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 u8 ad_actor_system[ETH_ALEN]; 141 }; 142 143 struct bond_parm_tbl { 144 char *modename; 145 int mode; 146 }; 147 148 struct netdev_notify_work { 149 struct delayed_work work; 150 struct net_device *dev; 151 struct netdev_bonding_info bonding_info; 152 }; 153 154 struct slave { 155 struct net_device *dev; /* first - useful for panic debug */ 156 struct bonding *bond; /* our master */ 157 int delay; 158 /* all three in jiffies */ 159 unsigned long last_link_up; 160 unsigned long last_rx; 161 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS]; 162 s8 link; /* one of BOND_LINK_XXXX */ 163 s8 new_link; 164 u8 backup:1, /* indicates backup slave. Value corresponds with 165 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */ 166 inactive:1, /* indicates inactive slave */ 167 should_notify:1; /* indicateds whether the state changed */ 168 u8 duplex; 169 u32 original_mtu; 170 u32 link_failure_count; 171 u32 speed; 172 u16 queue_id; 173 u8 perm_hwaddr[ETH_ALEN]; 174 struct ad_slave_info *ad_info; 175 struct tlb_slave_info tlb_info; 176 #ifdef CONFIG_NET_POLL_CONTROLLER 177 struct netpoll *np; 178 #endif 179 struct kobject kobj; 180 struct rtnl_link_stats64 slave_stats; 181 }; 182 183 struct bond_up_slave { 184 unsigned int count; 185 struct rcu_head rcu; 186 struct slave *arr[0]; 187 }; 188 189 /* 190 * Link pseudo-state only used internally by monitors 191 */ 192 #define BOND_LINK_NOCHANGE -1 193 194 /* 195 * Here are the locking policies for the two bonding locks: 196 * Get rcu_read_lock when reading or RTNL when writing slave list. 197 */ 198 struct bonding { 199 struct net_device *dev; /* first - useful for panic debug */ 200 struct slave __rcu *curr_active_slave; 201 struct slave __rcu *current_arp_slave; 202 struct slave __rcu *primary_slave; 203 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */ 204 bool force_primary; 205 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */ 206 int (*recv_probe)(const struct sk_buff *, struct bonding *, 207 struct slave *); 208 /* mode_lock is used for mode-specific locking needs, currently used by: 209 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and 210 * bond_3ad_state_machine_handler() concurrently and also 211 * the access to the state machine shared variables. 212 * TLB mode (5) - to sync the use and modifications of its hash table 213 * ALB mode (6) - to sync the use and modifications of its hash table 214 */ 215 spinlock_t mode_lock; 216 u8 send_peer_notif; 217 u8 igmp_retrans; 218 #ifdef CONFIG_PROC_FS 219 struct proc_dir_entry *proc_entry; 220 char proc_file_name[IFNAMSIZ]; 221 #endif /* CONFIG_PROC_FS */ 222 struct list_head bond_list; 223 u32 rr_tx_counter; 224 struct ad_bond_info ad_info; 225 struct alb_bond_info alb_info; 226 struct bond_params params; 227 struct workqueue_struct *wq; 228 struct delayed_work mii_work; 229 struct delayed_work arp_work; 230 struct delayed_work alb_work; 231 struct delayed_work ad_work; 232 struct delayed_work mcast_work; 233 struct delayed_work slave_arr_work; 234 #ifdef CONFIG_DEBUG_FS 235 /* debugging support via debugfs */ 236 struct dentry *debug_dir; 237 #endif /* CONFIG_DEBUG_FS */ 238 struct rtnl_link_stats64 bond_stats; 239 }; 240 241 #define bond_slave_get_rcu(dev) \ 242 ((struct slave *) rcu_dereference(dev->rx_handler_data)) 243 244 #define bond_slave_get_rtnl(dev) \ 245 ((struct slave *) rtnl_dereference(dev->rx_handler_data)) 246 247 void bond_queue_slave_event(struct slave *slave); 248 249 struct bond_vlan_tag { 250 __be16 vlan_proto; 251 unsigned short vlan_id; 252 }; 253 254 /** 255 * Returns NULL if the net_device does not belong to any of the bond's slaves 256 * 257 * Caller must hold bond lock for read 258 */ 259 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond, 260 struct net_device *slave_dev) 261 { 262 return