1 // SPDX-License-Identifier: GPL-1.0+ 2 /* 3 * originally based on the dummy device. 4 * 5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov. 6 * Based on dummy.c, and eql.c devices. 7 * 8 * bonding.c: an Ethernet Bonding driver 9 * 10 * This is useful to talk to a Cisco EtherChannel compatible equipment: 11 * Cisco 5500 12 * Sun Trunking (Solaris) 13 * Alteon AceDirector Trunks 14 * Linux Bonding 15 * and probably many L2 switches ... 16 * 17 * How it works: 18 * ifconfig bond0 ipaddress netmask up 19 * will setup a network device, with an ip address. No mac address 20 * will be assigned at this time. The hw mac address will come from 21 * the first slave bonded to the channel. All slaves will then use 22 * this hw mac address. 23 * 24 * ifconfig bond0 down 25 * will release all slaves, marking them as down. 26 * 27 * ifenslave bond0 eth0 28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either 29 * a: be used as initial mac address 30 * b: if a hw mac address already is there, eth0's hw mac address 31 * will then be set from bond0. 32 * 33 */ 34 35 #include <linux/kernel.h> 36 #include <linux/module.h> 37 #include <linux/types.h> 38 #include <linux/fcntl.h> 39 #include <linux/filter.h> 40 #include <linux/interrupt.h> 41 #include <linux/ptrace.h> 42 #include <linux/ioport.h> 43 #include <linux/in.h> 44 #include <net/ip.h> 45 #include <linux/ip.h> 46 #include <linux/icmp.h> 47 #include <linux/icmpv6.h> 48 #include <linux/tcp.h> 49 #include <linux/udp.h> 50 #include <linux/slab.h> 51 #include <linux/string.h> 52 #include <linux/init.h> 53 #include <linux/timer.h> 54 #include <linux/socket.h> 55 #include <linux/ctype.h> 56 #include <linux/inet.h> 57 #include <linux/bitops.h> 58 #include <linux/io.h> 59 #include <asm/dma.h> 60 #include <linux/uaccess.h> 61 #include <linux/errno.h> 62 #include <linux/netdevice.h> 63 #include <linux/inetdevice.h> 64 #include <linux/igmp.h> 65 #include <linux/etherdevice.h> 66 #include <linux/skbuff.h> 67 #include <net/sock.h> 68 #include <linux/rtnetlink.h> 69 #include <linux/smp.h> 70 #include <linux/if_ether.h> 71 #include <net/arp.h> 72 #include <linux/mii.h> 73 #include <linux/ethtool.h> 74 #include <linux/if_vlan.h> 75 #include <linux/if_bonding.h> 76 #include <linux/phy.h> 77 #include <linux/jiffies.h> 78 #include <linux/preempt.h> 79 #include <net/route.h> 80 #include <net/net_namespace.h> 81 #include <net/netns/generic.h> 82 #include <net/pkt_sched.h> 83 #include <linux/rculist.h> 84 #include <net/flow_dissector.h> 85 #include <net/xfrm.h> 86 #include <net/bonding.h> 87 #include <net/bond_3ad.h> 88 #include <net/bond_alb.h> 89 #if IS_ENABLED(CONFIG_TLS_DEVICE) 90 #include <net/tls.h> 91 #endif 92 #include <net/ip6_route.h> 93 #include <net/xdp.h> 94 95 #include "bonding_priv.h" 96 97 /*---------------------------- Module parameters ----------------------------*/ 98 99 /* monitor all links that often (in milliseconds). <=0 disables monitoring */ 100 101 static int max_bonds = BOND_DEFAULT_MAX_BONDS; 102 static int tx_queues = BOND_DEFAULT_TX_QUEUES; 103 static int num_peer_notif = 1; 104 static int miimon; 105 static int updelay; 106 static int downdelay; 107 static int use_carrier = 1; 108 static char *mode; 109 static char *primary; 110 static char *primary_reselect; 111 static char *lacp_rate; 112 static int min_links; 113 static char *ad_select; 114 static char *xmit_hash_policy; 115 static int arp_interval; 116 static char *arp_ip_target[BOND_MAX_ARP_TARGETS]; 117 static char *arp_validate; 118 static char *arp_all_targets; 119 static char *fail_over_mac; 120 static int all_slaves_active; 121 static struct bond_params bonding_defaults; 122 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP; 123 static int packets_per_slave = 1; 124 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 125 126 module_param(max_bonds, int, 0); 127 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices"); 128 module_param(tx_queues, int, 0); 129 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)"); 130 module_param_named(num_grat_arp, num_peer_notif, int, 0644); 131 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on " 132 "failover event (alias of num_unsol_na)"); 133 module_param_named(num_unsol_na, num_peer_notif, int, 0644); 134 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on " 135 "failover event (alias of num_grat_arp)"); 136 module_param(miimon, int, 0); 137 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds"); 138 module_param(updelay, int, 0); 139 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds"); 140 module_param(downdelay, int, 0); 141 MODULE_PARM_DESC(downdelay, "Delay before considering link down, " 142 "in milliseconds"); 143 module_param(use_carrier, int, 0); 144 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; " 145 "0 for off, 1 for on (default)"); 146 module_param(mode, charp, 0); 147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, " 148 "1 for active-backup, 2 for balance-xor, " 149 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, " 150 "6 for balance-alb"); 151 module_param(primary, charp, 0); 152 MODULE_PARM_DESC(primary, "Primary network device to use"); 153 module_param(primary_reselect, charp, 0); 154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave " 155 "once it comes up; " 156 "0 for always (default), " 157 "1 for only if speed of primary is " 158 "better, " 159 "2 for only on active slave " 160 "failure"); 161 module_param(lacp_rate, charp, 0); 162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; " 163 "0 for slow, 1 for fast"); 164 module_param(ad_select, charp, 0); 165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; " 166 "0 for stable (default), 1 for bandwidth, " 167 "2 for count"); 168 module_param(min_links, int, 0); 169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier"); 170 171 module_param(xmit_hash_policy, charp, 0); 172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; " 173 "0 for layer 2 (default), 1 for layer 3+4, " 174 "2 for layer 2+3, 3 for encap layer 2+3, " 175 "4 for encap layer 3+4, 5 for vlan+srcmac"); 176 module_param(arp_interval, int, 0); 177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds"); 178 module_param_array(arp_ip_target, charp, NULL, 0); 179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form"); 180 module_param(arp_validate, charp, 0); 181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; " 182 "0 for none (default), 1 for active, " 183 "2 for backup, 3 for all"); 184 module_param(arp_all_targets, charp, 0); 185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all"); 186 module_param(fail_over_mac, charp, 0); 187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to " 188 "the same MAC; 0 for none (default), " 189 "1 for active, 2 for follow"); 190 module_param(all_slaves_active, int, 0); 191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface " 192 "by setting active flag for all slaves; " 193 "0 for never (default), 1 for always."); 194 module_param(resend_igmp, int, 0); 195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on " 196 "link failure"); 197 module_param(packets_per_slave, int, 0); 198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr " 199 "mode; 0 for a random slave, 1 packet per " 200 "slave (default), >1 packets per slave."); 201 module_param(lp_interval, uint, 0); 202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where " 203 "the bonding driver sends learning packets to " 204 "each slaves peer switch. The default is 1."); 205 206 /*----------------------------- Global variables ----------------------------*/ 207 208 #ifdef CONFIG_NET_POLL_CONTROLLER 209 atomic_t netpoll_block_tx = ATOMIC_INIT(0); 210 #endif 211 212 unsigned int bond_net_id __read_mostly; 213 214 static const struct flow_dissector_key flow_keys_bonding_keys[] = { 215 { 216 .key_id = FLOW_DISSECTOR_KEY_CONTROL, 217 .offset = offsetof(struct flow_keys, control), 218 }, 219 { 220 .key_id = FLOW_DISSECTOR_KEY_BASIC, 221 .offset = offsetof(struct flow_keys, basic), 222 }, 223 { 224 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS, 225 .offset = offsetof(struct flow_keys, addrs.v4addrs), 226 }, 227 { 228 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS, 229 .offset = offsetof(struct flow_keys, addrs.v6addrs), 230 }, 231 { 232 .key_id = FLOW_DISSECTOR_KEY_TIPC, 233 .offset = offsetof(struct flow_keys, addrs.tipckey), 234 }, 235 { 236 .key_id = FLOW_DISSECTOR_KEY_PORTS, 237 .offset = offsetof(struct flow_keys, ports), 238 }, 239 { 240 .key_id = FLOW_DISSECTOR_KEY_ICMP, 241 .offset = offsetof(struct flow_keys, icmp), 242 }, 243 { 244 .key_id = FLOW_DISSECTOR_KEY_VLAN, 245 .offset = offsetof(struct flow_keys, vlan), 246 }, 247 { 248 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL, 249 .offset = offsetof(struct flow_keys, tags), 250 }, 251 { 252 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID, 253 .offset = offsetof(struct flow_keys, keyid), 254 }, 255 }; 256 257 static struct flow_dissector flow_keys_bonding __read_mostly; 258 259 /*-------------------------- Forward declarations ---------------------------*/ 260 261 static int bond_init(struct net_device *bond_dev); 262 static void bond_uninit(struct net_device *bond_dev); 263 static void bond_get_stats(struct net_device *bond_dev, 264 struct rtnl_link_stats64 *stats); 265 static void bond_slave_arr_handler(struct work_struct *work); 266 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 267 int mod); 268 static void bond_netdev_notify_work(struct work_struct *work); 269 270 /*---------------------------- General routines -----------------------------*/ 271 272 const char *bond_mode_name(int mode) 273 { 274 static const char *names[] = { 275 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)", 276 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)", 277 [BOND_MODE_XOR] = "load balancing (xor)", 278 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)", 279 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation", 280 [BOND_MODE_TLB] = "transmit load balancing", 281 [BOND_MODE_ALB] = "adaptive load balancing", 282 }; 283 284 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB) 285 return "unknown"; 286 287 return names[mode]; 288 } 289 290 /** 291 * bond_dev_queue_xmit - Prepare skb for xmit. 292 * 293 * @bond: bond device that got this skb for tx. 294 * @skb: hw accel VLAN tagged skb to transmit 295 * @slave_dev: slave that is supposed to xmit this skbuff 296 */ 297 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, 298 struct net_device *slave_dev) 299 { 300 skb->dev = slave_dev; 301 302 BUILD_BUG_ON(sizeof(skb->queue_mapping) != 303 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping)); 304 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping); 305 306 if (unlikely(netpoll_tx_running(bond->dev))) 307 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb); 308 309 return dev_queue_xmit(skb); 310 } 311 312 static bool bond_sk_check(struct bonding *bond) 313 { 314 switch (BOND_MODE(bond)) { 315 case BOND_MODE_8023AD: 316 case BOND_MODE_XOR: 317 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34) 318 return true; 319 fallthrough; 320 default: 321 return false; 322 } 323 } 324 325 static bool bond_xdp_check(struct bonding *bond) 326 { 327 switch (BOND_MODE(bond)) { 328 case BOND_MODE_ROUNDROBIN: 329 case BOND_MODE_ACTIVEBACKUP: 330 return true; 331 case BOND_MODE_8023AD: 332 case BOND_MODE_XOR: 333 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q 334 * payload is not in the packet due to hardware offload. 335 */ 336 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC) 337 return true; 338 fallthrough; 339 default: 340 return false; 341 } 342 } 343 344 /*---------------------------------- VLAN -----------------------------------*/ 345 346 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid, 347 * We don't protect the slave list iteration with a lock because: 348 * a. This operation is performed in IOCTL context, 349 * b. The operation is protected by the RTNL semaphore in the 8021q code, 350 * c. Holding a lock with BH disabled while directly calling a base driver 351 * entry point is generally a BAD idea. 352 * 353 * The design of synchronization/protection for this operation in the 8021q 354 * module is good for one or more VLAN devices over a single physical device 355 * and cannot be extended for a teaming solution like bonding, so there is a 356 * potential race condition here where a net device from the vlan group might 357 * be referenced (either by a base driver or the 8021q code) while it is being 358 * removed from the system. However, it turns out we're not making matters 359 * worse, and if it works for regular VLAN usage it will work here too. 360 */ 361 362 /** 363 * bond_vlan_rx_add_vid - Propagates adding an id to slaves 364 * @bond_dev: bonding net device that got called 365 * @proto: network protocol ID 366 * @vid: vlan id being added 367 */ 368 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, 369 __be16 proto, u16 vid) 370 { 371 struct bonding *bond = netdev_priv(bond_dev); 372 struct slave *slave, *rollback_slave; 373 struct list_head *iter; 374 int res; 375 376 bond_for_each_slave(bond, slave, iter) { 377 res = vlan_vid_add(slave->dev, proto, vid); 378 if (res) 379 goto unwind; 380 } 381 382 return 0; 383 384 unwind: 385 /* unwind to the slave that failed */ 386 bond_for_each_slave(bond, rollback_slave, iter) { 387 if (rollback_slave == slave) 388 break; 389 390 vlan_vid_del(rollback_slave->dev, proto, vid); 391 } 392 393 return res; 394 } 395 396 /** 397 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves 398 * @bond_dev: bonding net device that got called 399 * @proto: network protocol ID 400 * @vid: vlan id being removed 401 */ 402 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, 403 __be16 proto, u16 vid) 404 { 405 struct bonding *bond = netdev_priv(bond_dev); 406 struct list_head *iter; 407 struct slave *slave; 408 409 bond_for_each_slave(bond, slave, iter) 410 vlan_vid_del(slave->dev, proto, vid); 411 412 if (bond_is_lb(bond)) 413 bond_alb_clear_vlan(bond, vid); 414 415 return 0; 416 } 417 418 /*---------------------------------- XFRM -----------------------------------*/ 419 420 #ifdef CONFIG_XFRM_OFFLOAD 421 /** 422 * bond_ipsec_add_sa - program device with a security association 423 * @xs: pointer to transformer state struct 424 * @extack: extack point to fill failure reason 425 **/ 426 static int bond_ipsec_add_sa(struct xfrm_state *xs, 427 struct netlink_ext_ack *extack) 428 { 429 struct net_device *bond_dev = xs->xso.dev; 430 struct net_device *real_dev; 431 netdevice_tracker tracker; 432 struct bond_ipsec *ipsec; 433 struct bonding *bond; 434 struct slave *slave; 435 int err; 436 437 if (!bond_dev) 438 return -EINVAL; 439 440 rcu_read_lock(); 441 bond = netdev_priv(bond_dev); 442 slave = rcu_dereference(bond->curr_active_slave); 443 real_dev = slave ? slave->dev : NULL; 444 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 445 rcu_read_unlock(); 446 if (!real_dev) { 447 err = -ENODEV; 448 goto out; 449 } 450 451 if (!real_dev->xfrmdev_ops || 452 !real_dev->xfrmdev_ops->xdo_dev_state_add || 453 netif_is_bond_master(real_dev)) { 454 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload"); 455 err = -EINVAL; 456 goto out; 457 } 458 459 ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL); 460 if (!ipsec) { 461 err = -ENOMEM; 462 goto out; 463 } 464 465 xs->xso.real_dev = real_dev; 466 err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack); 467 if (!err) { 468 ipsec->xs = xs; 469 INIT_LIST_HEAD(&ipsec->list); 470 mutex_lock(&bond->ipsec_lock); 471 list_add(&ipsec->list, &bond->ipsec_list); 472 mutex_unlock(&bond->ipsec_lock); 473 } else { 474 kfree(ipsec); 475 } 476 out: 477 netdev_put(real_dev, &tracker); 478 return err; 479 } 480 481 static void bond_ipsec_add_sa_all(struct bonding *bond) 482 { 483 struct net_device *bond_dev = bond->dev; 484 struct net_device *real_dev; 485 struct bond_ipsec *ipsec; 486 struct slave *slave; 487 488 slave = rtnl_dereference(bond->curr_active_slave); 489 real_dev = slave ? slave->dev : NULL; 490 if (!real_dev) 491 return; 492 493 mutex_lock(&bond->ipsec_lock); 494 if (!real_dev->xfrmdev_ops || 495 !real_dev->xfrmdev_ops->xdo_dev_state_add || 496 netif_is_bond_master(real_dev)) { 497 if (!list_empty(&bond->ipsec_list)) 498 slave_warn(bond_dev, real_dev, 499 "%s: no slave xdo_dev_state_add\n", 500 __func__); 501 goto out; 502 } 503 504 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 505 /* If new state is added before ipsec_lock acquired */ 506 if (ipsec->xs->xso.real_dev == real_dev) 507 continue; 508 509 ipsec->xs->xso.real_dev = real_dev; 510 if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) { 511 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__); 512 ipsec->xs->xso.real_dev = NULL; 513 } 514 } 515 out: 516 mutex_unlock(&bond->ipsec_lock); 517 } 518 519 /** 520 * bond_ipsec_del_sa - clear out this specific SA 521 * @xs: pointer to transformer state struct 522 **/ 523 static void bond_ipsec_del_sa(struct xfrm_state *xs) 524 { 525 struct net_device *bond_dev = xs->xso.dev; 526 struct net_device *real_dev; 527 netdevice_tracker tracker; 528 struct bond_ipsec *ipsec; 529 struct bonding *bond; 530 struct slave *slave; 531 532 if (!bond_dev) 533 return; 534 535 rcu_read_lock(); 536 bond = netdev_priv(bond_dev); 537 slave = rcu_dereference(bond->curr_active_slave); 538 real_dev = slave ? slave->dev : NULL; 539 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 540 rcu_read_unlock(); 541 542 if (!slave) 543 goto out; 544 545 if (!xs->xso.real_dev) 546 goto out; 547 548 WARN_ON(xs->xso.real_dev != real_dev); 549 550 if (!real_dev->xfrmdev_ops || 551 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 552 netif_is_bond_master(real_dev)) { 553 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__); 554 goto out; 555 } 556 557 real_dev->xfrmdev_ops->xdo_dev_state_delete(xs); 558 out: 559 netdev_put(real_dev, &tracker); 560 mutex_lock(&bond->ipsec_lock); 561 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 562 if (ipsec->xs == xs) { 563 list_del(&ipsec->list); 564 kfree(ipsec); 565 break; 566 } 567 } 568 mutex_unlock(&bond->ipsec_lock); 569 } 570 571 static void bond_ipsec_del_sa_all(struct bonding *bond) 572 { 573 struct net_device *bond_dev = bond->dev; 574 struct net_device *real_dev; 575 struct bond_ipsec *ipsec; 576 struct slave *slave; 577 578 slave = rtnl_dereference(bond->curr_active_slave); 579 real_dev = slave ? slave->dev : NULL; 580 if (!real_dev) 581 return; 582 583 mutex_lock(&bond->ipsec_lock); 584 list_for_each_entry(ipsec, &bond->ipsec_list, list) { 585 if (!ipsec->xs->xso.real_dev) 586 continue; 587 588 if (!real_dev->xfrmdev_ops || 589 !real_dev->xfrmdev_ops->xdo_dev_state_delete || 590 netif_is_bond_master(real_dev)) { 591 slave_warn(bond_dev, real_dev, 592 "%s: no slave xdo_dev_state_delete\n", 593 __func__); 594 } else { 595 real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs); 596 if (real_dev->xfrmdev_ops->xdo_dev_state_free) 597 real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs); 598 } 599 } 600 mutex_unlock(&bond->ipsec_lock); 601 } 602 603 static void bond_ipsec_free_sa(struct xfrm_state *xs) 604 { 605 struct net_device *bond_dev = xs->xso.dev; 606 struct net_device *real_dev; 607 netdevice_tracker tracker; 608 struct bonding *bond; 609 struct slave *slave; 610 611 if (!bond_dev) 612 return; 613 614 rcu_read_lock(); 615 bond = netdev_priv(bond_dev); 616 slave = rcu_dereference(bond->curr_active_slave); 617 real_dev = slave ? slave->dev : NULL; 618 netdev_hold(real_dev, &tracker, GFP_ATOMIC); 619 rcu_read_unlock(); 620 621 if (!slave) 622 goto out; 623 624 if (!xs->xso.real_dev) 625 goto out; 626 627 WARN_ON(xs->xso.real_dev != real_dev); 628 629 if (real_dev && real_dev->xfrmdev_ops && 630 real_dev->xfrmdev_ops->xdo_dev_state_free) 631 real_dev->xfrmdev_ops->xdo_dev_state_free(xs); 632 out: 633 netdev_put(real_dev, &tracker); 634 } 635 636 /** 637 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload 638 * @skb: current data packet 639 * @xs: pointer to transformer state struct 640 **/ 641 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs) 642 { 643 struct net_device *bond_dev = xs->xso.dev; 644 struct net_device *real_dev; 645 struct slave *curr_active; 646 struct bonding *bond; 647 bool ok = false; 648 649 bond = netdev_priv(bond_dev); 650 rcu_read_lock(); 651 curr_active = rcu_dereference(bond->curr_active_slave); 652 if (!curr_active) 653 goto out; 654 real_dev = curr_active->dev; 655 656 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) 657 goto out; 658 659 if (!xs->xso.real_dev) 660 goto out; 661 662 if (!real_dev->xfrmdev_ops || 663 !real_dev->xfrmdev_ops->xdo_dev_offload_ok || 664 netif_is_bond_master(real_dev)) 665 goto out; 666 667 ok = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs); 668 out: 669 rcu_read_unlock(); 670 return ok; 671 } 672 673 static const struct xfrmdev_ops bond_xfrmdev_ops = { 674 .xdo_dev_state_add = bond_ipsec_add_sa, 675 .xdo_dev_state_delete = bond_ipsec_del_sa, 676 .xdo_dev_state_free = bond_ipsec_free_sa, 677 .xdo_dev_offload_ok = bond_ipsec_offload_ok, 678 }; 679 #endif /* CONFIG_XFRM_OFFLOAD */ 680 681 /*------------------------------- Link status -------------------------------*/ 682 683 /* Set the carrier state for the master according to the state of its 684 * slaves. If any slaves are up, the master is up. In 802.3ad mode, 685 * do special 802.3ad magic. 686 * 687 * Returns zero if carrier state does not change, nonzero if it does. 688 */ 689 int bond_set_carrier(struct bonding *bond) 690 { 691 struct list_head *iter; 692 struct slave *slave; 693 694 if (!bond_has_slaves(bond)) 695 goto down; 696 697 if (BOND_MODE(bond) == BOND_MODE_8023AD) 698 return bond_3ad_set_carrier(bond); 699 700 bond_for_each_slave(bond, slave, iter) { 701 if (slave->link == BOND_LINK_UP) { 702 if (!netif_carrier_ok(bond->dev)) { 703 netif_carrier_on(bond->dev); 704 return 1; 705 } 706 return 0; 707 } 708 } 709 710 down: 711 if (netif_carrier_ok(bond->dev)) { 712 netif_carrier_off(bond->dev); 713 return 1; 714 } 715 return 0; 716 } 717 718 /* Get link speed and duplex from the slave's base driver 719 * using ethtool. If for some reason the call fails or the 720 * values are invalid, set speed and duplex to -1, 721 * and return. Return 1 if speed or duplex settings are 722 * UNKNOWN; 0 otherwise. 723 */ 724 static int bond_update_speed_duplex(struct slave *slave) 725 { 726 struct net_device *slave_dev = slave->dev; 727 struct ethtool_link_ksettings ecmd; 728 int res; 729 730 slave->speed = SPEED_UNKNOWN; 731 slave->duplex = DUPLEX_UNKNOWN; 732 733 res = __ethtool_get_link_ksettings(slave_dev, &ecmd); 734 if (res < 0) 735 return 1; 736 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) 737 return 1; 738 switch (ecmd.base.duplex) { 739 case DUPLEX_FULL: 740 case DUPLEX_HALF: 741 break; 742 default: 743 return 1; 744 } 745 746 slave->speed = ecmd.base.speed; 747 slave->duplex = ecmd.base.duplex; 748 749 return 0; 750 } 751 752 const char *bond_slave_link_status(s8 link) 753 { 754 switch (link) { 755 case BOND_LINK_UP: 756 return "up"; 757 case BOND_LINK_FAIL: 758 return "going down"; 759 case BOND_LINK_DOWN: 760 return "down"; 761 case BOND_LINK_BACK: 762 return "going back"; 763 default: 764 return "unknown"; 765 } 766 } 767 768 /* if <dev> supports MII link status reporting, check its link status. 769 * 770 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(), 771 * depending upon the setting of the use_carrier parameter. 772 * 773 * Return either BMSR_LSTATUS, meaning that the link is up (or we 774 * can't tell and just pretend it is), or 0, meaning that the link is 775 * down. 776 * 777 * If reporting is non-zero, instead of faking link up, return -1 if 778 * both ETHTOOL and MII ioctls fail (meaning the device does not 779 * support them). If use_carrier is set, return whatever it says. 780 * It'd be nice if there was a good way to tell if a driver supports 781 * netif_carrier, but there really isn't. 782 */ 783 static int bond_check_dev_link(struct bonding *bond, 784 struct net_device *slave_dev, int reporting) 785 { 786 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 787 int (*ioctl)(struct net_device *, struct ifreq *, int); 788 struct ifreq ifr; 789 struct mii_ioctl_data *mii; 790 791 if (!reporting && !netif_running(slave_dev)) 792 return 0; 793 794 if (bond->params.use_carrier) 795 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0; 796 797 /* Try to get link status using Ethtool first. */ 798 if (slave_dev->ethtool_ops->get_link) 799 return slave_dev->ethtool_ops->get_link(slave_dev) ? 