) { 2974 /* Race avoidance with bond_close cancel of workqueue */ 2975 if (!rtnl_trylock()) { 2976 delay = 1; 2977 should_notify_peers = false; 2978 goto re_arm; 2979 } 2980 2981 bond_for_each_slave(bond, slave, iter) { 2982 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2983 } 2984 bond_miimon_commit(bond); 2985 2986 rtnl_unlock(); /* might sleep, hold no other locks */ 2987 } 2988 2989 re_arm: 2990 if (bond->params.miimon) 2991 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2992 2993 if (should_notify_peers) { 2994 if (!rtnl_trylock()) 2995 return; 2996 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2997 rtnl_unlock(); 2998 } 2999 } 3000 3001 static int bond_upper_dev_walk(struct net_device *upper, 3002 struct netdev_nested_priv *priv) 3003 { 3004 __be32 ip = *(__be32 *)priv->data; 3005 3006 return ip == bond_confirm_addr(upper, 0, ip); 3007 } 3008 3009 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 3010 { 3011 struct netdev_nested_priv priv = { 3012 .data = (void *)&ip, 3013 }; 3014 bool ret = false; 3015 3016 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 3017 return true; 3018 3019 rcu_read_lock(); 3020 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 3021 ret = true; 3022 rcu_read_unlock(); 3023 3024 return ret; 3025 } 3026 3027 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff) 3028 3029 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 3030 struct sk_buff *skb) 3031 { 3032 struct net_device *bond_dev = slave->bond->dev; 3033 struct net_device *slave_dev = slave->dev; 3034 struct bond_vlan_tag *outer_tag = tags; 3035 3036 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE) 3037 return true; 3038 3039 tags++; 3040 3041 /* Go through all the tags backwards and add them to the packet */ 3042 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) { 3043 if (!tags->vlan_id) { 3044 tags++; 3045 continue; 3046 } 3047 3048 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 3049 ntohs(outer_tag->vlan_proto), tags->vlan_id); 3050 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 3051 tags->vlan_id); 3052 if (!skb) { 3053 net_err_ratelimited("failed to insert inner VLAN tag\n"); 3054 return false; 3055 } 3056 3057 tags++; 3058 } 3059 /* Set the outer tag */ 3060 if (outer_tag->vlan_id) { 3061 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 3062 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 3063 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 3064 outer_tag->vlan_id); 3065 } 3066 3067 return true; 3068 } 3069 3070 /* We go to the (large) trouble of VLAN tagging ARP frames because 3071 * switches in VLAN mode (especially if ports are configured as 3072 * "native" to a VLAN) might not pass non-tagged frames. 3073 */ 3074 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 3075 __be32 src_ip, struct bond_vlan_tag *tags) 3076 { 3077 struct net_device *bond_dev = slave->bond->dev; 3078 struct net_device *slave_dev = slave->dev; 3079 struct sk_buff *skb; 3080 3081 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 3082 arp_op, &dest_ip, &src_ip); 3083 3084 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 3085 NULL, slave_dev->dev_addr, NULL); 3086 3087 if (!skb) { 3088 net_err_ratelimited("ARP packet allocation failed\n"); 3089 return; 3090 } 3091 3092 if (bond_handle_vlan(slave, tags, skb)) { 3093 slave_update_last_tx(slave); 3094 arp_xmit(skb); 3095 } 3096 3097 return; 3098 } 3099 3100 /* Validate the device path between the @start_dev and the @end_dev. 3101 * The path is valid if the @end_dev is reachable through device 3102 * stacking. 3103 * When the path is validated, collect any vlan information in the 3104 * path. 3105 */ 3106 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 3107 struct net_device *end_dev, 3108 int level) 3109 { 3110 struct bond_vlan_tag *tags; 3111 struct net_device *upper; 3112 struct list_head *iter; 3113 3114 if (start_dev == end_dev) { 3115 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 3116 if (!tags) 3117 return ERR_PTR(-ENOMEM); 3118 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE; 3119 return tags; 3120 } 3121 3122 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 3123 tags = bond_verify_device_path(upper, end_dev, level + 1); 3124 if (IS_ERR_OR_NULL(tags)) { 3125 if (IS_ERR(tags)) 3126 return tags; 3127 continue; 3128 } 3129 if (is_vlan_dev(upper)) { 3130 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 3131 tags[level].vlan_id = vlan_dev_vlan_id(upper); 3132 } 3133 3134 return tags; 3135 } 3136 3137 return NULL; 3138 } 3139 3140 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 3141 { 3142 struct rtable *rt; 3143 struct bond_vlan_tag *tags; 3144 __be32 *targets = bond->params.arp_targets, addr; 3145 int i; 3146 3147 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 3148 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 3149 __func__, &targets[i]); 3150 tags = NULL; 3151 3152 /* Find out through which dev should the packet go */ 3153 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0, 3154 RT_SCOPE_LINK); 3155 if (IS_ERR(rt)) { 3156 /* there's no route to target - try to send arp 3157 * probe to generate any traffic (arp_validate=0) 3158 */ 3159 if (bond->params.arp_validate) 3160 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 3161 bond->dev->name, 3162 &targets[i]); 3163 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 3164 0, tags); 3165 continue; 3166 } 3167 3168 /* bond device itself */ 3169 if (rt->dst.dev == bond->dev) 3170 goto found; 3171 3172 rcu_read_lock(); 3173 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 3174 rcu_read_unlock(); 3175 3176 if (!IS_ERR_OR_NULL(tags)) 3177 goto found; 3178 3179 /* Not our device - skip */ 3180 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 3181 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 3182 3183 ip_rt_put(rt); 3184 continue; 3185 3186 found: 3187 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 3188 ip_rt_put(rt); 3189 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 3190 kfree(tags); 3191 } 3192 } 3193 3194 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3195 { 3196 int i; 3197 3198 if (!sip || !bond_has_this_ip(bond, tip)) { 3199 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3200 __func__, &sip, &tip); 3201 return; 3202 } 3203 3204 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3205 if (i == -1) { 3206 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3207 __func__, &sip); 3208 return; 3209 } 3210 slave->last_rx = jiffies; 3211 slave->target_last_arp_rx[i] = jiffies; 3212 } 3213 3214 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3215 struct slave *slave) 3216 { 3217 struct arphdr *arp = (struct arphdr *)skb->data; 3218 struct slave *curr_active_slave, *curr_arp_slave; 3219 unsigned char *arp_ptr; 3220 __be32 sip, tip; 3221 unsigned int alen; 3222 3223 alen = arp_hdr_len(bond->dev); 3224 3225 if (alen > skb_headlen(skb)) { 3226 arp = kmalloc(alen, GFP_ATOMIC); 3227 if (!arp) 3228 goto out_unlock; 3229 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3230 goto out_unlock; 3231 } 3232 3233 if (arp->ar_hln != bond->dev->addr_len || 3234 skb->pkt_type == PACKET_OTHERHOST || 3235 skb->pkt_type == PACKET_LOOPBACK || 3236 arp->ar_hrd != htons(ARPHRD_ETHER) || 3237 arp->ar_pro != htons(ETH_P_IP) || 3238 arp->ar_pln != 4) 3239 goto out_unlock; 3240 3241 arp_ptr = (unsigned char *)(arp + 1); 3242 arp_ptr += bond->dev->addr_len; 3243 memcpy(&sip, arp_ptr, 4); 3244 arp_ptr += 4 + bond->dev->addr_len; 3245 memcpy(&tip, arp_ptr, 4); 3246 3247 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3248 __func__, slave->dev->name, bond_slave_state(slave), 3249 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3250 &sip, &tip); 3251 3252 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3253 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3254 3255 /* We 'trust' the received ARP enough to validate it if: 3256 * 3257 * (a) the slave receiving the ARP is active (which includes the 3258 * current ARP slave, if any), or 3259 * 3260 * (b) the receiving slave isn't active, but there is a currently 3261 * active slave and it received valid arp reply(s) after it became 3262 * the currently active slave, or 3263 * 3264 * (c) there is an ARP slave that sent an ARP during the prior ARP 3265 * interval, and we receive an ARP reply on any slave. We accept 3266 * these because switch FDB update delays may deliver the ARP 3267 * reply to a slave other than the sender of the ARP request. 3268 * 3269 * Note: for (b), backup slaves are receiving the broadcast ARP 3270 * request, not a reply. This request passes from the sending 3271 * slave through the L2 switch(es) to the receiving slave. Since 3272 * this is checking the request, sip/tip are swapped for 3273 * validation. 3274 * 3275 * This is done to avoid endless looping when we can't reach the 3276 * arp_ip_target and fool ourselves with our own arp requests. 3277 */ 3278 if (bond_is_active_slave(slave)) 3279 bond_validate_arp(bond, slave, sip, tip); 3280 else if (curr_active_slave && 3281 time_after(slave_last_rx(bond, curr_active_slave), 3282 curr_active_slave->last_link_up)) 3283 bond_validate_arp(bond, slave, tip, sip); 3284 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3285 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3286 bond_validate_arp(bond, slave, sip, tip); 3287 3288 out_unlock: 3289 if (arp != (struct arphdr *)skb->data) 3290 kfree(arp); 3291 return RX_HANDLER_ANOTHER; 3292 } 3293 3294 #if IS_ENABLED(CONFIG_IPV6) 3295 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3296 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3297 { 3298 struct net_device *bond_dev = slave->bond->dev; 3299 struct net_device *slave_dev = slave->dev; 3300 struct in6_addr mcaddr; 3301 struct sk_buff *skb; 3302 3303 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3304 daddr, saddr); 3305 3306 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3307 if (!skb) { 3308 net_err_ratelimited("NS packet allocation failed\n"); 3309 return; 3310 } 3311 3312 addrconf_addr_solict_mult(daddr, &mcaddr); 3313 if (bond_handle_vlan(slave, tags, skb)) { 3314 slave_update_last_tx(slave); 3315 ndisc_send_skb(skb, &mcaddr, saddr); 3316 } 3317 } 3318 3319 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3320 { 3321 struct in6_addr *targets = bond->params.ns_targets; 3322 struct bond_vlan_tag *tags; 3323 struct dst_entry *dst; 3324 struct in6_addr saddr; 3325 struct flowi6 fl6; 3326 int i; 3327 3328 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3329 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3330 __func__, &targets[i]); 3331 tags = NULL; 3332 3333 /* Find out through which dev should the packet go */ 3334 memset(&fl6, 0, sizeof(struct flowi6)); 3335 fl6.daddr = targets[i]; 3336 fl6.flowi6_oif = bond->dev->ifindex; 3337 3338 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3339 if (dst->error) { 3340 dst_release(dst); 3341 /* there's no route to target - try to send arp 3342 * probe to generate any traffic (arp_validate=0) 3343 */ 3344 if (bond->params.arp_validate) 3345 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3346 bond->dev->name, 3347 &targets[i]); 3348 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3349 continue; 3350 } 3351 3352 /* bond device itself */ 3353 if (dst->dev == bond->dev) 3354 goto found; 3355 3356 rcu_read_lock(); 3357 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3358 rcu_read_unlock(); 3359 3360 if (!IS_ERR_OR_NULL(tags)) 3361 goto found; 3362 3363 /* Not our device - skip */ 3364 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3365 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3366 3367 dst_release(dst); 3368 continue; 3369 3370 found: 3371 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3372 bond_ns_send(slave, &targets[i], &saddr, tags); 3373 else 3374 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3375 3376 dst_release(dst); 3377 kfree(tags); 3378 } 3379 } 3380 3381 static int bond_confirm_addr6(struct net_device *dev, 3382 struct netdev_nested_priv *priv) 3383 { 3384 struct in6_addr *addr = (struct in6_addr *)priv->data; 3385 3386 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3387 } 3388 3389 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3390 { 3391 struct netdev_nested_priv priv = { 3392 .data = addr, 3393 }; 3394 int ret = false; 3395 3396 if (bond_confirm_addr6(bond->dev, &priv)) 3397 return true; 3398 3399 rcu_read_lock(); 3400 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3401 ret = true; 3402 rcu_read_unlock(); 3403 3404 return ret; 3405 } 3406 3407 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3408 struct in6_addr *saddr, struct in6_addr *daddr) 3409 { 3410 int i; 3411 3412 /* Ignore NAs that: 3413 * 1. Source address is unspecified address. 3414 * 2. Dest address is neither all-nodes multicast address nor 3415 * exist on bond interface. 