netdev_lower_dev_get_private(bond->dev, slave_dev); 263 } 264 265 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave) 266 { 267 return slave->bond; 268 } 269 270 static inline bool bond_should_override_tx_queue(struct bonding *bond) 271 { 272 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 273 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN; 274 } 275 276 static inline bool bond_is_lb(const struct bonding *bond) 277 { 278 return BOND_MODE(bond) == BOND_MODE_TLB || 279 BOND_MODE(bond) == BOND_MODE_ALB; 280 } 281 282 static inline bool bond_is_nondyn_tlb(const struct bonding *bond) 283 { 284 return (BOND_MODE(bond) == BOND_MODE_TLB) && 285 (bond->params.tlb_dynamic_lb == 0); 286 } 287 288 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond) 289 { 290 return (BOND_MODE(bond) == BOND_MODE_8023AD || 291 BOND_MODE(bond) == BOND_MODE_XOR || 292 bond_is_nondyn_tlb(bond)); 293 } 294 295 static inline bool bond_mode_uses_arp(int mode) 296 { 297 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB && 298 mode != BOND_MODE_ALB; 299 } 300 301 static inline bool bond_mode_uses_primary(int mode) 302 { 303 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB || 304 mode == BOND_MODE_ALB; 305 } 306 307 static inline bool bond_uses_primary(struct bonding *bond) 308 { 309 return bond_mode_uses_primary(BOND_MODE(bond)); 310 } 311 312 static inline bool bond_slave_is_up(struct slave *slave) 313 { 314 return netif_running(slave->dev) && netif_carrier_ok(slave->dev); 315 } 316 317 static inline void bond_set_active_slave(struct slave *slave) 318 { 319 if (slave->backup) { 320 slave->backup = 0; 321 bond_queue_slave_event(slave); 322 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 323 } 324 } 325 326 static inline void bond_set_backup_slave(struct slave *slave) 327 { 328 if (!slave->backup) { 329 slave->backup = 1; 330 bond_queue_slave_event(slave); 331 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 332 } 333 } 334 335 static inline void bond_set_slave_state(struct slave *slave, 336 int slave_state, bool notify) 337 { 338 if (slave->backup == slave_state) 339 return; 340 341 slave->backup = slave_state; 342 if (notify) { 343 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 344 bond_queue_slave_event(slave); 345 slave->should_notify = 0; 346 } else { 347 if (slave->should_notify) 348 slave->should_notify = 0; 349 else 350 slave->should_notify = 1; 351 } 352 } 353 354 static inline void bond_slave_state_change(struct bonding *bond) 355 { 356 struct list_head *iter; 357 struct slave *tmp; 358 359 bond_for_each_slave(bond, tmp, iter) { 360 if (tmp->link == BOND_LINK_UP) 361 bond_set_active_slave(tmp); 362 else if (tmp->link == BOND_LINK_DOWN) 363 bond_set_backup_slave(tmp); 364 } 365 } 366 367 static inline void bond_slave_state_notify(struct bonding *bond) 368 { 369 struct list_head *iter; 370 struct slave *tmp; 371 372 bond_for_each_slave(bond, tmp, iter) { 373 if (tmp->should_notify) { 374 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC); 375 tmp->should_notify = 0; 376 } 377 } 378 } 379 380 static inline int bond_slave_state(struct slave *slave) 381 { 382 return slave->backup; 383 } 384 385 static inline bool bond_is_active_slave(struct slave *slave) 386 { 387 return !bond_slave_state(slave); 388 } 389 390 static inline bool bond_slave_can_tx(struct slave *slave) 391 { 392 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP && 393 bond_is_active_slave(slave); 394 } 395 396 #define BOND_PRI_RESELECT_ALWAYS 0 397 #define BOND_PRI_RESELECT_BETTER 1 398 #define BOND_PRI_RESELECT_FAILURE 2 399 400 #define BOND_FOM_NONE 0 401 #define BOND_FOM_ACTIVE 1 402 #define BOND_FOM_FOLLOW 2 403 404 #define BOND_ARP_TARGETS_ANY 0 405 #define BOND_ARP_TARGETS_ALL 1 406 407 #define BOND_ARP_VALIDATE_NONE 0 408 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE) 409 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP) 