800 BMSR_LSTATUS : 0; 801 802 /* Ethtool can't be used, fallback to MII ioctls. */ 803 ioctl = slave_ops->ndo_eth_ioctl; 804 if (ioctl) { 805 /* TODO: set pointer to correct ioctl on a per team member 806 * bases to make this more efficient. that is, once 807 * we determine the correct ioctl, we will always 808 * call it and not the others for that team 809 * member. 810 */ 811 812 /* We cannot assume that SIOCGMIIPHY will also read a 813 * register; not all network drivers (e.g., e100) 814 * support that. 815 */ 816 817 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */ 818 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ); 819 mii = if_mii(&ifr); 820 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) { 821 mii->reg_num = MII_BMSR; 822 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0) 823 return mii->val_out & BMSR_LSTATUS; 824 } 825 } 826 827 /* If reporting, report that either there's no ndo_eth_ioctl, 828 * or both SIOCGMIIREG and get_link failed (meaning that we 829 * cannot report link status). If not reporting, pretend 830 * we're ok. 831 */ 832 return reporting ? -1 : BMSR_LSTATUS; 833 } 834 835 /*----------------------------- Multicast list ------------------------------*/ 836 837 /* Push the promiscuity flag down to appropriate slaves */ 838 static int bond_set_promiscuity(struct bonding *bond, int inc) 839 { 840 struct list_head *iter; 841 int err = 0; 842 843 if (bond_uses_primary(bond)) { 844 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 845 846 if (curr_active) 847 err = dev_set_promiscuity(curr_active->dev, inc); 848 } else { 849 struct slave *slave; 850 851 bond_for_each_slave(bond, slave, iter) { 852 err = dev_set_promiscuity(slave->dev, inc); 853 if (err) 854 return err; 855 } 856 } 857 return err; 858 } 859 860 /* Push the allmulti flag down to all slaves */ 861 static int bond_set_allmulti(struct bonding *bond, int inc) 862 { 863 struct list_head *iter; 864 int err = 0; 865 866 if (bond_uses_primary(bond)) { 867 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave); 868 869 if (curr_active) 870 err = dev_set_allmulti(curr_active->dev, inc); 871 } else { 872 struct slave *slave; 873 874 bond_for_each_slave(bond, slave, iter) { 875 err = dev_set_allmulti(slave->dev, inc); 876 if (err) 877 return err; 878 } 879 } 880 return err; 881 } 882 883 /* Retrieve the list of registered multicast addresses for the bonding 884 * device and retransmit an IGMP JOIN request to the current active 885 * slave. 886 */ 887 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work) 888 { 889 struct bonding *bond = container_of(work, struct bonding, 890 mcast_work.work); 891 892 if (!rtnl_trylock()) { 893 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 894 return; 895 } 896 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev); 897 898 if (bond->igmp_retrans > 1) { 899 bond->igmp_retrans--; 900 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5); 901 } 902 rtnl_unlock(); 903 } 904 905 /* Flush bond's hardware addresses from slave */ 906 static void bond_hw_addr_flush(struct net_device *bond_dev, 907 struct net_device *slave_dev) 908 { 909 struct bonding *bond = netdev_priv(bond_dev); 910 911 dev_uc_unsync(slave_dev, bond_dev); 912 dev_mc_unsync(slave_dev, bond_dev); 913 914 if (BOND_MODE(bond) == BOND_MODE_8023AD) 915 dev_mc_del(slave_dev, lacpdu_mcast_addr); 916 } 917 918 /*--------------------------- Active slave change ---------------------------*/ 919 920 /* Update the hardware address list and promisc/allmulti for the new and 921 * old active slaves (if any). Modes that are not using primary keep all 922 * slaves up date at all times; only the modes that use primary need to call 923 * this function to swap these settings during a failover. 924 */ 925 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active, 926 struct slave *old_active) 927 { 928 if (old_active) { 929 if (bond->dev->flags & IFF_PROMISC) 930 dev_set_promiscuity(old_active->dev, -1); 931 932 if (bond->dev->flags & IFF_ALLMULTI) 933 dev_set_allmulti(old_active->dev, -1); 934 935 if (bond->dev->flags & IFF_UP) 936 bond_hw_addr_flush(bond->dev, old_active->dev); 937 938 bond_slave_ns_maddrs_add(bond, old_active); 939 } 940 941 if (new_active) { 942 /* FIXME: Signal errors upstream. */ 943 if (bond->dev->flags & IFF_PROMISC) 944 dev_set_promiscuity(new_active->dev, 1); 945 946 if (bond->dev->flags & IFF_ALLMULTI) 947 dev_set_allmulti(new_active->dev, 1); 948 949 if (bond->dev->flags & IFF_UP) { 950 netif_addr_lock_bh(bond->dev); 951 dev_uc_sync(new_active->dev, bond->dev); 952 dev_mc_sync(new_active->dev, bond->dev); 953 netif_addr_unlock_bh(bond->dev); 954 } 955 956 bond_slave_ns_maddrs_del(bond, new_active); 957 } 958 } 959 960 /** 961 * bond_set_dev_addr - clone slave's address to bond 962 * @bond_dev: bond net device 963 * @slave_dev: slave net device 964 * 965 * Should be called with RTNL held. 966 */ 967 static int bond_set_dev_addr(struct net_device *bond_dev, 968 struct net_device *slave_dev) 969 { 970 int err; 971 972 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n", 973 bond_dev, slave_dev, slave_dev->addr_len); 974 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL); 975 if (err) 976 return err; 977 978 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len); 979 bond_dev->addr_assign_type = NET_ADDR_STOLEN; 980 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev); 981 return 0; 982 } 983 984 static struct slave *bond_get_old_active(struct bonding *bond, 985 struct slave *new_active) 986 { 987 struct slave *slave; 988 struct list_head *iter; 989 990 bond_for_each_slave(bond, slave, iter) { 991 if (slave == new_active) 992 continue; 993 994 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr)) 995 return slave; 996 } 997 998 return NULL; 999 } 1000 1001 /* bond_do_fail_over_mac 1002 * 1003 * Perform special MAC address swapping for fail_over_mac settings 1004 * 1005 * Called with RTNL 1006 */ 1007 static void bond_do_fail_over_mac(struct bonding *bond, 1008 struct slave *new_active, 1009 struct slave *old_active) 1010 { 1011 u8 tmp_mac[MAX_ADDR_LEN]; 1012 struct sockaddr_storage ss; 1013 int rv; 1014 1015 switch (bond->params.fail_over_mac) { 1016 case BOND_FOM_ACTIVE: 1017 if (new_active) { 1018 rv = bond_set_dev_addr(bond->dev, new_active->dev); 1019 if (rv) 1020 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n", 1021 -rv); 1022 } 1023 break; 1024 case BOND_FOM_FOLLOW: 1025 /* if new_active && old_active, swap them 1026 * if just old_active, do nothing (going to no active slave) 1027 * if just new_active, set new_active to bond's MAC 1028 */ 1029 if (!new_active) 1030 return; 1031 1032 if (!old_active) 1033 old_active = bond_get_old_active(bond, new_active); 1034 1035 if (old_active) { 1036 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr, 1037 new_active->dev->addr_len); 1038 bond_hw_addr_copy(ss.__data, 1039 old_active->dev->dev_addr, 1040 old_active->dev->addr_len); 1041 ss.ss_family = new_active->dev->type; 1042 } else { 1043 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr, 1044 bond->dev->addr_len); 1045 ss.ss_family = bond->dev->type; 1046 } 1047 1048 rv = dev_set_mac_address(new_active->dev, 1049 (struct sockaddr *)&ss, NULL); 1050 if (rv) { 1051 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n", 1052 -rv); 1053 goto out; 1054 } 1055 1056 if (!old_active) 1057 goto out; 1058 1059 bond_hw_addr_copy(ss.__data, tmp_mac, 1060 new_active->dev->addr_len); 1061 ss.ss_family = old_active->dev->type; 1062 1063 rv = dev_set_mac_address(old_active->dev, 1064 (struct sockaddr *)&ss, NULL); 1065 if (rv) 1066 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n", 1067 -rv); 1068 out: 1069 break; 1070 default: 1071 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n", 1072 bond->params.fail_over_mac); 1073 break; 1074 } 1075 1076 } 1077 1078 /** 1079 * bond_choose_primary_or_current - select the primary or high priority slave 1080 * @bond: our bonding struct 1081 * 1082 * - Check if there is a primary link. If the primary link was set and is up, 1083 * go on and do link reselection. 1084 * 1085 * - If primary link is not set or down, find the highest priority link. 1086 * If the highest priority link is not current slave, set it as primary 1087 * link and do link reselection. 1088 */ 1089 static struct slave *bond_choose_primary_or_current(struct bonding *bond) 1090 { 1091 struct slave *prim = rtnl_dereference(bond->primary_slave); 1092 struct slave *curr = rtnl_dereference(bond->curr_active_slave); 1093 struct slave *slave, *hprio = NULL; 1094 struct list_head *iter; 1095 1096 if (!prim || prim->link != BOND_LINK_UP) { 1097 bond_for_each_slave(bond, slave, iter) { 1098 if (slave->link == BOND_LINK_UP) { 1099 hprio = hprio ?: slave; 1100 if (slave->prio > hprio->prio) 1101 hprio = slave; 1102 } 1103 } 1104 1105 if (hprio && hprio != curr) { 1106 prim = hprio; 1107 goto link_reselect; 1108 } 1109 1110 if (!curr || curr->link != BOND_LINK_UP) 1111 return NULL; 1112 return curr; 1113 } 1114 1115 if (bond->force_primary) { 1116 bond->force_primary = false; 1117 return prim; 1118 } 1119 1120 link_reselect: 1121 if (!curr || curr->link != BOND_LINK_UP) 1122 return prim; 1123 1124 /* At this point, prim and curr are both up */ 1125 switch (bond->params.primary_reselect) { 1126 case BOND_PRI_RESELECT_ALWAYS: 1127 return prim; 1128 case BOND_PRI_RESELECT_BETTER: 1129 if (prim->speed < curr->speed) 1130 return curr; 1131 if (prim->speed == curr->speed && prim->duplex <= curr->duplex) 1132 return curr; 1133 return prim; 1134 case BOND_PRI_RESELECT_FAILURE: 1135 return curr; 1136 default: 1137 netdev_err(bond->dev, "impossible primary_reselect %d\n", 1138 bond->params.primary_reselect); 1139 return curr; 1140 } 1141 } 1142 1143 /** 1144 * bond_find_best_slave - select the best available slave to be the active one 1145 * @bond: our bonding struct 1146 */ 1147 static struct slave *bond_find_best_slave(struct bonding *bond) 1148 { 1149 struct slave *slave, *bestslave = NULL; 1150 struct list_head *iter; 1151 int mintime = bond->params.updelay; 1152 1153 slave = bond_choose_primary_or_current(bond); 1154 if (slave) 1155 return slave; 1156 1157 bond_for_each_slave(bond, slave, iter) { 1158 if (slave->link == BOND_LINK_UP) 1159 return slave; 1160 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) && 1161 slave->delay < mintime) { 1162 mintime = slave->delay; 1163 bestslave = slave; 1164 } 1165 } 1166 1167 return bestslave; 1168 } 1169 1170 /* must be called in RCU critical section or with RTNL held */ 1171 static bool bond_should_notify_peers(struct bonding *bond) 1172 { 1173 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave); 1174 1175 if (!slave || !bond->send_peer_notif || 1176 bond->send_peer_notif % 1177 max(1, bond->params.peer_notif_delay) != 0 || 1178 !netif_carrier_ok(bond->dev) || 1179 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state)) 1180 return false; 1181 1182 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n", 1183 slave ? slave->dev->name : "NULL"); 1184 1185 return true; 1186 } 1187 1188 /** 1189 * bond_change_active_slave - change the active slave into the specified one 1190 * @bond: our bonding struct 1191 * @new_active: the new slave to make the active one 1192 * 1193 * Set the new slave to the bond's settings and unset them on the old 1194 * curr_active_slave. 1195 * Setting include flags, mc-list, promiscuity, allmulti, etc. 1196 * 1197 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP, 1198 * because it is apparently the best available slave we have, even though its 1199 * updelay hasn't timed out yet. 1200 * 1201 * Caller must hold RTNL. 1202 */ 1203 void bond_change_active_slave(struct bonding *bond, struct slave *new_active) 1204 { 1205 struct slave *old_active; 1206 1207 ASSERT_RTNL(); 1208 1209 old_active = rtnl_dereference(bond->curr_active_slave); 1210 1211 if (old_active == new_active) 1212 return; 1213 1214 #ifdef CONFIG_XFRM_OFFLOAD 1215 bond_ipsec_del_sa_all(bond); 1216 #endif /* CONFIG_XFRM_OFFLOAD */ 1217 1218 if (new_active) { 1219 new_active->last_link_up = jiffies; 1220 1221 if (new_active->link == BOND_LINK_BACK) { 1222 if (bond_uses_primary(bond)) { 1223 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n", 1224 (bond->params.updelay - new_active->delay) * bond->params.miimon); 1225 } 1226 1227 new_active->delay = 0; 1228 bond_set_slave_link_state(new_active, BOND_LINK_UP, 1229 BOND_SLAVE_NOTIFY_NOW); 1230 1231 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1232 bond_3ad_handle_link_change(new_active, BOND_LINK_UP); 1233 1234 if (bond_is_lb(bond)) 1235 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP); 1236 } else { 1237 if (bond_uses_primary(bond)) 1238 slave_info(bond->dev, new_active->dev, "making interface the new active one\n"); 1239 } 1240 } 1241 1242 if (bond_uses_primary(bond)) 1243 bond_hw_addr_swap(bond, new_active, old_active); 1244 1245 if (bond_is_lb(bond)) { 1246 bond_alb_handle_active_change(bond, new_active); 1247 if (old_active) 1248 bond_set_slave_inactive_flags(old_active, 1249 BOND_SLAVE_NOTIFY_NOW); 1250 if (new_active) 1251 bond_set_slave_active_flags(new_active, 1252 BOND_SLAVE_NOTIFY_NOW); 1253 } else { 1254 rcu_assign_pointer(bond->curr_active_slave, new_active); 1255 } 1256 1257 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 1258 if (old_active) 1259 bond_set_slave_inactive_flags(old_active, 1260 BOND_SLAVE_NOTIFY_NOW); 1261 1262 if (new_active) { 1263 bool should_notify_peers = false; 1264 1265 bond_set_slave_active_flags(new_active, 1266 BOND_SLAVE_NOTIFY_NOW); 1267 1268 if (bond->params.fail_over_mac) 1269 bond_do_fail_over_mac(bond, new_active, 1270 old_active); 1271 1272 if (netif_running(bond->dev)) { 1273 bond->send_peer_notif = 1274 bond->params.num_peer_notif * 1275 max(1, bond->params.peer_notif_delay); 1276 should_notify_peers = 1277 bond_should_notify_peers(bond); 1278 } 1279 1280 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev); 1281 if (should_notify_peers) { 1282 bond->send_peer_notif--; 1283 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 1284 bond->dev); 1285 } 1286 } 1287 } 1288 1289 #ifdef CONFIG_XFRM_OFFLOAD 1290 bond_ipsec_add_sa_all(bond); 1291 #endif /* CONFIG_XFRM_OFFLOAD */ 1292 1293 /* resend IGMP joins since active slave has changed or 1294 * all were sent on curr_active_slave. 1295 * resend only if bond is brought up with the affected 1296 * bonding modes and the retransmission is enabled 1297 */ 1298 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) && 1299 ((bond_uses_primary(bond) && new_active) || 1300 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) { 1301 bond->igmp_retrans = bond->params.resend_igmp; 1302 queue_delayed_work(bond->wq, &bond->mcast_work, 1); 1303 } 1304 } 1305 1306 /** 1307 * bond_select_active_slave - select a new active slave, if needed 1308 * @bond: our bonding struct 1309 * 1310 * This functions should be called when one of the following occurs: 1311 * - The old curr_active_slave has been released or lost its link. 1312 * - The primary_slave has got its link back. 1313 * - A slave has got its link back and there's no old curr_active_slave. 1314 * 1315 * Caller must hold RTNL. 1316 */ 1317 void bond_select_active_slave(struct bonding *bond) 1318 { 1319 struct slave *best_slave; 1320 int rv; 1321 1322 ASSERT_RTNL(); 1323 1324 best_slave = bond_find_best_slave(bond); 1325 if (best_slave != rtnl_dereference(bond->curr_active_slave)) { 1326 bond_change_active_slave(bond, best_slave); 1327 rv = bond_set_carrier(bond); 1328 if (!rv) 1329 return; 1330 1331 if (netif_carrier_ok(bond->dev)) 1332 netdev_info(bond->dev, "active interface up!\n"); 1333 else 1334 netdev_info(bond->dev, "now running without any active interface!\n"); 1335 } 1336 } 1337 1338 #ifdef CONFIG_NET_POLL_CONTROLLER 1339 static inline int slave_enable_netpoll(struct slave *slave) 1340 { 1341 struct netpoll *np; 1342 int err = 0; 1343 1344 np = kzalloc(sizeof(*np), GFP_KERNEL); 1345 err = -ENOMEM; 1346 if (!np) 1347 goto out; 1348 1349 err = __netpoll_setup(np, slave->dev); 1350 if (err) { 1351 kfree(np); 1352 goto out; 1353 } 1354 slave->np = np; 1355 out: 1356 return err; 1357 } 1358 static inline void slave_disable_netpoll(struct slave *slave) 1359 { 1360 struct netpoll *np = slave->np; 1361 1362 if (!np) 1363 return; 1364 1365 slave->np = NULL; 1366 1367 __netpoll_free(np); 1368 } 1369 1370 static void bond_poll_controller(struct net_device *bond_dev) 1371 { 1372 struct bonding *bond = netdev_priv(bond_dev); 1373 struct slave *slave = NULL; 1374 struct list_head *iter; 1375 struct ad_info ad_info; 1376 1377 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1378 if (bond_3ad_get_active_agg_info(bond, &ad_info)) 1379 return; 1380 1381 bond_for_each_slave_rcu(bond, slave, iter) { 1382 if (!bond_slave_is_up(slave)) 1383 continue; 1384 1385 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1386 struct aggregator *agg = 1387 SLAVE_AD_INFO(slave)->port.aggregator; 1388 1389 if (agg && 1390 agg->aggregator_identifier != ad_info.aggregator_id) 1391 continue; 1392 } 1393 1394 netpoll_poll_dev(slave->dev); 1395 } 1396 } 1397 1398 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1399 { 1400 struct bonding *bond = netdev_priv(bond_dev); 1401 struct list_head *iter; 1402 struct slave *slave; 1403 1404 bond_for_each_slave(bond, slave, iter) 1405 if (bond_slave_is_up(slave)) 1406 slave_disable_netpoll(slave); 1407 } 1408 1409 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni) 1410 { 1411 struct bonding *bond = netdev_priv(dev); 1412 struct list_head *iter; 1413 struct slave *slave; 1414 int err = 0; 1415 1416 bond_for_each_slave(bond, slave, iter) { 1417 err = slave_enable_netpoll(slave); 1418 if (err) { 1419 bond_netpoll_cleanup(dev); 1420 break; 1421 } 1422 } 1423 return err; 1424 } 1425 #else 1426 static inline int slave_enable_netpoll(struct slave *slave) 1427 { 1428 return 0; 1429 } 1430 static inline void slave_disable_netpoll(struct slave *slave) 1431 { 1432 } 1433 static void bond_netpoll_cleanup(struct net_device *bond_dev) 1434 { 1435 } 1436 #endif 1437 1438 /*---------------------------------- IOCTL ----------------------------------*/ 1439 1440 static netdev_features_t bond_fix_features(struct net_device *dev, 1441 netdev_features_t features) 1442 { 1443 struct bonding *bond = netdev_priv(dev); 1444 struct list_head *iter; 1445 netdev_features_t mask; 1446 struct slave *slave; 1447 1448 mask = features; 1449 1450 features &= ~NETIF_F_ONE_FOR_ALL; 1451 features |= NETIF_F_ALL_FOR_ALL; 1452 1453 bond_for_each_slave(bond, slave, iter) { 1454 features = netdev_increment_features(features, 1455 slave->dev->features, 1456 mask); 1457 } 1458 features = netdev_add_tso_features(features, mask); 1459 1460 return features; 1461 } 1462 1463 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1464 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \ 1465 NETIF_F_GSO_ENCAP_ALL | \ 1466 NETIF_F_HIGHDMA | NETIF_F_LRO) 1467 1468 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1469 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE) 1470 1471 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \ 1472 NETIF_F_GSO_SOFTWARE) 1473 1474 1475 static void bond_compute_features(struct bonding *bond) 1476 { 1477 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE | 1478 IFF_XMIT_DST_RELEASE_PERM; 1479 netdev_features_t vlan_features = BOND_VLAN_FEATURES; 1480 netdev_features_t enc_features = BOND_ENC_FEATURES; 1481 #ifdef CONFIG_XFRM_OFFLOAD 1482 netdev_features_t xfrm_features = BOND_XFRM_FEATURES; 1483 #endif /* CONFIG_XFRM_OFFLOAD */ 1484 netdev_features_t mpls_features = BOND_MPLS_FEATURES; 1485 struct net_device *bond_dev = bond->dev; 1486 struct list_head *iter; 1487 struct slave *slave; 1488 unsigned short max_hard_header_len = ETH_HLEN; 1489 unsigned int tso_max_size = TSO_MAX_SIZE; 1490 u16 tso_max_segs = TSO_MAX_SEGS; 1491 1492 if (!bond_has_slaves(bond)) 1493 goto done; 1494 vlan_features &= NETIF_F_ALL_FOR_ALL; 1495 mpls_features &= NETIF_F_ALL_FOR_ALL; 1496 1497 bond_for_each_slave(bond, slave, iter) { 1498 vlan_features = netdev_increment_features(vlan_features, 1499 slave->dev->vlan_features, BOND_VLAN_FEATURES); 1500 1501 enc_features = netdev_increment_features(enc_features, 1502 slave->dev->hw_enc_features, 1503 BOND_ENC_FEATURES); 1504 1505 #ifdef CONFIG_XFRM_OFFLOAD 1506 xfrm_features = netdev_increment_features(xfrm_features, 1507 slave->dev->hw_enc_features, 1508 BOND_XFRM_FEATURES); 1509 #endif /* CONFIG_XFRM_OFFLOAD */ 1510 1511 mpls_features = netdev_increment_features(mpls_features, 1512 slave->dev->mpls_features, 1513 BOND_MPLS_FEATURES); 1514 1515 dst_release_flag &= slave->dev->priv_flags; 1516 if (slave->dev->hard_header_len > max_hard_header_len) 1517 max_hard_header_len = slave->dev->hard_header_len; 1518 1519 tso_max_size = min(tso_max_size, slave->dev->tso_max_size); 1520 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs); 1521 } 1522 bond_dev->hard_header_len = max_hard_header_len; 1523 1524 done: 1525 bond_dev->vlan_features = vlan_features; 1526 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL | 1527 NETIF_F_HW_VLAN_CTAG_TX | 1528 NETIF_F_HW_VLAN_STAG_TX; 1529 #ifdef CONFIG_XFRM_OFFLOAD 1530 bond_dev->hw_enc_features |= xfrm_features; 1531 #endif /* CONFIG_XFRM_OFFLOAD */ 1532 bond_dev->mpls_features = mpls_features; 1533 netif_set_tso_max_segs(bond_dev, tso_max_segs); 1534 netif_set_tso_max_size(bond_dev, tso_max_size); 1535 1536 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; 1537 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) && 1538 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM)) 1539 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE; 1540 1541 netdev_change_features(bond_dev); 1542 } 1543 1544 static void bond_setup_by_slave(struct net_device *bond_dev, 1545 struct net_device *slave_dev) 1546 { 1547 bool was_up = !!(bond_dev->flags & IFF_UP); 1548 1549 dev_close(bond_dev); 1550 1551 bond_dev->header_ops = slave_dev->header_ops; 1552 1553 bond_dev->type = slave_dev->type; 1554 bond_dev->hard_header_len = slave_dev->hard_header_len; 1555 bond_dev->needed_headroom = slave_dev->needed_headroom; 1556 bond_dev->addr_len = slave_dev->addr_len; 1557 1558 memcpy(bond_dev->broadcast, slave_dev->broadcast, 1559 slave_dev->addr_len); 1560 1561 if (slave_dev->flags & IFF_POINTOPOINT) { 1562 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST); 1563 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP); 1564 } 1565 if (was_up) 1566 dev_open(bond_dev, NULL); 1567 } 1568 1569 /* On bonding slaves other than the currently active slave, suppress 1570 * duplicates except for alb non-mcast/bcast. 1571 */ 1572 static bool bond_should_deliver_exact_match(struct sk_buff *skb, 1573 struct slave *slave, 1574 struct bonding *bond) 1575 { 1576 if (bond_is_slave_inactive(slave)) { 1577 if (BOND_MODE(bond) == BOND_MODE_ALB && 1578 skb->pkt_type != PACKET_BROADCAST && 1579 skb->pkt_type != PACKET_MULTICAST) 1580 return false; 1581 return true; 1582 } 1583 return false; 1584 } 1585 1586 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb) 1587 { 1588 struct sk_buff *skb = *pskb; 1589 struct slave *slave; 1590 struct bonding *bond; 1591 int (*recv_probe)(const struct sk_buff *, struct bonding *, 1592 struct slave *); 1593 int ret = RX_HANDLER_ANOTHER; 1594 1595 skb = skb_share_check(skb, GFP_ATOMIC); 1596 if (unlikely(!skb)) 1597 return RX_HANDLER_CONSUMED; 1598 1599 *pskb = skb; 1600 1601 slave = bond_slave_get_rcu(skb->dev); 1602 bond = slave->bond; 1603 1604 recv_probe = READ_ONCE(bond->recv_probe); 1605 if (recv_probe) { 1606 ret = recv_probe(skb, bond, slave); 1607 if (ret == RX_HANDLER_CONSUMED) { 1608 consume_skb(skb); 1609 return ret; 1610 } 1611 } 1612 1613 /* 1614 * For packets determined by bond_should_deliver_exact_match() call to 1615 * be suppressed we want to make an exception for link-local packets. 1616 * This is necessary for e.g. LLDP daemons to be able to monitor 1617 * inactive slave links without being forced to bind to them 1618 * explicitly. 1619 * 1620 * At the same time, packets that are passed to the bonding master 1621 * (including link-local ones) can have their originating interface 1622 * determined via PACKET_ORIGDEV socket option. 