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 DRV_VERSION "3.7.1" 34 #define DRV_RELDATE "April 27, 2011" 35 #define DRV_NAME "bonding" 36 #define DRV_DESCRIPTION "Ethernet Channel Bonding Driver" 37 38 #define bond_version DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n" 39 40 #define BOND_MAX_ARP_TARGETS 16 41 42 #define BOND_DEFAULT_MIIMON 100 43 44 /* 45 * Less bad way to call ioctl from within the kernel; this needs to be 46 * done some other way to get the call out of interrupt context. 47 * Needs "ioctl" variable to be supplied by calling context. 48 */ 49 #define IOCTL(dev, arg, cmd) ({ \ 50 int res = 0; \ 51 mm_segment_t fs = get_fs(); \ 52 set_fs(get_ds()); \ 53 res = ioctl(dev, arg, cmd); \ 54 set_fs(fs); \ 55 res; }) 56 57 #define BOND_MODE(bond) ((bond)->params.mode) 58 59 /* slave list primitives */ 60 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower) 61 62 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond)) 63 64 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */ 65 #define bond_first_slave(bond) \ 66 (bond_has_slaves(bond) ? \ 67 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \ 68 NULL) 69 #define bond_last_slave(bond) \ 70 (bond_has_slaves(bond) ? \ 71 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \ 72 NULL) 73 74 /* Caller must have rcu_read_lock */ 75 #define bond_first_slave_rcu(bond) \ 76 netdev_lower_get_first_private_rcu(bond->dev) 77 78 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond)) 79 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond)) 80 81 /** 82 * bond_for_each_slave - iterate over all slaves 83 * @bond: the bond holding this list 84 * @pos: current slave 85 * @iter: list_head * iterator 86 * 87 * Caller must hold RTNL 88 */ 89 #define bond_for_each_slave(bond, pos, iter) \ 90 netdev_for_each_lower_private((bond)->dev, pos, iter) 91 92 /* Caller must have rcu_read_lock */ 93 #define bond_for_each_slave_rcu(bond, pos, iter) \ 94 netdev_for_each_lower_private_rcu((bond)->dev, pos, iter) 95 96 #ifdef CONFIG_NET_POLL_CONTROLLER 97 extern atomic_t netpoll_block_tx; 98 99 static inline void block_netpoll_tx(void) 100 { 101 atomic_inc(&netpoll_block_tx); 102 } 103 104 static inline void unblock_netpoll_tx(void) 105 { 106 atomic_dec(&netpoll_block_tx); 107 } 108 109 static inline int is_netpoll_tx_blocked(struct net_device *dev) 110 { 111 if (unlikely(netpoll_tx_running(dev))) 112 return atomic_read(&netpoll_block_tx); 113 return 0; 114 } 115 #else 116 #define block_netpoll_tx() 117 #define unblock_netpoll_tx() 118 #define is_netpoll_tx_blocked(dev) (0) 119 #endif 120 121 struct bond_params { 122 int mode; 123 int xmit_policy; 124 int miimon; 125 u8 num_peer_notif; 126 int arp_interval; 127 int arp_validate; 128 int arp_all_targets; 129 int use_carrier; 130 int fail_over_mac; 131 int updelay; 132 int downdelay; 133 int lacp_fast; 134 unsigned int min_links; 135 int ad_select; 136 char primary[IFNAMSIZ]; 137 int primary_reselect; 138 __be32 arp_targets[BOND_MAX_ARP_TARGETS]; 139 int tx_queues; 140 int all_slaves_active; 141 int resend_igmp; 142 int lp_interval; 143 int packets_per_slave; 144 int tlb_dynamic_lb; 145 struct reciprocal_value reciprocal_packets_per_slave; 146 }; 147 148 struct bond_parm_tbl { 149 char *modename; 150 int mode; 151 }; 152 153 struct netdev_notify_work { 154 struct delayed_work work; 155 struct net_device *dev; 156 struct netdev_bonding_info bonding_info; 157 }; 158 159 struct slave { 160 struct net_device *dev; /* first - useful for panic debug */ 161 struct bonding *bond; /* our master */ 162 int delay; 163 /* all three in jiffies */ 164 unsigned long last_link_up; 165 unsigned long last_rx; 166 unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS]; 167 s8 link; /* one