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/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/seq_file.h>
  69 #include <linux/smp.h>
  70 #include <linux/if_ether.h>
  71 #include <net/arp.h>
  72 #include <linux/mii.h>
  73 #include <linux/ethtool.h>
  74 #include <linux/if_vlan.h>
  75 #include <linux/if_bonding.h>
  76 #include <net/route.h>
  77 #include "bonding.h"
  78 #include "bond_3ad.h"
  79 #include "bond_alb.h"
  80
  81 /*---------------------------- Module parameters ----------------------------*/
  82
  83 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  84 #define BOND_LINK_MON_INTERV    0
  85 #define BOND_LINK_ARP_INTERV    0
  86
  87 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  88 static int miimon       = BOND_LINK_MON_INTERV;
  89 static int updelay      = 0;
  90 static int downdelay    = 0;
  91 static int use_carrier  = 1;
  92 static char *mode       = NULL;
  93 static char *primary    = NULL;
  94 static char *lacp_rate  = NULL;
  95 static char *xmit_hash_policy = NULL;
  96 static int arp_interval = BOND_LINK_ARP_INTERV;
  97 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
  98 static char *arp_validate = NULL;
  99 struct bond_params bonding_defaults;
 100
 101 module_param(max_bonds, int, 0);
 102 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 103 module_param(miimon, int, 0);
 104 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 105 module_param(updelay, int, 0);
 106 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 107 module_param(downdelay, int, 0);
 108 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 109                             "in milliseconds");
 110 module_param(use_carrier, int, 0);
 111 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 112                               "0 for off, 1 for on (default)");
 113 module_param(mode, charp, 0);
 114 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 115                        "1 for active-backup, 2 for balance-xor, "
 116                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 117                        "6 for balance-alb");
 118 module_param(primary, charp, 0);
 119 MODULE_PARM_DESC(primary, "Primary network device to use");
 120 module_param(lacp_rate, charp, 0);
 121 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 122                             "(slow/fast)");
 123 module_param(xmit_hash_policy, charp, 0);
 124 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 125                                    ", 1 for layer 3+4");
 126 module_param(arp_interval, int, 0);
 127 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 128 module_param_array(arp_ip_target, charp, NULL, 0);
 129 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 130 module_param(arp_validate, charp, 0);
 131 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 132
 133 /*----------------------------- Global variables ----------------------------*/
 134
 135 static const char * const version =
 136         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 137
 138 LIST_HEAD(bond_dev_list);
 139
 140 #ifdef CONFIG_PROC_FS
 141 static struct proc_dir_entry *bond_proc_dir = NULL;
 142 #endif
 143
 144 extern struct rw_semaphore bonding_rwsem;
 145 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 146 static int arp_ip_count = 0;
 147 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 148 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 149 static int lacp_fast    = 0;
 150
 151
 152 struct bond_parm_tbl bond_lacp_tbl[] = {
 153 {       "slow",         AD_LACP_SLOW},
 154 {       "fast",         AD_LACP_FAST},
 155 {       NULL,           -1},
 156 };
 157
 158 struct bond_parm_tbl bond_mode_tbl[] = {
 159 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 160 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 161 {       "balance-xor",          BOND_MODE_XOR},
 162 {       "broadcast",            BOND_MODE_BROADCAST},
 163 {       "802.3ad",              BOND_MODE_8023AD},
 164 {       "balance-tlb",          BOND_MODE_TLB},
 165 {       "balance-alb",          BOND_MODE_ALB},
 166 {       NULL,                   -1},
 167 };
 168
 169 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 170 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 171 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 172 {       NULL,                   -1},
 173 };
 174
 175 struct bond_parm_tbl arp_validate_tbl[] = {
 176 {       "none",                 BOND_ARP_VALIDATE_NONE},
 177 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 178 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 179 {       "all",                  BOND_ARP_VALIDATE_ALL},
 180 {       NULL,                   -1},
 181 };
 182
 183 /*-------------------------- Forward declarations ---------------------------*/
 184
 185 static void bond_send_gratuitous_arp(struct bonding *bond);
 186
 187 /*---------------------------- General routines -----------------------------*/
 188
 189 const char *bond_mode_name(int mode)
 190 {
 191         switch (mode) {
 192         case BOND_MODE_ROUNDROBIN :
 193                 return "load balancing (round-robin)";
 194         case BOND_MODE_ACTIVEBACKUP :
 195                 return "fault-tolerance (active-backup)";
 196         case BOND_MODE_XOR :
 197                 return "load balancing (xor)";
 198         case BOND_MODE_BROADCAST :
 199                 return "fault-tolerance (broadcast)";
 200         case BOND_MODE_8023AD:
 201                 return "IEEE 802.3ad Dynamic link aggregation";
 202         case BOND_MODE_TLB:
 203                 return "transmit load balancing";
 204         case BOND_MODE_ALB:
 205                 return "adaptive load balancing";
 206         default:
 207                 return "unknown";
 208         }
 209 }
 210
 211 /*---------------------------------- VLAN -----------------------------------*/
 212
 213 /**
 214  * bond_add_vlan - add a new vlan id on bond
 215  * @bond: bond that got the notification
 216  * @vlan_id: the vlan id to add
 217  *
 218  * Returns -ENOMEM if allocation failed.
 219  */
 220 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 221 {
 222         struct vlan_entry *vlan;
 223
 224         dprintk("bond: %s, vlan id %d\n",
 225                 (bond ? bond->dev->name: "None"), vlan_id);
 226
 227         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 228         if (!vlan) {
 229                 return -ENOMEM;
 230         }
 231
 232         INIT_LIST_HEAD(&vlan->vlan_list);
 233         vlan->vlan_id = vlan_id;
 234         vlan->vlan_ip = 0;
 235
 236         write_lock_bh(&bond->lock);
 237
 238         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 239
 240         write_unlock_bh(&bond->lock);
 241
 242         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 243
 244         return 0;
 245 }
 246
 247 /**
 248  * bond_del_vlan - delete a vlan id from bond
 249  * @bond: bond that got the notification
 250  * @vlan_id: the vlan id to delete
 251  *
 252  * returns -ENODEV if @vlan_id was not found in @bond.
 253  */
 254 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 255 {
 256         struct vlan_entry *vlan, *next;
 257         int res = -ENODEV;
 258
 259         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 260
 261         write_lock_bh(&bond->lock);
 262
 263         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 264                 if (vlan->vlan_id == vlan_id) {
 265                         list_del(&vlan->vlan_list);
 266
 267                         if ((bond->params.mode == BOND_MODE_TLB) ||
 268                             (bond->params.mode == BOND_MODE_ALB)) {
 269                                 bond_alb_clear_vlan(bond, vlan_id);
 270                         }
 271
 272                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 273                                 bond->dev->name);
 274
 275                         kfree(vlan);
 276
 277                         if (list_empty(&bond->vlan_list) &&
 278                             (bond->slave_cnt == 0)) {
 279                                 /* Last VLAN removed and no slaves, so
 280                                  * restore block on adding VLANs. This will
 281                                  * be removed once new slaves that are not
 282                                  * VLAN challenged will be added.
 283                                  */
 284                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 285                         }
 286
 287                         res = 0;
 288                         goto out;
 289                 }
 290         }
 291
 292         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 293                 bond->dev->name);
 294
 295 out:
 296         write_unlock_bh(&bond->lock);
 297         return res;
 298 }
 299
 300 /**
 301  * bond_has_challenged_slaves
 302  * @bond: the bond we're working on
 303  *
 304  * Searches the slave list. Returns 1 if a vlan challenged slave
 305  * was found, 0 otherwise.
 306  *
 307  * Assumes bond->lock is held.
 308  */
 309 static int bond_has_challenged_slaves(struct bonding *bond)
 310 {
 311         struct slave *slave;
 312         int i;
 313
 314         bond_for_each_slave(bond, slave, i) {
 315                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 316                         dprintk("found VLAN challenged slave - %s\n",
 317                                 slave->dev->name);
 318                         return 1;
 319                 }
 320         }
 321
 322         dprintk("no VLAN challenged slaves found\n");
 323         return 0;
 324 }
 325
 326 /**
 327  * bond_next_vlan - safely skip to the next item in the vlans list.
 328  * @bond: the bond we're working on
 329  * @curr: item we're advancing from
 330  *
 331  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 332  * or @curr->next otherwise (even if it is @curr itself again).
 333  *
 334  * Caller must hold bond->lock
 335  */
 336 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 337 {
 338         struct vlan_entry *next, *last;
 339
 340         if (list_empty(&bond->vlan_list)) {
 341                 return NULL;
 342         }
 343
 344         if (!curr) {
 345                 next = list_entry(bond->vlan_list.next,
 346                                   struct vlan_entry, vlan_list);
 347         } else {
 348                 last = list_entry(bond->vlan_list.prev,
 349                                   struct vlan_entry, vlan_list);
 350                 if (last == curr) {
 351                         next = list_entry(bond->vlan_list.next,
 352                                           struct vlan_entry, vlan_list);
 353                 } else {
 354                         next = list_entry(curr->vlan_list.next,
 355                                           struct vlan_entry, vlan_list);
 356                 }
 357         }
 358
 359         return next;
 360 }
 361
 362 /**
 363  * bond_dev_queue_xmit - Prepare skb for xmit.
 364  *
 365  * @bond: bond device that got this skb for tx.
 366  * @skb: hw accel VLAN tagged skb to transmit
 367  * @slave_dev: slave that is supposed to xmit this skbuff
 368  *
 369  * When the bond gets an skb to transmit that is
 370  * already hardware accelerated VLAN tagged, and it
 371  * needs to relay this skb to a slave that is not
 372  * hw accel capable, the skb needs to be "unaccelerated",
 373  * i.e. strip the hwaccel tag and re-insert it as part
 374  * of the payload.
 375  */
 376 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 377 {
 378         unsigned short vlan_id;
 379
 380         if (!list_empty(&bond->vlan_list) &&
 381             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 382             vlan_get_tag(skb, &vlan_id) == 0) {
 383                 skb->dev = slave_dev;
 384                 skb = vlan_put_tag(skb, vlan_id);
 385                 if (!skb) {
 386                         /* vlan_put_tag() frees the skb in case of error,
 387                          * so return success here so the calling functions
 388                          * won't attempt to free is again.
 389                          */
 390                         return 0;
 391                 }
 392         } else {
 393                 skb->dev = slave_dev;
 394         }
 395
 396         skb->priority = 1;
 397         dev_queue_xmit(skb);
 398
 399         return 0;
 400 }
 401
 402 /*
 403  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 404  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 405  * lock because:
 406  * a. This operation is performed in IOCTL context,
 407  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 408  * c. Holding a lock with BH disabled while directly calling a base driver
 409  *    entry point is generally a BAD idea.
 410  *
 411  * The design of synchronization/protection for this operation in the 8021q
 412  * module is good for one or more VLAN devices over a single physical device
 413  * and cannot be extended for a teaming solution like bonding, so there is a
 414  * potential race condition here where a net device from the vlan group might
 415  * be referenced (either by a base driver or the 8021q code) while it is being
 416  * removed from the system. However, it turns out we're not making matters
 417  * worse, and if it works for regular VLAN usage it will work here too.
 418 */
 419
 420 /**
 421  * bond_vlan_rx_register - Propagates registration to slaves
 422  * @bond_dev: bonding net device that got called
 423  * @grp: vlan group being registered
 424  */
 425 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 426 {
 427         struct bonding *bond = bond_dev->priv;
 428         struct slave *slave;
 429         int i;
 430
 431         bond->vlgrp = grp;
 432
 433         bond_for_each_slave(bond, slave, i) {
 434                 struct net_device *slave_dev = slave->dev;
 435
 436                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 437                     slave_dev->vlan_rx_register) {
 438                         slave_dev->vlan_rx_register(slave_dev, grp);
 439                 }
 440         }
 441 }
 442
 443 /**
 444  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 445  * @bond_dev: bonding net device that got called
 446  * @vid: vlan id being added
 447  */
 448 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 449 {
 450         struct bonding *bond = bond_dev->priv;
 451         struct slave *slave;
 452         int i, res;
 453
 454         bond_for_each_slave(bond, slave, i) {
 455                 struct net_device *slave_dev = slave->dev;
 456
 457                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 458                     slave_dev->vlan_rx_add_vid) {
 459                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 460                 }
 461         }
 462
 463         res = bond_add_vlan(bond, vid);
 464         if (res) {
 465                 printk(KERN_ERR DRV_NAME
 466                        ": %s: Error: Failed to add vlan id %d\n",
 467                        bond_dev->name, vid);
 468         }
 469 }
 470
 471 /**
 472  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 473  * @bond_dev: bonding net device that got called
 474  * @vid: vlan id being removed
 475  */
 476 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 477 {
 478         struct bonding *bond = bond_dev->priv;
 479         struct slave *slave;
 480         struct net_device *vlan_dev;
 481         int i, res;
 482
 483         bond_for_each_slave(bond, slave, i) {
 484                 struct net_device *slave_dev = slave->dev;
 485
 486                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 487                     slave_dev->vlan_rx_kill_vid) {
 488                         /* Save and then restore vlan_dev in the grp array,
 489                          * since the slave's driver might clear it.
 490                          */
 491                         vlan_dev = bond->vlgrp->vlan_devices[vid];
 492                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 493                         bond->vlgrp->vlan_devices[vid] = vlan_dev;
 494                 }
 495         }
 496
 497         res = bond_del_vlan(bond, vid);
 498         if (res) {
 499                 printk(KERN_ERR DRV_NAME
 500                        ": %s: Error: Failed to remove vlan id %d\n",
 501                        bond_dev->name, vid);
 502         }
 503 }
 504
 505 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 506 {
 507         struct vlan_entry *vlan;
 508
 509         write_lock_bh(&bond->lock);
 510
 511         if (list_empty(&bond->vlan_list)) {
 512                 goto out;
 513         }
 514
 515         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 516             slave_dev->vlan_rx_register) {
 517                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 518         }
 519
 520         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 521             !(slave_dev->vlan_rx_add_vid)) {
 522                 goto out;
 523         }
 524
 525         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 526                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 527         }
 528
 529 out:
 530         write_unlock_bh(&bond->lock);
 531 }
 532
 533 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 534 {
 535         struct vlan_entry *vlan;
 536         struct net_device *vlan_dev;
 537
 538         write_lock_bh(&bond->lock);
 539
 540         if (list_empty(&bond->vlan_list)) {
 541                 goto out;
 542         }
 543
 544         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 545             !(slave_dev->vlan_rx_kill_vid)) {
 546                 goto unreg;
 547         }
 548
 549         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 550                 /* Save and then restore vlan_dev in the grp array,
 551                  * since the slave's driver might clear it.
 552                  */
 553                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
 554                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 555                 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
 556         }
 557
 558 unreg:
 559         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 560             slave_dev->vlan_rx_register) {
 561                 slave_dev->vlan_rx_register(slave_dev, NULL);
 562         }
 563
 564 out:
 565         write_unlock_bh(&bond->lock);
 566 }
 567
 568 /*------------------------------- Link status -------------------------------*/
 569
 570 /*
 571  * Set the carrier state for the master according to the state of its
 572  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 573  * do special 802.3ad magic.
 574  *
 575  * Returns zero if carrier state does not change, nonzero if it does.
 576  */
 577 static int bond_set_carrier(struct bonding *bond)
 578 {
 579         struct slave *slave;
 580         int i;
 581
 582         if (bond->slave_cnt == 0)
 583                 goto down;
 584
 585         if (bond->params.mode == BOND_MODE_8023AD)
 586                 return bond_3ad_set_carrier(bond);
 587
 588         bond_for_each_slave(bond, slave, i) {
 589                 if (slave->link == BOND_LINK_UP) {
 590                         if (!netif_carrier_ok(bond->dev)) {
 591                                 netif_carrier_on(bond->dev);
 592                                 return 1;
 593                         }
 594                         return 0;
 595                 }
 596         }
 597
 598 down:
 599         if (netif_carrier_ok(bond->dev)) {
 600                 netif_carrier_off(bond->dev);
 601                 return 1;
 602         }
 603         return 0;
 604 }
 605
 606 /*
 607  * Get link speed and duplex from the slave's base driver
 608  * using ethtool. If for some reason the call fails or the
 609  * values are invalid, fake speed and duplex to 100/Full
 610  * and return error.
 611  */
 612 static int bond_update_speed_duplex(struct slave *slave)
 613 {
 614         struct net_device *slave_dev = slave->dev;
 615         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 616         struct ifreq ifr;
 617         struct ethtool_cmd etool;
 618
 619         /* Fake speed and duplex */
 620         slave->speed = SPEED_100;
 621         slave->duplex = DUPLEX_FULL;
 622
 623         if (slave_dev->ethtool_ops) {
 624                 int res;
 625
 626                 if (!slave_dev->ethtool_ops->get_settings) {
 627                         return -1;
 628                 }
 629
 630                 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 631                 if (res < 0) {
 632                         return -1;
 633                 }
 634
 635                 goto verify;
 636         }
 637
 638         ioctl = slave_dev->do_ioctl;
 639         strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 640         etool.cmd = ETHTOOL_GSET;
 641         ifr.ifr_data = (char*)&etool;
 642         if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
 643                 return -1;
 644         }
 645
 646 verify:
 647         switch (etool.speed) {
 648         case SPEED_10:
 649         case SPEED_100:
 650         case SPEED_1000:
 651         case SPEED_10000:
 652                 break;
 653         default:
 654                 return -1;
 655         }
 656
 657         switch (etool.duplex) {
 658         case DUPLEX_FULL:
 659         case DUPLEX_HALF:
 660                 break;
 661         default:
 662                 return -1;
 663         }
 664
 665         slave->speed = etool.speed;
 666         slave->duplex = etool.duplex;
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * if <dev> supports MII link status reporting, check its link status.
 673  *
 674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 675  * depening upon the setting of the use_carrier parameter.
 676  *
 677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 678  * can't tell and just pretend it is), or 0, meaning that the link is
 679  * down.
 680  *
 681  * If reporting is non-zero, instead of faking link up, return -1 if
 682  * both ETHTOOL and MII ioctls fail (meaning the device does not
 683  * support them).  If use_carrier is set, return whatever it says.
 684  * It'd be nice if there was a good way to tell if a driver supports
 685  * netif_carrier, but there really isn't.
 686  */
 687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 688 {
 689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 690         struct ifreq ifr;
 691         struct mii_ioctl_data *mii;
 692         struct ethtool_value etool;
 693
 694         if (bond->params.use_carrier) {
 695                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 696         }
 697
 698         ioctl = slave_dev->do_ioctl;
 699         if (ioctl) {
 700                 /* TODO: set pointer to correct ioctl on a per team member */
 701                 /*       bases to make this more efficient. that is, once  */
 702                 /*       we determine the correct ioctl, we will always    */
 703                 /*       call it and not the others for that team          */
 704                 /*       member.                                           */
 705
 706                 /*
 707                  * We cannot assume that SIOCGMIIPHY will also read a
 708                  * register; not all network drivers (e.g., e100)
 709                  * support that.