1623 */ 1624 if (bond_should_deliver_exact_match(skb, slave, bond)) { 1625 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) 1626 return RX_HANDLER_PASS; 1627 return RX_HANDLER_EXACT; 1628 } 1629 1630 skb->dev = bond->dev; 1631 1632 if (BOND_MODE(bond) == BOND_MODE_ALB && 1633 netif_is_bridge_port(bond->dev) && 1634 skb->pkt_type == PACKET_HOST) { 1635 1636 if (unlikely(skb_cow_head(skb, 1637 skb->data - skb_mac_header(skb)))) { 1638 kfree_skb(skb); 1639 return RX_HANDLER_CONSUMED; 1640 } 1641 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, 1642 bond->dev->addr_len); 1643 } 1644 1645 return ret; 1646 } 1647 1648 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond) 1649 { 1650 switch (BOND_MODE(bond)) { 1651 case BOND_MODE_ROUNDROBIN: 1652 return NETDEV_LAG_TX_TYPE_ROUNDROBIN; 1653 case BOND_MODE_ACTIVEBACKUP: 1654 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP; 1655 case BOND_MODE_BROADCAST: 1656 return NETDEV_LAG_TX_TYPE_BROADCAST; 1657 case BOND_MODE_XOR: 1658 case BOND_MODE_8023AD: 1659 return NETDEV_LAG_TX_TYPE_HASH; 1660 default: 1661 return NETDEV_LAG_TX_TYPE_UNKNOWN; 1662 } 1663 } 1664 1665 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond, 1666 enum netdev_lag_tx_type type) 1667 { 1668 if (type != NETDEV_LAG_TX_TYPE_HASH) 1669 return NETDEV_LAG_HASH_NONE; 1670 1671 switch (bond->params.xmit_policy) { 1672 case BOND_XMIT_POLICY_LAYER2: 1673 return NETDEV_LAG_HASH_L2; 1674 case BOND_XMIT_POLICY_LAYER34: 1675 return NETDEV_LAG_HASH_L34; 1676 case BOND_XMIT_POLICY_LAYER23: 1677 return NETDEV_LAG_HASH_L23; 1678 case BOND_XMIT_POLICY_ENCAP23: 1679 return NETDEV_LAG_HASH_E23; 1680 case BOND_XMIT_POLICY_ENCAP34: 1681 return NETDEV_LAG_HASH_E34; 1682 case BOND_XMIT_POLICY_VLAN_SRCMAC: 1683 return NETDEV_LAG_HASH_VLAN_SRCMAC; 1684 default: 1685 return NETDEV_LAG_HASH_UNKNOWN; 1686 } 1687 } 1688 1689 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave, 1690 struct netlink_ext_ack *extack) 1691 { 1692 struct netdev_lag_upper_info lag_upper_info; 1693 enum netdev_lag_tx_type type; 1694 int err; 1695 1696 type = bond_lag_tx_type(bond); 1697 lag_upper_info.tx_type = type; 1698 lag_upper_info.hash_type = bond_lag_hash_type(bond, type); 1699 1700 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave, 1701 &lag_upper_info, extack); 1702 if (err) 1703 return err; 1704 1705 slave->dev->flags |= IFF_SLAVE; 1706 return 0; 1707 } 1708 1709 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave) 1710 { 1711 netdev_upper_dev_unlink(slave->dev, bond->dev); 1712 slave->dev->flags &= ~IFF_SLAVE; 1713 } 1714 1715 static void slave_kobj_release(struct kobject *kobj) 1716 { 1717 struct slave *slave = to_slave(kobj); 1718 struct bonding *bond = bond_get_bond_by_slave(slave); 1719 1720 cancel_delayed_work_sync(&slave->notify_work); 1721 if (BOND_MODE(bond) == BOND_MODE_8023AD) 1722 kfree(SLAVE_AD_INFO(slave)); 1723 1724 kfree(slave); 1725 } 1726 1727 static struct kobj_type slave_ktype = { 1728 .release = slave_kobj_release, 1729 #ifdef CONFIG_SYSFS 1730 .sysfs_ops = &slave_sysfs_ops, 1731 #endif 1732 }; 1733 1734 static int bond_kobj_init(struct slave *slave) 1735 { 1736 int err; 1737 1738 err = kobject_init_and_add(&slave->kobj, &slave_ktype, 1739 &(slave->dev->dev.kobj), "bonding_slave"); 1740 if (err) 1741 kobject_put(&slave->kobj); 1742 1743 return err; 1744 } 1745 1746 static struct slave *bond_alloc_slave(struct bonding *bond, 1747 struct net_device *slave_dev) 1748 { 1749 struct slave *slave = NULL; 1750 1751 slave = kzalloc(sizeof(*slave), GFP_KERNEL); 1752 if (!slave) 1753 return NULL; 1754 1755 slave->bond = bond; 1756 slave->dev = slave_dev; 1757 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work); 1758 1759 if (bond_kobj_init(slave)) 1760 return NULL; 1761 1762 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 1763 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info), 1764 GFP_KERNEL); 1765 if (!SLAVE_AD_INFO(slave)) { 1766 kobject_put(&slave->kobj); 1767 return NULL; 1768 } 1769 } 1770 1771 return slave; 1772 } 1773 1774 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info) 1775 { 1776 info->bond_mode = BOND_MODE(bond); 1777 info->miimon = bond->params.miimon; 1778 info->num_slaves = bond->slave_cnt; 1779 } 1780 1781 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info) 1782 { 1783 strcpy(info->slave_name, slave->dev->name); 1784 info->link = slave->link; 1785 info->state = bond_slave_state(slave); 1786 info->link_failure_count = slave->link_failure_count; 1787 } 1788 1789 static void bond_netdev_notify_work(struct work_struct *_work) 1790 { 1791 struct slave *slave = container_of(_work, struct slave, 1792 notify_work.work); 1793 1794 if (rtnl_trylock()) { 1795 struct netdev_bonding_info binfo; 1796 1797 bond_fill_ifslave(slave, &binfo.slave); 1798 bond_fill_ifbond(slave->bond, &binfo.master); 1799 netdev_bonding_info_change(slave->dev, &binfo); 1800 rtnl_unlock(); 1801 } else { 1802 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1); 1803 } 1804 } 1805 1806 void bond_queue_slave_event(struct slave *slave) 1807 { 1808 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0); 1809 } 1810 1811 void bond_lower_state_changed(struct slave *slave) 1812 { 1813 struct netdev_lag_lower_state_info info; 1814 1815 info.link_up = slave->link == BOND_LINK_UP || 1816 slave->link == BOND_LINK_FAIL; 1817 info.tx_enabled = bond_is_active_slave(slave); 1818 netdev_lower_state_changed(slave->dev, &info); 1819 } 1820 1821 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \ 1822 if (extack) \ 1823 NL_SET_ERR_MSG(extack, errmsg); \ 1824 else \ 1825 netdev_err(bond_dev, "Error: %s\n", errmsg); \ 1826 } while (0) 1827 1828 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \ 1829 if (extack) \ 1830 NL_SET_ERR_MSG(extack, errmsg); \ 1831 else \ 1832 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \ 1833 } while (0) 1834 1835 /* The bonding driver uses ether_setup() to convert a master bond device 1836 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always 1837 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP 1838 * if they were set 1839 */ 1840 static void bond_ether_setup(struct net_device *bond_dev) 1841 { 1842 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP); 1843 1844 ether_setup(bond_dev); 1845 bond_dev->flags |= IFF_MASTER | flags; 1846 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1847 } 1848 1849 void bond_xdp_set_features(struct net_device *bond_dev) 1850 { 1851 struct bonding *bond = netdev_priv(bond_dev); 1852 xdp_features_t val = NETDEV_XDP_ACT_MASK; 1853 struct list_head *iter; 1854 struct slave *slave; 1855 1856 ASSERT_RTNL(); 1857 1858 if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) { 1859 xdp_clear_features_flag(bond_dev); 1860 return; 1861 } 1862 1863 bond_for_each_slave(bond, slave, iter) 1864 val &= slave->dev->xdp_features; 1865 1866 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY; 1867 1868 xdp_set_features_flag(bond_dev, val); 1869 } 1870 1871 /* enslave device <slave> to bond device <master> */ 1872 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev, 1873 struct netlink_ext_ack *extack) 1874 { 1875 struct bonding *bond = netdev_priv(bond_dev); 1876 const struct net_device_ops *slave_ops = slave_dev->netdev_ops; 1877 struct slave *new_slave = NULL, *prev_slave; 1878 struct sockaddr_storage ss; 1879 int link_reporting; 1880 int res = 0, i; 1881 1882 if (slave_dev->flags & IFF_MASTER && 1883 !netif_is_bond_master(slave_dev)) { 1884 BOND_NL_ERR(bond_dev, extack, 1885 "Device type (master device) cannot be enslaved"); 1886 return -EPERM; 1887 } 1888 1889 if (!bond->params.use_carrier && 1890 slave_dev->ethtool_ops->get_link == NULL && 1891 slave_ops->ndo_eth_ioctl == NULL) { 1892 slave_warn(bond_dev, slave_dev, "no link monitoring support\n"); 1893 } 1894 1895 /* already in-use? */ 1896 if (netdev_is_rx_handler_busy(slave_dev)) { 1897 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1898 "Device is in use and cannot be enslaved"); 1899 return -EBUSY; 1900 } 1901 1902 if (bond_dev == slave_dev) { 1903 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself."); 1904 return -EPERM; 1905 } 1906 1907 /* vlan challenged mutual exclusion */ 1908 /* no need to lock since we're protected by rtnl_lock */ 1909 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) { 1910 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n"); 1911 if (vlan_uses_dev(bond_dev)) { 1912 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1913 "Can not enslave VLAN challenged device to VLAN enabled bond"); 1914 return -EPERM; 1915 } else { 1916 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n"); 1917 } 1918 } else { 1919 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n"); 1920 } 1921 1922 if (slave_dev->features & NETIF_F_HW_ESP) 1923 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n"); 1924 1925 /* Old ifenslave binaries are no longer supported. These can 1926 * be identified with moderate accuracy by the state of the slave: 1927 * the current ifenslave will set the interface down prior to 1928 * enslaving it; the old ifenslave will not. 1929 */ 1930 if (slave_dev->flags & IFF_UP) { 1931 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1932 "Device can not be enslaved while up"); 1933 return -EPERM; 1934 } 1935 1936 /* set bonding device ether type by slave - bonding netdevices are 1937 * created with ether_setup, so when the slave type is not ARPHRD_ETHER 1938 * there is a need to override some of the type dependent attribs/funcs. 1939 * 1940 * bond ether type mutual exclusion - don't allow slaves of dissimilar 1941 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond 1942 */ 1943 if (!bond_has_slaves(bond)) { 1944 if (bond_dev->type != slave_dev->type) { 1945 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n", 1946 bond_dev->type, slave_dev->type); 1947 1948 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, 1949 bond_dev); 1950 res = notifier_to_errno(res); 1951 if (res) { 1952 slave_err(bond_dev, slave_dev, "refused to change device type\n"); 1953 return -EBUSY; 1954 } 1955 1956 /* Flush unicast and multicast addresses */ 1957 dev_uc_flush(bond_dev); 1958 dev_mc_flush(bond_dev); 1959 1960 if (slave_dev->type != ARPHRD_ETHER) 1961 bond_setup_by_slave(bond_dev, slave_dev); 1962 else 1963 bond_ether_setup(bond_dev); 1964 1965 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, 1966 bond_dev); 1967 } 1968 } else if (bond_dev->type != slave_dev->type) { 1969 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1970 "Device type is different from other slaves"); 1971 return -EINVAL; 1972 } 1973 1974 if (slave_dev->type == ARPHRD_INFINIBAND && 1975 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 1976 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1977 "Only active-backup mode is supported for infiniband slaves"); 1978 res = -EOPNOTSUPP; 1979 goto err_undo_flags; 1980 } 1981 1982 if (!slave_ops->ndo_set_mac_address || 1983 slave_dev->type == ARPHRD_INFINIBAND) { 1984 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n"); 1985 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP && 1986 bond->params.fail_over_mac != BOND_FOM_ACTIVE) { 1987 if (!bond_has_slaves(bond)) { 1988 bond->params.fail_over_mac = BOND_FOM_ACTIVE; 1989 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n"); 1990 } else { 1991 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 1992 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active"); 1993 res = -EOPNOTSUPP; 1994 goto err_undo_flags; 1995 } 1996 } 1997 } 1998 1999 call_netdevice_notifiers(NETDEV_JOIN, slave_dev); 2000 2001 /* If this is the first slave, then we need to set the master's hardware 2002 * address to be the same as the slave's. 2003 */ 2004 if (!bond_has_slaves(bond) && 2005 bond->dev->addr_assign_type == NET_ADDR_RANDOM) { 2006 res = bond_set_dev_addr(bond->dev, slave_dev); 2007 if (res) 2008 goto err_undo_flags; 2009 } 2010 2011 new_slave = bond_alloc_slave(bond, slave_dev); 2012 if (!new_slave) { 2013 res = -ENOMEM; 2014 goto err_undo_flags; 2015 } 2016 2017 /* Set the new_slave's queue_id to be zero. Queue ID mapping 2018 * is set via sysfs or module option if desired. 2019 */ 2020 new_slave->queue_id = 0; 2021 2022 /* Save slave's original mtu and then set it to match the bond */ 2023 new_slave->original_mtu = slave_dev->mtu; 2024 res = dev_set_mtu(slave_dev, bond->dev->mtu); 2025 if (res) { 2026 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res); 2027 goto err_free; 2028 } 2029 2030 /* Save slave's original ("permanent") mac address for modes 2031 * that need it, and for restoring it upon release, and then 2032 * set it to the master's address 2033 */ 2034 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr, 2035 slave_dev->addr_len); 2036 2037 if (!bond->params.fail_over_mac || 2038 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2039 /* Set slave to master's mac address. The application already 2040 * set the master's mac address to that of the first slave 2041 */ 2042 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 2043 ss.ss_family = slave_dev->type; 2044 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, 2045 extack); 2046 if (res) { 2047 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res); 2048 goto err_restore_mtu; 2049 } 2050 } 2051 2052 /* set no_addrconf flag before open to prevent IPv6 addrconf */ 2053 slave_dev->priv_flags |= IFF_NO_ADDRCONF; 2054 2055 /* open the slave since the application closed it */ 2056 res = dev_open(slave_dev, extack); 2057 if (res) { 2058 slave_err(bond_dev, slave_dev, "Opening slave failed\n"); 2059 goto err_restore_mac; 2060 } 2061 2062 slave_dev->priv_flags |= IFF_BONDING; 2063 /* initialize slave stats */ 2064 dev_get_stats(new_slave->dev, &new_slave->slave_stats); 2065 2066 if (bond_is_lb(bond)) { 2067 /* bond_alb_init_slave() must be called before all other stages since 2068 * it might fail and we do not want to have to undo everything 2069 */ 2070 res = bond_alb_init_slave(bond, new_slave); 2071 if (res) 2072 goto err_close; 2073 } 2074 2075 res = vlan_vids_add_by_dev(slave_dev, bond_dev); 2076 if (res) { 2077 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n"); 2078 goto err_close; 2079 } 2080 2081 prev_slave = bond_last_slave(bond); 2082 2083 new_slave->delay = 0; 2084 new_slave->link_failure_count = 0; 2085 2086 if (bond_update_speed_duplex(new_slave) && 2087 bond_needs_speed_duplex(bond)) 2088 new_slave->link = BOND_LINK_DOWN; 2089 2090 new_slave->last_rx = jiffies - 2091 (msecs_to_jiffies(bond->params.arp_interval) + 1); 2092 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 2093 new_slave->target_last_arp_rx[i] = new_slave->last_rx; 2094 2095 new_slave->last_tx = new_slave->last_rx; 2096 2097 if (bond->params.miimon && !bond->params.use_carrier) { 2098 link_reporting = bond_check_dev_link(bond, slave_dev, 1); 2099 2100 if ((link_reporting == -1) && !bond->params.arp_interval) { 2101 /* miimon is set but a bonded network driver 2102 * does not support ETHTOOL/MII and 2103 * arp_interval is not set. Note: if 2104 * use_carrier is enabled, we will never go 2105 * here (because netif_carrier is always 2106 * supported); thus, we don't need to change 2107 * the messages for netif_carrier. 2108 */ 2109 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"); 2110 } else if (link_reporting == -1) { 2111 /* unable get link status using mii/ethtool */ 2112 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"); 2113 } 2114 } 2115 2116 /* check for initial state */ 2117 new_slave->link = BOND_LINK_NOCHANGE; 2118 if (bond->params.miimon) { 2119 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) { 2120 if (bond->params.updelay) { 2121 bond_set_slave_link_state(new_slave, 2122 BOND_LINK_BACK, 2123 BOND_SLAVE_NOTIFY_NOW); 2124 new_slave->delay = bond->params.updelay; 2125 } else { 2126 bond_set_slave_link_state(new_slave, 2127 BOND_LINK_UP, 2128 BOND_SLAVE_NOTIFY_NOW); 2129 } 2130 } else { 2131 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN, 2132 BOND_SLAVE_NOTIFY_NOW); 2133 } 2134 } else if (bond->params.arp_interval) { 2135 bond_set_slave_link_state(new_slave, 2136 (netif_carrier_ok(slave_dev) ? 2137 BOND_LINK_UP : BOND_LINK_DOWN), 2138 BOND_SLAVE_NOTIFY_NOW); 2139 } else { 2140 bond_set_slave_link_state(new_slave, BOND_LINK_UP, 2141 BOND_SLAVE_NOTIFY_NOW); 2142 } 2143 2144 if (new_slave->link != BOND_LINK_DOWN) 2145 new_slave->last_link_up = jiffies; 2146 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n", 2147 new_slave->link == BOND_LINK_DOWN ? "DOWN" : 2148 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK")); 2149 2150 if (bond_uses_primary(bond) && bond->params.primary[0]) { 2151 /* if there is a primary slave, remember it */ 2152 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) { 2153 rcu_assign_pointer(bond->primary_slave, new_slave); 2154 bond->force_primary = true; 2155 } 2156 } 2157 2158 switch (BOND_MODE(bond)) { 2159 case BOND_MODE_ACTIVEBACKUP: 2160 bond_set_slave_inactive_flags(new_slave, 2161 BOND_SLAVE_NOTIFY_NOW); 2162 break; 2163 case BOND_MODE_8023AD: 2164 /* in 802.3ad mode, the internal mechanism 2165 * will activate the slaves in the selected 2166 * aggregator 2167 */ 2168 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2169 /* if this is the first slave */ 2170 if (!prev_slave) { 2171 SLAVE_AD_INFO(new_slave)->id = 1; 2172 /* Initialize AD with the number of times that the AD timer is called in 1 second 2173 * can be called only after the mac address of the bond is set 2174 */ 2175 bond_3ad_initialize(bond); 2176 } else { 2177 SLAVE_AD_INFO(new_slave)->id = 2178 SLAVE_AD_INFO(prev_slave)->id + 1; 2179 } 2180 2181 bond_3ad_bind_slave(new_slave); 2182 break; 2183 case BOND_MODE_TLB: 2184 case BOND_MODE_ALB: 2185 bond_set_active_slave(new_slave); 2186 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW); 2187 break; 2188 default: 2189 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n"); 2190 2191 /* always active in trunk mode */ 2192 bond_set_active_slave(new_slave); 2193 2194 /* In trunking mode there is little meaning to curr_active_slave 2195 * anyway (it holds no special properties of the bond device), 2196 * so we can change it without calling change_active_interface() 2197 */ 2198 if (!rcu_access_pointer(bond->curr_active_slave) && 2199 new_slave->link == BOND_LINK_UP) 2200 rcu_assign_pointer(bond->curr_active_slave, new_slave); 2201 2202 break; 2203 } /* switch(bond_mode) */ 2204 2205 #ifdef CONFIG_NET_POLL_CONTROLLER 2206 if (bond->dev->npinfo) { 2207 if (slave_enable_netpoll(new_slave)) { 2208 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n"); 2209 res = -EBUSY; 2210 goto err_detach; 2211 } 2212 } 2213 #endif 2214 2215 if (!(bond_dev->features & NETIF_F_LRO)) 2216 dev_disable_lro(slave_dev); 2217 2218 res = netdev_rx_handler_register(slave_dev, bond_handle_frame, 2219 new_slave); 2220 if (res) { 2221 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res); 2222 goto err_detach; 2223 } 2224 2225 res = bond_master_upper_dev_link(bond, new_slave, extack); 2226 if (res) { 2227 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res); 2228 goto err_unregister; 2229 } 2230 2231 bond_lower_state_changed(new_slave); 2232 2233 res = bond_sysfs_slave_add(new_slave); 2234 if (res) { 2235 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res); 2236 goto err_upper_unlink; 2237 } 2238 2239 /* If the mode uses primary, then the following is handled by 2240 * bond_change_active_slave(). 2241 */ 2242 if (!bond_uses_primary(bond)) { 2243 /* set promiscuity level to new slave */ 2244 if (bond_dev->flags & IFF_PROMISC) { 2245 res = dev_set_promiscuity(slave_dev, 1); 2246 if (res) 2247 goto err_sysfs_del; 2248 } 2249 2250 /* set allmulti level to new slave */ 2251 if (bond_dev->flags & IFF_ALLMULTI) { 2252 res = dev_set_allmulti(slave_dev, 1); 2253 if (res) { 2254 if (bond_dev->flags & IFF_PROMISC) 2255 dev_set_promiscuity(slave_dev, -1); 2256 goto err_sysfs_del; 2257 } 2258 } 2259 2260 if (bond_dev->flags & IFF_UP) { 2261 netif_addr_lock_bh(bond_dev); 2262 dev_mc_sync_multiple(slave_dev, bond_dev); 2263 dev_uc_sync_multiple(slave_dev, bond_dev); 2264 netif_addr_unlock_bh(bond_dev); 2265 2266 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2267 dev_mc_add(slave_dev, lacpdu_mcast_addr); 2268 } 2269 } 2270 2271 bond->slave_cnt++; 2272 bond_compute_features(bond); 2273 bond_set_carrier(bond); 2274 2275 /* Needs to be called before bond_select_active_slave(), which will 2276 * remove the maddrs if the slave is selected as active slave. 2277 */ 2278 bond_slave_ns_maddrs_add(bond, new_slave); 2279 2280 if (bond_uses_primary(bond)) { 2281 block_netpoll_tx(); 2282 bond_select_active_slave(bond); 2283 unblock_netpoll_tx(); 2284 } 2285 2286 if (bond_mode_can_use_xmit_hash(bond)) 2287 bond_update_slave_arr(bond, NULL); 2288 2289 if (!slave_dev->netdev_ops->ndo_bpf || 2290 !slave_dev->netdev_ops->ndo_xdp_xmit) { 2291 if (bond->xdp_prog) { 2292 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2293 "Slave does not support XDP"); 2294 res = -EOPNOTSUPP; 2295 goto err_sysfs_del; 2296 } 2297 } else if (bond->xdp_prog) { 2298 struct netdev_bpf xdp = { 2299 .command = XDP_SETUP_PROG, 2300 .flags = 0, 2301 .prog = bond->xdp_prog, 2302 .extack = extack, 2303 }; 2304 2305 if (dev_xdp_prog_count(slave_dev) > 0) { 2306 SLAVE_NL_ERR(bond_dev, slave_dev, extack, 2307 "Slave has XDP program loaded, please unload before enslaving"); 2308 res = -EOPNOTSUPP; 2309 goto err_sysfs_del; 2310 } 2311 2312 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 2313 if (res < 0) { 2314 /* ndo_bpf() sets extack error message */ 2315 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res); 2316 goto err_sysfs_del; 2317 } 2318 if (bond->xdp_prog) 2319 bpf_prog_inc(bond->xdp_prog); 2320 } 2321 2322 bond_xdp_set_features(bond_dev); 2323 2324 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n", 2325 bond_is_active_slave(new_slave) ? "an active" : "a backup", 2326 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down"); 2327 2328 /* enslave is successful */ 2329 bond_queue_slave_event(new_slave); 2330 return 0; 2331 2332 /* Undo stages on error */ 2333 err_sysfs_del: 2334 bond_sysfs_slave_del(new_slave); 2335 2336 err_upper_unlink: 2337 bond_upper_dev_unlink(bond, new_slave); 2338 2339 err_unregister: 2340 netdev_rx_handler_unregister(slave_dev); 2341 2342 err_detach: 2343 vlan_vids_del_by_dev(slave_dev, bond_dev); 2344 if (rcu_access_pointer(bond->primary_slave) == new_slave) 2345 RCU_INIT_POINTER(bond->primary_slave, NULL); 2346 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) { 2347 block_netpoll_tx(); 2348 bond_change_active_slave(bond, NULL); 2349 bond_select_active_slave(bond); 2350 unblock_netpoll_tx(); 2351 } 2352 /* either primary_slave or curr_active_slave might've changed */ 2353 synchronize_rcu(); 2354 slave_disable_netpoll(new_slave); 2355 2356 err_close: 2357 if (!netif_is_bond_master(slave_dev)) 2358 slave_dev->priv_flags &= ~IFF_BONDING; 2359 dev_close(slave_dev); 2360 2361 err_restore_mac: 2362 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2363 if (!bond->params.fail_over_mac || 2364 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2365 /* XXX TODO - fom follow mode needs to change master's 2366 * MAC if this slave's MAC is in use by the bond, or at 2367 * least print a warning. 2368 */ 2369 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr, 2370 new_slave->dev->addr_len); 2371 ss.ss_family = slave_dev->type; 2372 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2373 } 2374 2375 err_restore_mtu: 2376 dev_set_mtu(slave_dev, new_slave->original_mtu); 2377 2378 err_free: 2379 kobject_put(&new_slave->kobj); 2380 2381 err_undo_flags: 2382 /* Enslave of first slave has failed and we need to fix master's mac */ 2383 if (!bond_has_slaves(bond)) { 2384 if (ether_addr_equal_64bits(bond_dev->dev_addr, 2385 slave_dev->dev_addr)) 2386 eth_hw_addr_random(bond_dev); 2387 if (bond_dev->type != ARPHRD_ETHER) { 2388 dev_close(bond_dev); 2389 bond_ether_setup(bond_dev); 2390 } 2391 } 2392 2393 return res; 2394 } 2395 2396 /* Try to release the slave device <slave> from the bond device <master> 2397 * It is legal to access curr_active_slave without a lock because all the function 2398 * is RTNL-locked. If "all" is true it means that the function is being called 2399 * while destroying a bond interface and all slaves are being released. 2400 * 2401 * The rules for slave state should be: 2402 * for Active/Backup: 2403 * Active stays on all backups go down 2404 * for Bonded connections: 2405 * The first up interface should be left on and all others downed. 2406 */ 2407 static int __bond_release_one(struct net_device *bond_dev, 2408 struct net_device *slave_dev, 2409 bool all, bool unregister) 2410 { 2411 struct bonding *bond = netdev_priv(bond_dev); 2412 struct slave *slave, *oldcurrent; 2413 struct sockaddr_storage ss; 2414 int old_flags = bond_dev->flags; 2415 netdev_features_t old_features = bond_dev->features; 2416 2417 /* slave is not a slave or master is not master of this slave */ 2418 if (!