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 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <asm/system.h>
  55 #include <asm/io.h>
  56 #include <asm/dma.h>
  57 #include <asm/uaccess.h>
  58 #include <linux/errno.h>
  59 #include <linux/netdevice.h>
  60 #include <linux/inetdevice.h>
  61 #include <linux/igmp.h>
  62 #include <linux/etherdevice.h>
  63 #include <linux/skbuff.h>
  64 #include <net/sock.h>
  65 #include <linux/rtnetlink.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include "bonding.h"
  79 #include "bond_3ad.h"
  80 #include "bond_alb.h"
  81
  82 /*---------------------------- Module parameters ----------------------------*/
  83
  84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  85 #define BOND_LINK_MON_INTERV    0
  86 #define BOND_LINK_ARP_INTERV    0
  87
  88 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  89 static int num_grat_arp = 1;
  90 static int num_unsol_na = 1;
  91 static int miimon       = BOND_LINK_MON_INTERV;
  92 static int updelay      = 0;
  93 static int downdelay    = 0;
  94 static int use_carrier  = 1;
  95 static char *mode       = NULL;
  96 static char *primary    = NULL;
  97 static char *lacp_rate  = NULL;
  98 static char *ad_select  = NULL;
  99 static char *xmit_hash_policy = NULL;
 100 static int arp_interval = BOND_LINK_ARP_INTERV;
 101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 102 static char *arp_validate = NULL;
 103 static char *fail_over_mac = NULL;
 104 struct bond_params bonding_defaults;
 105
 106 module_param(max_bonds, int, 0);
 107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 108 module_param(num_grat_arp, int, 0644);
 109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 110 module_param(num_unsol_na, int, 0644);
 111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 112 module_param(miimon, int, 0);
 113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 114 module_param(updelay, int, 0);
 115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 116 module_param(downdelay, int, 0);
 117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 118                             "in milliseconds");
 119 module_param(use_carrier, int, 0);
 120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 121                               "0 for off, 1 for on (default)");
 122 module_param(mode, charp, 0);
 123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 124                        "1 for active-backup, 2 for balance-xor, "
 125                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 126                        "6 for balance-alb");
 127 module_param(primary, charp, 0);
 128 MODULE_PARM_DESC(primary, "Primary network device to use");
 129 module_param(lacp_rate, charp, 0);
 130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 131                             "(slow/fast)");
 132 module_param(ad_select, charp, 0);
 133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 134 module_param(xmit_hash_policy, charp, 0);
 135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 136                                    ", 1 for layer 3+4");
 137 module_param(arp_interval, int, 0);
 138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 139 module_param_array(arp_ip_target, charp, NULL, 0);
 140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 141 module_param(arp_validate, charp, 0);
 142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 143 module_param(fail_over_mac, charp, 0);
 144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 145
 146 /*----------------------------- Global variables ----------------------------*/
 147
 148 static const char * const version =
 149         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 150
 151 LIST_HEAD(bond_dev_list);
 152
 153 #ifdef CONFIG_PROC_FS
 154 static struct proc_dir_entry *bond_proc_dir = NULL;
 155 #endif
 156
 157 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 158 static int arp_ip_count = 0;
 159 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 160 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 161 static int lacp_fast    = 0;
 162
 163
 164 const struct bond_parm_tbl bond_lacp_tbl[] = {
 165 {       "slow",         AD_LACP_SLOW},
 166 {       "fast",         AD_LACP_FAST},
 167 {       NULL,           -1},
 168 };
 169
 170 const struct bond_parm_tbl bond_mode_tbl[] = {
 171 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 172 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 173 {       "balance-xor",          BOND_MODE_XOR},
 174 {       "broadcast",            BOND_MODE_BROADCAST},
 175 {       "802.3ad",              BOND_MODE_8023AD},
 176 {       "balance-tlb",          BOND_MODE_TLB},
