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_VLAN_ENCAP     2
  40 #define BOND_MAX_ARP_TARGETS    16
  41
  42 #define BOND_DEFAULT_MIIMON     100
  43
  44 #define IS_UP(dev)                                         \
  45               ((((dev)->flags & IFF_UP) == IFF_UP)      && \
  46                netif_running(dev)                       && \
  47                netif_carrier_ok(dev))
  48
  49 /*
  50  * Checks whether slave is ready for transmit.
  51  */
  52 #define SLAVE_IS_OK(slave)                              \
  53                     (((slave)->dev->flags & IFF_UP)  && \
  54                      netif_running((slave)->dev)     && \
  55                      ((slave)->link == BOND_LINK_UP) && \
  56                      bond_is_active_slave(slave))
  57
  58
  59 #define USES_PRIMARY(mode)                              \
  60                 (((mode) == BOND_MODE_ACTIVEBACKUP) ||  \
  61                  ((mode) == BOND_MODE_TLB)          ||  \
  62                  ((mode) == BOND_MODE_ALB))
  63
  64 #define BOND_NO_USES_ARP(mode)                          \
  65                 (((mode) == BOND_MODE_8023AD)   ||      \
  66                  ((mode) == BOND_MODE_TLB)      ||      \
  67                  ((mode) == BOND_MODE_ALB))
  68
  69 #define TX_QUEUE_OVERRIDE(mode)                         \
  70                         (((mode) == BOND_MODE_ACTIVEBACKUP) ||  \
  71                          ((mode) == BOND_MODE_ROUNDROBIN))
  72
  73 #define BOND_MODE_IS_LB(mode)                   \
  74                 (((mode) == BOND_MODE_TLB) ||   \
  75                  ((mode) == BOND_MODE_ALB))
  76
  77 #define IS_IP_TARGET_UNUSABLE_ADDRESS(a)        \
  78         ((htonl(INADDR_BROADCAST) == a) ||      \
  79          ipv4_is_zeronet(a))
  80 /*
  81  * Less bad way to call ioctl from within the kernel; this needs to be
  82  * done some other way to get the call out of interrupt context.
  83  * Needs "ioctl" variable to be supplied by calling context.
  84  */
  85 #define IOCTL(dev, arg, cmd) ({         \
  86         int res = 0;                    \
  87         mm_segment_t fs = get_fs();     \
  88         set_fs(get_ds());               \
  89         res = ioctl(dev, arg, cmd);     \
  90         set_fs(fs);                     \
  91         res; })
  92
  93 /* slave list primitives */
  94 #define bond_slave_list(bond) (&(bond)->dev->adj_list.lower)
  95
  96 #define bond_has_slaves(bond) !list_empty(bond_slave_list(bond))
  97
  98 /* IMPORTANT: bond_first/last_slave can return NULL in case of an empty list */
  99 #define bond_first_slave(bond) \
 100         (bond_has_slaves(bond) ? \
 101                 netdev_adjacent_get_private(bond_slave_list(bond)->next) : \
 102                 NULL)
 103 #define bond_last_slave(bond) \
 104         (bond_has_slaves(bond) ? \
 105                 netdev_adjacent_get_private(bond_slave_list(bond)->prev) : \
 106                 NULL)
 107
 108 /* Caller must have rcu_read_lock */
 109 #define bond_first_slave_rcu(bond) \
 110         netdev_lower_get_first_private_rcu(bond->dev)
 111
 112 #define bond_is_first_slave(bond, pos) (pos == bond_first_slave(bond))
 113 #define bond_is_last_slave(bond, pos) (pos == bond_last_slave(bond))
 114
 115 /**
 116  * bond_for_each_slave - iterate over all slaves
 117  * @bond:       the bond holding this list
 118  * @pos:        current slave
 119  * @iter:       list_head * iterator
 120  *
 121  * Caller must hold bond->lock
 122  */
 123 #define bond_for_each_slave(bond, pos, iter) \