(slave_dev->flags & IFF_SLAVE) || 2419 !netdev_has_upper_dev(slave_dev, bond_dev)) { 2420 slave_dbg(bond_dev, slave_dev, "cannot release slave\n"); 2421 return -EINVAL; 2422 } 2423 2424 block_netpoll_tx(); 2425 2426 slave = bond_get_slave_by_dev(bond, slave_dev); 2427 if (!slave) { 2428 /* not a slave of this bond */ 2429 slave_info(bond_dev, slave_dev, "interface not enslaved\n"); 2430 unblock_netpoll_tx(); 2431 return -EINVAL; 2432 } 2433 2434 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW); 2435 2436 bond_sysfs_slave_del(slave); 2437 2438 /* recompute stats just before removing the slave */ 2439 bond_get_stats(bond->dev, &bond->bond_stats); 2440 2441 if (bond->xdp_prog) { 2442 struct netdev_bpf xdp = { 2443 .command = XDP_SETUP_PROG, 2444 .flags = 0, 2445 .prog = NULL, 2446 .extack = NULL, 2447 }; 2448 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp)) 2449 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n"); 2450 } 2451 2452 /* unregister rx_handler early so bond_handle_frame wouldn't be called 2453 * for this slave anymore. 2454 */ 2455 netdev_rx_handler_unregister(slave_dev); 2456 2457 if (BOND_MODE(bond) == BOND_MODE_8023AD) 2458 bond_3ad_unbind_slave(slave); 2459 2460 bond_upper_dev_unlink(bond, slave); 2461 2462 if (bond_mode_can_use_xmit_hash(bond)) 2463 bond_update_slave_arr(bond, slave); 2464 2465 slave_info(bond_dev, slave_dev, "Releasing %s interface\n", 2466 bond_is_active_slave(slave) ? "active" : "backup"); 2467 2468 oldcurrent = rcu_access_pointer(bond->curr_active_slave); 2469 2470 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 2471 2472 if (!all && (!bond->params.fail_over_mac || 2473 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) { 2474 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) && 2475 bond_has_slaves(bond)) 2476 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", 2477 slave->perm_hwaddr); 2478 } 2479 2480 if (rtnl_dereference(bond->primary_slave) == slave) 2481 RCU_INIT_POINTER(bond->primary_slave, NULL); 2482 2483 if (oldcurrent == slave) 2484 bond_change_active_slave(bond, NULL); 2485 2486 /* Must be called after bond_change_active_slave () as the slave 2487 * might change from an active slave to a backup slave. Then it is 2488 * necessary to clear the maddrs on the backup slave. 2489 */ 2490 bond_slave_ns_maddrs_del(bond, slave); 2491 2492 if (bond_is_lb(bond)) { 2493 /* Must be called only after the slave has been 2494 * detached from the list and the curr_active_slave 2495 * has been cleared (if our_slave == old_current), 2496 * but before a new active slave is selected. 2497 */ 2498 bond_alb_deinit_slave(bond, slave); 2499 } 2500 2501 if (all) { 2502 RCU_INIT_POINTER(bond->curr_active_slave, NULL); 2503 } else if (oldcurrent == slave) { 2504 /* Note that we hold RTNL over this sequence, so there 2505 * is no concern that another slave add/remove event 2506 * will interfere. 2507 */ 2508 bond_select_active_slave(bond); 2509 } 2510 2511 bond_set_carrier(bond); 2512 if (!bond_has_slaves(bond)) 2513 eth_hw_addr_random(bond_dev); 2514 2515 unblock_netpoll_tx(); 2516 synchronize_rcu(); 2517 bond->slave_cnt--; 2518 2519 if (!bond_has_slaves(bond)) { 2520 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev); 2521 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev); 2522 } 2523 2524 bond_compute_features(bond); 2525 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) && 2526 (old_features & NETIF_F_VLAN_CHALLENGED)) 2527 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n"); 2528 2529 vlan_vids_del_by_dev(slave_dev, bond_dev); 2530 2531 /* If the mode uses primary, then this case was handled above by 2532 * bond_change_active_slave(..., NULL) 2533 */ 2534 if (!bond_uses_primary(bond)) { 2535 /* unset promiscuity level from slave 2536 * NOTE: The NETDEV_CHANGEADDR call above may change the value 2537 * of the IFF_PROMISC flag in the bond_dev, but we need the 2538 * value of that flag before that change, as that was the value 2539 * when this slave was attached, so we cache at the start of the 2540 * function and use it here. Same goes for ALLMULTI below 2541 */ 2542 if (old_flags & IFF_PROMISC) 2543 dev_set_promiscuity(slave_dev, -1); 2544 2545 /* unset allmulti level from slave */ 2546 if (old_flags & IFF_ALLMULTI) 2547 dev_set_allmulti(slave_dev, -1); 2548 2549 if (old_flags & IFF_UP) 2550 bond_hw_addr_flush(bond_dev, slave_dev); 2551 } 2552 2553 slave_disable_netpoll(slave); 2554 2555 /* close slave before restoring its mac address */ 2556 dev_close(slave_dev); 2557 2558 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF; 2559 2560 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE || 2561 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2562 /* restore original ("permanent") mac address */ 2563 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr, 2564 slave->dev->addr_len); 2565 ss.ss_family = slave_dev->type; 2566 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL); 2567 } 2568 2569 if (unregister) 2570 __dev_set_mtu(slave_dev, slave->original_mtu); 2571 else 2572 dev_set_mtu(slave_dev, slave->original_mtu); 2573 2574 if (!netif_is_bond_master(slave_dev)) 2575 slave_dev->priv_flags &= ~IFF_BONDING; 2576 2577 bond_xdp_set_features(bond_dev); 2578 kobject_put(&slave->kobj); 2579 2580 return 0; 2581 } 2582 2583 /* A wrapper used because of ndo_del_link */ 2584 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev) 2585 { 2586 return __bond_release_one(bond_dev, slave_dev, false, false); 2587 } 2588 2589 /* First release a slave and then destroy the bond if no more slaves are left. 2590 * Must be under rtnl_lock when this function is called. 2591 */ 2592 static int bond_release_and_destroy(struct net_device *bond_dev, 2593 struct net_device *slave_dev) 2594 { 2595 struct bonding *bond = netdev_priv(bond_dev); 2596 int ret; 2597 2598 ret = __bond_release_one(bond_dev, slave_dev, false, true); 2599 if (ret == 0 && !bond_has_slaves(bond) && 2600 bond_dev->reg_state != NETREG_UNREGISTERING) { 2601 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL; 2602 netdev_info(bond_dev, "Destroying bond\n"); 2603 bond_remove_proc_entry(bond); 2604 unregister_netdevice(bond_dev); 2605 } 2606 return ret; 2607 } 2608 2609 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info) 2610 { 2611 struct bonding *bond = netdev_priv(bond_dev); 2612 2613 bond_fill_ifbond(bond, info); 2614 } 2615 2616 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info) 2617 { 2618 struct bonding *bond = netdev_priv(bond_dev); 2619 struct list_head *iter; 2620 int i = 0, res = -ENODEV; 2621 struct slave *slave; 2622 2623 bond_for_each_slave(bond, slave, iter) { 2624 if (i++ == (int)info->slave_id) { 2625 res = 0; 2626 bond_fill_ifslave(slave, info); 2627 break; 2628 } 2629 } 2630 2631 return res; 2632 } 2633 2634 /*-------------------------------- Monitoring -------------------------------*/ 2635 2636 /* called with rcu_read_lock() */ 2637 static int bond_miimon_inspect(struct bonding *bond) 2638 { 2639 bool ignore_updelay = false; 2640 int link_state, commit = 0; 2641 struct list_head *iter; 2642 struct slave *slave; 2643 2644 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) { 2645 ignore_updelay = !rcu_dereference(bond->curr_active_slave); 2646 } else { 2647 struct bond_up_slave *usable_slaves; 2648 2649 usable_slaves = rcu_dereference(bond->usable_slaves); 2650 2651 if (usable_slaves && usable_slaves->count == 0) 2652 ignore_updelay = true; 2653 } 2654 2655 bond_for_each_slave_rcu(bond, slave, iter) { 2656 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2657 2658 link_state = bond_check_dev_link(bond, slave->dev, 0); 2659 2660 switch (slave->link) { 2661 case BOND_LINK_UP: 2662 if (link_state) 2663 continue; 2664 2665 bond_propose_link_state(slave, BOND_LINK_FAIL); 2666 commit++; 2667 slave->delay = bond->params.downdelay; 2668 if (slave->delay) { 2669 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n", 2670 (BOND_MODE(bond) == 2671 BOND_MODE_ACTIVEBACKUP) ? 2672 (bond_is_active_slave(slave) ? 2673 "active " : "backup ") : "", 2674 bond->params.downdelay * bond->params.miimon); 2675 } 2676 fallthrough; 2677 case BOND_LINK_FAIL: 2678 if (link_state) { 2679 /* recovered before downdelay expired */ 2680 bond_propose_link_state(slave, BOND_LINK_UP); 2681 slave->last_link_up = jiffies; 2682 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n", 2683 (bond->params.downdelay - slave->delay) * 2684 bond->params.miimon); 2685 commit++; 2686 continue; 2687 } 2688 2689 if (slave->delay <= 0) { 2690 bond_propose_link_state(slave, BOND_LINK_DOWN); 2691 commit++; 2692 continue; 2693 } 2694 2695 slave->delay--; 2696 break; 2697 2698 case BOND_LINK_DOWN: 2699 if (!link_state) 2700 continue; 2701 2702 bond_propose_link_state(slave, BOND_LINK_BACK); 2703 commit++; 2704 slave->delay = bond->params.updelay; 2705 2706 if (slave->delay) { 2707 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n", 2708 ignore_updelay ? 0 : 2709 bond->params.updelay * 2710 bond->params.miimon); 2711 } 2712 fallthrough; 2713 case BOND_LINK_BACK: 2714 if (!link_state) { 2715 bond_propose_link_state(slave, BOND_LINK_DOWN); 2716 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n", 2717 (bond->params.updelay - slave->delay) * 2718 bond->params.miimon); 2719 commit++; 2720 continue; 2721 } 2722 2723 if (ignore_updelay) 2724 slave->delay = 0; 2725 2726 if (slave->delay <= 0) { 2727 bond_propose_link_state(slave, BOND_LINK_UP); 2728 commit++; 2729 ignore_updelay = false; 2730 continue; 2731 } 2732 2733 slave->delay--; 2734 break; 2735 } 2736 } 2737 2738 return commit; 2739 } 2740 2741 static void bond_miimon_link_change(struct bonding *bond, 2742 struct slave *slave, 2743 char link) 2744 { 2745 switch (BOND_MODE(bond)) { 2746 case BOND_MODE_8023AD: 2747 bond_3ad_handle_link_change(slave, link); 2748 break; 2749 case BOND_MODE_TLB: 2750 case BOND_MODE_ALB: 2751 bond_alb_handle_link_change(bond, slave, link); 2752 break; 2753 case BOND_MODE_XOR: 2754 bond_update_slave_arr(bond, NULL); 2755 break; 2756 } 2757 } 2758 2759 static void bond_miimon_commit(struct bonding *bond) 2760 { 2761 struct slave *slave, *primary, *active; 2762 bool do_failover = false; 2763 struct list_head *iter; 2764 2765 ASSERT_RTNL(); 2766 2767 bond_for_each_slave(bond, slave, iter) { 2768 switch (slave->link_new_state) { 2769 case BOND_LINK_NOCHANGE: 2770 /* For 802.3ad mode, check current slave speed and 2771 * duplex again in case its port was disabled after 2772 * invalid speed/duplex reporting but recovered before 2773 * link monitoring could make a decision on the actual 2774 * link status 2775 */ 2776 if (BOND_MODE(bond) == BOND_MODE_8023AD && 2777 slave->link == BOND_LINK_UP) 2778 bond_3ad_adapter_speed_duplex_changed(slave); 2779 continue; 2780 2781 case BOND_LINK_UP: 2782 if (bond_update_speed_duplex(slave) && 2783 bond_needs_speed_duplex(bond)) { 2784 slave->link = BOND_LINK_DOWN; 2785 if (net_ratelimit()) 2786 slave_warn(bond->dev, slave->dev, 2787 "failed to get link speed/duplex\n"); 2788 continue; 2789 } 2790 bond_set_slave_link_state(slave, BOND_LINK_UP, 2791 BOND_SLAVE_NOTIFY_NOW); 2792 slave->last_link_up = jiffies; 2793 2794 primary = rtnl_dereference(bond->primary_slave); 2795 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 2796 /* prevent it from being the active one */ 2797 bond_set_backup_slave(slave); 2798 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) { 2799 /* make it immediately active */ 2800 bond_set_active_slave(slave); 2801 } 2802 2803 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n", 2804 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed, 2805 slave->duplex ? "full" : "half"); 2806 2807 bond_miimon_link_change(bond, slave, BOND_LINK_UP); 2808 2809 active = rtnl_dereference(bond->curr_active_slave); 2810 if (!active || slave == primary || slave->prio > active->prio) 2811 do_failover = true; 2812 2813 continue; 2814 2815 case BOND_LINK_DOWN: 2816 if (slave->link_failure_count < UINT_MAX) 2817 slave->link_failure_count++; 2818 2819 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 2820 BOND_SLAVE_NOTIFY_NOW); 2821 2822 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 2823 BOND_MODE(bond) == BOND_MODE_8023AD) 2824 bond_set_slave_inactive_flags(slave, 2825 BOND_SLAVE_NOTIFY_NOW); 2826 2827 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 2828 2829 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN); 2830 2831 if (slave == rcu_access_pointer(bond->curr_active_slave)) 2832 do_failover = true; 2833 2834 continue; 2835 2836 default: 2837 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n", 2838 slave->link_new_state); 2839 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 2840 2841 continue; 2842 } 2843 } 2844 2845 if (do_failover) { 2846 block_netpoll_tx(); 2847 bond_select_active_slave(bond); 2848 unblock_netpoll_tx(); 2849 } 2850 2851 bond_set_carrier(bond); 2852 } 2853 2854 /* bond_mii_monitor 2855 * 2856 * Really a wrapper that splits the mii monitor into two phases: an 2857 * inspection, then (if inspection indicates something needs to be done) 2858 * an acquisition of appropriate locks followed by a commit phase to 2859 * implement whatever link state changes are indicated. 2860 */ 2861 static void bond_mii_monitor(struct work_struct *work) 2862 { 2863 struct bonding *bond = container_of(work, struct bonding, 2864 mii_work.work); 2865 bool should_notify_peers = false; 2866 bool commit; 2867 unsigned long delay; 2868 struct slave *slave; 2869 struct list_head *iter; 2870 2871 delay = msecs_to_jiffies(bond->params.miimon); 2872 2873 if (!bond_has_slaves(bond)) 2874 goto re_arm; 2875 2876 rcu_read_lock(); 2877 should_notify_peers = bond_should_notify_peers(bond); 2878 commit = !!bond_miimon_inspect(bond); 2879 if (bond->send_peer_notif) { 2880 rcu_read_unlock(); 2881 if (rtnl_trylock()) { 2882 bond->send_peer_notif--; 2883 rtnl_unlock(); 2884 } 2885 } else { 2886 rcu_read_unlock(); 2887 } 2888 2889 if (commit) { 2890 /* Race avoidance with bond_close cancel of workqueue */ 2891 if (!rtnl_trylock()) { 2892 delay = 1; 2893 should_notify_peers = false; 2894 goto re_arm; 2895 } 2896 2897 bond_for_each_slave(bond, slave, iter) { 2898 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER); 2899 } 2900 bond_miimon_commit(bond); 2901 2902 rtnl_unlock(); /* might sleep, hold no other locks */ 2903 } 2904 2905 re_arm: 2906 if (bond->params.miimon) 2907 queue_delayed_work(bond->wq, &bond->mii_work, delay); 2908 2909 if (should_notify_peers) { 2910 if (!rtnl_trylock()) 2911 return; 2912 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev); 2913 rtnl_unlock(); 2914 } 2915 } 2916 2917 static int bond_upper_dev_walk(struct net_device *upper, 2918 struct netdev_nested_priv *priv) 2919 { 2920 __be32 ip = *(__be32 *)priv->data; 2921 2922 return ip == bond_confirm_addr(upper, 0, ip); 2923 } 2924 2925 static bool bond_has_this_ip(struct bonding *bond, __be32 ip) 2926 { 2927 struct netdev_nested_priv priv = { 2928 .data = (void *)&ip, 2929 }; 2930 bool ret = false; 2931 2932 if (ip == bond_confirm_addr(bond->dev, 0, ip)) 2933 return true; 2934 2935 rcu_read_lock(); 2936 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv)) 2937 ret = true; 2938 rcu_read_unlock(); 2939 2940 return ret; 2941 } 2942 2943 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff) 2944 2945 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags, 2946 struct sk_buff *skb) 2947 { 2948 struct net_device *bond_dev = slave->bond->dev; 2949 struct net_device *slave_dev = slave->dev; 2950 struct bond_vlan_tag *outer_tag = tags; 2951 2952 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE) 2953 return true; 2954 2955 tags++; 2956 2957 /* Go through all the tags backwards and add them to the packet */ 2958 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) { 2959 if (!tags->vlan_id) { 2960 tags++; 2961 continue; 2962 } 2963 2964 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n", 2965 ntohs(outer_tag->vlan_proto), tags->vlan_id); 2966 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto, 2967 tags->vlan_id); 2968 if (!skb) { 2969 net_err_ratelimited("failed to insert inner VLAN tag\n"); 2970 return false; 2971 } 2972 2973 tags++; 2974 } 2975 /* Set the outer tag */ 2976 if (outer_tag->vlan_id) { 2977 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n", 2978 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id); 2979 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto, 2980 outer_tag->vlan_id); 2981 } 2982 2983 return true; 2984 } 2985 2986 /* We go to the (large) trouble of VLAN tagging ARP frames because 2987 * switches in VLAN mode (especially if ports are configured as 2988 * "native" to a VLAN) might not pass non-tagged frames. 2989 */ 2990 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip, 2991 __be32 src_ip, struct bond_vlan_tag *tags) 2992 { 2993 struct net_device *bond_dev = slave->bond->dev; 2994 struct net_device *slave_dev = slave->dev; 2995 struct sk_buff *skb; 2996 2997 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n", 2998 arp_op, &dest_ip, &src_ip); 2999 3000 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip, 3001 NULL, slave_dev->dev_addr, NULL); 3002 3003 if (!skb) { 3004 net_err_ratelimited("ARP packet allocation failed\n"); 3005 return; 3006 } 3007 3008 if (bond_handle_vlan(slave, tags, skb)) { 3009 slave_update_last_tx(slave); 3010 arp_xmit(skb); 3011 } 3012 3013 return; 3014 } 3015 3016 /* Validate the device path between the @start_dev and the @end_dev. 3017 * The path is valid if the @end_dev is reachable through device 3018 * stacking. 3019 * When the path is validated, collect any vlan information in the 3020 * path. 3021 */ 3022 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 3023 struct net_device *end_dev, 3024 int level) 3025 { 3026 struct bond_vlan_tag *tags; 3027 struct net_device *upper; 3028 struct list_head *iter; 3029 3030 if (start_dev == end_dev) { 3031 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC); 3032 if (!tags) 3033 return ERR_PTR(-ENOMEM); 3034 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE; 3035 return tags; 3036 } 3037 3038 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) { 3039 tags = bond_verify_device_path(upper, end_dev, level + 1); 3040 if (IS_ERR_OR_NULL(tags)) { 3041 if (IS_ERR(tags)) 3042 return tags; 3043 continue; 3044 } 3045 if (is_vlan_dev(upper)) { 3046 tags[level].vlan_proto = vlan_dev_vlan_proto(upper); 3047 tags[level].vlan_id = vlan_dev_vlan_id(upper); 3048 } 3049 3050 return tags; 3051 } 3052 3053 return NULL; 3054 } 3055 3056 static void bond_arp_send_all(struct bonding *bond, struct slave *slave) 3057 { 3058 struct rtable *rt; 3059 struct bond_vlan_tag *tags; 3060 __be32 *targets = bond->params.arp_targets, addr; 3061 int i; 3062 3063 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) { 3064 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n", 3065 __func__, &targets[i]); 3066 tags = NULL; 3067 3068 /* Find out through which dev should the packet go */ 3069 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 3070 RTO_ONLINK, 0); 3071 if (IS_ERR(rt)) { 3072 /* there's no route to target - try to send arp 3073 * probe to generate any traffic (arp_validate=0) 3074 */ 3075 if (bond->params.arp_validate) 3076 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n", 3077 bond->dev->name, 3078 &targets[i]); 3079 bond_arp_send(slave, ARPOP_REQUEST, targets[i], 3080 0, tags); 3081 continue; 3082 } 3083 3084 /* bond device itself */ 3085 if (rt->dst.dev == bond->dev) 3086 goto found; 3087 3088 rcu_read_lock(); 3089 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0); 3090 rcu_read_unlock(); 3091 3092 if (!IS_ERR_OR_NULL(tags)) 3093 goto found; 3094 3095 /* Not our device - skip */ 3096 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n", 3097 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL"); 3098 3099 ip_rt_put(rt); 3100 continue; 3101 3102 found: 3103 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0); 3104 ip_rt_put(rt); 3105 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags); 3106 kfree(tags); 3107 } 3108 } 3109 3110 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip) 3111 { 3112 int i; 3113 3114 if (!sip || !bond_has_this_ip(bond, tip)) { 3115 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n", 3116 __func__, &sip, &tip); 3117 return; 3118 } 3119 3120 i = bond_get_targets_ip(bond->params.arp_targets, sip); 3121 if (i == -1) { 3122 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n", 3123 __func__, &sip); 3124 return; 3125 } 3126 slave->last_rx = jiffies; 3127 slave->target_last_arp_rx[i] = jiffies; 3128 } 3129 3130 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, 3131 struct slave *slave) 3132 { 3133 struct arphdr *arp = (struct arphdr *)skb->data; 3134 struct slave *curr_active_slave, *curr_arp_slave; 3135 unsigned char *arp_ptr; 3136 __be32 sip, tip; 3137 unsigned int alen; 3138 3139 alen = arp_hdr_len(bond->dev); 3140 3141 if (alen > skb_headlen(skb)) { 3142 arp = kmalloc(alen, GFP_ATOMIC); 3143 if (!arp) 3144 goto out_unlock; 3145 if (skb_copy_bits(skb, 0, arp, alen) < 0) 3146 goto out_unlock; 3147 } 3148 3149 if (arp->ar_hln != bond->dev->addr_len || 3150 skb->pkt_type == PACKET_OTHERHOST || 3151 skb->pkt_type == PACKET_LOOPBACK || 3152 arp->ar_hrd != htons(ARPHRD_ETHER) || 3153 arp->ar_pro != htons(ETH_P_IP) || 3154 arp->ar_pln != 4) 3155 goto out_unlock; 3156 3157 arp_ptr = (unsigned char *)(arp + 1); 3158 arp_ptr += bond->dev->addr_len; 3159 memcpy(&sip, arp_ptr, 4); 3160 arp_ptr += 4 + bond->dev->addr_len; 3161 memcpy(&tip, arp_ptr, 4); 3162 3163 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n", 3164 __func__, slave->dev->name, bond_slave_state(slave), 3165 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3166 &sip, &tip); 3167 3168 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3169 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3170 3171 /* We 'trust' the received ARP enough to validate it if: 3172 * 3173 * (a) the slave receiving the ARP is active (which includes the 3174 * current ARP slave, if any), or 3175 * 3176 * (b) the receiving slave isn't active, but there is a currently 3177 * active slave and it received valid arp reply(s) after it became 3178 * the currently active slave, or 3179 * 3180 * (c) there is an ARP slave that sent an ARP during the prior ARP 3181 * interval, and we receive an ARP reply on any slave. We accept 3182 * these because switch FDB update delays may deliver the ARP 3183 * reply to a slave other than the sender of the ARP request. 3184 * 3185 * Note: for (b), backup slaves are receiving the broadcast ARP 3186 * request, not a reply. This request passes from the sending 3187 * slave through the L2 switch(es) to the receiving slave. Since 3188 * this is checking the request, sip/tip are swapped for 3189 * validation. 3190 * 3191 * This is done to avoid endless looping when we can't reach the 3192 * arp_ip_target and fool ourselves with our own arp requests. 