git.oblomov.eu Git - linux-2.6/blob - drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 //#define BONDING_DEBUG 1
  35
  36 #include <linux/config.h>
  37 #include <linux/kernel.h>
  38 #include <linux/module.h>
  39 #include <linux/sched.h>
  40 #include <linux/types.h>
  41 #include <linux/fcntl.h>
  42 #include <linux/interrupt.h>
  43 #include <linux/ptrace.h>
  44 #include <linux/ioport.h>
  45 #include <linux/in.h>
  46 #include <net/ip.h>
  47 #include <linux/ip.h>
  48 #include <linux/tcp.h>
  49 #include <linux/udp.h>
  50 #include <linux/slab.h>
  51 #include <linux/string.h>
  52 #include <linux/init.h>
  53 #include <linux/timer.h>
  54 #include <linux/socket.h>
  55 #include <linux/ctype.h>
  56 #include <linux/inet.h>
  57 #include <linux/bitops.h>
  58 #include <asm/system.h>
  59 #include <asm/io.h>
  60 #include <asm/dma.h>
  61 #include <asm/uaccess.h>
  62 #include <linux/errno.h>
  63 #include <linux/netdevice.h>
  64 #include <linux/inetdevice.h>
  65 #include <linux/etherdevice.h>
  66 #include <linux/skbuff.h>
  67 #include <net/sock.h>
  68 #include <linux/rtnetlink.h>
  69 #include <linux/proc_fs.h>
  70 #include <linux/seq_file.h>
  71 #include <linux/smp.h>
  72 #include <linux/if_ether.h>
  73 #include <net/arp.h>
  74 #include <linux/mii.h>
  75 #include <linux/ethtool.h>
  76 #include <linux/if_vlan.h>
  77 #include <linux/if_bonding.h>
  78 #include <net/route.h>
  79 #include "bonding.h"
  80 #include "bond_3ad.h"
  81 #include "bond_alb.h"
  82
  83 /*---------------------------- Module parameters ----------------------------*/
  84
  85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  86 #define BOND_LINK_MON_INTERV    0
  87 #define BOND_LINK_ARP_INTERV    0
  88
  89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  90 static int miimon       = BOND_LINK_MON_INTERV;
  91 static int updelay      = 0;
  92 static int downdelay    = 0;
  93 static int use_carrier  = 1;
  94 static char *mode       = NULL;
  95 static char *primary    = NULL;
  96 static char *lacp_rate  = NULL;
  97 static char *xmit_hash_policy = NULL;
  98 static int arp_interval = BOND_LINK_ARP_INTERV;
  99 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 100 struct bond_params bonding_defaults;
 101
 102 module_param(max_bonds, int, 0);
 103 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 104 module_param(miimon, int, 0);
 105 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 106 module_param(updelay, int, 0);
 107 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 108 module_param(downdelay, int, 0);
 109 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 110                             "in milliseconds");
 111 module_param(use_carrier, int, 0);
 112 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 113                               "0 for off, 1 for on (default)");
 114 module_param(mode, charp, 0);
 115 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 116                        "1 for active-backup, 2 for balance-xor, "
 117                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 118                        "6 for balance-alb");
 119 module_param(primary, charp, 0);
 120 MODULE_PARM_DESC(primary, "Primary network device to use");
 121 module_param(lacp_rate, charp, 0);
 122 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 123                             "(slow/fast)");
 124 module_param(xmit_hash_policy, charp, 0);
 125 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 126                                    ", 1 for layer 3+4");
 127 module_param(arp_interval, int, 0);
 128 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 129 module_param_array(arp_ip_target, charp, NULL, 0);
 130 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 131
 132 /*----------------------------- Global variables ----------------------------*/
 133
 134 static const char * const version =
 135         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 136
 137 LIST_HEAD(bond_dev_list);
 138
 139 #ifdef CONFIG_PROC_FS
 140 static struct proc_dir_entry *bond_proc_dir = NULL;
 141 #endif
 142
 143 extern struct rw_semaphore bonding_rwsem;
 144 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 145 static int arp_ip_count = 0;
 146 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 147 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 148 static int lacp_fast    = 0;
 149
 150
 151 struct bond_parm_tbl bond_lacp_tbl[] = {
 152 {       "slow",         AD_LACP_SLOW},
 153 {       "fast",         AD_LACP_FAST},
 154 {       NULL,           -1},
 155 };
 156
 157 struct bond_parm_tbl bond_mode_tbl[] = {
 158 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 159 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 160 {       "balance-xor",          BOND_MODE_XOR},
 161 {       "broadcast",            BOND_MODE_BROADCAST},
 162 {       "802.3ad",              BOND_MODE_8023AD},
 163 {       "balance-tlb",          BOND_MODE_TLB},
 164 {       "balance-alb",          BOND_MODE_ALB},
 165 {       NULL,                   -1},
 166 };
 167
 168 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 169 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 170 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 171 {       NULL,                   -1},
 172 };
 173
 174 /*-------------------------- Forward declarations ---------------------------*/
 175
 176 static void bond_send_gratuitous_arp(struct bonding *bond);
 177
 178 /*---------------------------- General routines -----------------------------*/
 179
 180 const char *bond_mode_name(int mode)
 181 {
 182         switch (mode) {
 183         case BOND_MODE_ROUNDROBIN :
 184                 return "load balancing (round-robin)";
 185         case BOND_MODE_ACTIVEBACKUP :
 186                 return "fault-tolerance (active-backup)";
 187         case BOND_MODE_XOR :
 188                 return "load balancing (xor)";
 189         case BOND_MODE_BROADCAST :
 190                 return "fault-tolerance (broadcast)";
 191         case BOND_MODE_8023AD:
 192                 return "IEEE 802.3ad Dynamic link aggregation";
 193         case BOND_MODE_TLB:
 194                 return "transmit load balancing";
 195         case BOND_MODE_ALB:
 196                 return "adaptive load balancing";
 197         default:
 198                 return "unknown";
 199         }
 200 }
 201
 202 /*---------------------------------- VLAN -----------------------------------*/
 203
 204 /**
 205  * bond_add_vlan - add a new vlan id on bond
 206  * @bond: bond that got the notification
 207  * @vlan_id: the vlan id to add
 208  *
 209  * Returns -ENOMEM if allocation failed.
 210  */
 211 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 212 {
 213         struct vlan_entry *vlan;
 214
 215         dprintk("bond: %s, vlan id %d\n",
 216                 (bond ? bond->dev->name: "None"), vlan_id);
 217
 218         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 219         if (!vlan) {
 220                 return -ENOMEM;
 221         }
 222
 223         INIT_LIST_HEAD(&vlan->vlan_list);
 224         vlan->vlan_id = vlan_id;
 225         vlan->vlan_ip = 0;
 226
 227         write_lock_bh(&bond->lock);
 228
 229         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 230
 231         write_unlock_bh(&bond->lock);
 232
 233         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 234
 235         return 0;
 236 }
 237
 238 /**
 239  * bond_del_vlan - delete a vlan id from bond
 240  * @bond: bond that got the notification
 241  * @vlan_id: the vlan id to delete
 242  *
 243  * returns -ENODEV if @vlan_id was not found in @bond.
 244  */
 245 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 246 {
 247         struct vlan_entry *vlan, *next;
 248         int res = -ENODEV;
 249
 250         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 251
 252         write_lock_bh(&bond->lock);
 253
 254         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 255                 if (vlan->vlan_id == vlan_id) {
 256                         list_del(&vlan->vlan_list);
 257
 258                         if ((bond->params.mode == BOND_MODE_TLB) ||
 259                             (bond->params.mode == BOND_MODE_ALB)) {
 260                                 bond_alb_clear_vlan(bond, vlan_id);
 261                         }
 262
 263                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 264                                 bond->dev->name);
 265
 266                         kfree(vlan);
 267
 268                         if (list_empty(&bond->vlan_list) &&
 269                             (bond->slave_cnt == 0)) {
 270                                 /* Last VLAN removed and no slaves, so
 271                                  * restore block on adding VLANs. This will
 272                                  * be removed once new slaves that are not
 273                                  * VLAN challenged will be added.
 274                                  */
 275                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 276                         }
 277
 278                         res = 0;
 279                         goto out;
 280                 }
 281         }
 282
 283         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 284                 bond->dev->name);
 285
 286 out:
 287         write_unlock_bh(&bond->lock);
 288         return res;
 289 }
 290
 291 /**
 292  * bond_has_challenged_slaves
 293  * @bond: the bond we're working on
 294  *
 295  * Searches the slave list. Returns 1 if a vlan challenged slave
 296  * was found, 0 otherwise.
 297  *
 298  * Assumes bond->lock is held.
 299  */
 300 static int bond_has_challenged_slaves(struct bonding *bond)
 301 {
 302         struct slave *slave;
 303         int i;
 304
 305         bond_for_each_slave(bond, slave, i) {
 306                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 307                         dprintk("found VLAN challenged slave - %s\n",
 308                                 slave->dev->name);
 309                         return 1;
 310                 }
 311         }
 312
 313         dprintk("no VLAN challenged slaves found\n");
 314         return 0;
 315 }
 316
 317 /**
 318  * bond_next_vlan - safely skip to the next item in the vlans list.
 319  * @bond: the bond we're working on
 320  * @curr: item we're advancing from
 321  *
 322  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 323  * or @curr->next otherwise (even if it is @curr itself again).
 324  *
 325  * Caller must hold bond->lock
 326  */
 327 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 328 {
 329         struct vlan_entry *next, *last;
 330
 331         if (list_empty(&bond->vlan_list)) {
 332                 return NULL;
 333         }
 334
 335         if (!curr) {
 336                 next = list_entry(bond->vlan_list.next,
 337                                   struct vlan_entry, vlan_list);
 338         } else {
 339                 last = list_entry(bond->vlan_list.prev,
 340                                   struct vlan_entry, vlan_list);
 341                 if (last == curr) {
 342                         next = list_entry(bond->vlan_list.next,
 343                                           struct vlan_entry, vlan_list);
 344                 } else {
 345                         next = list_entry(curr->vlan_list.next,
 346                                           struct vlan_entry, vlan_list);
 347                 }
 348         }
 349
 350         return next;
 351 }
 352
 353 /**
 354  * bond_dev_queue_xmit - Prepare skb for xmit.
 355  *
 356  * @bond: bond device that got this skb for tx.
 357  * @skb: hw accel VLAN tagged skb to transmit
 358  * @slave_dev: slave that is supposed to xmit this skbuff
 359  *
 360  * When the bond gets an skb to transmit that is
 361  * already hardware accelerated VLAN tagged, and it
 362  * needs to relay this skb to a slave that is not
 363  * hw accel capable, the skb needs to be "unaccelerated",
 364  * i.e. strip the hwaccel tag and re-insert it as part
 365  * of the payload.
 366  */
 367 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 368 {
 369         unsigned short vlan_id;
 370
 371         if (!list_empty(&bond->vlan_list) &&
 372             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 373             vlan_get_tag(skb, &vlan_id) == 0) {
 374                 skb->dev = slave_dev;
 375                 skb = vlan_put_tag(skb, vlan_id);
 376                 if (!skb) {
 377                         /* vlan_put_tag() frees the skb in case of error,
 378                          * so return success here so the calling functions
 379                          * won't attempt to free is again.
 380                          */
 381                         return 0;
 382                 }
 383         } else {
 384                 skb->dev = slave_dev;
 385         }
 386
 387         skb->priority = 1;
 388         dev_queue_xmit(skb);
 389
 390         return 0;
 391 }
 392
 393 /*
 394  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 395  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 396  * lock because:
 397  * a. This operation is performed in IOCTL context,
 398  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 399  * c. Holding a lock with BH disabled while directly calling a base driver
 400  *    entry point is generally a BAD idea.
 401  *
 402  * The design of synchronization/protection for this operation in the 8021q
 403  * module is good for one or more VLAN devices over a single physical device
 404  * and cannot be extended for a teaming solution like bonding, so there is a
 405  * potential race condition here where a net device from the vlan group might
 406  * be referenced (either by a base driver or the 8021q code) while it is being
 407  * removed from the system. However, it turns out we're not making matters
 408  * worse, and if it works for regular VLAN usage it will work here too.
 409 */
 410
 411 /**
 412  * bond_vlan_rx_register - Propagates registration to slaves
 413  * @bond_dev: bonding net device that got called
 414  * @grp: vlan group being registered
 415  */
 416 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 417 {
 418         struct bonding *bond = bond_dev->priv;
 419         struct slave *slave;
 420         int i;
 421
 422         bond->vlgrp = grp;
 423
 424         bond_for_each_slave(bond, slave, i) {
 425                 struct net_device *slave_dev = slave->dev;
 426
 427                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 428                     slave_dev->vlan_rx_register) {
 429                         slave_dev->vlan_rx_register(slave_dev, grp);
 430                 }
 431         }
 432 }
 433
 434 /**
 435  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 436  * @bond_dev: bonding net device that got called
 437  * @vid: vlan id being added
 438  */
 439 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 440 {
 441         struct bonding *bond = bond_dev->priv;
 442         struct slave *slave;
 443         int i, res;
 444
 445         bond_for_each_slave(bond, slave, i) {
 446                 struct net_device *slave_dev = slave->dev;
 447
 448                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 449                     slave_dev->vlan_rx_add_vid) {
 450                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 451                 }
 452         }
 453
 454         res = bond_add_vlan(bond, vid);
 455         if (res) {
 456                 printk(KERN_ERR DRV_NAME
 457                        ": %s: Error: Failed to add vlan id %d\n",
 458                        bond_dev->name, vid);
 459         }
 460 }
 461
 462 /**
 463  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 464  * @bond_dev: bonding net device that got called
 465  * @vid: vlan id being removed
 466  */
 467 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 468 {
 469         struct bonding *bond = bond_dev->priv;
 470         struct slave *slave;
 471         struct net_device *vlan_dev;
 472         int i, res;
 473
 474         bond_for_each_slave(bond, slave, i) {
 475                 struct net_device *slave_dev = slave->dev;
 476
 477                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 478                     slave_dev->vlan_rx_kill_vid) {
 479                         /* Save and then restore vlan_dev in the grp array,
 480                          * since the slave's driver might clear it.
 481                          */
 482                         vlan_dev = bond->vlgrp->vlan_devices[vid];
 483                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 484                         bond->vlgrp->vlan_devices[vid] = vlan_dev;
 485                 }
 486         }
 487
 488         res = bond_del_vlan(bond, vid);
 489         if (res) {
 490                 printk(KERN_ERR DRV_NAME
 491                        ": %s: Error: Failed to remove vlan id %d\n",
 492                        bond_dev->name, vid);
 493         }
 494 }
 495
 496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 497 {
 498         struct vlan_entry *vlan;
 499
 500         write_lock_bh(&bond->lock);
 501
 502         if (list_empty(&bond->vlan_list)) {
 503                 goto out;
 504         }
 505
 506         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 507             slave_dev->vlan_rx_register) {
 508                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 509         }
 510
 511         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 512             !(slave_dev->vlan_rx_add_vid)) {
 513                 goto out;
 514         }
 515
 516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 517                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 518         }
 519
 520 out:
 521         write_unlock_bh(&bond->lock);
 522 }
 523
 524 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 525 {
 526         struct vlan_entry *vlan;
 527         struct net_device *vlan_dev;
 528
 529         write_lock_bh(&bond->lock);
 530
 531         if (list_empty(&bond->vlan_list)) {
 532                 goto out;
 533         }
 534
 535         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 536             !(slave_dev->vlan_rx_kill_vid)) {
 537                 goto unreg;
 538         }
 539
 540         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 541                 /* Save and then restore vlan_dev in the grp array,
 542                  * since the slave's driver might clear it.
 543                  */
 544                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
 545                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 546                 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
 547         }
 548
 549 unreg:
 550         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 551             slave_dev->vlan_rx_register) {
 552                 slave_dev->vlan_rx_register(slave_dev, NULL);
 553         }
 554
 555 out:
 556         write_unlock_bh(&bond->lock);
 557 }
 558
 559 /*------------------------------- Link status -------------------------------*/
 560
 561 /*
 562  * Set the carrier state for the master according to the state of its
 563  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 564  * do special 802.3ad magic.
 565  *
 566  * Returns zero if carrier state does not change, nonzero if it does.
 567  */
 568 static int bond_set_carrier(struct bonding *bond)
 569 {
 570         struct slave *slave;
 571         int i;
 572
 573         if (bond->slave_cnt == 0)
 574                 goto down;
 575
 576         if (bond->params.mode == BOND_MODE_8023AD)
 577                 return bond_3ad_set_carrier(bond);
 578
 579         bond_for_each_slave(bond, slave, i) {
 580                 if (slave->link == BOND_LINK_UP) {
 581                         if (!netif_carrier_ok(bond->dev)) {
 582                                 netif_carrier_on(bond->dev);
 583                                 return 1;
 584                         }
 585                         return 0;
 586                 }
 587         }
 588
 589 down:
 590         if (netif_carrier_ok(bond->dev)) {
 591                 netif_carrier_off(bond->dev);
 592                 return 1;
 593         }
 594         return 0;
 595 }
 596
 597 /*
 598  * Get link speed and duplex from the slave's base driver
 599  * using ethtool. If for some reason the call fails or the
 600  * values are invalid, fake speed and duplex to 100/Full
 601  * and return error.
 602  */
 603 static int bond_update_speed_duplex(struct slave *slave)
 604 {
 605         struct net_device *slave_dev = slave->dev;
 606         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 607         struct ifreq ifr;
 608         struct ethtool_cmd etool;
 609
 610         /* Fake speed and duplex */
 611         slave->speed = SPEED_100;
 612         slave->duplex = DUPLEX_FULL;
 613
 614         if (slave_dev->ethtool_ops) {
 615                 int res;
 616
 617                 if (!slave_dev->ethtool_ops->get_settings) {
 618                         return -1;
 619                 }
 620
 621                 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 622                 if (res < 0) {
 623                         return -1;
 624                 }
 625
 626                 goto verify;
 627         }
 628
 629         ioctl = slave_dev->do_ioctl;
 630         strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 631         etool.cmd = ETHTOOL_GSET;
 632         ifr.ifr_data = (char*)&etool;
 633         if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
 634                 return -1;
 635         }
 636
 637 verify:
 638         switch (etool.speed) {
 639         case SPEED_10:
 640         case SPEED_100:
 641         case SPEED_1000:
 642                 break;
 643         default:
 644                 return -1;
 645         }
 646
 647         switch (etool.duplex) {
 648         case DUPLEX_FULL:
 649         case DUPLEX_HALF:
 650                 break;
 651         default:
 652                 return -1;
 653         }
 654
 655         slave->speed = etool.speed;
 656         slave->duplex = etool.duplex;
 657
 658         return 0;
 659 }
 660
 661 /*
 662  * if <dev> supports MII link status reporting, check its link status.