3416 */ 3417 if (ipv6_addr_any(saddr) || 3418 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3419 !bond_has_this_ip6(bond, daddr))) { 3420 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3421 __func__, saddr, daddr); 3422 return; 3423 } 3424 3425 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3426 if (i == -1) { 3427 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3428 __func__, saddr); 3429 return; 3430 } 3431 slave->last_rx = jiffies; 3432 slave->target_last_arp_rx[i] = jiffies; 3433 } 3434 3435 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3436 struct slave *slave) 3437 { 3438 struct slave *curr_active_slave, *curr_arp_slave; 3439 struct in6_addr *saddr, *daddr; 3440 struct { 3441 struct ipv6hdr ip6; 3442 struct icmp6hdr icmp6; 3443 } *combined, _combined; 3444 3445 if (skb->pkt_type == PACKET_OTHERHOST || 3446 skb->pkt_type == PACKET_LOOPBACK) 3447 goto out; 3448 3449 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3450 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3451 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION && 3452 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)) 3453 goto out; 3454 3455 saddr = &combined->ip6.saddr; 3456 daddr = &combined->ip6.daddr; 3457 3458 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3459 __func__, slave->dev->name, bond_slave_state(slave), 3460 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3461 saddr, daddr); 3462 3463 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3464 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3465 3466 /* We 'trust' the received ARP enough to validate it if: 3467 * see bond_arp_rcv(). 3468 */ 3469 if (bond_is_active_slave(slave)) 3470 bond_validate_na(bond, slave, saddr, daddr); 3471 else if (curr_active_slave && 3472 time_after(slave_last_rx(bond, curr_active_slave), 3473 curr_active_slave->last_link_up)) 3474 bond_validate_na(bond, slave, daddr, saddr); 3475 else if (curr_arp_slave && 3476 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3477 bond_validate_na(bond, slave, saddr, daddr); 3478 3479 out: 3480 return RX_HANDLER_ANOTHER; 3481 } 3482 #endif 3483 3484 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3485 struct slave *slave) 3486 { 3487 #if IS_ENABLED(CONFIG_IPV6) 3488 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3489 #endif 3490 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3491 3492 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3493 __func__, skb->dev->name); 3494 3495 /* Use arp validate logic for both ARP and NS */ 3496 if (!slave_do_arp_validate(bond, slave)) { 3497 if ((slave_do_arp_validate_only(bond) && is_arp) || 3498 #if IS_ENABLED(CONFIG_IPV6) 3499 (slave_do_arp_validate_only(bond) && is_ipv6) || 3500 #endif 3501 !slave_do_arp_validate_only(bond)) 3502 slave->last_rx = jiffies; 3503 return RX_HANDLER_ANOTHER; 3504 } else if (is_arp) { 3505 return bond_arp_rcv(skb, bond, slave); 3506 #if IS_ENABLED(CONFIG_IPV6) 3507 } else if (is_ipv6) { 3508 return bond_na_rcv(skb, bond, slave); 3509 #endif 3510 } else { 3511 return RX_HANDLER_ANOTHER; 3512 } 3513 } 3514 3515 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3516 { 3517 bond_arp_send_all(bond, slave); 3518 #if IS_ENABLED(CONFIG_IPV6) 3519 bond_ns_send_all(bond, slave); 3520 #endif 3521 } 3522 3523 /* function to verify if we're in the arp_interval timeslice, returns true if 3524 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3525 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3526 */ 3527 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3528 int mod) 3529 { 3530 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3531 3532 return time_in_range(jiffies, 3533 last_act - delta_in_ticks, 3534 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3535 } 3536 3537 /* This function is called regularly to monitor each slave's link 3538 * ensuring that traffic is being sent and received when arp monitoring 3539 * is used in load-balancing mode. if the adapter has been dormant, then an 3540 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3541 * arp monitoring in active backup mode. 3542 */ 3543 static void bond_loadbalance_arp_mon(struct bonding *bond) 3544 { 3545 struct slave *slave, *oldcurrent; 3546 struct list_head *iter; 3547 int do_failover = 0, slave_state_changed = 0; 3548 3549 if (!bond_has_slaves(bond)) 3550 goto re_arm; 3551 3552 rcu_read_lock(); 3553 3554 oldcurrent = rcu_dereference(bond->curr_active_slave); 3555 /* see if any of the previous devices are up now (i.e. they have 3556 * xmt and rcv traffic). the curr_active_slave does not come into 3557 * the picture unless it is null. also, slave->last_link_up is not 3558 * needed here because we send an arp on each slave and give a slave 3559 * as long as it needs to get the tx/rx within the delta. 3560 * TODO: what about up/down delay in arp mode? it wasn't here before 3561 * so it can wait 3562 */ 3563 bond_for_each_slave_rcu(bond, slave, iter) { 3564 unsigned long last_tx = slave_last_tx(slave); 3565 3566 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3567 3568 if (slave->link != BOND_LINK_UP) { 3569 if (bond_time_in_interval(bond, last_tx, 1) && 3570 bond_time_in_interval(bond, slave->last_rx, 1)) { 3571 3572 bond_propose_link_state(slave, BOND_LINK_UP); 3573 slave_state_changed = 1; 3574 3575 /* primary_slave has no meaning in round-robin 3576 * mode. the window of a slave being up and 3577 * curr_active_slave being null after enslaving 3578 * is closed. 3579 */ 3580 if (!oldcurrent) { 3581 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3582 do_failover = 1; 3583 } else { 3584 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3585 } 3586 } 3587 } else { 3588 /* slave->link == BOND_LINK_UP */ 3589 3590 /* not all switches will respond to an arp request 3591 * when the source ip is 0, so don't take the link down 3592 * if we don't know our ip yet 3593 */ 3594 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3595 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3596 3597 bond_propose_link_state(slave, BOND_LINK_DOWN); 3598 slave_state_changed = 1; 3599 3600 if (slave->link_failure_count < UINT_MAX) 3601 slave->link_failure_count++; 3602 3603 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3604 3605 if (slave == oldcurrent) 3606 do_failover = 1; 3607 } 3608 } 3609 3610 /* note: if switch is in round-robin mode, all links 3611 * must tx arp to ensure all links rx an arp - otherwise 3612 * links may oscillate or not come up at all; if switch is 3613 * in something like xor mode, there is nothing we can 3614 * do - all replies will be rx'ed on same link causing slaves 3615 * to be unstable during low/no traffic periods 3616 */ 3617 if (bond_slave_is_up(slave)) 3618 bond_send_validate(bond, slave); 3619 } 3620 3621 rcu_read_unlock(); 3622 3623 if (do_failover || slave_state_changed) { 3624 if (!rtnl_trylock()) 3625 goto re_arm; 3626 3627 bond_for_each_slave(bond, slave, iter) { 3628 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3629 slave->link = slave->link_new_state; 3630 } 3631 3632 if (slave_state_changed) { 3633 bond_slave_state_change(bond); 3634 if (BOND_MODE(bond) == BOND_MODE_XOR) 3635 bond_update_slave_arr(bond, NULL); 3636 } 3637 if (do_failover) { 3638 block_netpoll_tx(); 3639 bond_select_active_slave(bond); 3640 unblock_netpoll_tx(); 3641 } 3642 rtnl_unlock(); 3643 } 3644 3645 re_arm: 3646 if (bond->params.arp_interval) 3647 queue_delayed_work(bond->wq, &bond->arp_work, 3648 msecs_to_jiffies(bond->params.arp_interval)); 3649 } 3650 3651 /* Called to inspect slaves for active-backup mode ARP monitor link state 3652 * changes. Sets proposed link state in slaves to specify what action 3653 * should take place for the slave. Returns 0 if no changes are found, >0 3654 * if changes to link states must be committed. 3655 * 3656 * Called with rcu_read_lock held. 3657 */ 3658 static int bond_ab_arp_inspect(struct bonding *bond) 3659 { 3660 unsigned long last_tx, last_rx; 3661 struct list_head *iter; 3662 struct slave *slave; 3663 int commit = 0; 3664 3665 bond_for_each_slave_rcu(bond, slave, iter) { 3666 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3667 last_rx = slave_last_rx(bond, slave); 3668 3669 if (slave->link != BOND_LINK_UP) { 3670 if (bond_time_in_interval(bond, last_rx, 1)) { 3671 bond_propose_link_state(slave, BOND_LINK_UP); 3672 commit++; 3673 } else if (slave->link == BOND_LINK_BACK) { 3674 bond_propose_link_state(slave, BOND_LINK_FAIL); 3675 commit++; 3676 } 3677 continue; 3678 } 3679 3680 /* Give slaves 2*delta after being enslaved or made 3681 * active. This avoids bouncing, as the last receive 3682 * times need a full ARP monitor cycle to be updated. 3683 */ 3684 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3685 continue; 3686 3687 /* Backup slave is down if: 3688 * - No current_arp_slave AND 3689 * - more than (missed_max+1)*delta since last receive AND 3690 * - the bond has an IP address 3691 * 3692 * Note: a non-null current_arp_slave indicates 3693 * the curr_active_slave went down and we are 3694 * searching for a new one; under this condition 3695 * we only take the curr_active_slave down - this 3696 * gives each slave a chance to tx/rx traffic 3697 * before being taken out 3698 */ 3699 if (!bond_is_active_slave(slave) && 3700 !rcu_access_pointer(bond->current_arp_slave) && 3701 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3702 bond_propose_link_state(slave, BOND_LINK_DOWN); 3703 commit++; 3704 } 3705 3706 /* Active slave is down if: 3707 * - more than missed_max*delta since transmitting OR 3708 * - (more than missed_max*delta since receive AND 3709 * the bond has an IP address) 3710 */ 3711 last_tx = slave_last_tx(slave); 3712 if (bond_is_active_slave(slave) && 3713 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3714 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3715 bond_propose_link_state(slave, BOND_LINK_DOWN); 3716 commit++; 3717 } 3718 } 3719 3720 return commit; 3721 } 3722 3723 /* Called to commit link state changes noted by inspection step of 3724 * active-backup mode ARP monitor. 3725 * 3726 * Called with RTNL hold. 3727 */ 3728 static void bond_ab_arp_commit(struct bonding *bond) 3729 { 3730 bool do_failover = false; 3731 struct list_head *iter; 3732 unsigned long last_tx; 3733 struct slave *slave; 3734 3735 bond_for_each_slave(bond, slave, iter) { 3736 switch (slave->link_new_state) { 3737 case BOND_LINK_NOCHANGE: 3738 continue; 3739 3740 case BOND_LINK_UP: 3741 last_tx = slave_last_tx(slave); 3742 if (rtnl_dereference(bond->curr_active_slave) != slave || 3743 (!rtnl_dereference(bond->curr_active_slave) && 3744 bond_time_in_interval(bond, last_tx, 1))) { 3745 struct slave *current_arp_slave; 3746 3747 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3748 bond_set_slave_link_state(slave, BOND_LINK_UP, 3749 BOND_SLAVE_NOTIFY_NOW); 3750 if (current_arp_slave) { 3751 bond_set_slave_inactive_flags( 3752 current_arp_slave, 3753 BOND_SLAVE_NOTIFY_NOW); 3754 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3755 } 3756 3757 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3758 3759 if (!rtnl_dereference(bond->curr_active_slave) || 3760 slave == rtnl_dereference(bond->primary_slave) || 3761 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio) 3762 do_failover = true; 3763 3764 } 3765 3766 continue; 3767 3768 case BOND_LINK_DOWN: 3769 if (slave->link_failure_count < UINT_MAX) 3770 slave->link_failure_count++; 3771 3772 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3773 BOND_SLAVE_NOTIFY_NOW); 3774 bond_set_slave_inactive_flags(slave, 3775 BOND_SLAVE_NOTIFY_NOW); 3776 3777 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3778 3779 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3780 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3781 do_failover = true; 3782 } 3783 3784 continue; 3785 3786 case BOND_LINK_FAIL: 3787 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3788 BOND_SLAVE_NOTIFY_NOW); 3789 bond_set_slave_inactive_flags(slave, 3790 BOND_SLAVE_NOTIFY_NOW); 3791 3792 /* A slave has just been enslaved and has become 3793 * the current active slave. 3794 */ 3795 if (rtnl_dereference(bond->curr_active_slave)) 3796 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3797 continue; 3798 3799 default: 3800 slave_err(bond->dev, slave->dev, 3801 "impossible: link_new_state %d on slave\n", 3802 slave->link_new_state); 3803 continue; 3804 } 3805 } 3806 3807 if (do_failover) { 3808 block_netpoll_tx(); 3809 bond_select_active_slave(bond); 3810 unblock_netpoll_tx(); 3811 } 3812 3813 bond_set_carrier(bond); 3814 } 3815 3816 /* Send ARP probes for active-backup mode ARP monitor. 