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/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <asm/system.h>
  57 #include <asm/io.h>
  58 #include <asm/dma.h>
  59 #include <asm/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.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 static char *arp_validate = NULL;
 101 static int fail_over_mac = 0;
 102 struct bond_params bonding_defaults;
 103
 104 module_param(max_bonds, int, 0);
 105 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 106 module_param(miimon, int, 0);
 107 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 108 module_param(updelay, int, 0);
 109 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 110 module_param(downdelay, int, 0);
 111 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 112                             "in milliseconds");
 113 module_param(use_carrier, int, 0);
 114 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 115                               "0 for off, 1 for on (default)");
 116 module_param(mode, charp, 0);
 117 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 118                        "1 for active-backup, 2 for balance-xor, "
 119                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 120                        "6 for balance-alb");
 121 module_param(primary, charp, 0);
 122 MODULE_PARM_DESC(primary, "Primary network device to use");
 123 module_param(lacp_rate, charp, 0);
 124 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 125                             "(slow/fast)");
 126 module_param(xmit_hash_policy, charp, 0);
 127 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 128                                    ", 1 for layer 3+4");
 129 module_param(arp_interval, int, 0);
 130 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 131 module_param_array(arp_ip_target, charp, NULL, 0);
 132 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 133 module_param(arp_validate, charp, 0);
 134 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 135 module_param(fail_over_mac, int, 0);
 136 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  0 of off (default), 1 for on.");
 137
 138 /*----------------------------- Global variables ----------------------------*/
 139
 140 static const char * const version =
 141         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 142
 143 LIST_HEAD(bond_dev_list);
 144
 145 #ifdef CONFIG_PROC_FS
 146 static struct proc_dir_entry *bond_proc_dir = NULL;
 147 #endif
 148
 149 extern struct rw_semaphore bonding_rwsem;
 150 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 151 static int arp_ip_count = 0;
 152 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 153 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 154 static int lacp_fast    = 0;
 155
 156
 157 struct bond_parm_tbl bond_lacp_tbl[] = {
 158 {       "slow",         AD_LACP_SLOW},
 159 {       "fast",         AD_LACP_FAST},
 160 {       NULL,           -1},
 161 };
 162
 163 struct bond_parm_tbl bond_mode_tbl[] = {
 164 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 165 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 166 {       "balance-xor",          BOND_MODE_XOR},
 167 {       "broadcast",            BOND_MODE_BROADCAST},
 168 {       "802.3ad",              BOND_MODE_8023AD},
 169 {       "balance-tlb",          BOND_MODE_TLB},
 170 {       "balance-alb",          BOND_MODE_ALB},
 171 {       NULL,                   -1},
 172 };
 173
 174 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 175 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 176 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 177 {       NULL,                   -1},
 178 };
 179
 180 struct bond_parm_tbl arp_validate_tbl[] = {
 181 {       "none",                 BOND_ARP_VALIDATE_NONE},
 182 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 183 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 184 {       "all",                  BOND_ARP_VALIDATE_ALL},
 185 {       NULL,                   -1},
 186 };
 187
 188 /*-------------------------- Forward declarations ---------------------------*/
 189
 190 static void bond_send_gratuitous_arp(struct bonding *bond);
 191
 192 /*---------------------------- General routines -----------------------------*/
 193
 194 static const char *bond_mode_name(int mode)
 195 {
 196         switch (mode) {
 197         case BOND_MODE_ROUNDROBIN :
 198                 return "load balancing (round-robin)";
 199         case BOND_MODE_ACTIVEBACKUP :
 200                 return "fault-tolerance (active-backup)";
 201         case BOND_MODE_XOR :
 202                 return "load balancing (xor)";
 203         case BOND_MODE_BROADCAST :
 204                 return "fault-tolerance (broadcast)";
 205         case BOND_MODE_8023AD:
 206                 return "IEEE 802.3ad Dynamic link aggregation";
 207         case BOND_MODE_TLB:
 208                 return "transmit load balancing";
 209         case BOND_MODE_ALB:
 210                 return "adaptive load balancing";
 211         default:
 212                 return "unknown";
 213         }
 214 }
 215
 216 /*---------------------------------- VLAN -----------------------------------*/
 217
 218 /**
 219  * bond_add_vlan - add a new vlan id on bond
 220  * @bond: bond that got the notification
 221  * @vlan_id: the vlan id to add
 222  *
 223  * Returns -ENOMEM if allocation failed.
 224  */
 225 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 226 {
 227         struct vlan_entry *vlan;
 228
 229         dprintk("bond: %s, vlan id %d\n",
 230                 (bond ? bond->dev->name: "None"), vlan_id);
 231
 232         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 233         if (!vlan) {
 234                 return -ENOMEM;
 235         }
 236
 237         INIT_LIST_HEAD(&vlan->vlan_list);
 238         vlan->vlan_id = vlan_id;
 239         vlan->vlan_ip = 0;
 240
 241         write_lock_bh(&bond->lock);
 242
 243         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 244
 245         write_unlock_bh(&bond->lock);
 246
 247         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 248
 249         return 0;
 250 }
 251
 252 /**
 253  * bond_del_vlan - delete a vlan id from bond
 254  * @bond: bond that got the notification
 255  * @vlan_id: the vlan id to delete
 256  *
 257  * returns -ENODEV if @vlan_id was not found in @bond.
 258  */
 259 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 260 {
 261         struct vlan_entry *vlan, *next;
 262         int res = -ENODEV;
 263
 264         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 265
 266         write_lock_bh(&bond->lock);
 267
 268         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 269                 if (vlan->vlan_id == vlan_id) {
 270                         list_del(&vlan->vlan_list);
 271
 272                         if ((bond->params.mode == BOND_MODE_TLB) ||
 273                             (bond->params.mode == BOND_MODE_ALB)) {
 274                                 bond_alb_clear_vlan(bond, vlan_id);
 275                         }
 276
 277                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 278                                 bond->dev->name);
 279
 280                         kfree(vlan);
 281
 282                         if (list_empty(&bond->vlan_list) &&
 283                             (bond->slave_cnt == 0)) {
 284                                 /* Last VLAN removed and no slaves, so
 285                                  * restore block on adding VLANs. This will
 286                                  * be removed once new slaves that are not
 287                                  * VLAN challenged will be added.
 288                                  */
 289                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 290                         }
 291
 292                         res = 0;
 293                         goto out;
 294                 }
 295         }
 296
 297         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 298                 bond->dev->name);
 299
 300 out:
 301         write_unlock_bh(&bond->lock);
 302         return res;
 303 }
 304
 305 /**
 306  * bond_has_challenged_slaves
 307  * @bond: the bond we're working on
 308  *
 309  * Searches the slave list. Returns 1 if a vlan challenged slave
 310  * was found, 0 otherwise.
 311  *
 312  * Assumes bond->lock is held.
 313  */
 314 static int bond_has_challenged_slaves(struct bonding *bond)
 315 {
 316         struct slave *slave;
 317         int i;
 318
 319         bond_for_each_slave(bond, slave, i) {
 320                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 321                         dprintk("found VLAN challenged slave - %s\n",
 322                                 slave->dev->name);
 323                         return 1;
 324                 }
 325         }
 326
 327         dprintk("no VLAN challenged slaves found\n");
 328         return 0;
 329 }
 330
 331 /**
 332  * bond_next_vlan - safely skip to the next item in the vlans list.
 333  * @bond: the bond we're working on
 334  * @curr: item we're advancing from
 335  *
 336  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 337  * or @curr->next otherwise (even if it is @curr itself again).
 338  *
 339  * Caller must hold bond->lock
 340  */
 341 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 342 {
 343         struct vlan_entry *next, *last;
 344
 345         if (list_empty(&bond->vlan_list)) {
 346                 return NULL;
 347         }
 348
 349         if (!curr) {
 350                 next = list_entry(bond->vlan_list.next,
 351                                   struct vlan_entry, vlan_list);
 352         } else {
 353                 last = list_entry(bond->vlan_list.prev,
 354                                   struct vlan_entry, vlan_list);
 355                 if (last == curr) {
 356                         next = list_entry(bond->vlan_list.next,
 357                                           struct vlan_entry, vlan_list);
 358                 } else {
 359                         next = list_entry(curr->vlan_list.next,
 360                                           struct vlan_entry, vlan_list);
 361                 }
 362         }
 363
 364         return next;
 365 }
 366
 367 /**
 368  * bond_dev_queue_xmit - Prepare skb for xmit.
 369  *
 370  * @bond: bond device that got this skb for tx.
 371  * @skb: hw accel VLAN tagged skb to transmit
 372  * @slave_dev: slave that is supposed to xmit this skbuff
 373  *
 374  * When the bond gets an skb to transmit that is
 375  * already hardware accelerated VLAN tagged, and it
 376  * needs to relay this skb to a slave that is not
 377  * hw accel capable, the skb needs to be "unaccelerated",
 378  * i.e. strip the hwaccel tag and re-insert it as part
 379  * of the payload.
 380  */
 381 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 382 {
 383         unsigned short vlan_id;
 384
 385         if (!list_empty(&bond->vlan_list) &&
 386             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 387             vlan_get_tag(skb, &vlan_id) == 0) {
 388                 skb->dev = slave_dev;
 389                 skb = vlan_put_tag(skb, vlan_id);
 390                 if (!skb) {
 391                         /* vlan_put_tag() frees the skb in case of error,
 392                          * so return success here so the calling functions
 393                          * won't attempt to free is again.
 394                          */
 395                         return 0;
 396                 }
 397         } else {
 398                 skb->dev = slave_dev;
 399         }
 400
 401         skb->priority = 1;
 402         dev_queue_xmit(skb);
 403
 404         return 0;
 405 }
 406
 407 /*
 408  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 409  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 410  * lock because:
 411  * a. This operation is performed in IOCTL context,
 412  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 413  * c. Holding a lock with BH disabled while directly calling a base driver
 414  *    entry point is generally a BAD idea.
 415  *
 416  * The design of synchronization/protection for this operation in the 8021q
 417  * module is good for one or more VLAN devices over a single physical device
 418  * and cannot be extended for a teaming solution like bonding, so there is a
 419  * potential race condition here where a net device from the vlan group might
 420  * be referenced (either by a base driver or the 8021q code) while it is being
 421  * removed from the system. However, it turns out we're not making matters
 422  * worse, and if it works for regular VLAN usage it will work here too.
 423 */
 424
 425 /**
 426  * bond_vlan_rx_register - Propagates registration to slaves
 427  * @bond_dev: bonding net device that got called
 428  * @grp: vlan group being registered
 429  */
 430 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 431 {
 432         struct bonding *bond = bond_dev->priv;
 433         struct slave *slave;
 434         int i;
 435
 436         bond->vlgrp = grp;
 437
 438         bond_for_each_slave(bond, slave, i) {
 439                 struct net_device *slave_dev = slave->dev;
 440
 441                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 442                     slave_dev->vlan_rx_register) {
 443                         slave_dev->vlan_rx_register(slave_dev, grp);
 444                 }
 445         }
 446 }
 447
 448 /**
 449  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 450  * @bond_dev: bonding net device that got called
 451  * @vid: vlan id being added
 452  */
 453 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 454 {
 455         struct bonding *bond = bond_dev->priv;
 456         struct slave *slave;
 457         int i, res;
 458
 459         bond_for_each_slave(bond, slave, i) {
 460                 struct net_device *slave_dev = slave->dev;
 461
 462                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 463                     slave_dev->vlan_rx_add_vid) {
 464                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 465                 }
 466         }
 467
 468         res = bond_add_vlan(bond, vid);
 469         if (res) {
 470                 printk(KERN_ERR DRV_NAME
 471                        ": %s: Error: Failed to add vlan id %d\n",
 472                        bond_dev->name, vid);
 473         }
 474 }
 475
 476 /**
 477  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 478  * @bond_dev: bonding net device that got called
 479  * @vid: vlan id being removed
 480  */
 481 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 482 {
 483         struct bonding *bond = bond_dev->priv;
 484         struct slave *slave;
 485         struct net_device *vlan_dev;
 486         int i, res;
 487
 488         bond_for_each_slave(bond, slave, i) {
 489                 struct net_device *slave_dev = slave->dev;
 490
 491                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 492                     slave_dev->vlan_rx_kill_vid) {
 493                         /* Save and then restore vlan_dev in the grp array,
 494                          * since the slave's driver might clear it.
 495                          */
 496                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 497                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 498                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 499                 }
 500         }
 501
 502         res = bond_del_vlan(bond, vid);
 503         if (res) {
 504                 printk(KERN_ERR DRV_NAME
 505                        ": %s: Error: Failed to remove vlan id %d\n",
 506                        bond_dev->name, vid);
 507         }
 508 }
 509
 510 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 511 {
 512         struct vlan_entry *vlan;
 513
 514         write_lock_bh(&bond->lock);
 515
 516         if (list_empty(&bond->vlan_list)) {
 517                 goto out;
 518         }
 519
 520         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 521             slave_dev->vlan_rx_register) {
 522                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 523         }
 524
 525         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 526             !(slave_dev->vlan_rx_add_vid)) {
 527                 goto out;
 528         }
 529
 530         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 531                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 532         }
 533
 534 out:
 535         write_unlock_bh(&bond->lock);
 536 }
 537
 538 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 539 {
 540         struct vlan_entry *vlan;
 541         struct net_device *vlan_dev;
 542
 543         write_lock_bh(&bond->lock);
 544
 545         if (list_empty(&bond->vlan_list)) {
 546                 goto out;
 547         }
 548
 549         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 550             !(slave_dev->vlan_rx_kill_vid)) {
 551                 goto unreg;
 552         }
 553
 554         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 555                 /* Save and then restore vlan_dev in the grp array,
 556                  * since the slave's driver might clear it.
 557                  */
 558                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 559                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 560                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 561         }
 562
 563 unreg:
 564         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 565             slave_dev->vlan_rx_register) {
 566                 slave_dev->vlan_rx_register(slave_dev, NULL);
 567         }
 568
 569 out:
 570         write_unlock_bh(&bond->lock);
 571 }
 572
 573 /*------------------------------- Link status -------------------------------*/
 574
 575 /*
 576  * Set the carrier state for the master according to the state of its
 577  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 578  * do special 802.3ad magic.
 579  *
 580  * Returns zero if carrier state does not change, nonzero if it does.
 581  */
 582 static int bond_set_carrier(struct bonding *bond)
 583 {
 584         struct slave *slave;
 585         int i;
 586
 587         if (bond->slave_cnt == 0)
 588                 goto down;
 589
 590         if (bond->params.mode == BOND_MODE_8023AD)
 591                 return bond_3ad_set_carrier(bond);
 592
 593         bond_for_each_slave(bond, slave, i) {
 594                 if (slave->link == BOND_LINK_UP) {
 595                         if (!netif_carrier_ok(bond->dev)) {
 596                                 netif_carrier_on(bond->dev);
 597                                 return 1;
 598                         }
 599                         return 0;
 600                 }
 601         }
 602
 603 down:
 604         if (netif_carrier_ok(bond->dev)) {
 605                 netif_carrier_off(bond->dev);
 606                 return 1;
 607         }
 608         return 0;
 609 }
 610
 611 /*
 612  * Get link speed and duplex from the slave's base driver
 613  * using ethtool. If for some reason the call fails or the
 614  * values are invalid, fake speed and duplex to 100/Full
 615  * and return error.
 616  */
 617 static int bond_update_speed_duplex(struct slave *slave)
 618 {
 619         struct net_device *slave_dev = slave->dev;
 620         struct ethtool_cmd etool;
 621         int res;
 622
 623         /* Fake speed and duplex */
 624         slave->speed = SPEED_100;
 625         slave->duplex = DUPLEX_FULL;
 626
 627         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 628                 return -1;
 629
 630         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 631         if (res < 0)
 632                 return -1;
 633
 634         switch (etool.speed) {
 635         case SPEED_10:
 636         case SPEED_100:
 637         case SPEED_1000:
 638         case SPEED_10000:
 639                 break;
 640         default:
 641                 return -1;
 642         }
 643
 644         switch (etool.duplex) {
 645         case DUPLEX_FULL:
 646         case DUPLEX_HALF:
 647                 break;
 648         default:
 649                 return -1;
 650         }
 651
 652         slave->speed = etool.speed;
 653         slave->duplex = etool.duplex;
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * if <dev> supports MII link status reporting, check its link status.
 660  *
 661  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 662  * depening upon the setting of the use_carrier parameter.
 663  *
 664  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 665  * can't tell and just pretend it is), or 0, meaning that the link is
 666  * down.
 667  *
 668  * If reporting is non-zero, instead of faking link up, return -1 if
 669  * both ETHTOOL and MII ioctls fail (meaning the device does not
 670  * support them).  If use_carrier is set, return whatever it says.
 671  * It'd be nice if there was a good way to tell if a driver supports
 672  * netif_carrier, but there really isn't.
 673  */
 674 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 675 {
 676         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 677         struct ifreq ifr;
 678         struct mii_ioctl_data *mii;
 679
 680         if (bond->params.use_carrier) {
 681                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 682         }
 683
 684         ioctl = slave_dev->do_ioctl;
 685         if (ioctl) {
 686                 /* TODO: set pointer to correct ioctl on a per team member */
 687                 /*       bases to make this more efficient. that is, once  */
 688                 /*       we determine the correct ioctl, we will always    */
 689                 /*       call it and not the others for that team          */
 690                 /*       member.                                           */
 691
 692                 /*
 693                  * We cannot assume that SIOCGMIIPHY will also read a
 694                  * register; not all network drivers (e.g., e100)
 695                  * support that.
 696                  */
 697
 698                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 699                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 700                 mii = if_mii(&ifr);
 701                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 702                         mii->reg_num = MII_BMSR;
 703                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 704                                 return (mii->val_out & BMSR_LSTATUS);
 705                         }
 706                 }
 707         }
 708
 709         /*
 710          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 711          * attempt to get link status from it if the above MII ioctls fail.
 712          */
 713         if (slave_dev->ethtool_ops) {
 714                 if (slave_dev->ethtool_ops->get_link) {
 715                         u32 link;
 716
 717                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 718
 719                         return link ? BMSR_LSTATUS : 0;
 720                 }
 721         }
 722
 723         /*
 724          * If reporting, report that either there's no dev->do_ioctl,
 725          * or both SIOCGMIIREG and get_link failed (meaning that we
 726          * cannot report link status).  If not reporting, pretend
 727          * we're ok.
