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 BOND_MODE_IS_LB(mode)                   \
  73                 (((mode) == BOND_MODE_TLB) ||   \
  74                  ((mode) == BOND_MODE_ALB))
  75
  76 #define IS_IP_TARGET_UNUSABLE_ADDRESS(a)        \
  77         ((htonl(INADDR_BROADCAST) == a) ||      \
  78          ipv4_is_zeronet(a))
  79 /*
  80  * Less bad way to call ioctl from within the kernel; this needs to be
  81  * done some other way to get the call out of interrupt context.
  82  * Needs "ioctl" variable to be supplied by calling context.
  83  */
  84 #define IOCTL(dev, arg, cmd) ({         \
  85         int res = 0;                    \
  86         mm_segment_t fs = get_fs();     \
  87         set_fs(get_ds());               \
  88         res = ioctl(dev, arg, cmd);     \
  89         set_fs(fs);                     \
  90         res; })
  91
  92 /* slave list primitives */
  93 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
  94
  95 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
  96
  97 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
  98 #define bond_first_slave(bond) \
  99         (bond_has_slaves(bond) ? \
 100                 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
 101                 NULL)
 102 #define bond_last_slave(bond) \
 103         (bond_has_slaves(bond) ? \
 104                 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
 105                 NULL)
 106
 107 /* Caller must have rcu_read_lock */
 108 #define bond_first_slave_rcu(bond) \
 109         netdev_lower_get_first_private_rcu(bond->dev)
 110
 111 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
 112 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
 113
 114 /**
 115  * bond_for_each_slave - iterate over all slaves
 116  * @bond:       the bond holding this list
 117  * @pos:        current slave
 118  * @iter:       list_head * iterator
 119  *
 120  * Caller must hold bond->lock
 121  */
 122 #define bond_for_each_slave(bond, pos, iter) \
 123         netdev_for_each_lower_private((bond)->dev, pos, iter)
 124
 125 /* Caller must have rcu_read_lock */
 126 #define bond_for_each_slave_rcu(bond, pos, iter) \
 127         netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
 128
 129 #ifdef CONFIG_NET_POLL_CONTROLLER
 130 extern atomic_t netpoll_block_tx;
 131
 132 static inline void block_netpoll_tx(void)
 133 {
 134         atomic_inc(&netpoll_block_tx);
 135 }
 136
 137 static inline void unblock_netpoll_tx(void)
 138 {
 139         atomic_dec(&netpoll_block_tx);
 140 }
 141
 142 static inline int is_netpoll_tx_blocked(struct net_device *dev)
 143 {
 144         if (unlikely(netpoll_tx_running(dev)))
 145                 return atomic_read(&netpoll_block_tx);
 146         return 0;
 147 }
 148 #else
 149 #define block_netpoll_tx()
 150 #define unblock_netpoll_tx()
 151 #define is_netpoll_tx_blocked(dev) (0)
 152 #endif
 153
 154 struct bond_params {
 155         int mode;
 156         int xmit_policy;
 157         int miimon;
 158         u8 num_peer_notif;
 159         int arp_interval;
 160         int arp_validate;
 161         int arp_all_targets;
 162         int use_carrier;
 163         int fail_over_mac;
 164         int updelay;
 165         int downdelay;
 166         int lacp_fast;
 167         unsigned int min_links;
 168         int ad_select;
 169         char primary[IFNAMSIZ];
 170         int primary_reselect;
 171         __be32 arp_targets[BOND_MAX_ARP_TARGETS];
 172         int tx_queues;
 173         int all_slaves_active;
 174         int resend_igmp;
 175         int lp_interval;
 176         int packets_per_slave;
 177         struct reciprocal_value reciprocal_packets_per_slave;
 178 };
 179
 180 struct bond_parm_tbl {
 181         char *modename;
 182         int mode;
 183 };
 184
 185 #define BOND_MAX_MODENAME_LEN 20
 186
 187 struct slave {
 188         struct net_device *dev; /* first - useful for panic debug */
 189         struct bonding *bond; /* our master */
 190         int    delay;
 191         /* all three in jiffies */
 192         unsigned long last_link_up;
 193         unsigned long last_rx;
 194         unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
 195         s8     link;    /* one of BOND_LINK_XXXX */
 196         s8     new_link;
 197         u8     backup:1,   /* indicates backup slave. Value corresponds with
 198                               BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
 199                inactive:1, /* indicates inactive slave */
 200                should_notify:1; /* indicateds whether the state changed */
 201         u8     duplex;
 202         u32    original_mtu;
 203         u32    link_failure_count;
 204         u32    speed;
 205         u16    queue_id;
 206         u8     perm_hwaddr[ETH_ALEN];
 207         struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
 208         struct tlb_slave_info tlb_info;
 209 #ifdef CONFIG_NET_POLL_CONTROLLER
 210         struct netpoll *np;
 211 #endif
 212         struct kobject kobj;
 213 };
 214
 215 /*
 216  * Link pseudo-state only used internally by monitors
 217  */
 218 #define BOND_LINK_NOCHANGE -1
 219
 220 /*
 221  * Here are the locking policies for the two bonding locks:
 222  *
 223  * 1) Get bond->lock when reading/writing slave list.
