1 /* 2 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'. 3 * 4 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes 5 * NCM: Network and Communications Management, Inc. 6 * 7 * BUT, I'm the one who modified it for ethernet, so: 8 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov 9 * 10 * This software may be used and distributed according to the terms 11 * of the GNU Public License, incorporated herein by reference. 12 * 13 */ 14 15 #ifndef _NET_BONDING_H 16 #define _NET_BONDING_H 17 18 #include <linux/timer.h> 19 #include <linux/proc_fs.h> 20 #include <linux/if_bonding.h> 21 #include <linux/cpumask.h> 22 #include <linux/in6.h> 23 #include <linux/netpoll.h> 24 #include <linux/inetdevice.h> 25 #include <linux/etherdevice.h> 26 #include <linux/reciprocal_div.h> 27 #include <linux/if_link.h> 28 29 #include <net/bond_3ad.h> 30 #include <net/bond_alb.h> 31 #include <net/bond_options.h> 32 33 #define BOND_MAX_ARP_TARGETS 16 34 35 #define BOND_DEFAULT_MIIMON 100 36 37 /* 38 * Less bad way to call ioctl from within the kernel; this needs to be 39 * done some other way to get the call out of interrupt context. 40 * Needs "ioctl" variable to be supplied by calling context. 41 */ 42 #define IOCTL(dev, arg, cmd) ({ \ 43 int res = 0; \ 44 mm_segment_t fs = get_fs(); \ 45 set_fs(get_ds()); \ 46 res = ioctl(dev, arg, cmd); \ 47 set_fs(fs); \ 48 res; }) 49 50 #define BOND_MODE(bond) ((bond)->params.mode) 51 52 /* slave list primitives */ 53 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower) 54 55 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond)) 56 57 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */ 58 #define bond_first_slave(bond) \ 59 (bond_has_slaves(bond) ? \ 60 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \ 61 NULL) 62 #define bond_last_slave(bond) \ 63 (bond_has_slaves(bond) ? \ 64 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \ 65 NULL) 66 67 /* Caller must have rcu_read_lock */ 68 #define bond_first_slave_rcu(bond) \ 69 netdev_lower_get_first_private_rcu(bond->dev) 70 71 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond)) 72 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond)) 73 74 /** 75 * bond_for_each_slave - iterate over all slaves 76 * @bond: the bond holding this list 77 * @pos: current slave 78 * @iter: list_head * iterator 79 * 80 * Caller must hold RTNL 81 */ 82 #define bond_for_each_slave(bond, pos, iter) \ 83 netdev_for_each_lower_private((bond)->dev, pos, iter) 84 85 /* Caller must have rcu_read_lock */ 86 #define bond_for_each_slave_rcu(bond, pos, iter) \ 87 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter) 88 89 #ifdef CONFIG_NET_POLL_CONTROLLER 90 extern atomic_t netpoll_block_tx; 91 92 static inline void block_netpoll_tx(void) 93 { 94 atomic_inc(&netpoll_block_tx); 95 } 96 97 static inline void unblock_netpoll_tx(void) 98 { 99 atomic_dec(&netpoll_block_tx); 100 } 101 102 static inline int is_netpoll_tx_blocked(struct net_device *dev) 103 { 104 if (unlikely(netpoll_tx_running(dev))) 105 return atomic_read(&netpoll_block_tx); 106 return 0; 107 } 108 #else 109 #define block_netpoll_tx() 110 #define unblock_netpoll_tx() 111 #define is_netpoll_tx_blocked(dev) (0) 112 #endif 113 114 struct bond_params { 115 int mode; 116 int xmit_policy; 117 int miimon; 118 u8 num_peer_notif; 119 int arp_interval; 120 int arp_validate; 121 int arp_all_targets; 122 int use_carrier; 123 int fail_over_mac; 124 int updelay; 125 int downdelay; 126 int lacp_fast; 127 unsigned int min_links; 128 int ad_select; 129 char primary[IFNAMSIZ]; 130 int primary_reselect; 131 __be32 arp_targets[BOND_MAX_ARP_TARGETS]; 132 int tx_queues; 133 int all_slaves_active; 134 int resend_igmp; 135 int lp_interval; 136 int packets_per_slave; 137 int tlb_dynamic_lb; 138 struct reciprocal_value reciprocal_packets_per_slave; 139 u16 ad_actor_sys_prio; 140 u16 ad_user_port_key; 141 u8 ad_actor_system[ETH_ALEN]; 