 728          */
 729         return (reporting ? -1 : BMSR_LSTATUS);
 730 }
 731
 732 /*----------------------------- Multicast list ------------------------------*/
 733
 734 /*
 735  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 736  */
 737 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 738 {
 739         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 740                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 741 }
 742
 743 /*
 744  * returns dmi entry if found, NULL otherwise
 745  */
 746 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 747 {
 748         struct dev_mc_list *idmi;
 749
 750         for (idmi = mc_list; idmi; idmi = idmi->next) {
 751                 if (bond_is_dmi_same(dmi, idmi)) {
 752                         return idmi;
 753                 }
 754         }
 755
 756         return NULL;
 757 }
 758
 759 /*
 760  * Push the promiscuity flag down to appropriate slaves
 761  */
 762 static void bond_set_promiscuity(struct bonding *bond, int inc)
 763 {
 764         if (USES_PRIMARY(bond->params.mode)) {
 765                 /* write lock already acquired */
 766                 if (bond->curr_active_slave) {
 767                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 768                 }
 769         } else {
 770                 struct slave *slave;
 771                 int i;
 772                 bond_for_each_slave(bond, slave, i) {
 773                         dev_set_promiscuity(slave->dev, inc);
 774                 }
 775         }
 776 }
 777
 778 /*
 779  * Push the allmulti flag down to all slaves
 780  */
 781 static void bond_set_allmulti(struct bonding *bond, int inc)
 782 {
 783         if (USES_PRIMARY(bond->params.mode)) {
 784                 /* write lock already acquired */
 785                 if (bond->curr_active_slave) {
 786                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 787                 }
 788         } else {
 789                 struct slave *slave;
 790                 int i;
 791                 bond_for_each_slave(bond, slave, i) {
 792                         dev_set_allmulti(slave->dev, inc);
 793                 }
 794         }
 795 }
 796
 797 /*
 798  * Add a Multicast address to slaves
 799  * according to mode
 800  */
 801 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 802 {
 803         if (USES_PRIMARY(bond->params.mode)) {
 804                 /* write lock already acquired */
 805                 if (bond->curr_active_slave) {
 806                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 807                 }
 808         } else {
 809                 struct slave *slave;
 810                 int i;
 811                 bond_for_each_slave(bond, slave, i) {
 812                         dev_mc_add(slave->dev, addr, alen, 0);
 813                 }
 814         }
 815 }
 816
 817 /*
 818  * Remove a multicast address from slave
 819  * according to mode
 820  */
 821 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 822 {
 823         if (USES_PRIMARY(bond->params.mode)) {
 824                 /* write lock already acquired */
 825                 if (bond->curr_active_slave) {
 826                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 827                 }
 828         } else {
 829                 struct slave *slave;
 830                 int i;
 831                 bond_for_each_slave(bond, slave, i) {
 832                         dev_mc_delete(slave->dev, addr, alen, 0);
 833                 }
 834         }
 835 }
 836
 837
 838 /*
 839  * Retrieve the list of registered multicast addresses for the bonding
 840  * device and retransmit an IGMP JOIN request to the current active
 841  * slave.
 842  */
 843 static void bond_resend_igmp_join_requests(struct bonding *bond)
 844 {
 845         struct in_device *in_dev;
 846         struct ip_mc_list *im;
 847
 848         rcu_read_lock();
 849         in_dev = __in_dev_get_rcu(bond->dev);
 850         if (in_dev) {
 851                 for (im = in_dev->mc_list; im; im = im->next) {
 852                         ip_mc_rejoin_group(im);
 853                 }
 854         }
 855
 856         rcu_read_unlock();
 857 }
 858
 859 /*
 860  * Totally destroys the mc_list in bond
 861  */
 862 static void bond_mc_list_destroy(struct bonding *bond)
 863 {
 864         struct dev_mc_list *dmi;
 865
 866         dmi = bond->mc_list;
 867         while (dmi) {
 868                 bond->mc_list = dmi->next;
 869                 kfree(dmi);
 870                 dmi = bond->mc_list;
 871         }
 872         bond->mc_list = NULL;
 873 }
 874
 875 /*
 876  * Copy all the Multicast addresses from src to the bonding device dst
 877  */
 878 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 879                              gfp_t gfp_flag)
 880 {
 881         struct dev_mc_list *dmi, *new_dmi;
 882
 883         for (dmi = mc_list; dmi; dmi = dmi->next) {
 884                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 885
 886                 if (!new_dmi) {
 887                         /* FIXME: Potential memory leak !!! */
 888                         return -ENOMEM;
 889                 }
 890
 891                 new_dmi->next = bond->mc_list;
 892                 bond->mc_list = new_dmi;
 893                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 894                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 895                 new_dmi->dmi_users = dmi->dmi_users;
 896                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 897         }
 898
 899         return 0;
 900 }
 901
 902 /*
 903  * flush all members of flush->mc_list from device dev->mc_list
 904  */
 905 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 906 {
 907         struct bonding *bond = bond_dev->priv;
 908         struct dev_mc_list *dmi;
 909
 910         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 911                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 912         }
 913
 914         if (bond->params.mode == BOND_MODE_8023AD) {
 915                 /* del lacpdu mc addr from mc list */
 916                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 917
 918                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 919         }
 920 }
 921
 922 /*--------------------------- Active slave change ---------------------------*/
 923
 924 /*
 925  * Update the mc list and multicast-related flags for the new and
 926  * old active slaves (if any) according to the multicast mode, and
 927  * promiscuous flags unconditionally.
 928  */
 929 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 930 {
 931         struct dev_mc_list *dmi;
 932
 933         if (!USES_PRIMARY(bond->params.mode)) {
 934                 /* nothing to do -  mc list is already up-to-date on
 935                  * all slaves
 936                  */
 937                 return;
 938         }
 939
 940         if (old_active) {
 941                 if (bond->dev->flags & IFF_PROMISC) {
 942                         dev_set_promiscuity(old_active->dev, -1);
 943                 }
 944
 945                 if (bond->dev->flags & IFF_ALLMULTI) {
 946                         dev_set_allmulti(old_active->dev, -1);
 947                 }
 948
 949                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 950                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 951                 }
 952         }
 953
 954         if (new_active) {
 955                 if (bond->dev->flags & IFF_PROMISC) {
 956                         dev_set_promiscuity(new_active->dev, 1);
 957                 }
 958
 959                 if (bond->dev->flags & IFF_ALLMULTI) {
 960                         dev_set_allmulti(new_active->dev, 1);
 961                 }
 962
 963                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 964                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 965                 }
 966                 bond_resend_igmp_join_requests(bond);
 967         }
 968 }
 969
 970 /**
 971  * find_best_interface - select the best available slave to be the active one
 972  * @bond: our bonding struct
 973  *
 974  * Warning: Caller must hold curr_slave_lock for writing.
 975  */
 976 static struct slave *bond_find_best_slave(struct bonding *bond)
 977 {
 978         struct slave *new_active, *old_active;
 979         struct slave *bestslave = NULL;
 980         int mintime = bond->params.updelay;
 981         int i;
 982
 983         new_active = old_active = bond->curr_active_slave;
 984
 985         if (!new_active) { /* there were no active slaves left */
 986                 if (bond->slave_cnt > 0) {  /* found one slave */
 987                         new_active = bond->first_slave;
 988                 } else {
 989                         return NULL; /* still no slave, return NULL */
 990                 }
 991         }
 992
 993         /* first try the primary link; if arping, a link must tx/rx traffic
 994          * before it can be considered the curr_active_slave - also, we would skip
 995          * slaves between the curr_active_slave and primary_slave that may be up
 996          * and able to arp
 997          */
 998         if ((bond->primary_slave) &&
 999             (!bond->params.arp_interval) &&
1000             (IS_UP(bond->primary_slave->dev))) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (IS_UP(new_active->dev)) {
1009                         if (new_active->link == BOND_LINK_UP) {
1010                                 return new_active;
1011                         } else if (new_active->link == BOND_LINK_BACK) {
1012                                 /* link up, but waiting for stabilization */
1013                                 if (new_active->delay < mintime) {
1014                                         mintime = new_active->delay;
1015                                         bestslave = new_active;
1016                                 }
1017                         }
1018                 }
1019         }
1020
1021         return bestslave;
1022 }
1023
1024 /**
1025  * change_active_interface - change the active slave into the specified one
1026  * @bond: our bonding struct
1027  * @new: the new slave to make the active one
1028  *
1029  * Set the new slave to the bond's settings and unset them on the old
1030  * curr_active_slave.
1031  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1032  *
1033  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1034  * because it is apparently the best available slave we have, even though its
1035  * updelay hasn't timed out yet.
1036  *
1037  * Warning: Caller must hold curr_slave_lock for writing.
1038  */
1039 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1040 {
1041         struct slave *old_active = bond->curr_active_slave;
1042
1043         if (old_active == new_active) {
1044                 return;
1045         }
1046
1047         if (new_active) {
1048                 if (new_active->link == BOND_LINK_BACK) {
1049                         if (USES_PRIMARY(bond->params.mode)) {
1050                                 printk(KERN_INFO DRV_NAME
1051                                        ": %s: making interface %s the new "
1052                                        "active one %d ms earlier.\n",
1053                                        bond->dev->name, new_active->dev->name,
1054                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1055                         }
1056
1057                         new_active->delay = 0;
1058                         new_active->link = BOND_LINK_UP;
1059                         new_active->jiffies = jiffies;
1060
1061                         if (bond->params.mode == BOND_MODE_8023AD) {
1062                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1063                         }
1064
1065                         if ((bond->params.mode == BOND_MODE_TLB) ||
1066                             (bond->params.mode == BOND_MODE_ALB)) {
1067                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1068                         }
1069                 } else {
1070                         if (USES_PRIMARY(bond->params.mode)) {
1071                                 printk(KERN_INFO DRV_NAME
1072                                        ": %s: making interface %s the new "
1073                                        "active one.\n",
1074                                        bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode)) {
1080                 bond_mc_swap(bond, new_active, old_active);
1081         }
1082
1083         if ((bond->params.mode == BOND_MODE_TLB) ||
1084             (bond->params.mode == BOND_MODE_ALB)) {
1085                 bond_alb_handle_active_change(bond, new_active);
1086                 if (old_active)
1087                         bond_set_slave_inactive_flags(old_active);
1088                 if (new_active)
1089                         bond_set_slave_active_flags(new_active);
1090         } else {
1091                 bond->curr_active_slave = new_active;
1092         }
1093
1094         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1095                 if (old_active) {
1096                         bond_set_slave_inactive_flags(old_active);
1097                 }
1098
1099                 if (new_active) {
1100                         bond_set_slave_active_flags(new_active);
1101                 }
1102
1103                 /* when bonding does not set the slave MAC address, the bond MAC
1104                  * address is the one of the active slave.
1105                  */
1106                 if (new_active && bond->params.fail_over_mac)
1107                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1108                                 new_active->dev->addr_len);
1109                 if (bond->curr_active_slave &&
1110                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1111                                         &bond->curr_active_slave->dev->state)) {
1112                         dprintk("delaying gratuitous arp on %s\n",
1113                                 bond->curr_active_slave->dev->name);
1114                         bond->send_grat_arp = 1;
1115                 } else
1116                         bond_send_gratuitous_arp(bond);
1117         }
1118 }
1119
1120 /**
1121  * bond_select_active_slave - select a new active slave, if needed
1122  * @bond: our bonding struct
1123  *
1124  * This functions shoud be called when one of the following occurs:
1125  * - The old curr_active_slave has been released or lost its link.
1126  * - The primary_slave has got its link back.
1127  * - A slave has got its link back and there's no old curr_active_slave.
1128  *
1129  * Warning: Caller must hold curr_slave_lock for writing.
1130  */
1131 void bond_select_active_slave(struct bonding *bond)
1132 {
1133         struct slave *best_slave;
1134         int rv;
1135
1136         best_slave = bond_find_best_slave(bond);
1137         if (best_slave != bond->curr_active_slave) {
1138                 bond_change_active_slave(bond, best_slave);
1139                 rv = bond_set_carrier(bond);
1140                 if (!rv)
1141                         return;
1142
1143                 if (netif_carrier_ok(bond->dev)) {
1144                         printk(KERN_INFO DRV_NAME
1145                                ": %s: first active interface up!\n",
1146                                bond->dev->name);
1147                 } else {
1148                         printk(KERN_INFO DRV_NAME ": %s: "
1149                                "now running without any active interface !\n",
1150                                bond->dev->name);
1151                 }
1152         }
1153 }
1154
1155 /*--------------------------- slave list handling ---------------------------*/
1156
1157 /*
1158  * This function attaches the slave to the end of list.
1159  *
1160  * bond->lock held for writing by caller.
1161  */
1162 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1163 {
1164         if (bond->first_slave == NULL) { /* attaching the first slave */
1165                 new_slave->next = new_slave;
1166                 new_slave->prev = new_slave;
1167                 bond->first_slave = new_slave;
1168         } else {
1169                 new_slave->next = bond->first_slave;
1170                 new_slave->prev = bond->first_slave->prev;
1171                 new_slave->next->prev = new_slave;
1172                 new_slave->prev->next = new_slave;
1173         }
1174
1175         bond->slave_cnt++;
1176 }
1177
1178 /*
1179  * This function detaches the slave from the list.
1180  * WARNING: no check is made to verify if the slave effectively
1181  * belongs to <bond>.
1182  * Nothing is freed on return, structures are just unchained.
1183  * If any slave pointer in bond was pointing to <slave>,
1184  * it should be changed by the calling function.
1185  *
1186  * bond->lock held for writing by caller.
1187  */
1188 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1189 {
1190         if (slave->next) {
1191                 slave->next->prev = slave->prev;
1192         }
1193
1194         if (slave->prev) {
1195                 slave->prev->next = slave->next;
1196         }
1197
1198         if (bond->first_slave == slave) { /* slave is the first slave */
1199                 if (bond->slave_cnt > 1) { /* there are more slave */
1200                         bond->first_slave = slave->next;
1201                 } else {
1202                         bond->first_slave = NULL; /* slave was the last one */
1203                 }
1204         }
1205
1206         slave->next = NULL;
1207         slave->prev = NULL;
1208         bond->slave_cnt--;
1209 }
1210
1211 /*---------------------------------- IOCTL ----------------------------------*/
1212
1213 static int bond_sethwaddr(struct net_device *bond_dev,
1214                           struct net_device *slave_dev)
1215 {
1216         dprintk("bond_dev=%p\n", bond_dev);
1217         dprintk("slave_dev=%p\n", slave_dev);
1218         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1219         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1220         return 0;
1221 }
1222
1223 #define BOND_VLAN_FEATURES \
1224         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1225          NETIF_F_HW_VLAN_FILTER)
1226
1227 /*
1228  * Compute the common dev->feature set available to all slaves.  Some
1229  * feature bits are managed elsewhere, so preserve those feature bits
1230  * on the master device.
1231  */
1232 static int bond_compute_features(struct bonding *bond)
1233 {
1234         struct slave *slave;
1235         struct net_device *bond_dev = bond->dev;
1236         unsigned long features = bond_dev->features;
1237         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1238                                                 bond_dev->hard_header_len);
1239         int i;
1240
1241         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1242         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1243                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1244
1245         bond_for_each_slave(bond, slave, i) {
1246                 features = netdev_compute_features(features,
1247                                                    slave->dev->features);
1248                 if (slave->dev->hard_header_len > max_hard_header_len)
1249                         max_hard_header_len = slave->dev->hard_header_len;
1250         }
1251
1252         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1253         bond_dev->features = features;
1254         bond_dev->hard_header_len = max_hard_header_len;
1255
1256         return 0;
1257 }
1258
1259
1260 static void bond_setup_by_slave(struct net_device *bond_dev,
1261                                 struct net_device *slave_dev)
1262 {
1263         struct bonding *bond = bond_dev->priv;
1264
1265         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1266         bond_dev->header_ops            = slave_dev->header_ops;
1267
1268         bond_dev->type              = slave_dev->type;
1269         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1270         bond_dev->addr_len          = slave_dev->addr_len;
1271
1272         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1273                 slave_dev->addr_len);
1274         bond->setup_by_slave = 1;
1275 }
1276
1277 /* enslave device <slave> to bond device <master> */
1278 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1279 {
1280         struct bonding *bond = bond_dev->priv;
1281         struct slave *new_slave = NULL;
1282         struct dev_mc_list *dmi;
1283         struct sockaddr addr;
1284         int link_reporting;
1285         int old_features = bond_dev->features;
1286         int res = 0;
1287
1288         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1289                 slave_dev->do_ioctl == NULL) {
1290                 printk(KERN_WARNING DRV_NAME
1291                        ": %s: Warning: no link monitoring support for %s\n",
1292                        bond_dev->name, slave_dev->name);
1293         }
1294
1295         /* bond must be initialized by bond_open() before enslaving */
1296         if (!(bond_dev->flags & IFF_UP)) {
1297                 printk(KERN_WARNING DRV_NAME
1298                         " %s: master_dev is not up in bond_enslave\n",
1299                         bond_dev->name);
1300         }
1301
1302         /* already enslaved */
1303         if (slave_dev->flags & IFF_SLAVE) {
1304                 dprintk("Error, Device was already enslaved\n");
1305                 return -EBUSY;
1306         }
1307
1308         /* vlan challenged mutual exclusion */
1309         /* no need to lock since we're protected by rtnl_lock */
1310         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1311                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1312                 if (!list_empty(&bond->vlan_list)) {
1313                         printk(KERN_ERR DRV_NAME
1314                                ": %s: Error: cannot enslave VLAN "
1315                                "challenged slave %s on VLAN enabled "
1316                                "bond %s\n", bond_dev->name, slave_dev->name,
1317                                bond_dev->name);
1318                         return -EPERM;
1319                 } else {
1320                         printk(KERN_WARNING DRV_NAME
1321                                ": %s: Warning: enslaved VLAN challenged "
1322                                "slave %s. Adding VLANs will be blocked as "
1323                                "long as %s is part of bond %s\n",
1324                                bond_dev->name, slave_dev->name, slave_dev->name,
1325                                bond_dev->name);
1326                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1327                 }
1328         } else {
1329                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1330                 if (bond->slave_cnt == 0) {
1331                         /* First slave, and it is not VLAN challenged,
1332                          * so remove the block of adding VLANs over the bond.
1333                          */
1334                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1335                 }
1336         }
1337
1338         /*
1339          * Old ifenslave binaries are no longer supported.  These can
1340          * be identified with moderate accurary by the state of the slave:
1341          * the current ifenslave will set the interface down prior to
1342          * enslaving it; the old ifenslave will not.
1343          */
1344         if ((slave_dev->flags & IFF_UP)) {
1345                 printk(KERN_ERR DRV_NAME ": %s is up. "
1346                        "This may be due to an out of date ifenslave.\n",
1347                        slave_dev->name);
1348                 res = -EPERM;
1349                 goto err_undo_flags;
1350         }
1351
1352         /* set bonding device ether type by slave - bonding netdevices are
1353          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1354          * there is a need to override some of the type dependent attribs/funcs.