 710                  */
 711
 712                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 713                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 714                 mii = if_mii(&ifr);
 715                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 716                         mii->reg_num = MII_BMSR;
 717                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 718                                 return (mii->val_out & BMSR_LSTATUS);
 719                         }
 720                 }
 721         }
 722
 723         /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
 724         /* for a period of time so we attempt to get link status   */
 725         /* from it last if the above MII ioctls fail...            */
 726         if (slave_dev->ethtool_ops) {
 727                 if (slave_dev->ethtool_ops->get_link) {
 728                         u32 link;
 729
 730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 731
 732                         return link ? BMSR_LSTATUS : 0;
 733                 }
 734         }
 735
 736         if (ioctl) {
 737                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 738                 etool.cmd = ETHTOOL_GLINK;
 739                 ifr.ifr_data = (char*)&etool;
 740                 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
 741                         if (etool.data == 1) {
 742                                 return BMSR_LSTATUS;
 743                         } else {
 744                                 dprintk("SIOCETHTOOL shows link down\n");
 745                                 return 0;
 746                         }
 747                 }
 748         }
 749
 750         /*
 751          * If reporting, report that either there's no dev->do_ioctl,
 752          * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
 753          * cannot report link status).  If not reporting, pretend
 754          * we're ok.
 755          */
 756         return (reporting ? -1 : BMSR_LSTATUS);
 757 }
 758
 759 /*----------------------------- Multicast list ------------------------------*/
 760
 761 /*
 762  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 763  */
 764 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 765 {
 766         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 767                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 768 }
 769
 770 /*
 771  * returns dmi entry if found, NULL otherwise
 772  */
 773 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 774 {
 775         struct dev_mc_list *idmi;
 776
 777         for (idmi = mc_list; idmi; idmi = idmi->next) {
 778                 if (bond_is_dmi_same(dmi, idmi)) {
 779                         return idmi;
 780                 }
 781         }
 782
 783         return NULL;
 784 }
 785
 786 /*
 787  * Push the promiscuity flag down to appropriate slaves
 788  */
 789 static void bond_set_promiscuity(struct bonding *bond, int inc)
 790 {
 791         if (USES_PRIMARY(bond->params.mode)) {
 792                 /* write lock already acquired */
 793                 if (bond->curr_active_slave) {
 794                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 795                 }
 796         } else {
 797                 struct slave *slave;
 798                 int i;
 799                 bond_for_each_slave(bond, slave, i) {
 800                         dev_set_promiscuity(slave->dev, inc);
 801                 }
 802         }
 803 }
 804
 805 /*
 806  * Push the allmulti flag down to all slaves
 807  */
 808 static void bond_set_allmulti(struct bonding *bond, int inc)
 809 {
 810         if (USES_PRIMARY(bond->params.mode)) {
 811                 /* write lock already acquired */
 812                 if (bond->curr_active_slave) {
 813                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 814                 }
 815         } else {
 816                 struct slave *slave;
 817                 int i;
 818                 bond_for_each_slave(bond, slave, i) {
 819                         dev_set_allmulti(slave->dev, inc);
 820                 }
 821         }
 822 }
 823
 824 /*
 825  * Add a Multicast address to slaves
 826  * according to mode
 827  */
 828 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 829 {
 830         if (USES_PRIMARY(bond->params.mode)) {
 831                 /* write lock already acquired */
 832                 if (bond->curr_active_slave) {
 833                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 834                 }
 835         } else {
 836                 struct slave *slave;
 837                 int i;
 838                 bond_for_each_slave(bond, slave, i) {
 839                         dev_mc_add(slave->dev, addr, alen, 0);
 840                 }
 841         }
 842 }
 843
 844 /*
 845  * Remove a multicast address from slave
 846  * according to mode
 847  */
 848 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 849 {
 850         if (USES_PRIMARY(bond->params.mode)) {
 851                 /* write lock already acquired */
 852                 if (bond->curr_active_slave) {
 853                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 854                 }
 855         } else {
 856                 struct slave *slave;
 857                 int i;
 858                 bond_for_each_slave(bond, slave, i) {
 859                         dev_mc_delete(slave->dev, addr, alen, 0);
 860                 }
 861         }
 862 }
 863
 864 /*
 865  * Totally destroys the mc_list in bond
 866  */
 867 static void bond_mc_list_destroy(struct bonding *bond)
 868 {
 869         struct dev_mc_list *dmi;
 870
 871         dmi = bond->mc_list;
 872         while (dmi) {
 873                 bond->mc_list = dmi->next;
 874                 kfree(dmi);
 875                 dmi = bond->mc_list;
 876         }
 877 }
 878
 879 /*
 880  * Copy all the Multicast addresses from src to the bonding device dst
 881  */
 882 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 883                              gfp_t gfp_flag)
 884 {
 885         struct dev_mc_list *dmi, *new_dmi;
 886
 887         for (dmi = mc_list; dmi; dmi = dmi->next) {
 888                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 889
 890                 if (!new_dmi) {
 891                         /* FIXME: Potential memory leak !!! */
 892                         return -ENOMEM;
 893                 }
 894
 895                 new_dmi->next = bond->mc_list;
 896                 bond->mc_list = new_dmi;
 897                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 898                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 899                 new_dmi->dmi_users = dmi->dmi_users;
 900                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 901         }
 902
 903         return 0;
 904 }
 905
 906 /*
 907  * flush all members of flush->mc_list from device dev->mc_list
 908  */
 909 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 910 {
 911         struct bonding *bond = bond_dev->priv;
 912         struct dev_mc_list *dmi;
 913
 914         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 915                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 916         }
 917
 918         if (bond->params.mode == BOND_MODE_8023AD) {
 919                 /* del lacpdu mc addr from mc list */
 920                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 921
 922                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 923         }
 924 }
 925
 926 /*--------------------------- Active slave change ---------------------------*/
 927
 928 /*
 929  * Update the mc list and multicast-related flags for the new and
 930  * old active slaves (if any) according to the multicast mode, and
 931  * promiscuous flags unconditionally.
 932  */
 933 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 934 {
 935         struct dev_mc_list *dmi;
 936
 937         if (!USES_PRIMARY(bond->params.mode)) {
 938                 /* nothing to do -  mc list is already up-to-date on
 939                  * all slaves
 940                  */
 941                 return;
 942         }
 943
 944         if (old_active) {
 945                 if (bond->dev->flags & IFF_PROMISC) {
 946                         dev_set_promiscuity(old_active->dev, -1);
 947                 }
 948
 949                 if (bond->dev->flags & IFF_ALLMULTI) {
 950                         dev_set_allmulti(old_active->dev, -1);
 951                 }
 952
 953                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 954                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 955                 }
 956         }
 957
 958         if (new_active) {
 959                 if (bond->dev->flags & IFF_PROMISC) {
 960                         dev_set_promiscuity(new_active->dev, 1);
 961                 }
 962
 963                 if (bond->dev->flags & IFF_ALLMULTI) {
 964                         dev_set_allmulti(new_active->dev, 1);
 965                 }
 966
 967                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 968                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 969                 }
 970         }
 971 }
 972
 973 /**
 974  * find_best_interface - select the best available slave to be the active one
 975  * @bond: our bonding struct
 976  *
 977  * Warning: Caller must hold curr_slave_lock for writing.
 978  */
 979 static struct slave *bond_find_best_slave(struct bonding *bond)
 980 {
 981         struct slave *new_active, *old_active;
 982         struct slave *bestslave = NULL;
 983         int mintime = bond->params.updelay;
 984         int i;
 985
 986         new_active = old_active = bond->curr_active_slave;
 987
 988         if (!new_active) { /* there were no active slaves left */
 989                 if (bond->slave_cnt > 0) {  /* found one slave */
 990                         new_active = bond->first_slave;
 991                 } else {
 992                         return NULL; /* still no slave, return NULL */
 993                 }
 994         }
 995
 996         /* first try the primary link; if arping, a link must tx/rx traffic
 997          * before it can be considered the curr_active_slave - also, we would skip
 998          * slaves between the curr_active_slave and primary_slave that may be up
 999          * and able to arp
1000          */
1001         if ((bond->primary_slave) &&
1002             (!bond->params.arp_interval) &&
1003             (IS_UP(bond->primary_slave->dev))) {
1004                 new_active = bond->primary_slave;
1005         }
1006
1007         /* remember where to stop iterating over the slaves */
1008         old_active = new_active;
1009
1010         bond_for_each_slave_from(bond, new_active, i, old_active) {
1011                 if (IS_UP(new_active->dev)) {
1012                         if (new_active->link == BOND_LINK_UP) {
1013                                 return new_active;
1014                         } else if (new_active->link == BOND_LINK_BACK) {
1015                                 /* link up, but waiting for stabilization */
1016                                 if (new_active->delay < mintime) {
1017                                         mintime = new_active->delay;
1018                                         bestslave = new_active;
1019                                 }
1020                         }
1021                 }
1022         }
1023
1024         return bestslave;
1025 }
1026
1027 /**
1028  * change_active_interface - change the active slave into the specified one
1029  * @bond: our bonding struct
1030  * @new: the new slave to make the active one
1031  *
1032  * Set the new slave to the bond's settings and unset them on the old
1033  * curr_active_slave.
1034  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035  *
1036  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037  * because it is apparently the best available slave we have, even though its
1038  * updelay hasn't timed out yet.
1039  *
1040  * Warning: Caller must hold curr_slave_lock for writing.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active) {
1047                 return;
1048         }
1049
1050         if (new_active) {
1051                 if (new_active->link == BOND_LINK_BACK) {
1052                         if (USES_PRIMARY(bond->params.mode)) {
1053                                 printk(KERN_INFO DRV_NAME
1054                                        ": %s: making interface %s the new "
1055                                        "active one %d ms earlier.\n",
1056                                        bond->dev->name, new_active->dev->name,
1057                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1058                         }
1059
1060                         new_active->delay = 0;
1061                         new_active->link = BOND_LINK_UP;
1062                         new_active->jiffies = jiffies;
1063
1064                         if (bond->params.mode == BOND_MODE_8023AD) {
1065                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1066                         }
1067
1068                         if ((bond->params.mode == BOND_MODE_TLB) ||
1069                             (bond->params.mode == BOND_MODE_ALB)) {
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                         }
1072                 } else {
1073                         if (USES_PRIMARY(bond->params.mode)) {
1074                                 printk(KERN_INFO DRV_NAME
1075                                        ": %s: making interface %s the new "
1076                                        "active one.\n",
1077                                        bond->dev->name, new_active->dev->name);
1078                         }
1079                 }
1080         }
1081
1082         if (USES_PRIMARY(bond->params.mode)) {
1083                 bond_mc_swap(bond, new_active, old_active);
1084         }
1085
1086         if ((bond->params.mode == BOND_MODE_TLB) ||
1087             (bond->params.mode == BOND_MODE_ALB)) {
1088                 bond_alb_handle_active_change(bond, new_active);
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091                 if (new_active)
1092                         bond_set_slave_active_flags(new_active);
1093         } else {
1094                 bond->curr_active_slave = new_active;
1095         }
1096
1097         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098                 if (old_active) {
1099                         bond_set_slave_inactive_flags(old_active);
1100                 }
1101
1102                 if (new_active) {
1103                         bond_set_slave_active_flags(new_active);
1104                 }
1105                 bond_send_gratuitous_arp(bond);
1106         }
1107 }
1108
1109 /**
1110  * bond_select_active_slave - select a new active slave, if needed
1111  * @bond: our bonding struct
1112  *
1113  * This functions shoud be called when one of the following occurs:
1114  * - The old curr_active_slave has been released or lost its link.
1115  * - The primary_slave has got its link back.
1116  * - A slave has got its link back and there's no old curr_active_slave.
1117  *
1118  * Warning: Caller must hold curr_slave_lock for writing.
1119  */
1120 void bond_select_active_slave(struct bonding *bond)
1121 {
1122         struct slave *best_slave;
1123         int rv;
1124
1125         best_slave = bond_find_best_slave(bond);
1126         if (best_slave != bond->curr_active_slave) {
1127                 bond_change_active_slave(bond, best_slave);
1128                 rv = bond_set_carrier(bond);
1129                 if (!rv)
1130                         return;
1131
1132                 if (netif_carrier_ok(bond->dev)) {
1133                         printk(KERN_INFO DRV_NAME
1134                                ": %s: first active interface up!\n",
1135                                bond->dev->name);
1136                 } else {
1137                         printk(KERN_INFO DRV_NAME ": %s: "
1138                                "now running without any active interface !\n",
1139                                bond->dev->name);
1140                 }
1141         }
1142 }
1143
1144 /*--------------------------- slave list handling ---------------------------*/
1145
1146 /*
1147  * This function attaches the slave to the end of list.
1148  *
1149  * bond->lock held for writing by caller.
1150  */
1151 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1152 {
1153         if (bond->first_slave == NULL) { /* attaching the first slave */
1154                 new_slave->next = new_slave;
1155                 new_slave->prev = new_slave;
1156                 bond->first_slave = new_slave;
1157         } else {
1158                 new_slave->next = bond->first_slave;
1159                 new_slave->prev = bond->first_slave->prev;
1160                 new_slave->next->prev = new_slave;
1161                 new_slave->prev->next = new_slave;
1162         }
1163
1164         bond->slave_cnt++;
1165 }
1166
1167 /*
1168  * This function detaches the slave from the list.
1169  * WARNING: no check is made to verify if the slave effectively
1170  * belongs to <bond>.
1171  * Nothing is freed on return, structures are just unchained.
1172  * If any slave pointer in bond was pointing to <slave>,
1173  * it should be changed by the calling function.
1174  *
1175  * bond->lock held for writing by caller.
1176  */
1177 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1178 {
1179         if (slave->next) {
1180                 slave->next->prev = slave->prev;
1181         }
1182
1183         if (slave->prev) {
1184                 slave->prev->next = slave->next;
1185         }
1186
1187         if (bond->first_slave == slave) { /* slave is the first slave */
1188                 if (bond->slave_cnt > 1) { /* there are more slave */
1189                         bond->first_slave = slave->next;
1190                 } else {
1191                         bond->first_slave = NULL; /* slave was the last one */
1192                 }
1193         }
1194
1195         slave->next = NULL;
1196         slave->prev = NULL;
1197         bond->slave_cnt--;
1198 }
1199
1200 /*---------------------------------- IOCTL ----------------------------------*/
1201
1202 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1203 {
1204         dprintk("bond_dev=%p\n", bond_dev);
1205         dprintk("slave_dev=%p\n", slave_dev);
1206         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1207         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1208         return 0;
1209 }
1210
1211 #define BOND_INTERSECT_FEATURES \
1212         (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1213
1214 /*
1215  * Compute the common dev->feature set available to all slaves.  Some
1216  * feature bits are managed elsewhere, so preserve feature bits set on
1217  * master device that are not part of the examined set.
1218  */
1219 static int bond_compute_features(struct bonding *bond)
1220 {
1221         unsigned long features = BOND_INTERSECT_FEATURES;
1222         struct slave *slave;
1223         struct net_device *bond_dev = bond->dev;
1224         unsigned short max_hard_header_len = ETH_HLEN;
1225         int i;
1226
1227         bond_for_each_slave(bond, slave, i) {
1228                 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1229                 if (slave->dev->hard_header_len > max_hard_header_len)
1230                         max_hard_header_len = slave->dev->hard_header_len;
1231         }
1232
1233         if ((features & NETIF_F_SG) &&
1234             !(features & NETIF_F_ALL_CSUM))
1235                 features &= ~NETIF_F_SG;
1236
1237         /*
1238          * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1239          * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1240          * implies that all slaves also support scatter-gather
1241          * (NETIF_F_SG), which implies that features also includes
1242          * NETIF_F_SG. So no need to check whether we have an
1243          * illegal combination of NETIF_F_{TSO,UFO} and
1244          * !NETIF_F_SG
1245          */
1246
1247         features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1248         bond_dev->features = features;
1249         bond_dev->hard_header_len = max_hard_header_len;
1250
1251         return 0;
1252 }
1253
1254 /* enslave device <slave> to bond device <master> */
1255 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1256 {
1257         struct bonding *bond = bond_dev->priv;
1258         struct slave *new_slave = NULL;
1259         struct dev_mc_list *dmi;
1260         struct sockaddr addr;
1261         int link_reporting;
1262         int old_features = bond_dev->features;
1263         int res = 0;
1264
1265         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1266                 slave_dev->do_ioctl == NULL) {
1267                 printk(KERN_WARNING DRV_NAME
1268                        ": %s: Warning: no link monitoring support for %s\n",
1269                        bond_dev->name, slave_dev->name);
1270         }
1271
1272         /* bond must be initialized by bond_open() before enslaving */
1273         if (!(bond_dev->flags & IFF_UP)) {
1274                 dprintk("Error, master_dev is not up\n");
1275                 return -EPERM;
1276         }
1277
1278         /* already enslaved */
1279         if (slave_dev->flags & IFF_SLAVE) {
1280                 dprintk("Error, Device was already enslaved\n");
1281                 return -EBUSY;
1282         }
1283
1284         /* vlan challenged mutual exclusion */
1285         /* no need to lock since we're protected by rtnl_lock */
1286         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1287                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1288                 if (!list_empty(&bond->vlan_list)) {
1289                         printk(KERN_ERR DRV_NAME
1290                                ": %s: Error: cannot enslave VLAN "
1291                                "challenged slave %s on VLAN enabled "
1292                                "bond %s\n", bond_dev->name, slave_dev->name,
1293                                bond_dev->name);
1294                         return -EPERM;
1295                 } else {
1296                         printk(KERN_WARNING DRV_NAME
1297                                ": %s: Warning: enslaved VLAN challenged "
1298                                "slave %s. Adding VLANs will be blocked as "
1299                                "long as %s is part of bond %s\n",
1300                                bond_dev->name, slave_dev->name, slave_dev->name,
1301                                bond_dev->name);
1302                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1303                 }
1304         } else {
1305                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1306                 if (bond->slave_cnt == 0) {
1307                         /* First slave, and it is not VLAN challenged,
1308                          * so remove the block of adding VLANs over the bond.
1309                          */
1310                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1311                 }
1312         }
1313
1314         /*
1315          * Old ifenslave binaries are no longer supported.  These can
1316          * be identified with moderate accurary by the state of the slave:
1317          * the current ifenslave will set the interface down prior to
1318          * enslaving it; the old ifenslave will not.