 663  *
 664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 665  * depening upon the setting of the use_carrier parameter.
 666  *
 667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 668  * can't tell and just pretend it is), or 0, meaning that the link is
 669  * down.
 670  *
 671  * If reporting is non-zero, instead of faking link up, return -1 if
 672  * both ETHTOOL and MII ioctls fail (meaning the device does not
 673  * support them).  If use_carrier is set, return whatever it says.
 674  * It'd be nice if there was a good way to tell if a driver supports
 675  * netif_carrier, but there really isn't.
 676  */
 677 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 678 {
 679         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 680         struct ifreq ifr;
 681         struct mii_ioctl_data *mii;
 682         struct ethtool_value etool;
 683
 684         if (bond->params.use_carrier) {
 685                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 686         }
 687
 688         ioctl = slave_dev->do_ioctl;
 689         if (ioctl) {
 690                 /* TODO: set pointer to correct ioctl on a per team member */
 691                 /*       bases to make this more efficient. that is, once  */
 692                 /*       we determine the correct ioctl, we will always    */
 693                 /*       call it and not the others for that team          */
 694                 /*       member.                                           */
 695
 696                 /*
 697                  * We cannot assume that SIOCGMIIPHY will also read a
 698                  * register; not all network drivers (e.g., e100)
 699                  * support that.
 700                  */
 701
 702                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 703                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 704                 mii = if_mii(&ifr);
 705                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 706                         mii->reg_num = MII_BMSR;
 707                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 708                                 return (mii->val_out & BMSR_LSTATUS);
 709                         }
 710                 }
 711         }
 712
 713         /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
 714         /* for a period of time so we attempt to get link status   */
 715         /* from it last if the above MII ioctls fail...            */
 716         if (slave_dev->ethtool_ops) {
 717                 if (slave_dev->ethtool_ops->get_link) {
 718                         u32 link;
 719
 720                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 721
 722                         return link ? BMSR_LSTATUS : 0;
 723                 }
 724         }
 725
 726         if (ioctl) {
 727                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 728                 etool.cmd = ETHTOOL_GLINK;
 729                 ifr.ifr_data = (char*)&etool;
 730                 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
 731                         if (etool.data == 1) {
 732                                 return BMSR_LSTATUS;
 733                         } else {
 734                                 dprintk("SIOCETHTOOL shows link down\n");
 735                                 return 0;
 736                         }
 737                 }
 738         }
 739
 740         /*
 741          * If reporting, report that either there's no dev->do_ioctl,
 742          * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
 743          * cannot report link status).  If not reporting, pretend
 744          * we're ok.
 745          */
 746         return (reporting ? -1 : BMSR_LSTATUS);
 747 }
 748
 749 /*----------------------------- Multicast list ------------------------------*/
 750
 751 /*
 752  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 753  */
 754 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 755 {
 756         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 757                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 758 }
 759
 760 /*
 761  * returns dmi entry if found, NULL otherwise
 762  */
 763 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 764 {
 765         struct dev_mc_list *idmi;
 766
 767         for (idmi = mc_list; idmi; idmi = idmi->next) {
 768                 if (bond_is_dmi_same(dmi, idmi)) {
 769                         return idmi;
 770                 }
 771         }
 772
 773         return NULL;
 774 }
 775
 776 /*
 777  * Push the promiscuity flag down to appropriate slaves
 778  */
 779 static void bond_set_promiscuity(struct bonding *bond, int inc)
 780 {
 781         if (USES_PRIMARY(bond->params.mode)) {
 782                 /* write lock already acquired */
 783                 if (bond->curr_active_slave) {
 784                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 785                 }
 786         } else {
 787                 struct slave *slave;
 788                 int i;
 789                 bond_for_each_slave(bond, slave, i) {
 790                         dev_set_promiscuity(slave->dev, inc);
 791                 }
 792         }
 793 }
 794
 795 /*
 796  * Push the allmulti flag down to all slaves
 797  */
 798 static void bond_set_allmulti(struct bonding *bond, int inc)
 799 {
 800         if (USES_PRIMARY(bond->params.mode)) {
 801                 /* write lock already acquired */
 802                 if (bond->curr_active_slave) {
 803                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 804                 }
 805         } else {
 806                 struct slave *slave;
 807                 int i;
 808                 bond_for_each_slave(bond, slave, i) {
 809                         dev_set_allmulti(slave->dev, inc);
 810                 }
 811         }
 812 }
 813
 814 /*
 815  * Add a Multicast address to slaves
 816  * according to mode
 817  */
 818 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 819 {
 820         if (USES_PRIMARY(bond->params.mode)) {
 821                 /* write lock already acquired */
 822                 if (bond->curr_active_slave) {
 823                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 824                 }
 825         } else {
 826                 struct slave *slave;
 827                 int i;
 828                 bond_for_each_slave(bond, slave, i) {
 829                         dev_mc_add(slave->dev, addr, alen, 0);
 830                 }
 831         }
 832 }
 833
 834 /*
 835  * Remove a multicast address from slave
 836  * according to mode
 837  */
 838 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 839 {
 840         if (USES_PRIMARY(bond->params.mode)) {
 841                 /* write lock already acquired */
 842                 if (bond->curr_active_slave) {
 843                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 844                 }
 845         } else {
 846                 struct slave *slave;
 847                 int i;
 848                 bond_for_each_slave(bond, slave, i) {
 849                         dev_mc_delete(slave->dev, addr, alen, 0);
 850                 }
 851         }
 852 }
 853
 854 /*
 855  * Totally destroys the mc_list in bond
 856  */
 857 static void bond_mc_list_destroy(struct bonding *bond)
 858 {
 859         struct dev_mc_list *dmi;
 860
 861         dmi = bond->mc_list;
 862         while (dmi) {
 863                 bond->mc_list = dmi->next;
 864                 kfree(dmi);
 865                 dmi = bond->mc_list;
 866         }
 867 }
 868
 869 /*
 870  * Copy all the Multicast addresses from src to the bonding device dst
 871  */
 872 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 873                              gfp_t gfp_flag)
 874 {
 875         struct dev_mc_list *dmi, *new_dmi;
 876
 877         for (dmi = mc_list; dmi; dmi = dmi->next) {
 878                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 879
 880                 if (!new_dmi) {
 881                         /* FIXME: Potential memory leak !!! */
 882                         return -ENOMEM;
 883                 }
 884
 885                 new_dmi->next = bond->mc_list;
 886                 bond->mc_list = new_dmi;
 887                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 888                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 889                 new_dmi->dmi_users = dmi->dmi_users;
 890                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 891         }
 892
 893         return 0;
 894 }
 895
 896 /*
 897  * flush all members of flush->mc_list from device dev->mc_list
 898  */
 899 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 900 {
 901         struct bonding *bond = bond_dev->priv;
 902         struct dev_mc_list *dmi;
 903
 904         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 905                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 906         }
 907
 908         if (bond->params.mode == BOND_MODE_8023AD) {
 909                 /* del lacpdu mc addr from mc list */
 910                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 911
 912                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 913         }
 914 }
 915
 916 /*--------------------------- Active slave change ---------------------------*/
 917
 918 /*
 919  * Update the mc list and multicast-related flags for the new and
 920  * old active slaves (if any) according to the multicast mode, and
 921  * promiscuous flags unconditionally.
 922  */
 923 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 924 {
 925         struct dev_mc_list *dmi;
 926
 927         if (!USES_PRIMARY(bond->params.mode)) {
 928                 /* nothing to do -  mc list is already up-to-date on
 929                  * all slaves
 930                  */
 931                 return;
 932         }
 933
 934         if (old_active) {
 935                 if (bond->dev->flags & IFF_PROMISC) {
 936                         dev_set_promiscuity(old_active->dev, -1);
 937                 }
 938
 939                 if (bond->dev->flags & IFF_ALLMULTI) {
 940                         dev_set_allmulti(old_active->dev, -1);
 941                 }
 942
 943                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 944                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 945                 }
 946         }
 947
 948         if (new_active) {
 949                 if (bond->dev->flags & IFF_PROMISC) {
 950                         dev_set_promiscuity(new_active->dev, 1);
 951                 }
 952
 953                 if (bond->dev->flags & IFF_ALLMULTI) {
 954                         dev_set_allmulti(new_active->dev, 1);
 955                 }
 956
 957                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 958                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 959                 }
 960         }
 961 }
 962
 963 /**
 964  * find_best_interface - select the best available slave to be the active one
 965  * @bond: our bonding struct
 966  *
 967  * Warning: Caller must hold curr_slave_lock for writing.
 968  */
 969 static struct slave *bond_find_best_slave(struct bonding *bond)
 970 {
 971         struct slave *new_active, *old_active;
 972         struct slave *bestslave = NULL;
 973         int mintime = bond->params.updelay;
 974         int i;
 975
 976         new_active = old_active = bond->curr_active_slave;
 977
 978         if (!new_active) { /* there were no active slaves left */
 979                 if (bond->slave_cnt > 0) {  /* found one slave */
 980                         new_active = bond->first_slave;
 981                 } else {
 982                         return NULL; /* still no slave, return NULL */
 983                 }
 984         }
 985
 986         /* first try the primary link; if arping, a link must tx/rx traffic
 987          * before it can be considered the curr_active_slave - also, we would skip
 988          * slaves between the curr_active_slave and primary_slave that may be up
 989          * and able to arp
 990          */
 991         if ((bond->primary_slave) &&
 992             (!bond->params.arp_interval) &&
 993             (IS_UP(bond->primary_slave->dev))) {
 994                 new_active = bond->primary_slave;
 995         }
 996
 997         /* remember where to stop iterating over the slaves */
 998         old_active = new_active;
 999
1000         bond_for_each_slave_from(bond, new_active, i, old_active) {
1001                 if (IS_UP(new_active->dev)) {
1002                         if (new_active->link == BOND_LINK_UP) {
1003                                 return new_active;
1004                         } else if (new_active->link == BOND_LINK_BACK) {
1005                                 /* link up, but waiting for stabilization */
1006                                 if (new_active->delay < mintime) {
1007                                         mintime = new_active->delay;
1008                                         bestslave = new_active;
1009                                 }
1010                         }
1011                 }
1012         }
1013
1014         return bestslave;
1015 }
1016
1017 /**
1018  * change_active_interface - change the active slave into the specified one
1019  * @bond: our bonding struct
1020  * @new: the new slave to make the active one
1021  *
1022  * Set the new slave to the bond's settings and unset them on the old
1023  * curr_active_slave.
1024  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1025  *
1026  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1027  * because it is apparently the best available slave we have, even though its
1028  * updelay hasn't timed out yet.
1029  *
1030  * Warning: Caller must hold curr_slave_lock for writing.
1031  */
1032 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1033 {
1034         struct slave *old_active = bond->curr_active_slave;
1035
1036         if (old_active == new_active) {
1037                 return;
1038         }
1039
1040         if (new_active) {
1041                 if (new_active->link == BOND_LINK_BACK) {
1042                         if (USES_PRIMARY(bond->params.mode)) {
1043                                 printk(KERN_INFO DRV_NAME
1044                                        ": %s: making interface %s the new "
1045                                        "active one %d ms earlier.\n",
1046                                        bond->dev->name, new_active->dev->name,
1047                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1048                         }
1049
1050                         new_active->delay = 0;
1051                         new_active->link = BOND_LINK_UP;
1052                         new_active->jiffies = jiffies;
1053
1054                         if (bond->params.mode == BOND_MODE_8023AD) {
1055                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1056                         }
1057
1058                         if ((bond->params.mode == BOND_MODE_TLB) ||
1059                             (bond->params.mode == BOND_MODE_ALB)) {
1060                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1061                         }
1062                 } else {
1063                         if (USES_PRIMARY(bond->params.mode)) {
1064                                 printk(KERN_INFO DRV_NAME
1065                                        ": %s: making interface %s the new "
1066                                        "active one.\n",
1067                                        bond->dev->name, new_active->dev->name);
1068                         }
1069                 }
1070         }
1071
1072         if (USES_PRIMARY(bond->params.mode)) {
1073                 bond_mc_swap(bond, new_active, old_active);
1074         }
1075
1076         if ((bond->params.mode == BOND_MODE_TLB) ||
1077             (bond->params.mode == BOND_MODE_ALB)) {
1078                 bond_alb_handle_active_change(bond, new_active);
1079                 if (old_active)
1080                         bond_set_slave_inactive_flags(old_active);
1081                 if (new_active)
1082                         bond_set_slave_active_flags(new_active);
1083         } else {
1084                 bond->curr_active_slave = new_active;
1085         }
1086
1087         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1088                 if (old_active) {
1089                         bond_set_slave_inactive_flags(old_active);
1090                 }
1091
1092                 if (new_active) {
1093                         bond_set_slave_active_flags(new_active);
1094                 }
1095                 bond_send_gratuitous_arp(bond);
1096         }
1097 }
1098
1099 /**
1100  * bond_select_active_slave - select a new active slave, if needed
1101  * @bond: our bonding struct
1102  *
1103  * This functions shoud be called when one of the following occurs:
1104  * - The old curr_active_slave has been released or lost its link.
1105  * - The primary_slave has got its link back.
1106  * - A slave has got its link back and there's no old curr_active_slave.
1107  *
1108  * Warning: Caller must hold curr_slave_lock for writing.
1109  */
1110 void bond_select_active_slave(struct bonding *bond)
1111 {
1112         struct slave *best_slave;
1113         int rv;
1114
1115         best_slave = bond_find_best_slave(bond);
1116         if (best_slave != bond->curr_active_slave) {
1117                 bond_change_active_slave(bond, best_slave);
1118                 rv = bond_set_carrier(bond);
1119                 if (!rv)
1120                         return;
1121
1122                 if (netif_carrier_ok(bond->dev)) {
1123                         printk(KERN_INFO DRV_NAME
1124                                ": %s: first active interface up!\n",
1125                                bond->dev->name);
1126                 } else {
1127                         printk(KERN_INFO DRV_NAME ": %s: "
1128                                "now running without any active interface !\n",
1129                                bond->dev->name);
1130                 }
1131         }
1132 }
1133
1134 /*--------------------------- slave list handling ---------------------------*/
1135
1136 /*
1137  * This function attaches the slave to the end of list.
1138  *
1139  * bond->lock held for writing by caller.
1140  */
1141 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1142 {
1143         if (bond->first_slave == NULL) { /* attaching the first slave */
1144                 new_slave->next = new_slave;
1145                 new_slave->prev = new_slave;
1146                 bond->first_slave = new_slave;
1147         } else {
1148                 new_slave->next = bond->first_slave;
1149                 new_slave->prev = bond->first_slave->prev;
1150                 new_slave->next->prev = new_slave;
1151                 new_slave->prev->next = new_slave;
1152         }
1153
1154         bond->slave_cnt++;
1155 }
1156
1157 /*
1158  * This function detaches the slave from the list.
1159  * WARNING: no check is made to verify if the slave effectively
1160  * belongs to <bond>.
1161  * Nothing is freed on return, structures are just unchained.
1162  * If any slave pointer in bond was pointing to <slave>,
1163  * it should be changed by the calling function.
1164  *
1165  * bond->lock held for writing by caller.
1166  */
1167 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1168 {
1169         if (slave->next) {
1170                 slave->next->prev = slave->prev;
1171         }
1172
1173         if (slave->prev) {
1174                 slave->prev->next = slave->next;
1175         }
1176
1177         if (bond->first_slave == slave) { /* slave is the first slave */
1178                 if (bond->slave_cnt > 1) { /* there are more slave */
1179                         bond->first_slave = slave->next;
1180                 } else {
1181                         bond->first_slave = NULL; /* slave was the last one */
1182                 }
1183         }
1184
1185         slave->next = NULL;
1186         slave->prev = NULL;
1187         bond->slave_cnt--;
1188 }
1189
1190 /*---------------------------------- IOCTL ----------------------------------*/
1191
1192 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1193 {
1194         dprintk("bond_dev=%p\n", bond_dev);
1195         dprintk("slave_dev=%p\n", slave_dev);
1196         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1197         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1198         return 0;
1199 }
1200
1201 #define BOND_INTERSECT_FEATURES \
1202         (NETIF_F_SG|NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM|\
1203         NETIF_F_TSO|NETIF_F_UFO)
1204
1205 /*
1206  * Compute the common dev->feature set available to all slaves.  Some
1207  * feature bits are managed elsewhere, so preserve feature bits set on
1208  * master device that are not part of the examined set.
