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 }; 140 141 struct bond_parm_tbl { 142 char *modename; 143 int mode; 144 }; 145 146 struct netdev_notify_work { 147 struct delayed_work work; 148 struct net_device *dev; 149 struct netdev_bonding_info bonding_info; 150 }; 151 152 struct slave { 153 struct net_device *dev; /* first - useful for panic debug */ 154 struct bonding *bond; /* our master */ 155 int delay; 156 /* all three in jiffies */ 157 unsigned long last_link_up; 158 unsigned long last_rx; 159 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS]; 160 s8 link; /* one of BOND_LINK_XXXX */ 161 s8 new_link; 162 u8 backup:1, /* indicates backup slave. Value corresponds with 163 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */ 164 inactive:1, /* indicates inactive slave */ 165 should_notify:1; /* indicateds whether the state changed */ 166 u8 duplex; 167 u32 original_mtu; 168 u32 link_failure_count; 169 u32 speed; 170 u16 queue_id; 171 u8 perm_hwaddr[ETH_ALEN]; 172 struct ad_slave_info *ad_info; 173 struct tlb_slave_info tlb_info; 174 #ifdef CONFIG_NET_POLL_CONTROLLER 175 struct netpoll *np; 176 #endif 177 struct kobject kobj; 178 struct rtnl_link_stats64 slave_stats; 179 }; 180 181 struct bond_up_slave { 182 unsigned int count; 183 struct rcu_head rcu; 184 struct slave *arr[0]; 185 }; 186 187 /* 188 * Link pseudo-state only used internally by monitors 189 */ 190 #define BOND_LINK_NOCHANGE -1 191 192 /* 193 * Here are the locking policies for the two bonding locks: 194 * Get rcu_read_lock when reading or RTNL when writing slave list. 195 */ 196 struct bonding { 197 struct net_device *dev; /* first - useful for panic debug */ 198 struct slave __rcu *curr_active_slave; 199 struct slave __rcu *current_arp_slave; 200 struct slave __rcu *primary_slave; 201 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */ 202 bool force_primary; 203 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */ 204 int (*recv_probe)(const struct sk_buff *, struct bonding *, 205 struct slave *); 206 /* mode_lock is used for mode-specific locking needs, currently used by: 207 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and 208 * bond_3ad_state_machine_handler() concurrently and also 209 * the access to the state machine shared variables. 210 * TLB mode (5) - to sync the use and modifications of its hash table 211 * ALB mode (6) - to sync the use and modifications of its hash table 212 */ 213 spinlock_t mode_lock; 214 u8 send_peer_notif; 215 u8 igmp_retrans; 216 #ifdef CONFIG_PROC_FS 217 struct proc_dir_entry *proc_entry; 218 char proc_file_name[IFNAMSIZ]; 219 #endif /* CONFIG_PROC_FS */ 220 struct list_head bond_list; 221 u32 rr_tx_counter; 222 struct ad_bond_info ad_info; 223 struct alb_bond_info alb_info; 224 struct bond_params params; 225 struct workqueue_struct *wq; 226 struct delayed_work mii_work; 227 struct delayed_work arp_work; 228 struct delayed_work alb_work; 229 struct delayed_work ad_work; 230 struct delayed_work mcast_work; 231 struct delayed_work slave_arr_work; 232 #ifdef CONFIG_DEBUG_FS 233 /* debugging support via debugfs */ 234 struct dentry *debug_dir; 235 #endif /* CONFIG_DEBUG_FS */ 236 struct rtnl_link_stats64 bond_stats; 237 }; 238 239 #define bond_slave_get_rcu(dev) \ 240 ((struct slave *) rcu_dereference(dev->rx_handler_data)) 241 242 #define bond_slave_get_rtnl(dev) \ 243 ((struct slave *) rtnl_dereference(dev->rx_handler_data)) 244 245 void bond_queue_slave_event(struct slave *slave); 246 247 struct bond_vlan_tag { 248 __be16 vlan_proto; 249 unsigned short vlan_id; 250 }; 251 252 /** 253 * Returns NULL if the net_device does not belong to any of the bond's slaves 254 * 255 * Caller must hold bond lock for read 256 */ 257 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond, 258 struct net_device *slave_dev) 259 { 260 return netdev_lower_dev_get_private(bond->dev, slave_dev); 