 124         netdev_for_each_lower_private((bond)->dev, pos, iter)
 125
 126 /* Caller must have rcu_read_lock */
 127 #define bond_for_each_slave_rcu(bond, pos, iter) \
 128         netdev_for_each_lower_private_rcu((bond)->dev, pos, iter)
 129
 130 #ifdef CONFIG_NET_POLL_CONTROLLER
 131 extern atomic_t netpoll_block_tx;
 132
 133 static inline void block_netpoll_tx(void)
 134 {
 135         atomic_inc(&netpoll_block_tx);
 136 }
 137
 138 static inline void unblock_netpoll_tx(void)
 139 {
 140         atomic_dec(&netpoll_block_tx);
 141 }
 142
 143 static inline int is_netpoll_tx_blocked(struct net_device *dev)
 144 {
 145         if (unlikely(netpoll_tx_running(dev)))
 146                 return atomic_read(&netpoll_block_tx);
 147         return 0;
 148 }
 149 #else
 150 #define block_netpoll_tx()
 151 #define unblock_netpoll_tx()
 152 #define is_netpoll_tx_blocked(dev) (0)
 153 #endif
 154
 155 struct bond_params {
 156         int mode;
 157         int xmit_policy;
 158         int miimon;
 159         u8 num_peer_notif;
 160         int arp_interval;
 161         int arp_validate;
 162         int arp_all_targets;
 163         int use_carrier;
 164         int fail_over_mac;
 165         int updelay;
 166         int downdelay;
 167         int lacp_fast;
 168         unsigned int min_links;
 169         int ad_select;
 170         char primary[IFNAMSIZ];
 171         int primary_reselect;
 172         __be32 arp_targets[BOND_MAX_ARP_TARGETS];
 173         int tx_queues;
 174         int all_slaves_active;
 175         int resend_igmp;
 176         int lp_interval;
 177         int packets_per_slave;
 178         struct reciprocal_value reciprocal_packets_per_slave;
 179 };
 180
 181 struct bond_parm_tbl {
 182         char *modename;
 183         int mode;
 184 };
 185
 186 #define BOND_MAX_MODENAME_LEN 20
 187
 188 struct slave {
 189         struct net_device *dev; /* first - useful for panic debug */
 190         struct bonding *bond; /* our master */
 191         int    delay;
 192         /* all three in jiffies */
 193         unsigned long last_link_up;
 194         unsigned long last_rx;
 195         unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
 196         s8     link;    /* one of BOND_LINK_XXXX */
 197         s8     new_link;
 198         u8     backup:1,   /* indicates backup slave. Value corresponds with
 199                               BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
 200                inactive:1, /* indicates inactive slave */
 201                should_notify:1; /* indicateds whether the state changed */
 202         u8     duplex;
 203         u32    original_mtu;
 204         u32    link_failure_count;
 205         u32    speed;
 206         u16    queue_id;
 207         u8     perm_hwaddr[ETH_ALEN];
 208         struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
 209         struct tlb_slave_info tlb_info;
 210 #ifdef CONFIG_NET_POLL_CONTROLLER
 211         struct netpoll *np;
 212 #endif
 213         struct kobject kobj;
 214 };
 215
 216 /*
 217  * Link pseudo-state only used internally by monitors
 218  */
 219 #define BOND_LINK_NOCHANGE -1
 220
 221 /*
 222  * Here are the locking policies for the two bonding locks:
 223  *
 224  * 1) Get bond->lock when reading/writing slave list.
 225  * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
 226  *    (It is unnecessary when the write-lock is put with bond->lock.)
 227  * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
 228  *    beforehand.