1355          *
1356          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1357          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1358          */
1359         if (bond->slave_cnt == 0) {
1360                 if (slave_dev->type != ARPHRD_ETHER)
1361                         bond_setup_by_slave(bond_dev, slave_dev);
1362         } else if (bond_dev->type != slave_dev->type) {
1363                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1364                         "from other slaves (%d), can not enslave it.\n",
1365                         slave_dev->name,
1366                         slave_dev->type, bond_dev->type);
1367                         res = -EINVAL;
1368                         goto err_undo_flags;
1369         }
1370
1371         if (slave_dev->set_mac_address == NULL) {
1372                 if (bond->slave_cnt == 0) {
1373                         printk(KERN_WARNING DRV_NAME
1374                                ": %s: Warning: The first slave device "
1375                                "specified does not support setting the MAC "
1376                                "address. Enabling the fail_over_mac option.",
1377                                bond_dev->name);
1378                         bond->params.fail_over_mac = 1;
1379                 } else if (!bond->params.fail_over_mac) {
1380                         printk(KERN_ERR DRV_NAME
1381                                 ": %s: Error: The slave device specified "
1382                                 "does not support setting the MAC address, "
1383                                 "but fail_over_mac is not enabled.\n"
1384                                 , bond_dev->name);
1385                         res = -EOPNOTSUPP;
1386                         goto err_undo_flags;
1387                 }
1388         }
1389
1390         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1391         if (!new_slave) {
1392                 res = -ENOMEM;
1393                 goto err_undo_flags;
1394         }
1395
1396         /* save slave's original flags before calling
1397          * netdev_set_master and dev_open
1398          */
1399         new_slave->original_flags = slave_dev->flags;
1400
1401         /*
1402          * Save slave's original ("permanent") mac address for modes
1403          * that need it, and for restoring it upon release, and then
1404          * set it to the master's address
1405          */
1406         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1407
1408         if (!bond->params.fail_over_mac) {
1409                 /*
1410                  * Set slave to master's mac address.  The application already
1411                  * set the master's mac address to that of the first slave
1412                  */
1413                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1414                 addr.sa_family = slave_dev->type;
1415                 res = dev_set_mac_address(slave_dev, &addr);
1416                 if (res) {
1417                         dprintk("Error %d calling set_mac_address\n", res);
1418                         goto err_free;
1419                 }
1420         }
1421
1422         res = netdev_set_master(slave_dev, bond_dev);
1423         if (res) {
1424                 dprintk("Error %d calling netdev_set_master\n", res);
1425                 goto err_close;
1426         }
1427         /* open the slave since the application closed it */
1428         res = dev_open(slave_dev);
1429         if (res) {
1430                 dprintk("Openning slave %s failed\n", slave_dev->name);
1431                 goto err_restore_mac;
1432         }
1433
1434         new_slave->dev = slave_dev;
1435         slave_dev->priv_flags |= IFF_BONDING;
1436
1437         if ((bond->params.mode == BOND_MODE_TLB) ||
1438             (bond->params.mode == BOND_MODE_ALB)) {
1439                 /* bond_alb_init_slave() must be called before all other stages since
1440                  * it might fail and we do not want to have to undo everything
1441                  */
1442                 res = bond_alb_init_slave(bond, new_slave);
1443                 if (res) {
1444                         goto err_unset_master;
1445                 }
1446         }
1447
1448         /* If the mode USES_PRIMARY, then the new slave gets the
1449          * master's promisc (and mc) settings only if it becomes the
1450          * curr_active_slave, and that is taken care of later when calling
1451          * bond_change_active()
1452          */
1453         if (!USES_PRIMARY(bond->params.mode)) {
1454                 /* set promiscuity level to new slave */
1455                 if (bond_dev->flags & IFF_PROMISC) {
1456                         dev_set_promiscuity(slave_dev, 1);
1457                 }
1458
1459                 /* set allmulti level to new slave */
1460                 if (bond_dev->flags & IFF_ALLMULTI) {
1461                         dev_set_allmulti(slave_dev, 1);
1462                 }
1463
1464                 /* upload master's mc_list to new slave */
1465                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1466                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1467                 }
1468         }
1469
1470         if (bond->params.mode == BOND_MODE_8023AD) {
1471                 /* add lacpdu mc addr to mc list */
1472                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1473
1474                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1475         }
1476
1477         bond_add_vlans_on_slave(bond, slave_dev);
1478
1479         write_lock_bh(&bond->lock);
1480
1481         bond_attach_slave(bond, new_slave);
1482
1483         new_slave->delay = 0;
1484         new_slave->link_failure_count = 0;
1485
1486         bond_compute_features(bond);
1487
1488         new_slave->last_arp_rx = jiffies;
1489
1490         if (bond->params.miimon && !bond->params.use_carrier) {
1491                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1492
1493                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1494                         /*
1495                          * miimon is set but a bonded network driver
1496                          * does not support ETHTOOL/MII and
1497                          * arp_interval is not set.  Note: if
1498                          * use_carrier is enabled, we will never go
1499                          * here (because netif_carrier is always
1500                          * supported); thus, we don't need to change
1501                          * the messages for netif_carrier.
1502                          */
1503                         printk(KERN_WARNING DRV_NAME
1504                                ": %s: Warning: MII and ETHTOOL support not "
1505                                "available for interface %s, and "
1506                                "arp_interval/arp_ip_target module parameters "
1507                                "not specified, thus bonding will not detect "
1508                                "link failures! see bonding.txt for details.\n",
1509                                bond_dev->name, slave_dev->name);
1510                 } else if (link_reporting == -1) {
1511                         /* unable get link status using mii/ethtool */
1512                         printk(KERN_WARNING DRV_NAME
1513                                ": %s: Warning: can't get link status from "
1514                                "interface %s; the network driver associated "
1515                                "with this interface does not support MII or "
1516                                "ETHTOOL link status reporting, thus miimon "
1517                                "has no effect on this interface.\n",
1518                                bond_dev->name, slave_dev->name);
1519                 }
1520         }
1521
1522         /* check for initial state */
1523         if (!bond->params.miimon ||
1524             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1525                 if (bond->params.updelay) {
1526                         dprintk("Initial state of slave_dev is "
1527                                 "BOND_LINK_BACK\n");
1528                         new_slave->link  = BOND_LINK_BACK;
1529                         new_slave->delay = bond->params.updelay;
1530                 } else {
1531                         dprintk("Initial state of slave_dev is "
1532                                 "BOND_LINK_UP\n");
1533                         new_slave->link  = BOND_LINK_UP;
1534                 }
1535                 new_slave->jiffies = jiffies;
1536         } else {
1537                 dprintk("Initial state of slave_dev is "
1538                         "BOND_LINK_DOWN\n");
1539                 new_slave->link  = BOND_LINK_DOWN;
1540         }
1541
1542         if (bond_update_speed_duplex(new_slave) &&
1543             (new_slave->link != BOND_LINK_DOWN)) {
1544                 printk(KERN_WARNING DRV_NAME
1545                        ": %s: Warning: failed to get speed and duplex from %s, "
1546                        "assumed to be 100Mb/sec and Full.\n",
1547                        bond_dev->name, new_slave->dev->name);
1548
1549                 if (bond->params.mode == BOND_MODE_8023AD) {
1550                         printk(KERN_WARNING DRV_NAME
1551                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1552                                "support in base driver for proper aggregator "
1553                                "selection.\n", bond_dev->name);
1554                 }
1555         }
1556
1557         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1558                 /* if there is a primary slave, remember it */
1559                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1560                         bond->primary_slave = new_slave;
1561                 }
1562         }
1563
1564         switch (bond->params.mode) {
1565         case BOND_MODE_ACTIVEBACKUP:
1566                 bond_set_slave_inactive_flags(new_slave);
1567                 bond_select_active_slave(bond);
1568                 break;
1569         case BOND_MODE_8023AD:
1570                 /* in 802.3ad mode, the internal mechanism
1571                  * will activate the slaves in the selected
1572                  * aggregator
1573                  */
1574                 bond_set_slave_inactive_flags(new_slave);
1575                 /* if this is the first slave */
1576                 if (bond->slave_cnt == 1) {
1577                         SLAVE_AD_INFO(new_slave).id = 1;
1578                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1579                          * can be called only after the mac address of the bond is set
1580                          */
1581                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1582                                             bond->params.lacp_fast);
1583                 } else {
1584                         SLAVE_AD_INFO(new_slave).id =
1585                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1586                 }
1587
1588                 bond_3ad_bind_slave(new_slave);
1589                 break;
1590         case BOND_MODE_TLB:
1591         case BOND_MODE_ALB:
1592                 new_slave->state = BOND_STATE_ACTIVE;
1593                 bond_set_slave_inactive_flags(new_slave);
1594                 break;
1595         default:
1596                 dprintk("This slave is always active in trunk mode\n");
1597
1598                 /* always active in trunk mode */
1599                 new_slave->state = BOND_STATE_ACTIVE;
1600
1601                 /* In trunking mode there is little meaning to curr_active_slave
1602                  * anyway (it holds no special properties of the bond device),
1603                  * so we can change it without calling change_active_interface()
1604                  */
1605                 if (!bond->curr_active_slave) {
1606                         bond->curr_active_slave = new_slave;
1607                 }
1608                 break;
1609         } /* switch(bond_mode) */
1610
1611         bond_set_carrier(bond);
1612
1613         write_unlock_bh(&bond->lock);
1614
1615         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1616         if (res)
1617                 goto err_unset_master;
1618
1619         printk(KERN_INFO DRV_NAME
1620                ": %s: enslaving %s as a%s interface with a%s link.\n",
1621                bond_dev->name, slave_dev->name,
1622                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1623                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1624
1625         /* enslave is successful */
1626         return 0;
1627
1628 /* Undo stages on error */
1629 err_unset_master:
1630         netdev_set_master(slave_dev, NULL);
1631
1632 err_close:
1633         dev_close(slave_dev);
1634
1635 err_restore_mac:
1636         if (!bond->params.fail_over_mac) {
1637                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1638                 addr.sa_family = slave_dev->type;
1639                 dev_set_mac_address(slave_dev, &addr);
1640         }
1641
1642 err_free:
1643         kfree(new_slave);
1644
1645 err_undo_flags:
1646         bond_dev->features = old_features;
1647
1648         return res;
1649 }
1650
1651 /*
1652  * Try to release the slave device <slave> from the bond device <master>
1653  * It is legal to access curr_active_slave without a lock because all the function
1654  * is write-locked.
1655  *
1656  * The rules for slave state should be:
1657  *   for Active/Backup:
1658  *     Active stays on all backups go down
1659  *   for Bonded connections:
1660  *     The first up interface should be left on and all others downed.
1661  */
1662 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1663 {
1664         struct bonding *bond = bond_dev->priv;
1665         struct slave *slave, *oldcurrent;
1666         struct sockaddr addr;
1667         int mac_addr_differ;
1668         DECLARE_MAC_BUF(mac);
1669
1670         /* slave is not a slave or master is not master of this slave */
1671         if (!(slave_dev->flags & IFF_SLAVE) ||
1672             (slave_dev->master != bond_dev)) {
1673                 printk(KERN_ERR DRV_NAME
1674                        ": %s: Error: cannot release %s.\n",
1675                        bond_dev->name, slave_dev->name);
1676                 return -EINVAL;
1677         }
1678
1679         write_lock_bh(&bond->lock);
1680
1681         slave = bond_get_slave_by_dev(bond, slave_dev);
1682         if (!slave) {
1683                 /* not a slave of this bond */
1684                 printk(KERN_INFO DRV_NAME
1685                        ": %s: %s not enslaved\n",
1686                        bond_dev->name, slave_dev->name);
1687                 write_unlock_bh(&bond->lock);
1688                 return -EINVAL;
1689         }
1690
1691         mac_addr_differ = memcmp(bond_dev->dev_addr,
1692                                  slave->perm_hwaddr,
1693                                  ETH_ALEN);
1694         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1695                 printk(KERN_WARNING DRV_NAME
1696                        ": %s: Warning: the permanent HWaddr of %s - "
1697                        "%s - is still in use by %s. "
1698                        "Set the HWaddr of %s to a different address "
1699                        "to avoid conflicts.\n",
1700                        bond_dev->name,
1701                        slave_dev->name,
1702                        print_mac(mac, slave->perm_hwaddr),
1703                        bond_dev->name,
1704                        slave_dev->name);
1705         }
1706
1707         /* Inform AD package of unbinding of slave. */
1708         if (bond->params.mode == BOND_MODE_8023AD) {
1709                 /* must be called before the slave is
1710                  * detached from the list
1711                  */
1712                 bond_3ad_unbind_slave(slave);
1713         }
1714
1715         printk(KERN_INFO DRV_NAME
1716                ": %s: releasing %s interface %s\n",
1717                bond_dev->name,
1718                (slave->state == BOND_STATE_ACTIVE)
1719                ? "active" : "backup",
1720                slave_dev->name);
1721
1722         oldcurrent = bond->curr_active_slave;
1723
1724         bond->current_arp_slave = NULL;
1725
1726         /* release the slave from its bond */
1727         bond_detach_slave(bond, slave);
1728
1729         bond_compute_features(bond);
1730
1731         if (bond->primary_slave == slave) {
1732                 bond->primary_slave = NULL;
1733         }
1734
1735         if (oldcurrent == slave) {
1736                 bond_change_active_slave(bond, NULL);
1737         }
1738
1739         if ((bond->params.mode == BOND_MODE_TLB) ||
1740             (bond->params.mode == BOND_MODE_ALB)) {
1741                 /* Must be called only after the slave has been
1742                  * detached from the list and the curr_active_slave
1743                  * has been cleared (if our_slave == old_current),
1744                  * but before a new active slave is selected.
1745                  */
1746                 bond_alb_deinit_slave(bond, slave);
1747         }
1748
1749         if (oldcurrent == slave) {
1750                 /*
1751                  * Note that we hold RTNL over this sequence, so there
1752                  * is no concern that another slave add/remove event
1753                  * will interfere.
1754                  */
1755                 write_unlock_bh(&bond->lock);
1756                 read_lock(&bond->lock);
1757                 write_lock_bh(&bond->curr_slave_lock);
1758
1759                 bond_select_active_slave(bond);
1760
1761                 write_unlock_bh(&bond->curr_slave_lock);
1762                 read_unlock(&bond->lock);
1763                 write_lock_bh(&bond->lock);
1764         }
1765
1766         if (bond->slave_cnt == 0) {
1767                 bond_set_carrier(bond);
1768
1769                 /* if the last slave was removed, zero the mac address
1770                  * of the master so it will be set by the application
1771                  * to the mac address of the first slave
1772                  */
1773                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1774
1775                 if (list_empty(&bond->vlan_list)) {
1776                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1777                 } else {
1778                         printk(KERN_WARNING DRV_NAME
1779                                ": %s: Warning: clearing HW address of %s while it "
1780                                "still has VLANs.\n",
1781                                bond_dev->name, bond_dev->name);
1782                         printk(KERN_WARNING DRV_NAME
1783                                ": %s: When re-adding slaves, make sure the bond's "
1784                                "HW address matches its VLANs'.\n",
1785                                bond_dev->name);
1786                 }
1787         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1788                    !bond_has_challenged_slaves(bond)) {
1789                 printk(KERN_INFO DRV_NAME
1790                        ": %s: last VLAN challenged slave %s "
1791                        "left bond %s. VLAN blocking is removed\n",
1792                        bond_dev->name, slave_dev->name, bond_dev->name);
1793                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1794         }
1795
1796         write_unlock_bh(&bond->lock);
1797
1798         /* must do this from outside any spinlocks */
1799         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1800
1801         bond_del_vlans_from_slave(bond, slave_dev);
1802
1803         /* If the mode USES_PRIMARY, then we should only remove its
1804          * promisc and mc settings if it was the curr_active_slave, but that was
1805          * already taken care of above when we detached the slave
1806          */
1807         if (!USES_PRIMARY(bond->params.mode)) {
1808                 /* unset promiscuity level from slave */
1809                 if (bond_dev->flags & IFF_PROMISC) {
1810                         dev_set_promiscuity(slave_dev, -1);
1811                 }
1812
1813                 /* unset allmulti level from slave */
1814                 if (bond_dev->flags & IFF_ALLMULTI) {
1815                         dev_set_allmulti(slave_dev, -1);
1816                 }
1817
1818                 /* flush master's mc_list from slave */
1819                 bond_mc_list_flush(bond_dev, slave_dev);
1820         }
1821
1822         netdev_set_master(slave_dev, NULL);
1823
1824         /* close slave before restoring its mac address */
1825         dev_close(slave_dev);
1826
1827         if (!bond->params.fail_over_mac) {
1828                 /* restore original ("permanent") mac address */
1829                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1830                 addr.sa_family = slave_dev->type;
1831                 dev_set_mac_address(slave_dev, &addr);
1832         }
1833
1834         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1835                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1836                                    IFF_SLAVE_NEEDARP);
1837
1838         kfree(slave);
1839
1840         return 0;  /* deletion OK */
1841 }
1842
1843 /*
1844 * Destroy a bonding device.
1845 * Must be under rtnl_lock when this function is called.
1846 */
1847 void bond_destroy(struct bonding *bond)
1848 {
1849         bond_deinit(bond->dev);
1850         bond_destroy_sysfs_entry(bond);
1851         unregister_netdevice(bond->dev);
1852 }
1853
1854 /*
1855 * First release a slave and than destroy the bond if no more slaves iare left.
1856 * Must be under rtnl_lock when this function is called.
1857 */
1858 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1859 {
1860         struct bonding *bond = bond_dev->priv;
1861         int ret;
1862
1863         ret = bond_release(bond_dev, slave_dev);
1864         if ((ret == 0) && (bond->slave_cnt == 0)) {
1865                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1866                        bond_dev->name, bond_dev->name);
1867                 bond_destroy(bond);
1868         }
1869         return ret;
1870 }
1871
1872 /*
1873  * This function releases all slaves.
1874  */
1875 static int bond_release_all(struct net_device *bond_dev)
1876 {
1877         struct bonding *bond = bond_dev->priv;
1878         struct slave *slave;
1879         struct net_device *slave_dev;
1880         struct sockaddr addr;
1881
1882         write_lock_bh(&bond->lock);
1883
1884         netif_carrier_off(bond_dev);
1885
1886         if (bond->slave_cnt == 0) {
1887                 goto out;
1888         }
1889
1890         bond->current_arp_slave = NULL;
1891         bond->primary_slave = NULL;
1892         bond_change_active_slave(bond, NULL);
1893
1894         while ((slave = bond->first_slave) != NULL) {
1895                 /* Inform AD package of unbinding of slave
1896                  * before slave is detached from the list.
1897                  */
1898                 if (bond->params.mode == BOND_MODE_8023AD) {
1899                         bond_3ad_unbind_slave(slave);
1900                 }
1901
1902                 slave_dev = slave->dev;
1903                 bond_detach_slave(bond, slave);
1904
1905                 if ((bond->params.mode == BOND_MODE_TLB) ||
1906                     (bond->params.mode == BOND_MODE_ALB)) {
1907                         /* must be called only after the slave
1908                          * has been detached from the list
1909                          */
1910                         bond_alb_deinit_slave(bond, slave);
1911                 }
1912
1913                 bond_compute_features(bond);
1914
1915                 /* now that the slave is detached, unlock and perform
1916                  * all the undo steps that should not be called from
1917                  * within a lock.