1319          */
1320         if ((slave_dev->flags & IFF_UP)) {
1321                 printk(KERN_ERR DRV_NAME ": %s is up. "
1322                        "This may be due to an out of date ifenslave.\n",
1323                        slave_dev->name);
1324                 res = -EPERM;
1325                 goto err_undo_flags;
1326         }
1327
1328         if (slave_dev->set_mac_address == NULL) {
1329                 printk(KERN_ERR DRV_NAME
1330                         ": %s: Error: The slave device you specified does "
1331                         "not support setting the MAC address. "
1332                         "Your kernel likely does not support slave "
1333                         "devices.\n", bond_dev->name);
1334                 res = -EOPNOTSUPP;
1335                 goto err_undo_flags;
1336         }
1337
1338         if (slave_dev->get_stats == NULL) {
1339                 printk(KERN_NOTICE DRV_NAME
1340                         ": %s: the driver for slave device %s does not provide "
1341                         "get_stats function, network statistics will be "
1342                         "inaccurate.\n", bond_dev->name, slave_dev->name);
1343         }
1344
1345         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1346         if (!new_slave) {
1347                 res = -ENOMEM;
1348                 goto err_undo_flags;
1349         }
1350
1351         /* save slave's original flags before calling
1352          * netdev_set_master and dev_open
1353          */
1354         new_slave->original_flags = slave_dev->flags;
1355
1356         /*
1357          * Save slave's original ("permanent") mac address for modes
1358          * that need it, and for restoring it upon release, and then
1359          * set it to the master's address
1360          */
1361         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1362
1363         /*
1364          * Set slave to master's mac address.  The application already
1365          * set the master's mac address to that of the first slave
1366          */
1367         memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1368         addr.sa_family = slave_dev->type;
1369         res = dev_set_mac_address(slave_dev, &addr);
1370         if (res) {
1371                 dprintk("Error %d calling set_mac_address\n", res);
1372                 goto err_free;
1373         }
1374
1375         /* open the slave since the application closed it */
1376         res = dev_open(slave_dev);
1377         if (res) {
1378                 dprintk("Openning slave %s failed\n", slave_dev->name);
1379                 goto err_restore_mac;
1380         }
1381
1382         res = netdev_set_master(slave_dev, bond_dev);
1383         if (res) {
1384                 dprintk("Error %d calling netdev_set_master\n", res);
1385                 goto err_close;
1386         }
1387
1388         new_slave->dev = slave_dev;
1389         slave_dev->priv_flags |= IFF_BONDING;
1390
1391         if ((bond->params.mode == BOND_MODE_TLB) ||
1392             (bond->params.mode == BOND_MODE_ALB)) {
1393                 /* bond_alb_init_slave() must be called before all other stages since
1394                  * it might fail and we do not want to have to undo everything
1395                  */
1396                 res = bond_alb_init_slave(bond, new_slave);
1397                 if (res) {
1398                         goto err_unset_master;
1399                 }
1400         }
1401
1402         /* If the mode USES_PRIMARY, then the new slave gets the
1403          * master's promisc (and mc) settings only if it becomes the
1404          * curr_active_slave, and that is taken care of later when calling
1405          * bond_change_active()
1406          */
1407         if (!USES_PRIMARY(bond->params.mode)) {
1408                 /* set promiscuity level to new slave */
1409                 if (bond_dev->flags & IFF_PROMISC) {
1410                         dev_set_promiscuity(slave_dev, 1);
1411                 }
1412
1413                 /* set allmulti level to new slave */
1414                 if (bond_dev->flags & IFF_ALLMULTI) {
1415                         dev_set_allmulti(slave_dev, 1);
1416                 }
1417
1418                 /* upload master's mc_list to new slave */
1419                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1420                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1421                 }
1422         }
1423
1424         if (bond->params.mode == BOND_MODE_8023AD) {
1425                 /* add lacpdu mc addr to mc list */
1426                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1427
1428                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1429         }
1430
1431         bond_add_vlans_on_slave(bond, slave_dev);
1432
1433         write_lock_bh(&bond->lock);
1434
1435         bond_attach_slave(bond, new_slave);
1436
1437         new_slave->delay = 0;
1438         new_slave->link_failure_count = 0;
1439
1440         bond_compute_features(bond);
1441
1442         new_slave->last_arp_rx = jiffies;
1443
1444         if (bond->params.miimon && !bond->params.use_carrier) {
1445                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1446
1447                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1448                         /*
1449                          * miimon is set but a bonded network driver
1450                          * does not support ETHTOOL/MII and
1451                          * arp_interval is not set.  Note: if
1452                          * use_carrier is enabled, we will never go
1453                          * here (because netif_carrier is always
1454                          * supported); thus, we don't need to change
1455                          * the messages for netif_carrier.
1456                          */
1457                         printk(KERN_WARNING DRV_NAME
1458                                ": %s: Warning: MII and ETHTOOL support not "
1459                                "available for interface %s, and "
1460                                "arp_interval/arp_ip_target module parameters "
1461                                "not specified, thus bonding will not detect "
1462                                "link failures! see bonding.txt for details.\n",
1463                                bond_dev->name, slave_dev->name);
1464                 } else if (link_reporting == -1) {
1465                         /* unable get link status using mii/ethtool */
1466                         printk(KERN_WARNING DRV_NAME
1467                                ": %s: Warning: can't get link status from "
1468                                "interface %s; the network driver associated "
1469                                "with this interface does not support MII or "
1470                                "ETHTOOL link status reporting, thus miimon "
1471                                "has no effect on this interface.\n",
1472                                bond_dev->name, slave_dev->name);
1473                 }
1474         }
1475
1476         /* check for initial state */
1477         if (!bond->params.miimon ||
1478             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1479                 if (bond->params.updelay) {
1480                         dprintk("Initial state of slave_dev is "
1481                                 "BOND_LINK_BACK\n");
1482                         new_slave->link  = BOND_LINK_BACK;
1483                         new_slave->delay = bond->params.updelay;
1484                 } else {
1485                         dprintk("Initial state of slave_dev is "
1486                                 "BOND_LINK_UP\n");
1487                         new_slave->link  = BOND_LINK_UP;
1488                 }
1489                 new_slave->jiffies = jiffies;
1490         } else {
1491                 dprintk("Initial state of slave_dev is "
1492                         "BOND_LINK_DOWN\n");
1493                 new_slave->link  = BOND_LINK_DOWN;
1494         }
1495
1496         if (bond_update_speed_duplex(new_slave) &&
1497             (new_slave->link != BOND_LINK_DOWN)) {
1498                 printk(KERN_WARNING DRV_NAME
1499                        ": %s: Warning: failed to get speed and duplex from %s, "
1500                        "assumed to be 100Mb/sec and Full.\n",
1501                        bond_dev->name, new_slave->dev->name);
1502
1503                 if (bond->params.mode == BOND_MODE_8023AD) {
1504                         printk(KERN_WARNING DRV_NAME
1505                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1506                                "support in base driver for proper aggregator "
1507                                "selection.\n", bond_dev->name);
1508                 }
1509         }
1510
1511         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1512                 /* if there is a primary slave, remember it */
1513                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1514                         bond->primary_slave = new_slave;
1515                 }
1516         }
1517
1518         switch (bond->params.mode) {
1519         case BOND_MODE_ACTIVEBACKUP:
1520                 bond_set_slave_inactive_flags(new_slave);
1521                 bond_select_active_slave(bond);
1522                 break;
1523         case BOND_MODE_8023AD:
1524                 /* in 802.3ad mode, the internal mechanism
1525                  * will activate the slaves in the selected
1526                  * aggregator
1527                  */
1528                 bond_set_slave_inactive_flags(new_slave);
1529                 /* if this is the first slave */
1530                 if (bond->slave_cnt == 1) {
1531                         SLAVE_AD_INFO(new_slave).id = 1;
1532                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1533                          * can be called only after the mac address of the bond is set
1534                          */
1535                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1536                                             bond->params.lacp_fast);
1537                 } else {
1538                         SLAVE_AD_INFO(new_slave).id =
1539                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1540                 }
1541
1542                 bond_3ad_bind_slave(new_slave);
1543                 break;
1544         case BOND_MODE_TLB:
1545         case BOND_MODE_ALB:
1546                 new_slave->state = BOND_STATE_ACTIVE;
1547                 if ((!bond->curr_active_slave) &&
1548                     (new_slave->link != BOND_LINK_DOWN)) {
1549                         /* first slave or no active slave yet, and this link
1550                          * is OK, so make this interface the active one
1551                          */
1552                         bond_change_active_slave(bond, new_slave);
1553                 } else {
1554                         bond_set_slave_inactive_flags(new_slave);
1555                 }
1556                 break;
1557         default:
1558                 dprintk("This slave is always active in trunk mode\n");
1559
1560                 /* always active in trunk mode */
1561                 new_slave->state = BOND_STATE_ACTIVE;
1562
1563                 /* In trunking mode there is little meaning to curr_active_slave
1564                  * anyway (it holds no special properties of the bond device),
1565                  * so we can change it without calling change_active_interface()
1566                  */
1567                 if (!bond->curr_active_slave) {
1568                         bond->curr_active_slave = new_slave;
1569                 }
1570                 break;
1571         } /* switch(bond_mode) */
1572
1573         bond_set_carrier(bond);
1574
1575         write_unlock_bh(&bond->lock);
1576
1577         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1578         if (res)
1579                 goto err_unset_master;
1580
1581         printk(KERN_INFO DRV_NAME
1582                ": %s: enslaving %s as a%s interface with a%s link.\n",
1583                bond_dev->name, slave_dev->name,
1584                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1585                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1586
1587         /* enslave is successful */
1588         return 0;
1589
1590 /* Undo stages on error */
1591 err_unset_master:
1592         netdev_set_master(slave_dev, NULL);
1593
1594 err_close:
1595         dev_close(slave_dev);
1596
1597 err_restore_mac:
1598         memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1599         addr.sa_family = slave_dev->type;
1600         dev_set_mac_address(slave_dev, &addr);
1601
1602 err_free:
1603         kfree(new_slave);
1604
1605 err_undo_flags:
1606         bond_dev->features = old_features;
1607
1608         return res;
1609 }
1610
1611 /*
1612  * Try to release the slave device <slave> from the bond device <master>
1613  * It is legal to access curr_active_slave without a lock because all the function
1614  * is write-locked.
1615  *
1616  * The rules for slave state should be:
1617  *   for Active/Backup:
1618  *     Active stays on all backups go down
1619  *   for Bonded connections:
1620  *     The first up interface should be left on and all others downed.
1621  */
1622 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1623 {
1624         struct bonding *bond = bond_dev->priv;
1625         struct slave *slave, *oldcurrent;
1626         struct sockaddr addr;
1627         int mac_addr_differ;
1628
1629         /* slave is not a slave or master is not master of this slave */
1630         if (!(slave_dev->flags & IFF_SLAVE) ||
1631             (slave_dev->master != bond_dev)) {
1632                 printk(KERN_ERR DRV_NAME
1633                        ": %s: Error: cannot release %s.\n",
1634                        bond_dev->name, slave_dev->name);
1635                 return -EINVAL;
1636         }
1637
1638         write_lock_bh(&bond->lock);
1639
1640         slave = bond_get_slave_by_dev(bond, slave_dev);
1641         if (!slave) {
1642                 /* not a slave of this bond */
1643                 printk(KERN_INFO DRV_NAME
1644                        ": %s: %s not enslaved\n",
1645                        bond_dev->name, slave_dev->name);
1646                 write_unlock_bh(&bond->lock);
1647                 return -EINVAL;
1648         }
1649
1650         mac_addr_differ = memcmp(bond_dev->dev_addr,
1651                                  slave->perm_hwaddr,
1652                                  ETH_ALEN);
1653         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1654                 printk(KERN_WARNING DRV_NAME
1655                        ": %s: Warning: the permanent HWaddr of %s "
1656                        "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1657                        "still in use by %s. Set the HWaddr of "
1658                        "%s to a different address to avoid "
1659                        "conflicts.\n",
1660                        bond_dev->name,
1661                        slave_dev->name,
1662                        slave->perm_hwaddr[0],
1663                        slave->perm_hwaddr[1],
1664                        slave->perm_hwaddr[2],
1665                        slave->perm_hwaddr[3],
1666                        slave->perm_hwaddr[4],
1667                        slave->perm_hwaddr[5],
1668                        bond_dev->name,
1669                        slave_dev->name);
1670         }
1671
1672         /* Inform AD package of unbinding of slave. */
1673         if (bond->params.mode == BOND_MODE_8023AD) {
1674                 /* must be called before the slave is
1675                  * detached from the list
1676                  */
1677                 bond_3ad_unbind_slave(slave);
1678         }
1679
1680         printk(KERN_INFO DRV_NAME
1681                ": %s: releasing %s interface %s\n",
1682                bond_dev->name,
1683                (slave->state == BOND_STATE_ACTIVE)
1684                ? "active" : "backup",
1685                slave_dev->name);
1686
1687         oldcurrent = bond->curr_active_slave;
1688
1689         bond->current_arp_slave = NULL;
1690
1691         /* release the slave from its bond */
1692         bond_detach_slave(bond, slave);
1693
1694         bond_compute_features(bond);
1695
1696         if (bond->primary_slave == slave) {
1697                 bond->primary_slave = NULL;
1698         }
1699
1700         if (oldcurrent == slave) {
1701                 bond_change_active_slave(bond, NULL);
1702         }
1703
1704         if ((bond->params.mode == BOND_MODE_TLB) ||
1705             (bond->params.mode == BOND_MODE_ALB)) {
1706                 /* Must be called only after the slave has been
1707                  * detached from the list and the curr_active_slave
1708                  * has been cleared (if our_slave == old_current),
1709                  * but before a new active slave is selected.
1710                  */
1711                 bond_alb_deinit_slave(bond, slave);
1712         }
1713
1714         if (oldcurrent == slave)
1715                 bond_select_active_slave(bond);
1716
1717         if (bond->slave_cnt == 0) {
1718                 bond_set_carrier(bond);
1719
1720                 /* if the last slave was removed, zero the mac address
1721                  * of the master so it will be set by the application
1722                  * to the mac address of the first slave
1723                  */
1724                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1725
1726                 if (list_empty(&bond->vlan_list)) {
1727                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1728                 } else {
1729                         printk(KERN_WARNING DRV_NAME
1730                                ": %s: Warning: clearing HW address of %s while it "
1731                                "still has VLANs.\n",
1732                                bond_dev->name, bond_dev->name);
1733                         printk(KERN_WARNING DRV_NAME
1734                                ": %s: When re-adding slaves, make sure the bond's "
1735                                "HW address matches its VLANs'.\n",
1736                                bond_dev->name);
1737                 }
1738         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1739                    !bond_has_challenged_slaves(bond)) {
1740                 printk(KERN_INFO DRV_NAME
1741                        ": %s: last VLAN challenged slave %s "
1742                        "left bond %s. VLAN blocking is removed\n",
1743                        bond_dev->name, slave_dev->name, bond_dev->name);
1744                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1745         }
1746
1747         write_unlock_bh(&bond->lock);
1748
1749         /* must do this from outside any spinlocks */
1750         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1751
1752         bond_del_vlans_from_slave(bond, slave_dev);
1753
1754         /* If the mode USES_PRIMARY, then we should only remove its
1755          * promisc and mc settings if it was the curr_active_slave, but that was
1756          * already taken care of above when we detached the slave
1757          */
1758         if (!USES_PRIMARY(bond->params.mode)) {
1759                 /* unset promiscuity level from slave */
1760                 if (bond_dev->flags & IFF_PROMISC) {
1761                         dev_set_promiscuity(slave_dev, -1);
1762                 }
1763
1764                 /* unset allmulti level from slave */
1765                 if (bond_dev->flags & IFF_ALLMULTI) {
1766                         dev_set_allmulti(slave_dev, -1);
1767                 }
1768
1769                 /* flush master's mc_list from slave */
1770                 bond_mc_list_flush(bond_dev, slave_dev);
1771         }
1772
1773         netdev_set_master(slave_dev, NULL);
1774
1775         /* close slave before restoring its mac address */
1776         dev_close(slave_dev);
1777
1778         /* restore original ("permanent") mac address */
1779         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1780         addr.sa_family = slave_dev->type;
1781         dev_set_mac_address(slave_dev, &addr);
1782
1783         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1784                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1785                                    IFF_SLAVE_NEEDARP);
1786
1787         kfree(slave);
1788
1789         return 0;  /* deletion OK */
1790 }
1791
1792 /*
1793  * This function releases all slaves.
1794  */
1795 static int bond_release_all(struct net_device *bond_dev)
1796 {
1797         struct bonding *bond = bond_dev->priv;
1798         struct slave *slave;
1799         struct net_device *slave_dev;
1800         struct sockaddr addr;
1801
1802         write_lock_bh(&bond->lock);
1803
1804         netif_carrier_off(bond_dev);
1805
1806         if (bond->slave_cnt == 0) {
1807                 goto out;
1808         }
1809
1810         bond->current_arp_slave = NULL;
1811         bond->primary_slave = NULL;
1812         bond_change_active_slave(bond, NULL);
1813
1814         while ((slave = bond->first_slave) != NULL) {
1815                 /* Inform AD package of unbinding of slave
1816                  * before slave is detached from the list.
1817                  */
1818                 if (bond->params.mode == BOND_MODE_8023AD) {
1819                         bond_3ad_unbind_slave(slave);
1820                 }
1821
1822                 slave_dev = slave->dev;
1823                 bond_detach_slave(bond, slave);
1824
1825                 if ((bond->params.mode == BOND_MODE_TLB) ||
1826                     (bond->params.mode == BOND_MODE_ALB)) {
1827                         /* must be called only after the slave
1828                          * has been detached from the list
1829                          */
1830                         bond_alb_deinit_slave(bond, slave);
1831                 }
1832
1833                 bond_compute_features(bond);
1834
1835                 /* now that the slave is detached, unlock and perform
1836                  * all the undo steps that should not be called from
1837                  * within a lock.