261 } 262 263 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave) 264 { 265 return slave->bond; 266 } 267 268 static inline bool bond_should_override_tx_queue(struct bonding *bond) 269 { 270 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 271 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN; 272 } 273 274 static inline bool bond_is_lb(const struct bonding *bond) 275 { 276 return BOND_MODE(bond) == BOND_MODE_TLB || 277 BOND_MODE(bond) == BOND_MODE_ALB; 278 } 279 280 static inline bool bond_is_nondyn_tlb(const struct bonding *bond) 281 { 282 return (BOND_MODE(bond) == BOND_MODE_TLB) && 283 (bond->params.tlb_dynamic_lb == 0); 284 } 285 286 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond) 287 { 288 return (BOND_MODE(bond) == BOND_MODE_8023AD || 289 BOND_MODE(bond) == BOND_MODE_XOR || 290 bond_is_nondyn_tlb(bond)); 291 } 292 293 static inline bool bond_mode_uses_arp(int mode) 294 { 295 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB && 296 mode != BOND_MODE_ALB; 297 } 298 299 static inline bool bond_mode_uses_primary(int mode) 300 { 301 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB || 302 mode == BOND_MODE_ALB; 303 } 304 305 static inline bool bond_uses_primary(struct bonding *bond) 306 { 307 return bond_mode_uses_primary(BOND_MODE(bond)); 308 } 309 310 static inline bool bond_slave_is_up(struct slave *slave) 311 { 312 return netif_running(slave->dev) && netif_carrier_ok(slave->dev); 313 } 314 315 static inline void bond_set_active_slave(struct slave *slave) 316 { 317 if (slave->backup) { 318 slave->backup = 0; 319 bond_queue_slave_event(slave); 320 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 321 } 322 } 323 324 static inline void bond_set_backup_slave(struct slave *slave) 325 { 326 if (!slave->backup) { 327 slave->backup = 1; 328 bond_queue_slave_event(slave); 329 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 330 } 331 } 332 333 static inline void bond_set_slave_state(struct slave *slave, 334 int slave_state, bool notify) 335 { 336 if (slave->backup == slave_state) 337 return; 338 339 slave->backup = slave_state; 340 if (notify) { 341 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 342 bond_queue_slave_event(slave); 343 slave->should_notify = 0; 344 } else { 345 if (slave->should_notify) 346 slave->should_notify = 0; 347 else 348 slave->should_notify = 1; 349 } 350 } 351 352 static inline void bond_slave_state_change(struct bonding *bond) 353 { 354 struct list_head *iter; 355 struct slave *tmp; 356 357 bond_for_each_slave(bond, tmp, iter) { 358 if (tmp->link == BOND_LINK_UP) 359 bond_set_active_slave(tmp); 360 else if (tmp->link == BOND_LINK_DOWN) 361 bond_set_backup_slave(tmp); 362 } 363 } 364 365 static inline void bond_slave_state_notify(struct bonding *bond) 366 { 367 struct list_head *iter; 368 struct slave *tmp; 369 370 bond_for_each_slave(bond, tmp, iter) { 371 if (tmp->should_notify) { 372 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC); 373 tmp->should_notify = 0; 374 } 375 } 376 } 377 378 static inline int bond_slave_state(struct slave *slave) 379 { 380 return slave->backup; 381 } 382 383 static inline bool bond_is_active_slave(struct slave *slave) 384 { 385 return !bond_slave_state(slave); 386 } 387 388 static inline bool bond_slave_can_tx(struct slave *slave) 389 { 390 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP && 391 bond_is_active_slave(slave); 392 } 393 394 #define BOND_PRI_RESELECT_ALWAYS 0 395 #define BOND_PRI_RESELECT_BETTER 1 396 #define BOND_PRI_RESELECT_FAILURE 2 397 398 #define BOND_FOM_NONE 0 399 #define BOND_FOM_ACTIVE 1 400 #define BOND_FOM_FOLLOW 2 401 402 #define BOND_ARP_TARGETS_ANY 0 403 #define BOND_ARP_TARGETS_ALL 1 404 405 #define BOND_ARP_VALIDATE_NONE 0 406 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE) 407 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP) 