 224  * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
 225  *    (It is unnecessary when the write-lock is put with bond->lock.)
 226  * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
 227  *    beforehand.
 228  */
 229 struct bonding {
 230         struct   net_device *dev; /* first - useful for panic debug */
 231         struct   slave *curr_active_slave;
 232         struct   slave *current_arp_slave;
 233         struct   slave *primary_slave;
 234         bool     force_primary;
 235         s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
 236         int     (*recv_probe)(const struct sk_buff *, struct bonding *,
 237                               struct slave *);
 238         rwlock_t lock;
 239         rwlock_t curr_slave_lock;
 240         u8       send_peer_notif;
 241         u8       igmp_retrans;
 242 #ifdef CONFIG_PROC_FS
 243         struct   proc_dir_entry *proc_entry;
 244         char     proc_file_name[IFNAMSIZ];
 245 #endif /* CONFIG_PROC_FS */
 246         struct   list_head bond_list;
 247         u32      rr_tx_counter;
 248         struct   ad_bond_info ad_info;
 249         struct   alb_bond_info alb_info;
 250         struct   bond_params params;
 251         struct   workqueue_struct *wq;
 252         struct   delayed_work mii_work;
 253         struct   delayed_work arp_work;
 254         struct   delayed_work alb_work;
 255         struct   delayed_work ad_work;
 256         struct   delayed_work mcast_work;
 257 #ifdef CONFIG_DEBUG_FS
 258         /* debugging support via debugfs */
 259         struct   dentry *debug_dir;
 260 #endif /* CONFIG_DEBUG_FS */
 261 };
 262
 263 #define bond_slave_get_rcu(dev) \
 264         ((struct slave *) rcu_dereference(dev->rx_handler_data))
 265
 266 #define bond_slave_get_rtnl(dev) \
 267         ((struct slave *) rtnl_dereference(dev->rx_handler_data))
 268
 269 struct bond_vlan_tag {
 270         __be16          vlan_proto;
 271         unsigned short  vlan_id;
 272 };
 273
 274 /**
 275  * Returns NULL if the net_device does not belong to any of the bond's slaves
 276  *
 277  * Caller must hold bond lock for read
 278  */
 279 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
 280                                                   struct net_device *slave_dev)
 281 {
 282         return netdev_lower_dev_get_private(bond->dev, slave_dev);
 283 }
 284
 285 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
 286 {
 287         if (!slave || !slave->bond)
 288                 return NULL;
 289         return slave->bond;
 290 }
 291
 292 static inline bool bond_is_lb(const struct bonding *bond)
 293 {
 294         return BOND_MODE_IS_LB(bond->params.mode);
 295 }
 296
 297 static inline void bond_set_active_slave(struct slave *slave)
 298 {
 299         if (slave->backup) {
 300                 slave->backup = 0;
 301                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 302         }
 303 }
 304
 305 static inline void bond_set_backup_slave(struct slave *slave)
 306 {
 307         if (!slave->backup) {
 308                 slave->backup = 1;
 309                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 310         }
 311 }
 312
 313 static inline void bond_set_slave_state(struct slave *slave,
 314                                         int slave_state, bool notify)
 315 {