142 }; 143 144 struct bond_parm_tbl { 145 char *modename; 146 int mode; 147 }; 148 149 struct netdev_notify_work { 150 struct delayed_work work; 151 struct net_device *dev; 152 struct netdev_bonding_info bonding_info; 153 }; 154 155 struct slave { 156 struct net_device *dev; /* first - useful for panic debug */ 157 struct bonding *bond; /* our master */ 158 int delay; 159 /* all three in jiffies */ 160 unsigned long last_link_up; 161 unsigned long last_rx; 162 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS]; 163 s8 link; /* one of BOND_LINK_XXXX */ 164 s8 new_link; 165 u8 backup:1, /* indicates backup slave. Value corresponds with 166 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */ 167 inactive:1, /* indicates inactive slave */ 168 should_notify:1; /* indicateds whether the state changed */ 169 u8 duplex; 170 u32 original_mtu; 171 u32 link_failure_count; 172 u32 speed; 173 u16 queue_id; 174 u8 perm_hwaddr[ETH_ALEN]; 175 struct ad_slave_info *ad_info; 176 struct tlb_slave_info tlb_info; 177 #ifdef CONFIG_NET_POLL_CONTROLLER 178 struct netpoll *np; 179 #endif 180 struct kobject kobj; 181 struct rtnl_link_stats64 slave_stats; 182 }; 183 184 struct bond_up_slave { 185 unsigned int count; 186 struct rcu_head rcu; 187 struct slave *arr[0]; 188 }; 189 190 /* 191 * Link pseudo-state only used internally by monitors 192 */ 193 #define BOND_LINK_NOCHANGE -1 194 195 /* 196 * Here are the locking policies for the two bonding locks: 197 * Get rcu_read_lock when reading or RTNL when writing slave list. 198 */ 199 struct bonding { 200 struct net_device *dev; /* first - useful for panic debug */ 201 struct slave __rcu *curr_active_slave; 202 struct slave __rcu *current_arp_slave; 203 struct slave __rcu *primary_slave; 204 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */ 205 bool force_primary; 206 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */ 207 int (*recv_probe)(const struct sk_buff *, struct bonding *, 208 struct slave *); 209 /* mode_lock is used for mode-specific locking needs, currently used by: 210 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and 211 * bond_3ad_state_machine_handler() concurrently and also 212 * the access to the state machine shared variables. 213 * TLB mode (5) - to sync the use and modifications of its hash table 214 * ALB mode (6) - to sync the use and modifications of its hash table 215 */ 216 spinlock_t mode_lock; 217 u8 send_peer_notif; 218 u8 igmp_retrans; 219 #ifdef CONFIG_PROC_FS 220 struct proc_dir_entry *proc_entry; 221 char proc_file_name[IFNAMSIZ]; 222 #endif /* CONFIG_PROC_FS */ 223 struct list_head bond_list; 224 u32 rr_tx_counter; 225 struct ad_bond_info ad_info; 226 struct alb_bond_info alb_info; 227 struct bond_params params; 228 struct workqueue_struct *wq; 229 struct delayed_work mii_work; 230 struct delayed_work arp_work; 231 struct delayed_work alb_work; 232 struct delayed_work ad_work; 233 struct delayed_work mcast_work; 234 struct delayed_work slave_arr_work; 235 #ifdef CONFIG_DEBUG_FS 236 /* debugging support via debugfs */ 237 struct dentry *debug_dir; 238 #endif /* CONFIG_DEBUG_FS */ 239 struct rtnl_link_stats64 bond_stats; 240 }; 241 242 #define bond_slave_get_rcu(dev) \ 243 ((struct slave *) rcu_dereference(dev->rx_handler_data)) 244 245 #define bond_slave_get_rtnl(dev) \ 246 ((struct slave *) rtnl_dereference(dev->rx_handler_data)) 247 248 void bond_queue_slave_event(struct slave *slave); 249 250 struct bond_vlan_tag { 251 __be16 vlan_proto; 252 unsigned short vlan_id; 253 }; 254 255 /** 256 * Returns NULL if the net_device does not belong to any of the bond's slaves 257 * 258 * Caller must hold bond lock for read 259 */ 260 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond, 261 struct net_device *slave_dev) 262 { 263 return netdev_lower_dev_get_private(bond->dev, slave_dev); 264 } 265 266 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave) 267 { 