of BOND_LINK_XXXX */ 168 s8 new_link; 169 u8 backup:1, /* indicates backup slave. Value corresponds with 170 BOND_STATE_ACTIVE and BOND_STATE_BACKUP */ 171 inactive:1, /* indicates inactive slave */ 172 should_notify:1; /* indicateds whether the state changed */ 173 u8 duplex; 174 u32 original_mtu; 175 u32 link_failure_count; 176 u32 speed; 177 u16 queue_id; 178 u8 perm_hwaddr[ETH_ALEN]; 179 struct ad_slave_info *ad_info; 180 struct tlb_slave_info tlb_info; 181 #ifdef CONFIG_NET_POLL_CONTROLLER 182 struct netpoll *np; 183 #endif 184 struct kobject kobj; 185 struct rtnl_link_stats64 slave_stats; 186 }; 187 188 struct bond_up_slave { 189 unsigned int count; 190 struct rcu_head rcu; 191 struct slave *arr[0]; 192 }; 193 194 /* 195 * Link pseudo-state only used internally by monitors 196 */ 197 #define BOND_LINK_NOCHANGE -1 198 199 /* 200 * Here are the locking policies for the two bonding locks: 201 * Get rcu_read_lock when reading or RTNL when writing slave list. 202 */ 203 struct bonding { 204 struct net_device *dev; /* first - useful for panic debug */ 205 struct slave __rcu *curr_active_slave; 206 struct slave __rcu *current_arp_slave; 207 struct slave __rcu *primary_slave; 208 struct bond_up_slave __rcu *slave_arr; /* Array of usable slaves */ 209 bool force_primary; 210 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */ 211 int (*recv_probe)(const struct sk_buff *, struct bonding *, 212 struct slave *); 213 /* mode_lock is used for mode-specific locking needs, currently used by: 214 * 3ad mode (4) - protect against running bond_3ad_unbind_slave() and 215 * bond_3ad_state_machine_handler() concurrently and also 216 * the access to the state machine shared variables. 217 * TLB mode (5) - to sync the use and modifications of its hash table 218 * ALB mode (6) - to sync the use and modifications of its hash table 219 */ 220 spinlock_t mode_lock; 221 u8 send_peer_notif; 222 u8 igmp_retrans; 223 #ifdef CONFIG_PROC_FS 224 struct proc_dir_entry *proc_entry; 225 char proc_file_name[IFNAMSIZ]; 226 #endif /* CONFIG_PROC_FS */ 227 struct list_head bond_list; 228 u32 rr_tx_counter; 229 struct ad_bond_info ad_info; 230 struct alb_bond_info alb_info; 231 struct bond_params params; 232 struct workqueue_struct *wq; 233 struct delayed_work mii_work; 234 struct delayed_work arp_work; 235 struct delayed_work alb_work; 236 struct delayed_work ad_work; 237 struct delayed_work mcast_work; 238 struct delayed_work slave_arr_work; 239 #ifdef CONFIG_DEBUG_FS 240 /* debugging support via debugfs */ 241 struct dentry *debug_dir; 242 #endif /* CONFIG_DEBUG_FS */ 243 struct rtnl_link_stats64 bond_stats; 244 }; 245 246 #define bond_slave_get_rcu(dev) \ 247 ((struct slave *) rcu_dereference(dev->rx_handler_data)) 248 249 #define bond_slave_get_rtnl(dev) \ 250 ((struct slave *) rtnl_dereference(dev->rx_handler_data)) 251 252 void bond_queue_slave_event(struct slave *slave); 253 254 struct bond_vlan_tag { 255 __be16 vlan_proto; 256 unsigned short vlan_id; 257 }; 258 259 /** 260 * Returns NULL if the net_device does not belong to any of the bond's slaves 261 * 262 * Caller must hold bond lock for read 263 */ 264 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond, 265 struct net_device *slave_dev) 266 { 267 return netdev_lower_dev_get_private(bond->dev, slave_dev); 268 } 269 270 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave) 271 { 272 return slave->bond; 273 } 274 275 static inline bool bond_should_override_tx_queue(struct bonding *bond) 276 { 277 return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP || 278 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN; 279 } 280 281 static inline bool bond_is_lb(const struct bonding *bond) 282 { 283 return BOND_MODE(bond) == BOND_MODE_TLB || 