408 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \ 409 BOND_ARP_VALIDATE_BACKUP) 410 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1) 411 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \ 412 BOND_ARP_FILTER) 413 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \ 414 BOND_ARP_FILTER) 415 416 #define BOND_SLAVE_NOTIFY_NOW true 417 #define BOND_SLAVE_NOTIFY_LATER false 418 419 static inline int slave_do_arp_validate(struct bonding *bond, 420 struct slave *slave) 421 { 422 return bond->params.arp_validate & (1 << bond_slave_state(slave)); 423 } 424 425 static inline int slave_do_arp_validate_only(struct bonding *bond) 426 { 427 return bond->params.arp_validate & BOND_ARP_FILTER; 428 } 429 430 static inline int bond_is_ip_target_ok(__be32 addr) 431 { 432 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr); 433 } 434 435 /* Get the oldest arp which we've received on this slave for bond's 436 * arp_targets. 437 */ 438 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond, 439 struct slave *slave) 440 { 441 int i = 1; 442 unsigned long ret = slave->target_last_arp_rx[0]; 443 444 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++) 445 if (time_before(slave->target_last_arp_rx[i], ret)) 446 ret = slave->target_last_arp_rx[i]; 447 448 return ret; 449 } 450 451 static inline unsigned long slave_last_rx(struct bonding *bond, 452 struct slave *slave) 453 { 454 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL) 455 return slave_oldest_target_arp_rx(bond, slave); 456 457 return slave->last_rx; 458 } 459 460 #ifdef CONFIG_NET_POLL_CONTROLLER 461 static inline void bond_netpoll_send_skb(const struct slave *slave, 462 struct sk_buff *skb) 463 { 464 struct netpoll *np = slave->np; 465 466 if (np) 467 netpoll_send_skb(np, skb); 468 } 469 #else 470 static inline void bond_netpoll_send_skb(const struct slave *slave, 471 struct sk_buff *skb) 472 { 473 } 474 #endif 475 476 static inline void bond_set_slave_inactive_flags(struct slave *slave, 477 bool notify) 478 { 479 if (!bond_is_lb(slave->bond)) 480 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify); 481 if (!slave->bond->params.all_slaves_active) 482 slave->inactive = 1; 483 } 484 485 static inline void bond_set_slave_active_flags(struct slave *slave, 486 bool notify) 487 { 488 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify); 489 slave->inactive = 0; 490 } 491 492 static inline bool bond_is_slave_inactive(struct slave *slave) 493 { 494 return slave->inactive; 495 } 496 497 static inline void bond_set_slave_link_state(struct slave *slave, int state) 498 { 499 slave->link = state; 500 bond_queue_slave_event(slave); 501 } 502 503 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local) 504 { 505 struct in_device *in_dev; 506 __be32 addr = 0; 507 508 rcu_read_lock(); 509 in_dev = __in_dev_get_rcu(dev); 510 511 if (in_dev) 512 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local, 513 RT_SCOPE_HOST); 514 rcu_read_unlock(); 515 return addr; 516 } 517 518 struct bond_net { 519 struct net *net; /* Associated network namespace */ 520 struct list_head dev_list; 521 #ifdef CONFIG_PROC_FS 522 struct proc_dir_entry *proc_dir; 523 #endif 524 struct class_attribute class_attr_bonding_masters; 525 }; 526 527 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave); 528 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev); 529 int bond_create(struct net *net, const char *name); 530 int bond_create_sysfs(struct bond_net *net); 531 void bond_destroy_sysfs(struct bond_net *net); 532 void bond_prepare_sysfs_group(struct bonding *bond); 533 int bond_sysfs_slave_add(struct slave *slave); 534 void bond_sysfs_slave_del(struct slave *slave); 535 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev); 536 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev); 537 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb); 