1209  */
1210 static int bond_compute_features(struct bonding *bond)
1211 {
1212         unsigned long features = BOND_INTERSECT_FEATURES;
1213         struct slave *slave;
1214         struct net_device *bond_dev = bond->dev;
1215         int i;
1216
1217         bond_for_each_slave(bond, slave, i)
1218                 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1219
1220         if ((features & NETIF_F_SG) &&
1221             !(features & (NETIF_F_IP_CSUM |
1222                           NETIF_F_NO_CSUM |
1223                           NETIF_F_HW_CSUM)))
1224                 features &= ~NETIF_F_SG;
1225
1226         /*
1227          * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1228          * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1229          * implies that all slaves also support scatter-gather
1230          * (NETIF_F_SG), which implies that features also includes
1231          * NETIF_F_SG. So no need to check whether we have an
1232          * illegal combination of NETIF_F_{TSO,UFO} and
1233          * !NETIF_F_SG
1234          */
1235
1236         features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1237         bond_dev->features = features;
1238
1239         return 0;
1240 }
1241
1242 /* enslave device <slave> to bond device <master> */
1243 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1244 {
1245         struct bonding *bond = bond_dev->priv;
1246         struct slave *new_slave = NULL;
1247         struct dev_mc_list *dmi;
1248         struct sockaddr addr;
1249         int link_reporting;
1250         int old_features = bond_dev->features;
1251         int res = 0;
1252
1253         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1254                 slave_dev->do_ioctl == NULL) {
1255                 printk(KERN_WARNING DRV_NAME
1256                        ": %s: Warning: no link monitoring support for %s\n",
1257                        bond_dev->name, slave_dev->name);
1258         }
1259
1260         /* bond must be initialized by bond_open() before enslaving */
1261         if (!(bond_dev->flags & IFF_UP)) {
1262                 dprintk("Error, master_dev is not up\n");
1263                 return -EPERM;
1264         }
1265
1266         /* already enslaved */
1267         if (slave_dev->flags & IFF_SLAVE) {
1268                 dprintk("Error, Device was already enslaved\n");
1269                 return -EBUSY;
1270         }
1271
1272         /* vlan challenged mutual exclusion */
1273         /* no need to lock since we're protected by rtnl_lock */
1274         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1275                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1276                 if (!list_empty(&bond->vlan_list)) {
1277                         printk(KERN_ERR DRV_NAME
1278                                ": %s: Error: cannot enslave VLAN "
1279                                "challenged slave %s on VLAN enabled "
1280                                "bond %s\n", bond_dev->name, slave_dev->name,
1281                                bond_dev->name);
1282                         return -EPERM;
1283                 } else {
1284                         printk(KERN_WARNING DRV_NAME
1285                                ": %s: Warning: enslaved VLAN challenged "
1286                                "slave %s. Adding VLANs will be blocked as "
1287                                "long as %s is part of bond %s\n",
1288                                bond_dev->name, slave_dev->name, slave_dev->name,
1289                                bond_dev->name);
1290                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1291                 }
1292         } else {
1293                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1294                 if (bond->slave_cnt == 0) {
1295                         /* First slave, and it is not VLAN challenged,
1296                          * so remove the block of adding VLANs over the bond.
1297                          */
1298                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1299                 }
1300         }
1301
1302         /*
1303          * Old ifenslave binaries are no longer supported.  These can
1304          * be identified with moderate accurary by the state of the slave:
1305          * the current ifenslave will set the interface down prior to
1306          * enslaving it; the old ifenslave will not.
1307          */
1308         if ((slave_dev->flags & IFF_UP)) {
1309                 printk(KERN_ERR DRV_NAME ": %s is up. "
1310                        "This may be due to an out of date ifenslave.\n",
1311                        slave_dev->name);
1312                 res = -EPERM;
1313                 goto err_undo_flags;
1314         }
1315
1316         if (slave_dev->set_mac_address == NULL) {
1317                 printk(KERN_ERR DRV_NAME
1318                         ": %s: Error: The slave device you specified does "
1319                         "not support setting the MAC address. "
1320                         "Your kernel likely does not support slave "
1321                         "devices.\n", bond_dev->name);
1322                 res = -EOPNOTSUPP;
1323                 goto err_undo_flags;
1324         }
1325
1326         new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1327         if (!new_slave) {
1328                 res = -ENOMEM;
1329                 goto err_undo_flags;
1330         }
1331
1332         memset(new_slave, 0, sizeof(struct slave));
1333
1334         /* save slave's original flags before calling
1335          * netdev_set_master and dev_open
1336          */
1337         new_slave->original_flags = slave_dev->flags;
1338
1339         /*
1340          * Save slave's original ("permanent") mac address for modes
1341          * that need it, and for restoring it upon release, and then
1342          * set it to the master's address
1343          */
1344         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1345
1346         /*
1347          * Set slave to master's mac address.  The application already
1348          * set the master's mac address to that of the first slave
1349          */
1350         memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1351         addr.sa_family = slave_dev->type;
1352         res = dev_set_mac_address(slave_dev, &addr);
1353         if (res) {
1354                 dprintk("Error %d calling set_mac_address\n", res);
1355                 goto err_free;
1356         }
1357
1358         /* open the slave since the application closed it */
1359         res = dev_open(slave_dev);
1360         if (res) {
1361                 dprintk("Openning slave %s failed\n", slave_dev->name);
1362                 goto err_restore_mac;
1363         }
1364
1365         res = netdev_set_master(slave_dev, bond_dev);
1366         if (res) {
1367                 dprintk("Error %d calling netdev_set_master\n", res);
1368                 goto err_close;
1369         }
1370
1371         new_slave->dev = slave_dev;
1372
1373         if ((bond->params.mode == BOND_MODE_TLB) ||
1374             (bond->params.mode == BOND_MODE_ALB)) {
1375                 /* bond_alb_init_slave() must be called before all other stages since
1376                  * it might fail and we do not want to have to undo everything
1377                  */
1378                 res = bond_alb_init_slave(bond, new_slave);
1379                 if (res) {
1380                         goto err_unset_master;
1381                 }
1382         }
1383
1384         /* If the mode USES_PRIMARY, then the new slave gets the
1385          * master's promisc (and mc) settings only if it becomes the
1386          * curr_active_slave, and that is taken care of later when calling
1387          * bond_change_active()
1388          */
1389         if (!USES_PRIMARY(bond->params.mode)) {
1390                 /* set promiscuity level to new slave */
1391                 if (bond_dev->flags & IFF_PROMISC) {
1392                         dev_set_promiscuity(slave_dev, 1);
1393                 }
1394
1395                 /* set allmulti level to new slave */
1396                 if (bond_dev->flags & IFF_ALLMULTI) {
1397                         dev_set_allmulti(slave_dev, 1);
1398                 }
1399
1400                 /* upload master's mc_list to new slave */
1401                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1402                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1403                 }
1404         }
1405
1406         if (bond->params.mode == BOND_MODE_8023AD) {
1407                 /* add lacpdu mc addr to mc list */
1408                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1409
1410                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1411         }
1412
1413         bond_add_vlans_on_slave(bond, slave_dev);
1414
1415         write_lock_bh(&bond->lock);
1416
1417         bond_attach_slave(bond, new_slave);
1418
1419         new_slave->delay = 0;
1420         new_slave->link_failure_count = 0;
1421
1422         bond_compute_features(bond);
1423
1424         if (bond->params.miimon && !bond->params.use_carrier) {
1425                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1426
1427                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1428                         /*
1429                          * miimon is set but a bonded network driver
1430                          * does not support ETHTOOL/MII and
1431                          * arp_interval is not set.  Note: if
1432                          * use_carrier is enabled, we will never go
1433                          * here (because netif_carrier is always
1434                          * supported); thus, we don't need to change
1435                          * the messages for netif_carrier.
1436                          */
1437                         printk(KERN_WARNING DRV_NAME
1438                                ": %s: Warning: MII and ETHTOOL support not "
1439                                "available for interface %s, and "
1440                                "arp_interval/arp_ip_target module parameters "
1441                                "not specified, thus bonding will not detect "
1442                                "link failures! see bonding.txt for details.\n",
1443                                bond_dev->name, slave_dev->name);
1444                 } else if (link_reporting == -1) {
1445                         /* unable get link status using mii/ethtool */
1446                         printk(KERN_WARNING DRV_NAME
1447                                ": %s: Warning: can't get link status from "
1448                                "interface %s; the network driver associated "
1449                                "with this interface does not support MII or "
1450                                "ETHTOOL link status reporting, thus miimon "
1451                                "has no effect on this interface.\n",
1452                                bond_dev->name, slave_dev->name);
1453                 }
1454         }
1455
1456         /* check for initial state */
1457         if (!bond->params.miimon ||
1458             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1459                 if (bond->params.updelay) {
1460                         dprintk("Initial state of slave_dev is "
1461                                 "BOND_LINK_BACK\n");
1462                         new_slave->link  = BOND_LINK_BACK;
1463                         new_slave->delay = bond->params.updelay;
1464                 } else {
1465                         dprintk("Initial state of slave_dev is "
1466                                 "BOND_LINK_UP\n");
1467                         new_slave->link  = BOND_LINK_UP;
1468                 }
1469                 new_slave->jiffies = jiffies;
1470         } else {
1471                 dprintk("Initial state of slave_dev is "
1472                         "BOND_LINK_DOWN\n");
1473                 new_slave->link  = BOND_LINK_DOWN;
1474         }
1475
1476         if (bond_update_speed_duplex(new_slave) &&
1477             (new_slave->link != BOND_LINK_DOWN)) {
1478                 printk(KERN_WARNING DRV_NAME
1479                        ": %s: Warning: failed to get speed and duplex from %s, "
1480                        "assumed to be 100Mb/sec and Full.\n",
1481                        bond_dev->name, new_slave->dev->name);
1482
1483                 if (bond->params.mode == BOND_MODE_8023AD) {
1484                         printk(KERN_WARNING DRV_NAME
1485                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1486                                "support in base driver for proper aggregator "
1487                                "selection.\n", bond_dev->name);
1488                 }
1489         }
1490
1491         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1492                 /* if there is a primary slave, remember it */
1493                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1494                         bond->primary_slave = new_slave;
1495                 }
1496         }
1497
1498         switch (bond->params.mode) {
1499         case BOND_MODE_ACTIVEBACKUP:
1500                 /* if we're in active-backup mode, we need one and
1501                  * only one active interface. The backup interfaces
1502                  * will have their SLAVE_INACTIVE flag set because we
1503                  * need them to be drop all packets. Thus, since we
1504                  * guarantee that curr_active_slave always point to
1505                  * the last usable interface, we just have to verify
1506                  * this interface's flag.
1507                  */
1508                 if (((!bond->curr_active_slave) ||
1509                      (bond->curr_active_slave->dev->priv_flags & IFF_SLAVE_INACTIVE)) &&
1510                     (new_slave->link != BOND_LINK_DOWN)) {
1511                         /* first slave or no active slave yet, and this link
1512                            is OK, so make this interface the active one */
1513                         bond_change_active_slave(bond, new_slave);
1514                         printk(KERN_INFO DRV_NAME
1515                                ": %s: first active interface up!\n",
1516                                bond->dev->name);
1517                         netif_carrier_on(bond->dev);
1518
1519                 } else {
1520                         dprintk("This is just a backup slave\n");
1521                         bond_set_slave_inactive_flags(new_slave);
1522                 }
1523                 break;
1524         case BOND_MODE_8023AD:
1525                 /* in 802.3ad mode, the internal mechanism
1526                  * will activate the slaves in the selected
1527                  * aggregator
1528                  */
1529                 bond_set_slave_inactive_flags(new_slave);
1530                 /* if this is the first slave */
1531                 if (bond->slave_cnt == 1) {
1532                         SLAVE_AD_INFO(new_slave).id = 1;
1533                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1534                          * can be called only after the mac address of the bond is set
1535                          */
1536                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1537                                             bond->params.lacp_fast);
1538                 } else {
1539                         SLAVE_AD_INFO(new_slave).id =
1540                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1541                 }
1542
1543                 bond_3ad_bind_slave(new_slave);
1544                 break;
1545         case BOND_MODE_TLB:
1546         case BOND_MODE_ALB:
1547                 new_slave->state = BOND_STATE_ACTIVE;
1548                 if ((!bond->curr_active_slave) &&
1549                     (new_slave->link != BOND_LINK_DOWN)) {
1550                         /* first slave or no active slave yet, and this link
1551                          * is OK, so make this interface the active one
1552                          */
1553                         bond_change_active_slave(bond, new_slave);
1554                 } else {
1555                         bond_set_slave_inactive_flags(new_slave);
1556                 }
1557                 break;
1558         default:
1559                 dprintk("This slave is always active in trunk mode\n");
1560
1561                 /* always active in trunk mode */
1562                 new_slave->state = BOND_STATE_ACTIVE;
1563
1564                 /* In trunking mode there is little meaning to curr_active_slave
1565                  * anyway (it holds no special properties of the bond device),
1566                  * so we can change it without calling change_active_interface()
1567                  */
1568                 if (!bond->curr_active_slave) {
1569                         bond->curr_active_slave = new_slave;
1570                 }
1571                 break;
1572         } /* switch(bond_mode) */
1573
1574         bond_set_carrier(bond);
1575
1576         write_unlock_bh(&bond->lock);
1577
1578         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1579         if (res)
1580                 goto err_unset_master;
1581
1582         printk(KERN_INFO DRV_NAME
1583                ": %s: enslaving %s as a%s interface with a%s link.\n",
1584                bond_dev->name, slave_dev->name,
1585                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1586                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1587
1588         /* enslave is successful */
1589         return 0;
1590
1591 /* Undo stages on error */
1592 err_unset_master:
1593         netdev_set_master(slave_dev, NULL);
1594
1595 err_close:
1596         dev_close(slave_dev);
1597
1598 err_restore_mac:
1599         memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1600         addr.sa_family = slave_dev->type;
1601         dev_set_mac_address(slave_dev, &addr);
1602
1603 err_free:
1604         kfree(new_slave);
1605
1606 err_undo_flags:
1607         bond_dev->features = old_features;
1608
1609         return res;
1610 }
1611
1612 /*
1613  * Try to release the slave device <slave> from the bond device <master>
1614  * It is legal to access curr_active_slave without a lock because all the function
1615  * is write-locked.
1616  *
1617  * The rules for slave state should be:
1618  *   for Active/Backup:
1619  *     Active stays on all backups go down
1620  *   for Bonded connections:
1621  *     The first up interface should be left on and all others downed.
1622  */
1623 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1624 {
1625         struct bonding *bond = bond_dev->priv;
1626         struct slave *slave, *oldcurrent;
1627         struct sockaddr addr;
1628         int mac_addr_differ;
1629
1630         /* slave is not a slave or master is not master of this slave */
1631         if (!(slave_dev->flags & IFF_SLAVE) ||
1632             (slave_dev->master != bond_dev)) {
1633                 printk(KERN_ERR DRV_NAME
1634                        ": %s: Error: cannot release %s.\n",
1635                        bond_dev->name, slave_dev->name);
1636                 return -EINVAL;
1637         }
1638
1639         write_lock_bh(&bond->lock);
1640
1641         slave = bond_get_slave_by_dev(bond, slave_dev);
1642         if (!slave) {
1643                 /* not a slave of this bond */
1644                 printk(KERN_INFO DRV_NAME
1645                        ": %s: %s not enslaved\n",
1646                        bond_dev->name, slave_dev->name);
1647                 write_unlock_bh(&bond->lock);
1648                 return -EINVAL;
1649         }
1650
1651         mac_addr_differ = memcmp(bond_dev->dev_addr,
1652                                  slave->perm_hwaddr,
1653                                  ETH_ALEN);
1654         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1655                 printk(KERN_WARNING DRV_NAME
1656                        ": %s: Warning: the permanent HWaddr of %s "
1657                        "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1658                        "still in use by %s. Set the HWaddr of "
1659                        "%s to a different address to avoid "
1660                        "conflicts.\n",
1661                        bond_dev->name,
1662                        slave_dev->name,
1663                        slave->perm_hwaddr[0],
1664                        slave->perm_hwaddr[1],
1665                        slave->perm_hwaddr[2],
1666                        slave->perm_hwaddr[3],
1667                        slave->perm_hwaddr[4],
1668                        slave->perm_hwaddr[5],
1669                        bond_dev->name,
1670                        slave_dev->name);
1671         }
1672
1673         /* Inform AD package of unbinding of slave. */
1674         if (bond->params.mode == BOND_MODE_8023AD) {
1675                 /* must be called before the slave is
1676                  * detached from the list
1677                  */
1678                 bond_3ad_unbind_slave(slave);
1679         }
1680
1681         printk(KERN_INFO DRV_NAME
1682                ": %s: releasing %s interface %s\n",
1683                bond_dev->name,
1684                (slave->state == BOND_STATE_ACTIVE)
1685                ? "active" : "backup",
1686                slave_dev->name);
1687
1688         oldcurrent = bond->curr_active_slave;
1689
1690         bond->current_arp_slave = NULL;
1691
1692         /* release the slave from its bond */
1693         bond_detach_slave(bond, slave);
1694
1695         bond_compute_features(bond);
1696
1697         if (bond->primary_slave == slave) {
1698                 bond->primary_slave = NULL;
1699         }
1700
1701         if (oldcurrent == slave) {
1702                 bond_change_active_slave(bond, NULL);
1703         }
1704
1705         if ((bond->params.mode == BOND_MODE_TLB) ||
1706             (bond->params.mode == BOND_MODE_ALB)) {
1707                 /* Must be called only after the slave has been
1708                  * detached from the list and the curr_active_slave
1709                  * has been cleared (if our_slave == old_current),
1710                  * but before a new active slave is selected.
1711                  */
1712                 bond_alb_deinit_slave(bond, slave);
1713         }
1714
1715         if (oldcurrent == slave)
1716                 bond_select_active_slave(bond);
1717
1718         if (bond->slave_cnt == 0) {
1719                 bond_set_carrier(bond);
1720
1721                 /* if the last slave was removed, zero the mac address
1722                  * of the master so it will be set by the application
1723                  * to the mac address of the first slave
1724                  */
1725                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1726
1727                 if (list_empty(&bond->vlan_list)) {
1728                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1729                 } else {
1730                         printk(KERN_WARNING DRV_NAME
1731                                ": %s: Warning: clearing HW address of %s while it "
1732                                "still has VLANs.\n",
1733                                bond_dev->name, bond_dev->name);
1734                         printk(KERN_WARNING DRV_NAME
1735                                ": %s: When re-adding slaves, make sure the bond's "
1736                                "HW address matches its VLANs'.\n",
1737                                bond_dev->name);
1738                 }
1739         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1740                    !bond_has_challenged_slaves(bond)) {
1741                 printk(KERN_INFO DRV_NAME
1742                        ": %s: last VLAN challenged slave %s "
1743                        "left bond %s. VLAN blocking is removed\n",
1744                        bond_dev->name, slave_dev->name, bond_dev->name);
1745                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1746         }
1747
1748         write_unlock_bh(&bond->lock);
1749
1750         /* must do this from outside any spinlocks */
1751         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1752
1753         bond_del_vlans_from_slave(bond, slave_dev);
1754
1755         /* If the mode USES_PRIMARY, then we should only remove its
1756          * promisc and mc settings if it was the curr_active_slave, but that was
1757          * already taken care of above when we detached the slave
1758          */
1759         if (!USES_PRIMARY(bond->params.mode)) {
1760                 /* unset promiscuity level from slave */
1761                 if (bond_dev->flags & IFF_PROMISC) {
1762                         dev_set_promiscuity(slave_dev, -1);
1763                 }
1764
1765                 /* unset allmulti level from slave */
1766                 if (bond_dev->flags & IFF_ALLMULTI) {
1767                         dev_set_allmulti(slave_dev, -1);
1768                 }
1769
1770                 /* flush master's mc_list from slave */
1771                 bond_mc_list_flush(bond_dev, slave_dev);
1772         }
1773
1774         netdev_set_master(slave_dev, NULL);
1775
1776         /* close slave before restoring its mac address */
1777         dev_close(slave_dev);
1778
1779         /* restore original ("permanent") mac address */
1780         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1781         addr.sa_family = slave_dev->type;
1782         dev_set_mac_address(slave_dev, &addr);
1783
1784         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1785                                    IFF_SLAVE_INACTIVE);
1786
1787         kfree(slave);
1788
1789         return 0;  /* deletion OK */
1790 }
1791
1792 /*
1793  * This function releases all slaves.