1918                  */
1919                 write_unlock_bh(&bond->lock);
1920
1921                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1922                 bond_del_vlans_from_slave(bond, slave_dev);
1923
1924                 /* If the mode USES_PRIMARY, then we should only remove its
1925                  * promisc and mc settings if it was the curr_active_slave, but that was
1926                  * already taken care of above when we detached the slave
1927                  */
1928                 if (!USES_PRIMARY(bond->params.mode)) {
1929                         /* unset promiscuity level from slave */
1930                         if (bond_dev->flags & IFF_PROMISC) {
1931                                 dev_set_promiscuity(slave_dev, -1);
1932                         }
1933
1934                         /* unset allmulti level from slave */
1935                         if (bond_dev->flags & IFF_ALLMULTI) {
1936                                 dev_set_allmulti(slave_dev, -1);
1937                         }
1938
1939                         /* flush master's mc_list from slave */
1940                         bond_mc_list_flush(bond_dev, slave_dev);
1941                 }
1942
1943                 netdev_set_master(slave_dev, NULL);
1944
1945                 /* close slave before restoring its mac address */
1946                 dev_close(slave_dev);
1947
1948                 if (!bond->params.fail_over_mac) {
1949                         /* restore original ("permanent") mac address*/
1950                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1951                         addr.sa_family = slave_dev->type;
1952                         dev_set_mac_address(slave_dev, &addr);
1953                 }
1954
1955                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1956                                            IFF_SLAVE_INACTIVE);
1957
1958                 kfree(slave);
1959
1960                 /* re-acquire the lock before getting the next slave */
1961                 write_lock_bh(&bond->lock);
1962         }
1963
1964         /* zero the mac address of the master so it will be
1965          * set by the application to the mac address of the
1966          * first slave
1967          */
1968         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1969
1970         if (list_empty(&bond->vlan_list)) {
1971                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1972         } else {
1973                 printk(KERN_WARNING DRV_NAME
1974                        ": %s: Warning: clearing HW address of %s while it "
1975                        "still has VLANs.\n",
1976                        bond_dev->name, bond_dev->name);
1977                 printk(KERN_WARNING DRV_NAME
1978                        ": %s: When re-adding slaves, make sure the bond's "
1979                        "HW address matches its VLANs'.\n",
1980                        bond_dev->name);
1981         }
1982
1983         printk(KERN_INFO DRV_NAME
1984                ": %s: released all slaves\n",
1985                bond_dev->name);
1986
1987 out:
1988         write_unlock_bh(&bond->lock);
1989
1990         return 0;
1991 }
1992
1993 /*
1994  * This function changes the active slave to slave <slave_dev>.
1995  * It returns -EINVAL in the following cases.
1996  *  - <slave_dev> is not found in the list.
1997  *  - There is not active slave now.
1998  *  - <slave_dev> is already active.
1999  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2000  *  - <slave_dev> is not running.
2001  * In these cases, this fuction does nothing.
2002  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2003  */
2004 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2005 {
2006         struct bonding *bond = bond_dev->priv;
2007         struct slave *old_active = NULL;
2008         struct slave *new_active = NULL;
2009         int res = 0;
2010
2011         if (!USES_PRIMARY(bond->params.mode)) {
2012                 return -EINVAL;
2013         }
2014
2015         /* Verify that master_dev is indeed the master of slave_dev */
2016         if (!(slave_dev->flags & IFF_SLAVE) ||
2017             (slave_dev->master != bond_dev)) {
2018                 return -EINVAL;
2019         }
2020
2021         read_lock(&bond->lock);
2022
2023         read_lock(&bond->curr_slave_lock);
2024         old_active = bond->curr_active_slave;
2025         read_unlock(&bond->curr_slave_lock);
2026
2027         new_active = bond_get_slave_by_dev(bond, slave_dev);
2028
2029         /*
2030          * Changing to the current active: do nothing; return success.
2031          */
2032         if (new_active && (new_active == old_active)) {
2033                 read_unlock(&bond->lock);
2034                 return 0;
2035         }
2036
2037         if ((new_active) &&
2038             (old_active) &&
2039             (new_active->link == BOND_LINK_UP) &&
2040             IS_UP(new_active->dev)) {
2041                 write_lock_bh(&bond->curr_slave_lock);
2042                 bond_change_active_slave(bond, new_active);
2043                 write_unlock_bh(&bond->curr_slave_lock);
2044         } else {
2045                 res = -EINVAL;
2046         }
2047
2048         read_unlock(&bond->lock);
2049
2050         return res;
2051 }
2052
2053 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2054 {
2055         struct bonding *bond = bond_dev->priv;
2056
2057         info->bond_mode = bond->params.mode;
2058         info->miimon = bond->params.miimon;
2059
2060         read_lock_bh(&bond->lock);
2061         info->num_slaves = bond->slave_cnt;
2062         read_unlock_bh(&bond->lock);
2063
2064         return 0;
2065 }
2066
2067 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2068 {
2069         struct bonding *bond = bond_dev->priv;
2070         struct slave *slave;
2071         int i, found = 0;
2072
2073         if (info->slave_id < 0) {
2074                 return -ENODEV;
2075         }
2076
2077         read_lock_bh(&bond->lock);
2078
2079         bond_for_each_slave(bond, slave, i) {
2080                 if (i == (int)info->slave_id) {
2081                         found = 1;
2082                         break;
2083                 }
2084         }
2085
2086         read_unlock_bh(&bond->lock);
2087
2088         if (found) {
2089                 strcpy(info->slave_name, slave->dev->name);
2090                 info->link = slave->link;
2091                 info->state = slave->state;
2092                 info->link_failure_count = slave->link_failure_count;
2093         } else {
2094                 return -ENODEV;
2095         }
2096
2097         return 0;
2098 }
2099
2100 /*-------------------------------- Monitoring -------------------------------*/
2101
2102 /*
2103  * if !have_locks, return nonzero if a failover is necessary.  if
2104  * have_locks, do whatever failover activities are needed.
2105  *
2106  * This is to separate the inspection and failover steps for locking
2107  * purposes; failover requires rtnl, but acquiring it for every
2108  * inspection is undesirable, so a wrapper first does inspection, and
2109  * the acquires the necessary locks and calls again to perform
2110  * failover if needed.  Since all locks are dropped, a complete
2111  * restart is needed between calls.
2112  */
2113 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2114 {
2115         struct slave *slave, *oldcurrent;
2116         int do_failover = 0;
2117         int i;
2118
2119         if (bond->slave_cnt == 0)
2120                 goto out;
2121
2122         /* we will try to read the link status of each of our slaves, and
2123          * set their IFF_RUNNING flag appropriately. For each slave not
2124          * supporting MII status, we won't do anything so that a user-space
2125          * program could monitor the link itself if needed.
2126          */
2127
2128         if (bond->send_grat_arp) {
2129                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2130                                 &bond->curr_active_slave->dev->state))
2131                         dprintk("Needs to send gratuitous arp but not yet\n");
2132                 else {
2133                         dprintk("sending delayed gratuitous arp on on %s\n",
2134                                 bond->curr_active_slave->dev->name);
2135                         bond_send_gratuitous_arp(bond);
2136                         bond->send_grat_arp = 0;
2137                 }
2138         }
2139         read_lock(&bond->curr_slave_lock);
2140         oldcurrent = bond->curr_active_slave;
2141         read_unlock(&bond->curr_slave_lock);
2142
2143         bond_for_each_slave(bond, slave, i) {
2144                 struct net_device *slave_dev = slave->dev;
2145                 int link_state;
2146                 u16 old_speed = slave->speed;
2147                 u8 old_duplex = slave->duplex;
2148
2149                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2150
2151                 switch (slave->link) {
2152                 case BOND_LINK_UP:      /* the link was up */
2153                         if (link_state == BMSR_LSTATUS) {
2154                                 if (!oldcurrent) {
2155                                         if (!have_locks)
2156                                                 return 1;
2157                                         do_failover = 1;
2158                                 }
2159                                 break;
2160                         } else { /* link going down */
2161                                 slave->link  = BOND_LINK_FAIL;
2162                                 slave->delay = bond->params.downdelay;
2163
2164                                 if (slave->link_failure_count < UINT_MAX) {
2165                                         slave->link_failure_count++;
2166                                 }
2167
2168                                 if (bond->params.downdelay) {
2169                                         printk(KERN_INFO DRV_NAME
2170                                                ": %s: link status down for %s "
2171                                                "interface %s, disabling it in "
2172                                                "%d ms.\n",
2173                                                bond->dev->name,
2174                                                IS_UP(slave_dev)
2175                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2176                                                   ? ((slave == oldcurrent)
2177                                                      ? "active " : "backup ")
2178                                                   : "")
2179                                                : "idle ",
2180                                                slave_dev->name,
2181                                                bond->params.downdelay * bond->params.miimon);
2182                                 }
2183                         }
2184                         /* no break ! fall through the BOND_LINK_FAIL test to
2185                            ensure proper action to be taken
2186                         */
2187                 case BOND_LINK_FAIL:    /* the link has just gone down */
2188                         if (link_state != BMSR_LSTATUS) {
2189                                 /* link stays down */
2190                                 if (slave->delay <= 0) {
2191                                         if (!have_locks)
2192                                                 return 1;
2193
2194                                         /* link down for too long time */
2195                                         slave->link = BOND_LINK_DOWN;
2196
2197                                         /* in active/backup mode, we must
2198                                          * completely disable this interface
2199                                          */
2200                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2201                                             (bond->params.mode == BOND_MODE_8023AD)) {
2202                                                 bond_set_slave_inactive_flags(slave);
2203                                         }
2204
2205                                         printk(KERN_INFO DRV_NAME
2206                                                ": %s: link status definitely "
2207                                                "down for interface %s, "
2208                                                "disabling it\n",
2209                                                bond->dev->name,
2210                                                slave_dev->name);
2211
2212                                         /* notify ad that the link status has changed */
2213                                         if (bond->params.mode == BOND_MODE_8023AD) {
2214                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2215                                         }
2216
2217                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2218                                             (bond->params.mode == BOND_MODE_ALB)) {
2219                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2220                                         }
2221
2222                                         if (slave == oldcurrent) {
2223                                                 do_failover = 1;
2224                                         }
2225                                 } else {
2226                                         slave->delay--;
2227                                 }
2228                         } else {
2229                                 /* link up again */
2230                                 slave->link  = BOND_LINK_UP;
2231                                 slave->jiffies = jiffies;
2232                                 printk(KERN_INFO DRV_NAME
2233                                        ": %s: link status up again after %d "
2234                                        "ms for interface %s.\n",
2235                                        bond->dev->name,
2236                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2237                                        slave_dev->name);
2238                         }
2239                         break;
2240                 case BOND_LINK_DOWN:    /* the link was down */
2241                         if (link_state != BMSR_LSTATUS) {
2242                                 /* the link stays down, nothing more to do */
2243                                 break;
2244                         } else {        /* link going up */
2245                                 slave->link  = BOND_LINK_BACK;
2246                                 slave->delay = bond->params.updelay;
2247
2248                                 if (bond->params.updelay) {
2249                                         /* if updelay == 0, no need to
2250                                            advertise about a 0 ms delay */
2251                                         printk(KERN_INFO DRV_NAME
2252                                                ": %s: link status up for "
2253                                                "interface %s, enabling it "
2254                                                "in %d ms.\n",
2255                                                bond->dev->name,
2256                                                slave_dev->name,
2257                                                bond->params.updelay * bond->params.miimon);
2258                                 }
2259                         }
2260                         /* no break ! fall through the BOND_LINK_BACK state in
2261                            case there's something to do.
2262                         */
2263                 case BOND_LINK_BACK:    /* the link has just come back */
2264                         if (link_state != BMSR_LSTATUS) {
2265                                 /* link down again */
2266                                 slave->link  = BOND_LINK_DOWN;
2267
2268                                 printk(KERN_INFO DRV_NAME
2269                                        ": %s: link status down again after %d "
2270                                        "ms for interface %s.\n",
2271                                        bond->dev->name,
2272                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2273                                        slave_dev->name);
2274                         } else {
2275                                 /* link stays up */
2276                                 if (slave->delay == 0) {
2277                                         if (!have_locks)
2278                                                 return 1;
2279
2280                                         /* now the link has been up for long time enough */
2281                                         slave->link = BOND_LINK_UP;
2282                                         slave->jiffies = jiffies;
2283
2284                                         if (bond->params.mode == BOND_MODE_8023AD) {
2285                                                 /* prevent it from being the active one */
2286                                                 slave->state = BOND_STATE_BACKUP;
2287                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2288                                                 /* make it immediately active */
2289                                                 slave->state = BOND_STATE_ACTIVE;
2290                                         } else if (slave != bond->primary_slave) {
2291                                                 /* prevent it from being the active one */
2292                                                 slave->state = BOND_STATE_BACKUP;
2293                                         }
2294
2295                                         printk(KERN_INFO DRV_NAME
2296                                                ": %s: link status definitely "
2297                                                "up for interface %s.\n",
2298                                                bond->dev->name,
2299                                                slave_dev->name);
2300
2301                                         /* notify ad that the link status has changed */
2302                                         if (bond->params.mode == BOND_MODE_8023AD) {
2303                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2304                                         }
2305
2306                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2307                                             (bond->params.mode == BOND_MODE_ALB)) {
2308                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2309                                         }
2310
2311                                         if ((!oldcurrent) ||
2312                                             (slave == bond->primary_slave)) {
2313                                                 do_failover = 1;
2314                                         }
2315                                 } else {
2316                                         slave->delay--;
2317                                 }
2318                         }
2319                         break;
2320                 default:
2321                         /* Should not happen */
2322                         printk(KERN_ERR DRV_NAME
2323                                ": %s: Error: %s Illegal value (link=%d)\n",
2324                                bond->dev->name,
2325                                slave->dev->name,
2326                                slave->link);
2327                         goto out;
2328                 } /* end of switch (slave->link) */
2329
2330                 bond_update_speed_duplex(slave);
2331
2332                 if (bond->params.mode == BOND_MODE_8023AD) {
2333                         if (old_speed != slave->speed) {
2334                                 bond_3ad_adapter_speed_changed(slave);
2335                         }
2336
2337                         if (old_duplex != slave->duplex) {
2338                                 bond_3ad_adapter_duplex_changed(slave);
2339                         }
2340                 }
2341
2342         } /* end of for */
2343
2344         if (do_failover) {
2345                 ASSERT_RTNL();
2346
2347                 write_lock_bh(&bond->curr_slave_lock);
2348
2349                 bond_select_active_slave(bond);
2350
2351                 write_unlock_bh(&bond->curr_slave_lock);
2352
2353         } else
2354                 bond_set_carrier(bond);
2355
2356 out:
2357         return 0;
2358 }
2359
2360 /*
2361  * bond_mii_monitor
2362  *
2363  * Really a wrapper that splits the mii monitor into two phases: an
2364  * inspection, then (if inspection indicates something needs to be
2365  * done) an acquisition of appropriate locks followed by another pass
2366  * to implement whatever link state changes are indicated.
2367  */
2368 void bond_mii_monitor(struct work_struct *work)
2369 {
2370         struct bonding *bond = container_of(work, struct bonding,
2371                                             mii_work.work);
2372         unsigned long delay;
2373
2374         read_lock(&bond->lock);
2375         if (bond->kill_timers) {
2376                 read_unlock(&bond->lock);
2377                 return;
2378         }
2379         if (__bond_mii_monitor(bond, 0)) {
2380                 read_unlock(&bond->lock);
2381                 rtnl_lock();
2382                 read_lock(&bond->lock);
2383                 __bond_mii_monitor(bond, 1);
2384                 rtnl_unlock();
2385         }
2386
2387         delay = ((bond->params.miimon * HZ) / 1000) ? : 1;
2388         read_unlock(&bond->lock);
2389         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2390 }
2391
2392 static __be32 bond_glean_dev_ip(struct net_device *dev)
2393 {
2394         struct in_device *idev;
2395         struct in_ifaddr *ifa;
2396         __be32 addr = 0;
2397
2398         if (!dev)
2399                 return 0;
2400
2401         rcu_read_lock();
2402         idev = __in_dev_get_rcu(dev);
2403         if (!idev)
2404                 goto out;
2405
2406         ifa = idev->ifa_list;
2407         if (!ifa)
2408                 goto out;
2409
2410         addr = ifa->ifa_local;
2411 out:
2412         rcu_read_unlock();
2413         return addr;
2414 }
2415
2416 static int bond_has_ip(struct bonding *bond)
2417 {
2418         struct vlan_entry *vlan, *vlan_next;
2419
2420         if (bond->master_ip)
2421                 return 1;
2422
2423         if (list_empty(&bond->vlan_list))
2424                 return 0;
2425
2426         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2427                                  vlan_list) {
2428                 if (vlan->vlan_ip)
2429                         return 1;
2430         }
2431
2432         return 0;
2433 }
2434
2435 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2436 {
2437         struct vlan_entry *vlan, *vlan_next;
2438
2439         if (ip == bond->master_ip)
2440                 return 1;
2441
2442         if (list_empty(&bond->vlan_list))
2443                 return 0;
2444
2445         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2446                                  vlan_list) {
2447                 if (ip == vlan->vlan_ip)
2448                         return 1;
2449         }
2450
2451         return 0;
2452 }
2453
2454 /*
2455  * We go to the (large) trouble of VLAN tagging ARP frames because
2456  * switches in VLAN mode (especially if ports are configured as
2457  * "native" to a VLAN) might not pass non-tagged frames.
2458  */
2459 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2460 {
2461         struct sk_buff *skb;
2462
2463         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2464                slave_dev->name, dest_ip, src_ip, vlan_id);
2465
2466         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2467                          NULL, slave_dev->dev_addr, NULL);
2468
2469         if (!skb) {
2470                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2471                 return;
2472         }
2473         if (vlan_id) {
2474                 skb = vlan_put_tag(skb, vlan_id);
2475                 if (!skb) {
2476                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2477                         return;
2478                 }
2479         }
2480         arp_xmit(skb);
2481 }
2482
2483
2484 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2485 {
2486         int i, vlan_id, rv;
2487         __be32 *targets = bond->params.arp_targets;
2488         struct vlan_entry *vlan, *vlan_next;
2489         struct net_device *vlan_dev;
2490         struct flowi fl;
2491         struct rtable *rt;
2492
2493         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2494                 if (!targets[i])
2495                         continue;
2496                 dprintk("basa: target %x\n", targets[i]);
2497                 if (list_empty(&bond->vlan_list)) {
2498                         dprintk("basa: empty vlan: arp_send\n");
2499                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2500                                       bond->master_ip, 0);
2501                         continue;
2502                 }
2503
2504                 /*
2505                  * If VLANs are configured, we do a route lookup to
2506                  * determine which VLAN interface would be used, so we
2507                  * can tag the ARP with the proper VLAN tag.