268 return slave->bond; 269 } 270 271 static inline bool bond_should_override_tx_queue(struct bonding *bond) 272 { 273 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 274 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN; 275 } 276 277 static inline bool bond_is_lb(const struct bonding *bond) 278 { 279 return BOND_MODE(bond) == BOND_MODE_TLB || 280 BOND_MODE(bond) == BOND_MODE_ALB; 281 } 282 283 static inline bool bond_is_nondyn_tlb(const struct bonding *bond) 284 { 285 return (BOND_MODE(bond) == BOND_MODE_TLB) && 286 (bond->params.tlb_dynamic_lb == 0); 287 } 288 289 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond) 290 { 291 return (BOND_MODE(bond) == BOND_MODE_8023AD || 292 BOND_MODE(bond) == BOND_MODE_XOR || 293 bond_is_nondyn_tlb(bond)); 294 } 295 296 static inline bool bond_mode_uses_arp(int mode) 297 { 298 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB && 299 mode != BOND_MODE_ALB; 300 } 301 302 static inline bool bond_mode_uses_primary(int mode) 303 { 304 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB || 305 mode == BOND_MODE_ALB; 306 } 307 308 static inline bool bond_uses_primary(struct bonding *bond) 309 { 310 return bond_mode_uses_primary(BOND_MODE(bond)); 311 } 312 313 static inline struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond) 314 { 315 struct slave *slave = rcu_dereference(bond->curr_active_slave); 316 317 return bond_uses_primary(bond) && slave ? slave->dev : NULL; 318 } 319 320 static inline bool bond_slave_is_up(struct slave *slave) 321 { 322 return netif_running(slave->dev) && netif_carrier_ok(slave->dev); 323 } 324 325 static inline void bond_set_active_slave(struct slave *slave) 326 { 327 if (slave->backup) { 328 slave->backup = 0; 329 bond_queue_slave_event(slave); 330 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 331 } 332 } 333 334 static inline void bond_set_backup_slave(struct slave *slave) 335 { 336 if (!slave->backup) { 337 slave->backup = 1; 338 bond_queue_slave_event(slave); 339 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 340 } 341 } 342 343 static inline void bond_set_slave_state(struct slave *slave, 344 int slave_state, bool notify) 345 { 346 if (slave->backup == slave_state) 347 return; 348 349 slave->backup = slave_state; 350 if (notify) { 351 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 352 bond_queue_slave_event(slave); 353 slave->should_notify = 0; 354 } else { 355 if (slave->should_notify) 356 slave->should_notify = 0; 357 else 358 slave->should_notify = 1; 359 } 360 } 361 362 static inline void bond_slave_state_change(struct bonding *bond) 363 { 364 struct list_head *iter; 365 struct slave *tmp; 366 367 bond_for_each_slave(bond, tmp, iter) { 368 if (tmp->link == BOND_LINK_UP) 369 bond_set_active_slave(tmp); 370 else if (tmp->link == BOND_LINK_DOWN) 371 bond_set_backup_slave(tmp); 372 } 373 } 374 375 static inline void bond_slave_state_notify(struct bonding *bond) 376 { 377 struct list_head *iter; 378 struct slave *tmp; 379 380 bond_for_each_slave(bond, tmp, iter) { 381 if (tmp->should_notify) { 382 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC); 383 tmp->should_notify = 0; 384 } 385 } 386 } 387 388 static inline int bond_slave_state(struct slave *slave) 389 { 390 return slave->backup; 391 } 392 393 static inline bool bond_is_active_slave(struct slave *slave) 394 { 395 return !bond_slave_state(slave); 396 } 397 398 static inline bool bond_slave_can_tx(struct slave *slave) 399 { 400 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP && 401 bond_is_active_slave(slave); 402 } 403 404 #define BOND_PRI_RESELECT_ALWAYS 0 405 #define BOND_PRI_RESELECT_BETTER 1 406 #define BOND_PRI_RESELECT_FAILURE 2 407 408 #define BOND_FOM_NONE 0 409 #define BOND_FOM_ACTIVE 1 410 #define BOND_FOM_FOLLOW 2 411 412 #define BOND_ARP_TARGETS_ANY 0 413 #define BOND_ARP_TARGETS_ALL 1 414 415 #define BOND_ARP_VALIDATE_NONE 0 416 