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 struct bond_params bonding_defaults;
 102
 103 module_param(max_bonds, int, 0);
 104 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 105 module_param(miimon, int, 0);
 106 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 107 module_param(updelay, int, 0);
 108 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 109 module_param(downdelay, int, 0);
 110 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 111                             "in milliseconds");
 112 module_param(use_carrier, int, 0);
 113 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 114                               "0 for off, 1 for on (default)");
 115 module_param(mode, charp, 0);
 116 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 117                        "1 for active-backup, 2 for balance-xor, "
 118                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 119                        "6 for balance-alb");
 120 module_param(primary, charp, 0);
 121 MODULE_PARM_DESC(primary, "Primary network device to use");
 122 module_param(lacp_rate, charp, 0);
 123 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 124                             "(slow/fast)");
 125 module_param(xmit_hash_policy, charp, 0);
 126 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 127                                    ", 1 for layer 3+4");
 128 module_param(arp_interval, int, 0);
 129 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 130 module_param_array(arp_ip_target, charp, NULL, 0);
 131 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 132 module_param(arp_validate, charp, 0);
 133 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 134
 135 /*----------------------------- Global variables ----------------------------*/
 136
 137 static const char * const version =
 138         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 139
 140 LIST_HEAD(bond_dev_list);
 141
 142 #ifdef CONFIG_PROC_FS
 143 static struct proc_dir_entry *bond_proc_dir = NULL;
 144 #endif
 145
 146 extern struct rw_semaphore bonding_rwsem;
 147 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 148 static int arp_ip_count = 0;
 149 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 150 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 151 static int lacp_fast    = 0;
 152
 153
 154 struct bond_parm_tbl bond_lacp_tbl[] = {
 155 {       "slow",         AD_LACP_SLOW},
 156 {       "fast",         AD_LACP_FAST},
 157 {       NULL,           -1},
 158 };
 159
 160 struct bond_parm_tbl bond_mode_tbl[] = {
 161 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 162 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 163 {       "balance-xor",          BOND_MODE_XOR},
 164 {       "broadcast",            BOND_MODE_BROADCAST},
 165 {       "802.3ad",              BOND_MODE_8023AD},
 166 {       "balance-tlb",          BOND_MODE_TLB},
 167 {       "balance-alb",          BOND_MODE_ALB},
 168 {       NULL,                   -1},
 169 };
 170
 171 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 172 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 173 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 174 {       NULL,                   -1},
 175 };
 176
 177 struct bond_parm_tbl arp_validate_tbl[] = {
 178 {       "none",                 BOND_ARP_VALIDATE_NONE},
 179 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 180 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 181 {       "all",                  BOND_ARP_VALIDATE_ALL},
 182 {       NULL,                   -1},
 183 };
 184
 185 /*-------------------------- Forward declarations ---------------------------*/
 186
 187 static void bond_send_gratuitous_arp(struct bonding *bond);
 188
 189 /*---------------------------- General routines -----------------------------*/
 190
 191 static const char *bond_mode_name(int mode)
 192 {
 193         switch (mode) {
 194         case BOND_MODE_ROUNDROBIN :
 195                 return "load balancing (round-robin)";
 196         case BOND_MODE_ACTIVEBACKUP :
 197                 return "fault-tolerance (active-backup)";
 198         case BOND_MODE_XOR :
 199                 return "load balancing (xor)";
 200         case BOND_MODE_BROADCAST :
 201                 return "fault-tolerance (broadcast)";
 202         case BOND_MODE_8023AD:
 203                 return "IEEE 802.3ad Dynamic link aggregation";
 204         case BOND_MODE_TLB:
 205                 return "transmit load balancing";
 206         case BOND_MODE_ALB:
 207                 return "adaptive load balancing";
 208         default:
 209                 return "unknown";
 210         }
 211 }
 212
 213 /*---------------------------------- VLAN -----------------------------------*/
 214
 215 /**
 216  * bond_add_vlan - add a new vlan id on bond
 217  * @bond: bond that got the notification
 218  * @vlan_id: the vlan id to add
 219  *
 220  * Returns -ENOMEM if allocation failed.
 221  */
 222 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 223 {
 224         struct vlan_entry *vlan;
 225
 226         dprintk("bond: %s, vlan id %d\n",
 227                 (bond ? bond->dev->name: "None"), vlan_id);
 228
 229         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 230         if (!vlan) {
 231                 return -ENOMEM;
 232         }
 233
 234         INIT_LIST_HEAD(&vlan->vlan_list);
 235         vlan->vlan_id = vlan_id;
 236         vlan->vlan_ip = 0;
 237
 238         write_lock_bh(&bond->lock);
 239
 240         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 241
 242         write_unlock_bh(&bond->lock);
 243
 244         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 245
 246         return 0;
 247 }
 248
 249 /**
 250  * bond_del_vlan - delete a vlan id from bond
 251  * @bond: bond that got the notification
 252  * @vlan_id: the vlan id to delete
 253  *
 254  * returns -ENODEV if @vlan_id was not found in @bond.
 255  */
 256 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 257 {
 258         struct vlan_entry *vlan, *next;
 259         int res = -ENODEV;
 260
 261         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 262
 263         write_lock_bh(&bond->lock);
 264
 265         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 266                 if (vlan->vlan_id == vlan_id) {
 267                         list_del(&vlan->vlan_list);
 268
 269                         if ((bond->params.mode == BOND_MODE_TLB) ||
 270                             (bond->params.mode == BOND_MODE_ALB)) {
 271                                 bond_alb_clear_vlan(bond, vlan_id);
 272                         }
 273
 274                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 275                                 bond->dev->name);
 276
 277                         kfree(vlan);
 278
 279                         if (list_empty(&bond->vlan_list) &&
 280                             (bond->slave_cnt == 0)) {
 281                                 /* Last VLAN removed and no slaves, so
 282                                  * restore block on adding VLANs. This will
 283                                  * be removed once new slaves that are not
 284                                  * VLAN challenged will be added.
 285                                  */
 286                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 287                         }
 288
 289                         res = 0;
 290                         goto out;
 291                 }
 292         }
 293
 294         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 295                 bond->dev->name);
 296
 297 out:
 298         write_unlock_bh(&bond->lock);
 299         return res;
 300 }
 301
 302 /**
 303  * bond_has_challenged_slaves
 304  * @bond: the bond we're working on
 305  *
 306  * Searches the slave list. Returns 1 if a vlan challenged slave
 307  * was found, 0 otherwise.
 308  *
 309  * Assumes bond->lock is held.
 310  */
 311 static int bond_has_challenged_slaves(struct bonding *bond)
 312 {
 313         struct slave *slave;
 314         int i;
 315
 316         bond_for_each_slave(bond, slave, i) {
 317                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 318                         dprintk("found VLAN challenged slave - %s\n",
 319                                 slave->dev->name);
 320                         return 1;
 321                 }
 322         }
 323
 324         dprintk("no VLAN challenged slaves found\n");
 325         return 0;
 326 }
 327
 328 /**
 329  * bond_next_vlan - safely skip to the next item in the vlans list.
 330  * @bond: the bond we're working on
 331  * @curr: item we're advancing from
 332  *
 333  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 334  * or @curr->next otherwise (even if it is @curr itself again).
 335  *
 336  * Caller must hold bond->lock
 337  */
 338 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 339 {
 340         struct vlan_entry *next, *last;
 341
 342         if (list_empty(&bond->vlan_list)) {
 343                 return NULL;
 344         }
 345
 346         if (!curr) {
 347                 next = list_entry(bond->vlan_list.next,
 348                                   struct vlan_entry, vlan_list);
 349         } else {
 350                 last = list_entry(bond->vlan_list.prev,
 351                                   struct vlan_entry, vlan_list);
 352                 if (last == curr) {
 353                         next = list_entry(bond->vlan_list.next,
 354                                           struct vlan_entry, vlan_list);
 355                 } else {
 356                         next = list_entry(curr->vlan_list.next,
 357                                           struct vlan_entry, vlan_list);
 358                 }
 359         }
 360
 361         return next;
 362 }
 363
 364 /**
 365  * bond_dev_queue_xmit - Prepare skb for xmit.
 366  *
 367  * @bond: bond device that got this skb for tx.
 368  * @skb: hw accel VLAN tagged skb to transmit
 369  * @slave_dev: slave that is supposed to xmit this skbuff
 370  *
 371  * When the bond gets an skb to transmit that is
 372  * already hardware accelerated VLAN tagged, and it
 373  * needs to relay this skb to a slave that is not
 374  * hw accel capable, the skb needs to be "unaccelerated",
 375  * i.e. strip the hwaccel tag and re-insert it as part
 376  * of the payload.
 377  */
 378 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 379 {
 380         unsigned short vlan_id;
 381
 382         if (!list_empty(&bond->vlan_list) &&
 383             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 384             vlan_get_tag(skb, &vlan_id) == 0) {
 385                 skb->dev = slave_dev;
 386                 skb = vlan_put_tag(skb, vlan_id);
 387                 if (!skb) {
 388                         /* vlan_put_tag() frees the skb in case of error,
 389                          * so return success here so the calling functions
 390                          * won't attempt to free is again.
 391                          */
 392                         return 0;
 393                 }
 394         } else {
 395                 skb->dev = slave_dev;
 396         }
 397
 398         skb->priority = 1;
 399         dev_queue_xmit(skb);
 400
 401         return 0;
 402 }
 403
 404 /*
 405  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 406  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 407  * lock because:
 408  * a. This operation is performed in IOCTL context,
 409  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 410  * c. Holding a lock with BH disabled while directly calling a base driver
 411  *    entry point is generally a BAD idea.
 412  *
 413  * The design of synchronization/protection for this operation in the 8021q
 414  * module is good for one or more VLAN devices over a single physical device
 415  * and cannot be extended for a teaming solution like bonding, so there is a
 416  * potential race condition here where a net device from the vlan group might
 417  * be referenced (either by a base driver or the 8021q code) while it is being
 418  * removed from the system. However, it turns out we're not making matters
 419  * worse, and if it works for regular VLAN usage it will work here too.
 420 */
 421
 422 /**
 423  * bond_vlan_rx_register - Propagates registration to slaves
 424  * @bond_dev: bonding net device that got called
 425  * @grp: vlan group being registered
 426  */
 427 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 428 {
 429         struct bonding *bond = bond_dev->priv;
 430         struct slave *slave;
 431         int i;
 432
 433         bond->vlgrp = grp;
 434
 435         bond_for_each_slave(bond, slave, i) {
 436                 struct net_device *slave_dev = slave->dev;
 437
 438                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 439                     slave_dev->vlan_rx_register) {
 440                         slave_dev->vlan_rx_register(slave_dev, grp);
 441                 }
 442         }
 443 }
 444
 445 /**
 446  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 447  * @bond_dev: bonding net device that got called
 448  * @vid: vlan id being added
 449  */
 450 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 451 {
 452         struct bonding *bond = bond_dev->priv;
 453         struct slave *slave;
 454         int i, res;
 455
 456         bond_for_each_slave(bond, slave, i) {
 457                 struct net_device *slave_dev = slave->dev;
 458
 459                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 460                     slave_dev->vlan_rx_add_vid) {
 461                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 462                 }
 463         }
 464
 465         res = bond_add_vlan(bond, vid);
 466         if (res) {
 467                 printk(KERN_ERR DRV_NAME
 468                        ": %s: Error: Failed to add vlan id %d\n",
 469                        bond_dev->name, vid);
 470         }
 471 }
 472
 473 /**
 474  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 475  * @bond_dev: bonding net device that got called
 476  * @vid: vlan id being removed
 477  */
 478 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 479 {
 480         struct bonding *bond = bond_dev->priv;
 481         struct slave *slave;
 482         struct net_device *vlan_dev;
 483         int i, res;
 484
 485         bond_for_each_slave(bond, slave, i) {
 486                 struct net_device *slave_dev = slave->dev;
 487
 488                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 489                     slave_dev->vlan_rx_kill_vid) {
 490                         /* Save and then restore vlan_dev in the grp array,
 491                          * since the slave's driver might clear it.
 492                          */
 493                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 494                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 495                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 496                 }
 497         }
 498
 499         res = bond_del_vlan(bond, vid);
 500         if (res) {
 501                 printk(KERN_ERR DRV_NAME
 502                        ": %s: Error: Failed to remove vlan id %d\n",
 503                        bond_dev->name, vid);
 504         }
 505 }
 506
 507 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 508 {
 509         struct vlan_entry *vlan;
 510
 511         write_lock_bh(&bond->lock);
 512
 513         if (list_empty(&bond->vlan_list)) {
 514                 goto out;
 515         }
 516
 517         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 518             slave_dev->vlan_rx_register) {
 519                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 520         }
 521
 522         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 523             !(slave_dev->vlan_rx_add_vid)) {
 524                 goto out;
 525         }
 526
 527         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 528                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 529         }
 530
 531 out:
 532         write_unlock_bh(&bond->lock);
 533 }
 534
 535 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 536 {
 537         struct vlan_entry *vlan;
 538         struct net_device *vlan_dev;
 539
 540         write_lock_bh(&bond->lock);
 541
 542         if (list_empty(&bond->vlan_list)) {
 543                 goto out;
 544         }
 545
 546         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 547             !(slave_dev->vlan_rx_kill_vid)) {
 548                 goto unreg;
 549         }
 550
 551         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 552                 /* Save and then restore vlan_dev in the grp array,
 553                  * since the slave's driver might clear it.
 554                  */
 555                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 556                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 557                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 558         }
 559
 560 unreg:
 561         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 562             slave_dev->vlan_rx_register) {
 563                 slave_dev->vlan_rx_register(slave_dev, NULL);
 564         }
 565
 566 out:
 567         write_unlock_bh(&bond->lock);
 568 }
 569
 570 /*------------------------------- Link status -------------------------------*/
 571
 572 /*
 573  * Set the carrier state for the master according to the state of its
 574  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 575  * do special 802.3ad magic.
 576  *
 577  * Returns zero if carrier state does not change, nonzero if it does.
 578  */
 579 static int bond_set_carrier(struct bonding *bond)
 580 {
 581         struct slave *slave;
 582         int i;
 583
 584         if (bond->slave_cnt == 0)
 585                 goto down;
 586
 587         if (bond->params.mode == BOND_MODE_8023AD)
 588                 return bond_3ad_set_carrier(bond);
 589
 590         bond_for_each_slave(bond, slave, i) {
 591                 if (slave->link == BOND_LINK_UP) {
 592                         if (!netif_carrier_ok(bond->dev)) {
 593                                 netif_carrier_on(bond->dev);
 594                                 return 1;
 595                         }
 596                         return 0;
 597                 }
 598         }
 599
 600 down:
 601         if (netif_carrier_ok(bond->dev)) {
 602                 netif_carrier_off(bond->dev);
 603                 return 1;
 604         }
 605         return 0;
 606 }
 607
 608 /*
 609  * Get link speed and duplex from the slave's base driver
 610  * using ethtool. If for some reason the call fails or the
 611  * values are invalid, fake speed and duplex to 100/Full
 612  * and return error.
 613  */
 614 static int bond_update_speed_duplex(struct slave *slave)
 615 {
 616         struct net_device *slave_dev = slave->dev;
 617         struct ethtool_cmd etool;
 618         int res;
 619
 620         /* Fake speed and duplex */
 621         slave->speed = SPEED_100;
 622         slave->duplex = DUPLEX_FULL;
 623
 624         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 625                 return -1;
 626
 627         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 628         if (res < 0)
 629                 return -1;
 630
 631         switch (etool.speed) {
 632         case SPEED_10:
 633         case SPEED_100:
 634         case SPEED_1000:
 635         case SPEED_10000:
 636                 break;
 637         default:
 638                 return -1;
 639         }
 640
 641         switch (etool.duplex) {
 642         case DUPLEX_FULL:
 643         case DUPLEX_HALF:
 644                 break;
 645         default:
 646                 return -1;
 647         }
 648
 649         slave->speed = etool.speed;
 650         slave->duplex = etool.duplex;
 651
 652         return 0;
 653 }
 654
 655 /*
 656  * if <dev> supports MII link status reporting, check its link status.
 657  *
 658  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 659  * depening upon the setting of the use_carrier parameter.
 660  *
 661  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 662  * can't tell and just pretend it is), or 0, meaning that the link is
 663  * down.
 664  *
 665  * If reporting is non-zero, instead of faking link up, return -1 if
 666  * both ETHTOOL and MII ioctls fail (meaning the device does not
 667  * support them).  If use_carrier is set, return whatever it says.
 668  * It'd be nice if there was a good way to tell if a driver supports
 669  * netif_carrier, but there really isn't.
 670  */
 671 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 672 {
 673         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 674         struct ifreq ifr;
 675         struct mii_ioctl_data *mii;
 676
 677         if (bond->params.use_carrier) {
 678                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 679         }
 680
 681         ioctl = slave_dev->do_ioctl;
 682         if (ioctl) {
 683                 /* TODO: set pointer to correct ioctl on a per team member */
 684                 /*       bases to make this more efficient. that is, once  */
 685                 /*       we determine the correct ioctl, we will always    */
 686                 /*       call it and not the others for that team          */
 687                 /*       member.                                           */
 688
 689                 /*
 690                  * We cannot assume that SIOCGMIIPHY will also read a
 691                  * register; not all network drivers (e.g., e100)
 692                  * support that.
 693                  */
 694
 695                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 696                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 697                 mii = if_mii(&ifr);
 698                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 699                         mii->reg_num = MII_BMSR;
 700                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 701                                 return (mii->val_out & BMSR_LSTATUS);
 702                         }
 703                 }
 704         }
 705
 706         /*
 707          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 708          * attempt to get link status from it if the above MII ioctls fail.
 709          */
 710         if (slave_dev->ethtool_ops) {
 711                 if (slave_dev->ethtool_ops->get_link) {
 712                         u32 link;
 713
 714                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 715
 716                         return link ? BMSR_LSTATUS : 0;
 717                 }
 718         }
 719
 720         /*
 721          * If reporting, report that either there's no dev->do_ioctl,
 722          * or both SIOCGMIIREG and get_link failed (meaning that we
 723          * cannot report link status).  If not reporting, pretend
 724          * we're ok.