3817 * 3818 * Called with rcu_read_lock held. 3819 */ 3820 static bool bond_ab_arp_probe(struct bonding *bond) 3821 { 3822 struct slave *slave, *before = NULL, *new_slave = NULL, 3823 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3824 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3825 struct list_head *iter; 3826 bool found = false; 3827 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3828 3829 if (curr_arp_slave && curr_active_slave) 3830 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3831 curr_arp_slave->dev->name, 3832 curr_active_slave->dev->name); 3833 3834 if (curr_active_slave) { 3835 bond_send_validate(bond, curr_active_slave); 3836 return should_notify_rtnl; 3837 } 3838 3839 /* if we don't have a curr_active_slave, search for the next available 3840 * backup slave from the current_arp_slave and make it the candidate 3841 * for becoming the curr_active_slave 3842 */ 3843 3844 if (!curr_arp_slave) { 3845 curr_arp_slave = bond_first_slave_rcu(bond); 3846 if (!curr_arp_slave) 3847 return should_notify_rtnl; 3848 } 3849 3850 bond_for_each_slave_rcu(bond, slave, iter) { 3851 if (!found && !before && bond_slave_is_up(slave)) 3852 before = slave; 3853 3854 if (found && !new_slave && bond_slave_is_up(slave)) 3855 new_slave = slave; 3856 /* if the link state is up at this point, we 3857 * mark it down - this can happen if we have 3858 * simultaneous link failures and 3859 * reselect_active_interface doesn't make this 3860 * one the current slave so it is still marked 3861 * up when it is actually down 3862 */ 3863 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3864 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3865 BOND_SLAVE_NOTIFY_LATER); 3866 if (slave->link_failure_count < UINT_MAX) 3867 slave->link_failure_count++; 3868 3869 bond_set_slave_inactive_flags(slave, 3870 BOND_SLAVE_NOTIFY_LATER); 3871 3872 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3873 } 3874 if (slave == curr_arp_slave) 3875 found = true; 3876 } 3877 3878 if (!new_slave && before) 3879 new_slave = before; 3880 3881 if (!new_slave) 3882 goto check_state; 3883 3884 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3885 BOND_SLAVE_NOTIFY_LATER); 3886 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3887 bond_send_validate(bond, new_slave); 3888 new_slave->last_link_up = jiffies; 3889 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3890 3891 check_state: 3892 bond_for_each_slave_rcu(bond, slave, iter) { 3893 if (slave->should_notify || slave->should_notify_link) { 3894 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3895 break; 3896 } 3897 } 3898 return should_notify_rtnl; 3899 } 3900 3901 static void bond_activebackup_arp_mon(struct bonding *bond) 3902 { 3903 bool should_notify_peers = false; 3904 bool should_notify_rtnl = false; 3905 int delta_in_ticks; 3906 3907 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3908 3909 if (!bond_has_slaves(bond)) 3910 goto re_arm; 3911 3912 rcu_read_lock(); 3913 3914 should_notify_peers = bond_should_notify_peers(bond); 3915 3916 if (bond_ab_arp_inspect(bond)) { 3917 rcu_read_unlock(); 3918 3919 /* Race avoidance with bond_close flush of workqueue */ 3920 if (!rtnl_trylock()) { 3921 delta_in_ticks = 1; 3922 should_notify_peers = false; 3923 goto re_arm; 3924 } 3925 3926 bond_ab_arp_commit(bond); 3927 3928 rtnl_unlock(); 3929 rcu_read_lock(); 3930 } 3931 3932 should_notify_rtnl = bond_ab_arp_probe(bond); 3933 rcu_read_unlock(); 3934 3935 re_arm: 3936 if (bond->params.arp_interval) 3937 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3938 3939 if (should_notify_peers || should_notify_rtnl) { 3940 if (!rtnl_trylock()) 3941 return; 3942 3943 if (should_notify_peers) { 3944 bond->send_peer_notif--; 3945 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3946 bond->dev); 3947 } 3948 if (should_notify_rtnl) { 3949 bond_slave_state_notify(bond); 3950 bond_slave_link_notify(bond); 3951 } 3952 3953 rtnl_unlock(); 3954 } 3955 } 3956 3957 static void bond_arp_monitor(struct work_struct *work) 3958 { 3959 struct bonding *bond = container_of(work, struct bonding, 3960 arp_work.work); 3961 3962 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3963 bond_activebackup_arp_mon(bond); 3964 else 3965 bond_loadbalance_arp_mon(bond); 3966 } 3967 3968 /*-------------------------- netdev event handling --------------------------*/ 3969 3970 /* Change device name */ 3971 static int bond_event_changename(struct bonding *bond) 3972 { 3973 bond_remove_proc_entry(bond); 3974 bond_create_proc_entry(bond); 3975 3976 bond_debug_reregister(bond); 3977 3978 return NOTIFY_DONE; 3979 } 3980 3981 static int bond_master_netdev_event(unsigned long event, 3982 struct net_device *bond_dev) 3983 { 3984 struct bonding *event_bond = netdev_priv(bond_dev); 3985 3986 netdev_dbg(bond_dev, "%s called\n", __func__); 3987 3988 switch (event) { 3989 case NETDEV_CHANGENAME: 3990 return bond_event_changename(event_bond); 3991 case NETDEV_UNREGISTER: 3992 bond_remove_proc_entry(event_bond); 3993 #ifdef CONFIG_XFRM_OFFLOAD 3994 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3995 #endif /* CONFIG_XFRM_OFFLOAD */ 3996 break; 3997 case NETDEV_REGISTER: 3998 bond_create_proc_entry(event_bond); 3999 break; 4000 default: 4001 break; 4002 } 4003 4004 return NOTIFY_DONE; 4005 } 4006 4007 static int bond_slave_netdev_event(unsigned long event, 4008 struct net_device *slave_dev) 4009 { 4010 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 4011 struct bonding *bond; 4012 struct net_device *bond_dev; 4013 4014 /* A netdev event can be generated while enslaving a device 4015 * before netdev_rx_handler_register is called in which case 4016 * slave will be NULL 4017 */ 4018 if (!slave) { 4019 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 4020 return NOTIFY_DONE; 4021 } 4022 4023 bond_dev = slave->bond->dev; 4024 bond = slave->bond; 4025 primary = rtnl_dereference(bond->primary_slave); 4026 4027 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 4028 4029 switch (event) { 4030 case NETDEV_UNREGISTER: 4031 if (bond_dev->type != ARPHRD_ETHER) 4032 bond_release_and_destroy(bond_dev, slave_dev); 4033 else 4034 __bond_release_one(bond_dev, slave_dev, false, true); 4035 break; 4036 case NETDEV_UP: 4037 case NETDEV_CHANGE: 4038 /* For 802.3ad mode only: 4039 * Getting invalid Speed/Duplex values here will put slave 4040 * in weird state. Mark it as link-fail if the link was 4041 * previously up or link-down if it hasn't yet come up, and 4042 * let link-monitoring (miimon) set it right when correct 4043 * speeds/duplex are available. 4044 */ 4045 if (bond_update_speed_duplex(slave) && 4046 BOND_MODE(bond) == BOND_MODE_8023AD) { 4047 if (slave->last_link_up) 4048 slave->link = BOND_LINK_FAIL; 4049 else 4050 slave->link = BOND_LINK_DOWN; 4051 } 4052 4053 if (BOND_MODE(bond) == BOND_MODE_8023AD) 4054 bond_3ad_adapter_speed_duplex_changed(slave); 4055 fallthrough; 4056 case NETDEV_DOWN: 4057 /* Refresh slave-array if applicable! 4058 * If the setup does not use miimon or arpmon (mode-specific!), 4059 * then these events will not cause the slave-array to be 4060 * refreshed. This will cause xmit to use a slave that is not 4061 * usable. Avoid such situation by refeshing the array at these 4062 * events. If these (miimon/arpmon) parameters are configured 4063 * then array gets refreshed twice and that should be fine! 4064 */ 4065 if (bond_mode_can_use_xmit_hash(bond)) 4066 bond_update_slave_arr(bond, NULL); 4067 break; 4068 case NETDEV_CHANGEMTU: 4069 /* TODO: Should slaves be allowed to 4070 * independently alter their MTU? For 4071 * an active-backup bond, slaves need 4072 * not be the same type of device, so 4073 * MTUs may vary. For other modes, 4074 * slaves arguably should have the 4075 * same MTUs. To do this, we'd need to 4076 * take over the slave's change_mtu 4077 * function for the duration of their 4078 * servitude. 4079 */ 4080 break; 4081 case NETDEV_CHANGENAME: 4082 /* we don't care if we don't have primary set */ 4083 if (!bond_uses_primary(bond) || 4084 !bond->params.primary[0]) 4085 break; 4086 4087 if (slave == primary) { 4088 /* slave's name changed - he's no longer primary */ 4089 RCU_INIT_POINTER(bond->primary_slave, NULL); 4090 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 4091 /* we have a new primary slave */ 4092 rcu_assign_pointer(bond->primary_slave, slave); 4093 } else { /* we didn't change primary - exit */ 4094 break; 4095 } 4096 4097 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 4098 primary ? slave_dev->name : "none"); 4099 4100 block_netpoll_tx(); 4101 bond_select_active_slave(bond); 4102 unblock_netpoll_tx(); 4103 break; 4104 case NETDEV_FEAT_CHANGE: 4105 if (!bond->notifier_ctx) { 4106 bond->notifier_ctx = true; 4107 bond_compute_features(bond); 4108 bond->notifier_ctx = false; 4109 } 4110 break; 4111 case NETDEV_RESEND_IGMP: 4112 /* Propagate to master device */ 4113 call_netdevice_notifiers(event, slave->bond->dev); 4114 break; 4115 case NETDEV_XDP_FEAT_CHANGE: 4116 bond_xdp_set_features(bond_dev); 4117 break; 4118 default: 4119 break; 4120 } 4121 4122 return NOTIFY_DONE; 4123 } 4124 4125 /* bond_netdev_event: handle netdev notifier chain events. 4126 * 4127 * This function receives events for the netdev chain. The caller (an 4128 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 4129 * locks for us to safely manipulate the slave devices (RTNL lock, 4130 * dev_probe_lock). 4131 */ 4132 static int bond_netdev_event(struct notifier_block *this, 4133 unsigned long event, void *ptr) 4134 { 4135 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 4136 4137 netdev_dbg(event_dev, "%s received %s\n", 4138 __func__, netdev_cmd_to_name(event)); 4139 4140 if (!(event_dev->priv_flags & IFF_BONDING)) 4141 return NOTIFY_DONE; 4142 4143 if (event_dev->flags & IFF_MASTER) { 4144 int ret; 4145 4146 ret = bond_master_netdev_event(event, event_dev); 4147 if (ret != NOTIFY_DONE) 4148 return ret; 4149 } 4150 4151 if (event_dev->flags & IFF_SLAVE) 4152 return bond_slave_netdev_event(event, event_dev); 4153 4154 return NOTIFY_DONE; 4155 } 4156 4157 static struct notifier_block bond_netdev_notifier = { 4158 .notifier_call = bond_netdev_event, 4159 }; 4160 4161 /*---------------------------- Hashing Policies -----------------------------*/ 4162 4163 /* Helper to access data in a packet, with or without a backing skb. 4164 * If skb is given the data is linearized if necessary via pskb_may_pull. 4165 */ 4166 static inline const void *bond_pull_data(struct sk_buff *skb, 4167 const void *data, int hlen, int n) 4168 { 4169 if (likely(n <= hlen)) 4170 return data; 4171 else if (skb && likely(pskb_may_pull(skb, n))) 4172 return skb->data; 4173 4174 return NULL; 4175 } 4176 4177 /* L2 hash helper */ 4178 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4179 { 4180 struct ethhdr *ep; 4181 4182 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4183 if (!data) 4184 return 0; 4185 4186 ep = (struct ethhdr *)(data + mhoff); 4187 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 4188 } 4189 4190 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 4191 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 4192 { 4193 const struct ipv6hdr *iph6; 4194 const struct iphdr *iph; 4195 4196 if (l2_proto == htons(ETH_P_IP)) { 4197 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 4198 if (!data) 4199 return false; 4200 4201 iph = (const struct iphdr *)(data + *nhoff); 4202 iph_to_flow_copy_v4addrs(fk, iph); 4203 *nhoff += iph->ihl << 2; 4204 if (!ip_is_fragment(iph)) 4205 *ip_proto = iph->protocol; 4206 } else if (l2_proto == htons(ETH_P_IPV6)) { 4207 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4208 if (!data) 4209 return false; 4210 4211 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4212 iph_to_flow_copy_v6addrs(fk, iph6); 4213 *nhoff += sizeof(*iph6); 4214 *ip_proto = iph6->nexthdr; 4215 } else { 4216 return false; 4217 } 4218 4219 if (l34 && *ip_proto >= 0) 4220 fk->ports.ports = skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4221 4222 return true; 4223 } 4224 4225 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4226 { 4227 u32 srcmac_vendor = 0, srcmac_dev = 0; 4228 struct ethhdr *mac_hdr; 4229 u16 vlan = 0; 4230 int i; 4231 4232 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4233 if (!data) 4234 return 0; 4235 mac_hdr = (struct ethhdr *)(data + mhoff); 4236 4237 for (i = 0; i < 3; i++) 4238 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4239 4240 for (i = 3; i < ETH_ALEN; i++) 4241 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4242 4243 if (skb && skb_vlan_tag_present(skb)) 4244 vlan = skb_vlan_tag_get(skb); 4245 4246 return vlan ^ srcmac_vendor ^ srcmac_dev; 4247 } 4248 4249 /* Extract the appropriate headers based on bond's xmit policy */ 4250 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4251 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4252 { 4253 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4254 int ip_proto = -1; 4255 4256 switch (bond->params.xmit_policy) { 4257 case BOND_XMIT_POLICY_ENCAP23: 4258 case BOND_XMIT_POLICY_ENCAP34: 4259 memset(fk, 0, sizeof(*fk)); 4260 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4261 fk, data, l2_proto, nhoff, hlen, 0); 4262 default: 4263 break; 4264 } 4265 4266 fk->ports.ports = 0; 4267 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4268 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4269 return false; 4270 4271 /* ICMP error packets contains at least 8 bytes of the header 4272 * of the packet which generated the error. Use this information 4273 * to correlate ICMP error packets within the same flow which 4274 * generated the error. 4275 */ 4276 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4277 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4278 if (ip_proto == IPPROTO_ICMP) { 4279 if (!icmp_is_err(fk->icmp.type)) 4280 return true; 4281 4282 nhoff += sizeof(struct icmphdr); 4283 } else if (ip_proto == IPPROTO_ICMPV6) { 4284 if (!icmpv6_is_err(fk->icmp.type)) 4285 return true; 4286 4287 nhoff += sizeof(struct icmp6hdr); 4288 } 4289 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4290 } 4291 4292 return true; 4293 } 4294 4295 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4296 { 4297 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4298 (__force u32)flow_get_u32_src(flow); 4299 hash ^= (hash >> 16); 4300 hash ^= (hash >> 8); 4301 4302 /* discard lowest hash bit to deal with the common even ports pattern */ 4303 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4304 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4305 return hash >> 1; 4306 4307 return hash; 4308 } 4309 4310 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4311 * the data as required, but this function can be used without it if the data is 4312 * known to be linear (e.g. with xdp_buff). 