410 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \ 411 BOND_ARP_VALIDATE_BACKUP) 412 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1) 413 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \ 414 BOND_ARP_FILTER) 415 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \ 416 BOND_ARP_FILTER) 417 418 #define BOND_SLAVE_NOTIFY_NOW true 419 #define BOND_SLAVE_NOTIFY_LATER false 420 421 static inline int slave_do_arp_validate(struct bonding *bond, 422 struct slave *slave) 423 { 424 return bond->params.arp_validate & (1 << bond_slave_state(slave)); 425 } 426 427 static inline int slave_do_arp_validate_only(struct bonding *bond) 428 { 429 return bond->params.arp_validate & BOND_ARP_FILTER; 430 } 431 432 static inline int bond_is_ip_target_ok(__be32 addr) 433 { 434 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr); 435 } 436 437 /* Get the oldest arp which we've received on this slave for bond's 438 * arp_targets. 439 */ 440 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond, 441 struct slave *slave) 442 { 443 int i = 1; 444 unsigned long ret = slave->target_last_arp_rx[0]; 445 446 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++) 447 if (time_before(slave->target_last_arp_rx[i], ret)) 448 ret = slave->target_last_arp_rx[i]; 449 450 return ret; 451 } 452 453 static inline unsigned long slave_last_rx(struct bonding *bond, 454 struct slave *slave) 455 { 456 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL) 457 return slave_oldest_target_arp_rx(bond, slave); 458 459 return slave->last_rx; 460 } 461 462 #ifdef CONFIG_NET_POLL_CONTROLLER 463 static inline void bond_netpoll_send_skb(const struct slave *slave, 464 struct sk_buff *skb) 465 { 466 struct netpoll *np = slave->np; 467 468 if (np) 469 netpoll_send_skb(np, skb); 470 } 471 #else 472 static inline void bond_netpoll_send_skb(const struct slave *slave, 473 struct sk_buff *skb) 474 { 475 } 476 #endif 477 478 static inline void bond_set_slave_inactive_flags(struct slave *slave, 479 bool notify) 480 { 481 if (!bond_is_lb(slave->bond)) 482 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify); 483 if (!slave->bond->params.all_slaves_active) 484 slave->inactive = 1; 485 } 486 487 static inline void bond_set_slave_active_flags(struct slave *slave, 488 bool notify) 489 { 490 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify); 491 slave->inactive = 0; 492 } 493 494 static inline bool bond_is_slave_inactive(struct slave *slave) 495 { 496 return slave->inactive; 497 } 498 499 static inline void bond_set_slave_link_state(struct slave *slave, int state) 500 { 501 slave->link = state; 502 bond_queue_slave_event(slave); 503 } 504 505 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local) 506 { 507 struct in_device *in_dev; 508 __be32 addr = 0; 509 510 rcu_read_lock(); 511 in_dev = __in_dev_get_rcu(dev); 512 513 if (in_dev) 514 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local, 515 RT_SCOPE_HOST); 516 rcu_read_unlock(); 517 return addr; 518 } 519 520 struct bond_net { 521 struct net *net; /* Associated network namespace */ 522 struct list_head dev_list; 523 #ifdef CONFIG_PROC_FS 524 struct proc_dir_entry *proc_dir; 525 #endif 526 struct class_attribute class_attr_bonding_masters; 527 }; 528 529 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave); 530 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev); 531 int bond_create(struct net *net, const char *name); 532 int bond_create_sysfs(struct bond_net *net); 533 void bond_destroy_sysfs(struct bond_net *net); 534 void bond_prepare_sysfs_group(struct bonding *bond); 535 int bond_sysfs_slave_add(struct slave *slave); 536 void bond_sysfs_slave_del(struct slave *slave); 537 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev); 538 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev); 539 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb); 