 725          */
 726         return (reporting ? -1 : BMSR_LSTATUS);
 727 }
 728
 729 /*----------------------------- Multicast list ------------------------------*/
 730
 731 /*
 732  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 733  */
 734 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 735 {
 736         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 737                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 738 }
 739
 740 /*
 741  * returns dmi entry if found, NULL otherwise
 742  */
 743 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 744 {
 745         struct dev_mc_list *idmi;
 746
 747         for (idmi = mc_list; idmi; idmi = idmi->next) {
 748                 if (bond_is_dmi_same(dmi, idmi)) {
 749                         return idmi;
 750                 }
 751         }
 752
 753         return NULL;
 754 }
 755
 756 /*
 757  * Push the promiscuity flag down to appropriate slaves
 758  */
 759 static void bond_set_promiscuity(struct bonding *bond, int inc)
 760 {
 761         if (USES_PRIMARY(bond->params.mode)) {
 762                 /* write lock already acquired */
 763                 if (bond->curr_active_slave) {
 764                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 765                 }
 766         } else {
 767                 struct slave *slave;
 768                 int i;
 769                 bond_for_each_slave(bond, slave, i) {
 770                         dev_set_promiscuity(slave->dev, inc);
 771                 }
 772         }
 773 }
 774
 775 /*
 776  * Push the allmulti flag down to all slaves
 777  */
 778 static void bond_set_allmulti(struct bonding *bond, int inc)
 779 {
 780         if (USES_PRIMARY(bond->params.mode)) {
 781                 /* write lock already acquired */
 782                 if (bond->curr_active_slave) {
 783                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 784                 }
 785         } else {
 786                 struct slave *slave;
 787                 int i;
 788                 bond_for_each_slave(bond, slave, i) {
 789                         dev_set_allmulti(slave->dev, inc);
 790                 }
 791         }
 792 }
 793
 794 /*
 795  * Add a Multicast address to slaves
 796  * according to mode
 797  */
 798 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 799 {
 800         if (USES_PRIMARY(bond->params.mode)) {
 801                 /* write lock already acquired */
 802                 if (bond->curr_active_slave) {
 803                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 804                 }
 805         } else {
 806                 struct slave *slave;
 807                 int i;
 808                 bond_for_each_slave(bond, slave, i) {
 809                         dev_mc_add(slave->dev, addr, alen, 0);
 810                 }
 811         }
 812 }
 813
 814 /*
 815  * Remove a multicast address from slave
 816  * according to mode
 817  */
 818 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 819 {
 820         if (USES_PRIMARY(bond->params.mode)) {
 821                 /* write lock already acquired */
 822                 if (bond->curr_active_slave) {
 823                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 824                 }
 825         } else {
 826                 struct slave *slave;
 827                 int i;
 828                 bond_for_each_slave(bond, slave, i) {
 829                         dev_mc_delete(slave->dev, addr, alen, 0);
 830                 }
 831         }
 832 }
 833
 834
 835 /*
 836  * Retrieve the list of registered multicast addresses for the bonding
 837  * device and retransmit an IGMP JOIN request to the current active
 838  * slave.
 839  */
 840 static void bond_resend_igmp_join_requests(struct bonding *bond)
 841 {
 842         struct in_device *in_dev;
 843         struct ip_mc_list *im;
 844
 845         rcu_read_lock();
 846         in_dev = __in_dev_get_rcu(bond->dev);
 847         if (in_dev) {
 848                 for (im = in_dev->mc_list; im; im = im->next) {
 849                         ip_mc_rejoin_group(im);
 850                 }
 851         }
 852
 853         rcu_read_unlock();
 854 }
 855
 856 /*
 857  * Totally destroys the mc_list in bond
 858  */
 859 static void bond_mc_list_destroy(struct bonding *bond)
 860 {
 861         struct dev_mc_list *dmi;
 862
 863         dmi = bond->mc_list;
 864         while (dmi) {
 865                 bond->mc_list = dmi->next;
 866                 kfree(dmi);
 867                 dmi = bond->mc_list;
 868         }
 869         bond->mc_list = NULL;
 870 }
 871
 872 /*
 873  * Copy all the Multicast addresses from src to the bonding device dst
 874  */
 875 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 876                              gfp_t gfp_flag)
 877 {
 878         struct dev_mc_list *dmi, *new_dmi;
 879
 880         for (dmi = mc_list; dmi; dmi = dmi->next) {
 881                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 882
 883                 if (!new_dmi) {
 884                         /* FIXME: Potential memory leak !!! */
 885                         return -ENOMEM;
 886                 }
 887
 888                 new_dmi->next = bond->mc_list;
 889                 bond->mc_list = new_dmi;
 890                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 891                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 892                 new_dmi->dmi_users = dmi->dmi_users;
 893                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 894         }
 895
 896         return 0;
 897 }
 898
 899 /*
 900  * flush all members of flush->mc_list from device dev->mc_list
 901  */
 902 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 903 {
 904         struct bonding *bond = bond_dev->priv;
 905         struct dev_mc_list *dmi;
 906
 907         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 908                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 909         }
 910
 911         if (bond->params.mode == BOND_MODE_8023AD) {
 912                 /* del lacpdu mc addr from mc list */
 913                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 914
 915                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 916         }
 917 }
 918
 919 /*--------------------------- Active slave change ---------------------------*/
 920
 921 /*
 922  * Update the mc list and multicast-related flags for the new and
 923  * old active slaves (if any) according to the multicast mode, and
 924  * promiscuous flags unconditionally.
 925  */
 926 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 927 {
 928         struct dev_mc_list *dmi;
 929
 930         if (!USES_PRIMARY(bond->params.mode)) {
 931                 /* nothing to do -  mc list is already up-to-date on
 932                  * all slaves
 933                  */
 934                 return;
 935         }
 936
 937         if (old_active) {
 938                 if (bond->dev->flags & IFF_PROMISC) {
 939                         dev_set_promiscuity(old_active->dev, -1);
 940                 }
 941
 942                 if (bond->dev->flags & IFF_ALLMULTI) {
 943                         dev_set_allmulti(old_active->dev, -1);
 944                 }
 945
 946                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 947                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 948                 }
 949         }
 950
 951         if (new_active) {
 952                 if (bond->dev->flags & IFF_PROMISC) {
 953                         dev_set_promiscuity(new_active->dev, 1);
 954                 }
 955
 956                 if (bond->dev->flags & IFF_ALLMULTI) {
 957                         dev_set_allmulti(new_active->dev, 1);
 958                 }
 959
 960                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 961                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 962                 }
 963                 bond_resend_igmp_join_requests(bond);
 964         }
 965 }
 966
 967 /**
 968  * find_best_interface - select the best available slave to be the active one
 969  * @bond: our bonding struct
 970  *
 971  * Warning: Caller must hold curr_slave_lock for writing.
 972  */
 973 static struct slave *bond_find_best_slave(struct bonding *bond)
 974 {
 975         struct slave *new_active, *old_active;
 976         struct slave *bestslave = NULL;
 977         int mintime = bond->params.updelay;
 978         int i;
 979
 980         new_active = old_active = bond->curr_active_slave;
 981
 982         if (!new_active) { /* there were no active slaves left */
 983                 if (bond->slave_cnt > 0) {  /* found one slave */
 984                         new_active = bond->first_slave;
 985                 } else {
 986                         return NULL; /* still no slave, return NULL */
 987                 }
 988         }
 989
 990         /* first try the primary link; if arping, a link must tx/rx traffic
 991          * before it can be considered the curr_active_slave - also, we would skip
 992          * slaves between the curr_active_slave and primary_slave that may be up
 993          * and able to arp
 994          */
 995         if ((bond->primary_slave) &&
 996             (!bond->params.arp_interval) &&
 997             (IS_UP(bond->primary_slave->dev))) {
 998                 new_active = bond->primary_slave;
 999         }
1000
1001         /* remember where to stop iterating over the slaves */
1002         old_active = new_active;
1003
1004         bond_for_each_slave_from(bond, new_active, i, old_active) {
1005                 if (IS_UP(new_active->dev)) {
1006                         if (new_active->link == BOND_LINK_UP) {
1007                                 return new_active;
1008                         } else if (new_active->link == BOND_LINK_BACK) {
1009                                 /* link up, but waiting for stabilization */
1010                                 if (new_active->delay < mintime) {
1011                                         mintime = new_active->delay;
1012                                         bestslave = new_active;
1013                                 }
1014                         }
1015                 }
1016         }
1017
1018         return bestslave;
1019 }
1020
1021 /**
1022  * change_active_interface - change the active slave into the specified one
1023  * @bond: our bonding struct
1024  * @new: the new slave to make the active one
1025  *
1026  * Set the new slave to the bond's settings and unset them on the old
1027  * curr_active_slave.
1028  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1029  *
1030  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1031  * because it is apparently the best available slave we have, even though its
1032  * updelay hasn't timed out yet.
1033  *
1034  * Warning: Caller must hold curr_slave_lock for writing.
1035  */
1036 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1037 {
1038         struct slave *old_active = bond->curr_active_slave;
1039
1040         if (old_active == new_active) {
1041                 return;
1042         }
1043
1044         if (new_active) {
1045                 if (new_active->link == BOND_LINK_BACK) {
1046                         if (USES_PRIMARY(bond->params.mode)) {
1047                                 printk(KERN_INFO DRV_NAME
1048                                        ": %s: making interface %s the new "
1049                                        "active one %d ms earlier.\n",
1050                                        bond->dev->name, new_active->dev->name,
1051                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1052                         }
1053
1054                         new_active->delay = 0;
1055                         new_active->link = BOND_LINK_UP;
1056                         new_active->jiffies = jiffies;
1057
1058                         if (bond->params.mode == BOND_MODE_8023AD) {
1059                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1060                         }
1061
1062                         if ((bond->params.mode == BOND_MODE_TLB) ||
1063                             (bond->params.mode == BOND_MODE_ALB)) {
1064                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1065                         }
1066                 } else {
1067                         if (USES_PRIMARY(bond->params.mode)) {
1068                                 printk(KERN_INFO DRV_NAME
1069                                        ": %s: making interface %s the new "
1070                                        "active one.\n",
1071                                        bond->dev->name, new_active->dev->name);
1072                         }
1073                 }
1074         }
1075
1076         if (USES_PRIMARY(bond->params.mode)) {
1077                 bond_mc_swap(bond, new_active, old_active);
1078         }
1079
1080         if ((bond->params.mode == BOND_MODE_TLB) ||
1081             (bond->params.mode == BOND_MODE_ALB)) {
1082                 bond_alb_handle_active_change(bond, new_active);
1083                 if (old_active)
1084                         bond_set_slave_inactive_flags(old_active);
1085                 if (new_active)
1086                         bond_set_slave_active_flags(new_active);
1087         } else {
1088                 bond->curr_active_slave = new_active;
1089         }
1090
1091         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1092                 if (old_active) {
1093                         bond_set_slave_inactive_flags(old_active);
1094                 }
1095
1096                 if (new_active) {
1097                         bond_set_slave_active_flags(new_active);
1098                 }
1099
1100                 /* when bonding does not set the slave MAC address, the bond MAC
1101                  * address is the one of the active slave.
1102                  */
1103                 if (new_active && !bond->do_set_mac_addr)
1104                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1105                                 new_active->dev->addr_len);
1106
1107                 bond_send_gratuitous_arp(bond);
1108         }
1109 }
1110
1111 /**
1112  * bond_select_active_slave - select a new active slave, if needed
1113  * @bond: our bonding struct
1114  *
1115  * This functions shoud be called when one of the following occurs:
1116  * - The old curr_active_slave has been released or lost its link.
1117  * - The primary_slave has got its link back.
1118  * - A slave has got its link back and there's no old curr_active_slave.
1119  *
1120  * Warning: Caller must hold curr_slave_lock for writing.
1121  */
1122 void bond_select_active_slave(struct bonding *bond)
1123 {
1124         struct slave *best_slave;
1125         int rv;
1126
1127         best_slave = bond_find_best_slave(bond);
1128         if (best_slave != bond->curr_active_slave) {
1129                 bond_change_active_slave(bond, best_slave);
1130                 rv = bond_set_carrier(bond);
1131                 if (!rv)
1132                         return;
1133
1134                 if (netif_carrier_ok(bond->dev)) {
1135                         printk(KERN_INFO DRV_NAME
1136                                ": %s: first active interface up!\n",
1137                                bond->dev->name);
1138                 } else {
1139                         printk(KERN_INFO DRV_NAME ": %s: "
1140                                "now running without any active interface !\n",
1141                                bond->dev->name);
1142                 }
1143         }
1144 }
1145
1146 /*--------------------------- slave list handling ---------------------------*/
1147
1148 /*
1149  * This function attaches the slave to the end of list.
1150  *
1151  * bond->lock held for writing by caller.
1152  */
1153 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1154 {
1155         if (bond->first_slave == NULL) { /* attaching the first slave */
1156                 new_slave->next = new_slave;
1157                 new_slave->prev = new_slave;
1158                 bond->first_slave = new_slave;
1159         } else {
1160                 new_slave->next = bond->first_slave;
1161                 new_slave->prev = bond->first_slave->prev;
1162                 new_slave->next->prev = new_slave;
1163                 new_slave->prev->next = new_slave;
1164         }
1165
1166         bond->slave_cnt++;
1167 }
1168
1169 /*
1170  * This function detaches the slave from the list.
1171  * WARNING: no check is made to verify if the slave effectively
1172  * belongs to <bond>.
1173  * Nothing is freed on return, structures are just unchained.
1174  * If any slave pointer in bond was pointing to <slave>,
1175  * it should be changed by the calling function.
1176  *
1177  * bond->lock held for writing by caller.
1178  */
1179 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1180 {
1181         if (slave->next) {
1182                 slave->next->prev = slave->prev;
1183         }
1184
1185         if (slave->prev) {
1186                 slave->prev->next = slave->next;
1187         }
1188
1189         if (bond->first_slave == slave) { /* slave is the first slave */
1190                 if (bond->slave_cnt > 1) { /* there are more slave */
1191                         bond->first_slave = slave->next;
1192                 } else {
1193                         bond->first_slave = NULL; /* slave was the last one */
1194                 }
1195         }
1196
1197         slave->next = NULL;
1198         slave->prev = NULL;
1199         bond->slave_cnt--;
1200 }
1201
1202 /*---------------------------------- IOCTL ----------------------------------*/
1203
1204 static int bond_sethwaddr(struct net_device *bond_dev,
1205                           struct net_device *slave_dev)
1206 {
1207         dprintk("bond_dev=%p\n", bond_dev);
1208         dprintk("slave_dev=%p\n", slave_dev);
1209         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1210         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1211         return 0;
1212 }
1213
1214 #define BOND_VLAN_FEATURES \
1215         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1216          NETIF_F_HW_VLAN_FILTER)
1217
1218 /*
1219  * Compute the common dev->feature set available to all slaves.  Some
1220  * feature bits are managed elsewhere, so preserve those feature bits
1221  * on the master device.
1222  */
1223 static int bond_compute_features(struct bonding *bond)
1224 {
1225         struct slave *slave;
1226         struct net_device *bond_dev = bond->dev;
1227         unsigned long features = bond_dev->features;
1228         unsigned short max_hard_header_len = ETH_HLEN;
1229         int i;
1230
1231         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1232         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1233                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1234
1235         bond_for_each_slave(bond, slave, i) {
1236                 features = netdev_compute_features(features,
1237                                                    slave->dev->features);
1238                 if (slave->dev->hard_header_len > max_hard_header_len)
1239                         max_hard_header_len = slave->dev->hard_header_len;
1240         }
1241
1242         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1243         bond_dev->features = features;
1244         bond_dev->hard_header_len = max_hard_header_len;
1245
1246         return 0;
1247 }
1248
1249
1250 static void bond_setup_by_slave(struct net_device *bond_dev,
1251                                 struct net_device *slave_dev)
1252 {
1253         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1254
1255         bond_dev->type              = slave_dev->type;
1256         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1257         bond_dev->addr_len          = slave_dev->addr_len;
1258
1259         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1260                 slave_dev->addr_len);
1261 }
1262
1263 /* enslave device <slave> to bond device <master> */
1264 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1265 {
1266         struct bonding *bond = bond_dev->priv;
1267         struct slave *new_slave = NULL;
1268         struct dev_mc_list *dmi;
1269         struct sockaddr addr;
1270         int link_reporting;
1271         int old_features = bond_dev->features;
1272         int res = 0;
1273
1274         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1275                 slave_dev->do_ioctl == NULL) {
1276                 printk(KERN_WARNING DRV_NAME
1277                        ": %s: Warning: no link monitoring support for %s\n",
1278                        bond_dev->name, slave_dev->name);
1279         }
1280
1281         /* bond must be initialized by bond_open() before enslaving */
1282         if (!(bond_dev->flags & IFF_UP)) {
1283                 dprintk("Error, master_dev is not up\n");
1284                 return -EPERM;
1285         }
1286
1287         /* already enslaved */
1288         if (slave_dev->flags & IFF_SLAVE) {
1289                 dprintk("Error, Device was already enslaved\n");
1290                 return -EBUSY;
1291         }
1292
1293         /* vlan challenged mutual exclusion */
1294         /* no need to lock since we're protected by rtnl_lock */
1295         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1296                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1297                 if (!list_empty(&bond->vlan_list)) {
1298                         printk(KERN_ERR DRV_NAME
1299                                ": %s: Error: cannot enslave VLAN "
1300                                "challenged slave %s on VLAN enabled "
1301                                "bond %s\n", bond_dev->name, slave_dev->name,
1302                                bond_dev->name);
1303                         return -EPERM;
1304                 } else {
1305                         printk(KERN_WARNING DRV_NAME
1306                                ": %s: Warning: enslaved VLAN challenged "
1307                                "slave %s. Adding VLANs will be blocked as "
1308                                "long as %s is part of bond %s\n",
1309                                bond_dev->name, slave_dev->name, slave_dev->name,
1310                                bond_dev->name);
1311                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1312                 }
1313         } else {
1314                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1315                 if (bond->slave_cnt == 0) {
1316                         /* First slave, and it is not VLAN challenged,
1317                          * so remove the block of adding VLANs over the bond.
1318                          */
1319                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1320                 }
1321         }
1322
1323         /*
1324          * Old ifenslave binaries are no longer supported.  These can
1325          * be identified with moderate accurary by the state of the slave:
1326          * the current ifenslave will set the interface down prior to
1327          * enslaving it; the old ifenslave will not.
1328          */
1329         if ((slave_dev->flags & IFF_UP)) {
1330                 printk(KERN_ERR DRV_NAME ": %s is up. "
1331                        "This may be due to an out of date ifenslave.\n",
1332                        slave_dev->name);
1333                 res = -EPERM;
1334                 goto err_undo_flags;
1335         }
1336
1337         /* set bonding device ether type by slave - bonding netdevices are
1338          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1339          * there is a need to override some of the type dependent attribs/funcs.