1838                  */
1839                 write_unlock_bh(&bond->lock);
1840
1841                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1842                 bond_del_vlans_from_slave(bond, slave_dev);
1843
1844                 /* If the mode USES_PRIMARY, then we should only remove its
1845                  * promisc and mc settings if it was the curr_active_slave, but that was
1846                  * already taken care of above when we detached the slave
1847                  */
1848                 if (!USES_PRIMARY(bond->params.mode)) {
1849                         /* unset promiscuity level from slave */
1850                         if (bond_dev->flags & IFF_PROMISC) {
1851                                 dev_set_promiscuity(slave_dev, -1);
1852                         }
1853
1854                         /* unset allmulti level from slave */
1855                         if (bond_dev->flags & IFF_ALLMULTI) {
1856                                 dev_set_allmulti(slave_dev, -1);
1857                         }
1858
1859                         /* flush master's mc_list from slave */
1860                         bond_mc_list_flush(bond_dev, slave_dev);
1861                 }
1862
1863                 netdev_set_master(slave_dev, NULL);
1864
1865                 /* close slave before restoring its mac address */
1866                 dev_close(slave_dev);
1867
1868                 /* restore original ("permanent") mac address*/
1869                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1870                 addr.sa_family = slave_dev->type;
1871                 dev_set_mac_address(slave_dev, &addr);
1872
1873                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1874                                            IFF_SLAVE_INACTIVE);
1875
1876                 kfree(slave);
1877
1878                 /* re-acquire the lock before getting the next slave */
1879                 write_lock_bh(&bond->lock);
1880         }
1881
1882         /* zero the mac address of the master so it will be
1883          * set by the application to the mac address of the
1884          * first slave
1885          */
1886         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1887
1888         if (list_empty(&bond->vlan_list)) {
1889                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1890         } else {
1891                 printk(KERN_WARNING DRV_NAME
1892                        ": %s: Warning: clearing HW address of %s while it "
1893                        "still has VLANs.\n",
1894                        bond_dev->name, bond_dev->name);
1895                 printk(KERN_WARNING DRV_NAME
1896                        ": %s: When re-adding slaves, make sure the bond's "
1897                        "HW address matches its VLANs'.\n",
1898                        bond_dev->name);
1899         }
1900
1901         printk(KERN_INFO DRV_NAME
1902                ": %s: released all slaves\n",
1903                bond_dev->name);
1904
1905 out:
1906         write_unlock_bh(&bond->lock);
1907
1908         return 0;
1909 }
1910
1911 /*
1912  * This function changes the active slave to slave <slave_dev>.
1913  * It returns -EINVAL in the following cases.
1914  *  - <slave_dev> is not found in the list.
1915  *  - There is not active slave now.
1916  *  - <slave_dev> is already active.
1917  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1918  *  - <slave_dev> is not running.
1919  * In these cases, this fuction does nothing.
1920  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1921  */
1922 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1923 {
1924         struct bonding *bond = bond_dev->priv;
1925         struct slave *old_active = NULL;
1926         struct slave *new_active = NULL;
1927         int res = 0;
1928
1929         if (!USES_PRIMARY(bond->params.mode)) {
1930                 return -EINVAL;
1931         }
1932
1933         /* Verify that master_dev is indeed the master of slave_dev */
1934         if (!(slave_dev->flags & IFF_SLAVE) ||
1935             (slave_dev->master != bond_dev)) {
1936                 return -EINVAL;
1937         }
1938
1939         write_lock_bh(&bond->lock);
1940
1941         old_active = bond->curr_active_slave;
1942         new_active = bond_get_slave_by_dev(bond, slave_dev);
1943
1944         /*
1945          * Changing to the current active: do nothing; return success.
1946          */
1947         if (new_active && (new_active == old_active)) {
1948                 write_unlock_bh(&bond->lock);
1949                 return 0;
1950         }
1951
1952         if ((new_active) &&
1953             (old_active) &&
1954             (new_active->link == BOND_LINK_UP) &&
1955             IS_UP(new_active->dev)) {
1956                 bond_change_active_slave(bond, new_active);
1957         } else {
1958                 res = -EINVAL;
1959         }
1960
1961         write_unlock_bh(&bond->lock);
1962
1963         return res;
1964 }
1965
1966 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1967 {
1968         struct bonding *bond = bond_dev->priv;
1969
1970         info->bond_mode = bond->params.mode;
1971         info->miimon = bond->params.miimon;
1972
1973         read_lock_bh(&bond->lock);
1974         info->num_slaves = bond->slave_cnt;
1975         read_unlock_bh(&bond->lock);
1976
1977         return 0;
1978 }
1979
1980 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1981 {
1982         struct bonding *bond = bond_dev->priv;
1983         struct slave *slave;
1984         int i, found = 0;
1985
1986         if (info->slave_id < 0) {
1987                 return -ENODEV;
1988         }
1989
1990         read_lock_bh(&bond->lock);
1991
1992         bond_for_each_slave(bond, slave, i) {
1993                 if (i == (int)info->slave_id) {
1994                         found = 1;
1995                         break;
1996                 }
1997         }
1998
1999         read_unlock_bh(&bond->lock);
2000
2001         if (found) {
2002                 strcpy(info->slave_name, slave->dev->name);
2003                 info->link = slave->link;
2004                 info->state = slave->state;
2005                 info->link_failure_count = slave->link_failure_count;
2006         } else {
2007                 return -ENODEV;
2008         }
2009
2010         return 0;
2011 }
2012
2013 /*-------------------------------- Monitoring -------------------------------*/
2014
2015 /* this function is called regularly to monitor each slave's link. */
2016 void bond_mii_monitor(struct net_device *bond_dev)
2017 {
2018         struct bonding *bond = bond_dev->priv;
2019         struct slave *slave, *oldcurrent;
2020         int do_failover = 0;
2021         int delta_in_ticks;
2022         int i;
2023
2024         read_lock(&bond->lock);
2025
2026         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2027
2028         if (bond->kill_timers) {
2029                 goto out;
2030         }
2031
2032         if (bond->slave_cnt == 0) {
2033                 goto re_arm;
2034         }
2035
2036         /* we will try to read the link status of each of our slaves, and
2037          * set their IFF_RUNNING flag appropriately. For each slave not
2038          * supporting MII status, we won't do anything so that a user-space
2039          * program could monitor the link itself if needed.
2040          */
2041
2042         read_lock(&bond->curr_slave_lock);
2043         oldcurrent = bond->curr_active_slave;
2044         read_unlock(&bond->curr_slave_lock);
2045
2046         bond_for_each_slave(bond, slave, i) {
2047                 struct net_device *slave_dev = slave->dev;
2048                 int link_state;
2049                 u16 old_speed = slave->speed;
2050                 u8 old_duplex = slave->duplex;
2051
2052                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2053
2054                 switch (slave->link) {
2055                 case BOND_LINK_UP:      /* the link was up */
2056                         if (link_state == BMSR_LSTATUS) {
2057                                 /* link stays up, nothing more to do */
2058                                 break;
2059                         } else { /* link going down */
2060                                 slave->link  = BOND_LINK_FAIL;
2061                                 slave->delay = bond->params.downdelay;
2062
2063                                 if (slave->link_failure_count < UINT_MAX) {
2064                                         slave->link_failure_count++;
2065                                 }
2066
2067                                 if (bond->params.downdelay) {
2068                                         printk(KERN_INFO DRV_NAME
2069                                                ": %s: link status down for %s "
2070                                                "interface %s, disabling it in "
2071                                                "%d ms.\n",
2072                                                bond_dev->name,
2073                                                IS_UP(slave_dev)
2074                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2075                                                   ? ((slave == oldcurrent)
2076                                                      ? "active " : "backup ")
2077                                                   : "")
2078                                                : "idle ",
2079                                                slave_dev->name,
2080                                                bond->params.downdelay * bond->params.miimon);
2081                                 }
2082                         }
2083                         /* no break ! fall through the BOND_LINK_FAIL test to
2084                            ensure proper action to be taken
2085                         */
2086                 case BOND_LINK_FAIL:    /* the link has just gone down */
2087                         if (link_state != BMSR_LSTATUS) {
2088                                 /* link stays down */
2089                                 if (slave->delay <= 0) {
2090                                         /* link down for too long time */
2091                                         slave->link = BOND_LINK_DOWN;
2092
2093                                         /* in active/backup mode, we must
2094                                          * completely disable this interface
2095                                          */
2096                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2097                                             (bond->params.mode == BOND_MODE_8023AD)) {
2098                                                 bond_set_slave_inactive_flags(slave);
2099                                         }
2100
2101                                         printk(KERN_INFO DRV_NAME
2102                                                ": %s: link status definitely "
2103                                                "down for interface %s, "
2104                                                "disabling it\n",
2105                                                bond_dev->name,
2106                                                slave_dev->name);
2107
2108                                         /* notify ad that the link status has changed */
2109                                         if (bond->params.mode == BOND_MODE_8023AD) {
2110                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2111                                         }
2112
2113                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2114                                             (bond->params.mode == BOND_MODE_ALB)) {
2115                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2116                                         }
2117
2118                                         if (slave == oldcurrent) {
2119                                                 do_failover = 1;
2120                                         }
2121                                 } else {
2122                                         slave->delay--;
2123                                 }
2124                         } else {
2125                                 /* link up again */
2126                                 slave->link  = BOND_LINK_UP;
2127                                 slave->jiffies = jiffies;
2128                                 printk(KERN_INFO DRV_NAME
2129                                        ": %s: link status up again after %d "
2130                                        "ms for interface %s.\n",
2131                                        bond_dev->name,
2132                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2133                                        slave_dev->name);
2134                         }
2135                         break;
2136                 case BOND_LINK_DOWN:    /* the link was down */
2137                         if (link_state != BMSR_LSTATUS) {
2138                                 /* the link stays down, nothing more to do */
2139                                 break;
2140                         } else {        /* link going up */
2141                                 slave->link  = BOND_LINK_BACK;
2142                                 slave->delay = bond->params.updelay;
2143
2144                                 if (bond->params.updelay) {
2145                                         /* if updelay == 0, no need to
2146                                            advertise about a 0 ms delay */
2147                                         printk(KERN_INFO DRV_NAME
2148                                                ": %s: link status up for "
2149                                                "interface %s, enabling it "
2150                                                "in %d ms.\n",
2151                                                bond_dev->name,
2152                                                slave_dev->name,
2153                                                bond->params.updelay * bond->params.miimon);
2154                                 }
2155                         }
2156                         /* no break ! fall through the BOND_LINK_BACK state in
2157                            case there's something to do.
2158                         */
2159                 case BOND_LINK_BACK:    /* the link has just come back */
2160                         if (link_state != BMSR_LSTATUS) {
2161                                 /* link down again */
2162                                 slave->link  = BOND_LINK_DOWN;
2163
2164                                 printk(KERN_INFO DRV_NAME
2165                                        ": %s: link status down again after %d "
2166                                        "ms for interface %s.\n",
2167                                        bond_dev->name,
2168                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2169                                        slave_dev->name);
2170                         } else {
2171                                 /* link stays up */
2172                                 if (slave->delay == 0) {
2173                                         /* now the link has been up for long time enough */
2174                                         slave->link = BOND_LINK_UP;
2175                                         slave->jiffies = jiffies;
2176
2177                                         if (bond->params.mode == BOND_MODE_8023AD) {
2178                                                 /* prevent it from being the active one */
2179                                                 slave->state = BOND_STATE_BACKUP;
2180                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2181                                                 /* make it immediately active */
2182                                                 slave->state = BOND_STATE_ACTIVE;
2183                                         } else if (slave != bond->primary_slave) {
2184                                                 /* prevent it from being the active one */
2185                                                 slave->state = BOND_STATE_BACKUP;
2186                                         }
2187
2188                                         printk(KERN_INFO DRV_NAME
2189                                                ": %s: link status definitely "
2190                                                "up for interface %s.\n",
2191                                                bond_dev->name,
2192                                                slave_dev->name);
2193
2194                                         /* notify ad that the link status has changed */
2195                                         if (bond->params.mode == BOND_MODE_8023AD) {
2196                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2197                                         }
2198
2199                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2200                                             (bond->params.mode == BOND_MODE_ALB)) {
2201                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2202                                         }
2203
2204                                         if ((!oldcurrent) ||
2205                                             (slave == bond->primary_slave)) {
2206                                                 do_failover = 1;
2207                                         }
2208                                 } else {
2209                                         slave->delay--;
2210                                 }
2211                         }
2212                         break;
2213                 default:
2214                         /* Should not happen */
2215                         printk(KERN_ERR DRV_NAME
2216                                ": %s: Error: %s Illegal value (link=%d)\n",
2217                                bond_dev->name,
2218                                slave->dev->name,
2219                                slave->link);
2220                         goto out;
2221                 } /* end of switch (slave->link) */
2222
2223                 bond_update_speed_duplex(slave);
2224
2225                 if (bond->params.mode == BOND_MODE_8023AD) {
2226                         if (old_speed != slave->speed) {
2227                                 bond_3ad_adapter_speed_changed(slave);
2228                         }
2229
2230                         if (old_duplex != slave->duplex) {
2231                                 bond_3ad_adapter_duplex_changed(slave);
2232                         }
2233                 }
2234
2235         } /* end of for */
2236
2237         if (do_failover) {
2238                 write_lock(&bond->curr_slave_lock);
2239
2240                 bond_select_active_slave(bond);
2241
2242                 write_unlock(&bond->curr_slave_lock);
2243         } else
2244                 bond_set_carrier(bond);
2245
2246 re_arm:
2247         if (bond->params.miimon) {
2248                 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2249         }
2250 out:
2251         read_unlock(&bond->lock);
2252 }
2253
2254
2255 static u32 bond_glean_dev_ip(struct net_device *dev)
2256 {
2257         struct in_device *idev;
2258         struct in_ifaddr *ifa;
2259         __be32 addr = 0;
2260
2261         if (!dev)
2262                 return 0;
2263
2264         rcu_read_lock();
2265         idev = __in_dev_get_rcu(dev);
2266         if (!idev)
2267                 goto out;
2268
2269         ifa = idev->ifa_list;
2270         if (!ifa)
2271                 goto out;
2272
2273         addr = ifa->ifa_local;
2274 out:
2275         rcu_read_unlock();
2276         return addr;
2277 }
2278
2279 static int bond_has_ip(struct bonding *bond)
2280 {
2281         struct vlan_entry *vlan, *vlan_next;
2282
2283         if (bond->master_ip)
2284                 return 1;
2285
2286         if (list_empty(&bond->vlan_list))
2287                 return 0;
2288
2289         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2290                                  vlan_list) {
2291                 if (vlan->vlan_ip)
2292                         return 1;
2293         }
2294
2295         return 0;
2296 }
2297
2298 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2299 {
2300         struct vlan_entry *vlan, *vlan_next;
2301
2302         if (ip == bond->master_ip)
2303                 return 1;
2304
2305         if (list_empty(&bond->vlan_list))
2306                 return 0;
2307
2308         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2309                                  vlan_list) {
2310                 if (ip == vlan->vlan_ip)
2311                         return 1;
2312         }
2313
2314         return 0;
2315 }
2316
2317 /*
2318  * We go to the (large) trouble of VLAN tagging ARP frames because
2319  * switches in VLAN mode (especially if ports are configured as
2320  * "native" to a VLAN) might not pass non-tagged frames.
2321  */
2322 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2323 {
2324         struct sk_buff *skb;
2325
2326         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2327                slave_dev->name, dest_ip, src_ip, vlan_id);
2328
2329         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2330                          NULL, slave_dev->dev_addr, NULL);
2331
2332         if (!skb) {
2333                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2334                 return;
2335         }
2336         if (vlan_id) {
2337                 skb = vlan_put_tag(skb, vlan_id);
2338                 if (!skb) {
2339                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2340                         return;
2341                 }
2342         }
2343         arp_xmit(skb);
2344 }
2345
2346
2347 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2348 {
2349         int i, vlan_id, rv;
2350         u32 *targets = bond->params.arp_targets;
2351         struct vlan_entry *vlan, *vlan_next;
2352         struct net_device *vlan_dev;
2353         struct flowi fl;
2354         struct rtable *rt;
2355
2356         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2357                 if (!targets[i])
2358                         continue;
2359                 dprintk("basa: target %x\n", targets[i]);
2360                 if (list_empty(&bond->vlan_list)) {
2361                         dprintk("basa: empty vlan: arp_send\n");
2362                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2363                                       bond->master_ip, 0);
2364                         continue;
2365                 }
2366
2367                 /*
2368                  * If VLANs are configured, we do a route lookup to
2369                  * determine which VLAN interface would be used, so we
2370                  * can tag the ARP with the proper VLAN tag.
2371                  */
2372                 memset(&fl, 0, sizeof(fl));
2373                 fl.fl4_dst = targets[i];
2374                 fl.fl4_tos = RTO_ONLINK;
2375
2376                 rv = ip_route_output_key(&rt, &fl);
2377                 if (rv) {
2378                         if (net_ratelimit()) {
2379                                 printk(KERN_WARNING DRV_NAME
2380                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2381                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2382                         }
2383                         continue;
2384                 }
2385
2386                 /*
2387                  * This target is not on a VLAN
2388                  */
2389                 if (rt->u.dst.dev == bond->dev) {
2390                         ip_rt_put(rt);
2391                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2392                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2393                                       bond->master_ip, 0);
2394                         continue;
2395                 }
2396
2397                 vlan_id = 0;
2398                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2399                                          vlan_list) {
2400                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2401                         if (vlan_dev == rt->u.dst.dev) {
2402                                 vlan_id = vlan->vlan_id;
2403                                 dprintk("basa: vlan match on %s %d\n",
2404                                        vlan_dev->name, vlan_id);
2405                                 break;
2406                         }
2407                 }
2408
2409                 if (vlan_id) {
2410                         ip_rt_put(rt);
2411                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2412                                       vlan->vlan_ip, vlan_id);
2413                         continue;
2414                 }
2415
2416                 if (net_ratelimit()) {
2417                         printk(KERN_WARNING DRV_NAME
2418                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2419                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2420                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2421                 }
2422                 ip_rt_put(rt);
2423         }
2424 }
2425
2426 /*
2427  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2428  * for each VLAN above us.