540 int bond_set_carrier(struct bonding *bond); 541 void bond_select_active_slave(struct bonding *bond); 542 void bond_change_active_slave(struct bonding *bond, struct slave *new_active); 543 void bond_create_debugfs(void); 544 void bond_destroy_debugfs(void); 545 void bond_debug_register(struct bonding *bond); 546 void bond_debug_unregister(struct bonding *bond); 547 void bond_debug_reregister(struct bonding *bond); 548 const char *bond_mode_name(int mode); 549 void bond_setup(struct net_device *bond_dev); 550 unsigned int bond_get_num_tx_queues(void); 551 int bond_netlink_init(void); 552 void bond_netlink_fini(void); 553 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond); 554 const char *bond_slave_link_status(s8 link); 555 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 556 struct net_device *end_dev, 557 int level); 558 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave); 559 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay); 560 561 #ifdef CONFIG_PROC_FS 562 void bond_create_proc_entry(struct bonding *bond); 563 void bond_remove_proc_entry(struct bonding *bond); 564 void bond_create_proc_dir(struct bond_net *bn); 565 void bond_destroy_proc_dir(struct bond_net *bn); 566 #else 567 static inline void bond_create_proc_entry(struct bonding *bond) 568 { 569 } 570 571 static inline void bond_remove_proc_entry(struct bonding *bond) 572 { 573 } 574 575 static inline void bond_create_proc_dir(struct bond_net *bn) 576 { 577 } 578 579 static inline void bond_destroy_proc_dir(struct bond_net *bn) 580 { 581 } 582 #endif 583 584 static inline struct slave *bond_slave_has_mac(struct bonding *bond, 585 const u8 *mac) 586 { 587 struct list_head *iter; 588 struct slave *tmp; 589 590 bond_for_each_slave(bond, tmp, iter) 591 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 592 return tmp; 593 594 return NULL; 595 } 596 597 /* Caller must hold rcu_read_lock() for read */ 598 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond, 599 const u8 *mac) 600 { 601 struct list_head *iter; 602 struct slave *tmp; 603 604 bond_for_each_slave_rcu(bond, tmp, iter) 605 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 606 return tmp; 607 608 return NULL; 609 } 610 611 /* Caller must hold rcu_read_lock() for read */ 612 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac) 613 { 614 struct list_head *iter; 615 struct slave *tmp; 616 struct netdev_hw_addr *ha; 617 618 bond_for_each_slave_rcu(bond, tmp, iter) 619 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 620 return true; 621 622 if (netdev_uc_empty(bond->dev)) 623 return false; 624 625 netdev_for_each_uc_addr(ha, bond->dev) 626 if (ether_addr_equal_64bits(mac, ha->addr)) 627 return true; 628 629 return false; 630 } 631 632 /* Check if the ip is present in arp ip list, or first free slot if ip == 0 633 * Returns -1 if not found, index if found 634 */ 635 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip) 636 { 637 int i; 638 639 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 640 if (targets[i] == ip) 641 return i; 642 else if (targets[i] == 0) 643 break; 644 645 return -1; 646 } 647 648 /* exported from bond_main.c */ 649 extern int bond_net_id; 650 extern const struct bond_parm_tbl bond_lacp_tbl[]; 651 extern const struct bond_parm_tbl xmit_hashtype_tbl[]; 652 extern const struct bond_parm_tbl arp_validate_tbl[]; 653 extern const struct bond_parm_tbl arp_all_targets_tbl[]; 654 extern const struct bond_parm_tbl fail_over_mac_tbl[]; 655 extern const struct bond_parm_tbl pri_reselect_tbl[]; 656 extern struct bond_parm_tbl ad_select_tbl[]; 657 658 /* exported from bond_netlink.c */ 659 extern struct rtnl_link_ops bond_link_ops; 660 661 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb) 662 { 663 atomic_long_inc(&dev->tx_dropped); 664 dev_kfree_skb_any(skb); 665 } 666 667 #endif /* _NET_BONDING_H */ 668