1340          *
1341          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1342          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1343          */
1344         if (bond->slave_cnt == 0) {
1345                 if (slave_dev->type != ARPHRD_ETHER)
1346                         bond_setup_by_slave(bond_dev, slave_dev);
1347         } else if (bond_dev->type != slave_dev->type) {
1348                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1349                         "from other slaves (%d), can not enslave it.\n",
1350                         slave_dev->name,
1351                         slave_dev->type, bond_dev->type);
1352                         res = -EINVAL;
1353                         goto err_undo_flags;
1354         }
1355
1356         if (slave_dev->set_mac_address == NULL) {
1357                 if (bond->slave_cnt == 0) {
1358                         printk(KERN_WARNING DRV_NAME
1359                                 ": %s: Warning: The first slave device you "
1360                                 "specified does not support setting the MAC "
1361                                 "address. This bond MAC address would be that "
1362                                 "of the active slave.\n", bond_dev->name);
1363                         bond->do_set_mac_addr = 0;
1364                 } else if (bond->do_set_mac_addr) {
1365                         printk(KERN_ERR DRV_NAME
1366                                 ": %s: Error: The slave device you specified "
1367                                 "does not support setting the MAC addres,."
1368                                 "but this bond uses this practice. \n"
1369                                 , bond_dev->name);
1370                         res = -EOPNOTSUPP;
1371                         goto err_undo_flags;
1372                 }
1373         }
1374
1375         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1376         if (!new_slave) {
1377                 res = -ENOMEM;
1378                 goto err_undo_flags;
1379         }
1380
1381         /* save slave's original flags before calling
1382          * netdev_set_master and dev_open
1383          */
1384         new_slave->original_flags = slave_dev->flags;
1385
1386         /*
1387          * Save slave's original ("permanent") mac address for modes
1388          * that need it, and for restoring it upon release, and then
1389          * set it to the master's address
1390          */
1391         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1392
1393         if (bond->do_set_mac_addr) {
1394                 /*
1395                  * Set slave to master's mac address.  The application already
1396                  * set the master's mac address to that of the first slave
1397                  */
1398                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1399                 addr.sa_family = slave_dev->type;
1400                 res = dev_set_mac_address(slave_dev, &addr);
1401                 if (res) {
1402                         dprintk("Error %d calling set_mac_address\n", res);
1403                         goto err_free;
1404                 }
1405         }
1406
1407         res = netdev_set_master(slave_dev, bond_dev);
1408         if (res) {
1409                 dprintk("Error %d calling netdev_set_master\n", res);
1410                 goto err_close;
1411         }
1412         /* open the slave since the application closed it */
1413         res = dev_open(slave_dev);
1414         if (res) {
1415                 dprintk("Openning slave %s failed\n", slave_dev->name);
1416                 goto err_restore_mac;
1417         }
1418
1419         new_slave->dev = slave_dev;
1420         slave_dev->priv_flags |= IFF_BONDING;
1421
1422         if ((bond->params.mode == BOND_MODE_TLB) ||
1423             (bond->params.mode == BOND_MODE_ALB)) {
1424                 /* bond_alb_init_slave() must be called before all other stages since
1425                  * it might fail and we do not want to have to undo everything
1426                  */
1427                 res = bond_alb_init_slave(bond, new_slave);
1428                 if (res) {
1429                         goto err_unset_master;
1430                 }
1431         }
1432
1433         /* If the mode USES_PRIMARY, then the new slave gets the
1434          * master's promisc (and mc) settings only if it becomes the
1435          * curr_active_slave, and that is taken care of later when calling
1436          * bond_change_active()
1437          */
1438         if (!USES_PRIMARY(bond->params.mode)) {
1439                 /* set promiscuity level to new slave */
1440                 if (bond_dev->flags & IFF_PROMISC) {
1441                         dev_set_promiscuity(slave_dev, 1);
1442                 }
1443
1444                 /* set allmulti level to new slave */
1445                 if (bond_dev->flags & IFF_ALLMULTI) {
1446                         dev_set_allmulti(slave_dev, 1);
1447                 }
1448
1449                 /* upload master's mc_list to new slave */
1450                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1451                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1452                 }
1453         }
1454
1455         if (bond->params.mode == BOND_MODE_8023AD) {
1456                 /* add lacpdu mc addr to mc list */
1457                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1458
1459                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1460         }
1461
1462         bond_add_vlans_on_slave(bond, slave_dev);
1463
1464         write_lock_bh(&bond->lock);
1465
1466         bond_attach_slave(bond, new_slave);
1467
1468         new_slave->delay = 0;
1469         new_slave->link_failure_count = 0;
1470
1471         bond_compute_features(bond);
1472
1473         new_slave->last_arp_rx = jiffies;
1474
1475         if (bond->params.miimon && !bond->params.use_carrier) {
1476                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1477
1478                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1479                         /*
1480                          * miimon is set but a bonded network driver
1481                          * does not support ETHTOOL/MII and
1482                          * arp_interval is not set.  Note: if
1483                          * use_carrier is enabled, we will never go
1484                          * here (because netif_carrier is always
1485                          * supported); thus, we don't need to change
1486                          * the messages for netif_carrier.
1487                          */
1488                         printk(KERN_WARNING DRV_NAME
1489                                ": %s: Warning: MII and ETHTOOL support not "
1490                                "available for interface %s, and "
1491                                "arp_interval/arp_ip_target module parameters "
1492                                "not specified, thus bonding will not detect "
1493                                "link failures! see bonding.txt for details.\n",
1494                                bond_dev->name, slave_dev->name);
1495                 } else if (link_reporting == -1) {
1496                         /* unable get link status using mii/ethtool */
1497                         printk(KERN_WARNING DRV_NAME
1498                                ": %s: Warning: can't get link status from "
1499                                "interface %s; the network driver associated "
1500                                "with this interface does not support MII or "
1501                                "ETHTOOL link status reporting, thus miimon "
1502                                "has no effect on this interface.\n",
1503                                bond_dev->name, slave_dev->name);
1504                 }
1505         }
1506
1507         /* check for initial state */
1508         if (!bond->params.miimon ||
1509             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1510                 if (bond->params.updelay) {
1511                         dprintk("Initial state of slave_dev is "
1512                                 "BOND_LINK_BACK\n");
1513                         new_slave->link  = BOND_LINK_BACK;
1514                         new_slave->delay = bond->params.updelay;
1515                 } else {
1516                         dprintk("Initial state of slave_dev is "
1517                                 "BOND_LINK_UP\n");
1518                         new_slave->link  = BOND_LINK_UP;
1519                 }
1520                 new_slave->jiffies = jiffies;
1521         } else {
1522                 dprintk("Initial state of slave_dev is "
1523                         "BOND_LINK_DOWN\n");
1524                 new_slave->link  = BOND_LINK_DOWN;
1525         }
1526
1527         if (bond_update_speed_duplex(new_slave) &&
1528             (new_slave->link != BOND_LINK_DOWN)) {
1529                 printk(KERN_WARNING DRV_NAME
1530                        ": %s: Warning: failed to get speed and duplex from %s, "
1531                        "assumed to be 100Mb/sec and Full.\n",
1532                        bond_dev->name, new_slave->dev->name);
1533
1534                 if (bond->params.mode == BOND_MODE_8023AD) {
1535                         printk(KERN_WARNING DRV_NAME
1536                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1537                                "support in base driver for proper aggregator "
1538                                "selection.\n", bond_dev->name);
1539                 }
1540         }
1541
1542         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1543                 /* if there is a primary slave, remember it */
1544                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1545                         bond->primary_slave = new_slave;
1546                 }
1547         }
1548
1549         switch (bond->params.mode) {
1550         case BOND_MODE_ACTIVEBACKUP:
1551                 bond_set_slave_inactive_flags(new_slave);
1552                 bond_select_active_slave(bond);
1553                 break;
1554         case BOND_MODE_8023AD:
1555                 /* in 802.3ad mode, the internal mechanism
1556                  * will activate the slaves in the selected
1557                  * aggregator
1558                  */
1559                 bond_set_slave_inactive_flags(new_slave);
1560                 /* if this is the first slave */
1561                 if (bond->slave_cnt == 1) {
1562                         SLAVE_AD_INFO(new_slave).id = 1;
1563                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1564                          * can be called only after the mac address of the bond is set
1565                          */
1566                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1567                                             bond->params.lacp_fast);
1568                 } else {
1569                         SLAVE_AD_INFO(new_slave).id =
1570                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1571                 }
1572
1573                 bond_3ad_bind_slave(new_slave);
1574                 break;
1575         case BOND_MODE_TLB:
1576         case BOND_MODE_ALB:
1577                 new_slave->state = BOND_STATE_ACTIVE;
1578                 if ((!bond->curr_active_slave) &&
1579                     (new_slave->link != BOND_LINK_DOWN)) {
1580                         /* first slave or no active slave yet, and this link
1581                          * is OK, so make this interface the active one
1582                          */
1583                         bond_change_active_slave(bond, new_slave);
1584                 } else {
1585                         bond_set_slave_inactive_flags(new_slave);
1586                 }
1587                 break;
1588         default:
1589                 dprintk("This slave is always active in trunk mode\n");
1590
1591                 /* always active in trunk mode */
1592                 new_slave->state = BOND_STATE_ACTIVE;
1593
1594                 /* In trunking mode there is little meaning to curr_active_slave
1595                  * anyway (it holds no special properties of the bond device),
1596                  * so we can change it without calling change_active_interface()
1597                  */
1598                 if (!bond->curr_active_slave) {
1599                         bond->curr_active_slave = new_slave;
1600                 }
1601                 break;
1602         } /* switch(bond_mode) */
1603
1604         bond_set_carrier(bond);
1605
1606         write_unlock_bh(&bond->lock);
1607
1608         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1609         if (res)
1610                 goto err_unset_master;
1611
1612         printk(KERN_INFO DRV_NAME
1613                ": %s: enslaving %s as a%s interface with a%s link.\n",
1614                bond_dev->name, slave_dev->name,
1615                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1616                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1617
1618         /* enslave is successful */
1619         return 0;
1620
1621 /* Undo stages on error */
1622 err_unset_master:
1623         netdev_set_master(slave_dev, NULL);
1624
1625 err_close:
1626         dev_close(slave_dev);
1627
1628 err_restore_mac:
1629         if (bond->do_set_mac_addr) {
1630                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1631                 addr.sa_family = slave_dev->type;
1632                 dev_set_mac_address(slave_dev, &addr);
1633         }
1634
1635 err_free:
1636         kfree(new_slave);
1637
1638 err_undo_flags:
1639         bond_dev->features = old_features;
1640
1641         return res;
1642 }
1643
1644 /*
1645  * Try to release the slave device <slave> from the bond device <master>
1646  * It is legal to access curr_active_slave without a lock because all the function
1647  * is write-locked.
1648  *
1649  * The rules for slave state should be:
1650  *   for Active/Backup:
1651  *     Active stays on all backups go down
1652  *   for Bonded connections:
1653  *     The first up interface should be left on and all others downed.
1654  */
1655 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1656 {
1657         struct bonding *bond = bond_dev->priv;
1658         struct slave *slave, *oldcurrent;
1659         struct sockaddr addr;
1660         int mac_addr_differ;
1661         DECLARE_MAC_BUF(mac);
1662
1663         /* slave is not a slave or master is not master of this slave */
1664         if (!(slave_dev->flags & IFF_SLAVE) ||
1665             (slave_dev->master != bond_dev)) {
1666                 printk(KERN_ERR DRV_NAME
1667                        ": %s: Error: cannot release %s.\n",
1668                        bond_dev->name, slave_dev->name);
1669                 return -EINVAL;
1670         }
1671
1672         write_lock_bh(&bond->lock);
1673
1674         slave = bond_get_slave_by_dev(bond, slave_dev);
1675         if (!slave) {
1676                 /* not a slave of this bond */
1677                 printk(KERN_INFO DRV_NAME
1678                        ": %s: %s not enslaved\n",
1679                        bond_dev->name, slave_dev->name);
1680                 write_unlock_bh(&bond->lock);
1681                 return -EINVAL;
1682         }
1683
1684         mac_addr_differ = memcmp(bond_dev->dev_addr,
1685                                  slave->perm_hwaddr,
1686                                  ETH_ALEN);
1687         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1688                 printk(KERN_WARNING DRV_NAME
1689                        ": %s: Warning: the permanent HWaddr of %s - "
1690                        "%s - is still in use by %s. "
1691                        "Set the HWaddr of %s to a different address "
1692                        "to avoid conflicts.\n",
1693                        bond_dev->name,
1694                        slave_dev->name,
1695                        print_mac(mac, slave->perm_hwaddr),
1696                        bond_dev->name,
1697                        slave_dev->name);
1698         }
1699
1700         /* Inform AD package of unbinding of slave. */
1701         if (bond->params.mode == BOND_MODE_8023AD) {
1702                 /* must be called before the slave is
1703                  * detached from the list
1704                  */
1705                 bond_3ad_unbind_slave(slave);
1706         }
1707
1708         printk(KERN_INFO DRV_NAME
1709                ": %s: releasing %s interface %s\n",
1710                bond_dev->name,
1711                (slave->state == BOND_STATE_ACTIVE)
1712                ? "active" : "backup",
1713                slave_dev->name);
1714
1715         oldcurrent = bond->curr_active_slave;
1716
1717         bond->current_arp_slave = NULL;
1718
1719         /* release the slave from its bond */
1720         bond_detach_slave(bond, slave);
1721
1722         bond_compute_features(bond);
1723
1724         if (bond->primary_slave == slave) {
1725                 bond->primary_slave = NULL;
1726         }
1727
1728         if (oldcurrent == slave) {
1729                 bond_change_active_slave(bond, NULL);
1730         }
1731
1732         if ((bond->params.mode == BOND_MODE_TLB) ||
1733             (bond->params.mode == BOND_MODE_ALB)) {
1734                 /* Must be called only after the slave has been
1735                  * detached from the list and the curr_active_slave
1736                  * has been cleared (if our_slave == old_current),
1737                  * but before a new active slave is selected.
1738                  */
1739                 bond_alb_deinit_slave(bond, slave);
1740         }
1741
1742         if (oldcurrent == slave)
1743                 bond_select_active_slave(bond);
1744
1745         if (bond->slave_cnt == 0) {
1746                 bond_set_carrier(bond);
1747
1748                 /* if the last slave was removed, zero the mac address
1749                  * of the master so it will be set by the application
1750                  * to the mac address of the first slave
1751                  */
1752                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1753
1754                 if (list_empty(&bond->vlan_list)) {
1755                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1756                 } else {
1757                         printk(KERN_WARNING DRV_NAME
1758                                ": %s: Warning: clearing HW address of %s while it "
1759                                "still has VLANs.\n",
1760                                bond_dev->name, bond_dev->name);
1761                         printk(KERN_WARNING DRV_NAME
1762                                ": %s: When re-adding slaves, make sure the bond's "
1763                                "HW address matches its VLANs'.\n",
1764                                bond_dev->name);
1765                 }
1766         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1767                    !bond_has_challenged_slaves(bond)) {
1768                 printk(KERN_INFO DRV_NAME
1769                        ": %s: last VLAN challenged slave %s "
1770                        "left bond %s. VLAN blocking is removed\n",
1771                        bond_dev->name, slave_dev->name, bond_dev->name);
1772                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1773         }
1774
1775         write_unlock_bh(&bond->lock);
1776
1777         /* must do this from outside any spinlocks */
1778         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1779
1780         bond_del_vlans_from_slave(bond, slave_dev);
1781
1782         /* If the mode USES_PRIMARY, then we should only remove its
1783          * promisc and mc settings if it was the curr_active_slave, but that was
1784          * already taken care of above when we detached the slave
1785          */
1786         if (!USES_PRIMARY(bond->params.mode)) {
1787                 /* unset promiscuity level from slave */
1788                 if (bond_dev->flags & IFF_PROMISC) {
1789                         dev_set_promiscuity(slave_dev, -1);
1790                 }
1791
1792                 /* unset allmulti level from slave */
1793                 if (bond_dev->flags & IFF_ALLMULTI) {
1794                         dev_set_allmulti(slave_dev, -1);
1795                 }
1796
1797                 /* flush master's mc_list from slave */
1798                 bond_mc_list_flush(bond_dev, slave_dev);
1799         }
1800
1801         netdev_set_master(slave_dev, NULL);
1802
1803         /* close slave before restoring its mac address */
1804         dev_close(slave_dev);
1805
1806         if (bond->do_set_mac_addr) {
1807                 /* restore original ("permanent") mac address */
1808                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1809                 addr.sa_family = slave_dev->type;
1810                 dev_set_mac_address(slave_dev, &addr);
1811         }
1812
1813         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1814                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1815                                    IFF_SLAVE_NEEDARP);
1816
1817         kfree(slave);
1818
1819         return 0;  /* deletion OK */
1820 }
1821
1822 /*
1823  * This function releases all slaves.
1824  */
1825 static int bond_release_all(struct net_device *bond_dev)
1826 {
1827         struct bonding *bond = bond_dev->priv;
1828         struct slave *slave;
1829         struct net_device *slave_dev;
1830         struct sockaddr addr;
1831
1832         write_lock_bh(&bond->lock);
1833
1834         netif_carrier_off(bond_dev);
1835
1836         if (bond->slave_cnt == 0) {
1837                 goto out;
1838         }
1839
1840         bond->current_arp_slave = NULL;
1841         bond->primary_slave = NULL;
1842         bond_change_active_slave(bond, NULL);
1843
1844         while ((slave = bond->first_slave) != NULL) {
1845                 /* Inform AD package of unbinding of slave
1846                  * before slave is detached from the list.
1847                  */
1848                 if (bond->params.mode == BOND_MODE_8023AD) {
1849                         bond_3ad_unbind_slave(slave);
1850                 }
1851
1852                 slave_dev = slave->dev;
1853                 bond_detach_slave(bond, slave);
1854
1855                 if ((bond->params.mode == BOND_MODE_TLB) ||
1856                     (bond->params.mode == BOND_MODE_ALB)) {
1857                         /* must be called only after the slave
1858                          * has been detached from the list
1859                          */
1860                         bond_alb_deinit_slave(bond, slave);
1861                 }
1862
1863                 bond_compute_features(bond);
1864
1865                 /* now that the slave is detached, unlock and perform
1866                  * all the undo steps that should not be called from
1867                  * within a lock.
