drivers/net/bonding/bonding.h

   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 _LINUX_BONDING_H
  16 #define _LINUX_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
  28 #include "bond_3ad.h"
  29 #include "bond_alb.h"
  30 #include "bond_options.h"
  31
  32 #define DRV_VERSION     "3.7.1"
  33 #define DRV_RELDATE     "April 27, 2011"
  34 #define DRV_NAME        "bonding"
  35 #define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
  36
  37 #define bond_version DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n"
  38
  39 #define BOND_MAX_ARP_TARGETS    16
  40
  41 #define BOND_DEFAULT_MIIMON     100
  42
  43 /*
  44  * Less bad way to call ioctl from within the kernel; this needs to be
  45  * done some other way to get the call out of interrupt context.
  46  * Needs "ioctl" variable to be supplied by calling context.
  47  */
  48 #define IOCTL(dev, arg, cmd) ({         \
  49         int res = 0;                    \
  50         mm_segment_t fs = get_fs();     \
  51         set_fs(get_ds());               \
  52         res = ioctl(dev, arg, cmd);     \
  53         set_fs(fs);                     \
  54         res; })
  55
  56 #define BOND_MODE(bond) ((bond)->params.mode)
  57
  58 /* slave list primitives */
  59 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
  60
  61 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
  62
  63 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
  64 #define bond_first_slave(bond) \
  65         (bond_has_slaves(bond) ? \
  66                 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
  67                 NULL)
  68 #define bond_last_slave(bond) \
  69         (bond_has_slaves(bond) ? \
  70                 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
  71                 NULL)
  72
  73 /* Caller must have rcu_read_lock */
  74 #define bond_first_slave_rcu(bond) \
  75         netdev_lower_get_first_private_rcu(bond->dev)
  76
  77 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
  78 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
  79
  80 /**
  81  * bond_for_each_slave - iterate over all slaves
  82  * @bond:       the bond holding this list
  83  * @pos:        current slave
  84  * @iter:       list_head * iterator
  85  *
  86  * Caller must hold bond->lock
  87  */
  88 #define bond_for_each_slave(bond, pos, iter) \
  89         netdev_for_each_lower_private((bond)->dev, pos, iter)
  90
  91 /* Caller must have rcu_read_lock */
  92 #define bond_for_each_slave_rcu(bond, pos, iter) \
  93         netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
  94
  95 #ifdef CONFIG_NET_POLL_CONTROLLER
  96 extern atomic_t netpoll_block_tx;
  97
  98 static inline void block_netpoll_tx(void)
  99 {
 100         atomic_inc(&netpoll_block_tx);
 101 }
 102
 103 static inline void unblock_netpoll_tx(void)
 104 {
 105         atomic_dec(&netpoll_block_tx);
 106 }
 107
 108 static inline int is_netpoll_tx_blocked(struct net_device *dev)
 109 {
 110         if (unlikely(netpoll_tx_running(dev)))
 111                 return atomic_read(&netpoll_block_tx);
 112         return 0;
 113 }
 114 #else
 115 #define block_netpoll_tx()
 116 #define unblock_netpoll_tx()
 117 #define is_netpoll_tx_blocked(dev) (0)
 118 #endif
 119
 120 struct bond_params {
 121         int mode;
 122         int xmit_policy;
 123         int miimon;
 124         u8 num_peer_notif;
 125         int arp_interval;
 126         int arp_validate;
 127         int arp_all_targets;
 128         int use_carrier;
 129         int fail_over_mac;
 130         int updelay;
 131         int downdelay;
 132         int lacp_fast;
 133         unsigned int min_links;
 134         int ad_select;
 135         char primary[IFNAMSIZ];
 136         int primary_reselect;
 137         __be32 arp_targets[BOND_MAX_ARP_TARGETS];
 138         int tx_queues;
 139         int all_slaves_active;
 140         int resend_igmp;
 141         int lp_interval;
 142         int packets_per_slave;
 143         int tlb_dynamic_lb;
 144         struct reciprocal_value reciprocal_packets_per_slave;
 145 };
 146
 147 struct bond_parm_tbl {
 148         char *modename;
 149         int mode;
 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 };
 179
 180 /*
 181  * Link pseudo-state only used internally by monitors
 182  */
 183 #define BOND_LINK_NOCHANGE -1
 184
 185 /*
 186  * Here are the locking policies for the two bonding locks:
 187  *
 188  * 1) Get bond->lock when reading/writing slave list.
 189  * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
 190  *    (It is unnecessary when the write-lock is put with bond->lock.)
 191  * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
 192  *    beforehand.