 177 {       "balance-alb",          BOND_MODE_ALB},
 178 {       NULL,                   -1},
 179 };
 180
 181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 182 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 183 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 184 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 185 {       NULL,                   -1},
 186 };
 187
 188 const struct bond_parm_tbl arp_validate_tbl[] = {
 189 {       "none",                 BOND_ARP_VALIDATE_NONE},
 190 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 191 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 192 {       "all",                  BOND_ARP_VALIDATE_ALL},
 193 {       NULL,                   -1},
 194 };
 195
 196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 197 {       "none",                 BOND_FOM_NONE},
 198 {       "active",               BOND_FOM_ACTIVE},
 199 {       "follow",               BOND_FOM_FOLLOW},
 200 {       NULL,                   -1},
 201 };
 202
 203 struct bond_parm_tbl ad_select_tbl[] = {
 204 {       "stable",       BOND_AD_STABLE},
 205 {       "bandwidth",    BOND_AD_BANDWIDTH},
 206 {       "count",        BOND_AD_COUNT},
 207 {       NULL,           -1},
 208 };
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static void bond_send_gratuitous_arp(struct bonding *bond);
 213 static void bond_deinit(struct net_device *bond_dev);
 214
 215 /*---------------------------- General routines -----------------------------*/
 216
 217 static const char *bond_mode_name(int mode)
 218 {
 219         static const char *names[] = {
 220                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 221                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 222                 [BOND_MODE_XOR] = "load balancing (xor)",
 223                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 224                 [BOND_MODE_8023AD]= "IEEE 802.3ad Dynamic link aggregation",
 225                 [BOND_MODE_TLB] = "transmit load balancing",
 226                 [BOND_MODE_ALB] = "adaptive load balancing",
 227         };
 228
 229         if (mode < 0 || mode > BOND_MODE_ALB)
 230                 return "unknown";
 231
 232         return names[mode];
 233 }
 234
 235 /*---------------------------------- VLAN -----------------------------------*/
 236
 237 /**
 238  * bond_add_vlan - add a new vlan id on bond
 239  * @bond: bond that got the notification
 240  * @vlan_id: the vlan id to add
 241  *
 242  * Returns -ENOMEM if allocation failed.
 243  */
 244 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 245 {
 246         struct vlan_entry *vlan;
 247
 248         pr_debug("bond: %s, vlan id %d\n",
 249                 (bond ? bond->dev->name: "None"), vlan_id);
 250
 251         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 252         if (!vlan) {
 253                 return -ENOMEM;
 254         }
 255
 256         INIT_LIST_HEAD(&vlan->vlan_list);
 257         vlan->vlan_id = vlan_id;
 258
 259         write_lock_bh(&bond->lock);
 260
 261         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 262
 263         write_unlock_bh(&bond->lock);
 264
 265         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 266
 267         return 0;
 268 }
 269
 270 /**
 271  * bond_del_vlan - delete a vlan id from bond
 272  * @bond: bond that got the notification
 273  * @vlan_id: the vlan id to delete
 274  *
 275  * returns -ENODEV if @vlan_id was not found in @bond.
 276  */
 277 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 278 {
 279         struct vlan_entry *vlan;
 280         int res = -ENODEV;
 281
 282         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 283
 284         write_lock_bh(&bond->lock);
 285
 286         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 287                 if (vlan->vlan_id == vlan_id) {
 288                         list_del(&vlan->vlan_list);
 289
 290                         if (bond_is_lb(bond))
 291                                 bond_alb_clear_vlan(bond, vlan_id);
 292
 293                         pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
 294                                 bond->dev->name);
 295
 296                         kfree(vlan);
 297
 298                         if (list_empty(&bond->vlan_list) &&
 299                             (bond->slave_cnt == 0)) {
 300                                 /* Last VLAN removed and no slaves, so
 301                                  * restore block on adding VLANs. This will
 302                                  * be removed once new slaves that are not
 303                                  * VLAN challenged will be added.