1794  */
1795 static int bond_release_all(struct net_device *bond_dev)
1796 {
1797         struct bonding *bond = bond_dev->priv;
1798         struct slave *slave;
1799         struct net_device *slave_dev;
1800         struct sockaddr addr;
1801
1802         write_lock_bh(&bond->lock);
1803
1804         netif_carrier_off(bond_dev);
1805
1806         if (bond->slave_cnt == 0) {
1807                 goto out;
1808         }
1809
1810         bond->current_arp_slave = NULL;
1811         bond->primary_slave = NULL;
1812         bond_change_active_slave(bond, NULL);
1813
1814         while ((slave = bond->first_slave) != NULL) {
1815                 /* Inform AD package of unbinding of slave
1816                  * before slave is detached from the list.
1817                  */
1818                 if (bond->params.mode == BOND_MODE_8023AD) {
1819                         bond_3ad_unbind_slave(slave);
1820                 }
1821
1822                 slave_dev = slave->dev;
1823                 bond_detach_slave(bond, slave);
1824
1825                 if ((bond->params.mode == BOND_MODE_TLB) ||
1826                     (bond->params.mode == BOND_MODE_ALB)) {
1827                         /* must be called only after the slave
1828                          * has been detached from the list
1829                          */
1830                         bond_alb_deinit_slave(bond, slave);
1831                 }
1832
1833                 bond_compute_features(bond);
1834
1835                 /* now that the slave is detached, unlock and perform
1836                  * all the undo steps that should not be called from
1837                  * within a lock.
1838                  */
1839                 write_unlock_bh(&bond->lock);
1840
1841                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1842                 bond_del_vlans_from_slave(bond, slave_dev);
1843
1844                 /* If the mode USES_PRIMARY, then we should only remove its
1845                  * promisc and mc settings if it was the curr_active_slave, but that was
1846                  * already taken care of above when we detached the slave
1847                  */
1848                 if (!USES_PRIMARY(bond->params.mode)) {
1849                         /* unset promiscuity level from slave */
1850                         if (bond_dev->flags & IFF_PROMISC) {
1851                                 dev_set_promiscuity(slave_dev, -1);
1852                         }
1853
1854                         /* unset allmulti level from slave */
1855                         if (bond_dev->flags & IFF_ALLMULTI) {
1856                                 dev_set_allmulti(slave_dev, -1);
1857                         }
1858
1859                         /* flush master's mc_list from slave */
1860                         bond_mc_list_flush(bond_dev, slave_dev);
1861                 }
1862
1863                 netdev_set_master(slave_dev, NULL);
1864
1865                 /* close slave before restoring its mac address */
1866                 dev_close(slave_dev);
1867
1868                 /* restore original ("permanent") mac address*/
1869                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1870                 addr.sa_family = slave_dev->type;
1871                 dev_set_mac_address(slave_dev, &addr);
1872
1873                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1874                                            IFF_SLAVE_INACTIVE);
1875
1876                 kfree(slave);
1877
1878                 /* re-acquire the lock before getting the next slave */
1879                 write_lock_bh(&bond->lock);
1880         }
1881
1882         /* zero the mac address of the master so it will be
1883          * set by the application to the mac address of the
1884          * first slave
1885          */
1886         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1887
1888         if (list_empty(&bond->vlan_list)) {
1889                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1890         } else {
1891                 printk(KERN_WARNING DRV_NAME
1892                        ": %s: Warning: clearing HW address of %s while it "
1893                        "still has VLANs.\n",
1894                        bond_dev->name, bond_dev->name);
1895                 printk(KERN_WARNING DRV_NAME
1896                        ": %s: When re-adding slaves, make sure the bond's "
1897                        "HW address matches its VLANs'.\n",
1898                        bond_dev->name);
1899         }
1900
1901         printk(KERN_INFO DRV_NAME
1902                ": %s: released all slaves\n",
1903                bond_dev->name);
1904
1905 out:
1906         write_unlock_bh(&bond->lock);
1907
1908         return 0;
1909 }
1910
1911 /*
1912  * This function changes the active slave to slave <slave_dev>.
1913  * It returns -EINVAL in the following cases.
1914  *  - <slave_dev> is not found in the list.
1915  *  - There is not active slave now.
1916  *  - <slave_dev> is already active.
1917  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1918  *  - <slave_dev> is not running.
1919  * In these cases, this fuction does nothing.
1920  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1921  */
1922 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1923 {
1924         struct bonding *bond = bond_dev->priv;
1925         struct slave *old_active = NULL;
1926         struct slave *new_active = NULL;
1927         int res = 0;
1928
1929         if (!USES_PRIMARY(bond->params.mode)) {
1930                 return -EINVAL;
1931         }
1932
1933         /* Verify that master_dev is indeed the master of slave_dev */
1934         if (!(slave_dev->flags & IFF_SLAVE) ||
1935             (slave_dev->master != bond_dev)) {
1936                 return -EINVAL;
1937         }
1938
1939         write_lock_bh(&bond->lock);
1940
1941         old_active = bond->curr_active_slave;
1942         new_active = bond_get_slave_by_dev(bond, slave_dev);
1943
1944         /*
1945          * Changing to the current active: do nothing; return success.
1946          */
1947         if (new_active && (new_active == old_active)) {
1948                 write_unlock_bh(&bond->lock);
1949                 return 0;
1950         }
1951
1952         if ((new_active) &&
1953             (old_active) &&
1954             (new_active->link == BOND_LINK_UP) &&
1955             IS_UP(new_active->dev)) {
1956                 bond_change_active_slave(bond, new_active);
1957         } else {
1958                 res = -EINVAL;
1959         }
1960
1961         write_unlock_bh(&bond->lock);
1962
1963         return res;
1964 }
1965
1966 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1967 {
1968         struct bonding *bond = bond_dev->priv;
1969
1970         info->bond_mode = bond->params.mode;
1971         info->miimon = bond->params.miimon;
1972
1973         read_lock_bh(&bond->lock);
1974         info->num_slaves = bond->slave_cnt;
1975         read_unlock_bh(&bond->lock);
1976
1977         return 0;
1978 }
1979
1980 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1981 {
1982         struct bonding *bond = bond_dev->priv;
1983         struct slave *slave;
1984         int i, found = 0;
1985
1986         if (info->slave_id < 0) {
1987                 return -ENODEV;
1988         }
1989
1990         read_lock_bh(&bond->lock);
1991
1992         bond_for_each_slave(bond, slave, i) {
1993                 if (i == (int)info->slave_id) {
1994                         found = 1;
1995                         break;
1996                 }
1997         }
1998
1999         read_unlock_bh(&bond->lock);
2000
2001         if (found) {
2002                 strcpy(info->slave_name, slave->dev->name);
2003                 info->link = slave->link;
2004                 info->state = slave->state;
2005                 info->link_failure_count = slave->link_failure_count;
2006         } else {
2007                 return -ENODEV;
2008         }
2009
2010         return 0;
2011 }
2012
2013 /*-------------------------------- Monitoring -------------------------------*/
2014
2015 /* this function is called regularly to monitor each slave's link. */
2016 void bond_mii_monitor(struct net_device *bond_dev)
2017 {
2018         struct bonding *bond = bond_dev->priv;
2019         struct slave *slave, *oldcurrent;
2020         int do_failover = 0;
2021         int delta_in_ticks;
2022         int i;
2023
2024         read_lock(&bond->lock);
2025
2026         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2027
2028         if (bond->kill_timers) {
2029                 goto out;
2030         }
2031
2032         if (bond->slave_cnt == 0) {
2033                 goto re_arm;
2034         }
2035
2036         /* we will try to read the link status of each of our slaves, and
2037          * set their IFF_RUNNING flag appropriately. For each slave not
2038          * supporting MII status, we won't do anything so that a user-space
2039          * program could monitor the link itself if needed.
2040          */
2041
2042         read_lock(&bond->curr_slave_lock);
2043         oldcurrent = bond->curr_active_slave;
2044         read_unlock(&bond->curr_slave_lock);
2045
2046         bond_for_each_slave(bond, slave, i) {
2047                 struct net_device *slave_dev = slave->dev;
2048                 int link_state;
2049                 u16 old_speed = slave->speed;
2050                 u8 old_duplex = slave->duplex;
2051
2052                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2053
2054                 switch (slave->link) {
2055                 case BOND_LINK_UP:      /* the link was up */
2056                         if (link_state == BMSR_LSTATUS) {
2057                                 /* link stays up, nothing more to do */
2058                                 break;
2059                         } else { /* link going down */
2060                                 slave->link  = BOND_LINK_FAIL;
2061                                 slave->delay = bond->params.downdelay;
2062
2063                                 if (slave->link_failure_count < UINT_MAX) {
2064                                         slave->link_failure_count++;
2065                                 }
2066
2067                                 if (bond->params.downdelay) {
2068                                         printk(KERN_INFO DRV_NAME
2069                                                ": %s: link status down for %s "
2070                                                "interface %s, disabling it in "
2071                                                "%d ms.\n",
2072                                                bond_dev->name,
2073                                                IS_UP(slave_dev)
2074                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2075                                                   ? ((slave == oldcurrent)
2076                                                      ? "active " : "backup ")
2077                                                   : "")
2078                                                : "idle ",
2079                                                slave_dev->name,
2080                                                bond->params.downdelay * bond->params.miimon);
2081                                 }
2082                         }
2083                         /* no break ! fall through the BOND_LINK_FAIL test to
2084                            ensure proper action to be taken
2085                         */
2086                 case BOND_LINK_FAIL:    /* the link has just gone down */
2087                         if (link_state != BMSR_LSTATUS) {
2088                                 /* link stays down */
2089                                 if (slave->delay <= 0) {
2090                                         /* link down for too long time */
2091                                         slave->link = BOND_LINK_DOWN;
2092
2093                                         /* in active/backup mode, we must
2094                                          * completely disable this interface
2095                                          */
2096                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2097                                             (bond->params.mode == BOND_MODE_8023AD)) {
2098                                                 bond_set_slave_inactive_flags(slave);
2099                                         }
2100
2101                                         printk(KERN_INFO DRV_NAME
2102                                                ": %s: link status definitely "
2103                                                "down for interface %s, "
2104                                                "disabling it\n",
2105                                                bond_dev->name,
2106                                                slave_dev->name);
2107
2108                                         /* notify ad that the link status has changed */
2109                                         if (bond->params.mode == BOND_MODE_8023AD) {
2110                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2111                                         }
2112
2113                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2114                                             (bond->params.mode == BOND_MODE_ALB)) {
2115                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2116                                         }
2117
2118                                         if (slave == oldcurrent) {
2119                                                 do_failover = 1;
2120                                         }
2121                                 } else {
2122                                         slave->delay--;
2123                                 }
2124                         } else {
2125                                 /* link up again */
2126                                 slave->link  = BOND_LINK_UP;
2127                                 slave->jiffies = jiffies;
2128                                 printk(KERN_INFO DRV_NAME
2129                                        ": %s: link status up again after %d "
2130                                        "ms for interface %s.\n",
2131                                        bond_dev->name,
2132                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2133                                        slave_dev->name);
2134                         }
2135                         break;
2136                 case BOND_LINK_DOWN:    /* the link was down */
2137                         if (link_state != BMSR_LSTATUS) {
2138                                 /* the link stays down, nothing more to do */
2139                                 break;
2140                         } else {        /* link going up */
2141                                 slave->link  = BOND_LINK_BACK;
2142                                 slave->delay = bond->params.updelay;
2143
2144                                 if (bond->params.updelay) {
2145                                         /* if updelay == 0, no need to
2146                                            advertise about a 0 ms delay */
2147                                         printk(KERN_INFO DRV_NAME
2148                                                ": %s: link status up for "
2149                                                "interface %s, enabling it "
2150                                                "in %d ms.\n",
2151                                                bond_dev->name,
2152                                                slave_dev->name,
2153                                                bond->params.updelay * bond->params.miimon);
2154                                 }
2155                         }
2156                         /* no break ! fall through the BOND_LINK_BACK state in
2157                            case there's something to do.
2158                         */
2159                 case BOND_LINK_BACK:    /* the link has just come back */
2160                         if (link_state != BMSR_LSTATUS) {
2161                                 /* link down again */
2162                                 slave->link  = BOND_LINK_DOWN;
2163
2164                                 printk(KERN_INFO DRV_NAME
2165                                        ": %s: link status down again after %d "
2166                                        "ms for interface %s.\n",
2167                                        bond_dev->name,
2168                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2169                                        slave_dev->name);
2170                         } else {
2171                                 /* link stays up */
2172                                 if (slave->delay == 0) {
2173                                         /* now the link has been up for long time enough */
2174                                         slave->link = BOND_LINK_UP;
2175                                         slave->jiffies = jiffies;
2176
2177                                         if (bond->params.mode == BOND_MODE_8023AD) {
2178                                                 /* prevent it from being the active one */
2179                                                 slave->state = BOND_STATE_BACKUP;
2180                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2181                                                 /* make it immediately active */
2182                                                 slave->state = BOND_STATE_ACTIVE;
2183                                         } else if (slave != bond->primary_slave) {
2184                                                 /* prevent it from being the active one */
2185                                                 slave->state = BOND_STATE_BACKUP;
2186                                         }
2187
2188                                         printk(KERN_INFO DRV_NAME
2189                                                ": %s: link status definitely "
2190                                                "up for interface %s.\n",
2191                                                bond_dev->name,
2192                                                slave_dev->name);
2193
2194                                         /* notify ad that the link status has changed */
2195                                         if (bond->params.mode == BOND_MODE_8023AD) {
2196                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2197                                         }
2198
2199                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2200                                             (bond->params.mode == BOND_MODE_ALB)) {
2201                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2202                                         }
2203
2204                                         if ((!oldcurrent) ||
2205                                             (slave == bond->primary_slave)) {
2206                                                 do_failover = 1;
2207                                         }
2208                                 } else {
2209                                         slave->delay--;
2210                                 }
2211                         }
2212                         break;
2213                 default:
2214                         /* Should not happen */
2215                         printk(KERN_ERR DRV_NAME
2216                                ": %s: Error: %s Illegal value (link=%d)\n",
2217                                bond_dev->name,
2218                                slave->dev->name,
2219                                slave->link);
2220                         goto out;
2221                 } /* end of switch (slave->link) */
2222
2223                 bond_update_speed_duplex(slave);
2224
2225                 if (bond->params.mode == BOND_MODE_8023AD) {
2226                         if (old_speed != slave->speed) {
2227                                 bond_3ad_adapter_speed_changed(slave);
2228                         }
2229
2230                         if (old_duplex != slave->duplex) {
2231                                 bond_3ad_adapter_duplex_changed(slave);
2232                         }
2233                 }
2234
2235         } /* end of for */
2236
2237         if (do_failover) {
2238                 write_lock(&bond->curr_slave_lock);
2239
2240                 bond_select_active_slave(bond);
2241
2242                 write_unlock(&bond->curr_slave_lock);
2243         } else
2244                 bond_set_carrier(bond);
2245
2246 re_arm:
2247         if (bond->params.miimon) {
2248                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2249         }
2250 out:
2251         read_unlock(&bond->lock);
2252 }
2253
2254
2255 static u32 bond_glean_dev_ip(struct net_device *dev)
2256 {
2257         struct in_device *idev;
2258         struct in_ifaddr *ifa;
2259         u32 addr = 0;
2260
2261         if (!dev)
2262                 return 0;
2263
2264         rcu_read_lock();
2265         idev = __in_dev_get_rcu(dev);
2266         if (!idev)
2267                 goto out;
2268
2269         ifa = idev->ifa_list;
2270         if (!ifa)
2271                 goto out;
2272
2273         addr = ifa->ifa_local;
2274 out:
2275         rcu_read_unlock();
2276         return addr;
2277 }
2278
2279 static int bond_has_ip(struct bonding *bond)
2280 {
2281         struct vlan_entry *vlan, *vlan_next;
2282
2283         if (bond->master_ip)
2284                 return 1;
2285
2286         if (list_empty(&bond->vlan_list))
2287                 return 0;
2288
2289         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2290                                  vlan_list) {
2291                 if (vlan->vlan_ip)
2292                         return 1;
2293         }
2294
2295         return 0;
2296 }
2297
2298 /*
2299  * We go to the (large) trouble of VLAN tagging ARP frames because
2300  * switches in VLAN mode (especially if ports are configured as
2301  * "native" to a VLAN) might not pass non-tagged frames.