1868                  */
1869                 write_unlock_bh(&bond->lock);
1870
1871                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1872                 bond_del_vlans_from_slave(bond, slave_dev);
1873
1874                 /* If the mode USES_PRIMARY, then we should only remove its
1875                  * promisc and mc settings if it was the curr_active_slave, but that was
1876                  * already taken care of above when we detached the slave
1877                  */
1878                 if (!USES_PRIMARY(bond->params.mode)) {
1879                         /* unset promiscuity level from slave */
1880                         if (bond_dev->flags & IFF_PROMISC) {
1881                                 dev_set_promiscuity(slave_dev, -1);
1882                         }
1883
1884                         /* unset allmulti level from slave */
1885                         if (bond_dev->flags & IFF_ALLMULTI) {
1886                                 dev_set_allmulti(slave_dev, -1);
1887                         }
1888
1889                         /* flush master's mc_list from slave */
1890                         bond_mc_list_flush(bond_dev, slave_dev);
1891                 }
1892
1893                 netdev_set_master(slave_dev, NULL);
1894
1895                 /* close slave before restoring its mac address */
1896                 dev_close(slave_dev);
1897
1898                 if (bond->do_set_mac_addr) {
1899                         /* restore original ("permanent") mac address*/
1900                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1901                         addr.sa_family = slave_dev->type;
1902                         dev_set_mac_address(slave_dev, &addr);
1903                 }
1904
1905                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1906                                            IFF_SLAVE_INACTIVE);
1907
1908                 kfree(slave);
1909
1910                 /* re-acquire the lock before getting the next slave */
1911                 write_lock_bh(&bond->lock);
1912         }
1913
1914         /* zero the mac address of the master so it will be
1915          * set by the application to the mac address of the
1916          * first slave
1917          */
1918         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1919
1920         if (list_empty(&bond->vlan_list)) {
1921                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1922         } else {
1923                 printk(KERN_WARNING DRV_NAME
1924                        ": %s: Warning: clearing HW address of %s while it "
1925                        "still has VLANs.\n",
1926                        bond_dev->name, bond_dev->name);
1927                 printk(KERN_WARNING DRV_NAME
1928                        ": %s: When re-adding slaves, make sure the bond's "
1929                        "HW address matches its VLANs'.\n",
1930                        bond_dev->name);
1931         }
1932
1933         printk(KERN_INFO DRV_NAME
1934                ": %s: released all slaves\n",
1935                bond_dev->name);
1936
1937 out:
1938         write_unlock_bh(&bond->lock);
1939
1940         return 0;
1941 }
1942
1943 /*
1944  * This function changes the active slave to slave <slave_dev>.
1945  * It returns -EINVAL in the following cases.
1946  *  - <slave_dev> is not found in the list.
1947  *  - There is not active slave now.
1948  *  - <slave_dev> is already active.
1949  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1950  *  - <slave_dev> is not running.
1951  * In these cases, this fuction does nothing.
1952  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1953  */
1954 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1955 {
1956         struct bonding *bond = bond_dev->priv;
1957         struct slave *old_active = NULL;
1958         struct slave *new_active = NULL;
1959         int res = 0;
1960
1961         if (!USES_PRIMARY(bond->params.mode)) {
1962                 return -EINVAL;
1963         }
1964
1965         /* Verify that master_dev is indeed the master of slave_dev */
1966         if (!(slave_dev->flags & IFF_SLAVE) ||
1967             (slave_dev->master != bond_dev)) {
1968                 return -EINVAL;
1969         }
1970
1971         write_lock_bh(&bond->lock);
1972
1973         old_active = bond->curr_active_slave;
1974         new_active = bond_get_slave_by_dev(bond, slave_dev);
1975
1976         /*
1977          * Changing to the current active: do nothing; return success.
1978          */
1979         if (new_active && (new_active == old_active)) {
1980                 write_unlock_bh(&bond->lock);
1981                 return 0;
1982         }
1983
1984         if ((new_active) &&
1985             (old_active) &&
1986             (new_active->link == BOND_LINK_UP) &&
1987             IS_UP(new_active->dev)) {
1988                 bond_change_active_slave(bond, new_active);
1989         } else {
1990                 res = -EINVAL;
1991         }
1992
1993         write_unlock_bh(&bond->lock);
1994
1995         return res;
1996 }
1997
1998 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1999 {
2000         struct bonding *bond = bond_dev->priv;
2001
2002         info->bond_mode = bond->params.mode;
2003         info->miimon = bond->params.miimon;
2004
2005         read_lock_bh(&bond->lock);
2006         info->num_slaves = bond->slave_cnt;
2007         read_unlock_bh(&bond->lock);
2008
2009         return 0;
2010 }
2011
2012 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2013 {
2014         struct bonding *bond = bond_dev->priv;
2015         struct slave *slave;
2016         int i, found = 0;
2017
2018         if (info->slave_id < 0) {
2019                 return -ENODEV;
2020         }
2021
2022         read_lock_bh(&bond->lock);
2023
2024         bond_for_each_slave(bond, slave, i) {
2025                 if (i == (int)info->slave_id) {
2026                         found = 1;
2027                         break;
2028                 }
2029         }
2030
2031         read_unlock_bh(&bond->lock);
2032
2033         if (found) {
2034                 strcpy(info->slave_name, slave->dev->name);
2035                 info->link = slave->link;
2036                 info->state = slave->state;
2037                 info->link_failure_count = slave->link_failure_count;
2038         } else {
2039                 return -ENODEV;
2040         }
2041
2042         return 0;
2043 }
2044
2045 /*-------------------------------- Monitoring -------------------------------*/
2046
2047 /* this function is called regularly to monitor each slave's link. */
2048 void bond_mii_monitor(struct net_device *bond_dev)
2049 {
2050         struct bonding *bond = bond_dev->priv;
2051         struct slave *slave, *oldcurrent;
2052         int do_failover = 0;
2053         int delta_in_ticks;
2054         int i;
2055
2056         read_lock(&bond->lock);
2057
2058         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2059
2060         if (bond->kill_timers) {
2061                 goto out;
2062         }
2063
2064         if (bond->slave_cnt == 0) {
2065                 goto re_arm;
2066         }
2067
2068         /* we will try to read the link status of each of our slaves, and
2069          * set their IFF_RUNNING flag appropriately. For each slave not
2070          * supporting MII status, we won't do anything so that a user-space
2071          * program could monitor the link itself if needed.
2072          */
2073
2074         read_lock(&bond->curr_slave_lock);
2075         oldcurrent = bond->curr_active_slave;
2076         read_unlock(&bond->curr_slave_lock);
2077
2078         bond_for_each_slave(bond, slave, i) {
2079                 struct net_device *slave_dev = slave->dev;
2080                 int link_state;
2081                 u16 old_speed = slave->speed;
2082                 u8 old_duplex = slave->duplex;
2083
2084                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2085
2086                 switch (slave->link) {
2087                 case BOND_LINK_UP:      /* the link was up */
2088                         if (link_state == BMSR_LSTATUS) {
2089                                 /* link stays up, nothing more to do */
2090                                 break;
2091                         } else { /* link going down */
2092                                 slave->link  = BOND_LINK_FAIL;
2093                                 slave->delay = bond->params.downdelay;
2094
2095                                 if (slave->link_failure_count < UINT_MAX) {
2096                                         slave->link_failure_count++;
2097                                 }
2098
2099                                 if (bond->params.downdelay) {
2100                                         printk(KERN_INFO DRV_NAME
2101                                                ": %s: link status down for %s "
2102                                                "interface %s, disabling it in "
2103                                                "%d ms.\n",
2104                                                bond_dev->name,
2105                                                IS_UP(slave_dev)
2106                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2107                                                   ? ((slave == oldcurrent)
2108                                                      ? "active " : "backup ")
2109                                                   : "")
2110                                                : "idle ",
2111                                                slave_dev->name,
2112                                                bond->params.downdelay * bond->params.miimon);
2113                                 }
2114                         }
2115                         /* no break ! fall through the BOND_LINK_FAIL test to
2116                            ensure proper action to be taken
2117                         */
2118                 case BOND_LINK_FAIL:    /* the link has just gone down */
2119                         if (link_state != BMSR_LSTATUS) {
2120                                 /* link stays down */
2121                                 if (slave->delay <= 0) {
2122                                         /* link down for too long time */
2123                                         slave->link = BOND_LINK_DOWN;
2124
2125                                         /* in active/backup mode, we must
2126                                          * completely disable this interface
2127                                          */
2128                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2129                                             (bond->params.mode == BOND_MODE_8023AD)) {
2130                                                 bond_set_slave_inactive_flags(slave);
2131                                         }
2132
2133                                         printk(KERN_INFO DRV_NAME
2134                                                ": %s: link status definitely "
2135                                                "down for interface %s, "
2136                                                "disabling it\n",
2137                                                bond_dev->name,
2138                                                slave_dev->name);
2139
2140                                         /* notify ad that the link status has changed */
2141                                         if (bond->params.mode == BOND_MODE_8023AD) {
2142                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2143                                         }
2144
2145                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2146                                             (bond->params.mode == BOND_MODE_ALB)) {
2147                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2148                                         }
2149
2150                                         if (slave == oldcurrent) {
2151                                                 do_failover = 1;
2152                                         }
2153                                 } else {
2154                                         slave->delay--;
2155                                 }
2156                         } else {
2157                                 /* link up again */
2158                                 slave->link  = BOND_LINK_UP;
2159                                 slave->jiffies = jiffies;
2160                                 printk(KERN_INFO DRV_NAME
2161                                        ": %s: link status up again after %d "
2162                                        "ms for interface %s.\n",
2163                                        bond_dev->name,
2164                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2165                                        slave_dev->name);
2166                         }
2167                         break;
2168                 case BOND_LINK_DOWN:    /* the link was down */
2169                         if (link_state != BMSR_LSTATUS) {
2170                                 /* the link stays down, nothing more to do */
2171                                 break;
2172                         } else {        /* link going up */
2173                                 slave->link  = BOND_LINK_BACK;
2174                                 slave->delay = bond->params.updelay;
2175
2176                                 if (bond->params.updelay) {
2177                                         /* if updelay == 0, no need to
2178                                            advertise about a 0 ms delay */
2179                                         printk(KERN_INFO DRV_NAME
2180                                                ": %s: link status up for "
2181                                                "interface %s, enabling it "
2182                                                "in %d ms.\n",
2183                                                bond_dev->name,
2184                                                slave_dev->name,
2185                                                bond->params.updelay * bond->params.miimon);
2186                                 }
2187                         }
2188                         /* no break ! fall through the BOND_LINK_BACK state in
2189                            case there's something to do.
2190                         */
2191                 case BOND_LINK_BACK:    /* the link has just come back */
2192                         if (link_state != BMSR_LSTATUS) {
2193                                 /* link down again */
2194                                 slave->link  = BOND_LINK_DOWN;
2195
2196                                 printk(KERN_INFO DRV_NAME
2197                                        ": %s: link status down again after %d "
2198                                        "ms for interface %s.\n",
2199                                        bond_dev->name,
2200                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2201                                        slave_dev->name);
2202                         } else {
2203                                 /* link stays up */
2204                                 if (slave->delay == 0) {
2205                                         /* now the link has been up for long time enough */
2206                                         slave->link = BOND_LINK_UP;
2207                                         slave->jiffies = jiffies;
2208
2209                                         if (bond->params.mode == BOND_MODE_8023AD) {
2210                                                 /* prevent it from being the active one */
2211                                                 slave->state = BOND_STATE_BACKUP;
2212                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2213                                                 /* make it immediately active */
2214                                                 slave->state = BOND_STATE_ACTIVE;
2215                                         } else if (slave != bond->primary_slave) {
2216                                                 /* prevent it from being the active one */
2217                                                 slave->state = BOND_STATE_BACKUP;
2218                                         }
2219
2220                                         printk(KERN_INFO DRV_NAME
2221                                                ": %s: link status definitely "
2222                                                "up for interface %s.\n",
2223                                                bond_dev->name,
2224                                                slave_dev->name);
2225
2226                                         /* notify ad that the link status has changed */
2227                                         if (bond->params.mode == BOND_MODE_8023AD) {
2228                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2229                                         }
2230
2231                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2232                                             (bond->params.mode == BOND_MODE_ALB)) {
2233                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2234                                         }
2235
2236                                         if ((!oldcurrent) ||
2237                                             (slave == bond->primary_slave)) {
2238                                                 do_failover = 1;
2239                                         }
2240                                 } else {
2241                                         slave->delay--;
2242                                 }
2243                         }
2244                         break;
2245                 default:
2246                         /* Should not happen */
2247                         printk(KERN_ERR DRV_NAME
2248                                ": %s: Error: %s Illegal value (link=%d)\n",
2249                                bond_dev->name,
2250                                slave->dev->name,
2251                                slave->link);
2252                         goto out;
2253                 } /* end of switch (slave->link) */
2254
2255                 bond_update_speed_duplex(slave);
2256
2257                 if (bond->params.mode == BOND_MODE_8023AD) {
2258                         if (old_speed != slave->speed) {
2259                                 bond_3ad_adapter_speed_changed(slave);
2260                         }
2261
2262                         if (old_duplex != slave->duplex) {
2263                                 bond_3ad_adapter_duplex_changed(slave);
2264                         }
2265                 }
2266
2267         } /* end of for */
2268
2269         if (do_failover) {
2270                 write_lock(&bond->curr_slave_lock);
2271
2272                 bond_select_active_slave(bond);
2273
2274                 write_unlock(&bond->curr_slave_lock);
2275         } else
2276                 bond_set_carrier(bond);
2277
2278 re_arm:
2279         if (bond->params.miimon) {
2280                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2281         }
2282 out:
2283         read_unlock(&bond->lock);
2284 }
2285
2286
2287 static __be32 bond_glean_dev_ip(struct net_device *dev)
2288 {
2289         struct in_device *idev;
2290         struct in_ifaddr *ifa;
2291         __be32 addr = 0;
2292
2293         if (!dev)
2294                 return 0;
2295
2296         rcu_read_lock();
2297         idev = __in_dev_get_rcu(dev);
2298         if (!idev)
2299                 goto out;
2300
2301         ifa = idev->ifa_list;
2302         if (!ifa)
2303                 goto out;
2304
2305         addr = ifa->ifa_local;
2306 out:
2307         rcu_read_unlock();
2308         return addr;
2309 }
2310
2311 static int bond_has_ip(struct bonding *bond)
2312 {
2313         struct vlan_entry *vlan, *vlan_next;
2314
2315         if (bond->master_ip)
2316                 return 1;
2317
2318         if (list_empty(&bond->vlan_list))
2319                 return 0;
2320
2321         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2322                                  vlan_list) {
2323                 if (vlan->vlan_ip)
2324                         return 1;
2325         }
2326
2327         return 0;
2328 }
2329
2330 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2331 {
2332         struct vlan_entry *vlan, *vlan_next;
2333
2334         if (ip == bond->master_ip)
2335                 return 1;
2336
2337         if (list_empty(&bond->vlan_list))
2338                 return 0;
2339
2340         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2341                                  vlan_list) {
2342                 if (ip == vlan->vlan_ip)
2343                         return 1;
2344         }
2345
2346         return 0;
2347 }
2348
2349 /*
2350  * We go to the (large) trouble of VLAN tagging ARP frames because
2351  * switches in VLAN mode (especially if ports are configured as
2352  * "native" to a VLAN) might not pass non-tagged frames.
2353  */
2354 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2355 {
2356         struct sk_buff *skb;
2357
2358         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2359                slave_dev->name, dest_ip, src_ip, vlan_id);
2360
2361         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2362                          NULL, slave_dev->dev_addr, NULL);
2363
2364         if (!skb) {
2365                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2366                 return;
2367         }
2368         if (vlan_id) {
2369                 skb = vlan_put_tag(skb, vlan_id);
2370                 if (!skb) {
2371                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2372                         return;
2373                 }
2374         }
2375         arp_xmit(skb);
2376 }
2377
2378
2379 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2380 {
2381         int i, vlan_id, rv;
2382         __be32 *targets = bond->params.arp_targets;
2383         struct vlan_entry *vlan, *vlan_next;
2384         struct net_device *vlan_dev;
2385         struct flowi fl;
2386         struct rtable *rt;
2387
2388         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2389                 if (!targets[i])
2390                         continue;
2391                 dprintk("basa: target %x\n", targets[i]);
2392                 if (list_empty(&bond->vlan_list)) {
2393                         dprintk("basa: empty vlan: arp_send\n");
2394                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2395                                       bond->master_ip, 0);
2396                         continue;
2397                 }
2398
2399                 /*
2400                  * If VLANs are configured, we do a route lookup to
2401                  * determine which VLAN interface would be used, so we
2402                  * can tag the ARP with the proper VLAN tag.
2403                  */
2404                 memset(&fl, 0, sizeof(fl));
2405                 fl.fl4_dst = targets[i];
2406                 fl.fl4_tos = RTO_ONLINK;
2407
2408                 rv = ip_route_output_key(&rt, &fl);
2409                 if (rv) {
2410                         if (net_ratelimit()) {
2411                                 printk(KERN_WARNING DRV_NAME
2412                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2413                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2414                         }
2415                         continue;
2416                 }
2417
2418                 /*
2419                  * This target is not on a VLAN
2420                  */
2421                 if (rt->u.dst.dev == bond->dev) {
2422                         ip_rt_put(rt);
2423                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2424                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2425                                       bond->master_ip, 0);
2426                         continue;
2427                 }
2428
2429                 vlan_id = 0;
2430                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2431                                          vlan_list) {
2432                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2433                         if (vlan_dev == rt->u.dst.dev) {
2434                                 vlan_id = vlan->vlan_id;
2435                                 dprintk("basa: vlan match on %s %d\n",
2436                                        vlan_dev->name, vlan_id);
2437                                 break;
2438                         }
2439                 }
2440
2441                 if (vlan_id) {
2442                         ip_rt_put(rt);
2443                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2444                                       vlan->vlan_ip, vlan_id);
2445                         continue;
2446                 }
2447
2448                 if (net_ratelimit()) {
2449                         printk(KERN_WARNING DRV_NAME
2450                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2451                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2452                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2453                 }
2454                 ip_rt_put(rt);
2455         }
2456 }
2457
2458 /*
2459  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2460  * for each VLAN above us.