 304                                  */
 305                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 306                         }
 307
 308                         res = 0;
 309                         goto out;
 310                 }
 311         }
 312
 313         pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 314                 bond->dev->name);
 315
 316 out:
 317         write_unlock_bh(&bond->lock);
 318         return res;
 319 }
 320
 321 /**
 322  * bond_has_challenged_slaves
 323  * @bond: the bond we're working on
 324  *
 325  * Searches the slave list. Returns 1 if a vlan challenged slave
 326  * was found, 0 otherwise.
 327  *
 328  * Assumes bond->lock is held.
 329  */
 330 static int bond_has_challenged_slaves(struct bonding *bond)
 331 {
 332         struct slave *slave;
 333         int i;
 334
 335         bond_for_each_slave(bond, slave, i) {
 336                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 337                         pr_debug("found VLAN challenged slave - %s\n",
 338                                 slave->dev->name);
 339                         return 1;
 340                 }
 341         }
 342
 343         pr_debug("no VLAN challenged slaves found\n");
 344         return 0;
 345 }
 346
 347 /**
 348  * bond_next_vlan - safely skip to the next item in the vlans list.
 349  * @bond: the bond we're working on
 350  * @curr: item we're advancing from
 351  *
 352  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 353  * or @curr->next otherwise (even if it is @curr itself again).
 354  *
 355  * Caller must hold bond->lock
 356  */
 357 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 358 {
 359         struct vlan_entry *next, *last;
 360
 361         if (list_empty(&bond->vlan_list)) {
 362                 return NULL;
 363         }
 364
 365         if (!curr) {
 366                 next = list_entry(bond->vlan_list.next,
 367                                   struct vlan_entry, vlan_list);
 368         } else {
 369                 last = list_entry(bond->vlan_list.prev,
 370                                   struct vlan_entry, vlan_list);
 371                 if (last == curr) {
 372                         next = list_entry(bond->vlan_list.next,
 373                                           struct vlan_entry, vlan_list);
 374                 } else {
 375                         next = list_entry(curr->vlan_list.next,
 376                                           struct vlan_entry, vlan_list);
 377                 }
 378         }
 379
 380         return next;
 381 }
 382
 383 /**
 384  * bond_dev_queue_xmit - Prepare skb for xmit.
 385  *
 386  * @bond: bond device that got this skb for tx.
 387  * @skb: hw accel VLAN tagged skb to transmit
 388  * @slave_dev: slave that is supposed to xmit this skbuff
 389  *
 390  * When the bond gets an skb to transmit that is
 391  * already hardware accelerated VLAN tagged, and it
 392  * needs to relay this skb to a slave that is not
 393  * hw accel capable, the skb needs to be "unaccelerated",
 394  * i.e. strip the hwaccel tag and re-insert it as part
 395  * of the payload.
 396  */
 397 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 398 {
 399         unsigned short uninitialized_var(vlan_id);
 400
 401         if (!list_empty(&bond->vlan_list) &&
 402             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 403             vlan_get_tag(skb, &vlan_id) == 0) {
 404                 skb->dev = slave_dev;
 405                 skb = vlan_put_tag(skb, vlan_id);
 406                 if (!skb) {
 407                         /* vlan_put_tag() frees the skb in case of error,
 408                          * so return success here so the calling functions
 409                          * won't attempt to free is again.
 410                          */
 411                         return 0;
 412                 }
 413         } else {
 414                 skb->dev = slave_dev;
 415         }
 416
 417         skb->priority = 1;
 418         dev_queue_xmit(skb);
 419
 420         return 0;
 421 }
 422
 423 /*
 424  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 425  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 426  * lock because:
 427  * a. This operation is performed in IOCTL context,
 428  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 429  * c. Holding a lock with BH disabled while directly calling a base driver
 430  *    entry point is generally a BAD idea.