2429  */
2430 static void bond_send_gratuitous_arp(struct bonding *bond)
2431 {
2432         struct slave *slave = bond->curr_active_slave;
2433         struct vlan_entry *vlan;
2434         struct net_device *vlan_dev;
2435
2436         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2437                                 slave ? slave->dev->name : "NULL");
2438         if (!slave)
2439                 return;
2440
2441         if (bond->master_ip) {
2442                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2443                                   bond->master_ip, 0);
2444         }
2445
2446         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2447                 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2448                 if (vlan->vlan_ip) {
2449                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2450                                       vlan->vlan_ip, vlan->vlan_id);
2451                 }
2452         }
2453 }
2454
2455 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2456 {
2457         int i;
2458         u32 *targets = bond->params.arp_targets;
2459
2460         targets = bond->params.arp_targets;
2461         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2462                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2463                         "%u.%u.%u.%u bhti(tip) %d\n",
2464                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2465                        bond_has_this_ip(bond, tip));
2466                 if (sip == targets[i]) {
2467                         if (bond_has_this_ip(bond, tip))
2468                                 slave->last_arp_rx = jiffies;
2469                         return;
2470                 }
2471         }
2472 }
2473
2474 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2475 {
2476         struct arphdr *arp;
2477         struct slave *slave;
2478         struct bonding *bond;
2479         unsigned char *arp_ptr;
2480         u32 sip, tip;
2481
2482         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2483                 goto out;
2484
2485         bond = dev->priv;
2486         read_lock(&bond->lock);
2487
2488         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2489                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2490                 orig_dev ? orig_dev->name : "NULL");
2491
2492         slave = bond_get_slave_by_dev(bond, orig_dev);
2493         if (!slave || !slave_do_arp_validate(bond, slave))
2494                 goto out_unlock;
2495
2496         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2497         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2498                                  (2 * dev->addr_len) +
2499                                  (2 * sizeof(u32)))))
2500                 goto out_unlock;
2501
2502         arp = skb->nh.arph;
2503         if (arp->ar_hln != dev->addr_len ||
2504             skb->pkt_type == PACKET_OTHERHOST ||
2505             skb->pkt_type == PACKET_LOOPBACK ||
2506             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2507             arp->ar_pro != htons(ETH_P_IP) ||
2508             arp->ar_pln != 4)
2509                 goto out_unlock;
2510
2511         arp_ptr = (unsigned char *)(arp + 1);
2512         arp_ptr += dev->addr_len;
2513         memcpy(&sip, arp_ptr, 4);
2514         arp_ptr += 4 + dev->addr_len;
2515         memcpy(&tip, arp_ptr, 4);
2516
2517         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2518                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2519                 slave->state, bond->params.arp_validate,
2520                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2521
2522         /*
2523          * Backup slaves won't see the ARP reply, but do come through
2524          * here for each ARP probe (so we swap the sip/tip to validate
2525          * the probe).  In a "redundant switch, common router" type of
2526          * configuration, the ARP probe will (hopefully) travel from
2527          * the active, through one switch, the router, then the other
2528          * switch before reaching the backup.
2529          */
2530         if (slave->state == BOND_STATE_ACTIVE)
2531                 bond_validate_arp(bond, slave, sip, tip);
2532         else
2533                 bond_validate_arp(bond, slave, tip, sip);
2534
2535 out_unlock:
2536         read_unlock(&bond->lock);
2537 out:
2538         dev_kfree_skb(skb);
2539         return NET_RX_SUCCESS;
2540 }
2541
2542 /*
2543  * this function is called regularly to monitor each slave's link
2544  * ensuring that traffic is being sent and received when arp monitoring
2545  * is used in load-balancing mode. if the adapter has been dormant, then an
2546  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2547  * arp monitoring in active backup mode.
2548  */
2549 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2550 {
2551         struct bonding *bond = bond_dev->priv;
2552         struct slave *slave, *oldcurrent;
2553         int do_failover = 0;
2554         int delta_in_ticks;
2555         int i;
2556
2557         read_lock(&bond->lock);
2558
2559         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2560
2561         if (bond->kill_timers) {
2562                 goto out;
2563         }
2564
2565         if (bond->slave_cnt == 0) {
2566                 goto re_arm;
2567         }
2568
2569         read_lock(&bond->curr_slave_lock);
2570         oldcurrent = bond->curr_active_slave;
2571         read_unlock(&bond->curr_slave_lock);
2572
2573         /* see if any of the previous devices are up now (i.e. they have
2574          * xmt and rcv traffic). the curr_active_slave does not come into
2575          * the picture unless it is null. also, slave->jiffies is not needed
2576          * here because we send an arp on each slave and give a slave as
2577          * long as it needs to get the tx/rx within the delta.
2578          * TODO: what about up/down delay in arp mode? it wasn't here before
2579          *       so it can wait
2580          */
2581         bond_for_each_slave(bond, slave, i) {
2582                 if (slave->link != BOND_LINK_UP) {
2583                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2584                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2585
2586                                 slave->link  = BOND_LINK_UP;
2587                                 slave->state = BOND_STATE_ACTIVE;
2588
2589                                 /* primary_slave has no meaning in round-robin
2590                                  * mode. the window of a slave being up and
2591                                  * curr_active_slave being null after enslaving
2592                                  * is closed.
2593                                  */
2594                                 if (!oldcurrent) {
2595                                         printk(KERN_INFO DRV_NAME
2596                                                ": %s: link status definitely "
2597                                                "up for interface %s, ",
2598                                                bond_dev->name,
2599                                                slave->dev->name);
2600                                         do_failover = 1;
2601                                 } else {
2602                                         printk(KERN_INFO DRV_NAME
2603                                                ": %s: interface %s is now up\n",
2604                                                bond_dev->name,
2605                                                slave->dev->name);
2606                                 }
2607                         }
2608                 } else {
2609                         /* slave->link == BOND_LINK_UP */
2610
2611                         /* not all switches will respond to an arp request
2612                          * when the source ip is 0, so don't take the link down
2613                          * if we don't know our ip yet
2614                          */
2615                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2616                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2617                              bond_has_ip(bond))) {
2618
2619                                 slave->link  = BOND_LINK_DOWN;
2620                                 slave->state = BOND_STATE_BACKUP;
2621
2622                                 if (slave->link_failure_count < UINT_MAX) {
2623                                         slave->link_failure_count++;
2624                                 }
2625
2626                                 printk(KERN_INFO DRV_NAME
2627                                        ": %s: interface %s is now down.\n",
2628                                        bond_dev->name,
2629                                        slave->dev->name);
2630
2631                                 if (slave == oldcurrent) {
2632                                         do_failover = 1;
2633                                 }
2634                         }
2635                 }
2636
2637                 /* note: if switch is in round-robin mode, all links
2638                  * must tx arp to ensure all links rx an arp - otherwise
2639                  * links may oscillate or not come up at all; if switch is
2640                  * in something like xor mode, there is nothing we can
2641                  * do - all replies will be rx'ed on same link causing slaves
2642                  * to be unstable during low/no traffic periods
2643                  */
2644                 if (IS_UP(slave->dev)) {
2645                         bond_arp_send_all(bond, slave);
2646                 }
2647         }
2648
2649         if (do_failover) {
2650                 write_lock(&bond->curr_slave_lock);
2651
2652                 bond_select_active_slave(bond);
2653
2654                 write_unlock(&bond->curr_slave_lock);
2655         }
2656
2657 re_arm:
2658         if (bond->params.arp_interval) {
2659                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2660         }
2661 out:
2662         read_unlock(&bond->lock);
2663 }
2664
2665 /*
2666  * When using arp monitoring in active-backup mode, this function is
2667  * called to determine if any backup slaves have went down or a new
2668  * current slave needs to be found.
2669  * The backup slaves never generate traffic, they are considered up by merely
2670  * receiving traffic. If the current slave goes down, each backup slave will
2671  * be given the opportunity to tx/rx an arp before being taken down - this
2672  * prevents all slaves from being taken down due to the current slave not
2673  * sending any traffic for the backups to receive. The arps are not necessarily
2674  * necessary, any tx and rx traffic will keep the current slave up. While any
2675  * rx traffic will keep the backup slaves up, the current slave is responsible
2676  * for generating traffic to keep them up regardless of any other traffic they
2677  * may have received.
2678  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2679  */
2680 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2681 {
2682         struct bonding *bond = bond_dev->priv;
2683         struct slave *slave;
2684         int delta_in_ticks;
2685         int i;
2686
2687         read_lock(&bond->lock);
2688
2689         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2690
2691         if (bond->kill_timers) {
2692                 goto out;
2693         }
2694
2695         if (bond->slave_cnt == 0) {
2696                 goto re_arm;
2697         }
2698
2699         /* determine if any slave has come up or any backup slave has
2700          * gone down
2701          * TODO: what about up/down delay in arp mode? it wasn't here before
2702          *       so it can wait
2703          */
2704         bond_for_each_slave(bond, slave, i) {
2705                 if (slave->link != BOND_LINK_UP) {
2706                         if ((jiffies - slave_last_rx(bond, slave)) <=
2707                              delta_in_ticks) {
2708
2709                                 slave->link = BOND_LINK_UP;
2710
2711                                 write_lock(&bond->curr_slave_lock);
2712
2713                                 if ((!bond->curr_active_slave) &&
2714                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2715                                         bond_change_active_slave(bond, slave);
2716                                         bond->current_arp_slave = NULL;
2717                                 } else if (bond->curr_active_slave != slave) {
2718                                         /* this slave has just come up but we
2719                                          * already have a current slave; this
2720                                          * can also happen if bond_enslave adds
2721                                          * a new slave that is up while we are
2722                                          * searching for a new slave
2723                                          */
2724                                         bond_set_slave_inactive_flags(slave);
2725                                         bond->current_arp_slave = NULL;
2726                                 }
2727
2728                                 bond_set_carrier(bond);
2729
2730                                 if (slave == bond->curr_active_slave) {
2731                                         printk(KERN_INFO DRV_NAME
2732                                                ": %s: %s is up and now the "
2733                                                "active interface\n",
2734                                                bond_dev->name,
2735                                                slave->dev->name);
2736                                         netif_carrier_on(bond->dev);
2737                                 } else {
2738                                         printk(KERN_INFO DRV_NAME
2739                                                ": %s: backup interface %s is "
2740                                                "now up\n",
2741                                                bond_dev->name,
2742                                                slave->dev->name);
2743                                 }
2744
2745                                 write_unlock(&bond->curr_slave_lock);
2746                         }
2747                 } else {
2748                         read_lock(&bond->curr_slave_lock);
2749
2750                         if ((slave != bond->curr_active_slave) &&
2751                             (!bond->current_arp_slave) &&
2752                             (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2753                              bond_has_ip(bond))) {
2754                                 /* a backup slave has gone down; three times
2755                                  * the delta allows the current slave to be
2756                                  * taken out before the backup slave.
2757                                  * note: a non-null current_arp_slave indicates
2758                                  * the curr_active_slave went down and we are
2759                                  * searching for a new one; under this
2760                                  * condition we only take the curr_active_slave
2761                                  * down - this gives each slave a chance to
2762                                  * tx/rx traffic before being taken out
2763                                  */
2764
2765                                 read_unlock(&bond->curr_slave_lock);
2766
2767                                 slave->link  = BOND_LINK_DOWN;
2768
2769                                 if (slave->link_failure_count < UINT_MAX) {
2770                                         slave->link_failure_count++;
2771                                 }
2772
2773                                 bond_set_slave_inactive_flags(slave);
2774
2775                                 printk(KERN_INFO DRV_NAME
2776                                        ": %s: backup interface %s is now down\n",
2777                                        bond_dev->name,
2778                                        slave->dev->name);
2779                         } else {
2780                                 read_unlock(&bond->curr_slave_lock);
2781                         }
2782                 }
2783         }
2784
2785         read_lock(&bond->curr_slave_lock);
2786         slave = bond->curr_active_slave;
2787         read_unlock(&bond->curr_slave_lock);
2788
2789         if (slave) {
2790                 /* if we have sent traffic in the past 2*arp_intervals but
2791                  * haven't xmit and rx traffic in that time interval, select
2792                  * a different slave. slave->jiffies is only updated when
2793                  * a slave first becomes the curr_active_slave - not necessarily
2794                  * after every arp; this ensures the slave has a full 2*delta
2795                  * before being taken out. if a primary is being used, check
2796                  * if it is up and needs to take over as the curr_active_slave
2797                  */
2798                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2799             (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2800              bond_has_ip(bond))) &&
2801                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2802
2803                         slave->link  = BOND_LINK_DOWN;
2804
2805                         if (slave->link_failure_count < UINT_MAX) {
2806                                 slave->link_failure_count++;
2807                         }
2808
2809                         printk(KERN_INFO DRV_NAME
2810                                ": %s: link status down for active interface "
2811                                "%s, disabling it\n",
2812                                bond_dev->name,
2813                                slave->dev->name);
2814
2815                         write_lock(&bond->curr_slave_lock);
2816
2817                         bond_select_active_slave(bond);
2818                         slave = bond->curr_active_slave;
2819
2820                         write_unlock(&bond->curr_slave_lock);
2821
2822                         bond->current_arp_slave = slave;
2823
2824                         if (slave) {
2825                                 slave->jiffies = jiffies;
2826                         }
2827                 } else if ((bond->primary_slave) &&
2828                            (bond->primary_slave != slave) &&
2829                            (bond->primary_slave->link == BOND_LINK_UP)) {
2830                         /* at this point, slave is the curr_active_slave */
2831                         printk(KERN_INFO DRV_NAME
2832                                ": %s: changing from interface %s to primary "
2833                                "interface %s\n",
2834                                bond_dev->name,
2835                                slave->dev->name,
2836                                bond->primary_slave->dev->name);
2837
2838                         /* primary is up so switch to it */
2839                         write_lock(&bond->curr_slave_lock);
2840                         bond_change_active_slave(bond, bond->primary_slave);
2841                         write_unlock(&bond->curr_slave_lock);
2842
2843                         slave = bond->primary_slave;
2844                         slave->jiffies = jiffies;
2845                 } else {
2846                         bond->current_arp_slave = NULL;
2847                 }
2848
2849                 /* the current slave must tx an arp to ensure backup slaves
2850                  * rx traffic
2851                  */
2852                 if (slave && bond_has_ip(bond)) {
2853                         bond_arp_send_all(bond, slave);
2854                 }
2855         }
2856
2857         /* if we don't have a curr_active_slave, search for the next available
2858          * backup slave from the current_arp_slave and make it the candidate
2859          * for becoming the curr_active_slave
2860          */
2861         if (!slave) {
2862                 if (!bond->current_arp_slave) {
2863                         bond->current_arp_slave = bond->first_slave;
2864                 }
2865
2866                 if (bond->current_arp_slave) {
2867                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2868
2869                         /* search for next candidate */
2870                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2871                                 if (IS_UP(slave->dev)) {
2872                                         slave->link = BOND_LINK_BACK;
2873                                         bond_set_slave_active_flags(slave);
2874                                         bond_arp_send_all(bond, slave);
2875                                         slave->jiffies = jiffies;
2876                                         bond->current_arp_slave = slave;
2877                                         break;
2878                                 }
2879
2880                                 /* if the link state is up at this point, we
2881                                  * mark it down - this can happen if we have
2882                                  * simultaneous link failures and
2883                                  * reselect_active_interface doesn't make this
2884                                  * one the current slave so it is still marked
2885                                  * up when it is actually down
2886                                  */
2887                                 if (slave->link == BOND_LINK_UP) {
2888                                         slave->link  = BOND_LINK_DOWN;
2889                                         if (slave->link_failure_count < UINT_MAX) {
2890                                                 slave->link_failure_count++;
2891                                         }
2892
2893                                         bond_set_slave_inactive_flags(slave);
2894
2895                                         printk(KERN_INFO DRV_NAME
2896                                                ": %s: backup interface %s is "
2897                                                "now down.\n",
2898                                                bond_dev->name,
2899                                                slave->dev->name);
2900                                 }
2901                         }
2902                 }
2903         }
2904
2905 re_arm:
2906         if (bond->params.arp_interval) {
2907                 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2908         }
2909 out:
2910         read_unlock(&bond->lock);
2911 }
2912
2913 /*------------------------------ proc/seq_file-------------------------------*/
2914
2915 #ifdef CONFIG_PROC_FS
2916
2917 #define SEQ_START_TOKEN ((void *)1)
2918
2919 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2920 {
2921         struct bonding *bond = seq->private;
2922         loff_t off = 0;
2923         struct slave *slave;
2924         int i;
2925
2926         /* make sure the bond won't be taken away */
2927         read_lock(&dev_base_lock);
2928         read_lock_bh(&bond->lock);
2929
2930         if (*pos == 0) {
2931                 return SEQ_START_TOKEN;
2932         }
2933
2934         bond_for_each_slave(bond, slave, i) {
2935                 if (++off == *pos) {
2936                         return slave;
2937                 }
2938         }
2939
2940         return NULL;
2941 }
2942
2943 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2944 {
2945         struct bonding *bond = seq->private;
2946         struct slave *slave = v;
2947
2948         ++*pos;
2949         if (v == SEQ_START_TOKEN) {
2950                 return bond->first_slave;
2951         }
2952
2953         slave = slave->next;
2954
2955         return (slave == bond->first_slave) ? NULL : slave;
2956 }
2957
2958 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2959 {
2960         struct bonding *bond = seq->private;
2961
2962         read_unlock_bh(&bond->lock);
2963         read_unlock(&dev_base_lock);
2964 }
2965
2966 static void bond_info_show_master(struct seq_file *seq)
2967 {
2968         struct bonding *bond = seq->private;
2969         struct slave *curr;
2970         int i;
2971         u32 target;
2972
2973         read_lock(&bond->curr_slave_lock);
2974         curr = bond->curr_active_slave;
2975         read_unlock(&bond->curr_slave_lock);
2976
2977         seq_printf(seq, "Bonding Mode: %s\n",
2978                    bond_mode_name(bond->params.mode));
2979
2980         if (bond->params.mode == BOND_MODE_XOR ||
2981                 bond->params.mode == BOND_MODE_8023AD) {
2982                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2983                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2984                         bond->params.xmit_policy);
2985         }
2986
2987         if (USES_PRIMARY(bond->params.mode)) {
2988                 seq_printf(seq, "Primary Slave: %s\n",
2989                            (bond->primary_slave) ?