 316         if (slave->backup == slave_state)
 317                 return;
 318
 319         slave->backup = slave_state;
 320         if (notify) {
 321                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 322                 slave->should_notify = 0;
 323         } else {
 324                 if (slave->should_notify)
 325                         slave->should_notify = 0;
 326                 else
 327                         slave->should_notify = 1;
 328         }
 329 }
 330
 331 static inline void bond_slave_state_change(struct bonding *bond)
 332 {
 333         struct list_head *iter;
 334         struct slave *tmp;
 335
 336         bond_for_each_slave(bond, tmp, iter) {
 337                 if (tmp->link == BOND_LINK_UP)
 338                         bond_set_active_slave(tmp);
 339                 else if (tmp->link == BOND_LINK_DOWN)
 340                         bond_set_backup_slave(tmp);
 341         }
 342 }
 343
 344 static inline void bond_slave_state_notify(struct bonding *bond)
 345 {
 346         struct list_head *iter;
 347         struct slave *tmp;
 348
 349         bond_for_each_slave(bond, tmp, iter) {
 350                 if (tmp->should_notify) {
 351                         rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
 352                         tmp->should_notify = 0;
 353                 }
 354         }
 355 }
 356
 357 static inline int bond_slave_state(struct slave *slave)
 358 {
 359         return slave->backup;
 360 }
 361
 362 static inline bool bond_is_active_slave(struct slave *slave)
 363 {
 364         return !bond_slave_state(slave);
 365 }
 366
 367 #define BOND_PRI_RESELECT_ALWAYS        0
 368 #define BOND_PRI_RESELECT_BETTER        1
 369 #define BOND_PRI_RESELECT_FAILURE       2
 370
 371 #define BOND_FOM_NONE                   0
 372 #define BOND_FOM_ACTIVE                 1
 373 #define BOND_FOM_FOLLOW                 2
 374
 375 #define BOND_ARP_TARGETS_ANY            0
 376 #define BOND_ARP_TARGETS_ALL            1
 377
 378 #define BOND_ARP_VALIDATE_NONE          0
 379 #define BOND_ARP_VALIDATE_ACTIVE        (1 << BOND_STATE_ACTIVE)
 380 #define BOND_ARP_VALIDATE_BACKUP        (1 << BOND_STATE_BACKUP)
 381 #define BOND_ARP_VALIDATE_ALL           (BOND_ARP_VALIDATE_ACTIVE | \
 382                                          BOND_ARP_VALIDATE_BACKUP)
 383 #define BOND_ARP_FILTER                 (BOND_ARP_VALIDATE_ALL + 1)
 384 #define BOND_ARP_FILTER_ACTIVE          (BOND_ARP_VALIDATE_ACTIVE | \
 385                                          BOND_ARP_FILTER)
 386 #define BOND_ARP_FILTER_BACKUP          (BOND_ARP_VALIDATE_BACKUP | \
 387                                          BOND_ARP_FILTER)
 388
 389 #define BOND_SLAVE_NOTIFY_NOW           true
 390 #define BOND_SLAVE_NOTIFY_LATER         false
 391
 392 static inline int slave_do_arp_validate(struct bonding *bond,
 393                                         struct slave *slave)
 394 {
 395         return bond->params.arp_validate & (1 << bond_slave_state(slave));
 396 }
 397
 398 static inline int slave_do_arp_validate_only(struct bonding *bond,
 399                                              struct slave *slave)
 400 {
 401         return bond->params.arp_validate & BOND_ARP_FILTER;
 402 }
 403
 404 /* Get the oldest arp which we've received on this slave for bond's
 405  * arp_targets.