 431  *
 432  * The design of synchronization/protection for this operation in the 8021q
 433  * module is good for one or more VLAN devices over a single physical device
 434  * and cannot be extended for a teaming solution like bonding, so there is a
 435  * potential race condition here where a net device from the vlan group might
 436  * be referenced (either by a base driver or the 8021q code) while it is being
 437  * removed from the system. However, it turns out we're not making matters
 438  * worse, and if it works for regular VLAN usage it will work here too.
 439 */
 440
 441 /**
 442  * bond_vlan_rx_register - Propagates registration to slaves
 443  * @bond_dev: bonding net device that got called
 444  * @grp: vlan group being registered
 445  */
 446 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 447 {
 448         struct bonding *bond = netdev_priv(bond_dev);
 449         struct slave *slave;
 450         int i;
 451
 452         bond->vlgrp = grp;
 453
 454         bond_for_each_slave(bond, slave, i) {
 455                 struct net_device *slave_dev = slave->dev;
 456                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 457
 458                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 459                     slave_ops->ndo_vlan_rx_register) {
 460                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 461                 }
 462         }
 463 }
 464
 465 /**
 466  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 467  * @bond_dev: bonding net device that got called
 468  * @vid: vlan id being added
 469  */
 470 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 471 {
 472         struct bonding *bond = netdev_priv(bond_dev);
 473         struct slave *slave;
 474         int i, res;
 475
 476         bond_for_each_slave(bond, slave, i) {
 477                 struct net_device *slave_dev = slave->dev;
 478                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 479
 480                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 481                     slave_ops->ndo_vlan_rx_add_vid) {
 482                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 483                 }
 484         }
 485
 486         res = bond_add_vlan(bond, vid);
 487         if (res) {
 488                 printk(KERN_ERR DRV_NAME
 489                        ": %s: Error: Failed to add vlan id %d\n",
 490                        bond_dev->name, vid);
 491         }
 492 }
 493
 494 /**
 495  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 496  * @bond_dev: bonding net device that got called
 497  * @vid: vlan id being removed
 498  */
 499 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 500 {
 501         struct bonding *bond = netdev_priv(bond_dev);
 502         struct slave *slave;
 503         struct net_device *vlan_dev;
 504         int i, res;
 505
 506         bond_for_each_slave(bond, slave, i) {
 507                 struct net_device *slave_dev = slave->dev;
 508                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 509
 510                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 511                     slave_ops->ndo_vlan_rx_kill_vid) {
 512                         /* Save and then restore vlan_dev in the grp array,
 513                          * since the slave's driver might clear it.
 514                          */
 515                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 516                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 517                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 518                 }
 519         }
 520
 521         res = bond_del_vlan(bond, vid);
 522         if (res) {
 523                 printk(KERN_ERR DRV_NAME
 524                        ": %s: Error: Failed to remove vlan id %d\n",
 525                        bond_dev->name, vid);
 526         }
 527 }
 528
 529 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 530 {
 531         struct vlan_entry *vlan;
 532         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 533
 534         write_lock_bh(&bond->lock);
 535
 536         if (list_empty(&bond->vlan_list))
 537                 goto out;
 538
 539         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 540             slave_ops->ndo_vlan_rx_register)
 541                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 542
 543         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 544             !(slave_ops->ndo_vlan_rx_add_vid))
 545                 goto out;
 546
 547         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 548                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 549
 550 out:
 551         write_unlock_bh(&bond->lock);
 552 }
 553
 554 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 555 {
 556         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 557         struct vlan_entry *vlan;
 558         struct net_device *vlan_dev;
 559
 560         write_lock_bh(&bond->lock);
 561
 562         if (list_empty(&bond->vlan_list))
 563                 goto out;
 564
 565         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 566             !(slave_ops->ndo_vlan_rx_kill_vid))
 567                 goto unreg;
 568
 569         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 570                 /* Save and then restore vlan_dev in the grp array,
 571                  * since the slave's driver might clear it.