2461  */
2462 static void bond_send_gratuitous_arp(struct bonding *bond)
2463 {
2464         struct slave *slave = bond->curr_active_slave;
2465         struct vlan_entry *vlan;
2466         struct net_device *vlan_dev;
2467
2468         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2469                                 slave ? slave->dev->name : "NULL");
2470         if (!slave)
2471                 return;
2472
2473         if (bond->master_ip) {
2474                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2475                                   bond->master_ip, 0);
2476         }
2477
2478         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2479                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2480                 if (vlan->vlan_ip) {
2481                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2482                                       vlan->vlan_ip, vlan->vlan_id);
2483                 }
2484         }
2485 }
2486
2487 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2488 {
2489         int i;
2490         __be32 *targets = bond->params.arp_targets;
2491
2492         targets = bond->params.arp_targets;
2493         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2494                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2495                         "%u.%u.%u.%u bhti(tip) %d\n",
2496                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2497                        bond_has_this_ip(bond, tip));
2498                 if (sip == targets[i]) {
2499                         if (bond_has_this_ip(bond, tip))
2500                                 slave->last_arp_rx = jiffies;
2501                         return;
2502                 }
2503         }
2504 }
2505
2506 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2507 {
2508         struct arphdr *arp;
2509         struct slave *slave;
2510         struct bonding *bond;
2511         unsigned char *arp_ptr;
2512         __be32 sip, tip;
2513
2514         if (dev->nd_net != &init_net)
2515                 goto out;
2516
2517         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2518                 goto out;
2519
2520         bond = dev->priv;
2521         read_lock(&bond->lock);
2522
2523         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2524                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2525                 orig_dev ? orig_dev->name : "NULL");
2526
2527         slave = bond_get_slave_by_dev(bond, orig_dev);
2528         if (!slave || !slave_do_arp_validate(bond, slave))
2529                 goto out_unlock;
2530
2531         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2532         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2533                                  (2 * dev->addr_len) +
2534                                  (2 * sizeof(u32)))))
2535                 goto out_unlock;
2536
2537         arp = arp_hdr(skb);
2538         if (arp->ar_hln != dev->addr_len ||
2539             skb->pkt_type == PACKET_OTHERHOST ||
2540             skb->pkt_type == PACKET_LOOPBACK ||
2541             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2542             arp->ar_pro != htons(ETH_P_IP) ||
2543             arp->ar_pln != 4)
2544                 goto out_unlock;
2545
2546         arp_ptr = (unsigned char *)(arp + 1);
2547         arp_ptr += dev->addr_len;
2548         memcpy(&sip, arp_ptr, 4);
2549         arp_ptr += 4 + dev->addr_len;
2550         memcpy(&tip, arp_ptr, 4);
2551
2552         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2553                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2554                 slave->state, bond->params.arp_validate,
2555                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2556
2557         /*
2558          * Backup slaves won't see the ARP reply, but do come through
2559          * here for each ARP probe (so we swap the sip/tip to validate
2560          * the probe).  In a "redundant switch, common router" type of
2561          * configuration, the ARP probe will (hopefully) travel from
2562          * the active, through one switch, the router, then the other
2563          * switch before reaching the backup.
2564          */
2565         if (slave->state == BOND_STATE_ACTIVE)
2566                 bond_validate_arp(bond, slave, sip, tip);
2567         else
2568                 bond_validate_arp(bond, slave, tip, sip);
2569
2570 out_unlock:
2571         read_unlock(&bond->lock);
2572 out:
2573         dev_kfree_skb(skb);
2574         return NET_RX_SUCCESS;
2575 }
2576
2577 /*
2578  * this function is called regularly to monitor each slave's link
2579  * ensuring that traffic is being sent and received when arp monitoring
2580  * is used in load-balancing mode. if the adapter has been dormant, then an
2581  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2582  * arp monitoring in active backup mode.
2583  */
2584 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2585 {
2586         struct bonding *bond = bond_dev->priv;
2587         struct slave *slave, *oldcurrent;
2588         int do_failover = 0;
2589         int delta_in_ticks;
2590         int i;
2591
2592         read_lock(&bond->lock);
2593
2594         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2595
2596         if (bond->kill_timers) {
2597                 goto out;
2598         }
2599
2600         if (bond->slave_cnt == 0) {
2601                 goto re_arm;
2602         }
2603
2604         read_lock(&bond->curr_slave_lock);
2605         oldcurrent = bond->curr_active_slave;
2606         read_unlock(&bond->curr_slave_lock);
2607
2608         /* see if any of the previous devices are up now (i.e. they have
2609          * xmt and rcv traffic). the curr_active_slave does not come into
2610          * the picture unless it is null. also, slave->jiffies is not needed
2611          * here because we send an arp on each slave and give a slave as
2612          * long as it needs to get the tx/rx within the delta.
2613          * TODO: what about up/down delay in arp mode? it wasn't here before
2614          *       so it can wait
2615          */
2616         bond_for_each_slave(bond, slave, i) {
2617                 if (slave->link != BOND_LINK_UP) {
2618                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2619                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2620
2621                                 slave->link  = BOND_LINK_UP;
2622                                 slave->state = BOND_STATE_ACTIVE;
2623
2624                                 /* primary_slave has no meaning in round-robin
2625                                  * mode. the window of a slave being up and
2626                                  * curr_active_slave being null after enslaving
2627                                  * is closed.
2628                                  */
2629                                 if (!oldcurrent) {
2630                                         printk(KERN_INFO DRV_NAME
2631                                                ": %s: link status definitely "
2632                                                "up for interface %s, ",
2633                                                bond_dev->name,
2634                                                slave->dev->name);
2635                                         do_failover = 1;
2636                                 } else {
2637                                         printk(KERN_INFO DRV_NAME
2638                                                ": %s: interface %s is now up\n",
2639                                                bond_dev->name,
2640                                                slave->dev->name);
2641                                 }
2642                         }
2643                 } else {
2644                         /* slave->link == BOND_LINK_UP */
2645
2646                         /* not all switches will respond to an arp request
2647                          * when the source ip is 0, so don't take the link down
2648                          * if we don't know our ip yet
2649                          */
2650                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2651                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2652                              bond_has_ip(bond))) {
2653
2654                                 slave->link  = BOND_LINK_DOWN;
2655                                 slave->state = BOND_STATE_BACKUP;
2656
2657                                 if (slave->link_failure_count < UINT_MAX) {
2658                                         slave->link_failure_count++;
2659                                 }
2660
2661                                 printk(KERN_INFO DRV_NAME
2662                                        ": %s: interface %s is now down.\n",
2663                                        bond_dev->name,
2664                                        slave->dev->name);
2665
2666                                 if (slave == oldcurrent) {
2667                                         do_failover = 1;
2668                                 }
2669                         }
2670                 }
2671
2672                 /* note: if switch is in round-robin mode, all links
2673                  * must tx arp to ensure all links rx an arp - otherwise
2674                  * links may oscillate or not come up at all; if switch is
2675                  * in something like xor mode, there is nothing we can
2676                  * do - all replies will be rx'ed on same link causing slaves
2677                  * to be unstable during low/no traffic periods
2678                  */
2679                 if (IS_UP(slave->dev)) {
2680                         bond_arp_send_all(bond, slave);
2681                 }
2682         }
2683
2684         if (do_failover) {
2685                 write_lock(&bond->curr_slave_lock);
2686
2687                 bond_select_active_slave(bond);
2688
2689                 write_unlock(&bond->curr_slave_lock);
2690         }
2691
2692 re_arm:
2693         if (bond->params.arp_interval) {
2694                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2695         }
2696 out:
2697         read_unlock(&bond->lock);
2698 }
2699
2700 /*
2701  * When using arp monitoring in active-backup mode, this function is
2702  * called to determine if any backup slaves have went down or a new
2703  * current slave needs to be found.
2704  * The backup slaves never generate traffic, they are considered up by merely
2705  * receiving traffic. If the current slave goes down, each backup slave will
2706  * be given the opportunity to tx/rx an arp before being taken down - this
2707  * prevents all slaves from being taken down due to the current slave not
2708  * sending any traffic for the backups to receive. The arps are not necessarily
2709  * necessary, any tx and rx traffic will keep the current slave up. While any
2710  * rx traffic will keep the backup slaves up, the current slave is responsible
2711  * for generating traffic to keep them up regardless of any other traffic they
2712  * may have received.
2713  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2714  */
2715 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2716 {
2717         struct bonding *bond = bond_dev->priv;
2718         struct slave *slave;
2719         int delta_in_ticks;
2720         int i;
2721
2722         read_lock(&bond->lock);
2723
2724         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2725
2726         if (bond->kill_timers) {
2727                 goto out;
2728         }
2729
2730         if (bond->slave_cnt == 0) {
2731                 goto re_arm;
2732         }
2733
2734         /* determine if any slave has come up or any backup slave has
2735          * gone down
2736          * TODO: what about up/down delay in arp mode? it wasn't here before
2737          *       so it can wait
2738          */
2739         bond_for_each_slave(bond, slave, i) {
2740                 if (slave->link != BOND_LINK_UP) {
2741                         if ((jiffies - slave_last_rx(bond, slave)) <=
2742                              delta_in_ticks) {
2743
2744                                 slave->link = BOND_LINK_UP;
2745
2746                                 write_lock(&bond->curr_slave_lock);
2747
2748                                 if ((!bond->curr_active_slave) &&
2749                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2750                                         bond_change_active_slave(bond, slave);
2751                                         bond->current_arp_slave = NULL;
2752                                 } else if (bond->curr_active_slave != slave) {
2753                                         /* this slave has just come up but we
2754                                          * already have a current slave; this
2755                                          * can also happen if bond_enslave adds
2756                                          * a new slave that is up while we are
2757                                          * searching for a new slave
2758                                          */
2759                                         bond_set_slave_inactive_flags(slave);
2760                                         bond->current_arp_slave = NULL;
2761                                 }
2762
2763                                 bond_set_carrier(bond);
2764
2765                                 if (slave == bond->curr_active_slave) {
2766                                         printk(KERN_INFO DRV_NAME
2767                                                ": %s: %s is up and now the "
2768                                                "active interface\n",
2769                                                bond_dev->name,
2770                                                slave->dev->name);
2771                                         netif_carrier_on(bond->dev);
2772                                 } else {
2773                                         printk(KERN_INFO DRV_NAME
2774                                                ": %s: backup interface %s is "
2775                                                "now up\n",
2776                                                bond_dev->name,
2777                                                slave->dev->name);
2778                                 }
2779
2780                                 write_unlock(&bond->curr_slave_lock);
2781                         }
2782                 } else {
2783                         read_lock(&bond->curr_slave_lock);
2784
2785                         if ((slave != bond->curr_active_slave) &&
2786                             (!bond->current_arp_slave) &&
2787                             (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2788                              bond_has_ip(bond))) {
2789                                 /* a backup slave has gone down; three times
2790                                  * the delta allows the current slave to be
2791                                  * taken out before the backup slave.
2792                                  * note: a non-null current_arp_slave indicates
2793                                  * the curr_active_slave went down and we are
2794                                  * searching for a new one; under this
2795                                  * condition we only take the curr_active_slave
2796                                  * down - this gives each slave a chance to
2797                                  * tx/rx traffic before being taken out
2798                                  */
2799
2800                                 read_unlock(&bond->curr_slave_lock);
2801
2802                                 slave->link  = BOND_LINK_DOWN;
2803
2804                                 if (slave->link_failure_count < UINT_MAX) {
2805                                         slave->link_failure_count++;
2806                                 }
2807
2808                                 bond_set_slave_inactive_flags(slave);
2809
2810                                 printk(KERN_INFO DRV_NAME
2811                                        ": %s: backup interface %s is now down\n",
2812                                        bond_dev->name,
2813                                        slave->dev->name);
2814                         } else {
2815                                 read_unlock(&bond->curr_slave_lock);
2816                         }
2817                 }
2818         }
2819
2820         read_lock(&bond->curr_slave_lock);
2821         slave = bond->curr_active_slave;
2822         read_unlock(&bond->curr_slave_lock);
2823
2824         if (slave) {
2825                 /* if we have sent traffic in the past 2*arp_intervals but
2826                  * haven't xmit and rx traffic in that time interval, select
2827                  * a different slave. slave->jiffies is only updated when
2828                  * a slave first becomes the curr_active_slave - not necessarily
2829                  * after every arp; this ensures the slave has a full 2*delta
2830                  * before being taken out. if a primary is being used, check
2831                  * if it is up and needs to take over as the curr_active_slave
2832                  */
2833                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2834             (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2835              bond_has_ip(bond))) &&
2836                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2837
2838                         slave->link  = BOND_LINK_DOWN;
2839
2840                         if (slave->link_failure_count < UINT_MAX) {
2841                                 slave->link_failure_count++;
2842                         }
2843
2844                         printk(KERN_INFO DRV_NAME
2845                                ": %s: link status down for active interface "
2846                                "%s, disabling it\n",
2847                                bond_dev->name,
2848                                slave->dev->name);
2849
2850                         write_lock(&bond->curr_slave_lock);
2851
2852                         bond_select_active_slave(bond);
2853                         slave = bond->curr_active_slave;
2854
2855                         write_unlock(&bond->curr_slave_lock);
2856
2857                         bond->current_arp_slave = slave;
2858
2859                         if (slave) {
2860                                 slave->jiffies = jiffies;
2861                         }
2862                 } else if ((bond->primary_slave) &&
2863                            (bond->primary_slave != slave) &&
2864                            (bond->primary_slave->link == BOND_LINK_UP)) {
2865                         /* at this point, slave is the curr_active_slave */
2866                         printk(KERN_INFO DRV_NAME
2867                                ": %s: changing from interface %s to primary "
2868                                "interface %s\n",
2869                                bond_dev->name,
2870                                slave->dev->name,
2871                                bond->primary_slave->dev->name);
2872
2873                         /* primary is up so switch to it */
2874                         write_lock(&bond->curr_slave_lock);
2875                         bond_change_active_slave(bond, bond->primary_slave);
2876                         write_unlock(&bond->curr_slave_lock);
2877
2878                         slave = bond->primary_slave;
2879                         slave->jiffies = jiffies;
2880                 } else {
2881                         bond->current_arp_slave = NULL;
2882                 }
2883
2884                 /* the current slave must tx an arp to ensure backup slaves
2885                  * rx traffic
2886                  */
2887                 if (slave && bond_has_ip(bond)) {
2888                         bond_arp_send_all(bond, slave);
2889                 }
2890         }
2891
2892         /* if we don't have a curr_active_slave, search for the next available
2893          * backup slave from the current_arp_slave and make it the candidate
2894          * for becoming the curr_active_slave
2895          */
2896         if (!slave) {
2897                 if (!bond->current_arp_slave) {
2898                         bond->current_arp_slave = bond->first_slave;
2899                 }
2900
2901                 if (bond->current_arp_slave) {
2902                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2903
2904                         /* search for next candidate */
2905                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2906                                 if (IS_UP(slave->dev)) {
2907                                         slave->link = BOND_LINK_BACK;
2908                                         bond_set_slave_active_flags(slave);
2909                                         bond_arp_send_all(bond, slave);
2910                                         slave->jiffies = jiffies;
2911                                         bond->current_arp_slave = slave;
2912                                         break;
2913                                 }
2914
2915                                 /* if the link state is up at this point, we
2916                                  * mark it down - this can happen if we have
2917                                  * simultaneous link failures and
2918                                  * reselect_active_interface doesn't make this
2919                                  * one the current slave so it is still marked
2920                                  * up when it is actually down
2921                                  */
2922                                 if (slave->link == BOND_LINK_UP) {
2923                                         slave->link  = BOND_LINK_DOWN;
2924                                         if (slave->link_failure_count < UINT_MAX) {
2925                                                 slave->link_failure_count++;
2926                                         }
2927
2928                                         bond_set_slave_inactive_flags(slave);
2929
2930                                         printk(KERN_INFO DRV_NAME
2931                                                ": %s: backup interface %s is "
2932                                                "now down.\n",
2933                                                bond_dev->name,
2934                                                slave->dev->name);
2935                                 }
2936                         }
2937                 }
2938         }
2939
2940 re_arm:
2941         if (bond->params.arp_interval) {
2942                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2943         }
2944 out:
2945         read_unlock(&bond->lock);
2946 }
2947
2948 /*------------------------------ proc/seq_file-------------------------------*/
2949
2950 #ifdef CONFIG_PROC_FS
2951
2952 #define SEQ_START_TOKEN ((void *)1)
2953
2954 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2955 {
2956         struct bonding *bond = seq->private;
2957         loff_t off = 0;
2958         struct slave *slave;
2959         int i;
2960
2961         /* make sure the bond won't be taken away */
2962         read_lock(&dev_base_lock);
2963         read_lock_bh(&bond->lock);
2964
2965         if (*pos == 0) {
2966                 return SEQ_START_TOKEN;
2967         }
2968
2969         bond_for_each_slave(bond, slave, i) {
2970                 if (++off == *pos) {
2971                         return slave;
2972                 }
2973         }
2974
2975         return NULL;
2976 }
2977
2978 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2979 {
2980         struct bonding *bond = seq->private;
2981         struct slave *slave = v;
2982
2983         ++*pos;
2984         if (v == SEQ_START_TOKEN) {
2985                 return bond->first_slave;
2986         }
2987
2988         slave = slave->next;
2989
2990         return (slave == bond->first_slave) ? NULL : slave;
2991 }
2992
2993 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2994 {
2995         struct bonding *bond = seq->private;
2996
2997         read_unlock_bh(&bond->lock);
2998         read_unlock(&dev_base_lock);
2999 }
3000
3001 static void bond_info_show_master(struct seq_file *seq)
3002 {
3003         struct bonding *bond = seq->private;
3004         struct slave *curr;
3005         int i;
3006         u32 target;
3007
3008         read_lock(&bond->curr_slave_lock);
3009         curr = bond->curr_active_slave;
3010         read_unlock(&bond->curr_slave_lock);
3011
3012         seq_printf(seq, "Bonding Mode: %s\n",
3013                    bond_mode_name(bond->params.mode));
3014
3015         if (bond->params.mode == BOND_MODE_XOR ||
3016                 bond->params.mode == BOND_MODE_8023AD) {
3017                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3018                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3019                         bond->params.xmit_policy);
3020         }
3021
3022         if (USES_PRIMARY(bond->params.mode)) {
3023                 seq_printf(seq, "Primary Slave: %s\n",
3024                            (bond->primary_slave) ?
