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