2302  */
2303 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2304 {
2305         struct sk_buff *skb;
2306
2307         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2308                slave_dev->name, dest_ip, src_ip, vlan_id);
2309
2310         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2311                          NULL, slave_dev->dev_addr, NULL);
2312
2313         if (!skb) {
2314                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2315                 return;
2316         }
2317         if (vlan_id) {
2318                 skb = vlan_put_tag(skb, vlan_id);
2319                 if (!skb) {
2320                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2321                         return;
2322                 }
2323         }
2324         arp_xmit(skb);
2325 }
2326
2327
2328 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2329 {
2330         int i, vlan_id, rv;
2331         u32 *targets = bond->params.arp_targets;
2332         struct vlan_entry *vlan, *vlan_next;
2333         struct net_device *vlan_dev;
2334         struct flowi fl;
2335         struct rtable *rt;
2336
2337         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2338                 if (!targets[i])
2339                         continue;
2340                 dprintk("basa: target %x\n", targets[i]);
2341                 if (list_empty(&bond->vlan_list)) {
2342                         dprintk("basa: empty vlan: arp_send\n");
2343                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2344                                       bond->master_ip, 0);
2345                         continue;
2346                 }
2347
2348                 /*
2349                  * If VLANs are configured, we do a route lookup to
2350                  * determine which VLAN interface would be used, so we
2351                  * can tag the ARP with the proper VLAN tag.
2352                  */
2353                 memset(&fl, 0, sizeof(fl));
2354                 fl.fl4_dst = targets[i];
2355                 fl.fl4_tos = RTO_ONLINK;
2356
2357                 rv = ip_route_output_key(&rt, &fl);
2358                 if (rv) {
2359                         if (net_ratelimit()) {
2360                                 printk(KERN_WARNING DRV_NAME
2361                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2362                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2363                         }
2364                         continue;
2365                 }
2366
2367                 /*
2368                  * This target is not on a VLAN
2369                  */
2370                 if (rt->u.dst.dev == bond->dev) {
2371                         ip_rt_put(rt);
2372                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2373                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2374                                       bond->master_ip, 0);
2375                         continue;
2376                 }
2377
2378                 vlan_id = 0;
2379                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2380                                          vlan_list) {
2381                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2382                         if (vlan_dev == rt->u.dst.dev) {
2383                                 vlan_id = vlan->vlan_id;
2384                                 dprintk("basa: vlan match on %s %d\n",
2385                                        vlan_dev->name, vlan_id);
2386                                 break;
2387                         }
2388                 }
2389
2390                 if (vlan_id) {
2391                         ip_rt_put(rt);
2392                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2393                                       vlan->vlan_ip, vlan_id);
2394                         continue;
2395                 }
2396
2397                 if (net_ratelimit()) {
2398                         printk(KERN_WARNING DRV_NAME
2399                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2400                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2401                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2402                 }
2403                 ip_rt_put(rt);
2404         }
2405 }
2406
2407 /*
2408  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2409  * for each VLAN above us.
2410  */
2411 static void bond_send_gratuitous_arp(struct bonding *bond)
2412 {
2413         struct slave *slave = bond->curr_active_slave;
2414         struct vlan_entry *vlan;
2415         struct net_device *vlan_dev;
2416
2417         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2418                                 slave ? slave->dev->name : "NULL");
2419         if (!slave)
2420                 return;
2421
2422         if (bond->master_ip) {
2423                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2424                                   bond->master_ip, 0);
2425         }
2426
2427         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2428                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2429                 if (vlan->vlan_ip) {
2430                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2431                                       vlan->vlan_ip, vlan->vlan_id);
2432                 }
2433         }
2434 }
2435
2436 /*
2437  * this function is called regularly to monitor each slave's link
2438  * ensuring that traffic is being sent and received when arp monitoring
2439  * is used in load-balancing mode. if the adapter has been dormant, then an
2440  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2441  * arp monitoring in active backup mode.
2442  */
2443 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2444 {
2445         struct bonding *bond = bond_dev->priv;
2446         struct slave *slave, *oldcurrent;
2447         int do_failover = 0;
2448         int delta_in_ticks;
2449         int i;
2450
2451         read_lock(&bond->lock);
2452
2453         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2454
2455         if (bond->kill_timers) {
2456                 goto out;
2457         }
2458
2459         if (bond->slave_cnt == 0) {
2460                 goto re_arm;
2461         }
2462
2463         read_lock(&bond->curr_slave_lock);
2464         oldcurrent = bond->curr_active_slave;
2465         read_unlock(&bond->curr_slave_lock);
2466
2467         /* see if any of the previous devices are up now (i.e. they have
2468          * xmt and rcv traffic). the curr_active_slave does not come into
2469          * the picture unless it is null. also, slave->jiffies is not needed
2470          * here because we send an arp on each slave and give a slave as
2471          * long as it needs to get the tx/rx within the delta.
2472          * TODO: what about up/down delay in arp mode? it wasn't here before
2473          *       so it can wait
2474          */
2475         bond_for_each_slave(bond, slave, i) {
2476                 if (slave->link != BOND_LINK_UP) {
2477                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2478                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2479
2480                                 slave->link  = BOND_LINK_UP;
2481                                 slave->state = BOND_STATE_ACTIVE;
2482
2483                                 /* primary_slave has no meaning in round-robin
2484                                  * mode. the window of a slave being up and
2485                                  * curr_active_slave being null after enslaving
2486                                  * is closed.
2487                                  */
2488                                 if (!oldcurrent) {
2489                                         printk(KERN_INFO DRV_NAME
2490                                                ": %s: link status definitely "
2491                                                "up for interface %s, ",
2492                                                bond_dev->name,
2493                                                slave->dev->name);
2494                                         do_failover = 1;
2495                                 } else {
2496                                         printk(KERN_INFO DRV_NAME
2497                                                ": %s: interface %s is now up\n",
2498                                                bond_dev->name,
2499                                                slave->dev->name);
2500                                 }
2501                         }
2502                 } else {
2503                         /* slave->link == BOND_LINK_UP */
2504
2505                         /* not all switches will respond to an arp request
2506                          * when the source ip is 0, so don't take the link down
2507                          * if we don't know our ip yet
2508                          */
2509                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2510                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2511                              bond_has_ip(bond))) {
2512
2513                                 slave->link  = BOND_LINK_DOWN;
2514                                 slave->state = BOND_STATE_BACKUP;
2515
2516                                 if (slave->link_failure_count < UINT_MAX) {
2517                                         slave->link_failure_count++;
2518                                 }
2519
2520                                 printk(KERN_INFO DRV_NAME
2521                                        ": %s: interface %s is now down.\n",
2522                                        bond_dev->name,
2523                                        slave->dev->name);
2524
2525                                 if (slave == oldcurrent) {
2526                                         do_failover = 1;
2527                                 }
2528                         }
2529                 }
2530
2531                 /* note: if switch is in round-robin mode, all links
2532                  * must tx arp to ensure all links rx an arp - otherwise
2533                  * links may oscillate or not come up at all; if switch is
2534                  * in something like xor mode, there is nothing we can
2535                  * do - all replies will be rx'ed on same link causing slaves
2536                  * to be unstable during low/no traffic periods
2537                  */
2538                 if (IS_UP(slave->dev)) {
2539                         bond_arp_send_all(bond, slave);
2540                 }
2541         }
2542
2543         if (do_failover) {
2544                 write_lock(&bond->curr_slave_lock);
2545
2546                 bond_select_active_slave(bond);
2547
2548                 write_unlock(&bond->curr_slave_lock);
2549         }
2550
2551 re_arm:
2552         if (bond->params.arp_interval) {
2553                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2554         }
2555 out:
2556         read_unlock(&bond->lock);
2557 }
2558
2559 /*
2560  * When using arp monitoring in active-backup mode, this function is
2561  * called to determine if any backup slaves have went down or a new
2562  * current slave needs to be found.
2563  * The backup slaves never generate traffic, they are considered up by merely
2564  * receiving traffic. If the current slave goes down, each backup slave will
2565  * be given the opportunity to tx/rx an arp before being taken down - this
2566  * prevents all slaves from being taken down due to the current slave not
2567  * sending any traffic for the backups to receive. The arps are not necessarily
2568  * necessary, any tx and rx traffic will keep the current slave up. While any
2569  * rx traffic will keep the backup slaves up, the current slave is responsible
2570  * for generating traffic to keep them up regardless of any other traffic they
2571  * may have received.
2572  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2573  */
2574 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2575 {
2576         struct bonding *bond = bond_dev->priv;
2577         struct slave *slave;
2578         int delta_in_ticks;
2579         int i;
2580
2581         read_lock(&bond->lock);
2582
2583         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2584
2585         if (bond->kill_timers) {
2586                 goto out;
2587         }
2588
2589         if (bond->slave_cnt == 0) {
2590                 goto re_arm;
2591         }
2592
2593         /* determine if any slave has come up or any backup slave has
2594          * gone down
2595          * TODO: what about up/down delay in arp mode? it wasn't here before
2596          *       so it can wait
2597          */
2598         bond_for_each_slave(bond, slave, i) {
2599                 if (slave->link != BOND_LINK_UP) {
2600                         if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2601
2602                                 slave->link = BOND_LINK_UP;
2603
2604                                 write_lock(&bond->curr_slave_lock);
2605
2606                                 if ((!bond->curr_active_slave) &&
2607                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2608                                         bond_change_active_slave(bond, slave);
2609                                         bond->current_arp_slave = NULL;
2610                                 } else if (bond->curr_active_slave != slave) {
2611                                         /* this slave has just come up but we
2612                                          * already have a current slave; this
2613                                          * can also happen if bond_enslave adds
2614                                          * a new slave that is up while we are
2615                                          * searching for a new slave
2616                                          */
2617                                         bond_set_slave_inactive_flags(slave);
2618                                         bond->current_arp_slave = NULL;
2619                                 }
2620
2621                                 bond_set_carrier(bond);
2622
2623                                 if (slave == bond->curr_active_slave) {
2624                                         printk(KERN_INFO DRV_NAME
2625                                                ": %s: %s is up and now the "
2626                                                "active interface\n",
2627                                                bond_dev->name,
2628                                                slave->dev->name);
2629                                         netif_carrier_on(bond->dev);
2630                                 } else {
2631                                         printk(KERN_INFO DRV_NAME
2632                                                ": %s: backup interface %s is "
2633                                                "now up\n",
2634                                                bond_dev->name,
2635                                                slave->dev->name);
2636                                 }
2637
2638                                 write_unlock(&bond->curr_slave_lock);
2639                         }
2640                 } else {
2641                         read_lock(&bond->curr_slave_lock);
2642
2643                         if ((slave != bond->curr_active_slave) &&
2644                             (!bond->current_arp_slave) &&
2645                             (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2646                              bond_has_ip(bond))) {
2647                                 /* a backup slave has gone down; three times
2648                                  * the delta allows the current slave to be
2649                                  * taken out before the backup slave.
2650                                  * note: a non-null current_arp_slave indicates
2651                                  * the curr_active_slave went down and we are
2652                                  * searching for a new one; under this
2653                                  * condition we only take the curr_active_slave
2654                                  * down - this gives each slave a chance to
2655                                  * tx/rx traffic before being taken out
2656                                  */
2657
2658                                 read_unlock(&bond->curr_slave_lock);
2659
2660                                 slave->link  = BOND_LINK_DOWN;
2661
2662                                 if (slave->link_failure_count < UINT_MAX) {
2663                                         slave->link_failure_count++;
2664                                 }
2665
2666                                 bond_set_slave_inactive_flags(slave);
2667
2668                                 printk(KERN_INFO DRV_NAME
2669                                        ": %s: backup interface %s is now down\n",
2670                                        bond_dev->name,
2671                                        slave->dev->name);
2672                         } else {
2673                                 read_unlock(&bond->curr_slave_lock);
2674                         }
2675                 }
2676         }
2677
2678         read_lock(&bond->curr_slave_lock);
2679         slave = bond->curr_active_slave;
2680         read_unlock(&bond->curr_slave_lock);
2681
2682         if (slave) {
2683                 /* if we have sent traffic in the past 2*arp_intervals but
2684                  * haven't xmit and rx traffic in that time interval, select
2685                  * a different slave. slave->jiffies is only updated when
2686                  * a slave first becomes the curr_active_slave - not necessarily
2687                  * after every arp; this ensures the slave has a full 2*delta
2688                  * before being taken out. if a primary is being used, check
2689                  * if it is up and needs to take over as the curr_active_slave
2690                  */
2691                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2692             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2693              bond_has_ip(bond))) &&
2694                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2695
2696                         slave->link  = BOND_LINK_DOWN;
2697
2698                         if (slave->link_failure_count < UINT_MAX) {
2699                                 slave->link_failure_count++;
2700                         }
2701
2702                         printk(KERN_INFO DRV_NAME
2703                                ": %s: link status down for active interface "
2704                                "%s, disabling it\n",
2705                                bond_dev->name,
2706                                slave->dev->name);
2707
2708                         write_lock(&bond->curr_slave_lock);
2709
2710                         bond_select_active_slave(bond);
2711                         slave = bond->curr_active_slave;
2712
2713                         write_unlock(&bond->curr_slave_lock);
2714
2715                         bond->current_arp_slave = slave;
2716
2717                         if (slave) {
2718                                 slave->jiffies = jiffies;
2719                         }
2720                 } else if ((bond->primary_slave) &&
2721                            (bond->primary_slave != slave) &&
2722                            (bond->primary_slave->link == BOND_LINK_UP)) {
2723                         /* at this point, slave is the curr_active_slave */
2724                         printk(KERN_INFO DRV_NAME
2725                                ": %s: changing from interface %s to primary "
2726                                "interface %s\n",
2727                                bond_dev->name,
2728                                slave->dev->name,
2729                                bond->primary_slave->dev->name);
2730
2731                         /* primary is up so switch to it */
2732                         write_lock(&bond->curr_slave_lock);
2733                         bond_change_active_slave(bond, bond->primary_slave);
2734                         write_unlock(&bond->curr_slave_lock);
2735
2736                         slave = bond->primary_slave;
2737                         slave->jiffies = jiffies;
2738                 } else {
2739                         bond->current_arp_slave = NULL;
2740                 }
2741
2742                 /* the current slave must tx an arp to ensure backup slaves
2743                  * rx traffic
2744                  */
2745                 if (slave && bond_has_ip(bond)) {
2746                         bond_arp_send_all(bond, slave);
2747                 }
2748         }
2749
2750         /* if we don't have a curr_active_slave, search for the next available
2751          * backup slave from the current_arp_slave and make it the candidate
2752          * for becoming the curr_active_slave
2753          */
2754         if (!slave) {
2755                 if (!bond->current_arp_slave) {
2756                         bond->current_arp_slave = bond->first_slave;
2757                 }
2758
2759                 if (bond->current_arp_slave) {
2760                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2761
2762                         /* search for next candidate */
2763                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2764                                 if (IS_UP(slave->dev)) {
2765                                         slave->link = BOND_LINK_BACK;
2766                                         bond_set_slave_active_flags(slave);
2767                                         bond_arp_send_all(bond, slave);
2768                                         slave->jiffies = jiffies;
2769                                         bond->current_arp_slave = slave;
2770                                         break;
2771                                 }
2772
2773                                 /* if the link state is up at this point, we
2774                                  * mark it down - this can happen if we have
2775                                  * simultaneous link failures and
2776                                  * reselect_active_interface doesn't make this
2777                                  * one the current slave so it is still marked
2778                                  * up when it is actually down
2779                                  */
2780                                 if (slave->link == BOND_LINK_UP) {
2781                                         slave->link  = BOND_LINK_DOWN;
2782                                         if (slave->link_failure_count < UINT_MAX) {
2783                                                 slave->link_failure_count++;
2784                                         }
2785
2786                                         bond_set_slave_inactive_flags(slave);
2787
2788                                         printk(KERN_INFO DRV_NAME
2789                                                ": %s: backup interface %s is "
2790                                                "now down.\n",
2791                                                bond_dev->name,
2792                                                slave->dev->name);
2793                                 }
2794                         }
2795                 }
2796         }
2797
2798 re_arm:
2799         if (bond->params.arp_interval) {
2800                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2801         }
2802 out:
2803         read_unlock(&bond->lock);
2804 }
2805
2806 /*------------------------------ proc/seq_file-------------------------------*/
2807
2808 #ifdef CONFIG_PROC_FS
2809
2810 #define SEQ_START_TOKEN ((void *)1)
2811
2812 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2813 {
2814         struct bonding *bond = seq->private;
2815         loff_t off = 0;
2816         struct slave *slave;
2817         int i;
2818
2819         /* make sure the bond won't be taken away */
2820         read_lock(&dev_base_lock);
2821         read_lock_bh(&bond->lock);
2822
2823         if (*pos == 0) {
2824                 return SEQ_START_TOKEN;
2825         }
2826
2827         bond_for_each_slave(bond, slave, i) {
2828                 if (++off == *pos) {
2829                         return slave;
2830                 }
2831         }
2832
2833         return NULL;
2834 }
2835
2836 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2837 {
2838         struct bonding *bond = seq->private;
2839         struct slave *slave = v;
2840
2841         ++*pos;
2842         if (v == SEQ_START_TOKEN) {
2843                 return bond->first_slave;
2844         }
2845
2846         slave = slave->next;
2847
2848         return (slave == bond->first_slave) ? NULL : slave;
2849 }
2850
2851 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2852 {
2853         struct bonding *bond = seq->private;
2854
2855         read_unlock_bh(&bond->lock);
2856         read_unlock(&dev_base_lock);
2857 }
2858
2859 static void bond_info_show_master(struct seq_file *seq)
2860 {
2861         struct bonding *bond = seq->private;
2862         struct slave *curr;
2863         int i;
2864         u32 target;
2865
2866         read_lock(&bond->curr_slave_lock);
2867         curr = bond->curr_active_slave;
2868         read_unlock(&bond->curr_slave_lock);
2869
2870         seq_printf(seq, "Bonding Mode: %s\n",
2871                    bond_mode_name(bond->params.mode));
2872
2873         if (bond->params.mode == BOND_MODE_XOR ||
2874                 bond->params.mode == BOND_MODE_8023AD) {
2875                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2876                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2877                         bond->params.xmit_policy);
2878         }
2879
2880         if (USES_PRIMARY(bond->params.mode)) {
2881                 seq_printf(seq, "Primary Slave: %s\n",
2882                            (bond->primary_slave) ?