4313 */ 4314 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4315 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4316 { 4317 struct flow_keys flow; 4318 u32 hash; 4319 4320 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4321 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4322 4323 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4324 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4325 return bond_eth_hash(skb, data, mhoff, hlen); 4326 4327 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4328 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4329 hash = bond_eth_hash(skb, data, mhoff, hlen); 4330 } else { 4331 if (flow.icmp.id) 4332 memcpy(&hash, &flow.icmp, sizeof(hash)); 4333 else 4334 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4335 } 4336 4337 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4338 } 4339 4340 /** 4341 * bond_xmit_hash - generate a hash value based on the xmit policy 4342 * @bond: bonding device 4343 * @skb: buffer to use for headers 4344 * 4345 * This function will extract the necessary headers from the skb buffer and use 4346 * them to generate a hash based on the xmit_policy set in the bonding device 4347 */ 4348 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4349 { 4350 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4351 skb->l4_hash) 4352 return skb->hash; 4353 4354 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4355 0, skb_network_offset(skb), 4356 skb_headlen(skb)); 4357 } 4358 4359 /** 4360 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4361 * @bond: bonding device 4362 * @xdp: buffer to use for headers 4363 * 4364 * The XDP variant of bond_xmit_hash. 4365 */ 4366 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4367 { 4368 struct ethhdr *eth; 4369 4370 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4371 return 0; 4372 4373 eth = (struct ethhdr *)xdp->data; 4374 4375 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4376 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4377 } 4378 4379 /*-------------------------- Device entry points ----------------------------*/ 4380 4381 void bond_work_init_all(struct bonding *bond) 4382 { 4383 INIT_DELAYED_WORK(&bond->mcast_work, 4384 bond_resend_igmp_join_requests_delayed); 4385 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4386 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4387 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4388 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4389 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4390 } 4391 4392 static void bond_work_cancel_all(struct bonding *bond) 4393 { 4394 cancel_delayed_work_sync(&bond->mii_work); 4395 cancel_delayed_work_sync(&bond->arp_work); 4396 cancel_delayed_work_sync(&bond->alb_work); 4397 cancel_delayed_work_sync(&bond->ad_work); 4398 cancel_delayed_work_sync(&bond->mcast_work); 4399 cancel_delayed_work_sync(&bond->slave_arr_work); 4400 } 4401 4402 static int bond_open(struct net_device *bond_dev) 4403 { 4404 struct bonding *bond = netdev_priv(bond_dev); 4405 struct list_head *iter; 4406 struct slave *slave; 4407 4408 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4409 bond->rr_tx_counter = alloc_percpu(u32); 4410 if (!bond->rr_tx_counter) 4411 return -ENOMEM; 4412 } 4413 4414 /* reset slave->backup and slave->inactive */ 4415 if (bond_has_slaves(bond)) { 4416 bond_for_each_slave(bond, slave, iter) { 4417 if (bond_uses_primary(bond) && 4418 slave != rcu_access_pointer(bond->curr_active_slave)) { 4419 bond_set_slave_inactive_flags(slave, 4420 BOND_SLAVE_NOTIFY_NOW); 4421 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4422 bond_set_slave_active_flags(slave, 4423 BOND_SLAVE_NOTIFY_NOW); 4424 } 4425 } 4426 } 4427 4428 if (bond_is_lb(bond)) { 4429 /* bond_alb_initialize must be called before the timer 4430 * is started. 4431 */ 4432 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4433 return -ENOMEM; 4434 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4435 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4436 } 4437 4438 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4439 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4440 4441 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4442 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4443 bond->recv_probe = bond_rcv_validate; 4444 } 4445 4446 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4447 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4448 /* register to receive LACPDUs */ 4449 bond->recv_probe = bond_3ad_lacpdu_recv; 4450 bond_3ad_initiate_agg_selection(bond, 1); 4451 4452 bond_for_each_slave(bond, slave, iter) 4453 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4454 } 4455 4456 if (bond_mode_can_use_xmit_hash(bond)) 4457 bond_update_slave_arr(bond, NULL); 4458 4459 return 0; 4460 } 4461 4462 static int bond_close(struct net_device *bond_dev) 4463 { 4464 struct bonding *bond = netdev_priv(bond_dev); 4465 struct slave *slave; 4466 4467 bond_work_cancel_all(bond); 4468 bond->send_peer_notif = 0; 4469 if (bond_is_lb(bond)) 4470 bond_alb_deinitialize(bond); 4471 bond->recv_probe = NULL; 4472 4473 if (bond_uses_primary(bond)) { 4474 rcu_read_lock(); 4475 slave = rcu_dereference(bond->curr_active_slave); 4476 if (slave) 4477 bond_hw_addr_flush(bond_dev, slave->dev); 4478 rcu_read_unlock(); 4479 } else { 4480 struct list_head *iter; 4481 4482 bond_for_each_slave(bond, slave, iter) 4483 bond_hw_addr_flush(bond_dev, slave->dev); 4484 } 4485 4486 return 0; 4487 } 4488 4489 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4490 * that some drivers can provide 32bit values only. 4491 */ 4492 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4493 const struct rtnl_link_stats64 *_new, 4494 const struct rtnl_link_stats64 *_old) 4495 { 4496 const u64 *new = (const u64 *)_new; 4497 const u64 *old = (const u64 *)_old; 4498 u64 *res = (u64 *)_res; 4499 int i; 4500 4501 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4502 u64 nv = new[i]; 4503 u64 ov = old[i]; 4504 s64 delta = nv - ov; 4505 4506 /* detects if this particular field is 32bit only */ 4507 if (((nv | ov) >> 32) == 0) 4508 delta = (s64)(s32)((u32)nv - (u32)ov); 4509 4510 /* filter anomalies, some drivers reset their stats 4511 * at down/up events. 4512 */ 4513 if (delta > 0) 4514 res[i] += delta; 4515 } 4516 } 4517 4518 #ifdef CONFIG_LOCKDEP 4519 static int bond_get_lowest_level_rcu(struct net_device *dev) 4520 { 4521 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4522 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4523 int cur = 0, max = 0; 4524 4525 now = dev; 4526 iter = &dev->adj_list.lower; 4527 4528 while (1) { 4529 next = NULL; 4530 while (1) { 4531 ldev = netdev_next_lower_dev_rcu(now, &iter); 4532 if (!ldev) 4533 break; 4534 4535 next = ldev; 4536 niter = &ldev->adj_list.lower; 4537 dev_stack[cur] = now; 4538 iter_stack[cur++] = iter; 4539 if (max <= cur) 4540 max = cur; 4541 break; 4542 } 4543 4544 if (!next) { 4545 if (!cur) 4546 return max; 4547 next = dev_stack[--cur]; 4548 niter = iter_stack[cur]; 4549 } 4550 4551 now = next; 4552 iter = niter; 4553 } 4554 4555 return max; 4556 } 4557 #endif 4558 4559 static void bond_get_stats(struct net_device *bond_dev, 4560 struct rtnl_link_stats64 *stats) 4561 { 4562 struct bonding *bond = netdev_priv(bond_dev); 4563 struct rtnl_link_stats64 temp; 4564 struct list_head *iter; 4565 struct slave *slave; 4566 int nest_level = 0; 4567 4568 4569 rcu_read_lock(); 4570 #ifdef CONFIG_LOCKDEP 4571 nest_level = bond_get_lowest_level_rcu(bond_dev); 4572 #endif 4573 4574 spin_lock_nested(&bond->stats_lock, nest_level); 4575 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4576 4577 bond_for_each_slave_rcu(bond, slave, iter) { 4578 const struct rtnl_link_stats64 *new = 4579 dev_get_stats(slave->dev, &temp); 4580 4581 bond_fold_stats(stats, new, &slave->slave_stats); 4582 4583 /* save off the slave stats for the next run */ 4584 memcpy(&slave->slave_stats, new, sizeof(*new)); 4585 } 4586 4587 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4588 spin_unlock(&bond->stats_lock); 4589 rcu_read_unlock(); 4590 } 4591 4592 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4593 { 4594 struct bonding *bond = netdev_priv(bond_dev); 4595 struct mii_ioctl_data *mii = NULL; 4596 4597 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4598 4599 switch (cmd) { 4600 case SIOCGMIIPHY: 4601 mii = if_mii(ifr); 4602 if (!mii) 4603 return -EINVAL; 4604 4605 mii->phy_id = 0; 4606 fallthrough; 4607 case SIOCGMIIREG: 4608 /* We do this again just in case we were called by SIOCGMIIREG 4609 * instead of SIOCGMIIPHY. 4610 */ 4611 mii = if_mii(ifr); 4612 if (!mii) 4613 return -EINVAL; 4614 4615 if (mii->reg_num == 1) { 4616 mii->val_out = 0; 4617 if (netif_carrier_ok(bond->dev)) 4618 mii->val_out = BMSR_LSTATUS; 4619 } 4620 4621 break; 4622 default: 4623 return -EOPNOTSUPP; 4624 } 4625 4626 return 0; 4627 } 4628 4629 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4630 { 4631 struct bonding *bond = netdev_priv(bond_dev); 4632 struct net_device *slave_dev = NULL; 4633 struct ifbond k_binfo; 4634 struct ifbond __user *u_binfo = NULL; 4635 struct ifslave k_sinfo; 4636 struct ifslave __user *u_sinfo = NULL; 4637 struct bond_opt_value newval; 4638 struct net *net; 4639 int res = 0; 4640 4641 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4642 4643 switch (cmd) { 4644 case SIOCBONDINFOQUERY: 4645 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4646 4647 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4648 return -EFAULT; 4649 4650 bond_info_query(bond_dev, &k_binfo); 4651 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4652 return -EFAULT; 4653 4654 return 0; 4655 case SIOCBONDSLAVEINFOQUERY: 4656 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4657 4658 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4659 return -EFAULT; 4660 4661 res = bond_slave_info_query(bond_dev, &k_sinfo); 4662 if (res == 0 && 4663 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4664 return -EFAULT; 4665 4666 return res; 4667 default: 4668 break; 4669 } 4670 4671 net = dev_net(bond_dev); 4672 4673 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4674 return -EPERM; 4675 4676 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4677 4678 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4679 4680 if (!slave_dev) 4681 return -ENODEV; 4682 4683 switch (cmd) { 4684 case SIOCBONDENSLAVE: 4685 res = bond_enslave(bond_dev, slave_dev, NULL); 4686 break; 4687 case SIOCBONDRELEASE: 4688 res = bond_release(bond_dev, slave_dev); 4689 break; 4690 case SIOCBONDSETHWADDR: 4691 res = bond_set_dev_addr(bond_dev, slave_dev); 4692 break; 4693 case SIOCBONDCHANGEACTIVE: 4694 bond_opt_initstr(&newval, slave_dev->name); 4695 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4696 &newval); 4697 break; 4698 default: 4699 res = -EOPNOTSUPP; 4700 } 4701 4702 return res; 4703 } 4704 4705 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4706 void __user *data, int cmd) 4707 { 4708 struct ifreq ifrdata = { .ifr_data = data }; 4709 4710 switch (cmd) { 4711 case BOND_INFO_QUERY_OLD: 4712 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4713 case BOND_SLAVE_INFO_QUERY_OLD: 4714 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4715 case BOND_ENSLAVE_OLD: 4716 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4717 case BOND_RELEASE_OLD: 4718 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4719 case BOND_SETHWADDR_OLD: 4720 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4721 case BOND_CHANGE_ACTIVE_OLD: 4722 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4723 } 4724 4725 return -EOPNOTSUPP; 4726 } 4727 4728 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4729 { 4730 struct bonding *bond = netdev_priv(bond_dev); 4731 4732 if (change & IFF_PROMISC) 4733 bond_set_promiscuity(bond, 4734 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4735 4736 if (change & IFF_ALLMULTI) 4737 bond_set_allmulti(bond, 4738 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4739 } 4740 4741 static void bond_set_rx_mode(struct net_device *bond_dev) 4742 { 4743 struct bonding *bond = netdev_priv(bond_dev); 4744 struct list_head *iter; 4745 struct slave *slave; 4746 4747 rcu_read_lock(); 4748 if (bond_uses_primary(bond)) { 4749 slave = rcu_dereference(bond->curr_active_slave); 4750 if (slave) { 4751 dev_uc_sync(slave->dev, bond_dev); 4752 dev_mc_sync(slave->dev, bond_dev); 4753 } 4754 } else { 4755 bond_for_each_slave_rcu(bond, slave, iter) { 4756 dev_uc_sync_multiple(slave->dev, bond_dev); 4757 dev_mc_sync_multiple(slave->dev, bond_dev); 4758 } 4759 } 4760 rcu_read_unlock(); 4761 } 4762 4763 static int bond_neigh_init(struct neighbour *n) 4764 { 4765 struct bonding *bond = netdev_priv(n->dev); 4766 const struct net_device_ops *slave_ops; 4767 struct neigh_parms parms; 4768 struct slave *slave; 4769 int ret = 0; 4770 4771 rcu_read_lock(); 4772 slave = bond_first_slave_rcu(bond); 4773 if (!slave) 4774 goto out; 4775 slave_ops = slave->dev->netdev_ops; 4776 if (!slave_ops->ndo_neigh_setup) 4777 goto out; 4778 4779 /* TODO: find another way [1] to implement this. 4780 * Passing a zeroed structure is fragile, 4781 * but at least we do not pass garbage. 4782 * 4783 * [1] One way would be that ndo_neigh_setup() never touch 4784 * struct neigh_parms, but propagate the new neigh_setup() 4785 * back to ___neigh_create() / neigh_parms_alloc() 4786 */ 4787 memset(&parms, 0, sizeof(parms)); 4788 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4789 4790 if (ret) 4791 goto out; 4792 4793 if (parms.neigh_setup) 4794 ret = parms.neigh_setup(n); 4795 out: 4796 rcu_read_unlock(); 4797 return ret; 4798 } 4799 4800 /* The bonding ndo_neigh_setup is called at init time beofre any 4801 * slave exists. So we must declare proxy setup function which will 4802 * be used at run time to resolve the actual slave neigh param setup. 