 572                  */
 573                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 574                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 575                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 576         }
 577
 578 unreg:
 579         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 580             slave_ops->ndo_vlan_rx_register)
 581                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 582
 583 out:
 584         write_unlock_bh(&bond->lock);
 585 }
 586
 587 /*------------------------------- Link status -------------------------------*/
 588
 589 /*
 590  * Set the carrier state for the master according to the state of its
 591  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 592  * do special 802.3ad magic.
 593  *
 594  * Returns zero if carrier state does not change, nonzero if it does.
 595  */
 596 static int bond_set_carrier(struct bonding *bond)
 597 {
 598         struct slave *slave;
 599         int i;
 600
 601         if (bond->slave_cnt == 0)
 602                 goto down;
 603
 604         if (bond->params.mode == BOND_MODE_8023AD)
 605                 return bond_3ad_set_carrier(bond);
 606
 607         bond_for_each_slave(bond, slave, i) {
 608                 if (slave->link == BOND_LINK_UP) {
 609                         if (!netif_carrier_ok(bond->dev)) {
 610                                 netif_carrier_on(bond->dev);
 611                                 return 1;
 612                         }
 613                         return 0;
 614                 }
 615         }
 616
 617 down:
 618         if (netif_carrier_ok(bond->dev)) {
 619                 netif_carrier_off(bond->dev);
 620                 return 1;
 621         }
 622         return 0;
 623 }
 624
 625 /*
 626  * Get link speed and duplex from the slave's base driver
 627  * using ethtool. If for some reason the call fails or the
 628  * values are invalid, fake speed and duplex to 100/Full
 629  * and return error.
 630  */
 631 static int bond_update_speed_duplex(struct slave *slave)
 632 {
 633         struct net_device *slave_dev = slave->dev;
 634         struct ethtool_cmd etool;
 635         int res;
 636
 637         /* Fake speed and duplex */
 638         slave->speed = SPEED_100;
 639         slave->duplex = DUPLEX_FULL;
 640
 641         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 642                 return -1;
 643
 644         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 645         if (res < 0)
 646                 return -1;
 647
 648         switch (etool.speed) {
 649         case SPEED_10:
 650         case SPEED_100:
 651         case SPEED_1000:
 652         case SPEED_10000:
 653                 break;
 654         default:
 655                 return -1;
 656         }
 657
 658         switch (etool.duplex) {
 659         case DUPLEX_FULL:
 660         case DUPLEX_HALF:
 661                 break;
 662         default:
 663                 return -1;
 664         }
 665
 666         slave->speed = etool.speed;
 667         slave->duplex = etool.duplex;
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * if <dev> supports MII link status reporting, check its link status.
 674  *
 675  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 676  * depening upon the setting of the use_carrier parameter.
 677  *
 678  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 679  * can't tell and just pretend it is), or 0, meaning that the link is
 680  * down.
 681  *
 682  * If reporting is non-zero, instead of faking link up, return -1 if
 683  * both ETHTOOL and MII ioctls fail (meaning the device does not
 684  * support them).  If use_carrier is set, return whatever it says.
 685  * It'd be nice if there was a good way to tell if a driver supports
 686  * netif_carrier, but there really isn't.
 687  */
 688 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 689 {
 690         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 691         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 692         struct ifreq ifr;
 693         struct mii_ioctl_data *mii;
 694
 695         if (bond->params.use_carrier)
 696                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 697
 698         ioctl = slave_ops->ndo_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         /*
 724          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 725          * attempt to get link status from it if the above MII ioctls fail.
 726          */
 727         if (slave_dev->ethtool_ops) {
 728                 if (slave_dev->ethtool_ops->get_link) {
 729                         u32 link;
 730
 731                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 732
 733                         return link ? BMSR_LSTATUS : 0;
 734                 }
 735         }
 736
 737         /*
 738          * If reporting, report that either there's no dev->do_ioctl,
 739          * or both SIOCGMIIREG and get_link failed (meaning that we
 740          * cannot report link status).  If not reporting, pretend
 741          * we're ok.