538 int bond_set_carrier(struct bonding *bond); 539 void bond_select_active_slave(struct bonding *bond); 540 void bond_change_active_slave(struct bonding *bond, struct slave *new_active); 541 void bond_create_debugfs(void); 542 void bond_destroy_debugfs(void); 543 void bond_debug_register(struct bonding *bond); 544 void bond_debug_unregister(struct bonding *bond); 545 void bond_debug_reregister(struct bonding *bond); 546 const char *bond_mode_name(int mode); 547 void bond_setup(struct net_device *bond_dev); 548 unsigned int bond_get_num_tx_queues(void); 549 int bond_netlink_init(void); 550 void bond_netlink_fini(void); 551 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond); 552 const char *bond_slave_link_status(s8 link); 553 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 554 struct net_device *end_dev, 555 int level); 556 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave); 557 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay); 558 559 #ifdef CONFIG_PROC_FS 560 void bond_create_proc_entry(struct bonding *bond); 561 void bond_remove_proc_entry(struct bonding *bond); 562 void bond_create_proc_dir(struct bond_net *bn); 563 void bond_destroy_proc_dir(struct bond_net *bn); 564 #else 565 static inline void bond_create_proc_entry(struct bonding *bond) 566 { 567 } 568 569 static inline void bond_remove_proc_entry(struct bonding *bond) 570 { 571 } 572 573 static inline void bond_create_proc_dir(struct bond_net *bn) 574 { 575 } 576 577 static inline void bond_destroy_proc_dir(struct bond_net *bn) 578 { 579 } 580 #endif 581 582 static inline struct slave *bond_slave_has_mac(struct bonding *bond, 583 const u8 *mac) 584 { 585 struct list_head *iter; 586 struct slave *tmp; 587 588 bond_for_each_slave(bond, tmp, iter) 589 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 590 return tmp; 591 592 return NULL; 593 } 594 595 /* Caller must hold rcu_read_lock() for read */ 596 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond, 597 const u8 *mac) 598 { 599 struct list_head *iter; 600 struct slave *tmp; 601 602 bond_for_each_slave_rcu(bond, tmp, iter) 603 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 604 return tmp; 605 606 return NULL; 607 } 608 609 /* Caller must hold rcu_read_lock() for read */ 610 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac) 611 { 612 struct list_head *iter; 613 struct slave *tmp; 614 struct netdev_hw_addr *ha; 615 616 bond_for_each_slave_rcu(bond, tmp, iter) 617 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 618 return true; 619 620 if (netdev_uc_empty(bond->dev)) 621 return false; 622 623 netdev_for_each_uc_addr(ha, bond->dev) 624 if (ether_addr_equal_64bits(mac, ha->addr)) 625 return true; 626 627 return false; 628 } 629 630 /* Check if the ip is present in arp ip list, or first free slot if ip == 0 631 * Returns -1 if not found, index if found 632 */ 633 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip) 634 { 635 int i; 636 637 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 638 if (targets[i] == ip) 639 return i; 640 else if (targets[i] == 0) 641 break; 642 643 return -1; 644 } 645 646 /* exported from bond_main.c */ 647 extern int bond_net_id; 648 extern const struct bond_parm_tbl bond_lacp_tbl[]; 649 extern const struct bond_parm_tbl xmit_hashtype_tbl[]; 650 extern const struct bond_parm_tbl arp_validate_tbl[]; 651 extern const struct bond_parm_tbl arp_all_targets_tbl[]; 652 extern const struct bond_parm_tbl fail_over_mac_tbl[]; 653 extern const struct bond_parm_tbl pri_reselect_tbl[]; 654 extern struct bond_parm_tbl ad_select_tbl[]; 655 656 /* exported from bond_netlink.c */ 657 extern struct rtnl_link_ops bond_link_ops; 658 659 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb) 660 { 661 atomic_long_inc(&dev->tx_dropped); 662 dev_kfree_skb_any(skb); 663 } 664 665 #endif /* _NET_BONDING_H */ 666