2883                            bond->primary_slave->dev->name : "None");
2884
2885                 seq_printf(seq, "Currently Active Slave: %s\n",
2886                            (curr) ? curr->dev->name : "None");
2887         }
2888
2889         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2890                    "up" : "down");
2891         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2892         seq_printf(seq, "Up Delay (ms): %d\n",
2893                    bond->params.updelay * bond->params.miimon);
2894         seq_printf(seq, "Down Delay (ms): %d\n",
2895                    bond->params.downdelay * bond->params.miimon);
2896
2897
2898         /* ARP information */
2899         if(bond->params.arp_interval > 0) {
2900                 int printed=0;
2901                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2902                                 bond->params.arp_interval);
2903
2904                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2905
2906                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2907                         if (!bond->params.arp_targets[i])
2908                                 continue;
2909                         if (printed)
2910                                 seq_printf(seq, ",");
2911                         target = ntohl(bond->params.arp_targets[i]);
2912                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
2913                         printed = 1;
2914                 }
2915                 seq_printf(seq, "\n");
2916         }
2917
2918         if (bond->params.mode == BOND_MODE_8023AD) {
2919                 struct ad_info ad_info;
2920
2921                 seq_puts(seq, "\n802.3ad info\n");
2922                 seq_printf(seq, "LACP rate: %s\n",
2923                            (bond->params.lacp_fast) ? "fast" : "slow");
2924
2925                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2926                         seq_printf(seq, "bond %s has no active aggregator\n",
2927                                    bond->dev->name);
2928                 } else {
2929                         seq_printf(seq, "Active Aggregator Info:\n");
2930
2931                         seq_printf(seq, "\tAggregator ID: %d\n",
2932                                    ad_info.aggregator_id);
2933                         seq_printf(seq, "\tNumber of ports: %d\n",
2934                                    ad_info.ports);
2935                         seq_printf(seq, "\tActor Key: %d\n",
2936                                    ad_info.actor_key);
2937                         seq_printf(seq, "\tPartner Key: %d\n",
2938                                    ad_info.partner_key);
2939                         seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2940                                    ad_info.partner_system[0],
2941                                    ad_info.partner_system[1],
2942                                    ad_info.partner_system[2],
2943                                    ad_info.partner_system[3],
2944                                    ad_info.partner_system[4],
2945                                    ad_info.partner_system[5]);
2946                 }
2947         }
2948 }
2949
2950 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
2951 {
2952         struct bonding *bond = seq->private;
2953
2954         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
2955         seq_printf(seq, "MII Status: %s\n",
2956                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
2957         seq_printf(seq, "Link Failure Count: %d\n",
2958                    slave->link_failure_count);
2959
2960         seq_printf(seq,
2961                    "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2962                    slave->perm_hwaddr[0], slave->perm_hwaddr[1],
2963                    slave->perm_hwaddr[2], slave->perm_hwaddr[3],
2964                    slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
2965
2966         if (bond->params.mode == BOND_MODE_8023AD) {
2967                 const struct aggregator *agg
2968                         = SLAVE_AD_INFO(slave).port.aggregator;
2969
2970                 if (agg) {
2971                         seq_printf(seq, "Aggregator ID: %d\n",
2972                                    agg->aggregator_identifier);
2973                 } else {
2974                         seq_puts(seq, "Aggregator ID: N/A\n");
2975                 }
2976         }
2977 }
2978
2979 static int bond_info_seq_show(struct seq_file *seq, void *v)
2980 {
2981         if (v == SEQ_START_TOKEN) {
2982                 seq_printf(seq, "%s\n", version);
2983                 bond_info_show_master(seq);
2984         } else {
2985                 bond_info_show_slave(seq, v);
2986         }
2987
2988         return 0;
2989 }
2990
2991 static struct seq_operations bond_info_seq_ops = {
2992         .start = bond_info_seq_start,
2993         .next  = bond_info_seq_next,
2994         .stop  = bond_info_seq_stop,
2995         .show  = bond_info_seq_show,
2996 };
2997
2998 static int bond_info_open(struct inode *inode, struct file *file)
2999 {
3000         struct seq_file *seq;
3001         struct proc_dir_entry *proc;
3002         int res;
3003
3004         res = seq_open(file, &bond_info_seq_ops);
3005         if (!res) {
3006                 /* recover the pointer buried in proc_dir_entry data */
3007                 seq = file->private_data;
3008                 proc = PDE(inode);
3009                 seq->private = proc->data;
3010         }
3011
3012         return res;
3013 }
3014
3015 static struct file_operations bond_info_fops = {
3016         .owner   = THIS_MODULE,
3017         .open    = bond_info_open,
3018         .read    = seq_read,
3019         .llseek  = seq_lseek,
3020         .release = seq_release,
3021 };
3022
3023 static int bond_create_proc_entry(struct bonding *bond)
3024 {
3025         struct net_device *bond_dev = bond->dev;
3026
3027         if (bond_proc_dir) {
3028                 bond->proc_entry = create_proc_entry(bond_dev->name,
3029                                                      S_IRUGO,
3030                                                      bond_proc_dir);
3031                 if (bond->proc_entry == NULL) {
3032                         printk(KERN_WARNING DRV_NAME
3033                                ": Warning: Cannot create /proc/net/%s/%s\n",
3034                                DRV_NAME, bond_dev->name);
3035                 } else {
3036                         bond->proc_entry->data = bond;
3037                         bond->proc_entry->proc_fops = &bond_info_fops;
3038                         bond->proc_entry->owner = THIS_MODULE;
3039                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3040                 }
3041         }
3042
3043         return 0;
3044 }
3045
3046 static void bond_remove_proc_entry(struct bonding *bond)
3047 {
3048         if (bond_proc_dir && bond->proc_entry) {
3049                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3050                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3051                 bond->proc_entry = NULL;
3052         }
3053 }
3054
3055 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3056  * Caller must hold rtnl_lock.
3057  */
3058 static void bond_create_proc_dir(void)
3059 {
3060         int len = strlen(DRV_NAME);
3061
3062         for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3063              bond_proc_dir = bond_proc_dir->next) {
3064                 if ((bond_proc_dir->namelen == len) &&
3065                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3066                         break;
3067                 }
3068         }
3069
3070         if (!bond_proc_dir) {
3071                 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3072                 if (bond_proc_dir) {
3073                         bond_proc_dir->owner = THIS_MODULE;
3074                 } else {
3075                         printk(KERN_WARNING DRV_NAME
3076                                 ": Warning: cannot create /proc/net/%s\n",
3077                                 DRV_NAME);
3078                 }
3079         }
3080 }
3081
3082 /* Destroy the bonding directory under /proc/net, if empty.
3083  * Caller must hold rtnl_lock.
3084  */
3085 static void bond_destroy_proc_dir(void)
3086 {
3087         struct proc_dir_entry *de;
3088
3089         if (!bond_proc_dir) {
3090                 return;
3091         }
3092
3093         /* verify that the /proc dir is empty */
3094         for (de = bond_proc_dir->subdir; de; de = de->next) {
3095                 /* ignore . and .. */
3096                 if (*(de->name) != '.') {
3097                         break;
3098                 }
3099         }
3100
3101         if (de) {
3102                 if (bond_proc_dir->owner == THIS_MODULE) {
3103                         bond_proc_dir->owner = NULL;
3104                 }
3105         } else {
3106                 remove_proc_entry(DRV_NAME, proc_net);
3107                 bond_proc_dir = NULL;
3108         }
3109 }
3110 #endif /* CONFIG_PROC_FS */
3111
3112 /*-------------------------- netdev event handling --------------------------*/
3113
3114 /*
3115  * Change device name
3116  */
3117 static int bond_event_changename(struct bonding *bond)
3118 {
3119 #ifdef CONFIG_PROC_FS
3120         bond_remove_proc_entry(bond);
3121         bond_create_proc_entry(bond);
3122 #endif
3123         down_write(&(bonding_rwsem));
3124         bond_destroy_sysfs_entry(bond);
3125         bond_create_sysfs_entry(bond);
3126         up_write(&(bonding_rwsem));
3127         return NOTIFY_DONE;
3128 }
3129
3130 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3131 {
3132         struct bonding *event_bond = bond_dev->priv;
3133
3134         switch (event) {
3135         case NETDEV_CHANGENAME:
3136                 return bond_event_changename(event_bond);
3137         case NETDEV_UNREGISTER:
3138                 /*
3139                  * TODO: remove a bond from the list?
3140                  */
3141                 break;
3142         default:
3143                 break;
3144         }
3145
3146         return NOTIFY_DONE;
3147 }
3148
3149 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3150 {
3151         struct net_device *bond_dev = slave_dev->master;
3152         struct bonding *bond = bond_dev->priv;
3153
3154         switch (event) {
3155         case NETDEV_UNREGISTER:
3156                 if (bond_dev) {
3157                         bond_release(bond_dev, slave_dev);
3158                 }
3159                 break;
3160         case NETDEV_CHANGE:
3161                 /*
3162                  * TODO: is this what we get if somebody
3163                  * sets up a hierarchical bond, then rmmod's
3164                  * one of the slave bonding devices?
3165                  */
3166                 break;
3167         case NETDEV_DOWN:
3168                 /*
3169                  * ... Or is it this?
3170                  */
3171                 break;
3172         case NETDEV_CHANGEMTU:
3173                 /*
3174                  * TODO: Should slaves be allowed to
3175                  * independently alter their MTU?  For
3176                  * an active-backup bond, slaves need
3177                  * not be the same type of device, so
3178                  * MTUs may vary.  For other modes,
3179                  * slaves arguably should have the
3180                  * same MTUs. To do this, we'd need to
3181                  * take over the slave's change_mtu
3182                  * function for the duration of their
3183                  * servitude.
3184                  */
3185                 break;
3186         case NETDEV_CHANGENAME:
3187                 /*
3188                  * TODO: handle changing the primary's name
3189                  */
3190                 break;
3191         case NETDEV_FEAT_CHANGE:
3192                 bond_compute_features(bond);
3193                 break;
3194         default:
3195                 break;
3196         }
3197
3198         return NOTIFY_DONE;
3199 }
3200
3201 /*
3202  * bond_netdev_event: handle netdev notifier chain events.
3203  *
3204  * This function receives events for the netdev chain.  The caller (an
3205  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3206  * locks for us to safely manipulate the slave devices (RTNL lock,
3207  * dev_probe_lock).
3208  */
3209 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3210 {
3211         struct net_device *event_dev = (struct net_device *)ptr;
3212
3213         dprintk("event_dev: %s, event: %lx\n",
3214                 (event_dev ? event_dev->name : "None"),
3215                 event);
3216
3217         if (event_dev->flags & IFF_MASTER) {
3218                 dprintk("IFF_MASTER\n");
3219                 return bond_master_netdev_event(event, event_dev);
3220         }
3221
3222         if (event_dev->flags & IFF_SLAVE) {
3223                 dprintk("IFF_SLAVE\n");
3224                 return bond_slave_netdev_event(event, event_dev);
3225         }
3226
3227         return NOTIFY_DONE;
3228 }
3229
3230 /*
3231  * bond_inetaddr_event: handle inetaddr notifier chain events.
3232  *
3233  * We keep track of device IPs primarily to use as source addresses in
3234  * ARP monitor probes (rather than spewing out broadcasts all the time).
3235  *
3236  * We track one IP for the main device (if it has one), plus one per VLAN.
3237  */
3238 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3239 {
3240         struct in_ifaddr *ifa = ptr;
3241         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3242         struct bonding *bond, *bond_next;
3243         struct vlan_entry *vlan, *vlan_next;
3244
3245         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3246                 if (bond->dev == event_dev) {
3247                         switch (event) {
3248                         case NETDEV_UP:
3249                                 bond->master_ip = ifa->ifa_local;
3250                                 return NOTIFY_OK;
3251                         case NETDEV_DOWN:
3252                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3253                                 return NOTIFY_OK;
3254                         default:
3255                                 return NOTIFY_DONE;
3256                         }
3257                 }
3258
3259                 if (list_empty(&bond->vlan_list))
3260                         continue;
3261
3262                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3263                                          vlan_list) {
3264                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3265                         if (vlan_dev == event_dev) {
3266                                 switch (event) {
3267                                 case NETDEV_UP:
3268                                         vlan->vlan_ip = ifa->ifa_local;
3269                                         return NOTIFY_OK;
3270                                 case NETDEV_DOWN:
3271                                         vlan->vlan_ip =
3272                                                 bond_glean_dev_ip(vlan_dev);
3273                                         return NOTIFY_OK;
3274                                 default:
3275                                         return NOTIFY_DONE;
3276                                 }
3277                         }
3278                 }
3279         }
3280         return NOTIFY_DONE;
3281 }
3282
3283 static struct notifier_block bond_netdev_notifier = {
3284         .notifier_call = bond_netdev_event,
3285 };
3286
3287 static struct notifier_block bond_inetaddr_notifier = {
3288         .notifier_call = bond_inetaddr_event,
3289 };
3290
3291 /*-------------------------- Packet type handling ---------------------------*/
3292
3293 /* register to receive lacpdus on a bond */
3294 static void bond_register_lacpdu(struct bonding *bond)
3295 {
3296         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3297
3298         /* initialize packet type */
3299         pk_type->type = PKT_TYPE_LACPDU;
3300         pk_type->dev = bond->dev;
3301         pk_type->func = bond_3ad_lacpdu_recv;
3302
3303         dev_add_pack(pk_type);
3304 }
3305
3306 /* unregister to receive lacpdus on a bond */
3307 static void bond_unregister_lacpdu(struct bonding *bond)
3308 {
3309         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3310 }
3311
3312 /*---------------------------- Hashing Policies -----------------------------*/
3313
3314 /*
3315  * Hash for the the output device based upon layer 3 and layer 4 data. If
3316  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3317  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3318  */
3319 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3320                                     struct net_device *bond_dev, int count)
3321 {
3322         struct ethhdr *data = (struct ethhdr *)skb->data;
3323         struct iphdr *iph = skb->nh.iph;
3324         u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3325         int layer4_xor = 0;
3326
3327         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3328                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3329                     (iph->protocol == IPPROTO_TCP ||
3330                      iph->protocol == IPPROTO_UDP)) {
3331                         layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3332                 }
3333                 return (layer4_xor ^
3334                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3335
3336         }
3337
3338         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3339 }
3340
3341 /*
3342  * Hash for the output device based upon layer 2 data
3343  */
3344 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3345                                    struct net_device *bond_dev, int count)
3346 {
3347         struct ethhdr *data = (struct ethhdr *)skb->data;
3348
3349         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3350 }
3351
3352 /*-------------------------- Device entry points ----------------------------*/
3353
3354 static int bond_open(struct net_device *bond_dev)
3355 {
3356         struct bonding *bond = bond_dev->priv;
3357         struct timer_list *mii_timer = &bond->mii_timer;
3358         struct timer_list *arp_timer = &bond->arp_timer;
3359
3360         bond->kill_timers = 0;
3361
3362         if ((bond->params.mode == BOND_MODE_TLB) ||
3363             (bond->params.mode == BOND_MODE_ALB)) {
3364                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3365
3366                 /* bond_alb_initialize must be called before the timer
3367                  * is started.
3368                  */
3369                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3370                         /* something went wrong - fail the open operation */
3371                         return -1;
3372                 }
3373
3374                 init_timer(alb_timer);
3375                 alb_timer->expires  = jiffies + 1;
3376                 alb_timer->data     = (unsigned long)bond;
3377                 alb_timer->function = (void *)&bond_alb_monitor;
3378                 add_timer(alb_timer);
3379         }
3380
3381         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3382                 init_timer(mii_timer);
3383                 mii_timer->expires  = jiffies + 1;
3384                 mii_timer->data     = (unsigned long)bond_dev;
3385                 mii_timer->function = (void *)&bond_mii_monitor;
3386                 add_timer(mii_timer);
3387         }
3388
3389         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3390                 init_timer(arp_timer);
3391                 arp_timer->expires  = jiffies + 1;
3392                 arp_timer->data     = (unsigned long)bond_dev;
3393                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3394                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3395                 } else {
3396                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3397                 }
3398                 add_timer(arp_timer);
3399         }
3400
3401         if (bond->params.mode == BOND_MODE_8023AD) {
3402                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3403                 init_timer(ad_timer);
3404                 ad_timer->expires  = jiffies + 1;
3405                 ad_timer->data     = (unsigned long)bond;
3406                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3407                 add_timer(ad_timer);
3408
3409                 /* register to receive LACPDUs */
3410                 bond_register_lacpdu(bond);
3411         }
3412
3413         return 0;
3414 }
3415
3416 static int bond_close(struct net_device *bond_dev)
3417 {
3418         struct bonding *bond = bond_dev->priv;
3419
3420         if (bond->params.mode == BOND_MODE_8023AD) {
3421                 /* Unregister the receive of LACPDUs */
3422                 bond_unregister_lacpdu(bond);
3423         }
3424
3425         write_lock_bh(&bond->lock);
3426
3427         bond_mc_list_destroy(bond);
3428
3429         /* signal timers not to re-arm */
3430         bond->kill_timers = 1;
3431
3432         write_unlock_bh(&bond->lock);
3433
3434         /* del_timer_sync must run without holding the bond->lock
3435          * because a running timer might be trying to hold it too
3436          */
3437
3438         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3439                 del_timer_sync(&bond->mii_timer);
3440         }
3441
3442         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3443                 del_timer_sync(&bond->arp_timer);
3444         }
3445
3446         switch (bond->params.mode) {
3447         case BOND_MODE_8023AD:
3448                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3449                 break;
3450         case BOND_MODE_TLB:
3451         case BOND_MODE_ALB:
3452                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3453                 break;
3454         default:
3455                 break;
3456         }
3457
3458         /* Release the bonded slaves */
3459         bond_release_all(bond_dev);
3460
3461         if ((bond->params.mode == BOND_MODE_TLB) ||
3462             (bond->params.mode == BOND_MODE_ALB)) {
3463                 /* Must be called only after all
3464                  * slaves have been released
3465                  */
3466                 bond_alb_deinitialize(bond);
3467         }
3468
3469         return 0;
3470 }
3471
3472 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3473 {
3474         struct bonding *bond = bond_dev->priv;
3475         struct net_device_stats *stats = &(bond->stats), *sstats;
3476         struct slave *slave;
3477         int i;
3478
3479         memset(stats, 0, sizeof(struct net_device_stats));
3480
3481         read_lock_bh(&bond->lock);
3482
3483         bond_for_each_slave(bond, slave, i) {
3484                 sstats = slave->dev->get_stats(slave->dev);
3485
3486                 stats->rx_packets += sstats->rx_packets;
3487                 stats->rx_bytes += sstats->rx_bytes;
3488                 stats->rx_errors += sstats->rx_errors;
3489                 stats->rx_dropped += sstats->rx_dropped;
3490
3491                 stats->tx_packets += sstats->tx_packets;
3492                 stats->tx_bytes += sstats->tx_bytes;
3493                 stats->tx_errors += sstats->tx_errors;
3494                 stats->tx_dropped += sstats->tx_dropped;
3495
3496                 stats->multicast += sstats->multicast;
3497                 stats->collisions += sstats->collisions;
3498
3499                 stats->rx_length_errors += sstats->rx_length_errors;
3500                 stats->rx_over_errors += sstats->rx_over_errors;
3501                 stats->rx_crc_errors += sstats->rx_crc_errors;
3502                 stats->rx_frame_errors += sstats->rx_frame_errors;
3503                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3504                 stats->rx_missed_errors += sstats->rx_missed_errors;
3505
3506                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3507                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3508                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3509                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3510                 stats->tx_window_errors += sstats->tx_window_errors;
3511         }
3512
3513         read_unlock_bh(&bond->lock);
3514
3515         return stats;
3516 }
3517
3518 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3519 {
3520         struct net_device *slave_dev = NULL;
3521         struct ifbond k_binfo;
3522         struct ifbond __user *u_binfo = NULL;
3523         struct ifslave k_sinfo;
3524         struct ifslave __user *u_sinfo = NULL;
3525         struct mii_ioctl_data *mii = NULL;
3526         int res = 0;
3527
3528         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3529                 bond_dev->name, cmd);
3530
3531         switch (cmd) {
3532         case SIOCGMIIPHY:
3533                 mii = if_mii(ifr);
3534                 if (!mii) {
3535                         return -EINVAL;
3536                 }
3537                 mii->phy_id = 0;
3538                 /* Fall Through */
3539         case SIOCGMIIREG:
3540                 /*
3541                  * We do this again just in case we were called by SIOCGMIIREG
3542                  * instead of SIOCGMIIPHY.