 229  */
 230 struct bonding {
 231         struct   net_device *dev; /* first - useful for panic debug */
 232         struct   slave *curr_active_slave;
 233         struct   slave *current_arp_slave;
 234         struct   slave *primary_slave;
 235         bool     force_primary;
 236         s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
 237         int     (*recv_probe)(const struct sk_buff *, struct bonding *,
 238                               struct slave *);
 239         rwlock_t lock;
 240         rwlock_t curr_slave_lock;
 241         u8       send_peer_notif;
 242         u8       igmp_retrans;
 243 #ifdef CONFIG_PROC_FS
 244         struct   proc_dir_entry *proc_entry;
 245         char     proc_file_name[IFNAMSIZ];
 246 #endif /* CONFIG_PROC_FS */
 247         struct   list_head bond_list;
 248         u32      rr_tx_counter;
 249         struct   ad_bond_info ad_info;
 250         struct   alb_bond_info alb_info;
 251         struct   bond_params params;
 252         struct   workqueue_struct *wq;
 253         struct   delayed_work mii_work;
 254         struct   delayed_work arp_work;
 255         struct   delayed_work alb_work;
 256         struct   delayed_work ad_work;
 257         struct   delayed_work mcast_work;
 258 #ifdef CONFIG_DEBUG_FS
 259         /* debugging support via debugfs */
 260         struct   dentry *debug_dir;
 261 #endif /* CONFIG_DEBUG_FS */
 262 };
 263
 264 #define bond_slave_get_rcu(dev) \
 265         ((struct slave *) rcu_dereference(dev->rx_handler_data))
 266
 267 #define bond_slave_get_rtnl(dev) \
 268         ((struct slave *) rtnl_dereference(dev->rx_handler_data))
 269
 270 struct bond_vlan_tag {
 271         __be16          vlan_proto;
 272         unsigned short  vlan_id;
 273 };
 274
 275 /**
 276  * Returns NULL if the net_device does not belong to any of the bond's slaves
 277  *
 278  * Caller must hold bond lock for read
 279  */
 280 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
 281                                                   struct net_device *slave_dev)
 282 {
 283         return netdev_lower_dev_get_private(bond->dev, slave_dev);
 284 }
 285
 286 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
 287 {
 288         if (!slave || !slave->bond)
 289                 return NULL;
 290         return slave->bond;
 291 }
 292
 293 static inline bool bond_is_lb(const struct bonding *bond)
 294 {
 295         return BOND_MODE_IS_LB(bond->params.mode);
 296 }
 297
 298 static inline void bond_set_active_slave(struct slave *slave)
 299 {
 300         if (slave->backup) {
 301                 slave->backup = 0;
 302                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 303         }
 304 }
 305
 306 static inline void bond_set_backup_slave(struct slave *slave)
 307 {
 308         if (!slave->backup) {
 309                 slave->backup = 1;
 310                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 311         }
 312 }
 313
 314 static inline void bond_set_slave_state(struct slave *slave,
 315                                         int slave_state, bool notify)
 316 {
 317         if (slave->backup == slave_state)
 318                 return;
 319
 320         slave->backup = slave_state;
 321         if (notify) {
 322                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_ATOMIC);
 323                 slave->should_notify = 0;
 324         } else {
 325                 if (slave->should_notify)
 326                         slave->should_notify = 0;
 327                 else
 328                         slave->should_notify = 1;
 329         }
 330 }
 331
 332 static inline void bond_slave_state_change(struct bonding *bond)
 333 {
 334         struct list_head *iter;
 335         struct slave *tmp;
 336
 337         bond_for_each_slave(bond, tmp, iter) {
 338                 if (tmp->link == BOND_LINK_UP)
 339                         bond_set_active_slave(tmp);
 340                 else if (tmp->link == BOND_LINK_DOWN)
 341                         bond_set_backup_slave(tmp);
 342         }
 343 }
 344
 345 static inline void bond_slave_state_notify(struct bonding *bond)
 346 {
 347         struct list_head *iter;
 348         struct slave *tmp;
 349
 350         bond_for_each_slave(bond, tmp, iter) {
 351                 if (tmp->should_notify) {
 352                         rtmsg_ifinfo(RTM_NEWLINK, tmp->dev, 0, GFP_ATOMIC);
 353                         tmp->should_notify = 0;
 354                 }
 355         }
 356 }
 357