3193 */ 3194 if (bond_is_active_slave(slave)) 3195 bond_validate_arp(bond, slave, sip, tip); 3196 else if (curr_active_slave && 3197 time_after(slave_last_rx(bond, curr_active_slave), 3198 curr_active_slave->last_link_up)) 3199 bond_validate_arp(bond, slave, tip, sip); 3200 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) && 3201 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3202 bond_validate_arp(bond, slave, sip, tip); 3203 3204 out_unlock: 3205 if (arp != (struct arphdr *)skb->data) 3206 kfree(arp); 3207 return RX_HANDLER_ANOTHER; 3208 } 3209 3210 #if IS_ENABLED(CONFIG_IPV6) 3211 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr, 3212 const struct in6_addr *saddr, struct bond_vlan_tag *tags) 3213 { 3214 struct net_device *bond_dev = slave->bond->dev; 3215 struct net_device *slave_dev = slave->dev; 3216 struct in6_addr mcaddr; 3217 struct sk_buff *skb; 3218 3219 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n", 3220 daddr, saddr); 3221 3222 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0); 3223 if (!skb) { 3224 net_err_ratelimited("NS packet allocation failed\n"); 3225 return; 3226 } 3227 3228 addrconf_addr_solict_mult(daddr, &mcaddr); 3229 if (bond_handle_vlan(slave, tags, skb)) { 3230 slave_update_last_tx(slave); 3231 ndisc_send_skb(skb, &mcaddr, saddr); 3232 } 3233 } 3234 3235 static void bond_ns_send_all(struct bonding *bond, struct slave *slave) 3236 { 3237 struct in6_addr *targets = bond->params.ns_targets; 3238 struct bond_vlan_tag *tags; 3239 struct dst_entry *dst; 3240 struct in6_addr saddr; 3241 struct flowi6 fl6; 3242 int i; 3243 3244 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) { 3245 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n", 3246 __func__, &targets[i]); 3247 tags = NULL; 3248 3249 /* Find out through which dev should the packet go */ 3250 memset(&fl6, 0, sizeof(struct flowi6)); 3251 fl6.daddr = targets[i]; 3252 fl6.flowi6_oif = bond->dev->ifindex; 3253 3254 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6); 3255 if (dst->error) { 3256 dst_release(dst); 3257 /* there's no route to target - try to send arp 3258 * probe to generate any traffic (arp_validate=0) 3259 */ 3260 if (bond->params.arp_validate) 3261 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n", 3262 bond->dev->name, 3263 &targets[i]); 3264 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3265 continue; 3266 } 3267 3268 /* bond device itself */ 3269 if (dst->dev == bond->dev) 3270 goto found; 3271 3272 rcu_read_lock(); 3273 tags = bond_verify_device_path(bond->dev, dst->dev, 0); 3274 rcu_read_unlock(); 3275 3276 if (!IS_ERR_OR_NULL(tags)) 3277 goto found; 3278 3279 /* Not our device - skip */ 3280 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n", 3281 &targets[i], dst->dev ? dst->dev->name : "NULL"); 3282 3283 dst_release(dst); 3284 continue; 3285 3286 found: 3287 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr)) 3288 bond_ns_send(slave, &targets[i], &saddr, tags); 3289 else 3290 bond_ns_send(slave, &targets[i], &in6addr_any, tags); 3291 3292 dst_release(dst); 3293 kfree(tags); 3294 } 3295 } 3296 3297 static int bond_confirm_addr6(struct net_device *dev, 3298 struct netdev_nested_priv *priv) 3299 { 3300 struct in6_addr *addr = (struct in6_addr *)priv->data; 3301 3302 return ipv6_chk_addr(dev_net(dev), addr, dev, 0); 3303 } 3304 3305 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr) 3306 { 3307 struct netdev_nested_priv priv = { 3308 .data = addr, 3309 }; 3310 int ret = false; 3311 3312 if (bond_confirm_addr6(bond->dev, &priv)) 3313 return true; 3314 3315 rcu_read_lock(); 3316 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv)) 3317 ret = true; 3318 rcu_read_unlock(); 3319 3320 return ret; 3321 } 3322 3323 static void bond_validate_na(struct bonding *bond, struct slave *slave, 3324 struct in6_addr *saddr, struct in6_addr *daddr) 3325 { 3326 int i; 3327 3328 /* Ignore NAs that: 3329 * 1. Source address is unspecified address. 3330 * 2. Dest address is neither all-nodes multicast address nor 3331 * exist on bond interface. 3332 */ 3333 if (ipv6_addr_any(saddr) || 3334 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) && 3335 !bond_has_this_ip6(bond, daddr))) { 3336 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n", 3337 __func__, saddr, daddr); 3338 return; 3339 } 3340 3341 i = bond_get_targets_ip6(bond->params.ns_targets, saddr); 3342 if (i == -1) { 3343 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n", 3344 __func__, saddr); 3345 return; 3346 } 3347 slave->last_rx = jiffies; 3348 slave->target_last_arp_rx[i] = jiffies; 3349 } 3350 3351 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond, 3352 struct slave *slave) 3353 { 3354 struct slave *curr_active_slave, *curr_arp_slave; 3355 struct in6_addr *saddr, *daddr; 3356 struct { 3357 struct ipv6hdr ip6; 3358 struct icmp6hdr icmp6; 3359 } *combined, _combined; 3360 3361 if (skb->pkt_type == PACKET_OTHERHOST || 3362 skb->pkt_type == PACKET_LOOPBACK) 3363 goto out; 3364 3365 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined); 3366 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP || 3367 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION && 3368 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)) 3369 goto out; 3370 3371 saddr = &combined->ip6.saddr; 3372 daddr = &combined->ip6.daddr; 3373 3374 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n", 3375 __func__, slave->dev->name, bond_slave_state(slave), 3376 bond->params.arp_validate, slave_do_arp_validate(bond, slave), 3377 saddr, daddr); 3378 3379 curr_active_slave = rcu_dereference(bond->curr_active_slave); 3380 curr_arp_slave = rcu_dereference(bond->current_arp_slave); 3381 3382 /* We 'trust' the received ARP enough to validate it if: 3383 * see bond_arp_rcv(). 3384 */ 3385 if (bond_is_active_slave(slave)) 3386 bond_validate_na(bond, slave, saddr, daddr); 3387 else if (curr_active_slave && 3388 time_after(slave_last_rx(bond, curr_active_slave), 3389 curr_active_slave->last_link_up)) 3390 bond_validate_na(bond, slave, daddr, saddr); 3391 else if (curr_arp_slave && 3392 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1)) 3393 bond_validate_na(bond, slave, saddr, daddr); 3394 3395 out: 3396 return RX_HANDLER_ANOTHER; 3397 } 3398 #endif 3399 3400 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond, 3401 struct slave *slave) 3402 { 3403 #if IS_ENABLED(CONFIG_IPV6) 3404 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6); 3405 #endif 3406 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP); 3407 3408 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n", 3409 __func__, skb->dev->name); 3410 3411 /* Use arp validate logic for both ARP and NS */ 3412 if (!slave_do_arp_validate(bond, slave)) { 3413 if ((slave_do_arp_validate_only(bond) && is_arp) || 3414 #if IS_ENABLED(CONFIG_IPV6) 3415 (slave_do_arp_validate_only(bond) && is_ipv6) || 3416 #endif 3417 !slave_do_arp_validate_only(bond)) 3418 slave->last_rx = jiffies; 3419 return RX_HANDLER_ANOTHER; 3420 } else if (is_arp) { 3421 return bond_arp_rcv(skb, bond, slave); 3422 #if IS_ENABLED(CONFIG_IPV6) 3423 } else if (is_ipv6) { 3424 return bond_na_rcv(skb, bond, slave); 3425 #endif 3426 } else { 3427 return RX_HANDLER_ANOTHER; 3428 } 3429 } 3430 3431 static void bond_send_validate(struct bonding *bond, struct slave *slave) 3432 { 3433 bond_arp_send_all(bond, slave); 3434 #if IS_ENABLED(CONFIG_IPV6) 3435 bond_ns_send_all(bond, slave); 3436 #endif 3437 } 3438 3439 /* function to verify if we're in the arp_interval timeslice, returns true if 3440 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval + 3441 * arp_interval/2) . the arp_interval/2 is needed for really fast networks. 3442 */ 3443 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act, 3444 int mod) 3445 { 3446 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3447 3448 return time_in_range(jiffies, 3449 last_act - delta_in_ticks, 3450 last_act + mod * delta_in_ticks + delta_in_ticks/2); 3451 } 3452 3453 /* This function is called regularly to monitor each slave's link 3454 * ensuring that traffic is being sent and received when arp monitoring 3455 * is used in load-balancing mode. if the adapter has been dormant, then an 3456 * arp is transmitted to generate traffic. see activebackup_arp_monitor for 3457 * arp monitoring in active backup mode. 3458 */ 3459 static void bond_loadbalance_arp_mon(struct bonding *bond) 3460 { 3461 struct slave *slave, *oldcurrent; 3462 struct list_head *iter; 3463 int do_failover = 0, slave_state_changed = 0; 3464 3465 if (!bond_has_slaves(bond)) 3466 goto re_arm; 3467 3468 rcu_read_lock(); 3469 3470 oldcurrent = rcu_dereference(bond->curr_active_slave); 3471 /* see if any of the previous devices are up now (i.e. they have 3472 * xmt and rcv traffic). the curr_active_slave does not come into 3473 * the picture unless it is null. also, slave->last_link_up is not 3474 * needed here because we send an arp on each slave and give a slave 3475 * as long as it needs to get the tx/rx within the delta. 3476 * TODO: what about up/down delay in arp mode? it wasn't here before 3477 * so it can wait 3478 */ 3479 bond_for_each_slave_rcu(bond, slave, iter) { 3480 unsigned long last_tx = slave_last_tx(slave); 3481 3482 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3483 3484 if (slave->link != BOND_LINK_UP) { 3485 if (bond_time_in_interval(bond, last_tx, 1) && 3486 bond_time_in_interval(bond, slave->last_rx, 1)) { 3487 3488 bond_propose_link_state(slave, BOND_LINK_UP); 3489 slave_state_changed = 1; 3490 3491 /* primary_slave has no meaning in round-robin 3492 * mode. the window of a slave being up and 3493 * curr_active_slave being null after enslaving 3494 * is closed. 3495 */ 3496 if (!oldcurrent) { 3497 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3498 do_failover = 1; 3499 } else { 3500 slave_info(bond->dev, slave->dev, "interface is now up\n"); 3501 } 3502 } 3503 } else { 3504 /* slave->link == BOND_LINK_UP */ 3505 3506 /* not all switches will respond to an arp request 3507 * when the source ip is 0, so don't take the link down 3508 * if we don't know our ip yet 3509 */ 3510 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3511 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) { 3512 3513 bond_propose_link_state(slave, BOND_LINK_DOWN); 3514 slave_state_changed = 1; 3515 3516 if (slave->link_failure_count < UINT_MAX) 3517 slave->link_failure_count++; 3518 3519 slave_info(bond->dev, slave->dev, "interface is now down\n"); 3520 3521 if (slave == oldcurrent) 3522 do_failover = 1; 3523 } 3524 } 3525 3526 /* note: if switch is in round-robin mode, all links 3527 * must tx arp to ensure all links rx an arp - otherwise 3528 * links may oscillate or not come up at all; if switch is 3529 * in something like xor mode, there is nothing we can 3530 * do - all replies will be rx'ed on same link causing slaves 3531 * to be unstable during low/no traffic periods 3532 */ 3533 if (bond_slave_is_up(slave)) 3534 bond_send_validate(bond, slave); 3535 } 3536 3537 rcu_read_unlock(); 3538 3539 if (do_failover || slave_state_changed) { 3540 if (!rtnl_trylock()) 3541 goto re_arm; 3542 3543 bond_for_each_slave(bond, slave, iter) { 3544 if (slave->link_new_state != BOND_LINK_NOCHANGE) 3545 slave->link = slave->link_new_state; 3546 } 3547 3548 if (slave_state_changed) { 3549 bond_slave_state_change(bond); 3550 if (BOND_MODE(bond) == BOND_MODE_XOR) 3551 bond_update_slave_arr(bond, NULL); 3552 } 3553 if (do_failover) { 3554 block_netpoll_tx(); 3555 bond_select_active_slave(bond); 3556 unblock_netpoll_tx(); 3557 } 3558 rtnl_unlock(); 3559 } 3560 3561 re_arm: 3562 if (bond->params.arp_interval) 3563 queue_delayed_work(bond->wq, &bond->arp_work, 3564 msecs_to_jiffies(bond->params.arp_interval)); 3565 } 3566 3567 /* Called to inspect slaves for active-backup mode ARP monitor link state 3568 * changes. Sets proposed link state in slaves to specify what action 3569 * should take place for the slave. Returns 0 if no changes are found, >0 3570 * if changes to link states must be committed. 3571 * 3572 * Called with rcu_read_lock held. 3573 */ 3574 static int bond_ab_arp_inspect(struct bonding *bond) 3575 { 3576 unsigned long last_tx, last_rx; 3577 struct list_head *iter; 3578 struct slave *slave; 3579 int commit = 0; 3580 3581 bond_for_each_slave_rcu(bond, slave, iter) { 3582 bond_propose_link_state(slave, BOND_LINK_NOCHANGE); 3583 last_rx = slave_last_rx(bond, slave); 3584 3585 if (slave->link != BOND_LINK_UP) { 3586 if (bond_time_in_interval(bond, last_rx, 1)) { 3587 bond_propose_link_state(slave, BOND_LINK_UP); 3588 commit++; 3589 } else if (slave->link == BOND_LINK_BACK) { 3590 bond_propose_link_state(slave, BOND_LINK_FAIL); 3591 commit++; 3592 } 3593 continue; 3594 } 3595 3596 /* Give slaves 2*delta after being enslaved or made 3597 * active. This avoids bouncing, as the last receive 3598 * times need a full ARP monitor cycle to be updated. 3599 */ 3600 if (bond_time_in_interval(bond, slave->last_link_up, 2)) 3601 continue; 3602 3603 /* Backup slave is down if: 3604 * - No current_arp_slave AND 3605 * - more than (missed_max+1)*delta since last receive AND 3606 * - the bond has an IP address 3607 * 3608 * Note: a non-null current_arp_slave indicates 3609 * the curr_active_slave went down and we are 3610 * searching for a new one; under this condition 3611 * we only take the curr_active_slave down - this 3612 * gives each slave a chance to tx/rx traffic 3613 * before being taken out 3614 */ 3615 if (!bond_is_active_slave(slave) && 3616 !rcu_access_pointer(bond->current_arp_slave) && 3617 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) { 3618 bond_propose_link_state(slave, BOND_LINK_DOWN); 3619 commit++; 3620 } 3621 3622 /* Active slave is down if: 3623 * - more than missed_max*delta since transmitting OR 3624 * - (more than missed_max*delta since receive AND 3625 * the bond has an IP address) 3626 */ 3627 last_tx = slave_last_tx(slave); 3628 if (bond_is_active_slave(slave) && 3629 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) || 3630 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) { 3631 bond_propose_link_state(slave, BOND_LINK_DOWN); 3632 commit++; 3633 } 3634 } 3635 3636 return commit; 3637 } 3638 3639 /* Called to commit link state changes noted by inspection step of 3640 * active-backup mode ARP monitor. 3641 * 3642 * Called with RTNL hold. 3643 */ 3644 static void bond_ab_arp_commit(struct bonding *bond) 3645 { 3646 bool do_failover = false; 3647 struct list_head *iter; 3648 unsigned long last_tx; 3649 struct slave *slave; 3650 3651 bond_for_each_slave(bond, slave, iter) { 3652 switch (slave->link_new_state) { 3653 case BOND_LINK_NOCHANGE: 3654 continue; 3655 3656 case BOND_LINK_UP: 3657 last_tx = slave_last_tx(slave); 3658 if (rtnl_dereference(bond->curr_active_slave) != slave || 3659 (!rtnl_dereference(bond->curr_active_slave) && 3660 bond_time_in_interval(bond, last_tx, 1))) { 3661 struct slave *current_arp_slave; 3662 3663 current_arp_slave = rtnl_dereference(bond->current_arp_slave); 3664 bond_set_slave_link_state(slave, BOND_LINK_UP, 3665 BOND_SLAVE_NOTIFY_NOW); 3666 if (current_arp_slave) { 3667 bond_set_slave_inactive_flags( 3668 current_arp_slave, 3669 BOND_SLAVE_NOTIFY_NOW); 3670 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3671 } 3672 3673 slave_info(bond->dev, slave->dev, "link status definitely up\n"); 3674 3675 if (!rtnl_dereference(bond->curr_active_slave) || 3676 slave == rtnl_dereference(bond->primary_slave) || 3677 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio) 3678 do_failover = true; 3679 3680 } 3681 3682 continue; 3683 3684 case BOND_LINK_DOWN: 3685 if (slave->link_failure_count < UINT_MAX) 3686 slave->link_failure_count++; 3687 3688 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3689 BOND_SLAVE_NOTIFY_NOW); 3690 bond_set_slave_inactive_flags(slave, 3691 BOND_SLAVE_NOTIFY_NOW); 3692 3693 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n"); 3694 3695 if (slave == rtnl_dereference(bond->curr_active_slave)) { 3696 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3697 do_failover = true; 3698 } 3699 3700 continue; 3701 3702 case BOND_LINK_FAIL: 3703 bond_set_slave_link_state(slave, BOND_LINK_FAIL, 3704 BOND_SLAVE_NOTIFY_NOW); 3705 bond_set_slave_inactive_flags(slave, 3706 BOND_SLAVE_NOTIFY_NOW); 3707 3708 /* A slave has just been enslaved and has become 3709 * the current active slave. 3710 */ 3711 if (rtnl_dereference(bond->curr_active_slave)) 3712 RCU_INIT_POINTER(bond->current_arp_slave, NULL); 3713 continue; 3714 3715 default: 3716 slave_err(bond->dev, slave->dev, 3717 "impossible: link_new_state %d on slave\n", 3718 slave->link_new_state); 3719 continue; 3720 } 3721 } 3722 3723 if (do_failover) { 3724 block_netpoll_tx(); 3725 bond_select_active_slave(bond); 3726 unblock_netpoll_tx(); 3727 } 3728 3729 bond_set_carrier(bond); 3730 } 3731 3732 /* Send ARP probes for active-backup mode ARP monitor. 3733 * 3734 * Called with rcu_read_lock held. 3735 */ 3736 static bool bond_ab_arp_probe(struct bonding *bond) 3737 { 3738 struct slave *slave, *before = NULL, *new_slave = NULL, 3739 *curr_arp_slave = rcu_dereference(bond->current_arp_slave), 3740 *curr_active_slave = rcu_dereference(bond->curr_active_slave); 3741 struct list_head *iter; 3742 bool found = false; 3743 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER; 3744 3745 if (curr_arp_slave && curr_active_slave) 3746 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n", 3747 curr_arp_slave->dev->name, 3748 curr_active_slave->dev->name); 3749 3750 if (curr_active_slave) { 3751 bond_send_validate(bond, curr_active_slave); 3752 return should_notify_rtnl; 3753 } 3754 3755 /* if we don't have a curr_active_slave, search for the next available 3756 * backup slave from the current_arp_slave and make it the candidate 3757 * for becoming the curr_active_slave 3758 */ 3759 3760 if (!curr_arp_slave) { 3761 curr_arp_slave = bond_first_slave_rcu(bond); 3762 if (!curr_arp_slave) 3763 return should_notify_rtnl; 3764 } 3765 3766 bond_for_each_slave_rcu(bond, slave, iter) { 3767 if (!found && !before && bond_slave_is_up(slave)) 3768 before = slave; 3769 3770 if (found && !new_slave && bond_slave_is_up(slave)) 3771 new_slave = slave; 3772 /* if the link state is up at this point, we 3773 * mark it down - this can happen if we have 3774 * simultaneous link failures and 3775 * reselect_active_interface doesn't make this 3776 * one the current slave so it is still marked 3777 * up when it is actually down 3778 */ 3779 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) { 3780 bond_set_slave_link_state(slave, BOND_LINK_DOWN, 3781 BOND_SLAVE_NOTIFY_LATER); 3782 if (slave->link_failure_count < UINT_MAX) 3783 slave->link_failure_count++; 3784 3785 bond_set_slave_inactive_flags(slave, 3786 BOND_SLAVE_NOTIFY_LATER); 3787 3788 slave_info(bond->dev, slave->dev, "backup interface is now down\n"); 3789 } 3790 if (slave == curr_arp_slave) 3791 found = true; 3792 } 3793 3794 if (!new_slave && before) 3795 new_slave = before; 3796 3797 if (!new_slave) 3798 goto check_state; 3799 3800 bond_set_slave_link_state(new_slave, BOND_LINK_BACK, 3801 BOND_SLAVE_NOTIFY_LATER); 3802 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER); 3803 bond_send_validate(bond, new_slave); 3804 new_slave->last_link_up = jiffies; 3805 rcu_assign_pointer(bond->current_arp_slave, new_slave); 3806 3807 check_state: 3808 bond_for_each_slave_rcu(bond, slave, iter) { 3809 if (slave->should_notify || slave->should_notify_link) { 3810 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW; 3811 break; 3812 } 3813 } 3814 return should_notify_rtnl; 3815 } 3816 3817 static void bond_activebackup_arp_mon(struct bonding *bond) 3818 { 3819 bool should_notify_peers = false; 3820 bool should_notify_rtnl = false; 3821 int delta_in_ticks; 3822 3823 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval); 3824 3825 if (!bond_has_slaves(bond)) 3826 goto re_arm; 3827 3828 rcu_read_lock(); 3829 3830 should_notify_peers = bond_should_notify_peers(bond); 3831 3832 if (bond_ab_arp_inspect(bond)) { 3833 rcu_read_unlock(); 3834 3835 /* Race avoidance with bond_close flush of workqueue */ 3836 if (!rtnl_trylock()) { 3837 delta_in_ticks = 1; 3838 should_notify_peers = false; 3839 goto re_arm; 3840 } 3841 3842 bond_ab_arp_commit(bond); 3843 3844 rtnl_unlock(); 3845 rcu_read_lock(); 3846 } 3847 3848 should_notify_rtnl = bond_ab_arp_probe(bond); 3849 rcu_read_unlock(); 3850 3851 re_arm: 3852 if (bond->params.arp_interval) 3853 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks); 3854 3855 if (should_notify_peers || should_notify_rtnl) { 3856 if (!rtnl_trylock()) 3857 return; 3858 3859 if (should_notify_peers) { 3860 bond->send_peer_notif--; 3861 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, 3862 bond->dev); 3863 } 3864 if (should_notify_rtnl) { 3865 bond_slave_state_notify(bond); 3866 bond_slave_link_notify(bond); 3867 } 3868 3869 rtnl_unlock(); 3870 } 3871 } 3872 3873 static void bond_arp_monitor(struct work_struct *work) 3874 { 3875 struct bonding *bond = container_of(work, struct bonding, 3876 arp_work.work); 3877 3878 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 3879 bond_activebackup_arp_mon(bond); 3880 else 3881 bond_loadbalance_arp_mon(bond); 3882 } 3883 3884 /*-------------------------- netdev event handling --------------------------*/ 3885 3886 /* Change device name */ 3887 static int bond_event_changename(struct bonding *bond) 3888 { 3889 bond_remove_proc_entry(bond); 3890 bond_create_proc_entry(bond); 3891 3892 bond_debug_reregister(bond); 3893 3894 return NOTIFY_DONE; 3895 } 3896 3897 static int bond_master_netdev_event(unsigned long event, 3898 struct net_device *bond_dev) 3899 { 3900 struct bonding *event_bond = netdev_priv(bond_dev); 3901 3902 netdev_dbg(bond_dev, "%s called\n", __func__); 3903 3904 switch (event) { 3905 case NETDEV_CHANGENAME: 3906 return bond_event_changename(event_bond); 3907 case NETDEV_UNREGISTER: 3908 bond_remove_proc_entry(event_bond); 3909 #ifdef CONFIG_XFRM_OFFLOAD 3910 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true); 3911 #endif /* CONFIG_XFRM_OFFLOAD */ 3912 break; 3913 case NETDEV_REGISTER: 3914 bond_create_proc_entry(event_bond); 3915 break; 3916 default: 3917 break; 3918 } 3919 3920 return NOTIFY_DONE; 3921 } 3922 3923 static int bond_slave_netdev_event(unsigned long event, 3924 struct net_device *slave_dev) 3925 { 3926 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary; 3927 struct bonding *bond; 3928 struct net_device *bond_dev; 3929 3930 /* A netdev event can be generated while enslaving a device 3931 * before netdev_rx_handler_register is called in which case 3932 * slave will be NULL 3933 */ 3934 if (!slave) { 3935 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__); 3936 return NOTIFY_DONE; 3937 } 3938 3939 bond_dev = slave->bond->dev; 3940 bond = slave->bond; 3941 primary = rtnl_dereference(bond->primary_slave); 3942 3943 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__); 3944 3945 switch (event) { 3946 case NETDEV_UNREGISTER: 3947 if (bond_dev->type != ARPHRD_ETHER) 3948 bond_release_and_destroy(bond_dev, slave_dev); 3949 else 3950 __bond_release_one(bond_dev, slave_dev, false, true); 3951 break; 3952 case NETDEV_UP: 3953 case NETDEV_CHANGE: 3954 /* For 802.3ad mode only: 3955 * Getting invalid Speed/Duplex values here will put slave 3956 * in weird state. Mark it as link-fail if the link was 3957 * previously up or link-down if it hasn't yet come up, and 3958 * let link-monitoring (miimon) set it right when correct 3959 * speeds/duplex are available. 3960 */ 3961 if (bond_update_speed_duplex(slave) && 3962 BOND_MODE(bond) == BOND_MODE_8023AD) { 3963 if (slave->last_link_up) 3964 slave->link = BOND_LINK_FAIL; 3965 else 3966 slave->link = BOND_LINK_DOWN; 3967 } 3968 3969 if (BOND_MODE(bond) == BOND_MODE_8023AD) 3970 bond_3ad_adapter_speed_duplex_changed(slave); 3971 fallthrough; 3972 case NETDEV_DOWN: 3973 /* Refresh slave-array if applicable! 3974 * If the setup does not use miimon or arpmon (mode-specific!), 3975 * then these events will not cause the slave-array to be 3976 * refreshed. This will cause xmit to use a slave that is not 3977 * usable. Avoid such situation by refeshing the array at these 3978 * events. If these (miimon/arpmon) parameters are configured 3979 * then array gets refreshed twice and that should be fine! 3980 */ 3981 if (bond_mode_can_use_xmit_hash(bond)) 3982 bond_update_slave_arr(bond, NULL); 3983 break; 3984 case NETDEV_CHANGEMTU: 3985 /* TODO: Should slaves be allowed to 3986 * independently alter their MTU? For 3987 * an active-backup bond, slaves need 3988 * not be the same type of device, so 3989 * MTUs may vary. For other modes, 3990 * slaves arguably should have the 3991 * same MTUs. To do this, we'd need to 3992 * take over the slave's change_mtu 3993 * function for the duration of their 3994 * servitude. 3995 */ 3996 break; 3997 case NETDEV_CHANGENAME: 3998 /* we don't care if we don't have primary set */ 3999 if (!bond_uses_primary(bond) || 4000 !bond->params.primary[0]) 4001 break; 4002 4003 if (slave == primary) { 4004 /* slave's name changed - he's no longer primary */ 4005 RCU_INIT_POINTER(bond->primary_slave, NULL); 4006 } else if (!strcmp(slave_dev->name, bond->params.primary)) { 4007 /* we have a new primary slave */ 4008 rcu_assign_pointer(bond->primary_slave, slave); 4009 } else { /* we didn't change primary - exit */ 4010 break; 4011 } 4012 4013 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n", 4014 primary ? slave_dev->name : "none"); 4015 4016 block_netpoll_tx(); 4017 bond_select_active_slave(bond); 4018 unblock_netpoll_tx(); 4019 break; 4020 case NETDEV_FEAT_CHANGE: 4021 if (!bond->notifier_ctx) { 4022 bond->notifier_ctx = true; 4023 bond_compute_features(bond); 4024 bond->notifier_ctx = false; 4025 } 4026 break; 4027 case NETDEV_RESEND_IGMP: 4028 /* Propagate to master device */ 4029 call_netdevice_notifiers(event, slave->bond->dev); 4030 break; 4031 case NETDEV_XDP_FEAT_CHANGE: 4032 bond_xdp_set_features(bond_dev); 4033 break; 4034 default: 4035 break; 4036 } 4037 4038 return NOTIFY_DONE; 4039 } 4040 4041 /* bond_netdev_event: handle netdev notifier chain events. 4042 * 4043 * This function receives events for the netdev chain. The caller (an 4044 * ioctl handler calling blocking_notifier_call_chain) holds the necessary 4045 * locks for us to safely manipulate the slave devices (RTNL lock, 4046 * dev_probe_lock). 