4803 * 4804 * It's also called by master devices (such as vlans) to setup their 4805 * underlying devices. In that case - do nothing, we're already set up from 4806 * our init. 4807 */ 4808 static int bond_neigh_setup(struct net_device *dev, 4809 struct neigh_parms *parms) 4810 { 4811 /* modify only our neigh_parms */ 4812 if (parms->dev == dev) 4813 parms->neigh_setup = bond_neigh_init; 4814 4815 return 0; 4816 } 4817 4818 /* Change the MTU of all of a master's slaves to match the master */ 4819 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4820 { 4821 struct bonding *bond = netdev_priv(bond_dev); 4822 struct slave *slave, *rollback_slave; 4823 struct list_head *iter; 4824 int res = 0; 4825 4826 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4827 4828 bond_for_each_slave(bond, slave, iter) { 4829 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4830 slave, slave->dev->netdev_ops->ndo_change_mtu); 4831 4832 res = dev_set_mtu(slave->dev, new_mtu); 4833 4834 if (res) { 4835 /* If we failed to set the slave's mtu to the new value 4836 * we must abort the operation even in ACTIVE_BACKUP 4837 * mode, because if we allow the backup slaves to have 4838 * different mtu values than the active slave we'll 4839 * need to change their mtu when doing a failover. That 4840 * means changing their mtu from timer context, which 4841 * is probably not a good idea. 4842 */ 4843 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4844 res, new_mtu); 4845 goto unwind; 4846 } 4847 } 4848 4849 WRITE_ONCE(bond_dev->mtu, new_mtu); 4850 4851 return 0; 4852 4853 unwind: 4854 /* unwind from head to the slave that failed */ 4855 bond_for_each_slave(bond, rollback_slave, iter) { 4856 int tmp_res; 4857 4858 if (rollback_slave == slave) 4859 break; 4860 4861 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4862 if (tmp_res) 4863 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4864 tmp_res); 4865 } 4866 4867 return res; 4868 } 4869 4870 /* Change HW address 4871 * 4872 * Note that many devices must be down to change the HW address, and 4873 * downing the master releases all slaves. We can make bonds full of 4874 * bonding devices to test this, however. 4875 */ 4876 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4877 { 4878 struct bonding *bond = netdev_priv(bond_dev); 4879 struct slave *slave, *rollback_slave; 4880 struct sockaddr_storage *ss = addr, tmp_ss; 4881 struct list_head *iter; 4882 int res = 0; 4883 4884 if (BOND_MODE(bond) == BOND_MODE_ALB) 4885 return bond_alb_set_mac_address(bond_dev, addr); 4886 4887 4888 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4889 4890 /* If fail_over_mac is enabled, do nothing and return success. 4891 * Returning an error causes ifenslave to fail. 4892 */ 4893 if (bond->params.fail_over_mac && 4894 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4895 return 0; 4896 4897 if (!is_valid_ether_addr(ss->__data)) 4898 return -EADDRNOTAVAIL; 4899 4900 bond_for_each_slave(bond, slave, iter) { 4901 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4902 __func__, slave); 4903 res = dev_set_mac_address(slave->dev, addr, NULL); 4904 if (res) { 4905 /* TODO: consider downing the slave 4906 * and retry ? 4907 * User should expect communications 4908 * breakage anyway until ARP finish 4909 * updating, so... 4910 */ 4911 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4912 __func__, res); 4913 goto unwind; 4914 } 4915 } 4916 4917 /* success */ 4918 dev_addr_set(bond_dev, ss->__data); 4919 return 0; 4920 4921 unwind: 4922 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4923 tmp_ss.ss_family = bond_dev->type; 4924 4925 /* unwind from head to the slave that failed */ 4926 bond_for_each_slave(bond, rollback_slave, iter) { 4927 int tmp_res; 4928 4929 if (rollback_slave == slave) 4930 break; 4931 4932 tmp_res = dev_set_mac_address(rollback_slave->dev, 4933 (struct sockaddr *)&tmp_ss, NULL); 4934 if (tmp_res) { 4935 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4936 __func__, tmp_res); 4937 } 4938 } 4939 4940 return res; 4941 } 4942 4943 /** 4944 * bond_get_slave_by_id - get xmit slave with slave_id 4945 * @bond: bonding device that is transmitting 4946 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4947 * 4948 * This function tries to get slave with slave_id but in case 4949 * it fails, it tries to find the first available slave for transmission. 4950 */ 4951 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4952 int slave_id) 4953 { 4954 struct list_head *iter; 4955 struct slave *slave; 4956 int i = slave_id; 4957 4958 /* Here we start from the slave with slave_id */ 4959 bond_for_each_slave_rcu(bond, slave, iter) { 4960 if (--i < 0) { 4961 if (bond_slave_can_tx(slave)) 4962 return slave; 4963 } 4964 } 4965 4966 /* Here we start from the first slave up to slave_id */ 4967 i = slave_id; 4968 bond_for_each_slave_rcu(bond, slave, iter) { 4969 if (--i < 0) 4970 break; 4971 if (bond_slave_can_tx(slave)) 4972 return slave; 4973 } 4974 /* no slave that can tx has been found */ 4975 return NULL; 4976 } 4977 4978 /** 4979 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4980 * @bond: bonding device to use 4981 * 4982 * Based on the value of the bonding device's packets_per_slave parameter 4983 * this function generates a slave id, which is usually used as the next 4984 * slave to transmit through. 4985 */ 4986 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4987 { 4988 u32 slave_id; 4989 struct reciprocal_value reciprocal_packets_per_slave; 4990 int packets_per_slave = bond->params.packets_per_slave; 4991 4992 switch (packets_per_slave) { 4993 case 0: 4994 slave_id = get_random_u32(); 4995 break; 4996 case 1: 4997 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4998 break; 4999 default: 5000 reciprocal_packets_per_slave = 5001 bond->params.reciprocal_packets_per_slave; 5002 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 5003 slave_id = reciprocal_divide(slave_id, 5004 reciprocal_packets_per_slave); 5005 break; 5006 } 5007 5008 return slave_id; 5009 } 5010 5011 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 5012 struct sk_buff *skb) 5013 { 5014 struct slave *slave; 5015 int slave_cnt; 5016 u32 slave_id; 5017 5018 /* Start with the curr_active_slave that joined the bond as the 5019 * default for sending IGMP traffic. For failover purposes one 5020 * needs to maintain some consistency for the interface that will 5021 * send the join/membership reports. The curr_active_slave found 5022 * will send all of this type of traffic. 5023 */ 5024 if (skb->protocol == htons(ETH_P_IP)) { 5025 int noff = skb_network_offset(skb); 5026 struct iphdr *iph; 5027 5028 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 5029 goto non_igmp; 5030 5031 iph = ip_hdr(skb); 5032 if (iph->protocol == IPPROTO_IGMP) { 5033 slave = rcu_dereference(bond->curr_active_slave); 5034 if (slave) 5035 return slave; 5036 return bond_get_slave_by_id(bond, 0); 5037 } 5038 } 5039 5040 non_igmp: 5041 slave_cnt = READ_ONCE(bond->slave_cnt); 5042 if (likely(slave_cnt)) { 5043 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 5044 return bond_get_slave_by_id(bond, slave_id); 5045 } 5046 return NULL; 5047 } 5048 5049 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 5050 struct xdp_buff *xdp) 5051 { 5052 struct slave *slave; 5053 int slave_cnt; 5054 u32 slave_id; 5055 const struct ethhdr *eth; 5056 void *data = xdp->data; 5057 5058 if (data + sizeof(struct ethhdr) > xdp->data_end) 5059 goto non_igmp; 5060 5061 eth = (struct ethhdr *)data; 5062 data += sizeof(struct ethhdr); 5063 5064 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 5065 if (eth->h_proto == htons(ETH_P_IP)) { 5066 const struct iphdr *iph; 5067 5068 if (data + sizeof(struct iphdr) > xdp->data_end) 5069 goto non_igmp; 5070 5071 iph = (struct iphdr *)data; 5072 5073 if (iph->protocol == IPPROTO_IGMP) { 5074 slave = rcu_dereference(bond->curr_active_slave); 5075 if (slave) 5076 return slave; 5077 return bond_get_slave_by_id(bond, 0); 5078 } 5079 } 5080 5081 non_igmp: 5082 slave_cnt = READ_ONCE(bond->slave_cnt); 5083 if (likely(slave_cnt)) { 5084 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 5085 return bond_get_slave_by_id(bond, slave_id); 5086 } 5087 return NULL; 5088 } 5089 5090 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 5091 struct net_device *bond_dev) 5092 { 5093 struct bonding *bond = netdev_priv(bond_dev); 5094 struct slave *slave; 5095 5096 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5097 if (likely(slave)) 5098 return bond_dev_queue_xmit(bond, skb, slave->dev); 5099 5100 return bond_tx_drop(bond_dev, skb); 5101 } 5102 5103 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 5104 { 5105 return rcu_dereference(bond->curr_active_slave); 5106 } 5107 5108 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 5109 * the bond has a usable interface. 5110 */ 5111 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 5112 struct net_device *bond_dev) 5113 { 5114 struct bonding *bond = netdev_priv(bond_dev); 5115 struct slave *slave; 5116 5117 slave = bond_xmit_activebackup_slave_get(bond); 5118 if (slave) 5119 return bond_dev_queue_xmit(bond, skb, slave->dev); 5120 5121 return bond_tx_drop(bond_dev, skb); 5122 } 5123 5124 /* Use this to update slave_array when (a) it's not appropriate to update 5125 * slave_array right away (note that update_slave_array() may sleep) 5126 * and / or (b) RTNL is not held. 5127 */ 5128 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 5129 { 5130 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 5131 } 5132 5133 /* Slave array work handler. Holds only RTNL */ 5134 static void bond_slave_arr_handler(struct work_struct *work) 5135 { 5136 struct bonding *bond = container_of(work, struct bonding, 5137 slave_arr_work.work); 5138 int ret; 5139 5140 if (!rtnl_trylock()) 5141 goto err; 5142 5143 ret = bond_update_slave_arr(bond, NULL); 5144 rtnl_unlock(); 5145 if (ret) { 5146 pr_warn_ratelimited("Failed to update slave array from WT\n"); 5147 goto err; 5148 } 5149 return; 5150 5151 err: 5152 bond_slave_arr_work_rearm(bond, 1); 5153 } 5154 5155 static void bond_skip_slave(struct bond_up_slave *slaves, 5156 struct slave *skipslave) 5157 { 5158 int idx; 5159 5160 /* Rare situation where caller has asked to skip a specific 5161 * slave but allocation failed (most likely!). BTW this is 5162 * only possible when the call is initiated from 5163 * __bond_release_one(). In this situation; overwrite the 5164 * skipslave entry in the array with the last entry from the 5165 * array to avoid a situation where the xmit path may choose 5166 * this to-be-skipped slave to send a packet out. 5167 */ 5168 for (idx = 0; slaves && idx < slaves->count; idx++) { 5169 if (skipslave == slaves->arr[idx]) { 5170 slaves->arr[idx] = 5171 slaves->arr[slaves->count - 1]; 5172 slaves->count--; 5173 break; 5174 } 5175 } 5176 } 5177 5178 static void bond_set_slave_arr(struct bonding *bond, 5179 struct bond_up_slave *usable_slaves, 5180 struct bond_up_slave *all_slaves) 5181 { 5182 struct bond_up_slave *usable, *all; 5183 5184 usable = rtnl_dereference(bond->usable_slaves); 5185 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5186 kfree_rcu(usable, rcu); 5187 5188 all = rtnl_dereference(bond->all_slaves); 5189 rcu_assign_pointer(bond->all_slaves, all_slaves); 5190 kfree_rcu(all, rcu); 5191 } 5192 5193 static void bond_reset_slave_arr(struct bonding *bond) 5194 { 5195 bond_set_slave_arr(bond, NULL, NULL); 5196 } 5197 5198 /* Build the usable slaves array in control path for modes that use xmit-hash 5199 * to determine the slave interface - 5200 * (a) BOND_MODE_8023AD 5201 * (b) BOND_MODE_XOR 5202 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5203 * 5204 * The caller is expected to hold RTNL only and NO other lock! 5205 */ 5206 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5207 { 5208 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5209 struct slave *slave; 5210 struct list_head *iter; 5211 int agg_id = 0; 5212 int ret = 0; 5213 5214 might_sleep(); 5215 5216 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5217 bond->slave_cnt), GFP_KERNEL); 5218 all_slaves = kzalloc(struct_size(all_slaves, arr, 5219 bond->slave_cnt), GFP_KERNEL); 5220 if (!usable_slaves || !all_slaves) { 5221 ret = -ENOMEM; 5222 goto out; 5223 } 5224 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5225 struct ad_info ad_info; 5226 5227 spin_lock_bh(&bond->mode_lock); 5228 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5229 spin_unlock_bh(&bond->mode_lock); 5230 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5231 /* No active aggragator means it's not safe to use 5232 * the previous array. 