3025                            bond->primary_slave->dev->name : "None");
3026
3027                 seq_printf(seq, "Currently Active Slave: %s\n",
3028                            (curr) ? curr->dev->name : "None");
3029         }
3030
3031         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3032                    "up" : "down");
3033         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3034         seq_printf(seq, "Up Delay (ms): %d\n",
3035                    bond->params.updelay * bond->params.miimon);
3036         seq_printf(seq, "Down Delay (ms): %d\n",
3037                    bond->params.downdelay * bond->params.miimon);
3038
3039
3040         /* ARP information */
3041         if(bond->params.arp_interval > 0) {
3042                 int printed=0;
3043                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3044                                 bond->params.arp_interval);
3045
3046                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3047
3048                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3049                         if (!bond->params.arp_targets[i])
3050                                 continue;
3051                         if (printed)
3052                                 seq_printf(seq, ",");
3053                         target = ntohl(bond->params.arp_targets[i]);
3054                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3055                         printed = 1;
3056                 }
3057                 seq_printf(seq, "\n");
3058         }
3059
3060         if (bond->params.mode == BOND_MODE_8023AD) {
3061                 struct ad_info ad_info;
3062                 DECLARE_MAC_BUF(mac);
3063
3064                 seq_puts(seq, "\n802.3ad info\n");
3065                 seq_printf(seq, "LACP rate: %s\n",
3066                            (bond->params.lacp_fast) ? "fast" : "slow");
3067
3068                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3069                         seq_printf(seq, "bond %s has no active aggregator\n",
3070                                    bond->dev->name);
3071                 } else {
3072                         seq_printf(seq, "Active Aggregator Info:\n");
3073
3074                         seq_printf(seq, "\tAggregator ID: %d\n",
3075                                    ad_info.aggregator_id);
3076                         seq_printf(seq, "\tNumber of ports: %d\n",
3077                                    ad_info.ports);
3078                         seq_printf(seq, "\tActor Key: %d\n",
3079                                    ad_info.actor_key);
3080                         seq_printf(seq, "\tPartner Key: %d\n",
3081                                    ad_info.partner_key);
3082                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3083                                    print_mac(mac, ad_info.partner_system));
3084                 }
3085         }
3086 }
3087
3088 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3089 {
3090         struct bonding *bond = seq->private;
3091         DECLARE_MAC_BUF(mac);
3092
3093         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3094         seq_printf(seq, "MII Status: %s\n",
3095                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3096         seq_printf(seq, "Link Failure Count: %u\n",
3097                    slave->link_failure_count);
3098
3099         seq_printf(seq,
3100                    "Permanent HW addr: %s\n",
3101                    print_mac(mac, slave->perm_hwaddr));
3102
3103         if (bond->params.mode == BOND_MODE_8023AD) {
3104                 const struct aggregator *agg
3105                         = SLAVE_AD_INFO(slave).port.aggregator;
3106
3107                 if (agg) {
3108                         seq_printf(seq, "Aggregator ID: %d\n",
3109                                    agg->aggregator_identifier);
3110                 } else {
3111                         seq_puts(seq, "Aggregator ID: N/A\n");
3112                 }
3113         }
3114 }
3115
3116 static int bond_info_seq_show(struct seq_file *seq, void *v)
3117 {
3118         if (v == SEQ_START_TOKEN) {
3119                 seq_printf(seq, "%s\n", version);
3120                 bond_info_show_master(seq);
3121         } else {
3122                 bond_info_show_slave(seq, v);
3123         }
3124
3125         return 0;
3126 }
3127
3128 static struct seq_operations bond_info_seq_ops = {
3129         .start = bond_info_seq_start,
3130         .next  = bond_info_seq_next,
3131         .stop  = bond_info_seq_stop,
3132         .show  = bond_info_seq_show,
3133 };
3134
3135 static int bond_info_open(struct inode *inode, struct file *file)
3136 {
3137         struct seq_file *seq;
3138         struct proc_dir_entry *proc;
3139         int res;
3140
3141         res = seq_open(file, &bond_info_seq_ops);
3142         if (!res) {
3143                 /* recover the pointer buried in proc_dir_entry data */
3144                 seq = file->private_data;
3145                 proc = PDE(inode);
3146                 seq->private = proc->data;
3147         }
3148
3149         return res;
3150 }
3151
3152 static const struct file_operations bond_info_fops = {
3153         .owner   = THIS_MODULE,
3154         .open    = bond_info_open,
3155         .read    = seq_read,
3156         .llseek  = seq_lseek,
3157         .release = seq_release,
3158 };
3159
3160 static int bond_create_proc_entry(struct bonding *bond)
3161 {
3162         struct net_device *bond_dev = bond->dev;
3163
3164         if (bond_proc_dir) {
3165                 bond->proc_entry = create_proc_entry(bond_dev->name,
3166                                                      S_IRUGO,
3167                                                      bond_proc_dir);
3168                 if (bond->proc_entry == NULL) {
3169                         printk(KERN_WARNING DRV_NAME
3170                                ": Warning: Cannot create /proc/net/%s/%s\n",
3171                                DRV_NAME, bond_dev->name);
3172                 } else {
3173                         bond->proc_entry->data = bond;
3174                         bond->proc_entry->proc_fops = &bond_info_fops;
3175                         bond->proc_entry->owner = THIS_MODULE;
3176                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3177                 }
3178         }
3179
3180         return 0;
3181 }
3182
3183 static void bond_remove_proc_entry(struct bonding *bond)
3184 {
3185         if (bond_proc_dir && bond->proc_entry) {
3186                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3187                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3188                 bond->proc_entry = NULL;
3189         }
3190 }
3191
3192 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3193  * Caller must hold rtnl_lock.
3194  */
3195 static void bond_create_proc_dir(void)
3196 {
3197         int len = strlen(DRV_NAME);
3198
3199         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3200              bond_proc_dir = bond_proc_dir->next) {
3201                 if ((bond_proc_dir->namelen == len) &&
3202                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3203                         break;
3204                 }
3205         }
3206
3207         if (!bond_proc_dir) {
3208                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3209                 if (bond_proc_dir) {
3210                         bond_proc_dir->owner = THIS_MODULE;
3211                 } else {
3212                         printk(KERN_WARNING DRV_NAME
3213                                 ": Warning: cannot create /proc/net/%s\n",
3214                                 DRV_NAME);
3215                 }
3216         }
3217 }
3218
3219 /* Destroy the bonding directory under /proc/net, if empty.
3220  * Caller must hold rtnl_lock.
3221  */
3222 static void bond_destroy_proc_dir(void)
3223 {
3224         struct proc_dir_entry *de;
3225
3226         if (!bond_proc_dir) {
3227                 return;
3228         }
3229
3230         /* verify that the /proc dir is empty */
3231         for (de = bond_proc_dir->subdir; de; de = de->next) {
3232                 /* ignore . and .. */
3233                 if (*(de->name) != '.') {
3234                         break;
3235                 }
3236         }
3237
3238         if (de) {
3239                 if (bond_proc_dir->owner == THIS_MODULE) {
3240                         bond_proc_dir->owner = NULL;
3241                 }
3242         } else {
3243                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3244                 bond_proc_dir = NULL;
3245         }
3246 }
3247 #endif /* CONFIG_PROC_FS */
3248
3249 /*-------------------------- netdev event handling --------------------------*/
3250
3251 /*
3252  * Change device name
3253  */
3254 static int bond_event_changename(struct bonding *bond)
3255 {
3256 #ifdef CONFIG_PROC_FS
3257         bond_remove_proc_entry(bond);
3258         bond_create_proc_entry(bond);
3259 #endif
3260         down_write(&(bonding_rwsem));
3261         bond_destroy_sysfs_entry(bond);
3262         bond_create_sysfs_entry(bond);
3263         up_write(&(bonding_rwsem));
3264         return NOTIFY_DONE;
3265 }
3266
3267 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3268 {
3269         struct bonding *event_bond = bond_dev->priv;
3270
3271         switch (event) {
3272         case NETDEV_CHANGENAME:
3273                 return bond_event_changename(event_bond);
3274         case NETDEV_UNREGISTER:
3275                 /*
3276                  * TODO: remove a bond from the list?
3277                  */
3278                 break;
3279         default:
3280                 break;
3281         }
3282
3283         return NOTIFY_DONE;
3284 }
3285
3286 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3287 {
3288         struct net_device *bond_dev = slave_dev->master;
3289         struct bonding *bond = bond_dev->priv;
3290
3291         switch (event) {
3292         case NETDEV_UNREGISTER:
3293                 if (bond_dev) {
3294                         bond_release(bond_dev, slave_dev);
3295                 }
3296                 break;
3297         case NETDEV_CHANGE:
3298                 /*
3299                  * TODO: is this what we get if somebody
3300                  * sets up a hierarchical bond, then rmmod's
3301                  * one of the slave bonding devices?
3302                  */
3303                 break;
3304         case NETDEV_DOWN:
3305                 /*
3306                  * ... Or is it this?
3307                  */
3308                 break;
3309         case NETDEV_CHANGEMTU:
3310                 /*
3311                  * TODO: Should slaves be allowed to
3312                  * independently alter their MTU?  For
3313                  * an active-backup bond, slaves need
3314                  * not be the same type of device, so
3315                  * MTUs may vary.  For other modes,
3316                  * slaves arguably should have the
3317                  * same MTUs. To do this, we'd need to
3318                  * take over the slave's change_mtu
3319                  * function for the duration of their
3320                  * servitude.
3321                  */
3322                 break;
3323         case NETDEV_CHANGENAME:
3324                 /*
3325                  * TODO: handle changing the primary's name
3326                  */
3327                 break;
3328         case NETDEV_FEAT_CHANGE:
3329                 bond_compute_features(bond);
3330                 break;
3331         default:
3332                 break;
3333         }
3334
3335         return NOTIFY_DONE;
3336 }
3337
3338 /*
3339  * bond_netdev_event: handle netdev notifier chain events.
3340  *
3341  * This function receives events for the netdev chain.  The caller (an
3342  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3343  * locks for us to safely manipulate the slave devices (RTNL lock,
3344  * dev_probe_lock).
3345  */
3346 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3347 {
3348         struct net_device *event_dev = (struct net_device *)ptr;
3349
3350         if (event_dev->nd_net != &init_net)
3351                 return NOTIFY_DONE;
3352
3353         dprintk("event_dev: %s, event: %lx\n",
3354                 (event_dev ? event_dev->name : "None"),
3355                 event);
3356
3357         if (!(event_dev->priv_flags & IFF_BONDING))
3358                 return NOTIFY_DONE;
3359
3360         if (event_dev->flags & IFF_MASTER) {
3361                 dprintk("IFF_MASTER\n");
3362                 return bond_master_netdev_event(event, event_dev);
3363         }
3364
3365         if (event_dev->flags & IFF_SLAVE) {
3366                 dprintk("IFF_SLAVE\n");
3367                 return bond_slave_netdev_event(event, event_dev);
3368         }
3369
3370         return NOTIFY_DONE;
3371 }
3372
3373 /*
3374  * bond_inetaddr_event: handle inetaddr notifier chain events.
3375  *
3376  * We keep track of device IPs primarily to use as source addresses in
3377  * ARP monitor probes (rather than spewing out broadcasts all the time).
3378  *
3379  * We track one IP for the main device (if it has one), plus one per VLAN.
3380  */
3381 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3382 {
3383         struct in_ifaddr *ifa = ptr;
3384         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3385         struct bonding *bond, *bond_next;
3386         struct vlan_entry *vlan, *vlan_next;
3387
3388         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3389                 if (bond->dev == event_dev) {
3390                         switch (event) {
3391                         case NETDEV_UP:
3392                                 bond->master_ip = ifa->ifa_local;
3393                                 return NOTIFY_OK;
3394                         case NETDEV_DOWN:
3395                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3396                                 return NOTIFY_OK;
3397                         default:
3398                                 return NOTIFY_DONE;
3399                         }
3400                 }
3401
3402                 if (list_empty(&bond->vlan_list))
3403                         continue;
3404
3405                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3406                                          vlan_list) {
3407                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3408                         if (vlan_dev == event_dev) {
3409                                 switch (event) {
3410                                 case NETDEV_UP:
3411                                         vlan->vlan_ip = ifa->ifa_local;
3412                                         return NOTIFY_OK;
3413                                 case NETDEV_DOWN:
3414                                         vlan->vlan_ip =
3415                                                 bond_glean_dev_ip(vlan_dev);
3416                                         return NOTIFY_OK;
3417                                 default:
3418                                         return NOTIFY_DONE;
3419                                 }
3420                         }
3421                 }
3422         }
3423         return NOTIFY_DONE;
3424 }
3425
3426 static struct notifier_block bond_netdev_notifier = {
3427         .notifier_call = bond_netdev_event,
3428 };
3429
3430 static struct notifier_block bond_inetaddr_notifier = {
3431         .notifier_call = bond_inetaddr_event,
3432 };
3433
3434 /*-------------------------- Packet type handling ---------------------------*/
3435
3436 /* register to receive lacpdus on a bond */
3437 static void bond_register_lacpdu(struct bonding *bond)
3438 {
3439         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3440
3441         /* initialize packet type */
3442         pk_type->type = PKT_TYPE_LACPDU;
3443         pk_type->dev = bond->dev;
3444         pk_type->func = bond_3ad_lacpdu_recv;
3445
3446         dev_add_pack(pk_type);
3447 }
3448
3449 /* unregister to receive lacpdus on a bond */
3450 static void bond_unregister_lacpdu(struct bonding *bond)
3451 {
3452         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3453 }
3454
3455 void bond_register_arp(struct bonding *bond)
3456 {
3457         struct packet_type *pt = &bond->arp_mon_pt;
3458
3459         if (pt->type)
3460                 return;
3461
3462         pt->type = htons(ETH_P_ARP);
3463         pt->dev = bond->dev;
3464         pt->func = bond_arp_rcv;
3465         dev_add_pack(pt);
3466 }
3467
3468 void bond_unregister_arp(struct bonding *bond)
3469 {
3470         struct packet_type *pt = &bond->arp_mon_pt;
3471
3472         dev_remove_pack(pt);
3473         pt->type = 0;
3474 }
3475
3476 /*---------------------------- Hashing Policies -----------------------------*/
3477
3478 /*
3479  * Hash for the output device based upon layer 3 and layer 4 data. If
3480  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3481  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3482  */
3483 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3484                                     struct net_device *bond_dev, int count)
3485 {
3486         struct ethhdr *data = (struct ethhdr *)skb->data;
3487         struct iphdr *iph = ip_hdr(skb);
3488         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3489         int layer4_xor = 0;
3490
3491         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3492                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3493                     (iph->protocol == IPPROTO_TCP ||
3494                      iph->protocol == IPPROTO_UDP)) {
3495                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3496                 }
3497                 return (layer4_xor ^
3498                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3499
3500         }
3501
3502         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3503 }
3504
3505 /*
3506  * Hash for the output device based upon layer 2 data
3507  */
3508 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3509                                    struct net_device *bond_dev, int count)
3510 {
3511         struct ethhdr *data = (struct ethhdr *)skb->data;
3512
3513         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3514 }
3515
3516 /*-------------------------- Device entry points ----------------------------*/
3517
3518 static int bond_open(struct net_device *bond_dev)
3519 {
3520         struct bonding *bond = bond_dev->priv;
3521         struct timer_list *mii_timer = &bond->mii_timer;
3522         struct timer_list *arp_timer = &bond->arp_timer;
3523
3524         bond->kill_timers = 0;
3525
3526         if ((bond->params.mode == BOND_MODE_TLB) ||
3527             (bond->params.mode == BOND_MODE_ALB)) {
3528                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3529
3530                 /* bond_alb_initialize must be called before the timer
3531                  * is started.