4047 */ 4048 static int bond_netdev_event(struct notifier_block *this, 4049 unsigned long event, void *ptr) 4050 { 4051 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 4052 4053 netdev_dbg(event_dev, "%s received %s\n", 4054 __func__, netdev_cmd_to_name(event)); 4055 4056 if (!(event_dev->priv_flags & IFF_BONDING)) 4057 return NOTIFY_DONE; 4058 4059 if (event_dev->flags & IFF_MASTER) { 4060 int ret; 4061 4062 ret = bond_master_netdev_event(event, event_dev); 4063 if (ret != NOTIFY_DONE) 4064 return ret; 4065 } 4066 4067 if (event_dev->flags & IFF_SLAVE) 4068 return bond_slave_netdev_event(event, event_dev); 4069 4070 return NOTIFY_DONE; 4071 } 4072 4073 static struct notifier_block bond_netdev_notifier = { 4074 .notifier_call = bond_netdev_event, 4075 }; 4076 4077 /*---------------------------- Hashing Policies -----------------------------*/ 4078 4079 /* Helper to access data in a packet, with or without a backing skb. 4080 * If skb is given the data is linearized if necessary via pskb_may_pull. 4081 */ 4082 static inline const void *bond_pull_data(struct sk_buff *skb, 4083 const void *data, int hlen, int n) 4084 { 4085 if (likely(n <= hlen)) 4086 return data; 4087 else if (skb && likely(pskb_may_pull(skb, n))) 4088 return skb->data; 4089 4090 return NULL; 4091 } 4092 4093 /* L2 hash helper */ 4094 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4095 { 4096 struct ethhdr *ep; 4097 4098 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4099 if (!data) 4100 return 0; 4101 4102 ep = (struct ethhdr *)(data + mhoff); 4103 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto); 4104 } 4105 4106 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data, 4107 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34) 4108 { 4109 const struct ipv6hdr *iph6; 4110 const struct iphdr *iph; 4111 4112 if (l2_proto == htons(ETH_P_IP)) { 4113 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph)); 4114 if (!data) 4115 return false; 4116 4117 iph = (const struct iphdr *)(data + *nhoff); 4118 iph_to_flow_copy_v4addrs(fk, iph); 4119 *nhoff += iph->ihl << 2; 4120 if (!ip_is_fragment(iph)) 4121 *ip_proto = iph->protocol; 4122 } else if (l2_proto == htons(ETH_P_IPV6)) { 4123 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6)); 4124 if (!data) 4125 return false; 4126 4127 iph6 = (const struct ipv6hdr *)(data + *nhoff); 4128 iph_to_flow_copy_v6addrs(fk, iph6); 4129 *nhoff += sizeof(*iph6); 4130 *ip_proto = iph6->nexthdr; 4131 } else { 4132 return false; 4133 } 4134 4135 if (l34 && *ip_proto >= 0) 4136 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen); 4137 4138 return true; 4139 } 4140 4141 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen) 4142 { 4143 u32 srcmac_vendor = 0, srcmac_dev = 0; 4144 struct ethhdr *mac_hdr; 4145 u16 vlan = 0; 4146 int i; 4147 4148 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr)); 4149 if (!data) 4150 return 0; 4151 mac_hdr = (struct ethhdr *)(data + mhoff); 4152 4153 for (i = 0; i < 3; i++) 4154 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i]; 4155 4156 for (i = 3; i < ETH_ALEN; i++) 4157 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i]; 4158 4159 if (skb && skb_vlan_tag_present(skb)) 4160 vlan = skb_vlan_tag_get(skb); 4161 4162 return vlan ^ srcmac_vendor ^ srcmac_dev; 4163 } 4164 4165 /* Extract the appropriate headers based on bond's xmit policy */ 4166 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data, 4167 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk) 4168 { 4169 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34; 4170 int ip_proto = -1; 4171 4172 switch (bond->params.xmit_policy) { 4173 case BOND_XMIT_POLICY_ENCAP23: 4174 case BOND_XMIT_POLICY_ENCAP34: 4175 memset(fk, 0, sizeof(*fk)); 4176 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding, 4177 fk, data, l2_proto, nhoff, hlen, 0); 4178 default: 4179 break; 4180 } 4181 4182 fk->ports.ports = 0; 4183 memset(&fk->icmp, 0, sizeof(fk->icmp)); 4184 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34)) 4185 return false; 4186 4187 /* ICMP error packets contains at least 8 bytes of the header 4188 * of the packet which generated the error. Use this information 4189 * to correlate ICMP error packets within the same flow which 4190 * generated the error. 4191 */ 4192 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) { 4193 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen); 4194 if (ip_proto == IPPROTO_ICMP) { 4195 if (!icmp_is_err(fk->icmp.type)) 4196 return true; 4197 4198 nhoff += sizeof(struct icmphdr); 4199 } else if (ip_proto == IPPROTO_ICMPV6) { 4200 if (!icmpv6_is_err(fk->icmp.type)) 4201 return true; 4202 4203 nhoff += sizeof(struct icmp6hdr); 4204 } 4205 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34); 4206 } 4207 4208 return true; 4209 } 4210 4211 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy) 4212 { 4213 hash ^= (__force u32)flow_get_u32_dst(flow) ^ 4214 (__force u32)flow_get_u32_src(flow); 4215 hash ^= (hash >> 16); 4216 hash ^= (hash >> 8); 4217 4218 /* discard lowest hash bit to deal with the common even ports pattern */ 4219 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 || 4220 xmit_policy == BOND_XMIT_POLICY_ENCAP34) 4221 return hash >> 1; 4222 4223 return hash; 4224 } 4225 4226 /* Generate hash based on xmit policy. If @skb is given it is used to linearize 4227 * the data as required, but this function can be used without it if the data is 4228 * known to be linear (e.g. with xdp_buff). 4229 */ 4230 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data, 4231 __be16 l2_proto, int mhoff, int nhoff, int hlen) 4232 { 4233 struct flow_keys flow; 4234 u32 hash; 4235 4236 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC) 4237 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen); 4238 4239 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 || 4240 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow)) 4241 return bond_eth_hash(skb, data, mhoff, hlen); 4242 4243 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 || 4244 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) { 4245 hash = bond_eth_hash(skb, data, mhoff, hlen); 4246 } else { 4247 if (flow.icmp.id) 4248 memcpy(&hash, &flow.icmp, sizeof(hash)); 4249 else 4250 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 4251 } 4252 4253 return bond_ip_hash(hash, &flow, bond->params.xmit_policy); 4254 } 4255 4256 /** 4257 * bond_xmit_hash - generate a hash value based on the xmit policy 4258 * @bond: bonding device 4259 * @skb: buffer to use for headers 4260 * 4261 * This function will extract the necessary headers from the skb buffer and use 4262 * them to generate a hash based on the xmit_policy set in the bonding device 4263 */ 4264 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb) 4265 { 4266 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 && 4267 skb->l4_hash) 4268 return skb->hash; 4269 4270 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol, 4271 0, skb_network_offset(skb), 4272 skb_headlen(skb)); 4273 } 4274 4275 /** 4276 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy 4277 * @bond: bonding device 4278 * @xdp: buffer to use for headers 4279 * 4280 * The XDP variant of bond_xmit_hash. 4281 */ 4282 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp) 4283 { 4284 struct ethhdr *eth; 4285 4286 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end) 4287 return 0; 4288 4289 eth = (struct ethhdr *)xdp->data; 4290 4291 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0, 4292 sizeof(struct ethhdr), xdp->data_end - xdp->data); 4293 } 4294 4295 /*-------------------------- Device entry points ----------------------------*/ 4296 4297 void bond_work_init_all(struct bonding *bond) 4298 { 4299 INIT_DELAYED_WORK(&bond->mcast_work, 4300 bond_resend_igmp_join_requests_delayed); 4301 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor); 4302 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor); 4303 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor); 4304 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler); 4305 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler); 4306 } 4307 4308 static void bond_work_cancel_all(struct bonding *bond) 4309 { 4310 cancel_delayed_work_sync(&bond->mii_work); 4311 cancel_delayed_work_sync(&bond->arp_work); 4312 cancel_delayed_work_sync(&bond->alb_work); 4313 cancel_delayed_work_sync(&bond->ad_work); 4314 cancel_delayed_work_sync(&bond->mcast_work); 4315 cancel_delayed_work_sync(&bond->slave_arr_work); 4316 } 4317 4318 static int bond_open(struct net_device *bond_dev) 4319 { 4320 struct bonding *bond = netdev_priv(bond_dev); 4321 struct list_head *iter; 4322 struct slave *slave; 4323 4324 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) { 4325 bond->rr_tx_counter = alloc_percpu(u32); 4326 if (!bond->rr_tx_counter) 4327 return -ENOMEM; 4328 } 4329 4330 /* reset slave->backup and slave->inactive */ 4331 if (bond_has_slaves(bond)) { 4332 bond_for_each_slave(bond, slave, iter) { 4333 if (bond_uses_primary(bond) && 4334 slave != rcu_access_pointer(bond->curr_active_slave)) { 4335 bond_set_slave_inactive_flags(slave, 4336 BOND_SLAVE_NOTIFY_NOW); 4337 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) { 4338 bond_set_slave_active_flags(slave, 4339 BOND_SLAVE_NOTIFY_NOW); 4340 } 4341 } 4342 } 4343 4344 if (bond_is_lb(bond)) { 4345 /* bond_alb_initialize must be called before the timer 4346 * is started. 4347 */ 4348 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB))) 4349 return -ENOMEM; 4350 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB) 4351 queue_delayed_work(bond->wq, &bond->alb_work, 0); 4352 } 4353 4354 if (bond->params.miimon) /* link check interval, in milliseconds. */ 4355 queue_delayed_work(bond->wq, &bond->mii_work, 0); 4356 4357 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */ 4358 queue_delayed_work(bond->wq, &bond->arp_work, 0); 4359 bond->recv_probe = bond_rcv_validate; 4360 } 4361 4362 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 4363 queue_delayed_work(bond->wq, &bond->ad_work, 0); 4364 /* register to receive LACPDUs */ 4365 bond->recv_probe = bond_3ad_lacpdu_recv; 4366 bond_3ad_initiate_agg_selection(bond, 1); 4367 4368 bond_for_each_slave(bond, slave, iter) 4369 dev_mc_add(slave->dev, lacpdu_mcast_addr); 4370 } 4371 4372 if (bond_mode_can_use_xmit_hash(bond)) 4373 bond_update_slave_arr(bond, NULL); 4374 4375 return 0; 4376 } 4377 4378 static int bond_close(struct net_device *bond_dev) 4379 { 4380 struct bonding *bond = netdev_priv(bond_dev); 4381 struct slave *slave; 4382 4383 bond_work_cancel_all(bond); 4384 bond->send_peer_notif = 0; 4385 if (bond_is_lb(bond)) 4386 bond_alb_deinitialize(bond); 4387 bond->recv_probe = NULL; 4388 4389 if (bond_uses_primary(bond)) { 4390 rcu_read_lock(); 4391 slave = rcu_dereference(bond->curr_active_slave); 4392 if (slave) 4393 bond_hw_addr_flush(bond_dev, slave->dev); 4394 rcu_read_unlock(); 4395 } else { 4396 struct list_head *iter; 4397 4398 bond_for_each_slave(bond, slave, iter) 4399 bond_hw_addr_flush(bond_dev, slave->dev); 4400 } 4401 4402 return 0; 4403 } 4404 4405 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but 4406 * that some drivers can provide 32bit values only. 4407 */ 4408 static void bond_fold_stats(struct rtnl_link_stats64 *_res, 4409 const struct rtnl_link_stats64 *_new, 4410 const struct rtnl_link_stats64 *_old) 4411 { 4412 const u64 *new = (const u64 *)_new; 4413 const u64 *old = (const u64 *)_old; 4414 u64 *res = (u64 *)_res; 4415 int i; 4416 4417 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) { 4418 u64 nv = new[i]; 4419 u64 ov = old[i]; 4420 s64 delta = nv - ov; 4421 4422 /* detects if this particular field is 32bit only */ 4423 if (((nv | ov) >> 32) == 0) 4424 delta = (s64)(s32)((u32)nv - (u32)ov); 4425 4426 /* filter anomalies, some drivers reset their stats 4427 * at down/up events. 4428 */ 4429 if (delta > 0) 4430 res[i] += delta; 4431 } 4432 } 4433 4434 #ifdef CONFIG_LOCKDEP 4435 static int bond_get_lowest_level_rcu(struct net_device *dev) 4436 { 4437 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1]; 4438 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1]; 4439 int cur = 0, max = 0; 4440 4441 now = dev; 4442 iter = &dev->adj_list.lower; 4443 4444 while (1) { 4445 next = NULL; 4446 while (1) { 4447 ldev = netdev_next_lower_dev_rcu(now, &iter); 4448 if (!ldev) 4449 break; 4450 4451 next = ldev; 4452 niter = &ldev->adj_list.lower; 4453 dev_stack[cur] = now; 4454 iter_stack[cur++] = iter; 4455 if (max <= cur) 4456 max = cur; 4457 break; 4458 } 4459 4460 if (!next) { 4461 if (!cur) 4462 return max; 4463 next = dev_stack[--cur]; 4464 niter = iter_stack[cur]; 4465 } 4466 4467 now = next; 4468 iter = niter; 4469 } 4470 4471 return max; 4472 } 4473 #endif 4474 4475 static void bond_get_stats(struct net_device *bond_dev, 4476 struct rtnl_link_stats64 *stats) 4477 { 4478 struct bonding *bond = netdev_priv(bond_dev); 4479 struct rtnl_link_stats64 temp; 4480 struct list_head *iter; 4481 struct slave *slave; 4482 int nest_level = 0; 4483 4484 4485 rcu_read_lock(); 4486 #ifdef CONFIG_LOCKDEP 4487 nest_level = bond_get_lowest_level_rcu(bond_dev); 4488 #endif 4489 4490 spin_lock_nested(&bond->stats_lock, nest_level); 4491 memcpy(stats, &bond->bond_stats, sizeof(*stats)); 4492 4493 bond_for_each_slave_rcu(bond, slave, iter) { 4494 const struct rtnl_link_stats64 *new = 4495 dev_get_stats(slave->dev, &temp); 4496 4497 bond_fold_stats(stats, new, &slave->slave_stats); 4498 4499 /* save off the slave stats for the next run */ 4500 memcpy(&slave->slave_stats, new, sizeof(*new)); 4501 } 4502 4503 memcpy(&bond->bond_stats, stats, sizeof(*stats)); 4504 spin_unlock(&bond->stats_lock); 4505 rcu_read_unlock(); 4506 } 4507 4508 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4509 { 4510 struct bonding *bond = netdev_priv(bond_dev); 4511 struct mii_ioctl_data *mii = NULL; 4512 4513 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd); 4514 4515 switch (cmd) { 4516 case SIOCGMIIPHY: 4517 mii = if_mii(ifr); 4518 if (!mii) 4519 return -EINVAL; 4520 4521 mii->phy_id = 0; 4522 fallthrough; 4523 case SIOCGMIIREG: 4524 /* We do this again just in case we were called by SIOCGMIIREG 4525 * instead of SIOCGMIIPHY. 4526 */ 4527 mii = if_mii(ifr); 4528 if (!mii) 4529 return -EINVAL; 4530 4531 if (mii->reg_num == 1) { 4532 mii->val_out = 0; 4533 if (netif_carrier_ok(bond->dev)) 4534 mii->val_out = BMSR_LSTATUS; 4535 } 4536 4537 break; 4538 default: 4539 return -EOPNOTSUPP; 4540 } 4541 4542 return 0; 4543 } 4544 4545 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd) 4546 { 4547 struct bonding *bond = netdev_priv(bond_dev); 4548 struct net_device *slave_dev = NULL; 4549 struct ifbond k_binfo; 4550 struct ifbond __user *u_binfo = NULL; 4551 struct ifslave k_sinfo; 4552 struct ifslave __user *u_sinfo = NULL; 4553 struct bond_opt_value newval; 4554 struct net *net; 4555 int res = 0; 4556 4557 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd); 4558 4559 switch (cmd) { 4560 case SIOCBONDINFOQUERY: 4561 u_binfo = (struct ifbond __user *)ifr->ifr_data; 4562 4563 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) 4564 return -EFAULT; 4565 4566 bond_info_query(bond_dev, &k_binfo); 4567 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) 4568 return -EFAULT; 4569 4570 return 0; 4571 case SIOCBONDSLAVEINFOQUERY: 4572 u_sinfo = (struct ifslave __user *)ifr->ifr_data; 4573 4574 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) 4575 return -EFAULT; 4576 4577 res = bond_slave_info_query(bond_dev, &k_sinfo); 4578 if (res == 0 && 4579 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) 4580 return -EFAULT; 4581 4582 return res; 4583 default: 4584 break; 4585 } 4586 4587 net = dev_net(bond_dev); 4588 4589 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 4590 return -EPERM; 4591 4592 slave_dev = __dev_get_by_name(net, ifr->ifr_slave); 4593 4594 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev); 4595 4596 if (!slave_dev) 4597 return -ENODEV; 4598 4599 switch (cmd) { 4600 case SIOCBONDENSLAVE: 4601 res = bond_enslave(bond_dev, slave_dev, NULL); 4602 break; 4603 case SIOCBONDRELEASE: 4604 res = bond_release(bond_dev, slave_dev); 4605 break; 4606 case SIOCBONDSETHWADDR: 4607 res = bond_set_dev_addr(bond_dev, slave_dev); 4608 break; 4609 case SIOCBONDCHANGEACTIVE: 4610 bond_opt_initstr(&newval, slave_dev->name); 4611 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE, 4612 &newval); 4613 break; 4614 default: 4615 res = -EOPNOTSUPP; 4616 } 4617 4618 return res; 4619 } 4620 4621 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr, 4622 void __user *data, int cmd) 4623 { 4624 struct ifreq ifrdata = { .ifr_data = data }; 4625 4626 switch (cmd) { 4627 case BOND_INFO_QUERY_OLD: 4628 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY); 4629 case BOND_SLAVE_INFO_QUERY_OLD: 4630 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY); 4631 case BOND_ENSLAVE_OLD: 4632 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE); 4633 case BOND_RELEASE_OLD: 4634 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE); 4635 case BOND_SETHWADDR_OLD: 4636 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR); 4637 case BOND_CHANGE_ACTIVE_OLD: 4638 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE); 4639 } 4640 4641 return -EOPNOTSUPP; 4642 } 4643 4644 static void bond_change_rx_flags(struct net_device *bond_dev, int change) 4645 { 4646 struct bonding *bond = netdev_priv(bond_dev); 4647 4648 if (change & IFF_PROMISC) 4649 bond_set_promiscuity(bond, 4650 bond_dev->flags & IFF_PROMISC ? 1 : -1); 4651 4652 if (change & IFF_ALLMULTI) 4653 bond_set_allmulti(bond, 4654 bond_dev->flags & IFF_ALLMULTI ? 1 : -1); 4655 } 4656 4657 static void bond_set_rx_mode(struct net_device *bond_dev) 4658 { 4659 struct bonding *bond = netdev_priv(bond_dev); 4660 struct list_head *iter; 4661 struct slave *slave; 4662 4663 rcu_read_lock(); 4664 if (bond_uses_primary(bond)) { 4665 slave = rcu_dereference(bond->curr_active_slave); 4666 if (slave) { 4667 dev_uc_sync(slave->dev, bond_dev); 4668 dev_mc_sync(slave->dev, bond_dev); 4669 } 4670 } else { 4671 bond_for_each_slave_rcu(bond, slave, iter) { 4672 dev_uc_sync_multiple(slave->dev, bond_dev); 4673 dev_mc_sync_multiple(slave->dev, bond_dev); 4674 } 4675 } 4676 rcu_read_unlock(); 4677 } 4678 4679 static int bond_neigh_init(struct neighbour *n) 4680 { 4681 struct bonding *bond = netdev_priv(n->dev); 4682 const struct net_device_ops *slave_ops; 4683 struct neigh_parms parms; 4684 struct slave *slave; 4685 int ret = 0; 4686 4687 rcu_read_lock(); 4688 slave = bond_first_slave_rcu(bond); 4689 if (!slave) 4690 goto out; 4691 slave_ops = slave->dev->netdev_ops; 4692 if (!slave_ops->ndo_neigh_setup) 4693 goto out; 4694 4695 /* TODO: find another way [1] to implement this. 4696 * Passing a zeroed structure is fragile, 4697 * but at least we do not pass garbage. 4698 * 4699 * [1] One way would be that ndo_neigh_setup() never touch 4700 * struct neigh_parms, but propagate the new neigh_setup() 4701 * back to ___neigh_create() / neigh_parms_alloc() 4702 */ 4703 memset(&parms, 0, sizeof(parms)); 4704 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms); 4705 4706 if (ret) 4707 goto out; 4708 4709 if (parms.neigh_setup) 4710 ret = parms.neigh_setup(n); 4711 out: 4712 rcu_read_unlock(); 4713 return ret; 4714 } 4715 4716 /* The bonding ndo_neigh_setup is called at init time beofre any 4717 * slave exists. So we must declare proxy setup function which will 4718 * be used at run time to resolve the actual slave neigh param setup. 4719 * 4720 * It's also called by master devices (such as vlans) to setup their 4721 * underlying devices. In that case - do nothing, we're already set up from 4722 * our init. 4723 */ 4724 static int bond_neigh_setup(struct net_device *dev, 4725 struct neigh_parms *parms) 4726 { 4727 /* modify only our neigh_parms */ 4728 if (parms->dev == dev) 4729 parms->neigh_setup = bond_neigh_init; 4730 4731 return 0; 4732 } 4733 4734 /* Change the MTU of all of a master's slaves to match the master */ 4735 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu) 4736 { 4737 struct bonding *bond = netdev_priv(bond_dev); 4738 struct slave *slave, *rollback_slave; 4739 struct list_head *iter; 4740 int res = 0; 4741 4742 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu); 4743 4744 bond_for_each_slave(bond, slave, iter) { 4745 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n", 4746 slave, slave->dev->netdev_ops->ndo_change_mtu); 4747 4748 res = dev_set_mtu(slave->dev, new_mtu); 4749 4750 if (res) { 4751 /* If we failed to set the slave's mtu to the new value 4752 * we must abort the operation even in ACTIVE_BACKUP 4753 * mode, because if we allow the backup slaves to have 4754 * different mtu values than the active slave we'll 4755 * need to change their mtu when doing a failover. That 4756 * means changing their mtu from timer context, which 4757 * is probably not a good idea. 4758 */ 4759 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n", 4760 res, new_mtu); 4761 goto unwind; 4762 } 4763 } 4764 4765 bond_dev->mtu = new_mtu; 4766 4767 return 0; 4768 4769 unwind: 4770 /* unwind from head to the slave that failed */ 4771 bond_for_each_slave(bond, rollback_slave, iter) { 4772 int tmp_res; 4773 4774 if (rollback_slave == slave) 4775 break; 4776 4777 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu); 4778 if (tmp_res) 4779 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n", 4780 tmp_res); 4781 } 4782 4783 return res; 4784 } 4785 4786 /* Change HW address 4787 * 4788 * Note that many devices must be down to change the HW address, and 4789 * downing the master releases all slaves. We can make bonds full of 4790 * bonding devices to test this, however. 4791 */ 4792 static int bond_set_mac_address(struct net_device *bond_dev, void *addr) 4793 { 4794 struct bonding *bond = netdev_priv(bond_dev); 4795 struct slave *slave, *rollback_slave; 4796 struct sockaddr_storage *ss = addr, tmp_ss; 4797 struct list_head *iter; 4798 int res = 0; 4799 4800 if (BOND_MODE(bond) == BOND_MODE_ALB) 4801 return bond_alb_set_mac_address(bond_dev, addr); 4802 4803 4804 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond); 4805 4806 /* If fail_over_mac is enabled, do nothing and return success. 4807 * Returning an error causes ifenslave to fail. 4808 */ 4809 if (bond->params.fail_over_mac && 4810 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 4811 return 0; 4812 4813 if (!is_valid_ether_addr(ss->__data)) 4814 return -EADDRNOTAVAIL; 4815 4816 bond_for_each_slave(bond, slave, iter) { 4817 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n", 4818 __func__, slave); 4819 res = dev_set_mac_address(slave->dev, addr, NULL); 4820 if (res) { 4821 /* TODO: consider downing the slave 4822 * and retry ? 4823 * User should expect communications 4824 * breakage anyway until ARP finish 4825 * updating, so... 4826 */ 4827 slave_dbg(bond_dev, slave->dev, "%s: err %d\n", 4828 __func__, res); 4829 goto unwind; 4830 } 4831 } 4832 4833 /* success */ 4834 dev_addr_set(bond_dev, ss->__data); 4835 return 0; 4836 4837 unwind: 4838 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len); 4839 tmp_ss.ss_family = bond_dev->type; 4840 4841 /* unwind from head to the slave that failed */ 4842 bond_for_each_slave(bond, rollback_slave, iter) { 4843 int tmp_res; 4844 4845 if (rollback_slave == slave) 4846 break; 4847 4848 tmp_res = dev_set_mac_address(rollback_slave->dev, 4849 (struct sockaddr *)&tmp_ss, NULL); 4850 if (tmp_res) { 4851 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n", 4852 __func__, tmp_res); 4853 } 4854 } 4855 4856 return res; 4857 } 4858 4859 /** 4860 * bond_get_slave_by_id - get xmit slave with slave_id 4861 * @bond: bonding device that is transmitting 4862 * @slave_id: slave id up to slave_cnt-1 through which to transmit 4863 * 4864 * This function tries to get slave with slave_id but in case 4865 * it fails, it tries to find the first available slave for transmission. 4866 */ 4867 static struct slave *bond_get_slave_by_id(struct bonding *bond, 4868 int slave_id) 4869 { 4870 struct list_head *iter; 4871 struct slave *slave; 4872 int i = slave_id; 4873 4874 /* Here we start from the slave with slave_id */ 4875 bond_for_each_slave_rcu(bond, slave, iter) { 4876 if (--i < 0) { 4877 if (bond_slave_can_tx(slave)) 4878 return slave; 4879 } 4880 } 4881 4882 /* Here we start from the first slave up to slave_id */ 4883 i = slave_id; 4884 bond_for_each_slave_rcu(bond, slave, iter) { 4885 if (--i < 0) 4886 break; 4887 if (bond_slave_can_tx(slave)) 4888 return slave; 4889 } 4890 /* no slave that can tx has been found */ 4891 return NULL; 4892 } 4893 4894 /** 4895 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave 4896 * @bond: bonding device to use 4897 * 4898 * Based on the value of the bonding device's packets_per_slave parameter 4899 * this function generates a slave id, which is usually used as the next 4900 * slave to transmit through. 