284 BOND_MODE(bond) == BOND_MODE_ALB; 285 } 286 287 static inline bool bond_is_nondyn_tlb(const struct bonding *bond) 288 { 289 return (BOND_MODE(bond) == BOND_MODE_TLB) && 290 (bond->params.tlb_dynamic_lb == 0); 291 } 292 293 static inline bool bond_mode_uses_xmit_hash(const struct bonding *bond) 294 { 295 return (BOND_MODE(bond) == BOND_MODE_8023AD || 296 BOND_MODE(bond) == BOND_MODE_XOR || 297 bond_is_nondyn_tlb(bond)); 298 } 299 300 static inline bool bond_mode_uses_arp(int mode) 301 { 302 return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB && 303 mode != BOND_MODE_ALB; 304 } 305 306 static inline bool bond_mode_uses_primary(int mode) 307 { 308 return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB || 309 mode == BOND_MODE_ALB; 310 } 311 312 static inline bool bond_uses_primary(struct bonding *bond) 313 { 314 return bond_mode_uses_primary(BOND_MODE(bond)); 315 } 316 317 static inline bool bond_slave_is_up(struct slave *slave) 318 { 319 return netif_running(slave->dev) && netif_carrier_ok(slave->dev); 320 } 321 322 static inline void bond_set_active_slave(struct slave *slave) 323 { 324 if (slave->backup) { 325 slave->backup = 0; 326 bond_queue_slave_event(slave); 327 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 328 } 329 } 330 331 static inline void bond_set_backup_slave(struct slave *slave) 332 { 333 if (!slave->backup) { 334 slave->backup = 1; 335 bond_queue_slave_event(slave); 336 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 337 } 338 } 339 340 static inline void bond_set_slave_state(struct slave *slave, 341 int slave_state, bool notify) 342 { 343 if (slave->backup == slave_state) 344 return; 345 346 slave->backup = slave_state; 347 if (notify) { 348 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC); 349 bond_queue_slave_event(slave); 350 slave->should_notify = 0; 351 } else { 352 if (slave->should_notify) 353 slave->should_notify = 0; 354 else 355 slave->should_notify = 1; 356 } 357 } 358 359 static inline void bond_slave_state_change(struct bonding *bond) 360 { 361 struct list_head *iter; 362 struct slave *tmp; 363 364 bond_for_each_slave(bond, tmp, iter) { 365 if (tmp->link == BOND_LINK_UP) 366 bond_set_active_slave(tmp); 367 else if (tmp->link == BOND_LINK_DOWN) 368 bond_set_backup_slave(tmp); 369 } 370 } 371 372 static inline void bond_slave_state_notify(struct bonding *bond) 373 { 374 struct list_head *iter; 375 struct slave *tmp; 376 377 bond_for_each_slave(bond, tmp, iter) { 378 if (tmp->should_notify) { 379 rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC); 380 tmp->should_notify = 0; 381 } 382 } 383 } 384 385 static inline int bond_slave_state(struct slave *slave) 386 { 387 return slave->backup; 388 } 389 390 static inline bool bond_is_active_slave(struct slave *slave) 391 { 392 return !bond_slave_state(slave); 393 } 394 395 static inline bool bond_slave_can_tx(struct slave *slave) 396 { 397 return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP && 398 bond_is_active_slave(slave); 399 } 400 401 #define BOND_PRI_RESELECT_ALWAYS 0 402 #define BOND_PRI_RESELECT_BETTER 1 403 #define BOND_PRI_RESELECT_FAILURE 2 404 405 #define BOND_FOM_NONE 0 406 #define BOND_FOM_ACTIVE 1 407 #define BOND_FOM_FOLLOW 2 408 409 #define BOND_ARP_TARGETS_ANY 0 410 #define BOND_ARP_TARGETS_ALL 1 411 412 #define BOND_ARP_VALIDATE_NONE 0 413 #define BOND_ARP_VALIDATE_ACTIVE (1 << BOND_STATE_ACTIVE) 414 #define BOND_ARP_VALIDATE_BACKUP (1 << BOND_STATE_BACKUP) 415 #define BOND_ARP_VALIDATE_ALL (BOND_ARP_VALIDATE_ACTIVE | \ 416 BOND_ARP_VALIDATE_BACKUP) 417 #define BOND_ARP_FILTER (BOND_ARP_VALIDATE_ALL + 1) 418 #define BOND_ARP_FILTER_ACTIVE (BOND_ARP_VALIDATE_ACTIVE | \ 419 BOND_ARP_FILTER) 420 #define BOND_ARP_FILTER_BACKUP (BOND_ARP_VALIDATE_BACKUP | \ 421 BOND_ARP_FILTER) 422 423 #define BOND_SLAVE_NOTIFY_NOW true 424 #define BOND_SLAVE_NOTIFY_LATER false 425 426 static inline int slave_do_arp_validate(struct bonding *bond, 427 struct slave *slave) 428 { 429 return bond->params.arp_validate & (1 << bond_slave_state(slave)); 430 } 431 432 static inline int slave_do_arp_validate_only(struct bonding *bond) 433 { 434 return bond->params.arp_validate & BOND_ARP_FILTER; 435 } 436 437 static inline int bond_is_ip_target_ok(__be32 addr) 438 { 439 return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr); 440 } 441 442 /* Get the oldest arp which we've received on this slave for bond's 443 * arp_targets. 444 */ 445 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond, 446 struct slave *slave) 447 { 448 int i = 1; 449 unsigned long ret = slave->target_last_arp_rx[0]; 450 451 for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++) 452 if (time_before(slave->target_last_arp_rx[i], ret)) 453 ret = slave->target_last_arp_rx[i]; 454 455 return ret; 456 } 457 458 static inline unsigned long slave_last_rx(struct bonding *bond, 459 struct slave *slave) 460 { 461 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL) 462 return slave_oldest_target_arp_rx(bond, slave); 463 464 return slave->last_rx; 465 } 466 467 #ifdef CONFIG_NET_POLL_CONTROLLER 468 static inline void bond_netpoll_send_skb(const struct slave *slave, 469 struct sk_buff *skb) 470 { 471 struct netpoll *np = slave->np; 472 473 if (np) 474 netpoll_send_skb(np, skb); 475 } 476 #else 477 static inline void bond_netpoll_send_skb(const struct slave *slave, 478 struct sk_buff *skb) 479 { 480 } 481 #endif 482 483 static inline void bond_set_slave_inactive_flags(struct slave *slave, 484 bool notify) 485 { 486 if (!bond_is_lb(slave->bond)) 487 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify); 488 if (!slave->bond->params.all_slaves_active) 489 slave->inactive = 1; 490 } 491 492 static inline void bond_set_slave_active_flags(struct slave *slave, 493 bool notify) 494 { 495 bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify); 496 slave->inactive = 0; 497 } 498 499 static inline bool bond_is_slave_inactive(struct slave *slave) 500 { 501 return slave->inactive; 502 } 503 504 static inline void bond_set_slave_link_state(struct slave *slave, int state) 505 { 506 slave->link = state; 507 bond_queue_slave_event(slave); 508 } 509 510 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local) 511 { 512 struct in_device *in_dev; 513 __be32 addr = 0; 514 515 rcu_read_lock(); 516 in_dev = __in_dev_get_rcu(dev); 517 518 if (in_dev) 519 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local, 520 RT_SCOPE_HOST); 521 rcu_read_unlock(); 522 return addr; 523 } 524 525 struct bond_net { 526 struct net *net; /* Associated network namespace */ 527 struct list_head dev_list; 528 #ifdef CONFIG_PROC_FS 529 struct proc_dir_entry *proc_dir; 530 #endif 531 struct class_attribute class_attr_bonding_masters; 532 }; 533 534 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave); 535 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev); 536 int bond_create(struct net *net, const char *name); 537 int bond_create_sysfs(struct bond_net *net); 538 void bond_destroy_sysfs(struct bond_net *net); 539 void bond_prepare_sysfs_group(struct bonding *bond); 540 int bond_sysfs_slave_add(struct slave *slave); 541 void bond_sysfs_slave_del(struct slave *slave); 542 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev); 543 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev); 544 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb); 545 int bond_set_carrier(struct bonding *bond); 546 void bond_select_active_slave(struct bonding *bond); 547 void bond_change_active_slave(struct bonding *bond, struct slave *new_active); 548 void bond_create_debugfs(void); 549 void bond_destroy_debugfs(void); 550 void bond_debug_register(struct bonding *bond); 551 void bond_debug_unregister(struct bonding *bond); 552 void bond_debug_reregister(struct bonding *bond); 553 const char *bond_mode_name(int mode); 554 void bond_setup(struct net_device *bond_dev); 555 unsigned int bond_get_num_tx_queues(void); 556 int bond_netlink_init(void); 557 void bond_netlink_fini(void); 558 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond); 559 const char *bond_slave_link_status(s8 link); 560 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev, 561 struct net_device *end_dev, 562 int level); 563 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave); 564 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay); 565 566 #ifdef CONFIG_PROC_FS 567 void bond_create_proc_entry(struct bonding *bond); 568 void bond_remove_proc_entry(struct bonding *bond); 569 void bond_create_proc_dir(struct bond_net *bn); 570 void bond_destroy_proc_dir(struct bond_net *bn); 571 #else 572 static inline void bond_create_proc_entry(struct bonding *bond) 573 { 574 } 575 576 static inline void bond_remove_proc_entry(struct bonding *bond) 577 { 578 } 579 580 static inline void bond_create_proc_dir(struct bond_net *bn) 581 { 582 } 583 584 static inline void bond_destroy_proc_dir(struct bond_net *bn) 585 { 586 } 587 #endif 588 589 static inline struct slave *bond_slave_has_mac(struct bonding *bond, 590 const u8 *mac) 591 { 592 struct list_head *iter; 593 struct slave *tmp; 594 595 bond_for_each_slave(bond, tmp, iter) 596 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 597 return tmp; 598 599 return NULL; 600 } 601 602 /* Caller must hold rcu_read_lock() for read */ 603 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond, 604 const u8 *mac) 605 { 606 struct list_head *iter; 607 struct slave *tmp; 608 609 bond_for_each_slave_rcu(bond, tmp, iter) 610 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 611 return tmp; 612 613 return NULL; 614 } 615 616 /* Caller must hold rcu_read_lock() for read */ 617 static inline bool bond_slave_has_mac_rx(struct bonding *bond, const u8 *mac) 618 { 619 struct list_head *iter; 620 struct slave *tmp; 621 struct netdev_hw_addr *ha; 622 623 bond_for_each_slave_rcu(bond, tmp, iter) 624 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr)) 625 return true; 626 627 if (netdev_uc_empty(bond->dev)) 628 return false; 629 630 netdev_for_each_uc_addr(ha, bond->dev) 631 if (ether_addr_equal_64bits(mac, ha->addr)) 632 return true; 633 634 return false; 635 } 636 637 /* Check if the ip is present in arp ip list, or first free slot if ip == 0 638 * Returns -1 if not found, index if found 639 */ 640 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip) 641 { 642 int i; 643 644 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++) 645 if (targets[i] == ip) 646 return i; 647 else if (targets[i] == 0) 648 break; 649 650 return -1; 651 } 652 653 /* exported from bond_main.c */ 654 extern int bond_net_id; 655 extern const struct bond_parm_tbl bond_lacp_tbl[]; 656 extern const struct bond_parm_tbl xmit_hashtype_tbl[]; 657 extern const struct bond_parm_tbl arp_validate_tbl[]; 658 extern const struct bond_parm_tbl arp_all_targets_tbl[]; 659 extern const struct bond_parm_tbl fail_over_mac_tbl[]; 660 extern const struct bond_parm_tbl pri_reselect_tbl[]; 661 extern struct bond_parm_tbl ad_select_tbl[]; 662 663 /* exported from bond_netlink.c */ 664 extern struct rtnl_link_ops bond_link_ops; 665 666 static inline void bond_tx_drop(struct net_device *dev, struct sk_buff *skb) 667 { 668 atomic_long_inc(&dev->tx_dropped); 669 dev_kfree_skb_any(skb); 670 } 671 672 #endif /* _NET_BONDING_H */ 673