3543                  */
3544                 mii = if_mii(ifr);
3545                 if (!mii) {
3546                         return -EINVAL;
3547                 }
3548
3549                 if (mii->reg_num == 1) {
3550                         struct bonding *bond = bond_dev->priv;
3551                         mii->val_out = 0;
3552                         read_lock_bh(&bond->lock);
3553                         read_lock(&bond->curr_slave_lock);
3554                         if (bond->curr_active_slave) {
3555                                 mii->val_out = BMSR_LSTATUS;
3556                         }
3557                         read_unlock(&bond->curr_slave_lock);
3558                         read_unlock_bh(&bond->lock);
3559                 }
3560
3561                 return 0;
3562         case BOND_INFO_QUERY_OLD:
3563         case SIOCBONDINFOQUERY:
3564                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3565
3566                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3567                         return -EFAULT;
3568                 }
3569
3570                 res = bond_info_query(bond_dev, &k_binfo);
3571                 if (res == 0) {
3572                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3573                                 return -EFAULT;
3574                         }
3575                 }
3576
3577                 return res;
3578         case BOND_SLAVE_INFO_QUERY_OLD:
3579         case SIOCBONDSLAVEINFOQUERY:
3580                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3581
3582                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3583                         return -EFAULT;
3584                 }
3585
3586                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3587                 if (res == 0) {
3588                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3589                                 return -EFAULT;
3590                         }
3591                 }
3592
3593                 return res;
3594         default:
3595                 /* Go on */
3596                 break;
3597         }
3598
3599         if (!capable(CAP_NET_ADMIN)) {
3600                 return -EPERM;
3601         }
3602
3603         down_write(&(bonding_rwsem));
3604         slave_dev = dev_get_by_name(ifr->ifr_slave);
3605
3606         dprintk("slave_dev=%p: \n", slave_dev);
3607
3608         if (!slave_dev) {
3609                 res = -ENODEV;
3610         } else {
3611                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3612                 switch (cmd) {
3613                 case BOND_ENSLAVE_OLD:
3614                 case SIOCBONDENSLAVE:
3615                         res = bond_enslave(bond_dev, slave_dev);
3616                         break;
3617                 case BOND_RELEASE_OLD:
3618                 case SIOCBONDRELEASE:
3619                         res = bond_release(bond_dev, slave_dev);
3620                         break;
3621                 case BOND_SETHWADDR_OLD:
3622                 case SIOCBONDSETHWADDR:
3623                         res = bond_sethwaddr(bond_dev, slave_dev);
3624                         break;
3625                 case BOND_CHANGE_ACTIVE_OLD:
3626                 case SIOCBONDCHANGEACTIVE:
3627                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3628                         break;
3629                 default:
3630                         res = -EOPNOTSUPP;
3631                 }
3632
3633                 dev_put(slave_dev);
3634         }
3635
3636         up_write(&(bonding_rwsem));
3637         return res;
3638 }
3639
3640 static void bond_set_multicast_list(struct net_device *bond_dev)
3641 {
3642         struct bonding *bond = bond_dev->priv;
3643         struct dev_mc_list *dmi;
3644
3645         write_lock_bh(&bond->lock);
3646
3647         /*
3648          * Do promisc before checking multicast_mode
3649          */
3650         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3651                 bond_set_promiscuity(bond, 1);
3652         }
3653
3654         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3655                 bond_set_promiscuity(bond, -1);
3656         }
3657
3658         /* set allmulti flag to slaves */
3659         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3660                 bond_set_allmulti(bond, 1);
3661         }
3662
3663         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3664                 bond_set_allmulti(bond, -1);
3665         }
3666
3667         bond->flags = bond_dev->flags;
3668
3669         /* looking for addresses to add to slaves' mc list */
3670         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3671                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3672                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3673                 }
3674         }
3675
3676         /* looking for addresses to delete from slaves' list */
3677         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3678                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3679                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3680                 }
3681         }
3682
3683         /* save master's multicast list */
3684         bond_mc_list_destroy(bond);
3685         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3686
3687         write_unlock_bh(&bond->lock);
3688 }
3689
3690 /*
3691  * Change the MTU of all of a master's slaves to match the master
3692  */
3693 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3694 {
3695         struct bonding *bond = bond_dev->priv;
3696         struct slave *slave, *stop_at;
3697         int res = 0;
3698         int i;
3699
3700         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3701                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3702
3703         /* Can't hold bond->lock with bh disabled here since
3704          * some base drivers panic. On the other hand we can't
3705          * hold bond->lock without bh disabled because we'll
3706          * deadlock. The only solution is to rely on the fact
3707          * that we're under rtnl_lock here, and the slaves
3708          * list won't change. This doesn't solve the problem
3709          * of setting the slave's MTU while it is
3710          * transmitting, but the assumption is that the base
3711          * driver can handle that.
3712          *
3713          * TODO: figure out a way to safely iterate the slaves
3714          * list, but without holding a lock around the actual
3715          * call to the base driver.
3716          */
3717
3718         bond_for_each_slave(bond, slave, i) {
3719                 dprintk("s %p s->p %p c_m %p\n", slave,
3720                         slave->prev, slave->dev->change_mtu);
3721
3722                 res = dev_set_mtu(slave->dev, new_mtu);
3723
3724                 if (res) {
3725                         /* If we failed to set the slave's mtu to the new value
3726                          * we must abort the operation even in ACTIVE_BACKUP
3727                          * mode, because if we allow the backup slaves to have
3728                          * different mtu values than the active slave we'll
3729                          * need to change their mtu when doing a failover. That
3730                          * means changing their mtu from timer context, which
3731                          * is probably not a good idea.
3732                          */
3733                         dprintk("err %d %s\n", res, slave->dev->name);
3734                         goto unwind;
3735                 }
3736         }
3737
3738         bond_dev->mtu = new_mtu;
3739
3740         return 0;
3741
3742 unwind:
3743         /* unwind from head to the slave that failed */
3744         stop_at = slave;
3745         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3746                 int tmp_res;
3747
3748                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3749                 if (tmp_res) {
3750                         dprintk("unwind err %d dev %s\n", tmp_res,
3751                                 slave->dev->name);
3752                 }
3753         }
3754
3755         return res;
3756 }
3757
3758 /*
3759  * Change HW address
3760  *
3761  * Note that many devices must be down to change the HW address, and
3762  * downing the master releases all slaves.  We can make bonds full of
3763  * bonding devices to test this, however.
3764  */
3765 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3766 {
3767         struct bonding *bond = bond_dev->priv;
3768         struct sockaddr *sa = addr, tmp_sa;
3769         struct slave *slave, *stop_at;
3770         int res = 0;
3771         int i;
3772
3773         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3774
3775         if (!is_valid_ether_addr(sa->sa_data)) {
3776                 return -EADDRNOTAVAIL;
3777         }
3778
3779         /* Can't hold bond->lock with bh disabled here since
3780          * some base drivers panic. On the other hand we can't
3781          * hold bond->lock without bh disabled because we'll
3782          * deadlock. The only solution is to rely on the fact
3783          * that we're under rtnl_lock here, and the slaves
3784          * list won't change. This doesn't solve the problem
3785          * of setting the slave's hw address while it is
3786          * transmitting, but the assumption is that the base
3787          * driver can handle that.
3788          *
3789          * TODO: figure out a way to safely iterate the slaves
3790          * list, but without holding a lock around the actual
3791          * call to the base driver.
3792          */
3793
3794         bond_for_each_slave(bond, slave, i) {
3795                 dprintk("slave %p %s\n", slave, slave->dev->name);
3796
3797                 if (slave->dev->set_mac_address == NULL) {
3798                         res = -EOPNOTSUPP;
3799                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3800                         goto unwind;
3801                 }
3802
3803                 res = dev_set_mac_address(slave->dev, addr);
3804                 if (res) {
3805                         /* TODO: consider downing the slave
3806                          * and retry ?
3807                          * User should expect communications
3808                          * breakage anyway until ARP finish
3809                          * updating, so...
3810                          */
3811                         dprintk("err %d %s\n", res, slave->dev->name);
3812                         goto unwind;
3813                 }
3814         }
3815
3816         /* success */
3817         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3818         return 0;
3819
3820 unwind:
3821         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3822         tmp_sa.sa_family = bond_dev->type;
3823
3824         /* unwind from head to the slave that failed */
3825         stop_at = slave;
3826         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3827                 int tmp_res;
3828
3829                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3830                 if (tmp_res) {
3831                         dprintk("unwind err %d dev %s\n", tmp_res,
3832                                 slave->dev->name);
3833                 }
3834         }
3835
3836         return res;
3837 }
3838
3839 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3840 {
3841         struct bonding *bond = bond_dev->priv;
3842         struct slave *slave, *start_at;
3843         int i;
3844         int res = 1;
3845
3846         read_lock(&bond->lock);
3847
3848         if (!BOND_IS_OK(bond)) {
3849                 goto out;
3850         }
3851
3852         read_lock(&bond->curr_slave_lock);
3853         slave = start_at = bond->curr_active_slave;
3854         read_unlock(&bond->curr_slave_lock);
3855
3856         if (!slave) {
3857                 goto out;
3858         }
3859
3860         bond_for_each_slave_from(bond, slave, i, start_at) {
3861                 if (IS_UP(slave->dev) &&
3862                     (slave->link == BOND_LINK_UP) &&
3863                     (slave->state == BOND_STATE_ACTIVE)) {
3864                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3865
3866                         write_lock(&bond->curr_slave_lock);
3867                         bond->curr_active_slave = slave->next;
3868                         write_unlock(&bond->curr_slave_lock);
3869
3870                         break;
3871                 }
3872         }
3873
3874
3875 out:
3876         if (res) {
3877                 /* no suitable interface, frame not sent */
3878                 dev_kfree_skb(skb);
3879         }
3880         read_unlock(&bond->lock);
3881         return 0;
3882 }
3883
3884 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
3885                                         struct bonding *bond,
3886                                         struct slave *slave)
3887 {
3888         struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
3889         struct ethhdr *eth_data;
3890         u8 *hwaddr;
3891         int res;
3892
3893         if (!skb2) {
3894                 printk(KERN_ERR DRV_NAME ": Error: "
3895                        "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3896                 return;
3897         }
3898
3899         skb2->mac.raw = (unsigned char *)skb2->data;
3900         eth_data = eth_hdr(skb2);
3901
3902         /* Pick an appropriate source MAC address
3903          *      -- use slave's perm MAC addr, unless used by bond
3904          *      -- otherwise, borrow active slave's perm MAC addr
3905          *         since that will not be used
3906          */
3907         hwaddr = slave->perm_hwaddr;
3908         if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
3909                 hwaddr = bond->curr_active_slave->perm_hwaddr;
3910
3911         /* Set source MAC address appropriately */
3912         memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
3913
3914         res = bond_dev_queue_xmit(bond, skb2, slave->dev);
3915         if (res)
3916                 dev_kfree_skb(skb2);
3917
3918         return;
3919 }
3920
3921 /*
3922  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3923  * the bond has a usable interface.
3924  */
3925 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3926 {
3927         struct bonding *bond = bond_dev->priv;
3928         int res = 1;
3929
3930         read_lock(&bond->lock);
3931         read_lock(&bond->curr_slave_lock);
3932
3933         if (!BOND_IS_OK(bond)) {
3934                 goto out;
3935         }
3936
3937         if (!bond->curr_active_slave)
3938                 goto out;
3939
3940         /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3941            for multicast traffic on snooping switches */
3942         if (skb->protocol == __constant_htons(ETH_P_IP) &&
3943             skb->nh.iph->protocol == IPPROTO_IGMP) {
3944                 struct slave *slave, *active_slave;
3945                 int i;
3946
3947                 active_slave = bond->curr_active_slave;
3948                 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
3949                                             active_slave->prev)
3950                         if (IS_UP(slave->dev) &&
3951                             (slave->link == BOND_LINK_UP))
3952                                 bond_activebackup_xmit_copy(skb, bond, slave);
3953         }
3954
3955         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
3956
3957 out:
3958         if (res) {
3959                 /* no suitable interface, frame not sent */
3960                 dev_kfree_skb(skb);
3961         }
3962         read_unlock(&bond->curr_slave_lock);
3963         read_unlock(&bond->lock);
3964         return 0;
3965 }
3966
3967 /*
3968  * In bond_xmit_xor() , we determine the output device by using a pre-
3969  * determined xmit_hash_policy(), If the selected device is not enabled,
3970  * find the next active slave.
3971  */
3972 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3973 {
3974         struct bonding *bond = bond_dev->priv;
3975         struct slave *slave, *start_at;
3976         int slave_no;
3977         int i;
3978         int res = 1;
3979
3980         read_lock(&bond->lock);
3981
3982         if (!BOND_IS_OK(bond)) {
3983                 goto out;
3984         }
3985
3986         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
3987
3988         bond_for_each_slave(bond, slave, i) {
3989                 slave_no--;
3990                 if (slave_no < 0) {
3991                         break;
3992                 }
3993         }
3994
3995         start_at = slave;
3996
3997         bond_for_each_slave_from(bond, slave, i, start_at) {
3998                 if (IS_UP(slave->dev) &&
3999                     (slave->link == BOND_LINK_UP) &&
4000                     (slave->state == BOND_STATE_ACTIVE)) {
4001                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4002                         break;
4003                 }
4004         }
4005
4006 out:
4007         if (res) {
4008                 /* no suitable interface, frame not sent */
4009                 dev_kfree_skb(skb);
4010         }
4011         read_unlock(&bond->lock);
4012         return 0;
4013 }
4014
4015 /*
4016  * in broadcast mode, we send everything to all usable interfaces.
4017  */
4018 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4019 {
4020         struct bonding *bond = bond_dev->priv;
4021         struct slave *slave, *start_at;
4022         struct net_device *tx_dev = NULL;
4023         int i;
4024         int res = 1;
4025
4026         read_lock(&bond->lock);
4027
4028         if (!BOND_IS_OK(bond)) {
4029                 goto out;
4030         }
4031
4032         read_lock(&bond->curr_slave_lock);
4033         start_at = bond->curr_active_slave;
4034         read_unlock(&bond->curr_slave_lock);
4035
4036         if (!start_at) {
4037                 goto out;
4038         }
4039
4040         bond_for_each_slave_from(bond, slave, i, start_at) {
4041                 if (IS_UP(slave->dev) &&
4042                     (slave->link == BOND_LINK_UP) &&
4043                     (slave->state == BOND_STATE_ACTIVE)) {
4044                         if (tx_dev) {
4045                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4046                                 if (!skb2) {
4047                                         printk(KERN_ERR DRV_NAME
4048                                                ": %s: Error: bond_xmit_broadcast(): "
4049                                                "skb_clone() failed\n",
4050                                                bond_dev->name);
4051                                         continue;
4052                                 }
4053
4054                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4055                                 if (res) {
4056                                         dev_kfree_skb(skb2);
4057                                         continue;
4058                                 }
4059                         }
4060                         tx_dev = slave->dev;
4061                 }
4062         }
4063
4064         if (tx_dev) {
4065                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4066         }
4067
4068 out:
4069         if (res) {
4070                 /* no suitable interface, frame not sent */
4071                 dev_kfree_skb(skb);
4072         }
4073         /* frame sent to all suitable interfaces */
4074         read_unlock(&bond->lock);
4075         return 0;
4076 }
4077
4078 /*------------------------- Device initialization ---------------------------*/
4079
4080 /*
4081  * set bond mode specific net device operations
4082  */
4083 void bond_set_mode_ops(struct bonding *bond, int mode)
4084 {
4085         struct net_device *bond_dev = bond->dev;
4086
4087         switch (mode) {
4088         case BOND_MODE_ROUNDROBIN:
4089                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4090                 break;
4091         case BOND_MODE_ACTIVEBACKUP:
4092                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4093                 break;
4094         case BOND_MODE_XOR:
4095                 bond_dev->hard_start_xmit = bond_xmit_xor;
4096                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4097                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4098                 else
4099                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4100                 break;
4101         case BOND_MODE_BROADCAST:
4102                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4103                 break;
4104         case BOND_MODE_8023AD:
4105                 bond_set_master_3ad_flags(bond);
4106                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4107                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4108                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4109                 else
4110                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4111                 break;
4112         case BOND_MODE_ALB:
4113                 bond_set_master_alb_flags(bond);
4114                 /* FALLTHRU */
4115         case BOND_MODE_TLB:
4116                 bond_dev->hard_start_xmit = bond_alb_xmit;
4117                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4118                 break;
4119         default:
4120                 /* Should never happen, mode already checked */
4121                 printk(KERN_ERR DRV_NAME
4122                        ": %s: Error: Unknown bonding mode %d\n",
4123                        bond_dev->name,
4124                        mode);
4125                 break;
4126         }
4127 }
4128
4129 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4130                                     struct ethtool_drvinfo *drvinfo)
4131 {
4132         strncpy(drvinfo->driver, DRV_NAME, 32);
4133         strncpy(drvinfo->version, DRV_VERSION, 32);
4134         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4135 }
4136
4137 static struct ethtool_ops bond_ethtool_ops = {
4138         .get_tx_csum            = ethtool_op_get_tx_csum,
4139         .get_tso                = ethtool_op_get_tso,
4140         .get_ufo                = ethtool_op_get_ufo,
4141         .get_sg                 = ethtool_op_get_sg,
4142         .get_drvinfo            = bond_ethtool_get_drvinfo,
4143 };
4144
4145 /*
4146  * Does not allocate but creates a /proc entry.