 742          */
 743         return (reporting ? -1 : BMSR_LSTATUS);
 744 }
 745
 746 /*----------------------------- Multicast list ------------------------------*/
 747
 748 /*
 749  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 750  */
 751 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 752 {
 753         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 754                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 755 }
 756
 757 /*
 758  * returns dmi entry if found, NULL otherwise
 759  */
 760 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 761 {
 762         struct dev_mc_list *idmi;
 763
 764         for (idmi = mc_list; idmi; idmi = idmi->next) {
 765                 if (bond_is_dmi_same(dmi, idmi)) {
 766                         return idmi;
 767                 }
 768         }
 769
 770         return NULL;
 771 }
 772
 773 /*
 774  * Push the promiscuity flag down to appropriate slaves
 775  */
 776 static int bond_set_promiscuity(struct bonding *bond, int inc)
 777 {
 778         int err = 0;
 779         if (USES_PRIMARY(bond->params.mode)) {
 780                 /* write lock already acquired */
 781                 if (bond->curr_active_slave) {
 782                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 783                                                   inc);
 784                 }
 785         } else {
 786                 struct slave *slave;
 787                 int i;
 788                 bond_for_each_slave(bond, slave, i) {
 789                         err = dev_set_promiscuity(slave->dev, inc);
 790                         if (err)
 791                                 return err;
 792                 }
 793         }
 794         return err;
 795 }
 796
 797 /*
 798  * Push the allmulti flag down to all slaves
 799  */
 800 static int bond_set_allmulti(struct bonding *bond, int inc)
 801 {
 802         int err = 0;
 803         if (USES_PRIMARY(bond->params.mode)) {
 804                 /* write lock already acquired */
 805                 if (bond->curr_active_slave) {
 806                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 807                                                inc);
 808                 }
 809         } else {
 810                 struct slave *slave;
 811                 int i;
 812                 bond_for_each_slave(bond, slave, i) {
 813                         err = dev_set_allmulti(slave->dev, inc);
 814                         if (err)
 815                                 return err;
 816                 }
 817         }
 818         return err;
 819 }
 820
 821 /*
 822  * Add a Multicast address to slaves
 823  * according to mode
 824  */
 825 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 826 {
 827         if (USES_PRIMARY(bond->params.mode)) {
 828                 /* write lock already acquired */
 829                 if (bond->curr_active_slave) {
 830                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 831                 }
 832         } else {
 833                 struct slave *slave;
 834                 int i;
 835                 bond_for_each_slave(bond, slave, i) {
 836                         dev_mc_add(slave->dev, addr, alen, 0);
 837                 }
 838         }
 839 }
 840
 841 /*
 842  * Remove a multicast address from slave
 843  * according to mode
 844  */
 845 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 846 {
 847         if (USES_PRIMARY(bond->params.mode)) {
 848                 /* write lock already acquired */
 849                 if (bond->curr_active_slave) {
 850                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 851                 }
 852         } else {
 853                 struct slave *slave;
 854                 int i;
 855                 bond_for_each_slave(bond, slave, i) {
 856                         dev_mc_delete(slave->dev, addr, alen, 0);
 857                 }
 858         }
 859 }
 860
 861
 862 /*
 863  * Retrieve the list of registered multicast addresses for the bonding
 864  * device and retransmit an IGMP JOIN request to the current active
 865  * slave.