2990                            bond->primary_slave->dev->name : "None");
2991
2992                 seq_printf(seq, "Currently Active Slave: %s\n",
2993                            (curr) ? curr->dev->name : "None");
2994         }
2995
2996         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2997                    "up" : "down");
2998         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2999         seq_printf(seq, "Up Delay (ms): %d\n",
3000                    bond->params.updelay * bond->params.miimon);
3001         seq_printf(seq, "Down Delay (ms): %d\n",
3002                    bond->params.downdelay * bond->params.miimon);
3003
3004
3005         /* ARP information */
3006         if(bond->params.arp_interval > 0) {
3007                 int printed=0;
3008                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3009                                 bond->params.arp_interval);
3010
3011                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3012
3013                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3014                         if (!bond->params.arp_targets[i])
3015                                 continue;
3016                         if (printed)
3017                                 seq_printf(seq, ",");
3018                         target = ntohl(bond->params.arp_targets[i]);
3019                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3020                         printed = 1;
3021                 }
3022                 seq_printf(seq, "\n");
3023         }
3024
3025         if (bond->params.mode == BOND_MODE_8023AD) {
3026                 struct ad_info ad_info;
3027
3028                 seq_puts(seq, "\n802.3ad info\n");
3029                 seq_printf(seq, "LACP rate: %s\n",
3030                            (bond->params.lacp_fast) ? "fast" : "slow");
3031
3032                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3033                         seq_printf(seq, "bond %s has no active aggregator\n",
3034                                    bond->dev->name);
3035                 } else {
3036                         seq_printf(seq, "Active Aggregator Info:\n");
3037
3038                         seq_printf(seq, "\tAggregator ID: %d\n",
3039                                    ad_info.aggregator_id);
3040                         seq_printf(seq, "\tNumber of ports: %d\n",
3041                                    ad_info.ports);
3042                         seq_printf(seq, "\tActor Key: %d\n",
3043                                    ad_info.actor_key);
3044                         seq_printf(seq, "\tPartner Key: %d\n",
3045                                    ad_info.partner_key);
3046                         seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3047                                    ad_info.partner_system[0],
3048                                    ad_info.partner_system[1],
3049                                    ad_info.partner_system[2],
3050                                    ad_info.partner_system[3],
3051                                    ad_info.partner_system[4],
3052                                    ad_info.partner_system[5]);
3053                 }
3054         }
3055 }
3056
3057 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3058 {
3059         struct bonding *bond = seq->private;
3060
3061         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3062         seq_printf(seq, "MII Status: %s\n",
3063                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3064         seq_printf(seq, "Link Failure Count: %u\n",
3065                    slave->link_failure_count);
3066
3067         seq_printf(seq,
3068                    "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3069                    slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3070                    slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3071                    slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3072
3073         if (bond->params.mode == BOND_MODE_8023AD) {
3074                 const struct aggregator *agg
3075                         = SLAVE_AD_INFO(slave).port.aggregator;
3076
3077                 if (agg) {
3078                         seq_printf(seq, "Aggregator ID: %d\n",
3079                                    agg->aggregator_identifier);
3080                 } else {
3081                         seq_puts(seq, "Aggregator ID: N/A\n");
3082                 }
3083         }
3084 }
3085
3086 static int bond_info_seq_show(struct seq_file *seq, void *v)
3087 {
3088         if (v == SEQ_START_TOKEN) {
3089                 seq_printf(seq, "%s\n", version);
3090                 bond_info_show_master(seq);
3091         } else {
3092                 bond_info_show_slave(seq, v);
3093         }
3094
3095         return 0;
3096 }
3097
3098 static struct seq_operations bond_info_seq_ops = {
3099         .start = bond_info_seq_start,
3100         .next  = bond_info_seq_next,
3101         .stop  = bond_info_seq_stop,
3102         .show  = bond_info_seq_show,
3103 };
3104
3105 static int bond_info_open(struct inode *inode, struct file *file)
3106 {
3107         struct seq_file *seq;
3108         struct proc_dir_entry *proc;
3109         int res;
3110
3111         res = seq_open(file, &bond_info_seq_ops);
3112         if (!res) {
3113                 /* recover the pointer buried in proc_dir_entry data */
3114                 seq = file->private_data;
3115                 proc = PDE(inode);
3116                 seq->private = proc->data;
3117         }
3118
3119         return res;
3120 }
3121
3122 static const struct file_operations bond_info_fops = {
3123         .owner   = THIS_MODULE,
3124         .open    = bond_info_open,
3125         .read    = seq_read,
3126         .llseek  = seq_lseek,
3127         .release = seq_release,
3128 };
3129
3130 static int bond_create_proc_entry(struct bonding *bond)
3131 {
3132         struct net_device *bond_dev = bond->dev;
3133
3134         if (bond_proc_dir) {
3135                 bond->proc_entry = create_proc_entry(bond_dev->name,
3136                                                      S_IRUGO,
3137                                                      bond_proc_dir);
3138                 if (bond->proc_entry == NULL) {
3139                         printk(KERN_WARNING DRV_NAME
3140                                ": Warning: Cannot create /proc/net/%s/%s\n",
3141                                DRV_NAME, bond_dev->name);
3142                 } else {
3143                         bond->proc_entry->data = bond;
3144                         bond->proc_entry->proc_fops = &bond_info_fops;
3145                         bond->proc_entry->owner = THIS_MODULE;
3146                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3147                 }
3148         }
3149
3150         return 0;
3151 }
3152
3153 static void bond_remove_proc_entry(struct bonding *bond)
3154 {
3155         if (bond_proc_dir && bond->proc_entry) {
3156                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3157                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3158                 bond->proc_entry = NULL;
3159         }
3160 }
3161
3162 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3163  * Caller must hold rtnl_lock.
3164  */
3165 static void bond_create_proc_dir(void)
3166 {
3167         int len = strlen(DRV_NAME);
3168
3169         for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3170              bond_proc_dir = bond_proc_dir->next) {
3171                 if ((bond_proc_dir->namelen == len) &&
3172                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3173                         break;
3174                 }
3175         }
3176
3177         if (!bond_proc_dir) {
3178                 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3179                 if (bond_proc_dir) {
3180                         bond_proc_dir->owner = THIS_MODULE;
3181                 } else {
3182                         printk(KERN_WARNING DRV_NAME
3183                                 ": Warning: cannot create /proc/net/%s\n",
3184                                 DRV_NAME);
3185                 }
3186         }
3187 }
3188
3189 /* Destroy the bonding directory under /proc/net, if empty.
3190  * Caller must hold rtnl_lock.
3191  */
3192 static void bond_destroy_proc_dir(void)
3193 {
3194         struct proc_dir_entry *de;
3195
3196         if (!bond_proc_dir) {
3197                 return;
3198         }
3199
3200         /* verify that the /proc dir is empty */
3201         for (de = bond_proc_dir->subdir; de; de = de->next) {
3202                 /* ignore . and .. */
3203                 if (*(de->name) != '.') {
3204                         break;
3205                 }
3206         }
3207
3208         if (de) {
3209                 if (bond_proc_dir->owner == THIS_MODULE) {
3210                         bond_proc_dir->owner = NULL;
3211                 }
3212         } else {
3213                 remove_proc_entry(DRV_NAME, proc_net);
3214                 bond_proc_dir = NULL;
3215         }
3216 }
3217 #endif /* CONFIG_PROC_FS */
3218
3219 /*-------------------------- netdev event handling --------------------------*/
3220
3221 /*
3222  * Change device name
3223  */
3224 static int bond_event_changename(struct bonding *bond)
3225 {
3226 #ifdef CONFIG_PROC_FS
3227         bond_remove_proc_entry(bond);
3228         bond_create_proc_entry(bond);
3229 #endif
3230         down_write(&(bonding_rwsem));
3231         bond_destroy_sysfs_entry(bond);
3232         bond_create_sysfs_entry(bond);
3233         up_write(&(bonding_rwsem));
3234         return NOTIFY_DONE;
3235 }
3236
3237 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3238 {
3239         struct bonding *event_bond = bond_dev->priv;
3240
3241         switch (event) {
3242         case NETDEV_CHANGENAME:
3243                 return bond_event_changename(event_bond);
3244         case NETDEV_UNREGISTER:
3245                 /*
3246                  * TODO: remove a bond from the list?
3247                  */
3248                 break;
3249         default:
3250                 break;
3251         }
3252
3253         return NOTIFY_DONE;
3254 }
3255
3256 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3257 {
3258         struct net_device *bond_dev = slave_dev->master;
3259         struct bonding *bond = bond_dev->priv;
3260
3261         switch (event) {
3262         case NETDEV_UNREGISTER:
3263                 if (bond_dev) {
3264                         bond_release(bond_dev, slave_dev);
3265                 }
3266                 break;
3267         case NETDEV_CHANGE:
3268                 /*
3269                  * TODO: is this what we get if somebody
3270                  * sets up a hierarchical bond, then rmmod's
3271                  * one of the slave bonding devices?
3272                  */
3273                 break;
3274         case NETDEV_DOWN:
3275                 /*
3276                  * ... Or is it this?
3277                  */
3278                 break;
3279         case NETDEV_CHANGEMTU:
3280                 /*
3281                  * TODO: Should slaves be allowed to
3282                  * independently alter their MTU?  For
3283                  * an active-backup bond, slaves need
3284                  * not be the same type of device, so
3285                  * MTUs may vary.  For other modes,
3286                  * slaves arguably should have the
3287                  * same MTUs. To do this, we'd need to
3288                  * take over the slave's change_mtu
3289                  * function for the duration of their
3290                  * servitude.
3291                  */
3292                 break;
3293         case NETDEV_CHANGENAME:
3294                 /*
3295                  * TODO: handle changing the primary's name
3296                  */
3297                 break;
3298         case NETDEV_FEAT_CHANGE:
3299                 bond_compute_features(bond);
3300                 break;
3301         default:
3302                 break;
3303         }
3304
3305         return NOTIFY_DONE;
3306 }
3307
3308 /*
3309  * bond_netdev_event: handle netdev notifier chain events.
3310  *
3311  * This function receives events for the netdev chain.  The caller (an
3312  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3313  * locks for us to safely manipulate the slave devices (RTNL lock,
3314  * dev_probe_lock).
3315  */
3316 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3317 {
3318         struct net_device *event_dev = (struct net_device *)ptr;
3319
3320         dprintk("event_dev: %s, event: %lx\n",
3321                 (event_dev ? event_dev->name : "None"),
3322                 event);
3323
3324         if (!(event_dev->priv_flags & IFF_BONDING))
3325                 return NOTIFY_DONE;
3326
3327         if (event_dev->flags & IFF_MASTER) {
3328                 dprintk("IFF_MASTER\n");
3329                 return bond_master_netdev_event(event, event_dev);
3330         }
3331
3332         if (event_dev->flags & IFF_SLAVE) {
3333                 dprintk("IFF_SLAVE\n");
3334                 return bond_slave_netdev_event(event, event_dev);
3335         }
3336
3337         return NOTIFY_DONE;
3338 }
3339
3340 /*
3341  * bond_inetaddr_event: handle inetaddr notifier chain events.
3342  *
3343  * We keep track of device IPs primarily to use as source addresses in
3344  * ARP monitor probes (rather than spewing out broadcasts all the time).
3345  *
3346  * We track one IP for the main device (if it has one), plus one per VLAN.
3347  */
3348 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3349 {
3350         struct in_ifaddr *ifa = ptr;
3351         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3352         struct bonding *bond, *bond_next;
3353         struct vlan_entry *vlan, *vlan_next;
3354
3355         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3356                 if (bond->dev == event_dev) {
3357                         switch (event) {
3358                         case NETDEV_UP:
3359                                 bond->master_ip = ifa->ifa_local;
3360                                 return NOTIFY_OK;
3361                         case NETDEV_DOWN:
3362                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3363                                 return NOTIFY_OK;
3364                         default:
3365                                 return NOTIFY_DONE;
3366                         }
3367                 }
3368
3369                 if (list_empty(&bond->vlan_list))
3370                         continue;
3371
3372                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3373                                          vlan_list) {
3374                         vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3375                         if (vlan_dev == event_dev) {
3376                                 switch (event) {
3377                                 case NETDEV_UP:
3378                                         vlan->vlan_ip = ifa->ifa_local;
3379                                         return NOTIFY_OK;
3380                                 case NETDEV_DOWN:
3381                                         vlan->vlan_ip =
3382                                                 bond_glean_dev_ip(vlan_dev);
3383                                         return NOTIFY_OK;
3384                                 default:
3385                                         return NOTIFY_DONE;
3386                                 }
3387                         }
3388                 }
3389         }
3390         return NOTIFY_DONE;
3391 }
3392
3393 static struct notifier_block bond_netdev_notifier = {
3394         .notifier_call = bond_netdev_event,
3395 };
3396
3397 static struct notifier_block bond_inetaddr_notifier = {
3398         .notifier_call = bond_inetaddr_event,
3399 };
3400
3401 /*-------------------------- Packet type handling ---------------------------*/
3402
3403 /* register to receive lacpdus on a bond */
3404 static void bond_register_lacpdu(struct bonding *bond)
3405 {
3406         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3407
3408         /* initialize packet type */
3409         pk_type->type = PKT_TYPE_LACPDU;
3410         pk_type->dev = bond->dev;
3411         pk_type->func = bond_3ad_lacpdu_recv;
3412
3413         dev_add_pack(pk_type);
3414 }
3415
3416 /* unregister to receive lacpdus on a bond */
3417 static void bond_unregister_lacpdu(struct bonding *bond)
3418 {
3419         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3420 }
3421
3422 void bond_register_arp(struct bonding *bond)
3423 {
3424         struct packet_type *pt = &bond->arp_mon_pt;
3425
3426         pt->type = htons(ETH_P_ARP);
3427         pt->dev = NULL; /*bond->dev;XXX*/
3428         pt->func = bond_arp_rcv;
3429         dev_add_pack(pt);
3430 }
3431
3432 void bond_unregister_arp(struct bonding *bond)
3433 {
3434         dev_remove_pack(&bond->arp_mon_pt);
3435 }
3436
3437 /*---------------------------- Hashing Policies -----------------------------*/
3438
3439 /*
3440  * Hash for the the output device based upon layer 3 and layer 4 data. If
3441  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3442  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3443  */
3444 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3445                                     struct net_device *bond_dev, int count)
3446 {
3447         struct ethhdr *data = (struct ethhdr *)skb->data;
3448         struct iphdr *iph = skb->nh.iph;
3449         u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3450         int layer4_xor = 0;
3451
3452         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3453                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3454                     (iph->protocol == IPPROTO_TCP ||
3455                      iph->protocol == IPPROTO_UDP)) {
3456                         layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3457                 }
3458                 return (layer4_xor ^
3459                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3460
3461         }
3462
3463         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3464 }
3465
3466 /*
3467  * Hash for the output device based upon layer 2 data
3468  */
3469 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3470                                    struct net_device *bond_dev, int count)
3471 {
3472         struct ethhdr *data = (struct ethhdr *)skb->data;
3473
3474         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3475 }
3476
3477 /*-------------------------- Device entry points ----------------------------*/
3478
3479 static int bond_open(struct net_device *bond_dev)
3480 {
3481         struct bonding *bond = bond_dev->priv;
3482         struct timer_list *mii_timer = &bond->mii_timer;
3483         struct timer_list *arp_timer = &bond->arp_timer;
3484
3485         bond->kill_timers = 0;
3486
3487         if ((bond->params.mode == BOND_MODE_TLB) ||
3488             (bond->params.mode == BOND_MODE_ALB)) {
3489                 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3490
3491                 /* bond_alb_initialize must be called before the timer
3492                  * is started.
3493                  */
3494                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3495                         /* something went wrong - fail the open operation */
3496                         return -1;
3497                 }
3498
3499                 init_timer(alb_timer);
3500                 alb_timer->expires  = jiffies + 1;
3501                 alb_timer->data     = (unsigned long)bond;
3502                 alb_timer->function = (void *)&bond_alb_monitor;
3503                 add_timer(alb_timer);
3504         }
3505
3506         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3507                 init_timer(mii_timer);
3508                 mii_timer->expires  = jiffies + 1;
3509                 mii_timer->data     = (unsigned long)bond_dev;
3510                 mii_timer->function = (void *)&bond_mii_monitor;
3511                 add_timer(mii_timer);
3512         }
3513
3514         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3515                 init_timer(arp_timer);
3516                 arp_timer->expires  = jiffies + 1;
3517                 arp_timer->data     = (unsigned long)bond_dev;
3518                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3519                         arp_timer->function = (void *)&bond_activebackup_arp_mon;
3520                 } else {
3521                         arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3522                 }
3523                 if (bond->params.arp_validate)
3524                         bond_register_arp(bond);
3525
3526                 add_timer(arp_timer);
3527         }
3528
3529         if (bond->params.mode == BOND_MODE_8023AD) {
3530                 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3531                 init_timer(ad_timer);
3532                 ad_timer->expires  = jiffies + 1;
3533                 ad_timer->data     = (unsigned long)bond;
3534                 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3535                 add_timer(ad_timer);
3536
3537                 /* register to receive LACPDUs */
3538                 bond_register_lacpdu(bond);
3539         }
3540
3541         return 0;
3542 }
3543
3544 static int bond_close(struct net_device *bond_dev)
3545 {
3546         struct bonding *bond = bond_dev->priv;
3547
3548         if (bond->params.mode == BOND_MODE_8023AD) {
3549                 /* Unregister the receive of LACPDUs */
3550                 bond_unregister_lacpdu(bond);
3551         }
3552
3553         if (bond->params.arp_validate)
3554                 bond_unregister_arp(bond);
3555
3556         write_lock_bh(&bond->lock);
3557
3558
3559         /* signal timers not to re-arm */
3560         bond->kill_timers = 1;
3561
3562         write_unlock_bh(&bond->lock);
3563
3564         /* del_timer_sync must run without holding the bond->lock
3565          * because a running timer might be trying to hold it too
3566          */
3567
3568         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3569                 del_timer_sync(&bond->mii_timer);
3570         }
3571
3572         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3573                 del_timer_sync(&bond->arp_timer);
3574         }
3575
3576         switch (bond->params.mode) {
3577         case BOND_MODE_8023AD:
3578                 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3579                 break;
3580         case BOND_MODE_TLB:
3581         case BOND_MODE_ALB:
3582                 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3583                 break;
3584         default:
3585                 break;
3586         }
3587
3588
3589         if ((bond->params.mode == BOND_MODE_TLB) ||
3590             (bond->params.mode == BOND_MODE_ALB)) {
3591                 /* Must be called only after all
3592                  * slaves have been released
3593                  */
3594                 bond_alb_deinitialize(bond);
3595         }
3596
3597         return 0;
3598 }
3599
3600 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3601 {
3602         struct bonding *bond = bond_dev->priv;
3603         struct net_device_stats *stats = &(bond->stats), *sstats;
3604         struct slave *slave;
3605         int i;
3606
3607         memset(stats, 0, sizeof(struct net_device_stats));
3608
3609         read_lock_bh(&bond->lock);
3610
3611         bond_for_each_slave(bond, slave, i) {
3612                 if (slave->dev->get_stats) {
3613                         sstats = slave->dev->get_stats(slave->dev);
3614
3615                         stats->rx_packets += sstats->rx_packets;
3616                         stats->rx_bytes += sstats->rx_bytes;
3617                         stats->rx_errors += sstats->rx_errors;
3618                         stats->rx_dropped += sstats->rx_dropped;
3619
3620                         stats->tx_packets += sstats->tx_packets;
3621                         stats->tx_bytes += sstats->tx_bytes;
3622                         stats->tx_errors += sstats->tx_errors;
3623                         stats->tx_dropped += sstats->tx_dropped;
3624
3625                         stats->multicast += sstats->multicast;
3626                         stats->collisions += sstats->collisions;
3627
3628                         stats->rx_length_errors += sstats->rx_length_errors;
3629                         stats->rx_over_errors += sstats->rx_over_errors;
3630                         stats->rx_crc_errors += sstats->rx_crc_errors;
3631                         stats->rx_frame_errors += sstats->rx_frame_errors;
3632                         stats->rx_fifo_errors += sstats->rx_fifo_errors;
3633                         stats->rx_missed_errors += sstats->rx_missed_errors;
3634
3635                         stats->tx_aborted_errors += sstats->tx_aborted_errors;
3636                         stats->tx_carrier_errors += sstats->tx_carrier_errors;
3637                         stats->tx_fifo_errors += sstats->tx_fifo_errors;
3638                         stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3639                         stats->tx_window_errors += sstats->tx_window_errors;
3640                 }
3641         }
3642
3643         read_unlock_bh(&bond->lock);
3644
3645         return stats;
3646 }
3647
3648 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3649 {
3650         struct net_device *slave_dev = NULL;
3651         struct ifbond k_binfo;
3652         struct ifbond __user *u_binfo = NULL;
3653         struct ifslave k_sinfo;
3654         struct ifslave __user *u_sinfo = NULL;
3655         struct mii_ioctl_data *mii = NULL;
3656         int res = 0;
3657
3658         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3659                 bond_dev->name, cmd);
3660
3661         switch (cmd) {
3662         case SIOCGMIIPHY:
3663                 mii = if_mii(ifr);
3664                 if (!mii) {
3665                         return -EINVAL;
3666                 }
3667                 mii->phy_id = 0;
3668                 /* Fall Through */
3669         case SIOCGMIIREG:
3670                 /*
3671                  * We do this again just in case we were called by SIOCGMIIREG
3672                  * instead of SIOCGMIIPHY.