 193  */
 194 struct bonding {
 195         struct   net_device *dev; /* first - useful for panic debug */
 196         struct   slave *curr_active_slave;
 197         struct   slave *current_arp_slave;
 198         struct   slave *primary_slave;
 199         bool     force_primary;
 200         s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
 201         int     (*recv_probe)(const struct sk_buff *, struct bonding *,
 202                               struct slave *);
 203         rwlock_t lock;
 204         rwlock_t curr_slave_lock;
 205         u8       send_peer_notif;
 206         u8       igmp_retrans;
 207 #ifdef CONFIG_PROC_FS
 208         struct   proc_dir_entry *proc_entry;
 209         char     proc_file_name[IFNAMSIZ];
 210 #endif /* CONFIG_PROC_FS */
 211         struct   list_head bond_list;
 212         u32      rr_tx_counter;
 213         struct   ad_bond_info ad_info;
 214         struct   alb_bond_info alb_info;
 215         struct   bond_params params;
 216         struct   workqueue_struct *wq;
 217         struct   delayed_work mii_work;
 218         struct   delayed_work arp_work;
 219         struct   delayed_work alb_work;
 220         struct   delayed_work ad_work;
 221         struct   delayed_work mcast_work;
 222 #ifdef CONFIG_DEBUG_FS
 223         /* debugging support via debugfs */
 224         struct   dentry *debug_dir;
 225 #endif /* CONFIG_DEBUG_FS */
 226 };
 227
 228 #define bond_slave_get_rcu(dev) \
 229         ((struct slave *) rcu_dereference(dev->rx_handler_data))
 230
 231 #define bond_slave_get_rtnl(dev) \
 232         ((struct slave *) rtnl_dereference(dev->rx_handler_data))
 233
 234 struct bond_vlan_tag {
 235         __be16          vlan_proto;
 236         unsigned short  vlan_id;
 237 };
 238
 239 /**
 240  * Returns NULL if the net_device does not belong to any of the bond's slaves
 241  *
 242  * Caller must hold bond lock for read
 243  */
 244 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
 245                                                   struct net_device *slave_dev)
 246 {
 247         return netdev_lower_dev_get_private(bond->dev, slave_dev);
 248 }
 249
 250 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
 251 {
 252         return slave->bond;
 253 }
 254
 255 static inline bool bond_should_override_tx_queue(struct bonding *bond)
 256 {
 257         return BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
 258                BOND_MODE(bond) == BOND_MODE_ROUNDROBIN;
 259 }
 260
 261 static inline bool bond_is_lb(const struct bonding *bond)
 262 {
 263         return BOND_MODE(bond) == BOND_MODE_TLB ||
 264                BOND_MODE(bond) == BOND_MODE_ALB;
 265 }
 266
 267 static inline bool bond_mode_uses_arp(int mode)
 268 {
 269         return mode != BOND_MODE_8023AD && mode != BOND_MODE_TLB &&
 270                mode != BOND_MODE_ALB;
 271 }
 272
 273 static inline bool bond_mode_uses_primary(int mode)
 274 {
 275         return mode == BOND_MODE_ACTIVEBACKUP || mode == BOND_MODE_TLB ||
 276                mode == BOND_MODE_ALB;
 277 }
 278
 279 static inline bool bond_uses_primary(struct bonding *bond)
 280 {
 281         return bond_mode_uses_primary(BOND_MODE(bond));
 282 }
 283
 284 static inline bool bond_slave_is_up(struct slave *slave)
 285 {
 286         return netif_running(slave->dev) && netif_carrier_ok(slave->dev);
 287 }
 288
 289 static inline void bond_set_active_slave(struct slave *slave)
 290 {
 291         if (slave->backup) {
 292                 slave->backup = 0;
 293                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 294         }
 295 }
 296
 297 static inline void bond_set_backup_slave(struct slave *slave)
 298 {
 299         if (!slave->backup) {
 300                 slave->backup = 1;
 301                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 302         }