3532                  */
3533                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3534                         /* something went wrong - fail the open operation */
3535                         return -1;
3536                 }
3537
3538                 init_timer(alb_timer);
3539                 alb_timer->expires  = jiffies + 1;
3540                 alb_timer->data     = (unsigned long)bond;
3541                 alb_timer->function = (void *)&bond_alb_monitor;
3542                 add_timer(alb_timer);
3543         }
3544
3545         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3546                 init_timer(mii_timer);
3547                 mii_timer->expires  = jiffies + 1;
3548                 mii_timer->data     = (unsigned long)bond_dev;
3549                 mii_timer->function = (void *)&bond_mii_monitor;
3550                 add_timer(mii_timer);
3551         }
3552
3553         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3554                 init_timer(arp_timer);
3555                 arp_timer->expires  = jiffies + 1;
3556                 arp_timer->data     = (unsigned long)bond_dev;
3557                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3558                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3559                 } else {
3560                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3561                 }
3562                 if (bond->params.arp_validate)
3563                         bond_register_arp(bond);
3564
3565                 add_timer(arp_timer);
3566         }
3567
3568         if (bond->params.mode == BOND_MODE_8023AD) {
3569                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3570                 init_timer(ad_timer);
3571                 ad_timer->expires  = jiffies + 1;
3572                 ad_timer->data     = (unsigned long)bond;
3573                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3574                 add_timer(ad_timer);
3575
3576                 /* register to receive LACPDUs */
3577                 bond_register_lacpdu(bond);
3578         }
3579
3580         return 0;
3581 }
3582
3583 static int bond_close(struct net_device *bond_dev)
3584 {
3585         struct bonding *bond = bond_dev->priv;
3586
3587         if (bond->params.mode == BOND_MODE_8023AD) {
3588                 /* Unregister the receive of LACPDUs */
3589                 bond_unregister_lacpdu(bond);
3590         }
3591
3592         if (bond->params.arp_validate)
3593                 bond_unregister_arp(bond);
3594
3595         write_lock_bh(&bond->lock);
3596
3597
3598         /* signal timers not to re-arm */
3599         bond->kill_timers = 1;
3600
3601         write_unlock_bh(&bond->lock);
3602
3603         /* del_timer_sync must run without holding the bond->lock
3604          * because a running timer might be trying to hold it too
3605          */
3606
3607         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3608                 del_timer_sync(&bond->mii_timer);
3609         }
3610
3611         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3612                 del_timer_sync(&bond->arp_timer);
3613         }
3614
3615         switch (bond->params.mode) {
3616         case BOND_MODE_8023AD:
3617                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3618                 break;
3619         case BOND_MODE_TLB:
3620         case BOND_MODE_ALB:
3621                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3622                 break;
3623         default:
3624                 break;
3625         }
3626
3627
3628         if ((bond->params.mode == BOND_MODE_TLB) ||
3629             (bond->params.mode == BOND_MODE_ALB)) {
3630                 /* Must be called only after all
3631                  * slaves have been released
3632                  */
3633                 bond_alb_deinitialize(bond);
3634         }
3635
3636         return 0;
3637 }
3638
3639 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3640 {
3641         struct bonding *bond = bond_dev->priv;
3642         struct net_device_stats *stats = &(bond->stats), *sstats;
3643         struct slave *slave;
3644         int i;
3645
3646         memset(stats, 0, sizeof(struct net_device_stats));
3647
3648         read_lock_bh(&bond->lock);
3649
3650         bond_for_each_slave(bond, slave, i) {
3651                 sstats = slave->dev->get_stats(slave->dev);
3652                 stats->rx_packets += sstats->rx_packets;
3653                 stats->rx_bytes += sstats->rx_bytes;
3654                 stats->rx_errors += sstats->rx_errors;
3655                 stats->rx_dropped += sstats->rx_dropped;
3656
3657                 stats->tx_packets += sstats->tx_packets;
3658                 stats->tx_bytes += sstats->tx_bytes;
3659                 stats->tx_errors += sstats->tx_errors;
3660                 stats->tx_dropped += sstats->tx_dropped;
3661
3662                 stats->multicast += sstats->multicast;
3663                 stats->collisions += sstats->collisions;
3664
3665                 stats->rx_length_errors += sstats->rx_length_errors;
3666                 stats->rx_over_errors += sstats->rx_over_errors;
3667                 stats->rx_crc_errors += sstats->rx_crc_errors;
3668                 stats->rx_frame_errors += sstats->rx_frame_errors;
3669                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3670                 stats->rx_missed_errors += sstats->rx_missed_errors;
3671
3672                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3673                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3674                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3675                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3676                 stats->tx_window_errors += sstats->tx_window_errors;
3677         }
3678
3679         read_unlock_bh(&bond->lock);
3680
3681         return stats;
3682 }
3683
3684 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3685 {
3686         struct net_device *slave_dev = NULL;
3687         struct ifbond k_binfo;
3688         struct ifbond __user *u_binfo = NULL;
3689         struct ifslave k_sinfo;
3690         struct ifslave __user *u_sinfo = NULL;
3691         struct mii_ioctl_data *mii = NULL;
3692         int res = 0;
3693
3694         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3695                 bond_dev->name, cmd);
3696
3697         switch (cmd) {
3698         case SIOCGMIIPHY:
3699                 mii = if_mii(ifr);
3700                 if (!mii) {
3701                         return -EINVAL;
3702                 }
3703                 mii->phy_id = 0;
3704                 /* Fall Through */
3705         case SIOCGMIIREG:
3706                 /*
3707                  * We do this again just in case we were called by SIOCGMIIREG
3708                  * instead of SIOCGMIIPHY.
3709                  */
3710                 mii = if_mii(ifr);
3711                 if (!mii) {
3712                         return -EINVAL;
3713                 }
3714
3715                 if (mii->reg_num == 1) {
3716                         struct bonding *bond = bond_dev->priv;
3717                         mii->val_out = 0;
3718                         read_lock_bh(&bond->lock);
3719                         read_lock(&bond->curr_slave_lock);
3720                         if (netif_carrier_ok(bond->dev)) {
3721                                 mii->val_out = BMSR_LSTATUS;
3722                         }
3723                         read_unlock(&bond->curr_slave_lock);
3724                         read_unlock_bh(&bond->lock);
3725                 }
3726
3727                 return 0;
3728         case BOND_INFO_QUERY_OLD:
3729         case SIOCBONDINFOQUERY:
3730                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3731
3732                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3733                         return -EFAULT;
3734                 }
3735
3736                 res = bond_info_query(bond_dev, &k_binfo);
3737                 if (res == 0) {
3738                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3739                                 return -EFAULT;
3740                         }
3741                 }
3742
3743                 return res;
3744         case BOND_SLAVE_INFO_QUERY_OLD:
3745         case SIOCBONDSLAVEINFOQUERY:
3746                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3747
3748                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3749                         return -EFAULT;
3750                 }
3751
3752                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3753                 if (res == 0) {
3754                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3755                                 return -EFAULT;
3756                         }
3757                 }
3758
3759                 return res;
3760         default:
3761                 /* Go on */
3762                 break;
3763         }
3764
3765         if (!capable(CAP_NET_ADMIN)) {
3766                 return -EPERM;
3767         }
3768
3769         down_write(&(bonding_rwsem));
3770         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3771
3772         dprintk("slave_dev=%p: \n", slave_dev);
3773
3774         if (!slave_dev) {
3775                 res = -ENODEV;
3776         } else {
3777                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3778                 switch (cmd) {
3779                 case BOND_ENSLAVE_OLD:
3780                 case SIOCBONDENSLAVE:
3781                         res = bond_enslave(bond_dev, slave_dev);
3782                         break;
3783                 case BOND_RELEASE_OLD:
3784                 case SIOCBONDRELEASE:
3785                         res = bond_release(bond_dev, slave_dev);
3786                         break;
3787                 case BOND_SETHWADDR_OLD:
3788                 case SIOCBONDSETHWADDR:
3789                         res = bond_sethwaddr(bond_dev, slave_dev);
3790                         break;
3791                 case BOND_CHANGE_ACTIVE_OLD:
3792                 case SIOCBONDCHANGEACTIVE:
3793                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3794                         break;
3795                 default:
3796                         res = -EOPNOTSUPP;
3797                 }
3798
3799                 dev_put(slave_dev);
3800         }
3801
3802         up_write(&(bonding_rwsem));
3803         return res;
3804 }
3805
3806 static void bond_set_multicast_list(struct net_device *bond_dev)
3807 {
3808         struct bonding *bond = bond_dev->priv;
3809         struct dev_mc_list *dmi;
3810
3811         write_lock_bh(&bond->lock);
3812
3813         /*
3814          * Do promisc before checking multicast_mode
3815          */
3816         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3817                 bond_set_promiscuity(bond, 1);
3818         }
3819
3820         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3821                 bond_set_promiscuity(bond, -1);
3822         }
3823
3824         /* set allmulti flag to slaves */
3825         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3826                 bond_set_allmulti(bond, 1);
3827         }
3828
3829         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3830                 bond_set_allmulti(bond, -1);
3831         }
3832
3833         bond->flags = bond_dev->flags;
3834
3835         /* looking for addresses to add to slaves' mc list */
3836         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3837                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3838                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3839                 }
3840         }
3841
3842         /* looking for addresses to delete from slaves' list */
3843         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3844                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3845                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3846                 }
3847         }
3848
3849         /* save master's multicast list */
3850         bond_mc_list_destroy(bond);
3851         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3852
3853         write_unlock_bh(&bond->lock);
3854 }
3855
3856 /*
3857  * Change the MTU of all of a master's slaves to match the master
3858  */
3859 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3860 {
3861         struct bonding *bond = bond_dev->priv;
3862         struct slave *slave, *stop_at;
3863         int res = 0;
3864         int i;
3865
3866         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3867                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3868
3869         /* Can't hold bond->lock with bh disabled here since
3870          * some base drivers panic. On the other hand we can't
3871          * hold bond->lock without bh disabled because we'll
3872          * deadlock. The only solution is to rely on the fact
3873          * that we're under rtnl_lock here, and the slaves
3874          * list won't change. This doesn't solve the problem
3875          * of setting the slave's MTU while it is
3876          * transmitting, but the assumption is that the base
3877          * driver can handle that.
3878          *
3879          * TODO: figure out a way to safely iterate the slaves
3880          * list, but without holding a lock around the actual
3881          * call to the base driver.
3882          */
3883
3884         bond_for_each_slave(bond, slave, i) {
3885                 dprintk("s %p s->p %p c_m %p\n", slave,
3886                         slave->prev, slave->dev->change_mtu);
3887
3888                 res = dev_set_mtu(slave->dev, new_mtu);
3889
3890                 if (res) {
3891                         /* If we failed to set the slave's mtu to the new value
3892                          * we must abort the operation even in ACTIVE_BACKUP
3893                          * mode, because if we allow the backup slaves to have
3894                          * different mtu values than the active slave we'll
3895                          * need to change their mtu when doing a failover. That
3896                          * means changing their mtu from timer context, which
3897                          * is probably not a good idea.
3898                          */
3899                         dprintk("err %d %s\n", res, slave->dev->name);
3900                         goto unwind;
3901                 }
3902         }
3903
3904         bond_dev->mtu = new_mtu;
3905
3906         return 0;
3907
3908 unwind:
3909         /* unwind from head to the slave that failed */
3910         stop_at = slave;
3911         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3912                 int tmp_res;
3913
3914                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3915                 if (tmp_res) {
3916                         dprintk("unwind err %d dev %s\n", tmp_res,
3917                                 slave->dev->name);
3918                 }
3919         }
3920
3921         return res;
3922 }
3923
3924 /*
3925  * Change HW address
3926  *
3927  * Note that many devices must be down to change the HW address, and
3928  * downing the master releases all slaves.  We can make bonds full of
3929  * bonding devices to test this, however.
3930  */
3931 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3932 {
3933         struct bonding *bond = bond_dev->priv;
3934         struct sockaddr *sa = addr, tmp_sa;
3935         struct slave *slave, *stop_at;
3936         int res = 0;
3937         int i;
3938
3939         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3940
3941         if (!bond->do_set_mac_addr)
3942                 return -EOPNOTSUPP;
3943
3944         if (!is_valid_ether_addr(sa->sa_data)) {
3945                 return -EADDRNOTAVAIL;
3946         }
3947
3948         /* Can't hold bond->lock with bh disabled here since
3949          * some base drivers panic. On the other hand we can't
3950          * hold bond->lock without bh disabled because we'll
3951          * deadlock. The only solution is to rely on the fact
3952          * that we're under rtnl_lock here, and the slaves
3953          * list won't change. This doesn't solve the problem
3954          * of setting the slave's hw address while it is
3955          * transmitting, but the assumption is that the base
3956          * driver can handle that.
3957          *
3958          * TODO: figure out a way to safely iterate the slaves
3959          * list, but without holding a lock around the actual
3960          * call to the base driver.
3961          */
3962
3963         bond_for_each_slave(bond, slave, i) {
3964                 dprintk("slave %p %s\n", slave, slave->dev->name);
3965
3966                 if (slave->dev->set_mac_address == NULL) {
3967                         res = -EOPNOTSUPP;
3968                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3969                         goto unwind;
3970                 }
3971
3972                 res = dev_set_mac_address(slave->dev, addr);
3973                 if (res) {
3974                         /* TODO: consider downing the slave
3975                          * and retry ?
3976                          * User should expect communications
3977                          * breakage anyway until ARP finish
3978                          * updating, so...
3979                          */
3980                         dprintk("err %d %s\n", res, slave->dev->name);
3981                         goto unwind;
3982                 }
3983         }
3984
3985         /* success */
3986         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3987         return 0;
3988
3989 unwind:
3990         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3991         tmp_sa.sa_family = bond_dev->type;
3992
3993         /* unwind from head to the slave that failed */
3994         stop_at = slave;
3995         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3996                 int tmp_res;
3997
3998                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3999                 if (tmp_res) {
4000                         dprintk("unwind err %d dev %s\n", tmp_res,
4001                                 slave->dev->name);
4002                 }
4003         }
4004
4005         return res;
4006 }
4007
4008 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4009 {
4010         struct bonding *bond = bond_dev->priv;
4011         struct slave *slave, *start_at;
4012         int i;
4013         int res = 1;
4014
4015         read_lock(&bond->lock);
4016
4017         if (!BOND_IS_OK(bond)) {
4018                 goto out;
4019         }
4020
4021         read_lock(&bond->curr_slave_lock);
4022         slave = start_at = bond->curr_active_slave;
4023         read_unlock(&bond->curr_slave_lock);
4024
4025         if (!slave) {
4026                 goto out;
4027         }
4028
4029         bond_for_each_slave_from(bond, slave, i, start_at) {
4030                 if (IS_UP(slave->dev) &&
4031                     (slave->link == BOND_LINK_UP) &&
4032                     (slave->state == BOND_STATE_ACTIVE)) {
4033                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4034
4035                         write_lock(&bond->curr_slave_lock);
4036                         bond->curr_active_slave = slave->next;
4037                         write_unlock(&bond->curr_slave_lock);
4038
4039                         break;
4040                 }
4041         }
4042
4043
4044 out:
4045         if (res) {
4046                 /* no suitable interface, frame not sent */
4047                 dev_kfree_skb(skb);
4048         }
4049         read_unlock(&bond->lock);
4050         return 0;
4051 }
4052
4053
4054 /*
4055  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4056  * the bond has a usable interface.
4057  */
4058 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4059 {
4060         struct bonding *bond = bond_dev->priv;
4061         int res = 1;
4062
4063         read_lock(&bond->lock);
4064         read_lock(&bond->curr_slave_lock);
4065
4066         if (!BOND_IS_OK(bond)) {
4067                 goto out;
4068         }
4069
4070         if (!bond->curr_active_slave)
4071                 goto out;
4072
4073         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4074
4075 out:
4076         if (res) {
4077                 /* no suitable interface, frame not sent */
4078                 dev_kfree_skb(skb);
4079         }
4080         read_unlock(&bond->curr_slave_lock);
4081         read_unlock(&bond->lock);
4082         return 0;
4083 }
4084
4085 /*
4086  * In bond_xmit_xor() , we determine the output device by using a pre-
4087  * determined xmit_hash_policy(), If the selected device is not enabled,
4088  * find the next active slave.
4089  */
4090 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4091 {
4092         struct bonding *bond = bond_dev->priv;
4093         struct slave *slave, *start_at;
4094         int slave_no;
4095         int i;
4096         int res = 1;
4097
4098         read_lock(&bond->lock);
4099
4100         if (!BOND_IS_OK(bond)) {
4101                 goto out;
4102         }
4103
4104         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4105
4106         bond_for_each_slave(bond, slave, i) {
4107                 slave_no--;
4108                 if (slave_no < 0) {
4109                         break;
4110                 }
4111         }
4112
4113         start_at = slave;
4114
4115         bond_for_each_slave_from(bond, slave, i, start_at) {
4116                 if (IS_UP(slave->dev) &&
4117                     (slave->link == BOND_LINK_UP) &&
4118                     (slave->state == BOND_STATE_ACTIVE)) {
4119                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4120                         break;
4121                 }
4122         }
4123
4124 out:
4125         if (res) {
4126                 /* no suitable interface, frame not sent */
4127                 dev_kfree_skb(skb);
4128         }
4129         read_unlock(&bond->lock);
4130         return 0;
4131 }
4132
4133 /*
4134  * in broadcast mode, we send everything to all usable interfaces.
4135  */
4136 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4137 {
4138         struct bonding *bond = bond_dev->priv;
4139         struct slave *slave, *start_at;
4140         struct net_device *tx_dev = NULL;
4141         int i;
4142         int res = 1;
4143
4144         read_lock(&bond->lock);
4145
4146         if (!BOND_IS_OK(bond)) {
4147                 goto out;
4148         }
4149
4150         read_lock(&bond->curr_slave_lock);
4151         start_at = bond->curr_active_slave;
4152         read_unlock(&bond->curr_slave_lock);
4153
4154         if (!start_at) {
4155                 goto out;
4156         }
4157
4158         bond_for_each_slave_from(bond, slave, i, start_at) {
4159                 if (IS_UP(slave->dev) &&
4160                     (slave->link == BOND_LINK_UP) &&
4161                     (slave->state == BOND_STATE_ACTIVE)) {
4162                         if (tx_dev) {
4163                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4164                                 if (!skb2) {
4165                                         printk(KERN_ERR DRV_NAME
4166                                                ": %s: Error: bond_xmit_broadcast(): "
4167                                                "skb_clone() failed\n",
4168                                                bond_dev->name);
4169                                         continue;
4170                                 }
4171
4172                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4173                                 if (res) {
4174                                         dev_kfree_skb(skb2);
4175                                         continue;
4176                                 }
4177                         }
4178                         tx_dev = slave->dev;
4179                 }
4180         }
4181
4182         if (tx_dev) {
4183                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4184         }
4185
4186 out:
4187         if (res) {
4188                 /* no suitable interface, frame not sent */
4189                 dev_kfree_skb(skb);
4190         }
4191         /* frame sent to all suitable interfaces */
4192         read_unlock(&bond->lock);
4193         return 0;
4194 }
4195
4196 /*------------------------- Device initialization ---------------------------*/
4197
4198 /*
4199  * set bond mode specific net device operations
4200  */
4201 void bond_set_mode_ops(struct bonding *bond, int mode)
4202 {
4203         struct net_device *bond_dev = bond->dev;
4204
4205         switch (mode) {
4206         case BOND_MODE_ROUNDROBIN:
4207                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4208                 break;
4209         case BOND_MODE_ACTIVEBACKUP:
4210                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4211                 break;
4212         case BOND_MODE_XOR:
4213                 bond_dev->hard_start_xmit = bond_xmit_xor;
4214                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4215                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4216                 else
4217                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4218                 break;
4219         case BOND_MODE_BROADCAST:
4220                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4221                 break;
4222         case BOND_MODE_8023AD:
4223                 bond_set_master_3ad_flags(bond);
4224                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4225                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4226                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4227                 else
4228                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4229                 break;
4230         case BOND_MODE_ALB:
4231                 bond_set_master_alb_flags(bond);
4232                 /* FALLTHRU */
4233         case BOND_MODE_TLB:
4234                 bond_dev->hard_start_xmit = bond_alb_xmit;
4235                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4236                 break;
4237         default:
4238                 /* Should never happen, mode already checked */
4239                 printk(KERN_ERR DRV_NAME
4240                        ": %s: Error: Unknown bonding mode %d\n",
4241                        bond_dev->name,
4242                        mode);
4243                 break;
4244         }
4245 }
4246
4247 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4248                                     struct ethtool_drvinfo *drvinfo)
4249 {
4250         strncpy(drvinfo->driver, DRV_NAME, 32);
4251         strncpy(drvinfo->version, DRV_VERSION, 32);
4252         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4253 }
4254
4255 static const struct ethtool_ops bond_ethtool_ops = {
4256         .get_drvinfo            = bond_ethtool_get_drvinfo,
4257 };
4258
4259 /*
4260  * Does not allocate but creates a /proc entry.
4261  * Allowed to fail.