3673                  */
3674                 mii = if_mii(ifr);
3675                 if (!mii) {
3676                         return -EINVAL;
3677                 }
3678
3679                 if (mii->reg_num == 1) {
3680                         struct bonding *bond = bond_dev->priv;
3681                         mii->val_out = 0;
3682                         read_lock_bh(&bond->lock);
3683                         read_lock(&bond->curr_slave_lock);
3684                         if (netif_carrier_ok(bond->dev)) {
3685                                 mii->val_out = BMSR_LSTATUS;
3686                         }
3687                         read_unlock(&bond->curr_slave_lock);
3688                         read_unlock_bh(&bond->lock);
3689                 }
3690
3691                 return 0;
3692         case BOND_INFO_QUERY_OLD:
3693         case SIOCBONDINFOQUERY:
3694                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3695
3696                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3697                         return -EFAULT;
3698                 }
3699
3700                 res = bond_info_query(bond_dev, &k_binfo);
3701                 if (res == 0) {
3702                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3703                                 return -EFAULT;
3704                         }
3705                 }
3706
3707                 return res;
3708         case BOND_SLAVE_INFO_QUERY_OLD:
3709         case SIOCBONDSLAVEINFOQUERY:
3710                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3711
3712                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3713                         return -EFAULT;
3714                 }
3715
3716                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3717                 if (res == 0) {
3718                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3719                                 return -EFAULT;
3720                         }
3721                 }
3722
3723                 return res;
3724         default:
3725                 /* Go on */
3726                 break;
3727         }
3728
3729         if (!capable(CAP_NET_ADMIN)) {
3730                 return -EPERM;
3731         }
3732
3733         down_write(&(bonding_rwsem));
3734         slave_dev = dev_get_by_name(ifr->ifr_slave);
3735
3736         dprintk("slave_dev=%p: \n", slave_dev);
3737
3738         if (!slave_dev) {
3739                 res = -ENODEV;
3740         } else {
3741                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3742                 switch (cmd) {
3743                 case BOND_ENSLAVE_OLD:
3744                 case SIOCBONDENSLAVE:
3745                         res = bond_enslave(bond_dev, slave_dev);
3746                         break;
3747                 case BOND_RELEASE_OLD:
3748                 case SIOCBONDRELEASE:
3749                         res = bond_release(bond_dev, slave_dev);
3750                         break;
3751                 case BOND_SETHWADDR_OLD:
3752                 case SIOCBONDSETHWADDR:
3753                         res = bond_sethwaddr(bond_dev, slave_dev);
3754                         break;
3755                 case BOND_CHANGE_ACTIVE_OLD:
3756                 case SIOCBONDCHANGEACTIVE:
3757                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3758                         break;
3759                 default:
3760                         res = -EOPNOTSUPP;
3761                 }
3762
3763                 dev_put(slave_dev);
3764         }
3765
3766         up_write(&(bonding_rwsem));
3767         return res;
3768 }
3769
3770 static void bond_set_multicast_list(struct net_device *bond_dev)
3771 {
3772         struct bonding *bond = bond_dev->priv;
3773         struct dev_mc_list *dmi;
3774
3775         write_lock_bh(&bond->lock);
3776
3777         /*
3778          * Do promisc before checking multicast_mode
3779          */
3780         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3781                 bond_set_promiscuity(bond, 1);
3782         }
3783
3784         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3785                 bond_set_promiscuity(bond, -1);
3786         }
3787
3788         /* set allmulti flag to slaves */
3789         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3790                 bond_set_allmulti(bond, 1);
3791         }
3792
3793         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3794                 bond_set_allmulti(bond, -1);
3795         }
3796
3797         bond->flags = bond_dev->flags;
3798
3799         /* looking for addresses to add to slaves' mc list */
3800         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3801                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3802                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3803                 }
3804         }
3805
3806         /* looking for addresses to delete from slaves' list */
3807         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3808                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3809                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3810                 }
3811         }
3812
3813         /* save master's multicast list */
3814         bond_mc_list_destroy(bond);
3815         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3816
3817         write_unlock_bh(&bond->lock);
3818 }
3819
3820 /*
3821  * Change the MTU of all of a master's slaves to match the master
3822  */
3823 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3824 {
3825         struct bonding *bond = bond_dev->priv;
3826         struct slave *slave, *stop_at;
3827         int res = 0;
3828         int i;
3829
3830         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3831                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3832
3833         /* Can't hold bond->lock with bh disabled here since
3834          * some base drivers panic. On the other hand we can't
3835          * hold bond->lock without bh disabled because we'll
3836          * deadlock. The only solution is to rely on the fact
3837          * that we're under rtnl_lock here, and the slaves
3838          * list won't change. This doesn't solve the problem
3839          * of setting the slave's MTU while it is
3840          * transmitting, but the assumption is that the base
3841          * driver can handle that.
3842          *
3843          * TODO: figure out a way to safely iterate the slaves
3844          * list, but without holding a lock around the actual
3845          * call to the base driver.
3846          */
3847
3848         bond_for_each_slave(bond, slave, i) {
3849                 dprintk("s %p s->p %p c_m %p\n", slave,
3850                         slave->prev, slave->dev->change_mtu);
3851
3852                 res = dev_set_mtu(slave->dev, new_mtu);
3853
3854                 if (res) {
3855                         /* If we failed to set the slave's mtu to the new value
3856                          * we must abort the operation even in ACTIVE_BACKUP
3857                          * mode, because if we allow the backup slaves to have
3858                          * different mtu values than the active slave we'll
3859                          * need to change their mtu when doing a failover. That
3860                          * means changing their mtu from timer context, which
3861                          * is probably not a good idea.
3862                          */
3863                         dprintk("err %d %s\n", res, slave->dev->name);
3864                         goto unwind;
3865                 }
3866         }
3867
3868         bond_dev->mtu = new_mtu;
3869
3870         return 0;
3871
3872 unwind:
3873         /* unwind from head to the slave that failed */
3874         stop_at = slave;
3875         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3876                 int tmp_res;
3877
3878                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3879                 if (tmp_res) {
3880                         dprintk("unwind err %d dev %s\n", tmp_res,
3881                                 slave->dev->name);
3882                 }
3883         }
3884
3885         return res;
3886 }
3887
3888 /*
3889  * Change HW address
3890  *
3891  * Note that many devices must be down to change the HW address, and
3892  * downing the master releases all slaves.  We can make bonds full of
3893  * bonding devices to test this, however.
3894  */
3895 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3896 {
3897         struct bonding *bond = bond_dev->priv;
3898         struct sockaddr *sa = addr, tmp_sa;
3899         struct slave *slave, *stop_at;
3900         int res = 0;
3901         int i;
3902
3903         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3904
3905         if (!is_valid_ether_addr(sa->sa_data)) {
3906                 return -EADDRNOTAVAIL;
3907         }
3908
3909         /* Can't hold bond->lock with bh disabled here since
3910          * some base drivers panic. On the other hand we can't
3911          * hold bond->lock without bh disabled because we'll
3912          * deadlock. The only solution is to rely on the fact
3913          * that we're under rtnl_lock here, and the slaves
3914          * list won't change. This doesn't solve the problem
3915          * of setting the slave's hw address while it is
3916          * transmitting, but the assumption is that the base
3917          * driver can handle that.
3918          *
3919          * TODO: figure out a way to safely iterate the slaves
3920          * list, but without holding a lock around the actual
3921          * call to the base driver.
3922          */
3923
3924         bond_for_each_slave(bond, slave, i) {
3925                 dprintk("slave %p %s\n", slave, slave->dev->name);
3926
3927                 if (slave->dev->set_mac_address == NULL) {
3928                         res = -EOPNOTSUPP;
3929                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3930                         goto unwind;
3931                 }
3932
3933                 res = dev_set_mac_address(slave->dev, addr);
3934                 if (res) {
3935                         /* TODO: consider downing the slave
3936                          * and retry ?
3937                          * User should expect communications
3938                          * breakage anyway until ARP finish
3939                          * updating, so...
3940                          */
3941                         dprintk("err %d %s\n", res, slave->dev->name);
3942                         goto unwind;
3943                 }
3944         }
3945
3946         /* success */
3947         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3948         return 0;
3949
3950 unwind:
3951         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3952         tmp_sa.sa_family = bond_dev->type;
3953
3954         /* unwind from head to the slave that failed */
3955         stop_at = slave;
3956         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3957                 int tmp_res;
3958
3959                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3960                 if (tmp_res) {
3961                         dprintk("unwind err %d dev %s\n", tmp_res,
3962                                 slave->dev->name);
3963                 }
3964         }
3965
3966         return res;
3967 }
3968
3969 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3970 {
3971         struct bonding *bond = bond_dev->priv;
3972         struct slave *slave, *start_at;
3973         int i;
3974         int res = 1;
3975
3976         read_lock(&bond->lock);
3977
3978         if (!BOND_IS_OK(bond)) {
3979                 goto out;
3980         }
3981
3982         read_lock(&bond->curr_slave_lock);
3983         slave = start_at = bond->curr_active_slave;
3984         read_unlock(&bond->curr_slave_lock);
3985
3986         if (!slave) {
3987                 goto out;
3988         }
3989
3990         bond_for_each_slave_from(bond, slave, i, start_at) {
3991                 if (IS_UP(slave->dev) &&
3992                     (slave->link == BOND_LINK_UP) &&
3993                     (slave->state == BOND_STATE_ACTIVE)) {
3994                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3995
3996                         write_lock(&bond->curr_slave_lock);
3997                         bond->curr_active_slave = slave->next;
3998                         write_unlock(&bond->curr_slave_lock);
3999
4000                         break;
4001                 }
4002         }
4003
4004
4005 out:
4006         if (res) {
4007                 /* no suitable interface, frame not sent */
4008                 dev_kfree_skb(skb);
4009         }
4010         read_unlock(&bond->lock);
4011         return 0;
4012 }
4013
4014 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
4015                                         struct bonding *bond,
4016                                         struct slave *slave)
4017 {
4018         struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
4019         struct ethhdr *eth_data;
4020         u8 *hwaddr;
4021         int res;
4022
4023         if (!skb2) {
4024                 printk(KERN_ERR DRV_NAME ": Error: "
4025                        "bond_activebackup_xmit_copy(): skb_copy() failed\n");
4026                 return;
4027         }
4028
4029         skb2->mac.raw = (unsigned char *)skb2->data;
4030         eth_data = eth_hdr(skb2);
4031
4032         /* Pick an appropriate source MAC address
4033          *      -- use slave's perm MAC addr, unless used by bond
4034          *      -- otherwise, borrow active slave's perm MAC addr
4035          *         since that will not be used
4036          */
4037         hwaddr = slave->perm_hwaddr;
4038         if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
4039                 hwaddr = bond->curr_active_slave->perm_hwaddr;
4040
4041         /* Set source MAC address appropriately */
4042         memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
4043
4044         res = bond_dev_queue_xmit(bond, skb2, slave->dev);
4045         if (res)
4046                 dev_kfree_skb(skb2);
4047
4048         return;
4049 }
4050
4051 /*
4052  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4053  * the bond has a usable interface.
4054  */
4055 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4056 {
4057         struct bonding *bond = bond_dev->priv;
4058         int res = 1;
4059
4060         read_lock(&bond->lock);
4061         read_lock(&bond->curr_slave_lock);
4062
4063         if (!BOND_IS_OK(bond)) {
4064                 goto out;
4065         }
4066
4067         if (!bond->curr_active_slave)
4068                 goto out;
4069
4070         /* Xmit IGMP frames on all slaves to ensure rapid fail-over
4071            for multicast traffic on snooping switches */
4072         if (skb->protocol == __constant_htons(ETH_P_IP) &&
4073             skb->nh.iph->protocol == IPPROTO_IGMP) {
4074                 struct slave *slave, *active_slave;
4075                 int i;
4076
4077                 active_slave = bond->curr_active_slave;
4078                 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
4079                                             active_slave->prev)
4080                         if (IS_UP(slave->dev) &&
4081                             (slave->link == BOND_LINK_UP))
4082                                 bond_activebackup_xmit_copy(skb, bond, slave);
4083         }
4084
4085         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4086
4087 out:
4088         if (res) {
4089                 /* no suitable interface, frame not sent */
4090                 dev_kfree_skb(skb);
4091         }
4092         read_unlock(&bond->curr_slave_lock);
4093         read_unlock(&bond->lock);
4094         return 0;
4095 }
4096
4097 /*
4098  * In bond_xmit_xor() , we determine the output device by using a pre-
4099  * determined xmit_hash_policy(), If the selected device is not enabled,
4100  * find the next active slave.
4101  */
4102 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4103 {
4104         struct bonding *bond = bond_dev->priv;
4105         struct slave *slave, *start_at;
4106         int slave_no;
4107         int i;
4108         int res = 1;
4109
4110         read_lock(&bond->lock);
4111
4112         if (!BOND_IS_OK(bond)) {
4113                 goto out;
4114         }
4115
4116         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4117
4118         bond_for_each_slave(bond, slave, i) {
4119                 slave_no--;
4120                 if (slave_no < 0) {
4121                         break;
4122                 }
4123         }
4124
4125         start_at = slave;
4126
4127         bond_for_each_slave_from(bond, slave, i, start_at) {
4128                 if (IS_UP(slave->dev) &&
4129                     (slave->link == BOND_LINK_UP) &&
4130                     (slave->state == BOND_STATE_ACTIVE)) {
4131                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4132                         break;
4133                 }
4134         }
4135
4136 out:
4137         if (res) {
4138                 /* no suitable interface, frame not sent */
4139                 dev_kfree_skb(skb);
4140         }
4141         read_unlock(&bond->lock);
4142         return 0;
4143 }
4144
4145 /*
4146  * in broadcast mode, we send everything to all usable interfaces.
4147  */
4148 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4149 {
4150         struct bonding *bond = bond_dev->priv;
4151         struct slave *slave, *start_at;
4152         struct net_device *tx_dev = NULL;
4153         int i;
4154         int res = 1;
4155
4156         read_lock(&bond->lock);
4157
4158         if (!BOND_IS_OK(bond)) {
4159                 goto out;
4160         }
4161
4162         read_lock(&bond->curr_slave_lock);
4163         start_at = bond->curr_active_slave;
4164         read_unlock(&bond->curr_slave_lock);
4165
4166         if (!start_at) {
4167                 goto out;
4168         }
4169
4170         bond_for_each_slave_from(bond, slave, i, start_at) {
4171                 if (IS_UP(slave->dev) &&
4172                     (slave->link == BOND_LINK_UP) &&
4173                     (slave->state == BOND_STATE_ACTIVE)) {
4174                         if (tx_dev) {
4175                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4176                                 if (!skb2) {
4177                                         printk(KERN_ERR DRV_NAME
4178                                                ": %s: Error: bond_xmit_broadcast(): "
4179                                                "skb_clone() failed\n",
4180                                                bond_dev->name);
4181                                         continue;
4182                                 }
4183
4184                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4185                                 if (res) {
4186                                         dev_kfree_skb(skb2);
4187                                         continue;
4188                                 }
4189                         }
4190                         tx_dev = slave->dev;
4191                 }
4192         }
4193
4194         if (tx_dev) {
4195                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4196         }
4197
4198 out:
4199         if (res) {
4200                 /* no suitable interface, frame not sent */
4201                 dev_kfree_skb(skb);
4202         }
4203         /* frame sent to all suitable interfaces */
4204         read_unlock(&bond->lock);
4205         return 0;
4206 }
4207
4208 /*------------------------- Device initialization ---------------------------*/
4209
4210 /*
4211  * set bond mode specific net device operations
4212  */
4213 void bond_set_mode_ops(struct bonding *bond, int mode)
4214 {
4215         struct net_device *bond_dev = bond->dev;
4216
4217         switch (mode) {
4218         case BOND_MODE_ROUNDROBIN:
4219                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4220                 break;
4221         case BOND_MODE_ACTIVEBACKUP:
4222                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4223                 break;
4224         case BOND_MODE_XOR:
4225                 bond_dev->hard_start_xmit = bond_xmit_xor;
4226                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4227                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4228                 else
4229                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4230                 break;
4231         case BOND_MODE_BROADCAST:
4232                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4233                 break;
4234         case BOND_MODE_8023AD:
4235                 bond_set_master_3ad_flags(bond);
4236                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4237                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4238                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4239                 else
4240                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4241                 break;
4242         case BOND_MODE_ALB:
4243                 bond_set_master_alb_flags(bond);
4244                 /* FALLTHRU */
4245         case BOND_MODE_TLB:
4246                 bond_dev->hard_start_xmit = bond_alb_xmit;
4247                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4248                 break;
4249         default:
4250                 /* Should never happen, mode already checked */
4251                 printk(KERN_ERR DRV_NAME
4252                        ": %s: Error: Unknown bonding mode %d\n",
4253                        bond_dev->name,
4254                        mode);
4255                 break;
4256         }
4257 }
4258
4259 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4260                                     struct ethtool_drvinfo *drvinfo)
4261 {
4262         strncpy(drvinfo->driver, DRV_NAME, 32);
4263         strncpy(drvinfo->version, DRV_VERSION, 32);
4264         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4265 }
4266
4267 static const struct ethtool_ops bond_ethtool_ops = {
4268         .get_tx_csum            = ethtool_op_get_tx_csum,
4269         .get_tso                = ethtool_op_get_tso,
4270         .get_ufo                = ethtool_op_get_ufo,
4271         .get_sg                 = ethtool_op_get_sg,
4272         .get_drvinfo            = bond_ethtool_get_drvinfo,
4273 };
4274
4275 /*
4276  * Does not allocate but creates a /proc entry.