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE) 417 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP) 418 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \ 419 BOND_ARP_VALIDATE_BACKUP) 420 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1) 421 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \ 422 BOND_ARP_FILTER) 423 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \ 424 BOND_ARP_FILTER) 425 426 #define BOND_SLAVE_NOTIFY_NOW true 427 #define BOND_SLAVE_NOTIFY_LATER false 428 429 static inline int slave_do_arp_validate(struct bonding *bond, 430 struct slave *slave) 431 { 432 return bond->params.arp_validate & (1 << bond_slave_state(slave)); 433 } 434 435 static inline int slave_do_arp_validate_only(struct bonding *bond) 436 { 437 return bond->params.arp_validate & BOND_ARP_FILTER; 438 } 439 440 static inline int bond_is_ip_target_ok(__be32 addr) 441 { 442 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr); 443 } 444 445 /* Get the oldest arp which we've received on this slave for bond's 446 * arp_targets. 447 */ 448 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond, 449 struct slave *slave) 450 { 451 int i = 1; 452 unsigned long ret = slave->target_last_arp_rx[0]; 453 454 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++) 455 if (time_before(slave->target_last_arp_rx[i], ret)) 456 ret = slave->target_last_arp_rx[i]; 457 458 return ret; 459 } 460 461 static inline unsigned long slave_last_rx(struct bonding *bond, 462 struct slave *slave) 463 { 464 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL) 465 return slave_oldest_target_arp_rx(bond, slave); 466 467 return slave->last_rx; 468 } 469 470 #ifdef CONFIG_NET_POLL_CONTROLLER 471 static inline void bond_netpoll_send_skb(const struct slave *slave, 472 struct sk_buff *skb) 473 { 474 struct netpoll *np = slave->np; 475 476 if (np) 477 netpoll_send_skb(np, skb); 478 } 479 #else 480 static inline void bond_netpoll_send_skb(const struct slave *slave, 481 struct sk_buff *skb) 482 { 483 } 484 #endif 485 486 static inline void bond_set_slave_inactive_flags(struct slave *slave, 487 bool notify) 488 { 489 if (!bond_is_lb(slave->bond)) 490 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify); 491 if (!slave->bond->params.all_slaves_active) 492 slave->inactive = 1; 493 } 494 495 static inline void bond_set_slave_active_flags(struct slave *slave, 496 bool notify) 497 { 498 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify); 499 slave->inactive = 0; 500 } 501 502 static inline bool bond_is_slave_inactive(struct slave *slave) 503 { 504 return slave->inactive; 505 } 506 507 static inline void bond_set_slave_link_state(struct slave *slave, int state) 508 { 509 slave->link = state; 510 bond_queue_slave_event(slave); 511 } 512 513 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local) 514 { 515 struct in_device *in_dev; 516 __be32 addr = 0; 517 518 rcu_read_lock(); 519 in_dev = __in_dev_get_rcu(dev); 520 521 if (in_dev) 522 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local, 523 RT_SCOPE_HOST); 524 rcu_read_unlock(); 525 return addr; 526 } 527 528 struct bond_net { 529 struct net *net; /* Associated network namespace */ 530 struct list_head dev_list; 531 #ifdef CONFIG_PROC_FS 532 struct proc_dir_entry *proc_dir; 533 #endif 534 struct class_attribute class_attr_bonding_masters; 535 }; 536 537 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave); 538 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev); 539 int bond_create(struct net *net, const char *name); 540 int bond_create_sysfs(struct bond_net *net); 541 void bond_destroy_sysfs(struct bond_net *net); 542 void bond_prepare_sysfs_group(struct bonding *bond); 543 int bond_sysfs_slave_add(struct slave *slave); 544 void bond_sysfs_slave_del(struct slave *slave); 545 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev); 546 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev); 547 