4901 */ 4902 static u32 bond_rr_gen_slave_id(struct bonding *bond) 4903 { 4904 u32 slave_id; 4905 struct reciprocal_value reciprocal_packets_per_slave; 4906 int packets_per_slave = bond->params.packets_per_slave; 4907 4908 switch (packets_per_slave) { 4909 case 0: 4910 slave_id = get_random_u32(); 4911 break; 4912 case 1: 4913 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4914 break; 4915 default: 4916 reciprocal_packets_per_slave = 4917 bond->params.reciprocal_packets_per_slave; 4918 slave_id = this_cpu_inc_return(*bond->rr_tx_counter); 4919 slave_id = reciprocal_divide(slave_id, 4920 reciprocal_packets_per_slave); 4921 break; 4922 } 4923 4924 return slave_id; 4925 } 4926 4927 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond, 4928 struct sk_buff *skb) 4929 { 4930 struct slave *slave; 4931 int slave_cnt; 4932 u32 slave_id; 4933 4934 /* Start with the curr_active_slave that joined the bond as the 4935 * default for sending IGMP traffic. For failover purposes one 4936 * needs to maintain some consistency for the interface that will 4937 * send the join/membership reports. The curr_active_slave found 4938 * will send all of this type of traffic. 4939 */ 4940 if (skb->protocol == htons(ETH_P_IP)) { 4941 int noff = skb_network_offset(skb); 4942 struct iphdr *iph; 4943 4944 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph)))) 4945 goto non_igmp; 4946 4947 iph = ip_hdr(skb); 4948 if (iph->protocol == IPPROTO_IGMP) { 4949 slave = rcu_dereference(bond->curr_active_slave); 4950 if (slave) 4951 return slave; 4952 return bond_get_slave_by_id(bond, 0); 4953 } 4954 } 4955 4956 non_igmp: 4957 slave_cnt = READ_ONCE(bond->slave_cnt); 4958 if (likely(slave_cnt)) { 4959 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 4960 return bond_get_slave_by_id(bond, slave_id); 4961 } 4962 return NULL; 4963 } 4964 4965 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond, 4966 struct xdp_buff *xdp) 4967 { 4968 struct slave *slave; 4969 int slave_cnt; 4970 u32 slave_id; 4971 const struct ethhdr *eth; 4972 void *data = xdp->data; 4973 4974 if (data + sizeof(struct ethhdr) > xdp->data_end) 4975 goto non_igmp; 4976 4977 eth = (struct ethhdr *)data; 4978 data += sizeof(struct ethhdr); 4979 4980 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */ 4981 if (eth->h_proto == htons(ETH_P_IP)) { 4982 const struct iphdr *iph; 4983 4984 if (data + sizeof(struct iphdr) > xdp->data_end) 4985 goto non_igmp; 4986 4987 iph = (struct iphdr *)data; 4988 4989 if (iph->protocol == IPPROTO_IGMP) { 4990 slave = rcu_dereference(bond->curr_active_slave); 4991 if (slave) 4992 return slave; 4993 return bond_get_slave_by_id(bond, 0); 4994 } 4995 } 4996 4997 non_igmp: 4998 slave_cnt = READ_ONCE(bond->slave_cnt); 4999 if (likely(slave_cnt)) { 5000 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt; 5001 return bond_get_slave_by_id(bond, slave_id); 5002 } 5003 return NULL; 5004 } 5005 5006 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb, 5007 struct net_device *bond_dev) 5008 { 5009 struct bonding *bond = netdev_priv(bond_dev); 5010 struct slave *slave; 5011 5012 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5013 if (likely(slave)) 5014 return bond_dev_queue_xmit(bond, skb, slave->dev); 5015 5016 return bond_tx_drop(bond_dev, skb); 5017 } 5018 5019 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond) 5020 { 5021 return rcu_dereference(bond->curr_active_slave); 5022 } 5023 5024 /* In active-backup mode, we know that bond->curr_active_slave is always valid if 5025 * the bond has a usable interface. 5026 */ 5027 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb, 5028 struct net_device *bond_dev) 5029 { 5030 struct bonding *bond = netdev_priv(bond_dev); 5031 struct slave *slave; 5032 5033 slave = bond_xmit_activebackup_slave_get(bond); 5034 if (slave) 5035 return bond_dev_queue_xmit(bond, skb, slave->dev); 5036 5037 return bond_tx_drop(bond_dev, skb); 5038 } 5039 5040 /* Use this to update slave_array when (a) it's not appropriate to update 5041 * slave_array right away (note that update_slave_array() may sleep) 5042 * and / or (b) RTNL is not held. 5043 */ 5044 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay) 5045 { 5046 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay); 5047 } 5048 5049 /* Slave array work handler. Holds only RTNL */ 5050 static void bond_slave_arr_handler(struct work_struct *work) 5051 { 5052 struct bonding *bond = container_of(work, struct bonding, 5053 slave_arr_work.work); 5054 int ret; 5055 5056 if (!rtnl_trylock()) 5057 goto err; 5058 5059 ret = bond_update_slave_arr(bond, NULL); 5060 rtnl_unlock(); 5061 if (ret) { 5062 pr_warn_ratelimited("Failed to update slave array from WT\n"); 5063 goto err; 5064 } 5065 return; 5066 5067 err: 5068 bond_slave_arr_work_rearm(bond, 1); 5069 } 5070 5071 static void bond_skip_slave(struct bond_up_slave *slaves, 5072 struct slave *skipslave) 5073 { 5074 int idx; 5075 5076 /* Rare situation where caller has asked to skip a specific 5077 * slave but allocation failed (most likely!). BTW this is 5078 * only possible when the call is initiated from 5079 * __bond_release_one(). In this situation; overwrite the 5080 * skipslave entry in the array with the last entry from the 5081 * array to avoid a situation where the xmit path may choose 5082 * this to-be-skipped slave to send a packet out. 5083 */ 5084 for (idx = 0; slaves && idx < slaves->count; idx++) { 5085 if (skipslave == slaves->arr[idx]) { 5086 slaves->arr[idx] = 5087 slaves->arr[slaves->count - 1]; 5088 slaves->count--; 5089 break; 5090 } 5091 } 5092 } 5093 5094 static void bond_set_slave_arr(struct bonding *bond, 5095 struct bond_up_slave *usable_slaves, 5096 struct bond_up_slave *all_slaves) 5097 { 5098 struct bond_up_slave *usable, *all; 5099 5100 usable = rtnl_dereference(bond->usable_slaves); 5101 rcu_assign_pointer(bond->usable_slaves, usable_slaves); 5102 kfree_rcu(usable, rcu); 5103 5104 all = rtnl_dereference(bond->all_slaves); 5105 rcu_assign_pointer(bond->all_slaves, all_slaves); 5106 kfree_rcu(all, rcu); 5107 } 5108 5109 static void bond_reset_slave_arr(struct bonding *bond) 5110 { 5111 bond_set_slave_arr(bond, NULL, NULL); 5112 } 5113 5114 /* Build the usable slaves array in control path for modes that use xmit-hash 5115 * to determine the slave interface - 5116 * (a) BOND_MODE_8023AD 5117 * (b) BOND_MODE_XOR 5118 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0 5119 * 5120 * The caller is expected to hold RTNL only and NO other lock! 5121 */ 5122 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave) 5123 { 5124 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL; 5125 struct slave *slave; 5126 struct list_head *iter; 5127 int agg_id = 0; 5128 int ret = 0; 5129 5130 might_sleep(); 5131 5132 usable_slaves = kzalloc(struct_size(usable_slaves, arr, 5133 bond->slave_cnt), GFP_KERNEL); 5134 all_slaves = kzalloc(struct_size(all_slaves, arr, 5135 bond->slave_cnt), GFP_KERNEL); 5136 if (!usable_slaves || !all_slaves) { 5137 ret = -ENOMEM; 5138 goto out; 5139 } 5140 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5141 struct ad_info ad_info; 5142 5143 spin_lock_bh(&bond->mode_lock); 5144 if (bond_3ad_get_active_agg_info(bond, &ad_info)) { 5145 spin_unlock_bh(&bond->mode_lock); 5146 pr_debug("bond_3ad_get_active_agg_info failed\n"); 5147 /* No active aggragator means it's not safe to use 5148 * the previous array. 5149 */ 5150 bond_reset_slave_arr(bond); 5151 goto out; 5152 } 5153 spin_unlock_bh(&bond->mode_lock); 5154 agg_id = ad_info.aggregator_id; 5155 } 5156 bond_for_each_slave(bond, slave, iter) { 5157 if (skipslave == slave) 5158 continue; 5159 5160 all_slaves->arr[all_slaves->count++] = slave; 5161 if (BOND_MODE(bond) == BOND_MODE_8023AD) { 5162 struct aggregator *agg; 5163 5164 agg = SLAVE_AD_INFO(slave)->port.aggregator; 5165 if (!agg || agg->aggregator_identifier != agg_id) 5166 continue; 5167 } 5168 if (!bond_slave_can_tx(slave)) 5169 continue; 5170 5171 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n", 5172 usable_slaves->count); 5173 5174 usable_slaves->arr[usable_slaves->count++] = slave; 5175 } 5176 5177 bond_set_slave_arr(bond, usable_slaves, all_slaves); 5178 return ret; 5179 out: 5180 if (ret != 0 && skipslave) { 5181 bond_skip_slave(rtnl_dereference(bond->all_slaves), 5182 skipslave); 5183 bond_skip_slave(rtnl_dereference(bond->usable_slaves), 5184 skipslave); 5185 } 5186 kfree_rcu(all_slaves, rcu); 5187 kfree_rcu(usable_slaves, rcu); 5188 5189 return ret; 5190 } 5191 5192 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond, 5193 struct sk_buff *skb, 5194 struct bond_up_slave *slaves) 5195 { 5196 struct slave *slave; 5197 unsigned int count; 5198 u32 hash; 5199 5200 hash = bond_xmit_hash(bond, skb); 5201 count = slaves ? READ_ONCE(slaves->count) : 0; 5202 if (unlikely(!count)) 5203 return NULL; 5204 5205 slave = slaves->arr[hash % count]; 5206 return slave; 5207 } 5208 5209 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond, 5210 struct xdp_buff *xdp) 5211 { 5212 struct bond_up_slave *slaves; 5213 unsigned int count; 5214 u32 hash; 5215 5216 hash = bond_xmit_hash_xdp(bond, xdp); 5217 slaves = rcu_dereference(bond->usable_slaves); 5218 count = slaves ? READ_ONCE(slaves->count) : 0; 5219 if (unlikely(!count)) 5220 return NULL; 5221 5222 return slaves->arr[hash % count]; 5223 } 5224 5225 /* Use this Xmit function for 3AD as well as XOR modes. The current 5226 * usable slave array is formed in the control path. The xmit function 5227 * just calculates hash and sends the packet out. 5228 */ 5229 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb, 5230 struct net_device *dev) 5231 { 5232 struct bonding *bond = netdev_priv(dev); 5233 struct bond_up_slave *slaves; 5234 struct slave *slave; 5235 5236 slaves = rcu_dereference(bond->usable_slaves); 5237 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5238 if (likely(slave)) 5239 return bond_dev_queue_xmit(bond, skb, slave->dev); 5240 5241 return bond_tx_drop(dev, skb); 5242 } 5243 5244 /* in broadcast mode, we send everything to all usable interfaces. */ 5245 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb, 5246 struct net_device *bond_dev) 5247 { 5248 struct bonding *bond = netdev_priv(bond_dev); 5249 struct slave *slave = NULL; 5250 struct list_head *iter; 5251 bool xmit_suc = false; 5252 bool skb_used = false; 5253 5254 bond_for_each_slave_rcu(bond, slave, iter) { 5255 struct sk_buff *skb2; 5256 5257 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)) 5258 continue; 5259 5260 if (bond_is_last_slave(bond, slave)) { 5261 skb2 = skb; 5262 skb_used = true; 5263 } else { 5264 skb2 = skb_clone(skb, GFP_ATOMIC); 5265 if (!skb2) { 5266 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n", 5267 bond_dev->name, __func__); 5268 continue; 5269 } 5270 } 5271 5272 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK) 5273 xmit_suc = true; 5274 } 5275 5276 if (!skb_used) 5277 dev_kfree_skb_any(skb); 5278 5279 if (xmit_suc) 5280 return NETDEV_TX_OK; 5281 5282 dev_core_stats_tx_dropped_inc(bond_dev); 5283 return NET_XMIT_DROP; 5284 } 5285 5286 /*------------------------- Device initialization ---------------------------*/ 5287 5288 /* Lookup the slave that corresponds to a qid */ 5289 static inline int bond_slave_override(struct bonding *bond, 5290 struct sk_buff *skb) 5291 { 5292 struct slave *slave = NULL; 5293 struct list_head *iter; 5294 5295 if (!skb_rx_queue_recorded(skb)) 5296 return 1; 5297 5298 /* Find out if any slaves have the same mapping as this skb. */ 5299 bond_for_each_slave_rcu(bond, slave, iter) { 5300 if (slave->queue_id == skb_get_queue_mapping(skb)) { 5301 if (bond_slave_is_up(slave) && 5302 slave->link == BOND_LINK_UP) { 5303 bond_dev_queue_xmit(bond, skb, slave->dev); 5304 return 0; 5305 } 5306 /* If the slave isn't UP, use default transmit policy. */ 5307 break; 5308 } 5309 } 5310 5311 return 1; 5312 } 5313 5314 5315 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb, 5316 struct net_device *sb_dev) 5317 { 5318 /* This helper function exists to help dev_pick_tx get the correct 5319 * destination queue. Using a helper function skips a call to 5320 * skb_tx_hash and will put the skbs in the queue we expect on their 5321 * way down to the bonding driver. 5322 */ 5323 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0; 5324 5325 /* Save the original txq to restore before passing to the driver */ 5326 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb); 5327 5328 if (unlikely(txq >= dev->real_num_tx_queues)) { 5329 do { 5330 txq -= dev->real_num_tx_queues; 5331 } while (txq >= dev->real_num_tx_queues); 5332 } 5333 return txq; 5334 } 5335 5336 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev, 5337 struct sk_buff *skb, 5338 bool all_slaves) 5339 { 5340 struct bonding *bond = netdev_priv(master_dev); 5341 struct bond_up_slave *slaves; 5342 struct slave *slave = NULL; 5343 5344 switch (BOND_MODE(bond)) { 5345 case BOND_MODE_ROUNDROBIN: 5346 slave = bond_xmit_roundrobin_slave_get(bond, skb); 5347 break; 5348 case BOND_MODE_ACTIVEBACKUP: 5349 slave = bond_xmit_activebackup_slave_get(bond); 5350 break; 5351 case BOND_MODE_8023AD: 5352 case BOND_MODE_XOR: 5353 if (all_slaves) 5354 slaves = rcu_dereference(bond->all_slaves); 5355 else 5356 slaves = rcu_dereference(bond->usable_slaves); 5357 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves); 5358 break; 5359 case BOND_MODE_BROADCAST: 5360 break; 5361 case BOND_MODE_ALB: 5362 slave = bond_xmit_alb_slave_get(bond, skb); 5363 break; 5364 case BOND_MODE_TLB: 5365 slave = bond_xmit_tlb_slave_get(bond, skb); 5366 break; 5367 default: 5368 /* Should never happen, mode already checked */ 5369 WARN_ONCE(true, "Unknown bonding mode"); 5370 break; 5371 } 5372 5373 if (slave) 5374 return slave->dev; 5375 return NULL; 5376 } 5377 5378 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow) 5379 { 5380 switch (sk->sk_family) { 5381 #if IS_ENABLED(CONFIG_IPV6) 5382 case AF_INET6: 5383 if (ipv6_only_sock(sk) || 5384 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) { 5385 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 5386 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr; 5387 flow->addrs.v6addrs.dst = sk->sk_v6_daddr; 5388 break; 5389 } 5390 fallthrough; 5391 #endif 5392 default: /* AF_INET */ 5393 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 5394 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr; 5395 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr; 5396 break; 5397 } 5398 5399 flow->ports.src = inet_sk(sk)->inet_sport; 5400 flow->ports.dst = inet_sk(sk)->inet_dport; 5401 } 5402 5403 /** 5404 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields 5405 * @sk: socket to use for headers 5406 * 5407 * This function will extract the necessary field from the socket and use 5408 * them to generate a hash based on the LAYER34 xmit_policy. 5409 * Assumes that sk is a TCP or UDP socket. 5410 */ 5411 static u32 bond_sk_hash_l34(struct sock *sk) 5412 { 5413 struct flow_keys flow; 5414 u32 hash; 5415 5416 bond_sk_to_flow(sk, &flow); 5417 5418 /* L4 */ 5419 memcpy(&hash, &flow.ports.ports, sizeof(hash)); 5420 /* L3 */ 5421 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34); 5422 } 5423 5424 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond, 5425 struct sock *sk) 5426 { 5427 struct bond_up_slave *slaves; 5428 struct slave *slave; 5429 unsigned int count; 5430 u32 hash; 5431 5432 slaves = rcu_dereference(bond->usable_slaves); 5433 count = slaves ? READ_ONCE(slaves->count) : 0; 5434 if (unlikely(!count)) 5435 return NULL; 5436 5437 hash = bond_sk_hash_l34(sk); 5438 slave = slaves->arr[hash % count]; 5439 5440 return slave->dev; 5441 } 5442 5443 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev, 5444 struct sock *sk) 5445 { 5446 struct bonding *bond = netdev_priv(dev); 5447 struct net_device *lower = NULL; 5448 5449 rcu_read_lock(); 5450 if (bond_sk_check(bond)) 5451 lower = __bond_sk_get_lower_dev(bond, sk); 5452 rcu_read_unlock(); 5453 5454 return lower; 5455 } 5456 5457 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5458 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb, 5459 struct net_device *dev) 5460 { 5461 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev); 5462 5463 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded 5464 * was true, if tls_device_down is running in parallel, but it's OK, 5465 * because bond_get_slave_by_dev has a NULL check. 5466 */ 5467 if (likely(bond_get_slave_by_dev(bond, tls_netdev))) 5468 return bond_dev_queue_xmit(bond, skb, tls_netdev); 5469 return bond_tx_drop(dev, skb); 5470 } 5471 #endif 5472 5473 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5474 { 5475 struct bonding *bond = netdev_priv(dev); 5476 5477 if (bond_should_override_tx_queue(bond) && 5478 !bond_slave_override(bond, skb)) 5479 return NETDEV_TX_OK; 5480 5481 #if IS_ENABLED(CONFIG_TLS_DEVICE) 5482 if (tls_is_skb_tx_device_offloaded(skb)) 5483 return bond_tls_device_xmit(bond, skb, dev); 5484 #endif 5485 5486 switch (BOND_MODE(bond)) { 5487 case BOND_MODE_ROUNDROBIN: 5488 return bond_xmit_roundrobin(skb, dev); 5489 case BOND_MODE_ACTIVEBACKUP: 5490 return bond_xmit_activebackup(skb, dev); 5491 case BOND_MODE_8023AD: 5492 case BOND_MODE_XOR: 5493 return bond_3ad_xor_xmit(skb, dev); 5494 case BOND_MODE_BROADCAST: 5495 return bond_xmit_broadcast(skb, dev); 5496 case BOND_MODE_ALB: 5497 return bond_alb_xmit(skb, dev); 5498 case BOND_MODE_TLB: 5499 return bond_tlb_xmit(skb, dev); 5500 default: 5501 /* Should never happen, mode already checked */ 5502 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond)); 5503 WARN_ON_ONCE(1); 5504 return bond_tx_drop(dev, skb); 5505 } 5506 } 5507 5508 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev) 5509 { 5510 struct bonding *bond = netdev_priv(dev); 5511 netdev_tx_t ret = NETDEV_TX_OK; 5512 5513 /* If we risk deadlock from transmitting this in the 5514 * netpoll path, tell netpoll to queue the frame for later tx 5515 */ 5516 if (unlikely(is_netpoll_tx_blocked(dev))) 5517 return NETDEV_TX_BUSY; 5518 5519 rcu_read_lock(); 5520 if (bond_has_slaves(bond)) 5521 ret = __bond_start_xmit(skb, dev); 5522 else 5523 ret = bond_tx_drop(dev, skb); 5524 rcu_read_unlock(); 5525 5526 return ret; 5527 } 5528 5529 static struct net_device * 5530 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp) 5531 { 5532 struct bonding *bond = netdev_priv(bond_dev); 5533 struct slave *slave; 5534 5535 /* Caller needs to hold rcu_read_lock() */ 5536 5537 switch (BOND_MODE(bond)) { 5538 case BOND_MODE_ROUNDROBIN: 5539 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp); 5540 break; 5541 5542 case BOND_MODE_ACTIVEBACKUP: 5543 slave = bond_xmit_activebackup_slave_get(bond); 5544 break; 5545 5546 case BOND_MODE_8023AD: 5547 case BOND_MODE_XOR: 5548 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp); 5549 break; 5550 5551 default: 5552 if (net_ratelimit()) 5553 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", 5554 BOND_MODE(bond)); 5555 return NULL; 5556 } 5557 5558 if (slave) 5559 return slave->dev; 5560 5561 return NULL; 5562 } 5563 5564 static int bond_xdp_xmit(struct net_device *bond_dev, 5565 int n, struct xdp_frame **frames, u32 flags) 5566 { 5567 int nxmit, err = -ENXIO; 5568 5569 rcu_read_lock(); 5570 5571 for (nxmit = 0; nxmit < n; nxmit++) { 5572 struct xdp_frame *frame = frames[nxmit]; 5573 struct xdp_frame *frames1[] = {frame}; 5574 struct net_device *slave_dev; 5575 struct xdp_buff xdp; 5576 5577 xdp_convert_frame_to_buff(frame, &xdp); 5578 5579 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp); 5580 if (!slave_dev) { 5581 err = -ENXIO; 5582 break; 5583 } 5584 5585 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags); 5586 if (err < 1) 5587 break; 5588 } 5589 5590 rcu_read_unlock(); 5591 5592 /* If error happened on the first frame then we can pass the error up, otherwise 5593 * report the number of frames that were xmitted. 5594 */ 5595 if (err < 0) 5596 return (nxmit == 0 ? err : nxmit); 5597 5598 return nxmit; 5599 } 5600 5601 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog, 5602 struct netlink_ext_ack *extack) 5603 { 5604 struct bonding *bond = netdev_priv(dev); 5605 struct list_head *iter; 5606 struct slave *slave, *rollback_slave; 5607 struct bpf_prog *old_prog; 5608 struct netdev_bpf xdp = { 5609 .command = XDP_SETUP_PROG, 5610 .flags = 0, 5611 .prog = prog, 5612 .extack = extack, 5613 }; 5614 int err; 5615 5616 ASSERT_RTNL(); 5617 5618 if (!bond_xdp_check(bond)) 5619 return -EOPNOTSUPP; 5620 5621 old_prog = bond->xdp_prog; 5622 bond->xdp_prog = prog; 5623 5624 bond_for_each_slave(bond, slave, iter) { 5625 struct net_device *slave_dev = slave->dev; 5626 5627 if (!slave_dev->netdev_ops->ndo_bpf || 5628 !slave_dev->netdev_ops->ndo_xdp_xmit) { 5629 SLAVE_NL_ERR(dev, slave_dev, extack, 5630 "Slave device does not support XDP"); 5631 err = -EOPNOTSUPP; 5632 goto err; 5633 } 5634 5635 if (dev_xdp_prog_count(slave_dev) > 0) { 5636 SLAVE_NL_ERR(dev, slave_dev, extack, 5637 "Slave has XDP program loaded, please unload before enslaving"); 5638 err = -EOPNOTSUPP; 5639 goto err; 5640 } 5641 5642 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5643 if (err < 0) { 5644 /* ndo_bpf() sets extack error message */ 5645 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err); 5646 goto err; 5647 } 5648 if (prog) 5649 bpf_prog_inc(prog); 5650 } 5651 5652 if (prog) { 5653 static_branch_inc(&bpf_master_redirect_enabled_key); 5654 } else if (old_prog) { 5655 bpf_prog_put(old_prog); 5656 static_branch_dec(&bpf_master_redirect_enabled_key); 5657 } 5658 5659 return 0; 5660 5661 err: 5662 /* unwind the program changes */ 5663 bond->xdp_prog = old_prog; 5664 xdp.prog = old_prog; 5665 xdp.extack = NULL; /* do not overwrite original error */ 5666 5667 bond_for_each_slave(bond, rollback_slave, iter) { 5668 struct net_device *slave_dev = rollback_slave->dev; 5669 int err_unwind; 5670 5671 if (slave == rollback_slave) 5672 break; 5673 5674 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp); 5675 if (err_unwind < 0) 5676 slave_err(dev, slave_dev, 5677 "Error %d when unwinding XDP program change\n", err_unwind); 5678 else if (xdp.prog) 5679 bpf_prog_inc(xdp.prog); 5680 } 5681 return err; 5682 } 5683 5684 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp) 5685 { 5686 switch (xdp->command) { 5687 case XDP_SETUP_PROG: 5688 return bond_xdp_set(dev, xdp->prog, xdp->extack); 5689 default: 5690 return -EINVAL; 5691 } 5692 } 5693 5694 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed) 5695 { 5696 if (speed == 0 || speed == SPEED_UNKNOWN) 5697 speed = slave->speed; 5698 else 5699 speed = min(speed, slave->speed); 5700 5701 return speed; 5702 } 5703 5704 /* Set the BOND_PHC_INDEX flag to notify user space */ 5705 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg) 5706 { 5707 struct ifreq *ifr = kernel_cfg->ifr; 5708 struct hwtstamp_config cfg; 5709 5710 if (kernel_cfg->copied_to_user) { 5711 /* Lower device has a legacy implementation */ 5712 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg))) 5713 return -EFAULT; 5714 5715 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5716 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg))) 5717 return -EFAULT; 5718 } else { 5719 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX; 5720 } 5721 5722 return 0; 5723 } 5724 5725 static int bond_hwtstamp_get(struct net_device *dev, 5726 struct kernel_hwtstamp_config *cfg) 5727 { 5728 struct bonding *bond = netdev_priv(dev); 5729 struct net_device *real_dev; 5730 int err; 5731 5732 real_dev = bond_option_active_slave_get_rcu(bond); 5733 if (!real_dev) 5734 return -EOPNOTSUPP; 5735 5736 err = generic_hwtstamp_get_lower(real_dev, cfg); 5737 if (err) 5738 return err; 5739 5740 return bond_set_phc_index_flag(cfg); 5741 } 5742 5743 static int bond_hwtstamp_set(struct net_device *dev, 5744 struct kernel_hwtstamp_config *cfg, 5745 struct netlink_ext_ack *extack) 5746 { 5747 struct bonding *bond = netdev_priv(dev); 5748 struct net_device *real_dev; 5749 int err; 5750 5751 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX)) 5752 return -EOPNOTSUPP; 5753 5754 real_dev = bond_option_active_slave_get_rcu(bond); 5755 if (!real_dev) 5756 return -EOPNOTSUPP; 5757 5758 err = generic_hwtstamp_set_lower(real_dev, cfg, extack); 5759 if (err) 5760 return err; 5761 5762 return bond_set_phc_index_flag(cfg); 5763 } 5764 5765 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev, 5766 struct ethtool_link_ksettings *cmd) 5767 { 5768 struct bonding *bond = netdev_priv(bond_dev); 5769 struct list_head *iter; 5770 struct slave *slave; 5771 u32 speed = 0; 5772 5773 cmd->base.duplex = DUPLEX_UNKNOWN; 5774 cmd->base.port = PORT_OTHER; 5775 5776 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we 5777 * do not need to check mode. Though link speed might not represent 5778 * the true receive or transmit bandwidth (not all modes are symmetric) 5779 * this is an accurate maximum. 