5233 */ 5234 bond_reset_slave_arr(bond); 5235 goto out; 5236 } 5237 spin_unlock_bh(&bond->mode_lock); 5238 agg_id = ad_info.aggregator_id; 5239 } 5240 bond_for_each_slave(bond, slave, iter) { 5241 if (skipslave == slave) 5242 continue; 5243 5244 all_slaves->arr[all_slaves->count++] = slave; 5245 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5246 struct aggregator *agg; 5247 5248 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5249 if (!agg || agg->aggregator_identifier != agg_id) 5250 continue; 5251 } 5252 if (!bond_slave_can_tx(slave)) 5253 continue; 5254 5255 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5256 usable_slaves->count); 5257 5258 usable_slaves->arr[usable_slaves->count++] = slave; 5259 } 5260 5261 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5262 return ret; 5263 out: 5264 if (ret != 0 && skipslave) { 5265 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5266 skipslave); 5267 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5268 skipslave); 5269 } 5270 kfree_rcu(all_slaves, rcu); 5271 kfree_rcu(usable_slaves, rcu); 5272 5273 return ret; 5274 } 5275 5276 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5277 struct sk_buff *skb, 5278 struct bond_up_slave *slaves) 5279 { 5280 struct slave *slave; 5281 unsigned int count; 5282 u32 hash; 5283 5284 hash = bond_xmit_hash(bond, skb); 5285 count = slaves ? READ_ONCE(slaves->count) : 0; 5286 if (unlikely(!count)) 5287 return NULL; 5288 5289 slave = slaves->arr[hash % count]; 5290 return slave; 5291 } 5292 5293 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5294 struct xdp_buff *xdp) 5295 { 5296 struct bond_up_slave *slaves; 5297 unsigned int count; 5298 u32 hash; 5299 5300 hash = bond_xmit_hash_xdp(bond, xdp); 5301 slaves = rcu_dereference(bond->usable_slaves); 5302 count = slaves ? READ_ONCE(slaves->count) : 0; 5303 if (unlikely(!count)) 5304 return NULL; 5305 5306 return slaves->arr[hash % count]; 5307 } 5308 5309 /* Use this Xmit function for 3AD as well as XOR modes. The current 5310 * usable slave array is formed in the control path. The xmit function 5311 * just calculates hash and sends the packet out. 5312 */ 5313 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5314 struct net_device *dev) 5315 { 5316 struct bonding *bond = netdev_priv(dev); 5317 struct bond_up_slave *slaves; 5318 struct slave *slave; 5319 5320 slaves = rcu_dereference(bond->usable_slaves); 5321 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5322 if (likely(slave)) 5323 return bond_dev_queue_xmit(bond, skb, slave->dev); 5324 5325 return bond_tx_drop(dev, skb); 5326 } 5327 5328 /* in broadcast mode, we send everything to all usable interfaces. */ 5329 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5330 struct net_device *bond_dev) 5331 { 5332 struct bonding *bond = netdev_priv(bond_dev); 5333 struct slave *slave = NULL; 5334 struct list_head *iter; 5335 bool xmit_suc = false; 5336 bool skb_used = false; 5337 5338 bond_for_each_slave_rcu(bond, slave, iter) { 5339 struct sk_buff *skb2; 5340 5341 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5342 continue; 5343 5344 if (bond_is_last_slave(bond, slave)) { 5345 skb2 = skb; 5346 skb_used = true; 5347 } else { 5348 skb2 = skb_clone(skb, GFP_ATOMIC); 5349 if (!skb2) { 5350 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5351 bond_dev->name, __func__); 5352 continue; 5353 } 5354 } 5355 5356 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5357 xmit_suc = true; 5358 } 5359 5360 if (!skb_used) 5361 dev_kfree_skb_any(skb); 5362 5363 if (xmit_suc) 5364 return NETDEV_TX_OK; 5365 5366 dev_core_stats_tx_dropped_inc(bond_dev); 5367 return NET_XMIT_DROP; 5368 } 5369 5370 /*------------------------- Device initialization ---------------------------*/ 5371 5372 /* Lookup the slave that corresponds to a qid */ 5373 static inline int bond_slave_override(struct bonding *bond, 5374 struct sk_buff *skb) 5375 { 5376 struct slave *slave = NULL; 5377 struct list_head *iter; 5378 5379 if (!skb_rx_queue_recorded(skb)) 5380 return 1; 5381 5382 /* Find out if any slaves have the same mapping as this skb. */ 5383 bond_for_each_slave_rcu(bond, slave, iter) { 5384 if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) { 5385 if (bond_slave_is_up(slave) && 5386 slave->link == BOND_LINK_UP) { 5387 bond_dev_queue_xmit(bond, skb, slave->dev); 5388 return 0; 5389 } 5390 /* If the slave isn't UP, use default transmit policy. */ 5391 break; 5392 } 5393 } 5394 5395 return 1; 5396 } 5397 5398 5399 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5400 struct net_device *sb_dev) 5401 { 5402 /* This helper function exists to help dev_pick_tx get the correct 5403 * destination queue. Using a helper function skips a call to 5404 * skb_tx_hash and will put the skbs in the queue we expect on their 5405 * way down to the bonding driver. 5406 */ 5407 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5408 5409 /* Save the original txq to restore before passing to the driver */ 5410 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5411 5412 if (unlikely(txq >= dev->real_num_tx_queues)) { 5413 do { 5414 txq -= dev->real_num_tx_queues; 5415 } while (txq >= dev->real_num_tx_queues); 5416 } 5417 return txq; 5418 } 5419 5420 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5421 struct sk_buff *skb, 5422 bool all_slaves) 5423 { 5424 struct bonding *bond = netdev_priv(master_dev); 5425 struct bond_up_slave *slaves; 5426 struct slave *slave = NULL; 5427 5428 switch (BOND_MODE(bond)) { 5429 case BOND_MODE_ROUNDROBIN: 5430 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5431 break; 5432 case BOND_MODE_ACTIVEBACKUP: 5433 slave = bond_xmit_activebackup_slave_get(bond); 5434 break; 5435 case BOND_MODE_8023AD: 5436 case BOND_MODE_XOR: 5437 if (all_slaves) 5438 slaves = rcu_dereference(bond->all_slaves); 5439 else 5440 slaves = rcu_dereference(bond->usable_slaves); 5441 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5442 break; 5443 case BOND_MODE_BROADCAST: 5444 break; 5445 case BOND_MODE_ALB: 5446 slave = bond_xmit_alb_slave_get(bond, skb); 5447 break; 5448 case BOND_MODE_TLB: 5449 slave = bond_xmit_tlb_slave_get(bond, skb); 5450 break; 5451 default: 5452 /* Should never happen, mode already checked */ 5453 WARN_ONCE(true, "Unknown bonding mode"); 5454 break; 5455 } 5456 5457 if (slave) 5458 return slave->dev; 5459 return NULL; 5460 } 5461 5462 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5463 { 5464 switch (sk->sk_family) { 5465 #if IS_ENABLED(CONFIG_IPV6) 5466 case AF_INET6: 5467 if (ipv6_only_sock(sk) || 5468 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5469 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5470 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5471 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5472 break; 5473 } 5474 fallthrough; 5475 #endif 5476 default: /* AF_INET */ 5477 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5478 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5479 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5480 break; 5481 } 5482 5483 flow->ports.src = inet_sk(sk)->inet_sport; 5484 flow->ports.dst = inet_sk(sk)->inet_dport; 5485 } 5486 5487 /** 5488 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5489 * @sk: socket to use for headers 5490 * 5491 * This function will extract the necessary field from the socket and use 5492 * them to generate a hash based on the LAYER34 xmit_policy. 5493 * Assumes that sk is a TCP or UDP socket. 5494 */ 5495 static u32 bond_sk_hash_l34(struct sock *sk) 5496 { 5497 struct flow_keys flow; 5498 u32 hash; 5499 5500 bond_sk_to_flow(sk, &flow); 5501 5502 /* L4 */ 5503 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5504 /* L3 */ 5505 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5506 } 5507 5508 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5509 struct sock *sk) 5510 { 5511 struct bond_up_slave *slaves; 5512 struct slave *slave; 5513 unsigned int count; 5514 u32 hash; 5515 5516 slaves = rcu_dereference(bond->usable_slaves); 5517 count = slaves ? READ_ONCE(slaves->count) : 0; 5518 if (unlikely(!count)) 5519 return NULL; 5520 5521 hash = bond_sk_hash_l34(sk); 5522 slave = slaves->arr[hash % count]; 5523 5524 return slave->dev; 5525 } 5526 5527 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5528 struct sock *sk) 5529 { 5530 struct bonding *bond = netdev_priv(dev); 5531 struct net_device *lower = NULL; 5532 5533 rcu_read_lock(); 5534 if (bond_sk_check(bond)) 5535 lower = __bond_sk_get_lower_dev(bond, sk); 5536 rcu_read_unlock(); 5537 5538 return lower; 5539 } 5540 5541 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5542 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5543 struct net_device *dev) 5544 { 5545 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5546 5547 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded 5548 * was true, if tls_device_down is running in parallel, but it's OK, 5549 * because bond_get_slave_by_dev has a NULL check. 5550 */ 5551 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5552 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5553 return bond_tx_drop(dev, skb); 5554 } 5555 #endif 5556 5557 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5558 { 5559 struct bonding *bond = netdev_priv(dev); 5560 5561 if (bond_should_override_tx_queue(bond) && 5562 !bond_slave_override(bond, skb)) 5563 return NETDEV_TX_OK; 5564 5565 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5566 if (tls_is_skb_tx_device_offloaded(skb)) 5567 return bond_tls_device_xmit(bond, skb, dev); 5568 #endif 5569 5570 switch (BOND_MODE(bond)) { 5571 case BOND_MODE_ROUNDROBIN: 5572 return bond_xmit_roundrobin(skb, dev); 5573 case BOND_MODE_ACTIVEBACKUP: 5574 return bond_xmit_activebackup(skb, dev); 5575 case BOND_MODE_8023AD: 5576 case BOND_MODE_XOR: 5577 return bond_3ad_xor_xmit(skb, dev); 5578 case BOND_MODE_BROADCAST: 5579 return bond_xmit_broadcast(skb, dev); 5580 case BOND_MODE_ALB: 5581 return bond_alb_xmit(skb, dev); 5582 case BOND_MODE_TLB: 5583 return bond_tlb_xmit(skb, dev); 5584 default: 5585 /* Should never happen, mode already checked */ 5586 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5587 WARN_ON_ONCE(1); 5588 return bond_tx_drop(dev, skb); 5589 } 5590 } 5591 5592 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5593 { 5594 struct bonding *bond = netdev_priv(dev); 5595 netdev_tx_t ret = NETDEV_TX_OK; 5596 5597 /* If we risk deadlock from transmitting this in the 5598 * netpoll path, tell netpoll to queue the frame for later tx 5599 */ 5600 if (unlikely(is_netpoll_tx_blocked(dev))) 5601 return NETDEV_TX_BUSY; 5602 5603 rcu_read_lock(); 5604 if (bond_has_slaves(bond)) 5605 ret = __bond_start_xmit(skb, dev); 5606 else 5607 ret = bond_tx_drop(dev, skb); 5608 rcu_read_unlock(); 5609 5610 return ret; 5611 } 5612 5613 static struct net_device * 5614 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5615 { 5616 struct bonding *bond = netdev_priv(bond_dev); 5617 struct slave *slave; 5618 5619 /* Caller needs to hold rcu_read_lock() */ 5620 5621 switch (BOND_MODE(bond)) { 5622 case BOND_MODE_ROUNDROBIN: 5623 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5624 break; 5625 5626 case BOND_MODE_ACTIVEBACKUP: 5627 slave = bond_xmit_activebackup_slave_get(bond); 5628 break; 5629 5630 case BOND_MODE_8023AD: 5631 case BOND_MODE_XOR: 5632 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5633 break; 5634 5635 default: 5636 if (net_ratelimit()) 5637 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", 5638 BOND_MODE(bond)); 5639 return NULL; 5640 } 5641 5642 if (slave) 5643 return slave->dev; 5644 5645 return NULL; 5646 } 5647 5648 static int bond_xdp_xmit(struct net_device *bond_dev, 5649 int n, struct xdp_frame **frames, u32 flags) 5650 { 5651 int nxmit, err = -ENXIO; 5652 5653 rcu_read_lock(); 5654 5655 for (nxmit = 0; nxmit < n; nxmit++) { 5656 struct xdp_frame *frame = frames[nxmit]; 5657 struct xdp_frame *frames1[] = {frame}; 5658 struct net_device *slave_dev; 5659 struct xdp_buff xdp; 5660 5661 xdp_convert_frame_to_buff(frame, &xdp); 5662 5663 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5664 if (!slave_dev) { 5665 err = -ENXIO; 5666 break; 5667 } 5668 5669 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5670 if (err < 1) 5671 break; 5672 } 5673 5674 rcu_read_unlock(); 5675 5676 /* If error happened on the first frame then we can pass the error up, otherwise 5677 * report the number of frames that were xmitted. 5678 */ 5679 if (err < 0) 5680 return (nxmit == 0 ? err : nxmit); 5681 5682 return nxmit; 5683 } 5684 5685 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5686 struct netlink_ext_ack *extack) 5687 { 5688 struct bonding *bond = netdev_priv(dev); 5689 struct list_head *iter; 5690 struct slave *slave, *rollback_slave; 5691 struct bpf_prog *old_prog; 5692 struct netdev_bpf xdp = { 5693 .command = XDP_SETUP_PROG, 5694 .flags = 0, 5695 .prog = prog, 5696 .extack = extack, 5697 }; 5698 int err; 5699 5700 ASSERT_RTNL(); 5701 5702 if (!bond_xdp_check(bond)) { 5703 BOND_NL_ERR(dev, extack, 5704 "No native XDP support for the current bonding mode"); 5705 return -EOPNOTSUPP; 5706 } 5707 5708 old_prog = bond->xdp_prog; 5709 bond->xdp_prog = prog; 5710 5711 bond_for_each_slave(bond, slave, iter) { 5712 struct net_device *slave_dev = slave->dev; 5713 5714 if (!slave_dev->netdev_ops->ndo_bpf || 5715 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5716 SLAVE_NL_ERR(dev, slave_dev, extack, 5717 "Slave device does not support XDP"); 5718 err = -EOPNOTSUPP; 5719 goto err; 5720 } 5721 5722 if (dev_xdp_prog_count(slave_dev) > 0) { 5723 SLAVE_NL_ERR(dev, slave_dev, extack, 5724 "Slave has XDP program loaded, please unload before enslaving"); 5725 err = -EOPNOTSUPP; 5726 goto err; 5727 } 5728 5729 err = dev_xdp_propagate(slave_dev, &xdp); 5730 if (err < 0) { 5731 /* ndo_bpf() sets extack error message */ 5732 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5733 goto err; 5734 } 5735 if (prog) 5736 bpf_prog_inc(prog); 5737 } 5738 5739 if (prog) { 5740 static_branch_inc(&bpf_master_redirect_enabled_key); 5741 } else if (old_prog) { 5742 bpf_prog_put(old_prog); 5743 static_branch_dec(&bpf_master_redirect_enabled_key); 5744 } 5745 5746 return 0; 5747 5748 err: 5749 /* unwind the program changes */ 5750 bond->xdp_prog = old_prog; 5751 xdp.prog = old_prog; 5752 xdp.extack = NULL; /* do not overwrite original error */ 5753 5754 bond_for_each_slave(bond, rollback_slave, iter) { 5755 struct net_device *slave_dev = rollback_slave->dev; 5756 int err_unwind; 5757 5758 if (slave == rollback_slave) 5759 break; 5760 5761 err_unwind = dev_xdp_propagate(slave_dev, &xdp); 5762 if (err_unwind < 0) 5763 slave_err(dev, slave_dev, 5764 "Error %d when unwinding XDP program change\n", err_unwind); 5765 else if (xdp.prog) 5766 bpf_prog_inc(xdp.prog); 5767 } 5768 return err; 5769 } 5770 5771 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5772 { 5773 switch (xdp->command) { 5774 case XDP_SETUP_PROG: 5775 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5776 default: 5777 return -EINVAL; 5778 } 5779 } 5780 5781 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5782 { 5783 if (speed == 0 || speed == SPEED_UNKNOWN) 5784 speed = slave->speed; 5785 else 5786 speed = min(speed, slave->speed); 5787 5788 return speed; 5789 } 5790 5791 /* Set the BOND_PHC_INDEX flag to notify user space */ 5792 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg) 5793 { 5794 struct ifreq *ifr = kernel_cfg->ifr; 5795 struct hwtstamp_config cfg; 5796 5797 if (kernel_cfg->copied_to_user) { 5798 /* Lower device has a legacy implementation */ 5799 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 5800 return -EFAULT; 5801 5802 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5803 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg))) 5804 return -EFAULT; 5805 } else { 5806 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5807 } 5808 5809 return 0; 5810 } 5811 5812 static int bond_hwtstamp_get(struct net_device *dev, 5813 struct kernel_hwtstamp_config *cfg) 5814 { 5815 struct bonding *bond = netdev_priv(dev); 5816 struct net_device *real_dev; 5817 int err; 5818 5819 real_dev = bond_option_active_slave_get_rcu(bond); 5820 if (!real_dev) 5821 return -EOPNOTSUPP; 5822 5823 err = generic_hwtstamp_get_lower(real_dev, cfg); 5824 if (err) 5825 return err; 5826 5827 return bond_set_phc_index_flag(cfg); 5828 } 5829 5830 static int bond_hwtstamp_set(struct net_device *dev, 5831 struct kernel_hwtstamp_config *cfg, 5832 struct netlink_ext_ack *extack) 5833 { 5834 struct bonding *bond = netdev_priv(dev); 5835 struct net_device *real_dev; 5836 int err; 5837 5838 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 5839 return -EOPNOTSUPP; 5840 5841 real_dev = bond_option_active_slave_get_rcu(bond); 5842 if (!real_dev) 5843 return -EOPNOTSUPP; 5844 5845 err = generic_hwtstamp_set_lower(real_dev, cfg, extack); 5846 if (err) 5847 return err; 5848 5849 return bond_set_phc_index_flag(cfg); 5850 } 5851 5852 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5853 struct ethtool_link_ksettings *cmd) 5854 { 5855 struct bonding *bond = netdev_priv(bond_dev); 5856 struct list_head *iter; 5857 struct slave *slave; 5858 u32 speed = 0; 5859 5860 cmd->base.duplex = DUPLEX_UNKNOWN; 5861 cmd->base.port = PORT_OTHER; 5862 5863 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5864 * do not need to check mode. Though link speed might not represent 5865 * the true receive or transmit bandwidth (not all modes are symmetric) 5866 * this is an accurate maximum. 5867 */ 5868 bond_for_each_slave(bond, slave, iter) { 5869 if (bond_slave_can_tx(slave)) { 5870 bond_update_speed_duplex(slave); 5871 if (slave->speed != SPEED_UNKNOWN) { 5872 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5873 speed = bond_mode_bcast_speed(slave, 5874 speed); 5875 else 5876 speed += slave->speed; 5877 } 5878 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5879 slave->duplex != DUPLEX_UNKNOWN) 5880 cmd->base.duplex = slave->duplex; 5881 } 5882 } 5883 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5884 5885 return 0; 5886 } 5887 5888 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5889 struct ethtool_drvinfo *drvinfo) 5890 { 5891 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5892 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5893 BOND_ABI_VERSION); 5894 } 5895 5896 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5897 struct kernel_ethtool_ts_info *info) 5898 { 5899 struct bonding *bond = netdev_priv(bond_dev); 5900 struct kernel_ethtool_ts_info ts_info; 5901 struct net_device *real_dev; 5902 bool sw_tx_support = false; 5903 struct list_head *iter; 5904 struct slave *slave; 5905 int ret = 0; 5906 5907 rcu_read_lock(); 5908 real_dev = bond_option_active_slave_get_rcu(bond); 5909 dev_hold(real_dev); 5910 rcu_read_unlock(); 5911 5912 if (real_dev) { 5913 ret = ethtool_get_ts_info_by_layer(real_dev, info); 5914 } else { 5915 /* Check if all slaves support software tx timestamping */ 5916 rcu_read_lock(); 5917 bond_for_each_slave_rcu(bond, slave, iter) { 5918 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info); 5919 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) { 5920 sw_tx_support = true; 5921 continue; 5922 } 5923 5924 sw_tx_support = false; 5925 break; 5926 } 5927 rcu_read_unlock(); 5928 } 5929 5930 if (sw_tx_support) 5931 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE; 5932 5933 dev_put(real_dev); 5934 return ret; 5935 } 5936 5937 static const struct ethtool_ops bond_ethtool_ops = { 5938 .get_drvinfo = bond_ethtool_get_drvinfo, 5939 .get_link = ethtool_op_get_link, 5940 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5941 .get_ts_info = bond_ethtool_get_ts_info, 5942 }; 5943 5944 static const struct net_device_ops bond_netdev_ops = { 5945 .ndo_init = bond_init, 5946 .ndo_uninit = bond_uninit, 5947 .ndo_open = bond_open, 5948 .ndo_stop = bond_close, 5949 .ndo_start_xmit = bond_start_xmit, 5950 .ndo_select_queue = bond_select_queue, 5951 .ndo_get_stats64 = bond_get_stats, 5952 .ndo_eth_ioctl = bond_eth_ioctl, 5953 .ndo_siocbond = bond_do_ioctl, 5954 .ndo_siocdevprivate = bond_siocdevprivate, 5955 .ndo_change_rx_flags = bond_change_rx_flags, 5956 .ndo_set_rx_mode = bond_set_rx_mode, 5957 .ndo_change_mtu = bond_change_mtu, 5958 .ndo_set_mac_address = bond_set_mac_address, 5959 .ndo_neigh_setup = bond_neigh_setup, 5960 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5961 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5962 #ifdef CONFIG_NET_POLL_CONTROLLER 5963 .ndo_netpoll_setup = bond_netpoll_setup, 5964 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5965 .ndo_poll_controller = bond_poll_controller, 5966 #endif 5967 .ndo_add_slave = bond_enslave, 5968 .ndo_del_slave = bond_release, 5969 .ndo_fix_features = bond_fix_features, 5970 .ndo_features_check = passthru_features_check, 5971 .ndo_get_xmit_slave = bond_xmit_get_slave, 5972 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5973 .ndo_bpf = bond_xdp, 5974 .ndo_xdp_xmit = bond_xdp_xmit, 5975 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5976 .ndo_hwtstamp_get = bond_hwtstamp_get, 5977 .ndo_hwtstamp_set = bond_hwtstamp_set, 5978 }; 5979 5980 static const struct device_type bond_type = { 5981 .name = "bond", 5982 }; 5983 5984 static void bond_destructor(struct net_device *bond_dev) 5985 { 5986 struct bonding *bond = netdev_priv(bond_dev); 5987 5988 if (bond->wq) 5989 destroy_workqueue(bond->wq); 5990 5991 free_percpu(bond->rr_tx_counter); 5992 } 5993 5994 void bond_setup(struct net_device *bond_dev) 5995 { 5996 struct bonding *bond = netdev_priv(bond_dev); 5997 5998 spin_lock_init(&bond->mode_lock); 5999 bond->params = bonding_defaults; 6000 6001 /* Initialize pointers */ 6002 bond->dev = bond_dev; 6003 6004 /* Initialize the device entry points */ 6005 ether_setup(bond_dev); 6006 bond_dev->max_mtu = ETH_MAX_MTU; 6007 bond_dev->netdev_ops = &bond_netdev_ops; 6008 bond_dev->ethtool_ops = &bond_ethtool_ops; 6009 6010 bond_dev->needs_free_netdev = true; 6011 bond_dev->priv_destructor = bond_destructor; 6012 6013 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 6014 6015 /* Initialize the device options */ 6016 bond_dev->flags |= IFF_MASTER; 6017 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 6018 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 6019 6020 #ifdef CONFIG_XFRM_OFFLOAD 6021 /* set up xfrm device ops (only supported in active-backup right now) */ 6022 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 6023 INIT_LIST_HEAD(&bond->ipsec_list); 6024 mutex_init(&bond->ipsec_lock); 6025 #endif /* CONFIG_XFRM_OFFLOAD */ 6026 6027 /* don't acquire bond device's netif_tx_lock when transmitting */ 6028 bond_dev->lltx = true; 6029 6030 /* Don't allow bond devices to change network namespaces. */ 6031 bond_dev->netns_immutable = true; 6032 6033 /* By default, we declare the bond to be fully 6034 * VLAN hardware accelerated capable. Special 6035 * care is taken in the various xmit functions 6036 * when there are slaves that are not hw accel 6037 * capable 6038 */ 6039 6040 bond_dev->hw_features = BOND_VLAN_FEATURES | 6041 NETIF_F_HW_VLAN_CTAG_RX | 6042 NETIF_F_HW_VLAN_CTAG_FILTER | 6043 NETIF_F_HW_VLAN_STAG_RX | 6044 NETIF_F_HW_VLAN_STAG_FILTER; 6045 6046 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 6047 bond_dev->features |= bond_dev->hw_features; 6048 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 6049 bond_dev->features |= NETIF_F_GSO_PARTIAL; 6050 #ifdef CONFIG_XFRM_OFFLOAD 6051 bond_dev->hw_features |= BOND_XFRM_FEATURES; 6052 /* Only enable XFRM features if this is an active-backup config */ 6053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 6054 bond_dev->features |= BOND_XFRM_FEATURES; 6055 #endif /* CONFIG_XFRM_OFFLOAD */ 6056 } 6057 6058 /* Destroy a bonding device. 6059 * Must be under rtnl_lock when this function is called. 6060 */ 6061 static void bond_uninit(struct net_device *bond_dev) 6062 { 6063 struct bonding *bond = netdev_priv(bond_dev); 6064 struct list_head *iter; 6065 struct slave *slave; 6066 6067 bond_netpoll_cleanup(bond_dev); 6068 6069 /* Release the bonded slaves */ 6070 bond_for_each_slave(bond, slave, iter) 6071 __bond_release_one(bond_dev, slave->dev, true, true); 6072 netdev_info(bond_dev, "Released all slaves\n"); 6073 6074 #ifdef CONFIG_XFRM_OFFLOAD 6075 mutex_destroy(&bond->ipsec_lock); 6076 #endif /* CONFIG_XFRM_OFFLOAD */ 6077 6078 bond_set_slave_arr(bond, NULL, NULL); 6079 6080 list_del_rcu(&bond->bond_list); 6081 6082 bond_debug_unregister(bond); 6083 } 6084 6085 /*------------------------- Module initialization ---------------------------*/ 6086 6087 static int __init bond_check_params(struct bond_params *params) 6088 { 6089 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 6090 struct bond_opt_value newval; 6091 const struct bond_opt_value *valptr; 6092 int arp_all_targets_value = 0; 6093 u16 ad_actor_sys_prio = 0; 6094 u16 ad_user_port_key = 0; 6095 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 6096 int arp_ip_count; 6097 int bond_mode = BOND_MODE_ROUNDROBIN; 6098 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 6099 int lacp_fast = 0; 6100 int tlb_dynamic_lb; 6101 6102 /* Convert string parameters. */ 6103 if (mode) { 6104 bond_opt_initstr(&newval, mode); 6105 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 6106 if (!valptr) { 6107 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 6108 return -EINVAL; 6109 } 6110 bond_mode = valptr->value; 6111 } 6112 6113 if (xmit_hash_policy) { 6114 if (bond_mode == BOND_MODE_ROUNDROBIN || 6115 bond_mode == BOND_MODE_ACTIVEBACKUP || 6116 bond_mode == BOND_MODE_BROADCAST) { 6117 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 6118 bond_mode_name(bond_mode)); 6119 } else { 6120 bond_opt_initstr(&newval, xmit_hash_policy); 6121 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 6122 &newval); 6123 if (!valptr) { 6124 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 6125 xmit_hash_policy); 6126 return -EINVAL; 6127 } 6128 xmit_hashtype = valptr->value; 6129 } 6130 } 6131 6132 if (lacp_rate) { 6133 if (bond_mode != BOND_MODE_8023AD) { 6134 pr_info("lacp_rate param is irrelevant in mode %s\n", 6135 bond_mode_name(bond_mode)); 6136 } else { 6137 bond_opt_initstr(&newval, lacp_rate); 6138 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 6139 &newval); 6140 if (!valptr) { 6141 pr_err("Error: Invalid lacp rate \"%s\"\n", 6142 lacp_rate); 6143 return -EINVAL; 6144 } 6145 lacp_fast = valptr->value; 6146 } 6147 } 6148 6149 if (ad_select) { 6150 bond_opt_initstr(&newval, ad_select); 6151 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 6152 &newval); 6153 if (!valptr) { 6154 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 6155 return -EINVAL; 6156 } 6157 params->ad_select = valptr->value; 6158 if (bond_mode != BOND_MODE_8023AD) 6159 pr_warn("ad_select param only affects 802.3ad mode\n"); 6160 } else { 6161 params->ad_select = BOND_AD_STABLE; 6162 } 6163 6164 if (max_bonds < 0) { 6165 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 6166 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 6167 max_bonds = BOND_DEFAULT_MAX_BONDS; 6168 } 6169 6170 if (miimon < 0) { 6171 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6172 miimon, INT_MAX); 6173 miimon = 0; 6174 } 6175 6176 if (updelay < 0) { 6177 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6178 updelay, INT_MAX); 6179 updelay = 0; 6180 } 6181 6182 if (downdelay < 0) { 6183 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6184 downdelay, INT_MAX); 6185 downdelay = 0; 6186 } 6187 6188 if ((use_carrier != 0) && (use_carrier != 1)) { 6189 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 6190 use_carrier); 6191 use_carrier = 1; 6192 } 6193 6194 if (num_peer_notif < 0 || num_peer_notif > 255) { 6195 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 6196 num_peer_notif); 6197 num_peer_notif = 1; 6198 } 6199 6200 /* reset values for 802.3ad/TLB/ALB */ 6201 if (!bond_mode_uses_arp(bond_mode)) { 6202 if (!miimon) { 6203 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"); 6204 pr_warn("Forcing miimon to 100msec\n"); 6205 miimon = BOND_DEFAULT_MIIMON; 6206 } 6207 } 6208 6209 if (tx_queues < 1 || tx_queues > 255) { 6210 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 6211 tx_queues, BOND_DEFAULT_TX_QUEUES); 6212 tx_queues = BOND_DEFAULT_TX_QUEUES; 6213 } 6214 6215 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 6216 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 6217 all_slaves_active); 6218 all_slaves_active = 0; 6219 } 6220 6221 if (resend_igmp < 0 || resend_igmp > 255) { 6222 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6223 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6224 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6225 } 6226 6227 bond_opt_initval(&newval, packets_per_slave); 6228 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6229 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6230 packets_per_slave, USHRT_MAX); 6231 packets_per_slave = 1; 6232 } 6233 6234 if (bond_mode == BOND_MODE_ALB) { 6235 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", 6236 updelay); 6237 } 6238 6239 if (!miimon) { 6240 if (updelay || downdelay) { 6241 /* just warn the user the up/down delay will have 6242 * no effect since miimon is zero... 6243 */ 6244 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", 6245 updelay, downdelay); 6246 } 6247 } else { 6248 /* don't allow arp monitoring */ 6249 if (arp_interval) { 6250 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6251 miimon, arp_interval); 6252 arp_interval = 0; 6253 } 6254 6255 if ((updelay % miimon) != 0) { 6256 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6257 updelay, miimon, (updelay / miimon) * miimon); 6258 } 6259 6260 updelay /= miimon; 6261 6262 if ((downdelay % miimon) != 0) { 6263 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6264 downdelay, miimon, 6265 (downdelay / miimon) * miimon); 6266 } 6267 6268 downdelay /= miimon; 6269 } 6270 6271 if (arp_interval < 0) { 6272 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6273 arp_interval, INT_MAX); 6274 arp_interval = 0; 6275 } 6276 6277 for (arp_ip_count = 0, i = 0; 6278 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6279 __be32 ip; 6280 6281 /* not a complete check, but good enough to catch mistakes */ 6282 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6283 !bond_is_ip_target_ok(ip)) { 6284 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6285 arp_ip_target[i]); 6286 arp_interval = 0; 6287 } else { 6288 if (bond_get_targets_ip(arp_target, ip) == -1) 6289 arp_target[arp_ip_count++] = ip; 6290 else 6291 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6292 &ip); 6293 } 6294 } 6295 6296 if (arp_interval && !arp_ip_count) { 6297 /* don't allow arping if no arp_ip_target given... */ 6298 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6299 arp_interval); 6300 arp_interval = 0; 6301 } 6302 6303 if (arp_validate) { 6304 if (!arp_interval) { 6305 pr_err("arp_validate requires arp_interval\n"); 6306 return -EINVAL; 6307 } 6308 6309 bond_opt_initstr(&newval, arp_validate); 6310 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6311 &newval); 6312 if (!valptr) { 6313 pr_err("Error: invalid arp_validate \"%s\"\n", 6314 arp_validate); 6315 return -EINVAL; 6316 } 6317 arp_validate_value = valptr->value; 6318 } else { 6319 arp_validate_value = 0; 6320 } 6321 6322 if (arp_all_targets) { 6323 bond_opt_initstr(&newval, arp_all_targets); 6324 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6325 &newval); 6326 if (!valptr) { 6327 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6328 arp_all_targets); 6329 arp_all_targets_value = 0; 6330 } else { 6331 arp_all_targets_value = valptr->value; 6332 } 6333 } 6334 6335 if (miimon) { 6336 pr_info("MII link monitoring set to %d ms\n", miimon); 6337 } else if (arp_interval) { 6338 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6339 arp_validate_value); 6340 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6341 arp_interval, valptr->string, arp_ip_count); 6342 6343 for (i = 0; i < arp_ip_count; i++) 6344 pr_cont(" %s", arp_ip_target[i]); 6345 6346 pr_cont("\n"); 6347 6348 } else if (max_bonds) { 6349 /* miimon and arp_interval not set, we need one so things 6350 * work as expected, see bonding.txt for details 6351 */ 6352 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"); 6353 } 6354 6355 if (primary && !bond_mode_uses_primary(bond_mode)) { 6356 /* currently, using a primary only makes sense 6357 * in active backup, TLB or ALB modes 6358 */ 6359 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6360 primary, bond_mode_name(bond_mode)); 6361 primary = NULL; 6362 } 6363 6364 if (primary && primary_reselect) { 6365 bond_opt_initstr(&newval, primary_reselect); 6366 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6367 &newval); 6368 if (!valptr) { 6369 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6370 primary_reselect); 6371 return -EINVAL; 6372 } 6373 primary_reselect_value = valptr->value; 6374 } else { 6375 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6376 } 6377 6378 if (fail_over_mac) { 6379 bond_opt_initstr(&newval, fail_over_mac); 6380 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6381 &newval); 6382 if (!valptr) { 6383 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6384 fail_over_mac); 6385 return -EINVAL; 6386 } 6387 fail_over_mac_value = valptr->value; 6388 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6389 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6390 } else { 6391 fail_over_mac_value = BOND_FOM_NONE; 6392 } 6393 6394 bond_opt_initstr(&newval, "default"); 6395 valptr = bond_opt_parse( 6396 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6397 &newval); 6398 if (!valptr) { 6399 pr_err("Error: No ad_actor_sys_prio default value"); 6400 return -EINVAL; 6401 } 6402 ad_actor_sys_prio = valptr->value; 6403 6404 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6405 &newval); 6406 if (!valptr) { 6407 pr_err("Error: No ad_user_port_key default value"); 6408 return -EINVAL; 6409 } 6410 ad_user_port_key = valptr->value; 6411 6412 bond_opt_initstr(&newval, "default"); 6413 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6414 if (!valptr) { 6415 pr_err("Error: No tlb_dynamic_lb default value"); 6416 return -EINVAL; 6417 } 6418 tlb_dynamic_lb = valptr->value; 6419 6420 if (lp_interval == 0) { 6421 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6422 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6423 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6424 } 6425 6426 /* fill params struct with the proper values */ 6427 params->mode = bond_mode; 6428 params->xmit_policy = xmit_hashtype; 6429 params->miimon = miimon; 6430 params->num_peer_notif = num_peer_notif; 6431 params->arp_interval = arp_interval; 6432 params->arp_validate = arp_validate_value; 6433 params->arp_all_targets = arp_all_targets_value; 6434 params->missed_max = 2; 6435 params->updelay = updelay; 6436 params->downdelay = downdelay; 6437 params->peer_notif_delay = 0; 6438 params->use_carrier = use_carrier; 6439 params->lacp_active = 1; 6440 params->lacp_fast = lacp_fast; 6441 params->primary[0] = 0; 6442 params->primary_reselect = primary_reselect_value; 6443 params->fail_over_mac = fail_over_mac_value; 6444 params->tx_queues = tx_queues; 6445 params->all_slaves_active = all_slaves_active; 6446 params->resend_igmp = resend_igmp; 6447 params->min_links = min_links; 6448 params->lp_interval = lp_interval; 6449 params->packets_per_slave = packets_per_slave; 6450 params->tlb_dynamic_lb = tlb_dynamic_lb; 6451 params->ad_actor_sys_prio = ad_actor_sys_prio; 6452 eth_zero_addr(params->ad_actor_system); 6453 params->ad_user_port_key = ad_user_port_key; 6454 params->coupled_control = 1; 6455 if (packets_per_slave > 0) { 6456 params->reciprocal_packets_per_slave = 6457 reciprocal_value(packets_per_slave); 6458 } else { 6459 /* reciprocal_packets_per_slave is unused if 6460 * packets_per_slave is 0 or 1, just initialize it 6461 */ 6462 params->reciprocal_packets_per_slave = 6463 (struct reciprocal_value) { 0 }; 6464 } 6465 6466 if (primary) 6467 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6468 6469 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6470 #if IS_ENABLED(CONFIG_IPV6) 6471 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6472 #endif 6473 6474 return 0; 6475 } 6476 6477 /* Called from registration process */ 6478 static int bond_init(struct net_device *bond_dev) 6479 { 6480 struct bonding *bond = netdev_priv(bond_dev); 6481 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6482 6483 netdev_dbg(bond_dev, "Begin bond_init\n"); 6484 6485 bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM, 6486 bond_dev->name); 6487 if (!bond->wq) 6488 return -ENOMEM; 6489 6490 bond->notifier_ctx = false; 6491 6492 spin_lock_init(&bond->stats_lock); 6493 netdev_lockdep_set_classes(bond_dev); 6494 6495 list_add_tail_rcu(&bond->bond_list, &bn->dev_list); 6496 6497 bond_prepare_sysfs_group(bond); 6498 6499 bond_debug_register(bond); 6500 6501 /* Ensure valid dev_addr */ 6502 if (is_zero_ether_addr(bond_dev->dev_addr) && 6503 bond_dev->addr_assign_type == NET_ADDR_PERM) 6504 eth_hw_addr_random(bond_dev); 6505 6506 return 0; 6507 } 6508 6509 unsigned int bond_get_num_tx_queues(void) 6510 { 6511 return tx_queues; 6512 } 6513 6514 /* Create a new bond based on the specified name and bonding parameters. 6515 * If name is NULL, obtain a suitable "bond%d" name for us. 6516 * Caller must NOT hold rtnl_lock; we need to release it here before we 6517 * set up our sysfs entries. 6518 */ 6519 int bond_create(struct net *net, const char *name) 6520 { 6521 struct net_device *bond_dev; 6522 struct bonding *bond; 6523 int res = -ENOMEM; 6524 6525 rtnl_lock(); 6526 6527 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6528 name ? name : "bond%d", NET_NAME_UNKNOWN, 6529 bond_setup, tx_queues); 6530 if (!bond_dev) 6531 goto out; 6532 6533 bond = netdev_priv(bond_dev); 6534 dev_net_set(bond_dev, net); 6535 bond_dev->rtnl_link_ops = &bond_link_ops; 6536 6537 res = register_netdevice(bond_dev); 6538 if (res < 0) { 6539 free_netdev(bond_dev); 6540 goto out; 6541 } 6542 6543 netif_carrier_off(bond_dev); 6544 6545 bond_work_init_all(bond); 6546 6547 out: 6548 rtnl_unlock(); 6549 return res; 6550 } 6551 6552 static int __net_init bond_net_init(struct net *net) 6553 { 6554 struct bond_net *bn = net_generic(net, bond_net_id); 6555 6556 bn->net = net; 6557 INIT_LIST_HEAD(&bn->dev_list); 6558 6559 bond_create_proc_dir(bn); 6560 bond_create_sysfs(bn); 6561 6562 return 0; 6563 } 6564 6565 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters 6566 * race condition in bond unloading") we need to remove sysfs files 6567 * before we remove our devices (done later in bond_net_exit_batch_rtnl()) 6568 */ 6569 static void __net_exit bond_net_pre_exit(struct net *net) 6570 { 6571 struct bond_net *bn = net_generic(net, bond_net_id); 6572 6573 bond_destroy_sysfs(bn); 6574 } 6575 6576 static void __net_exit bond_net_exit_batch_rtnl(struct list_head *net_list, 6577 struct list_head *dev_kill_list) 6578 { 6579 struct bond_net *bn; 6580 struct net *net; 6581 6582 /* Kill off any bonds created after unregistering bond rtnl ops */ 6583 list_for_each_entry(net, net_list, exit_list) { 6584 struct bonding *bond, *tmp_bond; 6585 6586 bn = net_generic(net, bond_net_id); 6587 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6588 unregister_netdevice_queue(bond->dev, dev_kill_list); 6589 } 6590 } 6591 6592 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory 6593 * only after all bonds are gone") bond_destroy_proc_dir() is called 6594 * after bond_net_exit_batch_rtnl() has completed. 6595 */ 6596 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6597 { 6598 struct bond_net *bn; 6599 struct net *net; 6600 6601 list_for_each_entry(net, net_list, exit_list) { 6602 bn = net_generic(net, bond_net_id); 6603 bond_destroy_proc_dir(bn); 6604 } 6605 } 6606 6607 static struct pernet_operations bond_net_ops = { 6608 .init = bond_net_init, 6609 .pre_exit = bond_net_pre_exit, 6610 .exit_batch_rtnl = bond_net_exit_batch_rtnl, 6611 .exit_batch = bond_net_exit_batch, 6612 .id = &bond_net_id, 6613 .size = sizeof(struct bond_net), 6614 }; 6615 6616 static int __init bonding_init(void) 6617 { 6618 int i; 6619 int res; 6620 6621 res = bond_check_params(&bonding_defaults); 6622 if (res) 6623 goto out; 6624 6625 bond_create_debugfs(); 6626 6627 res = register_pernet_subsys(&bond_net_ops); 6628 if (res) 6629 goto err_net_ops; 6630 6631 res = bond_netlink_init(); 6632 if (res) 6633 goto err_link; 6634 6635 for (i = 0; i < max_bonds; i++) { 6636 res = bond_create(&init_net, NULL); 6637 if (res) 6638 goto err; 6639 } 6640 6641 skb_flow_dissector_init(&flow_keys_bonding, 6642 flow_keys_bonding_keys, 6643 ARRAY_SIZE(flow_keys_bonding_keys)); 6644 6645 register_netdevice_notifier(&bond_netdev_notifier); 6646 out: 6647 return res; 6648 err: 6649 bond_netlink_fini(); 6650 err_link: 6651 unregister_pernet_subsys(&bond_net_ops); 6652 err_net_ops: 6653 bond_destroy_debugfs(); 6654 goto out; 6655 6656 } 6657 6658 static void __exit bonding_exit(void) 6659 { 6660 unregister_netdevice_notifier(&bond_netdev_notifier); 6661 6662 bond_netlink_fini(); 6663 unregister_pernet_subsys(&bond_net_ops); 6664 6665 bond_destroy_debugfs(); 6666 6667 #ifdef CONFIG_NET_POLL_CONTROLLER 6668 /* Make sure we don't have an imbalance on our netpoll blocking */ 6669 WARN_ON(atomic_read(&netpoll_block_tx)); 6670 #endif 6671 } 6672 6673 module_init(bonding_init); 6674 module_exit(bonding_exit); 6675 MODULE_LICENSE("GPL"); 6676 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6677 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6678