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb); 548 int bond_set_carrier(struct bonding *bond); 549 void bond_select_active_slave(struct bonding *bond); 550 void bond_change_active_slave(struct bonding *bond, struct slave *new_active); 551 void bond_create_debugfs(void); 552 void bond_destroy_debugfs(void); 553 void bond_debug_register(struct bonding *bond); 554 void bond_debug_unregister(struct bonding *bond); 555 void bond_debug_reregister(struct bonding *bond); 556 const char *bond_mode_name(int mode); 557 void bond_setup(struct net_device *bond_dev); 558 unsigned int bond_get_num_tx_queues(void); 559 int bond_netlink_init(void); 560 void bond_netlink_fini(void); 561 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond); 562 const char *bond_slave_link_status(s8 link); 563 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 564 struct net_device *end_dev, 565 int level); 566 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave); 567 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay); 568 569 #ifdef CONFIG_PROC_FS 570 void bond_create_proc_entry(struct bonding *bond); 571 void bond_remove_proc_entry(struct bonding *bond); 572 void bond_create_proc_dir(struct bond_net *bn); 573 void bond_destroy_proc_dir(struct bond_net *bn); 574 #else 575 static inline void bond_create_proc_entry(struct bonding *bond) 576 { 577 } 578 579 static inline void bond_remove_proc_entry(struct bonding *bond) 580 { 581 } 582 583 static inline void bond_create_proc_dir(struct bond_net *bn) 584 { 585 } 586 587 static inline void bond_destroy_proc_dir(struct bond_net *bn) 588 { 589 } 590 #endif 591 592 static inline struct slave *bond_slave_has_mac(struct bonding *bond, 593 const u8 *mac) 594 { 595 struct list_head *iter; 596 struct slave *tmp; 597 598 bond_for_each_slave(bond, tmp, iter) 599 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 600 return tmp; 601 602 return NULL; 603 } 604 605 /* Caller must hold rcu_read_lock() for read */ 606 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond, 607 const u8 *mac) 608 { 609 struct list_head *iter; 610 struct slave *tmp; 611 612 bond_for_each_slave_rcu(bond, tmp, iter) 613 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 614 return tmp; 615 616 return NULL; 617 } 618 619 /* Caller must hold rcu_read_lock() for read */ 620 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac) 621 { 622 struct list_head *iter; 623 struct slave *tmp; 624 struct netdev_hw_addr *ha; 625 626 bond_for_each_slave_rcu(bond, tmp, iter) 627 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 628 return true; 629 630 if (netdev_uc_empty(bond->dev)) 631 return false; 632 633 netdev_for_each_uc_addr(ha, bond->dev) 634 if (ether_addr_equal_64bits(mac, ha->addr)) 635 return true; 636 637 return false; 638 } 639 640 /* Check if the ip is present in arp ip list, or first free slot if ip == 0 641 * Returns -1 if not found, index if found 642 */ 643 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip) 644 { 645 int i; 646 647 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 648 if (targets[i] == ip) 649 return i; 650 else if (targets[i] == 0) 651 break; 652 653 return -1; 654 } 655 656 /* exported from bond_main.c */ 657 extern int bond_net_id; 658 extern const struct bond_parm_tbl bond_lacp_tbl[]; 659 extern const struct bond_parm_tbl xmit_hashtype_tbl[]; 660 extern const struct bond_parm_tbl arp_validate_tbl[]; 661 extern const struct bond_parm_tbl arp_all_targets_tbl[]; 662 extern const struct bond_parm_tbl fail_over_mac_tbl[]; 663 extern const struct bond_parm_tbl pri_reselect_tbl[]; 664 extern struct bond_parm_tbl ad_select_tbl[]; 665 666 /* exported from bond_netlink.c */ 667 extern struct rtnl_link_ops bond_link_ops; 668 669 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb) 670 { 671 atomic_long_inc(&dev->tx_dropped); 672 dev_kfree_skb_any(skb); 673 } 674 675 #endif /* _NET_BONDING_H */ 676