4262  */
4263 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4264 {
4265         struct bonding *bond = bond_dev->priv;
4266
4267         dprintk("Begin bond_init for %s\n", bond_dev->name);
4268
4269         /* initialize rwlocks */
4270         rwlock_init(&bond->lock);
4271         rwlock_init(&bond->curr_slave_lock);
4272
4273         bond->params = *params; /* copy params struct */
4274
4275         /* Initialize pointers */
4276         bond->first_slave = NULL;
4277         bond->curr_active_slave = NULL;
4278         bond->current_arp_slave = NULL;
4279         bond->primary_slave = NULL;
4280         bond->dev = bond_dev;
4281         INIT_LIST_HEAD(&bond->vlan_list);
4282
4283         /* Initialize the device entry points */
4284         bond_dev->open = bond_open;
4285         bond_dev->stop = bond_close;
4286         bond_dev->get_stats = bond_get_stats;
4287         bond_dev->do_ioctl = bond_do_ioctl;
4288         bond_dev->ethtool_ops = &bond_ethtool_ops;
4289         bond_dev->set_multicast_list = bond_set_multicast_list;
4290         bond_dev->change_mtu = bond_change_mtu;
4291         bond_dev->set_mac_address = bond_set_mac_address;
4292
4293         bond_set_mode_ops(bond, bond->params.mode);
4294
4295         bond_dev->destructor = free_netdev;
4296
4297         /* Initialize the device options */
4298         bond_dev->tx_queue_len = 0;
4299         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4300         bond_dev->priv_flags |= IFF_BONDING;
4301
4302         /* At first, we block adding VLANs. That's the only way to
4303          * prevent problems that occur when adding VLANs over an
4304          * empty bond. The block will be removed once non-challenged
4305          * slaves are enslaved.
4306          */
4307         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4308
4309         /* don't acquire bond device's netif_tx_lock when
4310          * transmitting */
4311         bond_dev->features |= NETIF_F_LLTX;
4312
4313         /* By default, we declare the bond to be fully
4314          * VLAN hardware accelerated capable. Special
4315          * care is taken in the various xmit functions
4316          * when there are slaves that are not hw accel
4317          * capable
4318          */
4319         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4320         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4321         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4322         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4323                                NETIF_F_HW_VLAN_RX |
4324                                NETIF_F_HW_VLAN_FILTER);
4325
4326 #ifdef CONFIG_PROC_FS
4327         bond_create_proc_entry(bond);
4328 #endif
4329
4330         /* set do_set_mac_addr to true on startup */
4331         bond->do_set_mac_addr = 1;
4332
4333         list_add_tail(&bond->bond_list, &bond_dev_list);
4334
4335         return 0;
4336 }
4337
4338 /* De-initialize device specific data.
4339  * Caller must hold rtnl_lock.
4340  */
4341 void bond_deinit(struct net_device *bond_dev)
4342 {
4343         struct bonding *bond = bond_dev->priv;
4344
4345         list_del(&bond->bond_list);
4346
4347 #ifdef CONFIG_PROC_FS
4348         bond_remove_proc_entry(bond);
4349 #endif
4350 }
4351
4352 /* Unregister and free all bond devices.
4353  * Caller must hold rtnl_lock.
4354  */
4355 static void bond_free_all(void)
4356 {
4357         struct bonding *bond, *nxt;
4358
4359         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4360                 struct net_device *bond_dev = bond->dev;
4361
4362                 bond_mc_list_destroy(bond);
4363                 /* Release the bonded slaves */
4364                 bond_release_all(bond_dev);
4365                 bond_deinit(bond_dev);
4366                 unregister_netdevice(bond_dev);
4367         }
4368
4369 #ifdef CONFIG_PROC_FS
4370         bond_destroy_proc_dir();
4371 #endif
4372 }
4373
4374 /*------------------------- Module initialization ---------------------------*/
4375
4376 /*
4377  * Convert string input module parms.  Accept either the
4378  * number of the mode or its string name.
4379  */
4380 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4381 {
4382         int i;
4383
4384         for (i = 0; tbl[i].modename; i++) {
4385                 if ((isdigit(*mode_arg) &&
4386                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4387                     (strncmp(mode_arg, tbl[i].modename,
4388                              strlen(tbl[i].modename)) == 0)) {
4389                         return tbl[i].mode;
4390                 }
4391         }
4392
4393         return -1;
4394 }
4395
4396 static int bond_check_params(struct bond_params *params)
4397 {
4398         int arp_validate_value;
4399
4400         /*
4401          * Convert string parameters.
4402          */
4403         if (mode) {
4404                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4405                 if (bond_mode == -1) {
4406                         printk(KERN_ERR DRV_NAME
4407                                ": Error: Invalid bonding mode \"%s\"\n",
4408                                mode == NULL ? "NULL" : mode);
4409                         return -EINVAL;
4410                 }
4411         }
4412
4413         if (xmit_hash_policy) {
4414                 if ((bond_mode != BOND_MODE_XOR) &&
4415                     (bond_mode != BOND_MODE_8023AD)) {
4416                         printk(KERN_INFO DRV_NAME
4417                                ": xor_mode param is irrelevant in mode %s\n",
4418                                bond_mode_name(bond_mode));
4419                 } else {
4420                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4421                                                         xmit_hashtype_tbl);
4422                         if (xmit_hashtype == -1) {
4423                                 printk(KERN_ERR DRV_NAME
4424                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4425                                 xmit_hash_policy == NULL ? "NULL" :
4426                                        xmit_hash_policy);
4427                                 return -EINVAL;
4428                         }
4429                 }
4430         }
4431
4432         if (lacp_rate) {
4433                 if (bond_mode != BOND_MODE_8023AD) {
4434                         printk(KERN_INFO DRV_NAME
4435                                ": lacp_rate param is irrelevant in mode %s\n",
4436                                bond_mode_name(bond_mode));
4437                 } else {
4438                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4439                         if (lacp_fast == -1) {
4440                                 printk(KERN_ERR DRV_NAME
4441                                        ": Error: Invalid lacp rate \"%s\"\n",
4442                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4443                                 return -EINVAL;
4444                         }
4445                 }
4446         }
4447
4448         if (max_bonds < 1 || max_bonds > INT_MAX) {
4449                 printk(KERN_WARNING DRV_NAME
4450                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4451                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4452                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4453                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4454         }
4455
4456         if (miimon < 0) {
4457                 printk(KERN_WARNING DRV_NAME
4458                        ": Warning: miimon module parameter (%d), "
4459                        "not in range 0-%d, so it was reset to %d\n",
4460                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4461                 miimon = BOND_LINK_MON_INTERV;
4462         }
4463
4464         if (updelay < 0) {
4465                 printk(KERN_WARNING DRV_NAME
4466                        ": Warning: updelay module parameter (%d), "
4467                        "not in range 0-%d, so it was reset to 0\n",
4468                        updelay, INT_MAX);
4469                 updelay = 0;
4470         }
4471
4472         if (downdelay < 0) {
4473                 printk(KERN_WARNING DRV_NAME
4474                        ": Warning: downdelay module parameter (%d), "
4475                        "not in range 0-%d, so it was reset to 0\n",
4476                        downdelay, INT_MAX);
4477                 downdelay = 0;
4478         }
4479
4480         if ((use_carrier != 0) && (use_carrier != 1)) {
4481                 printk(KERN_WARNING DRV_NAME
4482                        ": Warning: use_carrier module parameter (%d), "
4483                        "not of valid value (0/1), so it was set to 1\n",
4484                        use_carrier);
4485                 use_carrier = 1;
4486         }
4487
4488         /* reset values for 802.3ad */
4489         if (bond_mode == BOND_MODE_8023AD) {
4490                 if (!miimon) {
4491                         printk(KERN_WARNING DRV_NAME
4492                                ": Warning: miimon must be specified, "
4493                                "otherwise bonding will not detect link "
4494                                "failure, speed and duplex which are "
4495                                "essential for 802.3ad operation\n");
4496                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4497                         miimon = 100;
4498                 }
4499         }
4500
4501         /* reset values for TLB/ALB */
4502         if ((bond_mode == BOND_MODE_TLB) ||
4503             (bond_mode == BOND_MODE_ALB)) {
4504                 if (!miimon) {
4505                         printk(KERN_WARNING DRV_NAME
4506                                ": Warning: miimon must be specified, "
4507                                "otherwise bonding will not detect link "
4508                                "failure and link speed which are essential "
4509                                "for TLB/ALB load balancing\n");
4510                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4511                         miimon = 100;
4512                 }
4513         }
4514
4515         if (bond_mode == BOND_MODE_ALB) {
4516                 printk(KERN_NOTICE DRV_NAME
4517                        ": In ALB mode you might experience client "
4518                        "disconnections upon reconnection of a link if the "
4519                        "bonding module updelay parameter (%d msec) is "
4520                        "incompatible with the forwarding delay time of the "
4521                        "switch\n",
4522                        updelay);
4523         }
4524
4525         if (!miimon) {
4526                 if (updelay || downdelay) {
4527                         /* just warn the user the up/down delay will have
4528                          * no effect since miimon is zero...
4529                          */
4530                         printk(KERN_WARNING DRV_NAME
4531                                ": Warning: miimon module parameter not set "
4532                                "and updelay (%d) or downdelay (%d) module "
4533                                "parameter is set; updelay and downdelay have "
4534                                "no effect unless miimon is set\n",
4535                                updelay, downdelay);
4536                 }
4537         } else {
4538                 /* don't allow arp monitoring */
4539                 if (arp_interval) {
4540                         printk(KERN_WARNING DRV_NAME
4541                                ": Warning: miimon (%d) and arp_interval (%d) "
4542                                "can't be used simultaneously, disabling ARP "
4543                                "monitoring\n",
4544                                miimon, arp_interval);
4545                         arp_interval = 0;
4546                 }
4547
4548                 if ((updelay % miimon) != 0) {
4549                         printk(KERN_WARNING DRV_NAME
4550                                ": Warning: updelay (%d) is not a multiple "
4551                                "of miimon (%d), updelay rounded to %d ms\n",
4552                                updelay, miimon, (updelay / miimon) * miimon);
4553                 }
4554
4555                 updelay /= miimon;
4556
4557                 if ((downdelay % miimon) != 0) {
4558                         printk(KERN_WARNING DRV_NAME
4559                                ": Warning: downdelay (%d) is not a multiple "
4560                                "of miimon (%d), downdelay rounded to %d ms\n",
4561                                downdelay, miimon,
4562                                (downdelay / miimon) * miimon);
4563                 }
4564
4565                 downdelay /= miimon;
4566         }
4567
4568         if (arp_interval < 0) {
4569                 printk(KERN_WARNING DRV_NAME
4570                        ": Warning: arp_interval module parameter (%d) "
4571                        ", not in range 0-%d, so it was reset to %d\n",
4572                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4573                 arp_interval = BOND_LINK_ARP_INTERV;
4574         }
4575
4576         for (arp_ip_count = 0;
4577              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4578              arp_ip_count++) {
4579                 /* not complete check, but should be good enough to
4580                    catch mistakes */
4581                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4582                         printk(KERN_WARNING DRV_NAME
4583                                ": Warning: bad arp_ip_target module parameter "
4584                                "(%s), ARP monitoring will not be performed\n",
4585                                arp_ip_target[arp_ip_count]);
4586                         arp_interval = 0;
4587                 } else {
4588                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4589                         arp_target[arp_ip_count] = ip;
4590                 }
4591         }
4592
4593         if (arp_interval && !arp_ip_count) {
4594                 /* don't allow arping if no arp_ip_target given... */
4595                 printk(KERN_WARNING DRV_NAME
4596                        ": Warning: arp_interval module parameter (%d) "
4597                        "specified without providing an arp_ip_target "
4598                        "parameter, arp_interval was reset to 0\n",
4599                        arp_interval);
4600                 arp_interval = 0;
4601         }
4602
4603         if (arp_validate) {
4604                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4605                         printk(KERN_ERR DRV_NAME
4606                ": arp_validate only supported in active-backup mode\n");
4607                         return -EINVAL;
4608                 }
4609                 if (!arp_interval) {
4610                         printk(KERN_ERR DRV_NAME
4611                                ": arp_validate requires arp_interval\n");
4612                         return -EINVAL;
4613                 }
4614
4615                 arp_validate_value = bond_parse_parm(arp_validate,
4616                                                      arp_validate_tbl);
4617                 if (arp_validate_value == -1) {
4618                         printk(KERN_ERR DRV_NAME
4619                                ": Error: invalid arp_validate \"%s\"\n",
4620                                arp_validate == NULL ? "NULL" : arp_validate);
4621                         return -EINVAL;
4622                 }
4623         } else
4624                 arp_validate_value = 0;
4625
4626         if (miimon) {
4627                 printk(KERN_INFO DRV_NAME
4628                        ": MII link monitoring set to %d ms\n",
4629                        miimon);
4630         } else if (arp_interval) {
4631                 int i;
4632
4633                 printk(KERN_INFO DRV_NAME
4634                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4635                        arp_interval,
4636                        arp_validate_tbl[arp_validate_value].modename,
4637                        arp_ip_count);
4638
4639                 for (i = 0; i < arp_ip_count; i++)
4640                         printk (" %s", arp_ip_target[i]);
4641
4642                 printk("\n");
4643
4644         } else {
4645                 /* miimon and arp_interval not set, we need one so things
4646                  * work as expected, see bonding.txt for details
4647                  */
4648                 printk(KERN_WARNING DRV_NAME
4649                        ": Warning: either miimon or arp_interval and "
4650                        "arp_ip_target module parameters must be specified, "
4651                        "otherwise bonding will not detect link failures! see "
4652                        "bonding.txt for details.\n");
4653         }
4654
4655         if (primary && !USES_PRIMARY(bond_mode)) {
4656                 /* currently, using a primary only makes sense
4657                  * in active backup, TLB or ALB modes
4658                  */
4659                 printk(KERN_WARNING DRV_NAME
4660                        ": Warning: %s primary device specified but has no "
4661                        "effect in %s mode\n",
4662                        primary, bond_mode_name(bond_mode));
4663                 primary = NULL;
4664         }
4665
4666         /* fill params struct with the proper values */
4667         params->mode = bond_mode;
4668         params->xmit_policy = xmit_hashtype;
4669         params->miimon = miimon;
4670         params->arp_interval = arp_interval;
4671         params->arp_validate = arp_validate_value;
4672         params->updelay = updelay;
4673         params->downdelay = downdelay;
4674         params->use_carrier = use_carrier;
4675         params->lacp_fast = lacp_fast;
4676         params->primary[0] = 0;
4677
4678         if (primary) {
4679                 strncpy(params->primary, primary, IFNAMSIZ);
4680                 params->primary[IFNAMSIZ - 1] = 0;
4681         }
4682
4683         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4684
4685         return 0;
4686 }
4687
4688 static struct lock_class_key bonding_netdev_xmit_lock_key;
4689
4690 /* Create a new bond based on the specified name and bonding parameters.
4691  * If name is NULL, obtain a suitable "bond%d" name for us.
4692  * Caller must NOT hold rtnl_lock; we need to release it here before we
4693  * set up our sysfs entries.
4694  */
4695 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4696 {
4697         struct net_device *bond_dev;
4698         int res;
4699
4700         rtnl_lock();
4701         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4702                                 ether_setup);
4703         if (!bond_dev) {
4704                 printk(KERN_ERR DRV_NAME
4705                        ": %s: eek! can't alloc netdev!\n",
4706                        name);
4707                 res = -ENOMEM;
4708                 goto out_rtnl;
4709         }
4710
4711         if (!name) {
4712                 res = dev_alloc_name(bond_dev, "bond%d");
4713                 if (res < 0)
4714                         goto out_netdev;
4715         }
4716
4717         /* bond_init() must be called after dev_alloc_name() (for the
4718          * /proc files), but before register_netdevice(), because we
4719          * need to set function pointers.
4720          */
4721
4722         res = bond_init(bond_dev, params);
4723         if (res < 0) {
4724                 goto out_netdev;
4725         }
4726
4727         res = register_netdevice(bond_dev);
4728         if (res < 0) {
4729                 goto out_bond;
4730         }
4731
4732         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4733
4734         if (newbond)
4735                 *newbond = bond_dev->priv;
4736
4737         netif_carrier_off(bond_dev);
4738
4739         rtnl_unlock(); /* allows sysfs registration of net device */
4740         res = bond_create_sysfs_entry(bond_dev->priv);
4741         if (res < 0) {
4742                 rtnl_lock();
4743                 goto out_bond;
4744         }
4745
4746         return 0;
4747
4748 out_bond:
4749         bond_deinit(bond_dev);
4750 out_netdev:
4751         free_netdev(bond_dev);
4752 out_rtnl:
4753         rtnl_unlock();
4754         return res;
4755 }
4756
4757 static int __init bonding_init(void)
4758 {
4759         int i;
4760         int res;
4761
4762         printk(KERN_INFO "%s", version);
4763
4764         res = bond_check_params(&bonding_defaults);
4765         if (res) {
4766                 goto out;
4767         }
4768
4769 #ifdef CONFIG_PROC_FS
4770         bond_create_proc_dir();
4771 #endif
4772         for (i = 0; i < max_bonds; i++) {
4773                 res = bond_create(NULL, &bonding_defaults, NULL);
4774                 if (res)
4775                         goto err;
4776         }
4777
4778         res = bond_create_sysfs();
4779         if (res)
4780                 goto err;
4781
4782         register_netdevice_notifier(&bond_netdev_notifier);
4783         register_inetaddr_notifier(&bond_inetaddr_notifier);
4784
4785         goto out;
4786 err:
4787         rtnl_lock();
4788         bond_free_all();
4789         bond_destroy_sysfs();
4790         rtnl_unlock();
4791 out:
4792         return res;
4793
4794 }
4795
4796 static void __exit bonding_exit(void)
4797 {
4798         unregister_netdevice_notifier(&bond_netdev_notifier);
4799         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4800
4801         rtnl_lock();
4802         bond_free_all();
4803         bond_destroy_sysfs();
4804         rtnl_unlock();
4805 }
4806
4807 module_init(bonding_init);
4808 module_exit(bonding_exit);
4809 MODULE_LICENSE("GPL");
4810 MODULE_VERSION(DRV_VERSION);
4811 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4812 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4813 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4814
4815 /*
4816  * Local variables:
4817  *  c-indent-level: 8
4818  *  c-basic-offset: 8
4819  *  tab-width: 8
4820  * End:
4821  */
4822