 358 static inline int bond_slave_state(struct slave *slave)
 359 {
 360         return slave->backup;
 361 }
 362
 363 static inline bool bond_is_active_slave(struct slave *slave)
 364 {
 365         return !bond_slave_state(slave);
 366 }
 367
 368 #define BOND_PRI_RESELECT_ALWAYS        0
 369 #define BOND_PRI_RESELECT_BETTER        1
 370 #define BOND_PRI_RESELECT_FAILURE       2
 371
 372 #define BOND_FOM_NONE                   0
 373 #define BOND_FOM_ACTIVE                 1
 374 #define BOND_FOM_FOLLOW                 2
 375
 376 #define BOND_ARP_TARGETS_ANY            0
 377 #define BOND_ARP_TARGETS_ALL            1
 378
 379 #define BOND_ARP_VALIDATE_NONE          0
 380 #define BOND_ARP_VALIDATE_ACTIVE        (1 << BOND_STATE_ACTIVE)
 381 #define BOND_ARP_VALIDATE_BACKUP        (1 << BOND_STATE_BACKUP)
 382 #define BOND_ARP_VALIDATE_ALL           (BOND_ARP_VALIDATE_ACTIVE | \
 383                                          BOND_ARP_VALIDATE_BACKUP)
 384 #define BOND_ARP_FILTER                 (BOND_ARP_VALIDATE_ALL + 1)
 385 #define BOND_ARP_FILTER_ACTIVE          (BOND_ARP_VALIDATE_ACTIVE | \
 386                                          BOND_ARP_FILTER)
 387 #define BOND_ARP_FILTER_BACKUP          (BOND_ARP_VALIDATE_BACKUP | \
 388                                          BOND_ARP_FILTER)
 389
 390 #define BOND_SLAVE_NOTIFY_NOW           true
 391 #define BOND_SLAVE_NOTIFY_LATER         false
 392
 393 static inline int slave_do_arp_validate(struct bonding *bond,
 394                                         struct slave *slave)
 395 {
 396         return bond->params.arp_validate & (1 << bond_slave_state(slave));
 397 }
 398
 399 static inline int slave_do_arp_validate_only(struct bonding *bond,
 400                                              struct slave *slave)
 401 {
 402         return bond->params.arp_validate & BOND_ARP_FILTER;
 403 }
 404
 405 /* Get the oldest arp which we've received on this slave for bond's
 406  * arp_targets.
 407  */
 408 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
 409                                                        struct slave *slave)
 410 {
 411         int i = 1;
 412         unsigned long ret = slave->target_last_arp_rx[0];
 413
 414         for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
 415                 if (time_before(slave->target_last_arp_rx[i], ret))
 416                         ret = slave->target_last_arp_rx[i];
 417
 418         return ret;
 419 }
 420
 421 static inline unsigned long slave_last_rx(struct bonding *bond,
 422                                         struct slave *slave)
 423 {
 424         if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
 425                 return slave_oldest_target_arp_rx(bond, slave);
 426
 427         return slave->last_rx;
 428 }
 429
 430 #ifdef CONFIG_NET_POLL_CONTROLLER
 431 static inline void bond_netpoll_send_skb(const struct slave *slave,
 432                                          struct sk_buff *skb)
 433 {
 434         struct netpoll *np = slave->np;
 435
 436         if (np)
 437                 netpoll_send_skb(np, skb);
 438 }
 439 #else
 440 static inline void bond_netpoll_send_skb(const struct slave *slave,
 441                                          struct sk_buff *skb)
 442 {
 443 }
 444 #endif
 445
 446 static inline void bond_set_slave_inactive_flags(struct slave *slave,
 447                                                  bool notify)
 448 {
 449         if (!bond_is_lb(slave->bond))
 450                 bond_set_slave_state(slave, BOND_STATE_BACKUP, notify);
 451         if (!slave->bond->params.all_slaves_active)
 452                 slave->inactive = 1;
 453 }
 454
 455 static inline void bond_set_slave_active_flags(struct slave *slave,
 456                                                bool notify)
 457 {
 458         bond_set_slave_state(slave, BOND_STATE_ACTIVE, notify);
 459         slave->inactive = 0;
 460 }
 461
 462 static inline bool bond_is_slave_inactive(struct slave *slave)
 463 {
 464         return slave->inactive;
 465 }
 466
 467 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
 468 {
 469         struct in_device *in_dev;
 470         __be32 addr = 0;
 471
 472         rcu_read_lock();
 473         in_dev = __in_dev_get_rcu(dev);
 474
 475         if (in_dev)
 476                 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
 477                                          RT_SCOPE_HOST);
 478         rcu_read_unlock();
 479         return addr;
 480 }
 481
 482 static inline bool slave_can_tx(struct slave *slave)