5780 */ 5781 bond_for_each_slave(bond, slave, iter) { 5782 if (bond_slave_can_tx(slave)) { 5783 bond_update_speed_duplex(slave); 5784 if (slave->speed != SPEED_UNKNOWN) { 5785 if (BOND_MODE(bond) == BOND_MODE_BROADCAST) 5786 speed = bond_mode_bcast_speed(slave, 5787 speed); 5788 else 5789 speed += slave->speed; 5790 } 5791 if (cmd->base.duplex == DUPLEX_UNKNOWN && 5792 slave->duplex != DUPLEX_UNKNOWN) 5793 cmd->base.duplex = slave->duplex; 5794 } 5795 } 5796 cmd->base.speed = speed ? : SPEED_UNKNOWN; 5797 5798 return 0; 5799 } 5800 5801 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev, 5802 struct ethtool_drvinfo *drvinfo) 5803 { 5804 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver)); 5805 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d", 5806 BOND_ABI_VERSION); 5807 } 5808 5809 static int bond_ethtool_get_ts_info(struct net_device *bond_dev, 5810 struct ethtool_ts_info *info) 5811 { 5812 struct bonding *bond = netdev_priv(bond_dev); 5813 struct ethtool_ts_info ts_info; 5814 const struct ethtool_ops *ops; 5815 struct net_device *real_dev; 5816 bool sw_tx_support = false; 5817 struct phy_device *phydev; 5818 struct list_head *iter; 5819 struct slave *slave; 5820 int ret = 0; 5821 5822 rcu_read_lock(); 5823 real_dev = bond_option_active_slave_get_rcu(bond); 5824 dev_hold(real_dev); 5825 rcu_read_unlock(); 5826 5827 if (real_dev) { 5828 ops = real_dev->ethtool_ops; 5829 phydev = real_dev->phydev; 5830 5831 if (phy_has_tsinfo(phydev)) { 5832 ret = phy_ts_info(phydev, info); 5833 goto out; 5834 } else if (ops->get_ts_info) { 5835 ret = ops->get_ts_info(real_dev, info); 5836 goto out; 5837 } 5838 } else { 5839 /* Check if all slaves support software tx timestamping */ 5840 rcu_read_lock(); 5841 bond_for_each_slave_rcu(bond, slave, iter) { 5842 ret = -1; 5843 ops = slave->dev->ethtool_ops; 5844 phydev = slave->dev->phydev; 5845 5846 if (phy_has_tsinfo(phydev)) 5847 ret = phy_ts_info(phydev, &ts_info); 5848 else if (ops->get_ts_info) 5849 ret = ops->get_ts_info(slave->dev, &ts_info); 5850 5851 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) { 5852 sw_tx_support = true; 5853 continue; 5854 } 5855 5856 sw_tx_support = false; 5857 break; 5858 } 5859 rcu_read_unlock(); 5860 } 5861 5862 ret = 0; 5863 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE | 5864 SOF_TIMESTAMPING_SOFTWARE; 5865 if (sw_tx_support) 5866 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE; 5867 5868 info->phc_index = -1; 5869 5870 out: 5871 dev_put(real_dev); 5872 return ret; 5873 } 5874 5875 static const struct ethtool_ops bond_ethtool_ops = { 5876 .get_drvinfo = bond_ethtool_get_drvinfo, 5877 .get_link = ethtool_op_get_link, 5878 .get_link_ksettings = bond_ethtool_get_link_ksettings, 5879 .get_ts_info = bond_ethtool_get_ts_info, 5880 }; 5881 5882 static const struct net_device_ops bond_netdev_ops = { 5883 .ndo_init = bond_init, 5884 .ndo_uninit = bond_uninit, 5885 .ndo_open = bond_open, 5886 .ndo_stop = bond_close, 5887 .ndo_start_xmit = bond_start_xmit, 5888 .ndo_select_queue = bond_select_queue, 5889 .ndo_get_stats64 = bond_get_stats, 5890 .ndo_eth_ioctl = bond_eth_ioctl, 5891 .ndo_siocbond = bond_do_ioctl, 5892 .ndo_siocdevprivate = bond_siocdevprivate, 5893 .ndo_change_rx_flags = bond_change_rx_flags, 5894 .ndo_set_rx_mode = bond_set_rx_mode, 5895 .ndo_change_mtu = bond_change_mtu, 5896 .ndo_set_mac_address = bond_set_mac_address, 5897 .ndo_neigh_setup = bond_neigh_setup, 5898 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid, 5899 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid, 5900 #ifdef CONFIG_NET_POLL_CONTROLLER 5901 .ndo_netpoll_setup = bond_netpoll_setup, 5902 .ndo_netpoll_cleanup = bond_netpoll_cleanup, 5903 .ndo_poll_controller = bond_poll_controller, 5904 #endif 5905 .ndo_add_slave = bond_enslave, 5906 .ndo_del_slave = bond_release, 5907 .ndo_fix_features = bond_fix_features, 5908 .ndo_features_check = passthru_features_check, 5909 .ndo_get_xmit_slave = bond_xmit_get_slave, 5910 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev, 5911 .ndo_bpf = bond_xdp, 5912 .ndo_xdp_xmit = bond_xdp_xmit, 5913 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave, 5914 .ndo_hwtstamp_get = bond_hwtstamp_get, 5915 .ndo_hwtstamp_set = bond_hwtstamp_set, 5916 }; 5917 5918 static const struct device_type bond_type = { 5919 .name = "bond", 5920 }; 5921 5922 static void bond_destructor(struct net_device *bond_dev) 5923 { 5924 struct bonding *bond = netdev_priv(bond_dev); 5925 5926 if (bond->wq) 5927 destroy_workqueue(bond->wq); 5928 5929 free_percpu(bond->rr_tx_counter); 5930 } 5931 5932 void bond_setup(struct net_device *bond_dev) 5933 { 5934 struct bonding *bond = netdev_priv(bond_dev); 5935 5936 spin_lock_init(&bond->mode_lock); 5937 bond->params = bonding_defaults; 5938 5939 /* Initialize pointers */ 5940 bond->dev = bond_dev; 5941 5942 /* Initialize the device entry points */ 5943 ether_setup(bond_dev); 5944 bond_dev->max_mtu = ETH_MAX_MTU; 5945 bond_dev->netdev_ops = &bond_netdev_ops; 5946 bond_dev->ethtool_ops = &bond_ethtool_ops; 5947 5948 bond_dev->needs_free_netdev = true; 5949 bond_dev->priv_destructor = bond_destructor; 5950 5951 SET_NETDEV_DEVTYPE(bond_dev, &bond_type); 5952 5953 /* Initialize the device options */ 5954 bond_dev->flags |= IFF_MASTER; 5955 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE; 5956 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING); 5957 5958 #ifdef CONFIG_XFRM_OFFLOAD 5959 /* set up xfrm device ops (only supported in active-backup right now) */ 5960 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops; 5961 INIT_LIST_HEAD(&bond->ipsec_list); 5962 mutex_init(&bond->ipsec_lock); 5963 #endif /* CONFIG_XFRM_OFFLOAD */ 5964 5965 /* don't acquire bond device's netif_tx_lock when transmitting */ 5966 bond_dev->features |= NETIF_F_LLTX; 5967 5968 /* By default, we declare the bond to be fully 5969 * VLAN hardware accelerated capable. Special 5970 * care is taken in the various xmit functions 5971 * when there are slaves that are not hw accel 5972 * capable 5973 */ 5974 5975 /* Don't allow bond devices to change network namespaces. */ 5976 bond_dev->features |= NETIF_F_NETNS_LOCAL; 5977 5978 bond_dev->hw_features = BOND_VLAN_FEATURES | 5979 NETIF_F_HW_VLAN_CTAG_RX | 5980 NETIF_F_HW_VLAN_CTAG_FILTER | 5981 NETIF_F_HW_VLAN_STAG_RX | 5982 NETIF_F_HW_VLAN_STAG_FILTER; 5983 5984 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL; 5985 bond_dev->features |= bond_dev->hw_features; 5986 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX; 5987 #ifdef CONFIG_XFRM_OFFLOAD 5988 bond_dev->hw_features |= BOND_XFRM_FEATURES; 5989 /* Only enable XFRM features if this is an active-backup config */ 5990 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) 5991 bond_dev->features |= BOND_XFRM_FEATURES; 5992 #endif /* CONFIG_XFRM_OFFLOAD */ 5993 } 5994 5995 /* Destroy a bonding device. 5996 * Must be under rtnl_lock when this function is called. 5997 */ 5998 static void bond_uninit(struct net_device *bond_dev) 5999 { 6000 struct bonding *bond = netdev_priv(bond_dev); 6001 struct list_head *iter; 6002 struct slave *slave; 6003 6004 bond_netpoll_cleanup(bond_dev); 6005 6006 /* Release the bonded slaves */ 6007 bond_for_each_slave(bond, slave, iter) 6008 __bond_release_one(bond_dev, slave->dev, true, true); 6009 netdev_info(bond_dev, "Released all slaves\n"); 6010 6011 #ifdef CONFIG_XFRM_OFFLOAD 6012 mutex_destroy(&bond->ipsec_lock); 6013 #endif /* CONFIG_XFRM_OFFLOAD */ 6014 6015 bond_set_slave_arr(bond, NULL, NULL); 6016 6017 list_del(&bond->bond_list); 6018 6019 bond_debug_unregister(bond); 6020 } 6021 6022 /*------------------------- Module initialization ---------------------------*/ 6023 6024 static int __init bond_check_params(struct bond_params *params) 6025 { 6026 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i; 6027 struct bond_opt_value newval; 6028 const struct bond_opt_value *valptr; 6029 int arp_all_targets_value = 0; 6030 u16 ad_actor_sys_prio = 0; 6031 u16 ad_user_port_key = 0; 6032 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 }; 6033 int arp_ip_count; 6034 int bond_mode = BOND_MODE_ROUNDROBIN; 6035 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2; 6036 int lacp_fast = 0; 6037 int tlb_dynamic_lb; 6038 6039 /* Convert string parameters. */ 6040 if (mode) { 6041 bond_opt_initstr(&newval, mode); 6042 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval); 6043 if (!valptr) { 6044 pr_err("Error: Invalid bonding mode \"%s\"\n", mode); 6045 return -EINVAL; 6046 } 6047 bond_mode = valptr->value; 6048 } 6049 6050 if (xmit_hash_policy) { 6051 if (bond_mode == BOND_MODE_ROUNDROBIN || 6052 bond_mode == BOND_MODE_ACTIVEBACKUP || 6053 bond_mode == BOND_MODE_BROADCAST) { 6054 pr_info("xmit_hash_policy param is irrelevant in mode %s\n", 6055 bond_mode_name(bond_mode)); 6056 } else { 6057 bond_opt_initstr(&newval, xmit_hash_policy); 6058 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH), 6059 &newval); 6060 if (!valptr) { 6061 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n", 6062 xmit_hash_policy); 6063 return -EINVAL; 6064 } 6065 xmit_hashtype = valptr->value; 6066 } 6067 } 6068 6069 if (lacp_rate) { 6070 if (bond_mode != BOND_MODE_8023AD) { 6071 pr_info("lacp_rate param is irrelevant in mode %s\n", 6072 bond_mode_name(bond_mode)); 6073 } else { 6074 bond_opt_initstr(&newval, lacp_rate); 6075 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE), 6076 &newval); 6077 if (!valptr) { 6078 pr_err("Error: Invalid lacp rate \"%s\"\n", 6079 lacp_rate); 6080 return -EINVAL; 6081 } 6082 lacp_fast = valptr->value; 6083 } 6084 } 6085 6086 if (ad_select) { 6087 bond_opt_initstr(&newval, ad_select); 6088 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT), 6089 &newval); 6090 if (!valptr) { 6091 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select); 6092 return -EINVAL; 6093 } 6094 params->ad_select = valptr->value; 6095 if (bond_mode != BOND_MODE_8023AD) 6096 pr_warn("ad_select param only affects 802.3ad mode\n"); 6097 } else { 6098 params->ad_select = BOND_AD_STABLE; 6099 } 6100 6101 if (max_bonds < 0) { 6102 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n", 6103 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS); 6104 max_bonds = BOND_DEFAULT_MAX_BONDS; 6105 } 6106 6107 if (miimon < 0) { 6108 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6109 miimon, INT_MAX); 6110 miimon = 0; 6111 } 6112 6113 if (updelay < 0) { 6114 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6115 updelay, INT_MAX); 6116 updelay = 0; 6117 } 6118 6119 if (downdelay < 0) { 6120 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6121 downdelay, INT_MAX); 6122 downdelay = 0; 6123 } 6124 6125 if ((use_carrier != 0) && (use_carrier != 1)) { 6126 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n", 6127 use_carrier); 6128 use_carrier = 1; 6129 } 6130 6131 if (num_peer_notif < 0 || num_peer_notif > 255) { 6132 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n", 6133 num_peer_notif); 6134 num_peer_notif = 1; 6135 } 6136 6137 /* reset values for 802.3ad/TLB/ALB */ 6138 if (!bond_mode_uses_arp(bond_mode)) { 6139 if (!miimon) { 6140 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"); 6141 pr_warn("Forcing miimon to 100msec\n"); 6142 miimon = BOND_DEFAULT_MIIMON; 6143 } 6144 } 6145 6146 if (tx_queues < 1 || tx_queues > 255) { 6147 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n", 6148 tx_queues, BOND_DEFAULT_TX_QUEUES); 6149 tx_queues = BOND_DEFAULT_TX_QUEUES; 6150 } 6151 6152 if ((all_slaves_active != 0) && (all_slaves_active != 1)) { 6153 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n", 6154 all_slaves_active); 6155 all_slaves_active = 0; 6156 } 6157 6158 if (resend_igmp < 0 || resend_igmp > 255) { 6159 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n", 6160 resend_igmp, BOND_DEFAULT_RESEND_IGMP); 6161 resend_igmp = BOND_DEFAULT_RESEND_IGMP; 6162 } 6163 6164 bond_opt_initval(&newval, packets_per_slave); 6165 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) { 6166 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n", 6167 packets_per_slave, USHRT_MAX); 6168 packets_per_slave = 1; 6169 } 6170 6171 if (bond_mode == BOND_MODE_ALB) { 6172 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", 6173 updelay); 6174 } 6175 6176 if (!miimon) { 6177 if (updelay || downdelay) { 6178 /* just warn the user the up/down delay will have 6179 * no effect since miimon is zero... 6180 */ 6181 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", 6182 updelay, downdelay); 6183 } 6184 } else { 6185 /* don't allow arp monitoring */ 6186 if (arp_interval) { 6187 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n", 6188 miimon, arp_interval); 6189 arp_interval = 0; 6190 } 6191 6192 if ((updelay % miimon) != 0) { 6193 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n", 6194 updelay, miimon, (updelay / miimon) * miimon); 6195 } 6196 6197 updelay /= miimon; 6198 6199 if ((downdelay % miimon) != 0) { 6200 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n", 6201 downdelay, miimon, 6202 (downdelay / miimon) * miimon); 6203 } 6204 6205 downdelay /= miimon; 6206 } 6207 6208 if (arp_interval < 0) { 6209 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n", 6210 arp_interval, INT_MAX); 6211 arp_interval = 0; 6212 } 6213 6214 for (arp_ip_count = 0, i = 0; 6215 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) { 6216 __be32 ip; 6217 6218 /* not a complete check, but good enough to catch mistakes */ 6219 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) || 6220 !bond_is_ip_target_ok(ip)) { 6221 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n", 6222 arp_ip_target[i]); 6223 arp_interval = 0; 6224 } else { 6225 if (bond_get_targets_ip(arp_target, ip) == -1) 6226 arp_target[arp_ip_count++] = ip; 6227 else 6228 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n", 6229 &ip); 6230 } 6231 } 6232 6233 if (arp_interval && !arp_ip_count) { 6234 /* don't allow arping if no arp_ip_target given... */ 6235 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n", 6236 arp_interval); 6237 arp_interval = 0; 6238 } 6239 6240 if (arp_validate) { 6241 if (!arp_interval) { 6242 pr_err("arp_validate requires arp_interval\n"); 6243 return -EINVAL; 6244 } 6245 6246 bond_opt_initstr(&newval, arp_validate); 6247 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE), 6248 &newval); 6249 if (!valptr) { 6250 pr_err("Error: invalid arp_validate \"%s\"\n", 6251 arp_validate); 6252 return -EINVAL; 6253 } 6254 arp_validate_value = valptr->value; 6255 } else { 6256 arp_validate_value = 0; 6257 } 6258 6259 if (arp_all_targets) { 6260 bond_opt_initstr(&newval, arp_all_targets); 6261 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS), 6262 &newval); 6263 if (!valptr) { 6264 pr_err("Error: invalid arp_all_targets_value \"%s\"\n", 6265 arp_all_targets); 6266 arp_all_targets_value = 0; 6267 } else { 6268 arp_all_targets_value = valptr->value; 6269 } 6270 } 6271 6272 if (miimon) { 6273 pr_info("MII link monitoring set to %d ms\n", miimon); 6274 } else if (arp_interval) { 6275 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE, 6276 arp_validate_value); 6277 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):", 6278 arp_interval, valptr->string, arp_ip_count); 6279 6280 for (i = 0; i < arp_ip_count; i++) 6281 pr_cont(" %s", arp_ip_target[i]); 6282 6283 pr_cont("\n"); 6284 6285 } else if (max_bonds) { 6286 /* miimon and arp_interval not set, we need one so things 6287 * work as expected, see bonding.txt for details 6288 */ 6289 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"); 6290 } 6291 6292 if (primary && !bond_mode_uses_primary(bond_mode)) { 6293 /* currently, using a primary only makes sense 6294 * in active backup, TLB or ALB modes 6295 */ 6296 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n", 6297 primary, bond_mode_name(bond_mode)); 6298 primary = NULL; 6299 } 6300 6301 if (primary && primary_reselect) { 6302 bond_opt_initstr(&newval, primary_reselect); 6303 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT), 6304 &newval); 6305 if (!valptr) { 6306 pr_err("Error: Invalid primary_reselect \"%s\"\n", 6307 primary_reselect); 6308 return -EINVAL; 6309 } 6310 primary_reselect_value = valptr->value; 6311 } else { 6312 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS; 6313 } 6314 6315 if (fail_over_mac) { 6316 bond_opt_initstr(&newval, fail_over_mac); 6317 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC), 6318 &newval); 6319 if (!valptr) { 6320 pr_err("Error: invalid fail_over_mac \"%s\"\n", 6321 fail_over_mac); 6322 return -EINVAL; 6323 } 6324 fail_over_mac_value = valptr->value; 6325 if (bond_mode != BOND_MODE_ACTIVEBACKUP) 6326 pr_warn("Warning: fail_over_mac only affects active-backup mode\n"); 6327 } else { 6328 fail_over_mac_value = BOND_FOM_NONE; 6329 } 6330 6331 bond_opt_initstr(&newval, "default"); 6332 valptr = bond_opt_parse( 6333 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO), 6334 &newval); 6335 if (!valptr) { 6336 pr_err("Error: No ad_actor_sys_prio default value"); 6337 return -EINVAL; 6338 } 6339 ad_actor_sys_prio = valptr->value; 6340 6341 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY), 6342 &newval); 6343 if (!valptr) { 6344 pr_err("Error: No ad_user_port_key default value"); 6345 return -EINVAL; 6346 } 6347 ad_user_port_key = valptr->value; 6348 6349 bond_opt_initstr(&newval, "default"); 6350 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval); 6351 if (!valptr) { 6352 pr_err("Error: No tlb_dynamic_lb default value"); 6353 return -EINVAL; 6354 } 6355 tlb_dynamic_lb = valptr->value; 6356 6357 if (lp_interval == 0) { 6358 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n", 6359 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL); 6360 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL; 6361 } 6362 6363 /* fill params struct with the proper values */ 6364 params->mode = bond_mode; 6365 params->xmit_policy = xmit_hashtype; 6366 params->miimon = miimon; 6367 params->num_peer_notif = num_peer_notif; 6368 params->arp_interval = arp_interval; 6369 params->arp_validate = arp_validate_value; 6370 params->arp_all_targets = arp_all_targets_value; 6371 params->missed_max = 2; 6372 params->updelay = updelay; 6373 params->downdelay = downdelay; 6374 params->peer_notif_delay = 0; 6375 params->use_carrier = use_carrier; 6376 params->lacp_active = 1; 6377 params->lacp_fast = lacp_fast; 6378 params->primary[0] = 0; 6379 params->primary_reselect = primary_reselect_value; 6380 params->fail_over_mac = fail_over_mac_value; 6381 params->tx_queues = tx_queues; 6382 params->all_slaves_active = all_slaves_active; 6383 params->resend_igmp = resend_igmp; 6384 params->min_links = min_links; 6385 params->lp_interval = lp_interval; 6386 params->packets_per_slave = packets_per_slave; 6387 params->tlb_dynamic_lb = tlb_dynamic_lb; 6388 params->ad_actor_sys_prio = ad_actor_sys_prio; 6389 eth_zero_addr(params->ad_actor_system); 6390 params->ad_user_port_key = ad_user_port_key; 6391 if (packets_per_slave > 0) { 6392 params->reciprocal_packets_per_slave = 6393 reciprocal_value(packets_per_slave); 6394 } else { 6395 /* reciprocal_packets_per_slave is unused if 6396 * packets_per_slave is 0 or 1, just initialize it 6397 */ 6398 params->reciprocal_packets_per_slave = 6399 (struct reciprocal_value) { 0 }; 6400 } 6401 6402 if (primary) 6403 strscpy_pad(params->primary, primary, sizeof(params->primary)); 6404 6405 memcpy(params->arp_targets, arp_target, sizeof(arp_target)); 6406 #if IS_ENABLED(CONFIG_IPV6) 6407 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS); 6408 #endif 6409 6410 return 0; 6411 } 6412 6413 /* Called from registration process */ 6414 static int bond_init(struct net_device *bond_dev) 6415 { 6416 struct bonding *bond = netdev_priv(bond_dev); 6417 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id); 6418 6419 netdev_dbg(bond_dev, "Begin bond_init\n"); 6420 6421 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM); 6422 if (!bond->wq) 6423 return -ENOMEM; 6424 6425 bond->notifier_ctx = false; 6426 6427 spin_lock_init(&bond->stats_lock); 6428 netdev_lockdep_set_classes(bond_dev); 6429 6430 list_add_tail(&bond->bond_list, &bn->dev_list); 6431 6432 bond_prepare_sysfs_group(bond); 6433 6434 bond_debug_register(bond); 6435 6436 /* Ensure valid dev_addr */ 6437 if (is_zero_ether_addr(bond_dev->dev_addr) && 6438 bond_dev->addr_assign_type == NET_ADDR_PERM) 6439 eth_hw_addr_random(bond_dev); 6440 6441 return 0; 6442 } 6443 6444 unsigned int bond_get_num_tx_queues(void) 6445 { 6446 return tx_queues; 6447 } 6448 6449 /* Create a new bond based on the specified name and bonding parameters. 6450 * If name is NULL, obtain a suitable "bond%d" name for us. 6451 * Caller must NOT hold rtnl_lock; we need to release it here before we 6452 * set up our sysfs entries. 6453 */ 6454 int bond_create(struct net *net, const char *name) 6455 { 6456 struct net_device *bond_dev; 6457 struct bonding *bond; 6458 int res = -ENOMEM; 6459 6460 rtnl_lock(); 6461 6462 bond_dev = alloc_netdev_mq(sizeof(struct bonding), 6463 name ? name : "bond%d", NET_NAME_UNKNOWN, 6464 bond_setup, tx_queues); 6465 if (!bond_dev) 6466 goto out; 6467 6468 bond = netdev_priv(bond_dev); 6469 dev_net_set(bond_dev, net); 6470 bond_dev->rtnl_link_ops = &bond_link_ops; 6471 6472 res = register_netdevice(bond_dev); 6473 if (res < 0) { 6474 free_netdev(bond_dev); 6475 goto out; 6476 } 6477 6478 netif_carrier_off(bond_dev); 6479 6480 bond_work_init_all(bond); 6481 6482 out: 6483 rtnl_unlock(); 6484 return res; 6485 } 6486 6487 static int __net_init bond_net_init(struct net *net) 6488 { 6489 struct bond_net *bn = net_generic(net, bond_net_id); 6490 6491 bn->net = net; 6492 INIT_LIST_HEAD(&bn->dev_list); 6493 6494 bond_create_proc_dir(bn); 6495 bond_create_sysfs(bn); 6496 6497 return 0; 6498 } 6499 6500 static void __net_exit bond_net_exit_batch(struct list_head *net_list) 6501 { 6502 struct bond_net *bn; 6503 struct net *net; 6504 LIST_HEAD(list); 6505 6506 list_for_each_entry(net, net_list, exit_list) { 6507 bn = net_generic(net, bond_net_id); 6508 bond_destroy_sysfs(bn); 6509 } 6510 6511 /* Kill off any bonds created after unregistering bond rtnl ops */ 6512 rtnl_lock(); 6513 list_for_each_entry(net, net_list, exit_list) { 6514 struct bonding *bond, *tmp_bond; 6515 6516 bn = net_generic(net, bond_net_id); 6517 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list) 6518 unregister_netdevice_queue(bond->dev, &list); 6519 } 6520 unregister_netdevice_many(&list); 6521 rtnl_unlock(); 6522 6523 list_for_each_entry(net, net_list, exit_list) { 6524 bn = net_generic(net, bond_net_id); 6525 bond_destroy_proc_dir(bn); 6526 } 6527 } 6528 6529 static struct pernet_operations bond_net_ops = { 6530 .init = bond_net_init, 6531 .exit_batch = bond_net_exit_batch, 6532 .id = &bond_net_id, 6533 .size = sizeof(struct bond_net), 6534 }; 6535 6536 static int __init bonding_init(void) 6537 { 6538 int i; 6539 int res; 6540 6541 res = bond_check_params(&bonding_defaults); 6542 if (res) 6543 goto out; 6544 6545 bond_create_debugfs(); 6546 6547 res = register_pernet_subsys(&bond_net_ops); 6548 if (res) 6549 goto err_net_ops; 6550 6551 res = bond_netlink_init(); 6552 if (res) 6553 goto err_link; 6554 6555 for (i = 0; i < max_bonds; i++) { 6556 res = bond_create(&init_net, NULL); 6557 if (res) 6558 goto err; 6559 } 6560 6561 skb_flow_dissector_init(&flow_keys_bonding, 6562 flow_keys_bonding_keys, 6563 ARRAY_SIZE(flow_keys_bonding_keys)); 6564 6565 register_netdevice_notifier(&bond_netdev_notifier); 6566 out: 6567 return res; 6568 err: 6569 bond_netlink_fini(); 6570 err_link: 6571 unregister_pernet_subsys(&bond_net_ops); 6572 err_net_ops: 6573 bond_destroy_debugfs(); 6574 goto out; 6575 6576 } 6577 6578 static void __exit bonding_exit(void) 6579 { 6580 unregister_netdevice_notifier(&bond_netdev_notifier); 6581 6582 bond_netlink_fini(); 6583 unregister_pernet_subsys(&bond_net_ops); 6584 6585 bond_destroy_debugfs(); 6586 6587 #ifdef CONFIG_NET_POLL_CONTROLLER 6588 /* Make sure we don't have an imbalance on our netpoll blocking */ 6589 WARN_ON(atomic_read(&netpoll_block_tx)); 6590 #endif 6591 } 6592 6593 module_init(bonding_init); 6594 module_exit(bonding_exit); 6595 MODULE_LICENSE("GPL"); 6596 MODULE_DESCRIPTION(DRV_DESCRIPTION); 6597 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others"); 6598