2508                  */
2509                 memset(&fl, 0, sizeof(fl));
2510                 fl.fl4_dst = targets[i];
2511                 fl.fl4_tos = RTO_ONLINK;
2512
2513                 rv = ip_route_output_key(&rt, &fl);
2514                 if (rv) {
2515                         if (net_ratelimit()) {
2516                                 printk(KERN_WARNING DRV_NAME
2517                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2518                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2519                         }
2520                         continue;
2521                 }
2522
2523                 /*
2524                  * This target is not on a VLAN
2525                  */
2526                 if (rt->u.dst.dev == bond->dev) {
2527                         ip_rt_put(rt);
2528                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2529                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2530                                       bond->master_ip, 0);
2531                         continue;
2532                 }
2533
2534                 vlan_id = 0;
2535                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2536                                          vlan_list) {
2537                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2538                         if (vlan_dev == rt->u.dst.dev) {
2539                                 vlan_id = vlan->vlan_id;
2540                                 dprintk("basa: vlan match on %s %d\n",
2541                                        vlan_dev->name, vlan_id);
2542                                 break;
2543                         }
2544                 }
2545
2546                 if (vlan_id) {
2547                         ip_rt_put(rt);
2548                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2549                                       vlan->vlan_ip, vlan_id);
2550                         continue;
2551                 }
2552
2553                 if (net_ratelimit()) {
2554                         printk(KERN_WARNING DRV_NAME
2555                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2556                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2557                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2558                 }
2559                 ip_rt_put(rt);
2560         }
2561 }
2562
2563 /*
2564  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2565  * for each VLAN above us.
2566  */
2567 static void bond_send_gratuitous_arp(struct bonding *bond)
2568 {
2569         struct slave *slave = bond->curr_active_slave;
2570         struct vlan_entry *vlan;
2571         struct net_device *vlan_dev;
2572
2573         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2574                                 slave ? slave->dev->name : "NULL");
2575         if (!slave)
2576                 return;
2577
2578         if (bond->master_ip) {
2579                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2580                                 bond->master_ip, 0);
2581         }
2582
2583         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2584                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2585                 if (vlan->vlan_ip) {
2586                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2587                                       vlan->vlan_ip, vlan->vlan_id);
2588                 }
2589         }
2590 }
2591
2592 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2593 {
2594         int i;
2595         __be32 *targets = bond->params.arp_targets;
2596
2597         targets = bond->params.arp_targets;
2598         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2599                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2600                         "%u.%u.%u.%u bhti(tip) %d\n",
2601                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2602                        bond_has_this_ip(bond, tip));
2603                 if (sip == targets[i]) {
2604                         if (bond_has_this_ip(bond, tip))
2605                                 slave->last_arp_rx = jiffies;
2606                         return;
2607                 }
2608         }
2609 }
2610
2611 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2612 {
2613         struct arphdr *arp;
2614         struct slave *slave;
2615         struct bonding *bond;
2616         unsigned char *arp_ptr;
2617         __be32 sip, tip;
2618
2619         if (dev->nd_net != &init_net)
2620                 goto out;
2621
2622         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2623                 goto out;
2624
2625         bond = dev->priv;
2626         read_lock(&bond->lock);
2627
2628         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2629                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2630                 orig_dev ? orig_dev->name : "NULL");
2631
2632         slave = bond_get_slave_by_dev(bond, orig_dev);
2633         if (!slave || !slave_do_arp_validate(bond, slave))
2634                 goto out_unlock;
2635
2636         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2637         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2638                                  (2 * dev->addr_len) +
2639                                  (2 * sizeof(u32)))))
2640                 goto out_unlock;
2641
2642         arp = arp_hdr(skb);
2643         if (arp->ar_hln != dev->addr_len ||
2644             skb->pkt_type == PACKET_OTHERHOST ||
2645             skb->pkt_type == PACKET_LOOPBACK ||
2646             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2647             arp->ar_pro != htons(ETH_P_IP) ||
2648             arp->ar_pln != 4)
2649                 goto out_unlock;
2650
2651         arp_ptr = (unsigned char *)(arp + 1);
2652         arp_ptr += dev->addr_len;
2653         memcpy(&sip, arp_ptr, 4);
2654         arp_ptr += 4 + dev->addr_len;
2655         memcpy(&tip, arp_ptr, 4);
2656
2657         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2658                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2659                 slave->state, bond->params.arp_validate,
2660                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2661
2662         /*
2663          * Backup slaves won't see the ARP reply, but do come through
2664          * here for each ARP probe (so we swap the sip/tip to validate
2665          * the probe).  In a "redundant switch, common router" type of
2666          * configuration, the ARP probe will (hopefully) travel from
2667          * the active, through one switch, the router, then the other
2668          * switch before reaching the backup.
2669          */
2670         if (slave->state == BOND_STATE_ACTIVE)
2671                 bond_validate_arp(bond, slave, sip, tip);
2672         else
2673                 bond_validate_arp(bond, slave, tip, sip);
2674
2675 out_unlock:
2676         read_unlock(&bond->lock);
2677 out:
2678         dev_kfree_skb(skb);
2679         return NET_RX_SUCCESS;
2680 }
2681
2682 /*
2683  * this function is called regularly to monitor each slave's link
2684  * ensuring that traffic is being sent and received when arp monitoring
2685  * is used in load-balancing mode. if the adapter has been dormant, then an
2686  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2687  * arp monitoring in active backup mode.
2688  */
2689 void bond_loadbalance_arp_mon(struct work_struct *work)
2690 {
2691         struct bonding *bond = container_of(work, struct bonding,
2692                                             arp_work.work);
2693         struct slave *slave, *oldcurrent;
2694         int do_failover = 0;
2695         int delta_in_ticks;
2696         int i;
2697
2698         read_lock(&bond->lock);
2699
2700         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2701
2702         if (bond->kill_timers) {
2703                 goto out;
2704         }
2705
2706         if (bond->slave_cnt == 0) {
2707                 goto re_arm;
2708         }
2709
2710         read_lock(&bond->curr_slave_lock);
2711         oldcurrent = bond->curr_active_slave;
2712         read_unlock(&bond->curr_slave_lock);
2713
2714         /* see if any of the previous devices are up now (i.e. they have
2715          * xmt and rcv traffic). the curr_active_slave does not come into
2716          * the picture unless it is null. also, slave->jiffies is not needed
2717          * here because we send an arp on each slave and give a slave as
2718          * long as it needs to get the tx/rx within the delta.
2719          * TODO: what about up/down delay in arp mode? it wasn't here before
2720          *       so it can wait
2721          */
2722         bond_for_each_slave(bond, slave, i) {
2723                 if (slave->link != BOND_LINK_UP) {
2724                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2725                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2726
2727                                 slave->link  = BOND_LINK_UP;
2728                                 slave->state = BOND_STATE_ACTIVE;
2729
2730                                 /* primary_slave has no meaning in round-robin
2731                                  * mode. the window of a slave being up and
2732                                  * curr_active_slave being null after enslaving
2733                                  * is closed.
2734                                  */
2735                                 if (!oldcurrent) {
2736                                         printk(KERN_INFO DRV_NAME
2737                                                ": %s: link status definitely "
2738                                                "up for interface %s, ",
2739                                                bond->dev->name,
2740                                                slave->dev->name);
2741                                         do_failover = 1;
2742                                 } else {
2743                                         printk(KERN_INFO DRV_NAME
2744                                                ": %s: interface %s is now up\n",
2745                                                bond->dev->name,
2746                                                slave->dev->name);
2747                                 }
2748                         }
2749                 } else {
2750                         /* slave->link == BOND_LINK_UP */
2751
2752                         /* not all switches will respond to an arp request
2753                          * when the source ip is 0, so don't take the link down
2754                          * if we don't know our ip yet
2755                          */
2756                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2757                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2758                              bond_has_ip(bond))) {
2759
2760                                 slave->link  = BOND_LINK_DOWN;
2761                                 slave->state = BOND_STATE_BACKUP;
2762
2763                                 if (slave->link_failure_count < UINT_MAX) {
2764                                         slave->link_failure_count++;
2765                                 }
2766
2767                                 printk(KERN_INFO DRV_NAME
2768                                        ": %s: interface %s is now down.\n",
2769                                        bond->dev->name,
2770                                        slave->dev->name);
2771
2772                                 if (slave == oldcurrent) {
2773                                         do_failover = 1;
2774                                 }
2775                         }
2776                 }
2777
2778                 /* note: if switch is in round-robin mode, all links
2779                  * must tx arp to ensure all links rx an arp - otherwise
2780                  * links may oscillate or not come up at all; if switch is
2781                  * in something like xor mode, there is nothing we can
2782                  * do - all replies will be rx'ed on same link causing slaves
2783                  * to be unstable during low/no traffic periods
2784                  */
2785                 if (IS_UP(slave->dev)) {
2786                         bond_arp_send_all(bond, slave);
2787                 }
2788         }
2789
2790         if (do_failover) {
2791                 rtnl_lock();
2792                 write_lock_bh(&bond->curr_slave_lock);
2793
2794                 bond_select_active_slave(bond);
2795
2796                 write_unlock_bh(&bond->curr_slave_lock);
2797                 rtnl_unlock();
2798
2799         }
2800
2801 re_arm:
2802         if (bond->params.arp_interval)
2803                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2804 out:
2805         read_unlock(&bond->lock);
2806 }
2807
2808 /*
2809  * When using arp monitoring in active-backup mode, this function is
2810  * called to determine if any backup slaves have went down or a new
2811  * current slave needs to be found.
2812  * The backup slaves never generate traffic, they are considered up by merely
2813  * receiving traffic. If the current slave goes down, each backup slave will
2814  * be given the opportunity to tx/rx an arp before being taken down - this
2815  * prevents all slaves from being taken down due to the current slave not
2816  * sending any traffic for the backups to receive. The arps are not necessarily
2817  * necessary, any tx and rx traffic will keep the current slave up. While any
2818  * rx traffic will keep the backup slaves up, the current slave is responsible
2819  * for generating traffic to keep them up regardless of any other traffic they
2820  * may have received.
2821  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2822  */
2823 void bond_activebackup_arp_mon(struct work_struct *work)
2824 {
2825         struct bonding *bond = container_of(work, struct bonding,
2826                                             arp_work.work);
2827         struct slave *slave;
2828         int delta_in_ticks;
2829         int i;
2830
2831         read_lock(&bond->lock);
2832
2833         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2834
2835         if (bond->kill_timers) {
2836                 goto out;
2837         }
2838
2839         if (bond->slave_cnt == 0) {
2840                 goto re_arm;
2841         }
2842
2843         /* determine if any slave has come up or any backup slave has
2844          * gone down
2845          * TODO: what about up/down delay in arp mode? it wasn't here before
2846          *       so it can wait
2847          */
2848         bond_for_each_slave(bond, slave, i) {
2849                 if (slave->link != BOND_LINK_UP) {
2850                         if ((jiffies - slave_last_rx(bond, slave)) <=
2851                              delta_in_ticks) {
2852
2853                                 slave->link = BOND_LINK_UP;
2854
2855                                 rtnl_lock();
2856
2857                                 write_lock_bh(&bond->curr_slave_lock);
2858
2859                                 if ((!bond->curr_active_slave) &&
2860                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2861                                         bond_change_active_slave(bond, slave);
2862                                         bond->current_arp_slave = NULL;
2863                                 } else if (bond->curr_active_slave != slave) {
2864                                         /* this slave has just come up but we
2865                                          * already have a current slave; this
2866                                          * can also happen if bond_enslave adds
2867                                          * a new slave that is up while we are
2868                                          * searching for a new slave
2869                                          */
2870                                         bond_set_slave_inactive_flags(slave);
2871                                         bond->current_arp_slave = NULL;
2872                                 }
2873
2874                                 bond_set_carrier(bond);
2875
2876                                 if (slave == bond->curr_active_slave) {
2877                                         printk(KERN_INFO DRV_NAME
2878                                                ": %s: %s is up and now the "
2879                                                "active interface\n",
2880                                                bond->dev->name,
2881                                                slave->dev->name);
2882                                         netif_carrier_on(bond->dev);
2883                                 } else {
2884                                         printk(KERN_INFO DRV_NAME
2885                                                ": %s: backup interface %s is "
2886                                                "now up\n",
2887                                                bond->dev->name,
2888                                                slave->dev->name);
2889                                 }
2890
2891                                 write_unlock_bh(&bond->curr_slave_lock);
2892                                 rtnl_unlock();
2893                         }
2894                 } else {
2895                         read_lock(&bond->curr_slave_lock);
2896
2897                         if ((slave != bond->curr_active_slave) &&
2898                             (!bond->current_arp_slave) &&
2899                             (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2900                              bond_has_ip(bond))) {
2901                                 /* a backup slave has gone down; three times
2902                                  * the delta allows the current slave to be
2903                                  * taken out before the backup slave.
2904                                  * note: a non-null current_arp_slave indicates
2905                                  * the curr_active_slave went down and we are
2906                                  * searching for a new one; under this
2907                                  * condition we only take the curr_active_slave
2908                                  * down - this gives each slave a chance to
2909                                  * tx/rx traffic before being taken out
2910                                  */
2911
2912                                 read_unlock(&bond->curr_slave_lock);
2913
2914                                 slave->link  = BOND_LINK_DOWN;
2915
2916                                 if (slave->link_failure_count < UINT_MAX) {
2917                                         slave->link_failure_count++;
2918                                 }
2919
2920                                 bond_set_slave_inactive_flags(slave);
2921
2922                                 printk(KERN_INFO DRV_NAME
2923                                        ": %s: backup interface %s is now down\n",
2924                                        bond->dev->name,
2925                                        slave->dev->name);
2926                         } else {
2927                                 read_unlock(&bond->curr_slave_lock);
2928                         }
2929                 }
2930         }
2931
2932         read_lock(&bond->curr_slave_lock);
2933         slave = bond->curr_active_slave;
2934         read_unlock(&bond->curr_slave_lock);
2935
2936         if (slave) {
2937                 /* if we have sent traffic in the past 2*arp_intervals but
2938                  * haven't xmit and rx traffic in that time interval, select
2939                  * a different slave. slave->jiffies is only updated when
2940                  * a slave first becomes the curr_active_slave - not necessarily
2941                  * after every arp; this ensures the slave has a full 2*delta
2942                  * before being taken out. if a primary is being used, check
2943                  * if it is up and needs to take over as the curr_active_slave
2944                  */
2945                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2946             (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2947              bond_has_ip(bond))) &&
2948                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2949
2950                         slave->link  = BOND_LINK_DOWN;
2951
2952                         if (slave->link_failure_count < UINT_MAX) {
2953                                 slave->link_failure_count++;
2954                         }
2955
2956                         printk(KERN_INFO DRV_NAME
2957                                ": %s: link status down for active interface "
2958                                "%s, disabling it\n",
2959                                bond->dev->name,
2960                                slave->dev->name);
2961
2962                         rtnl_lock();
2963                         write_lock_bh(&bond->curr_slave_lock);
2964
2965                         bond_select_active_slave(bond);
2966                         slave = bond->curr_active_slave;
2967
2968                         write_unlock_bh(&bond->curr_slave_lock);
2969
2970                         rtnl_unlock();
2971
2972                         bond->current_arp_slave = slave;
2973
2974                         if (slave) {
2975                                 slave->jiffies = jiffies;
2976                         }
2977                 } else if ((bond->primary_slave) &&
2978                            (bond->primary_slave != slave) &&
2979                            (bond->primary_slave->link == BOND_LINK_UP)) {
2980                         /* at this point, slave is the curr_active_slave */
2981                         printk(KERN_INFO DRV_NAME
2982                                ": %s: changing from interface %s to primary "
2983                                "interface %s\n",
2984                                bond->dev->name,
2985                                slave->dev->name,
2986                                bond->primary_slave->dev->name);
2987
2988                         /* primary is up so switch to it */
2989                         rtnl_lock();
2990                         write_lock_bh(&bond->curr_slave_lock);
2991                         bond_change_active_slave(bond, bond->primary_slave);
2992                         write_unlock_bh(&bond->curr_slave_lock);
2993
2994                         rtnl_unlock();
2995
2996                         slave = bond->primary_slave;
2997                         slave->jiffies = jiffies;
2998                 } else {
2999                         bond->current_arp_slave = NULL;
3000                 }
3001
3002                 /* the current slave must tx an arp to ensure backup slaves
3003                  * rx traffic
3004                  */
3005                 if (slave && bond_has_ip(bond)) {
3006                         bond_arp_send_all(bond, slave);
3007                 }
3008         }
3009
3010         /* if we don't have a curr_active_slave, search for the next available
3011          * backup slave from the current_arp_slave and make it the candidate
3012          * for becoming the curr_active_slave
3013          */
3014         if (!slave) {
3015                 if (!bond->current_arp_slave) {
3016                         bond->current_arp_slave = bond->first_slave;
3017                 }
3018
3019                 if (bond->current_arp_slave) {
3020                         bond_set_slave_inactive_flags(bond->current_arp_slave);
3021
3022                         /* search for next candidate */
3023                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3024                                 if (IS_UP(slave->dev)) {
3025                                         slave->link = BOND_LINK_BACK;
3026                                         bond_set_slave_active_flags(slave);
3027                                         bond_arp_send_all(bond, slave);
3028                                         slave->jiffies = jiffies;
3029                                         bond->current_arp_slave = slave;
3030                                         break;
3031                                 }
3032
3033                                 /* if the link state is up at this point, we
3034                                  * mark it down - this can happen if we have
3035                                  * simultaneous link failures and
3036                                  * reselect_active_interface doesn't make this
3037                                  * one the current slave so it is still marked
3038                                  * up when it is actually down
3039                                  */
3040                                 if (slave->link == BOND_LINK_UP) {
3041                                         slave->link  = BOND_LINK_DOWN;
3042                                         if (slave->link_failure_count < UINT_MAX) {
3043                                                 slave->link_failure_count++;
3044                                         }
3045
3046                                         bond_set_slave_inactive_flags(slave);
3047
3048                                         printk(KERN_INFO DRV_NAME
3049                                                ": %s: backup interface %s is "
3050                                                "now down.\n",
3051                                                bond->dev->name,
3052                                                slave->dev->name);
3053                                 }
3054                         }
3055                 }
3056         }
3057
3058 re_arm:
3059         if (bond->params.arp_interval) {
3060                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3061         }
3062 out:
3063         read_unlock(&bond->lock);
3064 }
3065
3066 /*------------------------------ proc/seq_file-------------------------------*/
3067
3068 #ifdef CONFIG_PROC_FS
3069
3070 #define SEQ_START_TOKEN ((void *)1)
3071
3072 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3073 {
3074         struct bonding *bond = seq->private;
3075         loff_t off = 0;
3076         struct slave *slave;
3077         int i;
3078
3079         /* make sure the bond won't be taken away */
3080         read_lock(&dev_base_lock);
3081         read_lock_bh(&bond->lock);
3082
3083         if (*pos == 0) {
3084                 return SEQ_START_TOKEN;
3085         }
3086
3087         bond_for_each_slave(bond, slave, i) {
3088                 if (++off == *pos) {
3089                         return slave;
3090                 }
3091         }
3092
3093         return NULL;
3094 }
3095
3096 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3097 {
3098         struct bonding *bond = seq->private;
3099         struct slave *slave = v;
3100
3101         ++*pos;
3102         if (v == SEQ_START_TOKEN) {
3103                 return bond->first_slave;
3104         }
3105
3106         slave = slave->next;
3107
3108         return (slave == bond->first_slave) ? NULL : slave;
3109 }
3110
3111 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3112 {
3113         struct bonding *bond = seq->private;
3114
3115         read_unlock_bh(&bond->lock);
3116         read_unlock(&dev_base_lock);
3117 }
3118
3119 static void bond_info_show_master(struct seq_file *seq)
3120 {
3121         struct bonding *bond = seq->private;
3122         struct slave *curr;
3123         int i;
3124         u32 target;
3125
3126         read_lock(&bond->curr_slave_lock);
3127         curr = bond->curr_active_slave;
3128         read_unlock(&bond->curr_slave_lock);
3129
3130         seq_printf(seq, "Bonding Mode: %s",
3131                    bond_mode_name(bond->params.mode));
3132
3133         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3134             bond->params.fail_over_mac)
3135                 seq_printf(seq, " (fail_over_mac)");
3136
3137         seq_printf(seq, "\n");
3138
3139         if (bond->params.mode == BOND_MODE_XOR ||
3140                 bond->params.mode == BOND_MODE_8023AD) {
3141                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3142                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3143                         bond->params.xmit_policy);
3144         }
3145
3146         if (USES_PRIMARY(bond->params.mode)) {
3147                 seq_printf(seq, "Primary Slave: %s\n",
3148                            (bond->primary_slave) ?