 483 {
 484         if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP &&
 485             bond_is_active_slave(slave))
 486                 return true;
 487         else
 488                 return false;
 489 }
 490
 491 struct bond_net;
 492
 493 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
 494 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
 495 int bond_create(struct net *net, const char *name);
 496 int bond_create_sysfs(struct bond_net *net);
 497 void bond_destroy_sysfs(struct bond_net *net);
 498 void bond_prepare_sysfs_group(struct bonding *bond);
 499 int bond_sysfs_slave_add(struct slave *slave);
 500 void bond_sysfs_slave_del(struct slave *slave);
 501 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
 502 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
 503 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count);
 504 void bond_select_active_slave(struct bonding *bond);
 505 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
 506 void bond_create_debugfs(void);
 507 void bond_destroy_debugfs(void);
 508 void bond_debug_register(struct bonding *bond);
 509 void bond_debug_unregister(struct bonding *bond);
 510 void bond_debug_reregister(struct bonding *bond);
 511 const char *bond_mode_name(int mode);
 512 void bond_setup(struct net_device *bond_dev);
 513 unsigned int bond_get_num_tx_queues(void);
 514 int bond_netlink_init(void);
 515 void bond_netlink_fini(void);
 516 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
 517 struct net_device *bond_option_active_slave_get(struct bonding *bond);
 518 const char *bond_slave_link_status(s8 link);
 519
 520 struct bond_net {
 521         struct net *            net;    /* Associated network namespace */
 522         struct list_head        dev_list;
 523 #ifdef CONFIG_PROC_FS
 524         struct proc_dir_entry * proc_dir;
 525 #endif
 526         struct class_attribute  class_attr_bonding_masters;
 527 };
 528
 529 #ifdef CONFIG_PROC_FS
 530 void bond_create_proc_entry(struct bonding *bond);
 531 void bond_remove_proc_entry(struct bonding *bond);
 532 void bond_create_proc_dir(struct bond_net *bn);
 533 void bond_destroy_proc_dir(struct bond_net *bn);
 534 #else
 535 static inline void bond_create_proc_entry(struct bonding *bond)
 536 {
 537 }
 538
 539 static inline void bond_remove_proc_entry(struct bonding *bond)
 540 {
 541 }
 542
 543 static inline void bond_create_proc_dir(struct bond_net *bn)
 544 {
 545 }
 546
 547 static inline void bond_destroy_proc_dir(struct bond_net *bn)
 548 {
 549 }
 550 #endif
 551
 552 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
 553                                                const u8 *mac)
 554 {
 555         struct list_head *iter;
 556         struct slave *tmp;
 557
 558         bond_for_each_slave(bond, tmp, iter)
 559                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 560                         return tmp;
 561
 562         return NULL;
 563 }
 564
 565 /* Caller must hold rcu_read_lock() for read */
 566 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
 567                                                const u8 *mac)
 568 {
 569         struct list_head *iter;
 570         struct slave *tmp;
 571
 572         bond_for_each_slave_rcu(bond, tmp, iter)
 573                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 574                         return tmp;
 575
 576         return NULL;
 577 }
 578
 579 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
 580  * Returns -1 if not found, index if found
 581  */
 582 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
 583 {
 584         int i;
 585
 586         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
 587                 if (targets[i] == ip)
 588                         return i;
 589                 else if (targets[i] == 0)
 590                         break;
 591
 592         return -1;
 593 }
 594
 595 /* exported from bond_main.c */
 596 extern int bond_net_id;
 597 extern const struct bond_parm_tbl bond_lacp_tbl[];
 598 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
 599 extern const struct bond_parm_tbl arp_validate_tbl[];
 600 extern const struct bond_parm_tbl arp_all_targets_tbl[];
 601 extern const struct bond_parm_tbl fail_over_mac_tbl[];
 602 extern const struct bond_parm_tbl pri_reselect_tbl[];
 603 extern struct bond_parm_tbl ad_select_tbl[];
 604
 605 /* exported from bond_netlink.c */
 606 extern struct rtnl_link_ops bond_link_ops;
 607
 608 #endif /* _LINUX_BONDING_H */