 866  */
 867 static void bond_resend_igmp_join_requests(struct bonding *bond)
 868 {
 869         struct in_device *in_dev;
 870         struct ip_mc_list *im;
 871
 872         rcu_read_lock();
 873         in_dev = __in_dev_get_rcu(bond->dev);
 874         if (in_dev) {
 875                 for (im = in_dev->mc_list; im; im = im->next) {
 876                         ip_mc_rejoin_group(im);
 877                 }
 878         }
 879
 880         rcu_read_unlock();
 881 }
 882
 883 /*
 884  * Totally destroys the mc_list in bond
 885  */
 886 static void bond_mc_list_destroy(struct bonding *bond)
 887 {
 888         struct dev_mc_list *dmi;
 889
 890         dmi = bond->mc_list;
 891         while (dmi) {
 892                 bond->mc_list = dmi->next;
 893                 kfree(dmi);
 894                 dmi = bond->mc_list;
 895         }
 896         bond->mc_list = NULL;
 897 }
 898
 899 /*
 900  * Copy all the Multicast addresses from src to the bonding device dst
 901  */
 902 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 903                              gfp_t gfp_flag)
 904 {
 905         struct dev_mc_list *dmi, *new_dmi;
 906
 907         for (dmi = mc_list; dmi; dmi = dmi->next) {
 908                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 909
 910                 if (!new_dmi) {
 911                         /* FIXME: Potential memory leak !!! */
 912                         return -ENOMEM;
 913                 }
 914
 915                 new_dmi->next = bond->mc_list;
 916                 bond->mc_list = new_dmi;
 917                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 918                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 919                 new_dmi->dmi_users = dmi->dmi_users;
 920                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 921         }
 922
 923         return 0;
 924 }
 925
 926 /*
 927  * flush all members of flush->mc_list from device dev->mc_list
 928  */
 929 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 930 {
 931         struct bonding *bond = netdev_priv(bond_dev);
 932         struct dev_mc_list *dmi;
 933
 934         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 935                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 936         }
 937
 938         if (bond->params.mode == BOND_MODE_8023AD) {
 939                 /* del lacpdu mc addr from mc list */
 940                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 941
 942                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 943         }
 944 }
 945
 946 /*--------------------------- Active slave change ---------------------------*/
 947
 948 /*
 949  * Update the mc list and multicast-related flags for the new and
 950  * old active slaves (if any) according to the multicast mode, and
 951  * promiscuous flags unconditionally.
 952  */
 953 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 954 {
 955         struct dev_mc_list *dmi;
 956
 957         if (!USES_PRIMARY(bond->params.mode)) {
 958                 /* nothing to do -  mc list is already up-to-date on
 959                  * all slaves
 960                  */
 961                 return;
 962         }
 963
 964         if (old_active) {
 965                 if (bond->dev->flags & IFF_PROMISC) {
 966                         dev_set_promiscuity(old_active->dev, -1);
 967                 }
 968
 969                 if (bond->dev->flags & IFF_ALLMULTI) {
 970                         dev_set_allmulti(old_active->dev, -1);
 971                 }
 972
 973                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 974                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 975                 }
 976         }
 977
 978         if (new_active) {
 979                 /* FIXME: Signal errors upstream. */
 980                 if (bond->dev->flags & IFF_PROMISC) {
 981                         dev_set_promiscuity(new_active->dev, 1);
 982                 }
 983
 984                 if (bond->dev->flags & IFF_ALLMULTI) {
 985                         dev_set_allmulti(new_active->dev, 1);
 986                 }
 987
 988                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 989                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 990                 }
 991                 bond_resend_igmp_join_requests(bond);
 992         }
 993 }
 994
 995 /*
 996  * bond_do_fail_over_mac
 997  *
 998  * Perform special MAC address swapping for fail_over_mac settings
 999  *
1000  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1001  */
1002 static void bond_do_fail_over_mac(struct bonding *bond,
1003                                   struct slave *new_active,
1004                                   struct slave *old_active)
1005         __releases(&bond->curr_slave_lock)
