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         unsigned long jiffies;
 192         unsigned long last_arp_rx;
 193         unsigned long target_last_arp_rx[BOND_MAX_ARP_TARGETS];
 194         s8     link;    /* one of BOND_LINK_XXXX */
 195         s8     new_link;
 196         u8     backup:1,   /* indicates backup slave. Value corresponds with
 197                               BOND_STATE_ACTIVE and BOND_STATE_BACKUP */
 198                inactive:1; /* indicates inactive slave */
 199         u8     duplex;
 200         u32    original_mtu;
 201         u32    link_failure_count;
 202         u32    speed;
 203         u16    queue_id;
 204         u8     perm_hwaddr[ETH_ALEN];
 205         struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
 206         struct tlb_slave_info tlb_info;
 207 #ifdef CONFIG_NET_POLL_CONTROLLER
 208         struct netpoll *np;
 209 #endif
 210         struct kobject kobj;
 211 };
 212
 213 /*
 214  * Link pseudo-state only used internally by monitors
 215  */
 216 #define BOND_LINK_NOCHANGE -1
 217
 218 /*
 219  * Here are the locking policies for the two bonding locks:
 220  *
 221  * 1) Get bond->lock when reading/writing slave list.
 222  * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
 223  *    (It is unnecessary when the write-lock is put with bond->lock.)
 224  * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
 225  *    beforehand.
 226  */
 227 struct bonding {
 228         struct   net_device *dev; /* first - useful for panic debug */
 229         struct   slave *curr_active_slave;
 230         struct   slave *current_arp_slave;
 231         struct   slave *primary_slave;
 232         bool     force_primary;
 233         s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
 234         int     (*recv_probe)(const struct sk_buff *, struct bonding *,
 235                               struct slave *);
 236         rwlock_t lock;
 237         rwlock_t curr_slave_lock;
 238         u8       send_peer_notif;
 239         u8       igmp_retrans;
 240 #ifdef CONFIG_PROC_FS
 241         struct   proc_dir_entry *proc_entry;
 242         char     proc_file_name[IFNAMSIZ];
 243 #endif /* CONFIG_PROC_FS */
 244         struct   list_head bond_list;
 245         u32      rr_tx_counter;
 246         struct   ad_bond_info ad_info;
 247         struct   alb_bond_info alb_info;
 248         struct   bond_params params;
 249         struct   workqueue_struct *wq;
 250         struct   delayed_work mii_work;
 251         struct   delayed_work arp_work;
 252         struct   delayed_work alb_work;
 253         struct   delayed_work ad_work;
 254         struct   delayed_work mcast_work;
 255 #ifdef CONFIG_DEBUG_FS
 256         /* debugging support via debugfs */
 257         struct   dentry *debug_dir;
 258 #endif /* CONFIG_DEBUG_FS */
 259 };
 260
 261 #define bond_slave_get_rcu(dev) \
 262         ((struct slave *) rcu_dereference(dev->rx_handler_data))
 263
 264 #define bond_slave_get_rtnl(dev) \
 265         ((struct slave *) rtnl_dereference(dev->rx_handler_data))
 266
 267 /**
 268  * Returns NULL if the net_device does not belong to any of the bond's slaves
 269  *
 270  * Caller must hold bond lock for read
 271  */
 272 static inline struct slave *bond_get_slave_by_dev(struct bonding *bond,
 273                                                   struct net_device *slave_dev)
 274 {
 275         return netdev_lower_dev_get_private(bond->dev, slave_dev);
 276 }
 277
 278 static inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
 279 {
 280         if (!slave || !slave->bond)
 281                 return NULL;
 282         return slave->bond;
 283 }
 284
 285 static inline bool bond_is_lb(const struct bonding *bond)
 286 {
 287         return BOND_MODE_IS_LB(bond->params.mode);
 288 }
 289
 290 static inline void bond_set_active_slave(struct slave *slave)
 291 {
 292         if (slave->backup) {
 293                 slave->backup = 0;
 294                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_KERNEL);
 295         }
 296 }
 297
 298 static inline void bond_set_backup_slave(struct slave *slave)
 299 {
 300         if (!slave->backup) {
 301                 slave->backup = 1;
 302                 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, 0, GFP_KERNEL);
 303         }
 304 }
 305
 306 static inline void bond_slave_state_change(struct bonding *bond)
 307 {
 308         struct list_head *iter;
 309         struct slave *tmp;
 310
 311         bond_for_each_slave(bond, tmp, iter) {
 312                 if (tmp->link == BOND_LINK_UP)