3149                            bond->primary_slave->dev->name : "None");
3150
3151                 seq_printf(seq, "Currently Active Slave: %s\n",
3152                            (curr) ? curr->dev->name : "None");
3153         }
3154
3155         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3156                    "up" : "down");
3157         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3158         seq_printf(seq, "Up Delay (ms): %d\n",
3159                    bond->params.updelay * bond->params.miimon);
3160         seq_printf(seq, "Down Delay (ms): %d\n",
3161                    bond->params.downdelay * bond->params.miimon);
3162
3163
3164         /* ARP information */
3165         if(bond->params.arp_interval > 0) {
3166                 int printed=0;
3167                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3168                                 bond->params.arp_interval);
3169
3170                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3171
3172                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3173                         if (!bond->params.arp_targets[i])
3174                                 continue;
3175                         if (printed)
3176                                 seq_printf(seq, ",");
3177                         target = ntohl(bond->params.arp_targets[i]);
3178                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3179                         printed = 1;
3180                 }
3181                 seq_printf(seq, "\n");
3182         }
3183
3184         if (bond->params.mode == BOND_MODE_8023AD) {
3185                 struct ad_info ad_info;
3186                 DECLARE_MAC_BUF(mac);
3187
3188                 seq_puts(seq, "\n802.3ad info\n");
3189                 seq_printf(seq, "LACP rate: %s\n",
3190                            (bond->params.lacp_fast) ? "fast" : "slow");
3191
3192                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3193                         seq_printf(seq, "bond %s has no active aggregator\n",
3194                                    bond->dev->name);
3195                 } else {
3196                         seq_printf(seq, "Active Aggregator Info:\n");
3197
3198                         seq_printf(seq, "\tAggregator ID: %d\n",
3199                                    ad_info.aggregator_id);
3200                         seq_printf(seq, "\tNumber of ports: %d\n",
3201                                    ad_info.ports);
3202                         seq_printf(seq, "\tActor Key: %d\n",
3203                                    ad_info.actor_key);
3204                         seq_printf(seq, "\tPartner Key: %d\n",
3205                                    ad_info.partner_key);
3206                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3207                                    print_mac(mac, ad_info.partner_system));
3208                 }
3209         }
3210 }
3211
3212 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3213 {
3214         struct bonding *bond = seq->private;
3215         DECLARE_MAC_BUF(mac);
3216
3217         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3218         seq_printf(seq, "MII Status: %s\n",
3219                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3220         seq_printf(seq, "Link Failure Count: %u\n",
3221                    slave->link_failure_count);
3222
3223         seq_printf(seq,
3224                    "Permanent HW addr: %s\n",
3225                    print_mac(mac, slave->perm_hwaddr));
3226
3227         if (bond->params.mode == BOND_MODE_8023AD) {
3228                 const struct aggregator *agg
3229                         = SLAVE_AD_INFO(slave).port.aggregator;
3230
3231                 if (agg) {
3232                         seq_printf(seq, "Aggregator ID: %d\n",
3233                                    agg->aggregator_identifier);
3234                 } else {
3235                         seq_puts(seq, "Aggregator ID: N/A\n");
3236                 }
3237         }
3238 }
3239
3240 static int bond_info_seq_show(struct seq_file *seq, void *v)
3241 {
3242         if (v == SEQ_START_TOKEN) {
3243                 seq_printf(seq, "%s\n", version);
3244                 bond_info_show_master(seq);
3245         } else {
3246                 bond_info_show_slave(seq, v);
3247         }
3248
3249         return 0;
3250 }
3251
3252 static struct seq_operations bond_info_seq_ops = {
3253         .start = bond_info_seq_start,
3254         .next  = bond_info_seq_next,
3255         .stop  = bond_info_seq_stop,
3256         .show  = bond_info_seq_show,
3257 };
3258
3259 static int bond_info_open(struct inode *inode, struct file *file)
3260 {
3261         struct seq_file *seq;
3262         struct proc_dir_entry *proc;
3263         int res;
3264
3265         res = seq_open(file, &bond_info_seq_ops);
3266         if (!res) {
3267                 /* recover the pointer buried in proc_dir_entry data */
3268                 seq = file->private_data;
3269                 proc = PDE(inode);
3270                 seq->private = proc->data;
3271         }
3272
3273         return res;
3274 }
3275
3276 static const struct file_operations bond_info_fops = {
3277         .owner   = THIS_MODULE,
3278         .open    = bond_info_open,
3279         .read    = seq_read,
3280         .llseek  = seq_lseek,
3281         .release = seq_release,
3282 };
3283
3284 static int bond_create_proc_entry(struct bonding *bond)
3285 {
3286         struct net_device *bond_dev = bond->dev;
3287
3288         if (bond_proc_dir) {
3289                 bond->proc_entry = create_proc_entry(bond_dev->name,
3290                                                      S_IRUGO,
3291                                                      bond_proc_dir);
3292                 if (bond->proc_entry == NULL) {
3293                         printk(KERN_WARNING DRV_NAME
3294                                ": Warning: Cannot create /proc/net/%s/%s\n",
3295                                DRV_NAME, bond_dev->name);
3296                 } else {
3297                         bond->proc_entry->data = bond;
3298                         bond->proc_entry->proc_fops = &bond_info_fops;
3299                         bond->proc_entry->owner = THIS_MODULE;
3300                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3301                 }
3302         }
3303
3304         return 0;
3305 }
3306
3307 static void bond_remove_proc_entry(struct bonding *bond)
3308 {
3309         if (bond_proc_dir && bond->proc_entry) {
3310                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3311                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3312                 bond->proc_entry = NULL;
3313         }
3314 }
3315
3316 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3317  * Caller must hold rtnl_lock.
3318  */
3319 static void bond_create_proc_dir(void)
3320 {
3321         int len = strlen(DRV_NAME);
3322
3323         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3324              bond_proc_dir = bond_proc_dir->next) {
3325                 if ((bond_proc_dir->namelen == len) &&
3326                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3327                         break;
3328                 }
3329         }
3330
3331         if (!bond_proc_dir) {
3332                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3333                 if (bond_proc_dir) {
3334                         bond_proc_dir->owner = THIS_MODULE;
3335                 } else {
3336                         printk(KERN_WARNING DRV_NAME
3337                                 ": Warning: cannot create /proc/net/%s\n",
3338                                 DRV_NAME);
3339                 }
3340         }
3341 }
3342
3343 /* Destroy the bonding directory under /proc/net, if empty.
3344  * Caller must hold rtnl_lock.
3345  */
3346 static void bond_destroy_proc_dir(void)
3347 {
3348         struct proc_dir_entry *de;
3349
3350         if (!bond_proc_dir) {
3351                 return;
3352         }
3353
3354         /* verify that the /proc dir is empty */
3355         for (de = bond_proc_dir->subdir; de; de = de->next) {
3356                 /* ignore . and .. */
3357                 if (*(de->name) != '.') {
3358                         break;
3359                 }
3360         }
3361
3362         if (de) {
3363                 if (bond_proc_dir->owner == THIS_MODULE) {
3364                         bond_proc_dir->owner = NULL;
3365                 }
3366         } else {
3367                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3368                 bond_proc_dir = NULL;
3369         }
3370 }
3371 #endif /* CONFIG_PROC_FS */
3372
3373 /*-------------------------- netdev event handling --------------------------*/
3374
3375 /*
3376  * Change device name
3377  */
3378 static int bond_event_changename(struct bonding *bond)
3379 {
3380 #ifdef CONFIG_PROC_FS
3381         bond_remove_proc_entry(bond);
3382         bond_create_proc_entry(bond);
3383 #endif
3384         down_write(&(bonding_rwsem));
3385         bond_destroy_sysfs_entry(bond);
3386         bond_create_sysfs_entry(bond);
3387         up_write(&(bonding_rwsem));
3388         return NOTIFY_DONE;
3389 }
3390
3391 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3392 {
3393         struct bonding *event_bond = bond_dev->priv;
3394
3395         switch (event) {
3396         case NETDEV_CHANGENAME:
3397                 return bond_event_changename(event_bond);
3398         case NETDEV_UNREGISTER:
3399                 /*
3400                  * TODO: remove a bond from the list?
3401                  */
3402                 break;
3403         default:
3404                 break;
3405         }
3406
3407         return NOTIFY_DONE;
3408 }
3409
3410 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3411 {
3412         struct net_device *bond_dev = slave_dev->master;
3413         struct bonding *bond = bond_dev->priv;
3414
3415         switch (event) {
3416         case NETDEV_UNREGISTER:
3417                 if (bond_dev) {
3418                         if (bond->setup_by_slave)
3419                                 bond_release_and_destroy(bond_dev, slave_dev);
3420                         else
3421                                 bond_release(bond_dev, slave_dev);
3422                 }
3423                 break;
3424         case NETDEV_CHANGE:
3425                 /*
3426                  * TODO: is this what we get if somebody
3427                  * sets up a hierarchical bond, then rmmod's
3428                  * one of the slave bonding devices?
3429                  */
3430                 break;
3431         case NETDEV_DOWN:
3432                 /*
3433                  * ... Or is it this?
3434                  */
3435                 break;
3436         case NETDEV_CHANGEMTU:
3437                 /*
3438                  * TODO: Should slaves be allowed to
3439                  * independently alter their MTU?  For
3440                  * an active-backup bond, slaves need
3441                  * not be the same type of device, so
3442                  * MTUs may vary.  For other modes,
3443                  * slaves arguably should have the
3444                  * same MTUs. To do this, we'd need to
3445                  * take over the slave's change_mtu
3446                  * function for the duration of their
3447                  * servitude.
3448                  */
3449                 break;
3450         case NETDEV_CHANGENAME:
3451                 /*
3452                  * TODO: handle changing the primary's name
3453                  */
3454                 break;
3455         case NETDEV_FEAT_CHANGE:
3456                 bond_compute_features(bond);
3457                 break;
3458         default:
3459                 break;
3460         }
3461
3462         return NOTIFY_DONE;
3463 }
3464
3465 /*
3466  * bond_netdev_event: handle netdev notifier chain events.
3467  *
3468  * This function receives events for the netdev chain.  The caller (an
3469  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3470  * locks for us to safely manipulate the slave devices (RTNL lock,
3471  * dev_probe_lock).
3472  */
3473 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3474 {
3475         struct net_device *event_dev = (struct net_device *)ptr;
3476
3477         if (event_dev->nd_net != &init_net)
3478                 return NOTIFY_DONE;
3479
3480         dprintk("event_dev: %s, event: %lx\n",
3481                 (event_dev ? event_dev->name : "None"),
3482                 event);
3483
3484         if (!(event_dev->priv_flags & IFF_BONDING))
3485                 return NOTIFY_DONE;
3486
3487         if (event_dev->flags & IFF_MASTER) {
3488                 dprintk("IFF_MASTER\n");
3489                 return bond_master_netdev_event(event, event_dev);
3490         }
3491
3492         if (event_dev->flags & IFF_SLAVE) {
3493                 dprintk("IFF_SLAVE\n");
3494                 return bond_slave_netdev_event(event, event_dev);
3495         }
3496
3497         return NOTIFY_DONE;
3498 }
3499
3500 /*
3501  * bond_inetaddr_event: handle inetaddr notifier chain events.
3502  *
3503  * We keep track of device IPs primarily to use as source addresses in
3504  * ARP monitor probes (rather than spewing out broadcasts all the time).
3505  *
3506  * We track one IP for the main device (if it has one), plus one per VLAN.
3507  */
3508 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3509 {
3510         struct in_ifaddr *ifa = ptr;
3511         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3512         struct bonding *bond, *bond_next;
3513         struct vlan_entry *vlan, *vlan_next;
3514
3515         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3516                 if (bond->dev == event_dev) {
3517                         switch (event) {
3518                         case NETDEV_UP:
3519                                 bond->master_ip = ifa->ifa_local;
3520                                 return NOTIFY_OK;
3521                         case NETDEV_DOWN:
3522                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3523                                 return NOTIFY_OK;
3524                         default:
3525                                 return NOTIFY_DONE;
3526                         }
3527                 }
3528
3529                 if (list_empty(&bond->vlan_list))
3530                         continue;
3531
3532                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3533                                          vlan_list) {
3534                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3535                         if (vlan_dev == event_dev) {
3536                                 switch (event) {
3537                                 case NETDEV_UP:
3538                                         vlan->vlan_ip = ifa->ifa_local;
3539                                         return NOTIFY_OK;
3540                                 case NETDEV_DOWN:
3541                                         vlan->vlan_ip =
3542                                                 bond_glean_dev_ip(vlan_dev);
3543                                         return NOTIFY_OK;
3544                                 default:
3545                                         return NOTIFY_DONE;
3546                                 }
3547                         }
3548                 }
3549         }
3550         return NOTIFY_DONE;
3551 }
3552
3553 static struct notifier_block bond_netdev_notifier = {
3554         .notifier_call = bond_netdev_event,
3555 };
3556
3557 static struct notifier_block bond_inetaddr_notifier = {
3558         .notifier_call = bond_inetaddr_event,
3559 };
3560
3561 /*-------------------------- Packet type handling ---------------------------*/
3562
3563 /* register to receive lacpdus on a bond */
3564 static void bond_register_lacpdu(struct bonding *bond)
3565 {
3566         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3567
3568         /* initialize packet type */
3569         pk_type->type = PKT_TYPE_LACPDU;
3570         pk_type->dev = bond->dev;
3571         pk_type->func = bond_3ad_lacpdu_recv;
3572
3573         dev_add_pack(pk_type);
3574 }
3575
3576 /* unregister to receive lacpdus on a bond */
3577 static void bond_unregister_lacpdu(struct bonding *bond)
3578 {
3579         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3580 }
3581
3582 void bond_register_arp(struct bonding *bond)
3583 {
3584         struct packet_type *pt = &bond->arp_mon_pt;
3585
3586         if (pt->type)
3587                 return;
3588
3589         pt->type = htons(ETH_P_ARP);
3590         pt->dev = bond->dev;
3591         pt->func = bond_arp_rcv;
3592         dev_add_pack(pt);
3593 }
3594
3595 void bond_unregister_arp(struct bonding *bond)
3596 {
3597         struct packet_type *pt = &bond->arp_mon_pt;
3598
3599         dev_remove_pack(pt);
3600         pt->type = 0;
3601 }
3602
3603 /*---------------------------- Hashing Policies -----------------------------*/
3604
3605 /*
3606  * Hash for the output device based upon layer 3 and layer 4 data. If
3607  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3608  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3609  */
3610 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3611                                     struct net_device *bond_dev, int count)
3612 {
3613         struct ethhdr *data = (struct ethhdr *)skb->data;
3614         struct iphdr *iph = ip_hdr(skb);
3615         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3616         int layer4_xor = 0;
3617
3618         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3619                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3620                     (iph->protocol == IPPROTO_TCP ||
3621                      iph->protocol == IPPROTO_UDP)) {
3622                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3623                 }
3624                 return (layer4_xor ^
3625                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3626
3627         }
3628
3629         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3630 }
3631
3632 /*
3633  * Hash for the output device based upon layer 2 data
3634  */
3635 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3636                                    struct net_device *bond_dev, int count)
3637 {
3638         struct ethhdr *data = (struct ethhdr *)skb->data;
3639
3640         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3641 }
3642
3643 /*-------------------------- Device entry points ----------------------------*/
3644
3645 static int bond_open(struct net_device *bond_dev)
3646 {
3647         struct bonding *bond = bond_dev->priv;
3648
3649         bond->kill_timers = 0;
3650
3651         if ((bond->params.mode == BOND_MODE_TLB) ||
3652             (bond->params.mode == BOND_MODE_ALB)) {
3653                 /* bond_alb_initialize must be called before the timer
3654                  * is started.