4277  * Allowed to fail.
4278  */
4279 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4280 {
4281         struct bonding *bond = bond_dev->priv;
4282
4283         dprintk("Begin bond_init for %s\n", bond_dev->name);
4284
4285         /* initialize rwlocks */
4286         rwlock_init(&bond->lock);
4287         rwlock_init(&bond->curr_slave_lock);
4288
4289         bond->params = *params; /* copy params struct */
4290
4291         /* Initialize pointers */
4292         bond->first_slave = NULL;
4293         bond->curr_active_slave = NULL;
4294         bond->current_arp_slave = NULL;
4295         bond->primary_slave = NULL;
4296         bond->dev = bond_dev;
4297         INIT_LIST_HEAD(&bond->vlan_list);
4298
4299         /* Initialize the device entry points */
4300         bond_dev->open = bond_open;
4301         bond_dev->stop = bond_close;
4302         bond_dev->get_stats = bond_get_stats;
4303         bond_dev->do_ioctl = bond_do_ioctl;
4304         bond_dev->ethtool_ops = &bond_ethtool_ops;
4305         bond_dev->set_multicast_list = bond_set_multicast_list;
4306         bond_dev->change_mtu = bond_change_mtu;
4307         bond_dev->set_mac_address = bond_set_mac_address;
4308
4309         bond_set_mode_ops(bond, bond->params.mode);
4310
4311         bond_dev->destructor = free_netdev;
4312
4313         /* Initialize the device options */
4314         bond_dev->tx_queue_len = 0;
4315         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4316         bond_dev->priv_flags |= IFF_BONDING;
4317
4318         /* At first, we block adding VLANs. That's the only way to
4319          * prevent problems that occur when adding VLANs over an
4320          * empty bond. The block will be removed once non-challenged
4321          * slaves are enslaved.
4322          */
4323         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4324
4325         /* don't acquire bond device's netif_tx_lock when
4326          * transmitting */
4327         bond_dev->features |= NETIF_F_LLTX;
4328
4329         /* By default, we declare the bond to be fully
4330          * VLAN hardware accelerated capable. Special
4331          * care is taken in the various xmit functions
4332          * when there are slaves that are not hw accel
4333          * capable
4334          */
4335         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4336         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4337         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4338         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4339                                NETIF_F_HW_VLAN_RX |
4340                                NETIF_F_HW_VLAN_FILTER);
4341
4342 #ifdef CONFIG_PROC_FS
4343         bond_create_proc_entry(bond);
4344 #endif
4345
4346         list_add_tail(&bond->bond_list, &bond_dev_list);
4347
4348         return 0;
4349 }
4350
4351 /* De-initialize device specific data.
4352  * Caller must hold rtnl_lock.
4353  */
4354 void bond_deinit(struct net_device *bond_dev)
4355 {
4356         struct bonding *bond = bond_dev->priv;
4357
4358         list_del(&bond->bond_list);
4359
4360 #ifdef CONFIG_PROC_FS
4361         bond_remove_proc_entry(bond);
4362 #endif
4363 }
4364
4365 /* Unregister and free all bond devices.
4366  * Caller must hold rtnl_lock.
4367  */
4368 static void bond_free_all(void)
4369 {
4370         struct bonding *bond, *nxt;
4371
4372         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4373                 struct net_device *bond_dev = bond->dev;
4374
4375                 bond_mc_list_destroy(bond);
4376                 /* Release the bonded slaves */
4377                 bond_release_all(bond_dev);
4378                 unregister_netdevice(bond_dev);
4379                 bond_deinit(bond_dev);
4380         }
4381
4382 #ifdef CONFIG_PROC_FS
4383         bond_destroy_proc_dir();
4384 #endif
4385 }
4386
4387 /*------------------------- Module initialization ---------------------------*/
4388
4389 /*
4390  * Convert string input module parms.  Accept either the
4391  * number of the mode or its string name.
4392  */
4393 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4394 {
4395         int i;
4396
4397         for (i = 0; tbl[i].modename; i++) {
4398                 if ((isdigit(*mode_arg) &&
4399                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4400                     (strncmp(mode_arg, tbl[i].modename,
4401                              strlen(tbl[i].modename)) == 0)) {
4402                         return tbl[i].mode;
4403                 }
4404         }
4405
4406         return -1;
4407 }
4408
4409 static int bond_check_params(struct bond_params *params)
4410 {
4411         int arp_validate_value;
4412
4413         /*
4414          * Convert string parameters.
4415          */
4416         if (mode) {
4417                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4418                 if (bond_mode == -1) {
4419                         printk(KERN_ERR DRV_NAME
4420                                ": Error: Invalid bonding mode \"%s\"\n",
4421                                mode == NULL ? "NULL" : mode);
4422                         return -EINVAL;
4423                 }
4424         }
4425
4426         if (xmit_hash_policy) {
4427                 if ((bond_mode != BOND_MODE_XOR) &&
4428                     (bond_mode != BOND_MODE_8023AD)) {
4429                         printk(KERN_INFO DRV_NAME
4430                                ": xor_mode param is irrelevant in mode %s\n",
4431                                bond_mode_name(bond_mode));
4432                 } else {
4433                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4434                                                         xmit_hashtype_tbl);
4435                         if (xmit_hashtype == -1) {
4436                                 printk(KERN_ERR DRV_NAME
4437                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4438                                 xmit_hash_policy == NULL ? "NULL" :
4439                                        xmit_hash_policy);
4440                                 return -EINVAL;
4441                         }
4442                 }
4443         }
4444
4445         if (lacp_rate) {
4446                 if (bond_mode != BOND_MODE_8023AD) {
4447                         printk(KERN_INFO DRV_NAME
4448                                ": lacp_rate param is irrelevant in mode %s\n",
4449                                bond_mode_name(bond_mode));
4450                 } else {
4451                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4452                         if (lacp_fast == -1) {
4453                                 printk(KERN_ERR DRV_NAME
4454                                        ": Error: Invalid lacp rate \"%s\"\n",
4455                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4456                                 return -EINVAL;
4457                         }
4458                 }
4459         }
4460
4461         if (max_bonds < 1 || max_bonds > INT_MAX) {
4462                 printk(KERN_WARNING DRV_NAME
4463                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4464                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4465                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4466                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4467         }
4468
4469         if (miimon < 0) {
4470                 printk(KERN_WARNING DRV_NAME
4471                        ": Warning: miimon module parameter (%d), "
4472                        "not in range 0-%d, so it was reset to %d\n",
4473                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4474                 miimon = BOND_LINK_MON_INTERV;
4475         }
4476
4477         if (updelay < 0) {
4478                 printk(KERN_WARNING DRV_NAME
4479                        ": Warning: updelay module parameter (%d), "
4480                        "not in range 0-%d, so it was reset to 0\n",
4481                        updelay, INT_MAX);
4482                 updelay = 0;
4483         }
4484
4485         if (downdelay < 0) {
4486                 printk(KERN_WARNING DRV_NAME
4487                        ": Warning: downdelay module parameter (%d), "
4488                        "not in range 0-%d, so it was reset to 0\n",
4489                        downdelay, INT_MAX);
4490                 downdelay = 0;
4491         }
4492
4493         if ((use_carrier != 0) && (use_carrier != 1)) {
4494                 printk(KERN_WARNING DRV_NAME
4495                        ": Warning: use_carrier module parameter (%d), "
4496                        "not of valid value (0/1), so it was set to 1\n",
4497                        use_carrier);
4498                 use_carrier = 1;
4499         }
4500
4501         /* reset values for 802.3ad */
4502         if (bond_mode == BOND_MODE_8023AD) {
4503                 if (!miimon) {
4504                         printk(KERN_WARNING DRV_NAME
4505                                ": Warning: miimon must be specified, "
4506                                "otherwise bonding will not detect link "
4507                                "failure, speed and duplex which are "
4508                                "essential for 802.3ad operation\n");
4509                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4510                         miimon = 100;
4511                 }
4512         }
4513
4514         /* reset values for TLB/ALB */
4515         if ((bond_mode == BOND_MODE_TLB) ||
4516             (bond_mode == BOND_MODE_ALB)) {
4517                 if (!miimon) {
4518                         printk(KERN_WARNING DRV_NAME
4519                                ": Warning: miimon must be specified, "
4520                                "otherwise bonding will not detect link "
4521                                "failure and link speed which are essential "
4522                                "for TLB/ALB load balancing\n");
4523                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4524                         miimon = 100;
4525                 }
4526         }
4527
4528         if (bond_mode == BOND_MODE_ALB) {
4529                 printk(KERN_NOTICE DRV_NAME
4530                        ": In ALB mode you might experience client "
4531                        "disconnections upon reconnection of a link if the "
4532                        "bonding module updelay parameter (%d msec) is "
4533                        "incompatible with the forwarding delay time of the "
4534                        "switch\n",
4535                        updelay);
4536         }
4537
4538         if (!miimon) {
4539                 if (updelay || downdelay) {
4540                         /* just warn the user the up/down delay will have
4541                          * no effect since miimon is zero...
4542                          */
4543                         printk(KERN_WARNING DRV_NAME
4544                                ": Warning: miimon module parameter not set "
4545                                "and updelay (%d) or downdelay (%d) module "
4546                                "parameter is set; updelay and downdelay have "
4547                                "no effect unless miimon is set\n",
4548                                updelay, downdelay);
4549                 }
4550         } else {
4551                 /* don't allow arp monitoring */
4552                 if (arp_interval) {
4553                         printk(KERN_WARNING DRV_NAME
4554                                ": Warning: miimon (%d) and arp_interval (%d) "
4555                                "can't be used simultaneously, disabling ARP "
4556                                "monitoring\n",
4557                                miimon, arp_interval);
4558                         arp_interval = 0;
4559                 }
4560
4561                 if ((updelay % miimon) != 0) {
4562                         printk(KERN_WARNING DRV_NAME
4563                                ": Warning: updelay (%d) is not a multiple "
4564                                "of miimon (%d), updelay rounded to %d ms\n",
4565                                updelay, miimon, (updelay / miimon) * miimon);
4566                 }
4567
4568                 updelay /= miimon;
4569
4570                 if ((downdelay % miimon) != 0) {
4571                         printk(KERN_WARNING DRV_NAME
4572                                ": Warning: downdelay (%d) is not a multiple "
4573                                "of miimon (%d), downdelay rounded to %d ms\n",
4574                                downdelay, miimon,
4575                                (downdelay / miimon) * miimon);
4576                 }
4577
4578                 downdelay /= miimon;
4579         }
4580
4581         if (arp_interval < 0) {
4582                 printk(KERN_WARNING DRV_NAME
4583                        ": Warning: arp_interval module parameter (%d) "
4584                        ", not in range 0-%d, so it was reset to %d\n",
4585                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4586                 arp_interval = BOND_LINK_ARP_INTERV;
4587         }
4588
4589         for (arp_ip_count = 0;
4590              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4591              arp_ip_count++) {
4592                 /* not complete check, but should be good enough to
4593                    catch mistakes */
4594                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4595                         printk(KERN_WARNING DRV_NAME
4596                                ": Warning: bad arp_ip_target module parameter "
4597                                "(%s), ARP monitoring will not be performed\n",
4598                                arp_ip_target[arp_ip_count]);
4599                         arp_interval = 0;
4600                 } else {
4601                         u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4602                         arp_target[arp_ip_count] = ip;
4603                 }
4604         }
4605
4606         if (arp_interval && !arp_ip_count) {
4607                 /* don't allow arping if no arp_ip_target given... */
4608                 printk(KERN_WARNING DRV_NAME
4609                        ": Warning: arp_interval module parameter (%d) "
4610                        "specified without providing an arp_ip_target "
4611                        "parameter, arp_interval was reset to 0\n",
4612                        arp_interval);
4613                 arp_interval = 0;
4614         }
4615
4616         if (arp_validate) {
4617                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4618                         printk(KERN_ERR DRV_NAME
4619                ": arp_validate only supported in active-backup mode\n");
4620                         return -EINVAL;
4621                 }
4622                 if (!arp_interval) {
4623                         printk(KERN_ERR DRV_NAME
4624                                ": arp_validate requires arp_interval\n");
4625                         return -EINVAL;
4626                 }
4627
4628                 arp_validate_value = bond_parse_parm(arp_validate,
4629                                                      arp_validate_tbl);
4630                 if (arp_validate_value == -1) {
4631                         printk(KERN_ERR DRV_NAME
4632                                ": Error: invalid arp_validate \"%s\"\n",
4633                                arp_validate == NULL ? "NULL" : arp_validate);
4634                         return -EINVAL;
4635                 }
4636         } else
4637                 arp_validate_value = 0;
4638
4639         if (miimon) {
4640                 printk(KERN_INFO DRV_NAME
4641                        ": MII link monitoring set to %d ms\n",
4642                        miimon);
4643         } else if (arp_interval) {
4644                 int i;
4645
4646                 printk(KERN_INFO DRV_NAME
4647                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4648                        arp_interval,
4649                        arp_validate_tbl[arp_validate_value].modename,
4650                        arp_ip_count);
4651
4652                 for (i = 0; i < arp_ip_count; i++)
4653                         printk (" %s", arp_ip_target[i]);
4654
4655                 printk("\n");
4656
4657         } else {
4658                 /* miimon and arp_interval not set, we need one so things
4659                  * work as expected, see bonding.txt for details
4660                  */
4661                 printk(KERN_WARNING DRV_NAME
4662                        ": Warning: either miimon or arp_interval and "
4663                        "arp_ip_target module parameters must be specified, "
4664                        "otherwise bonding will not detect link failures! see "
4665                        "bonding.txt for details.\n");
4666         }
4667
4668         if (primary && !USES_PRIMARY(bond_mode)) {
4669                 /* currently, using a primary only makes sense
4670                  * in active backup, TLB or ALB modes
4671                  */
4672                 printk(KERN_WARNING DRV_NAME
4673                        ": Warning: %s primary device specified but has no "
4674                        "effect in %s mode\n",
4675                        primary, bond_mode_name(bond_mode));
4676                 primary = NULL;
4677         }
4678
4679         /* fill params struct with the proper values */
4680         params->mode = bond_mode;
4681         params->xmit_policy = xmit_hashtype;
4682         params->miimon = miimon;
4683         params->arp_interval = arp_interval;
4684         params->arp_validate = arp_validate_value;
4685         params->updelay = updelay;
4686         params->downdelay = downdelay;
4687         params->use_carrier = use_carrier;
4688         params->lacp_fast = lacp_fast;
4689         params->primary[0] = 0;
4690
4691         if (primary) {
4692                 strncpy(params->primary, primary, IFNAMSIZ);
4693                 params->primary[IFNAMSIZ - 1] = 0;
4694         }
4695
4696         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4697
4698         return 0;
4699 }
4700
4701 static struct lock_class_key bonding_netdev_xmit_lock_key;
4702
4703 /* Create a new bond based on the specified name and bonding parameters.
4704  * If name is NULL, obtain a suitable "bond%d" name for us.
4705  * Caller must NOT hold rtnl_lock; we need to release it here before we
4706  * set up our sysfs entries.
4707  */
4708 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4709 {
4710         struct net_device *bond_dev;
4711         int res;
4712
4713         rtnl_lock();
4714         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4715                                 ether_setup);
4716         if (!bond_dev) {
4717                 printk(KERN_ERR DRV_NAME
4718                        ": %s: eek! can't alloc netdev!\n",
4719                        name);
4720                 res = -ENOMEM;
4721                 goto out_rtnl;
4722         }
4723
4724         if (!name) {
4725                 res = dev_alloc_name(bond_dev, "bond%d");
4726                 if (res < 0)
4727                         goto out_netdev;
4728         }
4729
4730         /* bond_init() must be called after dev_alloc_name() (for the
4731          * /proc files), but before register_netdevice(), because we
4732          * need to set function pointers.
4733          */
4734
4735         res = bond_init(bond_dev, params);
4736         if (res < 0) {
4737                 goto out_netdev;
4738         }
4739
4740         SET_MODULE_OWNER(bond_dev);
4741
4742         res = register_netdevice(bond_dev);
4743         if (res < 0) {
4744                 goto out_bond;
4745         }
4746
4747         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4748
4749         if (newbond)
4750                 *newbond = bond_dev->priv;
4751
4752         netif_carrier_off(bond_dev);
4753
4754         rtnl_unlock(); /* allows sysfs registration of net device */
4755         res = bond_create_sysfs_entry(bond_dev->priv);
4756         if (res < 0) {
4757                 rtnl_lock();
4758                 goto out_bond;
4759         }
4760
4761         return 0;
4762
4763 out_bond:
4764         bond_deinit(bond_dev);
4765 out_netdev:
4766         free_netdev(bond_dev);
4767 out_rtnl:
4768         rtnl_unlock();
4769         return res;
4770 }
4771
4772 static int __init bonding_init(void)
4773 {
4774         int i;
4775         int res;
4776
4777         printk(KERN_INFO "%s", version);
4778
4779         res = bond_check_params(&bonding_defaults);
4780         if (res) {
4781                 goto out;
4782         }
4783
4784 #ifdef CONFIG_PROC_FS
4785         bond_create_proc_dir();
4786 #endif
4787         for (i = 0; i < max_bonds; i++) {
4788                 res = bond_create(NULL, &bonding_defaults, NULL);
4789                 if (res)
4790                         goto err;
4791         }
4792
4793         res = bond_create_sysfs();
4794         if (res)
4795                 goto err;
4796
4797         register_netdevice_notifier(&bond_netdev_notifier);
4798         register_inetaddr_notifier(&bond_inetaddr_notifier);
4799
4800         goto out;
4801 err:
4802         rtnl_lock();
4803         bond_free_all();
4804         bond_destroy_sysfs();
4805         rtnl_unlock();
4806 out:
4807         return res;
4808
4809 }
4810
4811 static void __exit bonding_exit(void)
4812 {
4813         unregister_netdevice_notifier(&bond_netdev_notifier);
4814         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4815
4816         rtnl_lock();
4817         bond_free_all();
4818         bond_destroy_sysfs();
4819         rtnl_unlock();
4820 }
4821
4822 module_init(bonding_init);
4823 module_exit(bonding_exit);
4824 MODULE_LICENSE("GPL");
4825 MODULE_VERSION(DRV_VERSION);
4826 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4827 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4828 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4829
4830 /*
4831  * Local variables:
4832  *  c-indent-level: 8
4833  *  c-basic-offset: 8
4834  *  tab-width: 8
4835  * End:
4836  */
4837