 313                         bond_set_active_slave(tmp);
 314                 else if (tmp->link == BOND_LINK_DOWN)
 315                         bond_set_backup_slave(tmp);
 316         }
 317 }
 318
 319 static inline int bond_slave_state(struct slave *slave)
 320 {
 321         return slave->backup;
 322 }
 323
 324 static inline bool bond_is_active_slave(struct slave *slave)
 325 {
 326         return !bond_slave_state(slave);
 327 }
 328
 329 #define BOND_PRI_RESELECT_ALWAYS        0
 330 #define BOND_PRI_RESELECT_BETTER        1
 331 #define BOND_PRI_RESELECT_FAILURE       2
 332
 333 #define BOND_FOM_NONE                   0
 334 #define BOND_FOM_ACTIVE                 1
 335 #define BOND_FOM_FOLLOW                 2
 336
 337 #define BOND_ARP_TARGETS_ANY            0
 338 #define BOND_ARP_TARGETS_ALL            1
 339
 340 #define BOND_ARP_VALIDATE_NONE          0
 341 #define BOND_ARP_VALIDATE_ACTIVE        (1 << BOND_STATE_ACTIVE)
 342 #define BOND_ARP_VALIDATE_BACKUP        (1 << BOND_STATE_BACKUP)
 343 #define BOND_ARP_VALIDATE_ALL           (BOND_ARP_VALIDATE_ACTIVE | \
 344                                          BOND_ARP_VALIDATE_BACKUP)
 345
 346 static inline int slave_do_arp_validate(struct bonding *bond,
 347                                         struct slave *slave)
 348 {
 349         return bond->params.arp_validate & (1 << bond_slave_state(slave));
 350 }
 351
 352 /* Get the oldest arp which we've received on this slave for bond's
 353  * arp_targets.
 354  */
 355 static inline unsigned long slave_oldest_target_arp_rx(struct bonding *bond,
 356                                                        struct slave *slave)
 357 {
 358         int i = 1;
 359         unsigned long ret = slave->target_last_arp_rx[0];
 360
 361         for (; (i < BOND_MAX_ARP_TARGETS) && bond->params.arp_targets[i]; i++)
 362                 if (time_before(slave->target_last_arp_rx[i], ret))
 363                         ret = slave->target_last_arp_rx[i];
 364
 365         return ret;
 366 }
 367
 368 static inline unsigned long slave_last_rx(struct bonding *bond,
 369                                         struct slave *slave)
 370 {
 371         if (slave_do_arp_validate(bond, slave)) {
 372                 if (bond->params.arp_all_targets == BOND_ARP_TARGETS_ALL)
 373                         return slave_oldest_target_arp_rx(bond, slave);
 374                 else
 375                         return slave->last_arp_rx;
 376         }
 377
 378         return slave->dev->last_rx;
 379 }
 380
 381 #ifdef CONFIG_NET_POLL_CONTROLLER
 382 static inline void bond_netpoll_send_skb(const struct slave *slave,
 383                                          struct sk_buff *skb)
 384 {
 385         struct netpoll *np = slave->np;
 386
 387         if (np)
 388                 netpoll_send_skb(np, skb);
 389 }
 390 #else
 391 static inline void bond_netpoll_send_skb(const struct slave *slave,
 392                                          struct sk_buff *skb)
 393 {
 394 }
 395 #endif
 396
 397 static inline void bond_set_slave_inactive_flags(struct slave *slave)
 398 {
 399         if (!bond_is_lb(slave->bond))
 400                 bond_set_backup_slave(slave);
 401         if (!slave->bond->params.all_slaves_active)
 402                 slave->inactive = 1;
 403 }
 404
 405 static inline void bond_set_slave_active_flags(struct slave *slave)
 406 {
 407         bond_set_active_slave(slave);
 408         slave->inactive = 0;
 409 }
 410
 411 static inline bool bond_is_slave_inactive(struct slave *slave)
 412 {
 413         return slave->inactive;
 414 }
 415
 416 static inline __be32 bond_confirm_addr(struct net_device *dev, __be32 dst, __be32 local)
 417 {
 418         struct in_device *in_dev;
 419         __be32 addr = 0;
 420
 421         rcu_read_lock();
 422         in_dev = __in_dev_get_rcu(dev);
 423
 424         if (in_dev)
 425                 addr = inet_confirm_addr(dev_net(dev), in_dev, dst, local,
 426                                          RT_SCOPE_HOST);
 427         rcu_read_unlock();
 428         return addr;
 429 }
 430
 431 static inline bool slave_can_tx(struct slave *slave)
 432 {
 433         if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP &&
 434             bond_is_active_slave(slave))
 435                 return true;
 436         else
 437                 return false;
 438 }
 439
 440 struct bond_net;
 441
 442 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond, struct slave *slave);
 443 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