 303 }
 304
 305 static inline void bond_set_slave_state(struct slave *slave,
 306                                         int slave_state, bool notify)
 307 {
 308         if (slave->backup == slave_state)
 309                 return;
 310
 311         slave->backup = slave_state;
 312         if (notify) {
 313                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 314                 slave->should_notify = 0;
 315         } else {
 316                 if (slave->should_notify)
 317                         slave->should_notify = 0;
 318                 else
 319                         slave->should_notify = 1;
 320         }
 321 }
 322
 323 static inline void bond_slave_state_change(struct bonding *bond)
 324 {
 325         struct list_head *iter;
 326         struct slave *tmp;
 327
 328         bond_for_each_slave(bond, tmp, iter) {
 329                 if (tmp->link == BOND_LINK_UP)
 330                         bond_set_active_slave(tmp);
 331                 else if (tmp->link == BOND_LINK_DOWN)
 332                         bond_set_backup_slave(tmp);
 333         }
 334 }
 335
 336 static inline void bond_slave_state_notify(struct bonding *bond)
 337 {
 338         struct list_head *iter;
 339         struct slave *tmp;
 340
 341         bond_for_each_slave(bond, tmp, iter) {
 342                 if (tmp->should_notify) {
 343                         rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
 344                         tmp->should_notify = 0;
 345                 }
 346         }
 347 }
 348
 349 static inline int bond_slave_state(struct slave *slave)
 350 {
 351         return slave->backup;
 352 }
 353
 354 static inline bool bond_is_active_slave(struct slave *slave)
 355 {
 356         return !bond_slave_state(slave);
 357 }
 358
 359 static inline bool bond_slave_can_tx(struct slave *slave)
 360 {
 361         return bond_slave_is_up(slave) && slave->link == BOND_LINK_UP &&
 362                bond_is_active_slave(slave);
 363 }
 364
 365 #define BOND_PRI_RESELECT_ALWAYS        0
 366 #define BOND_PRI_RESELECT_BETTER        1
 367 #define BOND_PRI_RESELECT_FAILURE       2
 368
 369 #define BOND_FOM_NONE                   0
 370 #define BOND_FOM_ACTIVE                 1
 371 #define BOND_FOM_FOLLOW                 2
 372
 373 #define BOND_ARP_TARGETS_ANY            0
 374 #define BOND_ARP_TARGETS_ALL            1
 375
 376 #define BOND_ARP_VALIDATE_NONE          0
 377 #define BOND_ARP_VALIDATE_ACTIVE        (1 << BOND_STATE_ACTIVE)
 378 #define BOND_ARP_VALIDATE_BACKUP        (1 << BOND_STATE_BACKUP)
 379 #define BOND_ARP_VALIDATE_ALL           (BOND_ARP_VALIDATE_ACTIVE | \
 380                                          BOND_ARP_VALIDATE_BACKUP)
 381 #define BOND_ARP_FILTER                 (BOND_ARP_VALIDATE_ALL + 1)
 382 #define BOND_ARP_FILTER_ACTIVE          (BOND_ARP_VALIDATE_ACTIVE | \
 383                                          BOND_ARP_FILTER)
 384 #define BOND_ARP_FILTER_BACKUP          (BOND_ARP_VALIDATE_BACKUP | \
 385                                          BOND_ARP_FILTER)
 386
 387 #define BOND_SLAVE_NOTIFY_NOW           true
 388 #define BOND_SLAVE_NOTIFY_LATER         false
 389
 390 static inline int slave_do_arp_validate(struct bonding *bond,
 391                                         struct slave *slave)
 392 {
 393         return bond->params.arp_validate & (1 << bond_slave_state(slave));
 394 }
 395
 396 static inline int slave_do_arp_validate_only(struct bonding *bond)
 397 {
 398         return bond->params.arp_validate & BOND_ARP_FILTER;
 399 }
 400
 401 static inline int bond_is_ip_target_ok(__be32 addr)
 402 {
 403         return !ipv4_is_lbcast(addr) && !ipv4_is_zeronet(addr);
 404 }
 405
 406 /* Get the oldest arp which we've received on this slave for bond's
 407  * arp_targets.