3655                  */
3656                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3657                         /* something went wrong - fail the open operation */
3658                         return -1;
3659                 }
3660
3661                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3662                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3663         }
3664
3665         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3666                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3667                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3668         }
3669
3670         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3671                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3672                         INIT_DELAYED_WORK(&bond->arp_work,
3673                                           bond_activebackup_arp_mon);
3674                 else
3675                         INIT_DELAYED_WORK(&bond->arp_work,
3676                                           bond_loadbalance_arp_mon);
3677
3678                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3679                 if (bond->params.arp_validate)
3680                         bond_register_arp(bond);
3681         }
3682
3683         if (bond->params.mode == BOND_MODE_8023AD) {
3684                 INIT_DELAYED_WORK(&bond->ad_work, bond_alb_monitor);
3685                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3686                 /* register to receive LACPDUs */
3687                 bond_register_lacpdu(bond);
3688         }
3689
3690         return 0;
3691 }
3692
3693 static int bond_close(struct net_device *bond_dev)
3694 {
3695         struct bonding *bond = bond_dev->priv;
3696
3697         if (bond->params.mode == BOND_MODE_8023AD) {
3698                 /* Unregister the receive of LACPDUs */
3699                 bond_unregister_lacpdu(bond);
3700         }
3701
3702         if (bond->params.arp_validate)
3703                 bond_unregister_arp(bond);
3704
3705         write_lock_bh(&bond->lock);
3706
3707
3708         /* signal timers not to re-arm */
3709         bond->kill_timers = 1;
3710
3711         write_unlock_bh(&bond->lock);
3712
3713         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3714                 cancel_delayed_work(&bond->mii_work);
3715         }
3716
3717         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3718                 cancel_delayed_work(&bond->arp_work);
3719         }
3720
3721         switch (bond->params.mode) {
3722         case BOND_MODE_8023AD:
3723                 cancel_delayed_work(&bond->ad_work);
3724                 break;
3725         case BOND_MODE_TLB:
3726         case BOND_MODE_ALB:
3727                 cancel_delayed_work(&bond->alb_work);
3728                 break;
3729         default:
3730                 break;
3731         }
3732
3733
3734         if ((bond->params.mode == BOND_MODE_TLB) ||
3735             (bond->params.mode == BOND_MODE_ALB)) {
3736                 /* Must be called only after all
3737                  * slaves have been released
3738                  */
3739                 bond_alb_deinitialize(bond);
3740         }
3741
3742         return 0;
3743 }
3744
3745 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3746 {
3747         struct bonding *bond = bond_dev->priv;
3748         struct net_device_stats *stats = &(bond->stats), *sstats;
3749         struct slave *slave;
3750         int i;
3751
3752         memset(stats, 0, sizeof(struct net_device_stats));
3753
3754         read_lock_bh(&bond->lock);
3755
3756         bond_for_each_slave(bond, slave, i) {
3757                 sstats = slave->dev->get_stats(slave->dev);
3758                 stats->rx_packets += sstats->rx_packets;
3759                 stats->rx_bytes += sstats->rx_bytes;
3760                 stats->rx_errors += sstats->rx_errors;
3761                 stats->rx_dropped += sstats->rx_dropped;
3762
3763                 stats->tx_packets += sstats->tx_packets;
3764                 stats->tx_bytes += sstats->tx_bytes;
3765                 stats->tx_errors += sstats->tx_errors;
3766                 stats->tx_dropped += sstats->tx_dropped;
3767
3768                 stats->multicast += sstats->multicast;
3769                 stats->collisions += sstats->collisions;
3770
3771                 stats->rx_length_errors += sstats->rx_length_errors;
3772                 stats->rx_over_errors += sstats->rx_over_errors;
3773                 stats->rx_crc_errors += sstats->rx_crc_errors;
3774                 stats->rx_frame_errors += sstats->rx_frame_errors;
3775                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3776                 stats->rx_missed_errors += sstats->rx_missed_errors;
3777
3778                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3779                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3780                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3781                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3782                 stats->tx_window_errors += sstats->tx_window_errors;
3783         }
3784
3785         read_unlock_bh(&bond->lock);
3786
3787         return stats;
3788 }
3789
3790 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3791 {
3792         struct net_device *slave_dev = NULL;
3793         struct ifbond k_binfo;
3794         struct ifbond __user *u_binfo = NULL;
3795         struct ifslave k_sinfo;
3796         struct ifslave __user *u_sinfo = NULL;
3797         struct mii_ioctl_data *mii = NULL;
3798         int res = 0;
3799
3800         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3801                 bond_dev->name, cmd);
3802
3803         switch (cmd) {
3804         case SIOCGMIIPHY:
3805                 mii = if_mii(ifr);
3806                 if (!mii) {
3807                         return -EINVAL;
3808                 }
3809                 mii->phy_id = 0;
3810                 /* Fall Through */
3811         case SIOCGMIIREG:
3812                 /*
3813                  * We do this again just in case we were called by SIOCGMIIREG
3814                  * instead of SIOCGMIIPHY.
3815                  */
3816                 mii = if_mii(ifr);
3817                 if (!mii) {
3818                         return -EINVAL;
3819                 }
3820
3821                 if (mii->reg_num == 1) {
3822                         struct bonding *bond = bond_dev->priv;
3823                         mii->val_out = 0;
3824                         read_lock_bh(&bond->lock);
3825                         read_lock(&bond->curr_slave_lock);
3826                         if (netif_carrier_ok(bond->dev)) {
3827                                 mii->val_out = BMSR_LSTATUS;
3828                         }
3829                         read_unlock(&bond->curr_slave_lock);
3830                         read_unlock_bh(&bond->lock);
3831                 }
3832
3833                 return 0;
3834         case BOND_INFO_QUERY_OLD:
3835         case SIOCBONDINFOQUERY:
3836                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3837
3838                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3839                         return -EFAULT;
3840                 }
3841
3842                 res = bond_info_query(bond_dev, &k_binfo);
3843                 if (res == 0) {
3844                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3845                                 return -EFAULT;
3846                         }
3847                 }
3848
3849                 return res;
3850         case BOND_SLAVE_INFO_QUERY_OLD:
3851         case SIOCBONDSLAVEINFOQUERY:
3852                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3853
3854                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3855                         return -EFAULT;
3856                 }
3857
3858                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3859                 if (res == 0) {
3860                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3861                                 return -EFAULT;
3862                         }
3863                 }
3864
3865                 return res;
3866         default:
3867                 /* Go on */
3868                 break;
3869         }
3870
3871         if (!capable(CAP_NET_ADMIN)) {
3872                 return -EPERM;
3873         }
3874
3875         down_write(&(bonding_rwsem));
3876         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3877
3878         dprintk("slave_dev=%p: \n", slave_dev);
3879
3880         if (!slave_dev) {
3881                 res = -ENODEV;
3882         } else {
3883                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3884                 switch (cmd) {
3885                 case BOND_ENSLAVE_OLD:
3886                 case SIOCBONDENSLAVE:
3887                         res = bond_enslave(bond_dev, slave_dev);
3888                         break;
3889                 case BOND_RELEASE_OLD:
3890                 case SIOCBONDRELEASE:
3891                         res = bond_release(bond_dev, slave_dev);
3892                         break;
3893                 case BOND_SETHWADDR_OLD:
3894                 case SIOCBONDSETHWADDR:
3895                         res = bond_sethwaddr(bond_dev, slave_dev);
3896                         break;
3897                 case BOND_CHANGE_ACTIVE_OLD:
3898                 case SIOCBONDCHANGEACTIVE:
3899                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3900                         break;
3901                 default:
3902                         res = -EOPNOTSUPP;
3903                 }
3904
3905                 dev_put(slave_dev);
3906         }
3907
3908         up_write(&(bonding_rwsem));
3909         return res;
3910 }
3911
3912 static void bond_set_multicast_list(struct net_device *bond_dev)
3913 {
3914         struct bonding *bond = bond_dev->priv;
3915         struct dev_mc_list *dmi;
3916
3917         write_lock_bh(&bond->lock);
3918
3919         /*
3920          * Do promisc before checking multicast_mode
3921          */
3922         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3923                 bond_set_promiscuity(bond, 1);
3924         }
3925
3926         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3927                 bond_set_promiscuity(bond, -1);
3928         }
3929
3930         /* set allmulti flag to slaves */
3931         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3932                 bond_set_allmulti(bond, 1);
3933         }
3934
3935         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3936                 bond_set_allmulti(bond, -1);
3937         }
3938
3939         bond->flags = bond_dev->flags;
3940
3941         /* looking for addresses to add to slaves' mc list */
3942         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3943                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3944                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3945                 }
3946         }
3947
3948         /* looking for addresses to delete from slaves' list */
3949         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3950                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3951                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3952                 }
3953         }
3954
3955         /* save master's multicast list */
3956         bond_mc_list_destroy(bond);
3957         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3958
3959         write_unlock_bh(&bond->lock);
3960 }
3961
3962 /*
3963  * Change the MTU of all of a master's slaves to match the master
3964  */
3965 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3966 {
3967         struct bonding *bond = bond_dev->priv;
3968         struct slave *slave, *stop_at;
3969         int res = 0;
3970         int i;
3971
3972         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3973                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3974
3975         /* Can't hold bond->lock with bh disabled here since
3976          * some base drivers panic. On the other hand we can't
3977          * hold bond->lock without bh disabled because we'll
3978          * deadlock. The only solution is to rely on the fact
3979          * that we're under rtnl_lock here, and the slaves
3980          * list won't change. This doesn't solve the problem
3981          * of setting the slave's MTU while it is
3982          * transmitting, but the assumption is that the base
3983          * driver can handle that.
3984          *
3985          * TODO: figure out a way to safely iterate the slaves
3986          * list, but without holding a lock around the actual
3987          * call to the base driver.
3988          */
3989
3990         bond_for_each_slave(bond, slave, i) {
3991                 dprintk("s %p s->p %p c_m %p\n", slave,
3992                         slave->prev, slave->dev->change_mtu);
3993
3994                 res = dev_set_mtu(slave->dev, new_mtu);
3995
3996                 if (res) {
3997                         /* If we failed to set the slave's mtu to the new value
3998                          * we must abort the operation even in ACTIVE_BACKUP
3999                          * mode, because if we allow the backup slaves to have
4000                          * different mtu values than the active slave we'll
4001                          * need to change their mtu when doing a failover. That
4002                          * means changing their mtu from timer context, which
4003                          * is probably not a good idea.
4004                          */
4005                         dprintk("err %d %s\n", res, slave->dev->name);
4006                         goto unwind;
4007                 }
4008         }
4009
4010         bond_dev->mtu = new_mtu;
4011
4012         return 0;
4013
4014 unwind:
4015         /* unwind from head to the slave that failed */
4016         stop_at = slave;
4017         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4018                 int tmp_res;
4019
4020                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4021                 if (tmp_res) {
4022                         dprintk("unwind err %d dev %s\n", tmp_res,
4023                                 slave->dev->name);
4024                 }
4025         }
4026
4027         return res;
4028 }
4029
4030 /*
4031  * Change HW address
4032  *
4033  * Note that many devices must be down to change the HW address, and
4034  * downing the master releases all slaves.  We can make bonds full of
4035  * bonding devices to test this, however.
4036  */
4037 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4038 {
4039         struct bonding *bond = bond_dev->priv;
4040         struct sockaddr *sa = addr, tmp_sa;
4041         struct slave *slave, *stop_at;
4042         int res = 0;
4043         int i;
4044
4045         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4046
4047         /*
4048          * If fail_over_mac is enabled, do nothing and return success.
4049          * Returning an error causes ifenslave to fail.
4050          */
4051         if (bond->params.fail_over_mac)
4052                 return 0;
4053
4054         if (!is_valid_ether_addr(sa->sa_data)) {
4055                 return -EADDRNOTAVAIL;
4056         }
4057
4058         /* Can't hold bond->lock with bh disabled here since
4059          * some base drivers panic. On the other hand we can't
4060          * hold bond->lock without bh disabled because we'll
4061          * deadlock. The only solution is to rely on the fact
4062          * that we're under rtnl_lock here, and the slaves
4063          * list won't change. This doesn't solve the problem
4064          * of setting the slave's hw address while it is
4065          * transmitting, but the assumption is that the base
4066          * driver can handle that.
4067          *
4068          * TODO: figure out a way to safely iterate the slaves
4069          * list, but without holding a lock around the actual
4070          * call to the base driver.
4071          */
4072
4073         bond_for_each_slave(bond, slave, i) {
4074                 dprintk("slave %p %s\n", slave, slave->dev->name);
4075
4076                 if (slave->dev->set_mac_address == NULL) {
4077                         res = -EOPNOTSUPP;
4078                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4079                         goto unwind;
4080                 }
4081
4082                 res = dev_set_mac_address(slave->dev, addr);
4083                 if (res) {
4084                         /* TODO: consider downing the slave
4085                          * and retry ?
4086                          * User should expect communications
4087                          * breakage anyway until ARP finish
4088                          * updating, so...
4089                          */
4090                         dprintk("err %d %s\n", res, slave->dev->name);
4091                         goto unwind;
4092                 }
4093         }
4094
4095         /* success */
4096         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4097         return 0;
4098
4099 unwind:
4100         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4101         tmp_sa.sa_family = bond_dev->type;
4102
4103         /* unwind from head to the slave that failed */
4104         stop_at = slave;
4105         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4106                 int tmp_res;
4107
4108                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4109                 if (tmp_res) {
4110                         dprintk("unwind err %d dev %s\n", tmp_res,
4111                                 slave->dev->name);
4112                 }
4113         }
4114
4115         return res;
4116 }
4117
4118 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4119 {
4120         struct bonding *bond = bond_dev->priv;
4121         struct slave *slave, *start_at;
4122         int i, slave_no, res = 1;
4123
4124         read_lock(&bond->lock);
4125
4126         if (!BOND_IS_OK(bond)) {
4127                 goto out;
4128         }
4129
4130         /*
4131          * Concurrent TX may collide on rr_tx_counter; we accept that
4132          * as being rare enough not to justify using an atomic op here
4133          */
4134         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4135
4136         bond_for_each_slave(bond, slave, i) {
4137                 slave_no--;
4138                 if (slave_no < 0) {
4139                         break;
4140                 }
4141         }
4142
4143         start_at = slave;
4144         bond_for_each_slave_from(bond, slave, i, start_at) {
4145                 if (IS_UP(slave->dev) &&
4146                     (slave->link == BOND_LINK_UP) &&
4147                     (slave->state == BOND_STATE_ACTIVE)) {
4148                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4149                         break;
4150                 }
4151         }
4152
4153 out:
4154         if (res) {
4155                 /* no suitable interface, frame not sent */
4156                 dev_kfree_skb(skb);
4157         }
4158         read_unlock(&bond->lock);
4159         return 0;
4160 }
4161
4162
4163 /*
4164  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4165  * the bond has a usable interface.
4166  */
4167 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4168 {
4169         struct bonding *bond = bond_dev->priv;
4170         int res = 1;
4171
4172         read_lock(&bond->lock);
4173         read_lock(&bond->curr_slave_lock);
4174
4175         if (!BOND_IS_OK(bond)) {
4176                 goto out;
4177         }
4178
4179         if (!bond->curr_active_slave)
4180                 goto out;
4181
4182         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4183
4184 out:
4185         if (res) {
4186                 /* no suitable interface, frame not sent */
4187                 dev_kfree_skb(skb);
4188         }
4189         read_unlock(&bond->curr_slave_lock);
4190         read_unlock(&bond->lock);
4191         return 0;
4192 }
4193
4194 /*
4195  * In bond_xmit_xor() , we determine the output device by using a pre-
4196  * determined xmit_hash_policy(), If the selected device is not enabled,
4197  * find the next active slave.
4198  */
4199 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4200 {
4201         struct bonding *bond = bond_dev->priv;
4202         struct slave *slave, *start_at;
4203         int slave_no;
4204         int i;
4205         int res = 1;
4206
4207         read_lock(&bond->lock);
4208
4209         if (!BOND_IS_OK(bond)) {
4210                 goto out;
4211         }
4212
4213         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4214
4215         bond_for_each_slave(bond, slave, i) {
4216                 slave_no--;
4217                 if (slave_no < 0) {
4218                         break;
4219                 }
4220         }
4221
4222         start_at = slave;
4223
4224         bond_for_each_slave_from(bond, slave, i, start_at) {
4225                 if (IS_UP(slave->dev) &&
4226                     (slave->link == BOND_LINK_UP) &&
4227                     (slave->state == BOND_STATE_ACTIVE)) {
4228                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4229                         break;
4230                 }
4231         }
4232
4233 out:
4234         if (res) {
4235                 /* no suitable interface, frame not sent */
4236                 dev_kfree_skb(skb);
4237         }
4238         read_unlock(&bond->lock);
4239         return 0;
4240 }
4241
4242 /*
4243  * in broadcast mode, we send everything to all usable interfaces.
4244  */
4245 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4246 {
4247         struct bonding *bond = bond_dev->priv;
4248         struct slave *slave, *start_at;
4249         struct net_device *tx_dev = NULL;
4250         int i;
4251         int res = 1;
4252
4253         read_lock(&bond->lock);
4254
4255         if (!BOND_IS_OK(bond)) {
4256                 goto out;
4257         }
4258
4259         read_lock(&bond->curr_slave_lock);
4260         start_at = bond->curr_active_slave;
4261         read_unlock(&bond->curr_slave_lock);
4262
4263         if (!start_at) {
4264                 goto out;
4265         }
4266
4267         bond_for_each_slave_from(bond, slave, i, start_at) {
4268                 if (IS_UP(slave->dev) &&
4269                     (slave->link == BOND_LINK_UP) &&
4270                     (slave->state == BOND_STATE_ACTIVE)) {
4271                         if (tx_dev) {
4272                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4273                                 if (!skb2) {
4274                                         printk(KERN_ERR DRV_NAME
4275                                                ": %s: Error: bond_xmit_broadcast(): "
4276                                                "skb_clone() failed\n",
4277                                                bond_dev->name);
4278                                         continue;
4279                                 }
4280
4281                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4282                                 if (res) {
4283                                         dev_kfree_skb(skb2);
4284                                         continue;
4285                                 }
4286                         }
4287                         tx_dev = slave->dev;
4288                 }
4289         }
4290
4291         if (tx_dev) {
4292                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4293         }
4294
4295 out:
4296         if (res) {
4297                 /* no suitable interface, frame not sent */
4298                 dev_kfree_skb(skb);
4299         }
4300         /* frame sent to all suitable interfaces */
4301         read_unlock(&bond->lock);
4302         return 0;
4303 }
4304
4305 /*------------------------- Device initialization ---------------------------*/
4306
4307 /*
4308  * set bond mode specific net device operations
4309  */
4310 void bond_set_mode_ops(struct bonding *bond, int mode)
4311 {
4312         struct net_device *bond_dev = bond->dev;
4313
4314         switch (mode) {
4315         case BOND_MODE_ROUNDROBIN:
4316                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4317                 break;
4318         case BOND_MODE_ACTIVEBACKUP:
4319                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4320                 break;
4321         case BOND_MODE_XOR:
4322                 bond_dev->hard_start_xmit = bond_xmit_xor;
4323                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4324                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4325                 else
4326                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4327                 break;
4328         case BOND_MODE_BROADCAST:
4329                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4330                 break;
4331         case BOND_MODE_8023AD:
4332                 bond_set_master_3ad_flags(bond);
4333                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4334                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4335                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4336                 else
4337                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4338                 break;
4339         case BOND_MODE_ALB:
4340                 bond_set_master_alb_flags(bond);
4341                 /* FALLTHRU */
4342         case BOND_MODE_TLB:
4343                 bond_dev->hard_start_xmit = bond_alb_xmit;
4344                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4345                 break;
4346         default:
4347                 /* Should never happen, mode already checked */
4348                 printk(KERN_ERR DRV_NAME
4349                        ": %s: Error: Unknown bonding mode %d\n",
4350                        bond_dev->name,
4351                        mode);
4352                 break;
4353         }
4354 }
4355
4356 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4357                                     struct ethtool_drvinfo *drvinfo)
4358 {
4359         strncpy(drvinfo->driver, DRV_NAME, 32);
4360         strncpy(drvinfo->version, DRV_VERSION, 32);
4361         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4362 }
4363
4364 static const struct ethtool_ops bond_ethtool_ops = {
4365         .get_drvinfo            = bond_ethtool_get_drvinfo,
4366 };
4367
4368 /*
4369  * Does not allocate but creates a /proc entry.
4370  * Allowed to fail.