 406  */
 407 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
 408                                                        struct slave *slave)
 409 {
 410         int i = 1;
 411         unsigned long ret = slave->target_last_arp_rx[0];
 412
 413         for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
 414                 if (time_before(slave->target_last_arp_rx[i], ret))
 415                         ret = slave->target_last_arp_rx[i];
 416
 417         return ret;
 418 }
 419
 420 static inline unsigned long slave_last_rx(struct bonding *bond,
 421                                         struct slave *slave)
 422 {
 423         if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
 424                 return slave_oldest_target_arp_rx(bond, slave);
 425
 426         return slave->last_rx;
 427 }
 428
 429 #ifdef CONFIG_NET_POLL_CONTROLLER
 430 static inline void bond_netpoll_send_skb(const struct slave *slave,
 431                                          struct sk_buff *skb)
 432 {
 433         struct netpoll *np = slave->np;
 434
 435         if (np)
 436                 netpoll_send_skb(np, skb);
 437 }
 438 #else
 439 static inline void bond_netpoll_send_skb(const struct slave *slave,
 440                                          struct sk_buff *skb)
 441 {
 442 }
 443 #endif
 444
 445 static inline void bond_set_slave_inactive_flags(struct slave *slave,
 446                                                  bool notify)
 447 {
 448         if (!bond_is_lb(slave->bond))
 449                 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
 450         if (!slave->bond->params.all_slaves_active)
 451                 slave->inactive = 1;
 452 }
 453
 454 static inline void bond_set_slave_active_flags(struct slave *slave,
 455                                                bool notify)
 456 {
 457         bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
 458         slave->inactive = 0;
 459 }
 460
 461 static inline bool bond_is_slave_inactive(struct slave *slave)
 462 {
 463         return slave->inactive;
 464 }
 465
 466 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
 467 {
 468         struct in_device *in_dev;
 469         __be32 addr = 0;
 470
 471         rcu_read_lock();
 472         in_dev = __in_dev_get_rcu(dev);
 473
 474         if (in_dev)
 475                 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
 476                                          RT_SCOPE_HOST);
 477         rcu_read_unlock();
 478         return addr;
 479 }
 480
 481 static inline bool slave_can_tx(struct slave *slave)
 482 {
 483         if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP &&
 484             bond_is_active_slave(slave))
 485                 return true;
 486         else
 487                 return false;
 488 }
 489
 490 struct bond_net;
 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 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count);
 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 struct bond_net {
 520         struct net *            net;    /* Associated network namespace */
 521         struct list_head        dev_list;
 522 #ifdef CONFIG_PROC_FS
 523         struct proc_dir_entry * proc_dir;
 524 #endif
 525         struct class_attribute  class_attr_bonding_masters;
 526 };
 527
 528 #ifdef CONFIG_PROC_FS
 529 void bond_create_proc_entry(struct bonding *bond);
 530 void bond_remove_proc_entry(struct bonding *bond);
 531 void bond_create_proc_dir(struct bond_net *bn);
 532 void bond_destroy_proc_dir(struct bond_net *bn);
 533 #else
 534 static inline void bond_create_proc_entry(struct bonding *bond)
 535 {
 536 }
 537
 538 static inline void bond_remove_proc_entry(struct bonding *bond)
 539 {
 540 }
 541
 542 static inline void bond_create_proc_dir(struct bond_net *bn)
 543 {
 544 }
 545
 546 static inline void bond_destroy_proc_dir(struct bond_net *bn)
 547 {
 548 }
 549 #endif
 550
 551 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
 552                                                const u8 *mac)
 553 {
 554         struct list_head *iter;
 555         struct slave *tmp;
 556
 557         bond_for_each_slave(bond, tmp, iter)
 558                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 559                         return tmp;
 560
 561         return NULL;
 562 }
 563
 564 /* Caller must hold rcu_read_lock() for read */
 565 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
 566                                                const u8 *mac)
 567 {
 568         struct list_head *iter;
 569         struct slave *tmp;
 570
 571         bond_for_each_slave_rcu(bond, tmp, iter)
 572                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 573                         return tmp;
 574
 575         return NULL;
 576 }
 577
 578 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
 579  * Returns -1 if not found, index if found
 580  */
 581 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
 582 {
 583         int i;
 584
 585         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
 586                 if (targets[i] == ip)
 587                         return i;
 588                 else if (targets[i] == 0)
 589                         break;
 590
 591         return -1;
 592 }
 593
 594 /* exported from bond_main.c */
 595 extern int bond_net_id;
 596 extern const struct bond_parm_tbl bond_lacp_tbl[];
 597 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
 598 extern const struct bond_parm_tbl arp_validate_tbl[];
 599 extern const struct bond_parm_tbl arp_all_targets_tbl[];
 600 extern const struct bond_parm_tbl fail_over_mac_tbl[];
 601 extern const struct bond_parm_tbl pri_reselect_tbl[];
 602 extern struct bond_parm_tbl ad_select_tbl[];
 603
 604 /* exported from bond_netlink.c */
 605 extern struct rtnl_link_ops bond_link_ops;
 606
 607 #endif /* _LINUX_BONDING_H */