 444 int bond_create(struct net *net, const char *name);
 445 int bond_create_sysfs(struct bond_net *net);
 446 void bond_destroy_sysfs(struct bond_net *net);
 447 void bond_prepare_sysfs_group(struct bonding *bond);
 448 int bond_sysfs_slave_add(struct slave *slave);
 449 void bond_sysfs_slave_del(struct slave *slave);
 450 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev);
 451 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev);
 452 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count);
 453 int bond_parse_parm(const char *mode_arg, const struct bond_parm_tbl *tbl);
 454 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl);
 455 void bond_select_active_slave(struct bonding *bond);
 456 void bond_change_active_slave(struct bonding *bond, struct slave *new_active);
 457 void bond_create_debugfs(void);
 458 void bond_destroy_debugfs(void);
 459 void bond_debug_register(struct bonding *bond);
 460 void bond_debug_unregister(struct bonding *bond);
 461 void bond_debug_reregister(struct bonding *bond);
 462 const char *bond_mode_name(int mode);
 463 void bond_setup(struct net_device *bond_dev);
 464 unsigned int bond_get_num_tx_queues(void);
 465 int bond_netlink_init(void);
 466 void bond_netlink_fini(void);
 467 int bond_option_arp_ip_target_add(struct bonding *bond, __be32 target);
 468 int bond_option_arp_ip_target_rem(struct bonding *bond, __be32 target);
 469 struct net_device *bond_option_active_slave_get_rcu(struct bonding *bond);
 470 struct net_device *bond_option_active_slave_get(struct bonding *bond);
 471 const char *bond_slave_link_status(s8 link);
 472
 473 struct bond_net {
 474         struct net *            net;    /* Associated network namespace */
 475         struct list_head        dev_list;
 476 #ifdef CONFIG_PROC_FS
 477         struct proc_dir_entry * proc_dir;
 478 #endif
 479         struct class_attribute  class_attr_bonding_masters;
 480 };
 481
 482 #ifdef CONFIG_PROC_FS
 483 void bond_create_proc_entry(struct bonding *bond);
 484 void bond_remove_proc_entry(struct bonding *bond);
 485 void bond_create_proc_dir(struct bond_net *bn);
 486 void bond_destroy_proc_dir(struct bond_net *bn);
 487 #else
 488 static inline void bond_create_proc_entry(struct bonding *bond)
 489 {
 490 }
 491
 492 static inline void bond_remove_proc_entry(struct bonding *bond)
 493 {
 494 }
 495
 496 static inline void bond_create_proc_dir(struct bond_net *bn)
 497 {
 498 }
 499
 500 static inline void bond_destroy_proc_dir(struct bond_net *bn)
 501 {
 502 }
 503 #endif
 504
 505 static inline struct slave *bond_slave_has_mac(struct bonding *bond,
 506                                                const u8 *mac)
 507 {
 508         struct list_head *iter;
 509         struct slave *tmp;
 510
 511         bond_for_each_slave(bond, tmp, iter)
 512                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 513                         return tmp;
 514
 515         return NULL;
 516 }
 517
 518 /* Caller must hold rcu_read_lock() for read */
 519 static inline struct slave *bond_slave_has_mac_rcu(struct bonding *bond,
 520                                                const u8 *mac)
 521 {
 522         struct list_head *iter;
 523         struct slave *tmp;
 524
 525         bond_for_each_slave_rcu(bond, tmp, iter)
 526                 if (ether_addr_equal_64bits(mac, tmp->dev->dev_addr))
 527                         return tmp;
 528
 529         return NULL;
 530 }
 531
 532 /* Check if the ip is present in arp ip list, or first free slot if ip == 0
 533  * Returns -1 if not found, index if found
 534  */
 535 static inline int bond_get_targets_ip(__be32 *targets, __be32 ip)
 536 {
 537         int i;
 538
 539         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
 540                 if (targets[i] == ip)
 541                         return i;
 542                 else if (targets[i] == 0)
 543                         break;
 544
 545         return -1;
 546 }
 547
 548 /* exported from bond_main.c */
 549 extern int bond_net_id;
 550 extern const struct bond_parm_tbl bond_lacp_tbl[];
 551 extern const struct bond_parm_tbl xmit_hashtype_tbl[];
 552 extern const struct bond_parm_tbl arp_validate_tbl[];
 553 extern const struct bond_parm_tbl arp_all_targets_tbl[];
 554 extern const struct bond_parm_tbl fail_over_mac_tbl[];
 555 extern const struct bond_parm_tbl pri_reselect_tbl[];
 556 extern struct bond_parm_tbl ad_select_tbl[];
 557
 558 /* exported from bond_netlink.c */
 559 extern struct rtnl_link_ops bond_link_ops;
 560
 561 #endif /* _LINUX_BONDING_H */