 408  */
 409 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
 410                                                        struct slave *slave)
 411 {
 412         int i = 1;
 413         unsigned long ret = slave->target_last_arp_rx[0];
 414
 415         for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
 416                 if (time_before(slave->target_last_arp_rx[i], ret))
 417                         ret = slave->target_last_arp_rx[i];
 418
 419         return ret;
 420 }
 421
 422 static inline unsigned long slave_last_rx(struct bonding *bond,
 423                                         struct slave *slave)
 424 {
 425         if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
 426                 return slave_oldest_target_arp_rx(bond, slave);
 427
 428         return slave->last_rx;
 429 }
 430
 431 #ifdef CONFIG_NET_POLL_CONTROLLER
 432 static inline void bond_netpoll_send_skb(const struct slave *slave,
 433                                          struct sk_buff *skb)
 434 {
 435         struct netpoll *np = slave->np;
 436
 437         if (np)
 438                 netpoll_send_skb(np, skb);
 439 }
 440 #else
 441 static inline void bond_netpoll_send_skb(const struct slave *slave,
 442                                          struct sk_buff *skb)
 443 {
 444 }
 445 #endif
 446
 447 static inline void bond_set_slave_inactive_flags(struct slave *slave,
 448                                                  bool notify)
 449 {
 450         if (!bond_is_lb(slave->bond))
 451                 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
 452         if (!slave->bond->params.all_slaves_active)
 453                 slave->inactive = 1;
 454 }
 455
 456 static inline void bond_set_slave_active_flags(struct slave *slave,
 457                                                bool notify)
 458 {
 459         bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
 460         slave->inactive = 0;
 461 }
 462
 463 static inline bool bond_is_slave_inactive(struct slave *slave)
 464 {
 465         return slave->inactive;
 466 }
 467
 468 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
 469 {
 470         struct in_device *in_dev;
 471         __be32 addr = 0;
 472
 473         rcu_read_lock();
 474         in_dev = __in_dev_get_rcu(dev);
 475
 476         if (in_dev)
 477                 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
 478                                          RT_SCOPE_HOST);
 479         rcu_read_unlock();
 480         return addr;
 481 }
 482
 483 struct bond_net {
 484         struct net              *net;   /* Associated network namespace */
 485         struct list_head        dev_list;
 486 #ifdef CONFIG_PROC_FS
 487         struct proc_dir_entry   *proc_dir;
 488 #endif
 489         struct class_attribute  class_attr_bonding_masters;
 490 };
 491
 492 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
 493 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
 494 int bond_create(struct net *net, const char *name);
 495 int bond_create_sysfs(struct bond_net *net);
 496 void bond_destroy_sysfs(struct bond_net *net);
 497 void bond_prepare_sysfs_group(struct bonding *bond);
 498 int bond_sysfs_slave_add(struct slave *slave);
 499 void bond_sysfs_slave_del(struct slave *slave);
 500 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
 501 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
 502 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb);
 503 void bond_select_active_slave(struct bonding *bond);
 504 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
 505 void bond_create_debugfs(void);
 506 void bond_destroy_debugfs(void);
 507 void bond_debug_register(struct bonding *bond);
 508 void bond_debug_unregister(struct bonding *bond);
 509 void bond_debug_reregister(struct bonding *bond);
 510 const char *bond_mode_name(int mode);
 511 void bond_setup(struct net_device *bond_dev);
 512 unsigned int bond_get_num_tx_queues(void);
 513 int bond_netlink_init(void);
 514 void bond_netlink_fini(void);
 515 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
 516 struct net_device *bond_option_active_slave_get(struct bonding *bond);
 517 const char *bond_slave_link_status(s8 link);
 518
 519 #ifdef CONFIG_PROC_FS
 520 void bond_create_proc_entry(struct bonding *bond);
 521 void bond_remove_proc_entry(struct bonding *bond);
 522 void bond_create_proc_dir(struct bond_net *bn);
 523 void bond_destroy_proc_dir(struct bond_net *bn);
 524 #else
 525 static inline void bond_create_proc_entry(struct bonding *bond)
 526 {
 527 }
 528
 529 static inline void bond_remove_proc_entry(struct bonding *bond)
 530 {
 531 }
 532
 533 static inline void bond_create_proc_dir(struct bond_net *bn)
 534 {
 535 }
 536
 537 static inline void bond_destroy_proc_dir(struct bond_net *bn)
 538 {
 539 }
 540 #endif
 541
 542 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
 543                                                const u8 *mac)
 544 {
 545         struct list_head *iter;
 546         struct slave *tmp;
 547
 548         bond_for_each_slave(bond, tmp, iter)
 549                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 550                         return tmp;
 551
 552         return NULL;
 553 }
 554
 555 /* Caller must hold rcu_read_lock() for read */
 556 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
 557                                                const u8 *mac)
 558 {
 559         struct list_head *iter;
 560         struct slave *tmp;
 561
 562         bond_for_each_slave_rcu(bond, tmp, iter)
 563                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 564                         return tmp;
 565
 566         return NULL;
 567 }
 568
 569 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
 570  * Returns -1 if not found, index if found
 571  */
 572 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
 573 {
 574         int i;
 575
 576         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
 577                 if (targets[i] == ip)
 578                         return i;
 579                 else if (targets[i] == 0)
 580                         break;
 581
 582         return -1;
 583 }
 584
 585 /* exported from bond_main.c */
 586 extern int bond_net_id;
 587 extern const struct bond_parm_tbl bond_lacp_tbl[];
 588 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
 589 extern const struct bond_parm_tbl arp_validate_tbl[];
 590 extern const struct bond_parm_tbl arp_all_targets_tbl[];
 591 extern const struct bond_parm_tbl fail_over_mac_tbl[];
 592 extern const struct bond_parm_tbl pri_reselect_tbl[];
 593 extern struct bond_parm_tbl ad_select_tbl[];
 594
 595 /* exported from bond_netlink.c */
 596 extern struct rtnl_link_ops bond_link_ops;
 597
 598 #endif /* _LINUX_BONDING_H */