4371  */
4372 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4373 {
4374         struct bonding *bond = bond_dev->priv;
4375
4376         dprintk("Begin bond_init for %s\n", bond_dev->name);
4377
4378         /* initialize rwlocks */
4379         rwlock_init(&bond->lock);
4380         rwlock_init(&bond->curr_slave_lock);
4381
4382         bond->params = *params; /* copy params struct */
4383
4384         bond->wq = create_singlethread_workqueue(bond_dev->name);
4385         if (!bond->wq)
4386                 return -ENOMEM;
4387
4388         /* Initialize pointers */
4389         bond->first_slave = NULL;
4390         bond->curr_active_slave = NULL;
4391         bond->current_arp_slave = NULL;
4392         bond->primary_slave = NULL;
4393         bond->dev = bond_dev;
4394         bond->send_grat_arp = 0;
4395         bond->setup_by_slave = 0;
4396         INIT_LIST_HEAD(&bond->vlan_list);
4397
4398         /* Initialize the device entry points */
4399         bond_dev->open = bond_open;
4400         bond_dev->stop = bond_close;
4401         bond_dev->get_stats = bond_get_stats;
4402         bond_dev->do_ioctl = bond_do_ioctl;
4403         bond_dev->ethtool_ops = &bond_ethtool_ops;
4404         bond_dev->set_multicast_list = bond_set_multicast_list;
4405         bond_dev->change_mtu = bond_change_mtu;
4406         bond_dev->set_mac_address = bond_set_mac_address;
4407
4408         bond_set_mode_ops(bond, bond->params.mode);
4409
4410         bond_dev->destructor = free_netdev;
4411
4412         /* Initialize the device options */
4413         bond_dev->tx_queue_len = 0;
4414         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4415         bond_dev->priv_flags |= IFF_BONDING;
4416
4417         /* At first, we block adding VLANs. That's the only way to
4418          * prevent problems that occur when adding VLANs over an
4419          * empty bond. The block will be removed once non-challenged
4420          * slaves are enslaved.
4421          */
4422         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4423
4424         /* don't acquire bond device's netif_tx_lock when
4425          * transmitting */
4426         bond_dev->features |= NETIF_F_LLTX;
4427
4428         /* By default, we declare the bond to be fully
4429          * VLAN hardware accelerated capable. Special
4430          * care is taken in the various xmit functions
4431          * when there are slaves that are not hw accel
4432          * capable
4433          */
4434         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4435         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4436         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4437         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4438                                NETIF_F_HW_VLAN_RX |
4439                                NETIF_F_HW_VLAN_FILTER);
4440
4441 #ifdef CONFIG_PROC_FS
4442         bond_create_proc_entry(bond);
4443 #endif
4444         list_add_tail(&bond->bond_list, &bond_dev_list);
4445
4446         return 0;
4447 }
4448
4449 /* De-initialize device specific data.
4450  * Caller must hold rtnl_lock.
4451  */
4452 void bond_deinit(struct net_device *bond_dev)
4453 {
4454         struct bonding *bond = bond_dev->priv;
4455
4456         list_del(&bond->bond_list);
4457
4458 #ifdef CONFIG_PROC_FS
4459         bond_remove_proc_entry(bond);
4460 #endif
4461 }
4462
4463 /* Unregister and free all bond devices.
4464  * Caller must hold rtnl_lock.
4465  */
4466 static void bond_free_all(void)
4467 {
4468         struct bonding *bond, *nxt;
4469
4470         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4471                 struct net_device *bond_dev = bond->dev;
4472
4473                 bond_mc_list_destroy(bond);
4474                 /* Release the bonded slaves */
4475                 bond_release_all(bond_dev);
4476                 bond_deinit(bond_dev);
4477                 unregister_netdevice(bond_dev);
4478         }
4479
4480 #ifdef CONFIG_PROC_FS
4481         bond_destroy_proc_dir();
4482 #endif
4483 }
4484
4485 /*------------------------- Module initialization ---------------------------*/
4486
4487 /*
4488  * Convert string input module parms.  Accept either the
4489  * number of the mode or its string name.
4490  */
4491 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4492 {
4493         int i;
4494
4495         for (i = 0; tbl[i].modename; i++) {
4496                 if ((isdigit(*mode_arg) &&
4497                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4498                     (strncmp(mode_arg, tbl[i].modename,
4499                              strlen(tbl[i].modename)) == 0)) {
4500                         return tbl[i].mode;
4501                 }
4502         }
4503
4504         return -1;
4505 }
4506
4507 static int bond_check_params(struct bond_params *params)
4508 {
4509         int arp_validate_value;
4510
4511         /*
4512          * Convert string parameters.
4513          */
4514         if (mode) {
4515                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4516                 if (bond_mode == -1) {
4517                         printk(KERN_ERR DRV_NAME
4518                                ": Error: Invalid bonding mode \"%s\"\n",
4519                                mode == NULL ? "NULL" : mode);
4520                         return -EINVAL;
4521                 }
4522         }
4523
4524         if (xmit_hash_policy) {
4525                 if ((bond_mode != BOND_MODE_XOR) &&
4526                     (bond_mode != BOND_MODE_8023AD)) {
4527                         printk(KERN_INFO DRV_NAME
4528                                ": xor_mode param is irrelevant in mode %s\n",
4529                                bond_mode_name(bond_mode));
4530                 } else {
4531                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4532                                                         xmit_hashtype_tbl);
4533                         if (xmit_hashtype == -1) {
4534                                 printk(KERN_ERR DRV_NAME
4535                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4536                                 xmit_hash_policy == NULL ? "NULL" :
4537                                        xmit_hash_policy);
4538                                 return -EINVAL;
4539                         }
4540                 }
4541         }
4542
4543         if (lacp_rate) {
4544                 if (bond_mode != BOND_MODE_8023AD) {
4545                         printk(KERN_INFO DRV_NAME
4546                                ": lacp_rate param is irrelevant in mode %s\n",
4547                                bond_mode_name(bond_mode));
4548                 } else {
4549                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4550                         if (lacp_fast == -1) {
4551                                 printk(KERN_ERR DRV_NAME
4552                                        ": Error: Invalid lacp rate \"%s\"\n",
4553                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4554                                 return -EINVAL;
4555                         }
4556                 }
4557         }
4558
4559         if (max_bonds < 1 || max_bonds > INT_MAX) {
4560                 printk(KERN_WARNING DRV_NAME
4561                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4562                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4563                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4564                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4565         }
4566
4567         if (miimon < 0) {
4568                 printk(KERN_WARNING DRV_NAME
4569                        ": Warning: miimon module parameter (%d), "
4570                        "not in range 0-%d, so it was reset to %d\n",
4571                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4572                 miimon = BOND_LINK_MON_INTERV;
4573         }
4574
4575         if (updelay < 0) {
4576                 printk(KERN_WARNING DRV_NAME
4577                        ": Warning: updelay module parameter (%d), "
4578                        "not in range 0-%d, so it was reset to 0\n",
4579                        updelay, INT_MAX);
4580                 updelay = 0;
4581         }
4582
4583         if (downdelay < 0) {
4584                 printk(KERN_WARNING DRV_NAME
4585                        ": Warning: downdelay module parameter (%d), "
4586                        "not in range 0-%d, so it was reset to 0\n",
4587                        downdelay, INT_MAX);
4588                 downdelay = 0;
4589         }
4590
4591         if ((use_carrier != 0) && (use_carrier != 1)) {
4592                 printk(KERN_WARNING DRV_NAME
4593                        ": Warning: use_carrier module parameter (%d), "
4594                        "not of valid value (0/1), so it was set to 1\n",
4595                        use_carrier);
4596                 use_carrier = 1;
4597         }
4598
4599         /* reset values for 802.3ad */
4600         if (bond_mode == BOND_MODE_8023AD) {
4601                 if (!miimon) {
4602                         printk(KERN_WARNING DRV_NAME
4603                                ": Warning: miimon must be specified, "
4604                                "otherwise bonding will not detect link "
4605                                "failure, speed and duplex which are "
4606                                "essential for 802.3ad operation\n");
4607                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4608                         miimon = 100;
4609                 }
4610         }
4611
4612         /* reset values for TLB/ALB */
4613         if ((bond_mode == BOND_MODE_TLB) ||
4614             (bond_mode == BOND_MODE_ALB)) {
4615                 if (!miimon) {
4616                         printk(KERN_WARNING DRV_NAME
4617                                ": Warning: miimon must be specified, "
4618                                "otherwise bonding will not detect link "
4619                                "failure and link speed which are essential "
4620                                "for TLB/ALB load balancing\n");
4621                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4622                         miimon = 100;
4623                 }
4624         }
4625
4626         if (bond_mode == BOND_MODE_ALB) {
4627                 printk(KERN_NOTICE DRV_NAME
4628                        ": In ALB mode you might experience client "
4629                        "disconnections upon reconnection of a link if the "
4630                        "bonding module updelay parameter (%d msec) is "
4631                        "incompatible with the forwarding delay time of the "
4632                        "switch\n",
4633                        updelay);
4634         }
4635
4636         if (!miimon) {
4637                 if (updelay || downdelay) {
4638                         /* just warn the user the up/down delay will have
4639                          * no effect since miimon is zero...
4640                          */
4641                         printk(KERN_WARNING DRV_NAME
4642                                ": Warning: miimon module parameter not set "
4643                                "and updelay (%d) or downdelay (%d) module "
4644                                "parameter is set; updelay and downdelay have "
4645                                "no effect unless miimon is set\n",
4646                                updelay, downdelay);
4647                 }
4648         } else {
4649                 /* don't allow arp monitoring */
4650                 if (arp_interval) {
4651                         printk(KERN_WARNING DRV_NAME
4652                                ": Warning: miimon (%d) and arp_interval (%d) "
4653                                "can't be used simultaneously, disabling ARP "
4654                                "monitoring\n",
4655                                miimon, arp_interval);
4656                         arp_interval = 0;
4657                 }
4658
4659                 if ((updelay % miimon) != 0) {
4660                         printk(KERN_WARNING DRV_NAME
4661                                ": Warning: updelay (%d) is not a multiple "
4662                                "of miimon (%d), updelay rounded to %d ms\n",
4663                                updelay, miimon, (updelay / miimon) * miimon);
4664                 }
4665
4666                 updelay /= miimon;
4667
4668                 if ((downdelay % miimon) != 0) {
4669                         printk(KERN_WARNING DRV_NAME
4670                                ": Warning: downdelay (%d) is not a multiple "
4671                                "of miimon (%d), downdelay rounded to %d ms\n",
4672                                downdelay, miimon,
4673                                (downdelay / miimon) * miimon);
4674                 }
4675
4676                 downdelay /= miimon;
4677         }
4678
4679         if (arp_interval < 0) {
4680                 printk(KERN_WARNING DRV_NAME
4681                        ": Warning: arp_interval module parameter (%d) "
4682                        ", not in range 0-%d, so it was reset to %d\n",
4683                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4684                 arp_interval = BOND_LINK_ARP_INTERV;
4685         }
4686
4687         for (arp_ip_count = 0;
4688              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4689              arp_ip_count++) {
4690                 /* not complete check, but should be good enough to
4691                    catch mistakes */
4692                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4693                         printk(KERN_WARNING DRV_NAME
4694                                ": Warning: bad arp_ip_target module parameter "
4695                                "(%s), ARP monitoring will not be performed\n",
4696                                arp_ip_target[arp_ip_count]);
4697                         arp_interval = 0;
4698                 } else {
4699                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4700                         arp_target[arp_ip_count] = ip;
4701                 }
4702         }
4703
4704         if (arp_interval && !arp_ip_count) {
4705                 /* don't allow arping if no arp_ip_target given... */
4706                 printk(KERN_WARNING DRV_NAME
4707                        ": Warning: arp_interval module parameter (%d) "
4708                        "specified without providing an arp_ip_target "
4709                        "parameter, arp_interval was reset to 0\n",
4710                        arp_interval);
4711                 arp_interval = 0;
4712         }
4713
4714         if (arp_validate) {
4715                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4716                         printk(KERN_ERR DRV_NAME
4717                ": arp_validate only supported in active-backup mode\n");
4718                         return -EINVAL;
4719                 }
4720                 if (!arp_interval) {
4721                         printk(KERN_ERR DRV_NAME
4722                                ": arp_validate requires arp_interval\n");
4723                         return -EINVAL;
4724                 }
4725
4726                 arp_validate_value = bond_parse_parm(arp_validate,
4727                                                      arp_validate_tbl);
4728                 if (arp_validate_value == -1) {
4729                         printk(KERN_ERR DRV_NAME
4730                                ": Error: invalid arp_validate \"%s\"\n",
4731                                arp_validate == NULL ? "NULL" : arp_validate);
4732                         return -EINVAL;
4733                 }
4734         } else
4735                 arp_validate_value = 0;
4736
4737         if (miimon) {
4738                 printk(KERN_INFO DRV_NAME
4739                        ": MII link monitoring set to %d ms\n",
4740                        miimon);
4741         } else if (arp_interval) {
4742                 int i;
4743
4744                 printk(KERN_INFO DRV_NAME
4745                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4746                        arp_interval,
4747                        arp_validate_tbl[arp_validate_value].modename,
4748                        arp_ip_count);
4749
4750                 for (i = 0; i < arp_ip_count; i++)
4751                         printk (" %s", arp_ip_target[i]);
4752
4753                 printk("\n");
4754
4755         } else {
4756                 /* miimon and arp_interval not set, we need one so things
4757                  * work as expected, see bonding.txt for details
4758                  */
4759                 printk(KERN_WARNING DRV_NAME
4760                        ": Warning: either miimon or arp_interval and "
4761                        "arp_ip_target module parameters must be specified, "
4762                        "otherwise bonding will not detect link failures! see "
4763                        "bonding.txt for details.\n");
4764         }
4765
4766         if (primary && !USES_PRIMARY(bond_mode)) {
4767                 /* currently, using a primary only makes sense
4768                  * in active backup, TLB or ALB modes
4769                  */
4770                 printk(KERN_WARNING DRV_NAME
4771                        ": Warning: %s primary device specified but has no "
4772                        "effect in %s mode\n",
4773                        primary, bond_mode_name(bond_mode));
4774                 primary = NULL;
4775         }
4776
4777         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4778                 printk(KERN_WARNING DRV_NAME
4779                        ": Warning: fail_over_mac only affects "
4780                        "active-backup mode.\n");
4781
4782         /* fill params struct with the proper values */
4783         params->mode = bond_mode;
4784         params->xmit_policy = xmit_hashtype;
4785         params->miimon = miimon;
4786         params->arp_interval = arp_interval;
4787         params->arp_validate = arp_validate_value;
4788         params->updelay = updelay;
4789         params->downdelay = downdelay;
4790         params->use_carrier = use_carrier;
4791         params->lacp_fast = lacp_fast;
4792         params->primary[0] = 0;
4793         params->fail_over_mac = fail_over_mac;
4794
4795         if (primary) {
4796                 strncpy(params->primary, primary, IFNAMSIZ);
4797                 params->primary[IFNAMSIZ - 1] = 0;
4798         }
4799
4800         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4801
4802         return 0;
4803 }
4804
4805 static struct lock_class_key bonding_netdev_xmit_lock_key;
4806
4807 /* Create a new bond based on the specified name and bonding parameters.
4808  * If name is NULL, obtain a suitable "bond%d" name for us.
4809  * Caller must NOT hold rtnl_lock; we need to release it here before we
4810  * set up our sysfs entries.
4811  */
4812 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4813 {
4814         struct net_device *bond_dev;
4815         int res;
4816
4817         rtnl_lock();
4818         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4819                                 ether_setup);
4820         if (!bond_dev) {
4821                 printk(KERN_ERR DRV_NAME
4822                        ": %s: eek! can't alloc netdev!\n",
4823                        name);
4824                 res = -ENOMEM;
4825                 goto out_rtnl;
4826         }
4827
4828         if (!name) {
4829                 res = dev_alloc_name(bond_dev, "bond%d");
4830                 if (res < 0)
4831                         goto out_netdev;
4832         }
4833
4834         /* bond_init() must be called after dev_alloc_name() (for the
4835          * /proc files), but before register_netdevice(), because we
4836          * need to set function pointers.
4837          */
4838
4839         res = bond_init(bond_dev, params);
4840         if (res < 0) {
4841                 goto out_netdev;
4842         }
4843
4844         res = register_netdevice(bond_dev);
4845         if (res < 0) {
4846                 goto out_bond;
4847         }
4848
4849         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4850
4851         if (newbond)
4852                 *newbond = bond_dev->priv;
4853
4854         netif_carrier_off(bond_dev);
4855
4856         rtnl_unlock(); /* allows sysfs registration of net device */
4857         res = bond_create_sysfs_entry(bond_dev->priv);
4858         if (res < 0) {
4859                 rtnl_lock();
4860                 goto out_bond;
4861         }
4862
4863         return 0;
4864
4865 out_bond:
4866         bond_deinit(bond_dev);
4867 out_netdev:
4868         free_netdev(bond_dev);
4869 out_rtnl:
4870         rtnl_unlock();
4871         return res;
4872 }
4873
4874 static void bond_work_cancel_all(struct bonding *bond)
4875 {
4876         write_lock_bh(&bond->lock);
4877         bond->kill_timers = 1;
4878         write_unlock_bh(&bond->lock);
4879
4880         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4881                 cancel_delayed_work(&bond->mii_work);
4882
4883         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4884                 cancel_delayed_work(&bond->arp_work);
4885
4886         if (bond->params.mode == BOND_MODE_ALB &&
4887             delayed_work_pending(&bond->alb_work))
4888                 cancel_delayed_work(&bond->alb_work);
4889
4890         if (bond->params.mode == BOND_MODE_8023AD &&
4891             delayed_work_pending(&bond->ad_work))
4892                 cancel_delayed_work(&bond->ad_work);
4893 }
4894
4895 static int __init bonding_init(void)
4896 {
4897         int i;
4898         int res;
4899         struct bonding *bond, *nxt;
4900
4901         printk(KERN_INFO "%s", version);
4902
4903         res = bond_check_params(&bonding_defaults);
4904         if (res) {
4905                 goto out;
4906         }
4907
4908 #ifdef CONFIG_PROC_FS
4909         bond_create_proc_dir();
4910 #endif
4911         for (i = 0; i < max_bonds; i++) {
4912                 res = bond_create(NULL, &bonding_defaults, NULL);
4913                 if (res)
4914                         goto err;
4915         }
4916
4917         res = bond_create_sysfs();
4918         if (res)
4919                 goto err;
4920
4921         register_netdevice_notifier(&bond_netdev_notifier);
4922         register_inetaddr_notifier(&bond_inetaddr_notifier);
4923
4924         goto out;
4925 err:
4926         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4927                 bond_work_cancel_all(bond);
4928                 destroy_workqueue(bond->wq);
4929         }
4930
4931         rtnl_lock();
4932         bond_free_all();
4933         bond_destroy_sysfs();
4934         rtnl_unlock();
4935 out:
4936         return res;
4937
4938 }
4939
4940 static void __exit bonding_exit(void)
4941 {
4942         unregister_netdevice_notifier(&bond_netdev_notifier);
4943         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4944
4945         rtnl_lock();
4946         bond_free_all();
4947         bond_destroy_sysfs();
4948         rtnl_unlock();
4949 }
4950
4951 module_init(bonding_init);
4952 module_exit(bonding_exit);
4953 MODULE_LICENSE("GPL");
4954 MODULE_VERSION(DRV_VERSION);
4955 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4956 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4957 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4958
4959 /*
4960  * Local variables:
4961  *  c-indent-level: 8
4962  *  c-basic-offset: 8
4963  *  tab-width: 8
4964  * End:
4965  */
4966