4147  * Allowed to fail.
4148  */
4149 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4150 {
4151         struct bonding *bond = bond_dev->priv;
4152
4153         dprintk("Begin bond_init for %s\n", bond_dev->name);
4154
4155         /* initialize rwlocks */
4156         rwlock_init(&bond->lock);
4157         rwlock_init(&bond->curr_slave_lock);
4158
4159         bond->params = *params; /* copy params struct */
4160
4161         /* Initialize pointers */
4162         bond->first_slave = NULL;
4163         bond->curr_active_slave = NULL;
4164         bond->current_arp_slave = NULL;
4165         bond->primary_slave = NULL;
4166         bond->dev = bond_dev;
4167         INIT_LIST_HEAD(&bond->vlan_list);
4168
4169         /* Initialize the device entry points */
4170         bond_dev->open = bond_open;
4171         bond_dev->stop = bond_close;
4172         bond_dev->get_stats = bond_get_stats;
4173         bond_dev->do_ioctl = bond_do_ioctl;
4174         bond_dev->ethtool_ops = &bond_ethtool_ops;
4175         bond_dev->set_multicast_list = bond_set_multicast_list;
4176         bond_dev->change_mtu = bond_change_mtu;
4177         bond_dev->set_mac_address = bond_set_mac_address;
4178
4179         bond_set_mode_ops(bond, bond->params.mode);
4180
4181         bond_dev->destructor = free_netdev;
4182
4183         /* Initialize the device options */
4184         bond_dev->tx_queue_len = 0;
4185         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4186
4187         /* At first, we block adding VLANs. That's the only way to
4188          * prevent problems that occur when adding VLANs over an
4189          * empty bond. The block will be removed once non-challenged
4190          * slaves are enslaved.
4191          */
4192         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4193
4194         /* don't acquire bond device's xmit_lock when
4195          * transmitting */
4196         bond_dev->features |= NETIF_F_LLTX;
4197
4198         /* By default, we declare the bond to be fully
4199          * VLAN hardware accelerated capable. Special
4200          * care is taken in the various xmit functions
4201          * when there are slaves that are not hw accel
4202          * capable
4203          */
4204         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4205         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4206         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4207         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4208                                NETIF_F_HW_VLAN_RX |
4209                                NETIF_F_HW_VLAN_FILTER);
4210
4211 #ifdef CONFIG_PROC_FS
4212         bond_create_proc_entry(bond);
4213 #endif
4214
4215         list_add_tail(&bond->bond_list, &bond_dev_list);
4216
4217         return 0;
4218 }
4219
4220 /* De-initialize device specific data.
4221  * Caller must hold rtnl_lock.
4222  */
4223 void bond_deinit(struct net_device *bond_dev)
4224 {
4225         struct bonding *bond = bond_dev->priv;
4226
4227         list_del(&bond->bond_list);
4228
4229 #ifdef CONFIG_PROC_FS
4230         bond_remove_proc_entry(bond);
4231 #endif
4232 }
4233
4234 /* Unregister and free all bond devices.
4235  * Caller must hold rtnl_lock.
4236  */
4237 static void bond_free_all(void)
4238 {
4239         struct bonding *bond, *nxt;
4240
4241         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4242                 struct net_device *bond_dev = bond->dev;
4243
4244                 unregister_netdevice(bond_dev);
4245                 bond_deinit(bond_dev);
4246         }
4247
4248 #ifdef CONFIG_PROC_FS
4249         bond_destroy_proc_dir();
4250 #endif
4251 }
4252
4253 /*------------------------- Module initialization ---------------------------*/
4254
4255 /*
4256  * Convert string input module parms.  Accept either the
4257  * number of the mode or its string name.
4258  */
4259 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4260 {
4261         int i;
4262
4263         for (i = 0; tbl[i].modename; i++) {
4264                 if ((isdigit(*mode_arg) &&
4265                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4266                     (strncmp(mode_arg, tbl[i].modename,
4267                              strlen(tbl[i].modename)) == 0)) {
4268                         return tbl[i].mode;
4269                 }
4270         }
4271
4272         return -1;
4273 }
4274
4275 static int bond_check_params(struct bond_params *params)
4276 {
4277         /*
4278          * Convert string parameters.
4279          */
4280         if (mode) {
4281                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4282                 if (bond_mode == -1) {
4283                         printk(KERN_ERR DRV_NAME
4284                                ": Error: Invalid bonding mode \"%s\"\n",
4285                                mode == NULL ? "NULL" : mode);
4286                         return -EINVAL;
4287                 }
4288         }
4289
4290         if (xmit_hash_policy) {
4291                 if ((bond_mode != BOND_MODE_XOR) &&
4292                     (bond_mode != BOND_MODE_8023AD)) {
4293                         printk(KERN_INFO DRV_NAME
4294                                ": xor_mode param is irrelevant in mode %s\n",
4295                                bond_mode_name(bond_mode));
4296                 } else {
4297                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4298                                                         xmit_hashtype_tbl);
4299                         if (xmit_hashtype == -1) {
4300                                 printk(KERN_ERR DRV_NAME
4301                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4302                                 xmit_hash_policy == NULL ? "NULL" :
4303                                        xmit_hash_policy);
4304                                 return -EINVAL;
4305                         }
4306                 }
4307         }
4308
4309         if (lacp_rate) {
4310                 if (bond_mode != BOND_MODE_8023AD) {
4311                         printk(KERN_INFO DRV_NAME
4312                                ": lacp_rate param is irrelevant in mode %s\n",
4313                                bond_mode_name(bond_mode));
4314                 } else {
4315                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4316                         if (lacp_fast == -1) {
4317                                 printk(KERN_ERR DRV_NAME
4318                                        ": Error: Invalid lacp rate \"%s\"\n",
4319                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4320                                 return -EINVAL;
4321                         }
4322                 }
4323         }
4324
4325         if (max_bonds < 1 || max_bonds > INT_MAX) {
4326                 printk(KERN_WARNING DRV_NAME
4327                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4328                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4329                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4330                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4331         }
4332
4333         if (miimon < 0) {
4334                 printk(KERN_WARNING DRV_NAME
4335                        ": Warning: miimon module parameter (%d), "
4336                        "not in range 0-%d, so it was reset to %d\n",
4337                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4338                 miimon = BOND_LINK_MON_INTERV;
4339         }
4340
4341         if (updelay < 0) {
4342                 printk(KERN_WARNING DRV_NAME
4343                        ": Warning: updelay module parameter (%d), "
4344                        "not in range 0-%d, so it was reset to 0\n",
4345                        updelay, INT_MAX);
4346                 updelay = 0;
4347         }
4348
4349         if (downdelay < 0) {
4350                 printk(KERN_WARNING DRV_NAME
4351                        ": Warning: downdelay module parameter (%d), "
4352                        "not in range 0-%d, so it was reset to 0\n",
4353                        downdelay, INT_MAX);
4354                 downdelay = 0;
4355         }
4356
4357         if ((use_carrier != 0) && (use_carrier != 1)) {
4358                 printk(KERN_WARNING DRV_NAME
4359                        ": Warning: use_carrier module parameter (%d), "
4360                        "not of valid value (0/1), so it was set to 1\n",
4361                        use_carrier);
4362                 use_carrier = 1;
4363         }
4364
4365         /* reset values for 802.3ad */
4366         if (bond_mode == BOND_MODE_8023AD) {
4367                 if (!miimon) {
4368                         printk(KERN_WARNING DRV_NAME
4369                                ": Warning: miimon must be specified, "
4370                                "otherwise bonding will not detect link "
4371                                "failure, speed and duplex which are "
4372                                "essential for 802.3ad operation\n");
4373                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4374                         miimon = 100;
4375                 }
4376         }
4377
4378         /* reset values for TLB/ALB */
4379         if ((bond_mode == BOND_MODE_TLB) ||
4380             (bond_mode == BOND_MODE_ALB)) {
4381                 if (!miimon) {
4382                         printk(KERN_WARNING DRV_NAME
4383                                ": Warning: miimon must be specified, "
4384                                "otherwise bonding will not detect link "
4385                                "failure and link speed which are essential "
4386                                "for TLB/ALB load balancing\n");
4387                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4388                         miimon = 100;
4389                 }
4390         }
4391
4392         if (bond_mode == BOND_MODE_ALB) {
4393                 printk(KERN_NOTICE DRV_NAME
4394                        ": In ALB mode you might experience client "
4395                        "disconnections upon reconnection of a link if the "
4396                        "bonding module updelay parameter (%d msec) is "
4397                        "incompatible with the forwarding delay time of the "
4398                        "switch\n",
4399                        updelay);
4400         }
4401
4402         if (!miimon) {
4403                 if (updelay || downdelay) {
4404                         /* just warn the user the up/down delay will have
4405                          * no effect since miimon is zero...
4406                          */
4407                         printk(KERN_WARNING DRV_NAME
4408                                ": Warning: miimon module parameter not set "
4409                                "and updelay (%d) or downdelay (%d) module "
4410                                "parameter is set; updelay and downdelay have "
4411                                "no effect unless miimon is set\n",
4412                                updelay, downdelay);
4413                 }
4414         } else {
4415                 /* don't allow arp monitoring */
4416                 if (arp_interval) {
4417                         printk(KERN_WARNING DRV_NAME
4418                                ": Warning: miimon (%d) and arp_interval (%d) "
4419                                "can't be used simultaneously, disabling ARP "
4420                                "monitoring\n",
4421                                miimon, arp_interval);
4422                         arp_interval = 0;
4423                 }
4424
4425                 if ((updelay % miimon) != 0) {
4426                         printk(KERN_WARNING DRV_NAME
4427                                ": Warning: updelay (%d) is not a multiple "
4428                                "of miimon (%d), updelay rounded to %d ms\n",
4429                                updelay, miimon, (updelay / miimon) * miimon);
4430                 }
4431
4432                 updelay /= miimon;
4433
4434                 if ((downdelay % miimon) != 0) {
4435                         printk(KERN_WARNING DRV_NAME
4436                                ": Warning: downdelay (%d) is not a multiple "
4437                                "of miimon (%d), downdelay rounded to %d ms\n",
4438                                downdelay, miimon,
4439                                (downdelay / miimon) * miimon);
4440                 }
4441
4442                 downdelay /= miimon;
4443         }
4444
4445         if (arp_interval < 0) {
4446                 printk(KERN_WARNING DRV_NAME
4447                        ": Warning: arp_interval module parameter (%d) "
4448                        ", not in range 0-%d, so it was reset to %d\n",
4449                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4450                 arp_interval = BOND_LINK_ARP_INTERV;
4451         }
4452
4453         for (arp_ip_count = 0;
4454              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4455              arp_ip_count++) {
4456                 /* not complete check, but should be good enough to
4457                    catch mistakes */
4458                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4459                         printk(KERN_WARNING DRV_NAME
4460                                ": Warning: bad arp_ip_target module parameter "
4461                                "(%s), ARP monitoring will not be performed\n",
4462                                arp_ip_target[arp_ip_count]);
4463                         arp_interval = 0;
4464                 } else {
4465                         u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4466                         arp_target[arp_ip_count] = ip;
4467                 }
4468         }
4469
4470         if (arp_interval && !arp_ip_count) {
4471                 /* don't allow arping if no arp_ip_target given... */
4472                 printk(KERN_WARNING DRV_NAME
4473                        ": Warning: arp_interval module parameter (%d) "
4474                        "specified without providing an arp_ip_target "
4475                        "parameter, arp_interval was reset to 0\n",
4476                        arp_interval);
4477                 arp_interval = 0;
4478         }
4479
4480         if (miimon) {
4481                 printk(KERN_INFO DRV_NAME
4482                        ": MII link monitoring set to %d ms\n",
4483                        miimon);
4484         } else if (arp_interval) {
4485                 int i;
4486
4487                 printk(KERN_INFO DRV_NAME
4488                        ": ARP monitoring set to %d ms with %d target(s):",
4489                        arp_interval, arp_ip_count);
4490
4491                 for (i = 0; i < arp_ip_count; i++)
4492                         printk (" %s", arp_ip_target[i]);
4493
4494                 printk("\n");
4495
4496         } else {
4497                 /* miimon and arp_interval not set, we need one so things
4498                  * work as expected, see bonding.txt for details
4499                  */
4500                 printk(KERN_WARNING DRV_NAME
4501                        ": Warning: either miimon or arp_interval and "
4502                        "arp_ip_target module parameters must be specified, "
4503                        "otherwise bonding will not detect link failures! see "
4504                        "bonding.txt for details.\n");
4505         }
4506
4507         if (primary && !USES_PRIMARY(bond_mode)) {
4508                 /* currently, using a primary only makes sense
4509                  * in active backup, TLB or ALB modes
4510                  */
4511                 printk(KERN_WARNING DRV_NAME
4512                        ": Warning: %s primary device specified but has no "
4513                        "effect in %s mode\n",
4514                        primary, bond_mode_name(bond_mode));
4515                 primary = NULL;
4516         }
4517
4518         /* fill params struct with the proper values */
4519         params->mode = bond_mode;
4520         params->xmit_policy = xmit_hashtype;
4521         params->miimon = miimon;
4522         params->arp_interval = arp_interval;
4523         params->updelay = updelay;
4524         params->downdelay = downdelay;
4525         params->use_carrier = use_carrier;
4526         params->lacp_fast = lacp_fast;
4527         params->primary[0] = 0;
4528
4529         if (primary) {
4530                 strncpy(params->primary, primary, IFNAMSIZ);
4531                 params->primary[IFNAMSIZ - 1] = 0;
4532         }
4533
4534         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4535
4536         return 0;
4537 }
4538
4539 /* Create a new bond based on the specified name and bonding parameters.
4540  * Caller must NOT hold rtnl_lock; we need to release it here before we
4541  * set up our sysfs entries.
4542  */
4543 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4544 {
4545         struct net_device *bond_dev;
4546         int res;
4547
4548         rtnl_lock();
4549         bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4550         if (!bond_dev) {
4551                 printk(KERN_ERR DRV_NAME
4552                        ": %s: eek! can't alloc netdev!\n",
4553                        name);
4554                 res = -ENOMEM;
4555                 goto out_rtnl;
4556         }
4557
4558         /* bond_init() must be called after dev_alloc_name() (for the
4559          * /proc files), but before register_netdevice(), because we
4560          * need to set function pointers.
4561          */
4562
4563         res = bond_init(bond_dev, params);
4564         if (res < 0) {
4565                 goto out_netdev;
4566         }
4567
4568         SET_MODULE_OWNER(bond_dev);
4569
4570         res = register_netdevice(bond_dev);
4571         if (res < 0) {
4572                 goto out_bond;
4573         }
4574         if (newbond)
4575                 *newbond = bond_dev->priv;
4576
4577         netif_carrier_off(bond_dev);
4578
4579         rtnl_unlock(); /* allows sysfs registration of net device */
4580         res = bond_create_sysfs_entry(bond_dev->priv);
4581         goto done;
4582 out_bond:
4583         bond_deinit(bond_dev);
4584 out_netdev:
4585         free_netdev(bond_dev);
4586 out_rtnl:
4587         rtnl_unlock();
4588 done:
4589         return res;
4590 }
4591
4592 static int __init bonding_init(void)
4593 {
4594         int i;
4595         int res;
4596         char new_bond_name[8];  /* Enough room for 999 bonds at init. */
4597
4598         printk(KERN_INFO "%s", version);
4599
4600         res = bond_check_params(&bonding_defaults);
4601         if (res) {
4602                 goto out;
4603         }
4604
4605 #ifdef CONFIG_PROC_FS
4606         bond_create_proc_dir();
4607 #endif
4608         for (i = 0; i < max_bonds; i++) {
4609                 sprintf(new_bond_name, "bond%d",i);
4610                 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4611                 if (res)
4612                         goto err;
4613         }
4614
4615         res = bond_create_sysfs();
4616         if (res)
4617                 goto err;
4618
4619         register_netdevice_notifier(&bond_netdev_notifier);
4620         register_inetaddr_notifier(&bond_inetaddr_notifier);
4621
4622         goto out;
4623 err:
4624         rtnl_lock();
4625         bond_free_all();
4626         bond_destroy_sysfs();
4627         rtnl_unlock();
4628 out:
4629         return res;
4630
4631 }
4632
4633 static void __exit bonding_exit(void)
4634 {
4635         unregister_netdevice_notifier(&bond_netdev_notifier);
4636         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4637
4638         rtnl_lock();
4639         bond_free_all();
4640         bond_destroy_sysfs();
4641         rtnl_unlock();
4642 }
4643
4644 module_init(bonding_init);
4645 module_exit(bonding_exit);
4646 MODULE_LICENSE("GPL");
4647 MODULE_VERSION(DRV_VERSION);
4648 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4649 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4650 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4651
4652 /*
4653  * Local variables:
4654  *  c-indent-level: 8
4655  *  c-basic-offset: 8
4656  *  tab-width: 8
4657  * End:
4658  */
4659