1006         __releases(&bond->lock)
1007         __acquires(&bond->lock)
1008         __acquires(&bond->curr_slave_lock)
1009 {
1010         u8 tmp_mac[ETH_ALEN];
1011         struct sockaddr saddr;
1012         int rv;
1013
1014         switch (bond->params.fail_over_mac) {
1015         case BOND_FOM_ACTIVE:
1016                 if (new_active)
1017                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1018                                new_active->dev->addr_len);
1019                 break;
1020         case BOND_FOM_FOLLOW:
1021                 /*
1022                  * if new_active && old_active, swap them
1023                  * if just old_active, do nothing (going to no active slave)
1024                  * if just new_active, set new_active to bond's MAC
1025                  */
1026                 if (!new_active)
1027                         return;
1028
1029                 write_unlock_bh(&bond->curr_slave_lock);
1030                 read_unlock(&bond->lock);
1031
1032                 if (old_active) {
1033                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1034                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1035                                ETH_ALEN);
1036                         saddr.sa_family = new_active->dev->type;
1037                 } else {
1038                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1039                         saddr.sa_family = bond->dev->type;
1040                 }
1041
1042                 rv = dev_set_mac_address(new_active->dev, &saddr);
1043                 if (rv) {
1044                         printk(KERN_ERR DRV_NAME
1045                                ": %s: Error %d setting MAC of slave %s\n",
1046                                bond->dev->name, -rv, new_active->dev->name);
1047                         goto out;
1048                 }
1049
1050                 if (!old_active)
1051                         goto out;
1052
1053                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1054                 saddr.sa_family = old_active->dev->type;
1055
1056                 rv = dev_set_mac_address(old_active->dev, &saddr);
1057                 if (rv)
1058                         printk(KERN_ERR DRV_NAME
1059                                ": %s: Error %d setting MAC of slave %s\n",
1060                                bond->dev->name, -rv, new_active->dev->name);
1061 out:
1062                 read_lock(&bond->lock);
1063                 write_lock_bh(&bond->curr_slave_lock);
1064                 break;
1065         default:
1066                 printk(KERN_ERR DRV_NAME
1067                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1068                        bond->dev->name, bond->params.fail_over_mac);
1069                 break;
1070         }
1071
1072 }
1073
1074
1075 /**
1076  * find_best_interface - select the best available slave to be the active one
1077  * @bond: our bonding struct
1078  *
1079  * Warning: Caller must hold curr_slave_lock for writing.
1080  */
1081 static struct slave *bond_find_best_slave(struct bonding *bond)
1082 {
1083         struct slave *new_active, *old_active;
1084         struct slave *bestslave = NULL;
1085         int mintime = bond->params.updelay;
1086         int i;
1087
1088         new_active = old_active = bond->curr_active_slave;
1089
1090         if (!new_active) { /* there were no active slaves left */
1091                 if (bond->slave_cnt > 0) {  /* found one slave */
1092                         new_active = bond->first_slave;
1093                 } else {
1094                         return NULL; /* still no slave, return NULL */
1095                 }
1096         }
1097
1098         /* first try the primary link; if arping, a link must tx/rx traffic
1099          * before it can be considered the curr_active_slave - also, we would skip
1100          * slaves between the curr_active_slave and primary_slave that may be up
1101          * and able to arp
1102          */
1103         if ((bond->primary_slave) &&
1104             (!bond->params.arp_interval) &&
1105             (IS_UP(bond->primary_slave->dev))) {
1106                 new_active = bond->primary_slave;
1107         }
1108
1109         /* remember where to stop iterating over the slaves */
1110         old_active = new_active;
1111
1112         bond_for_each_slave_from(bond, new_active, i, old_active) {
1113                 if (IS_UP(new_active->dev)) {
1114                         if (new_active->link == BOND_LINK_UP) {
1115                                 return new_active;
1116                         } else if (new_active->link == BOND_LINK_BACK) {
1117                                 /* link up, but waiting for stabilization */
1118                                 if (new_active->delay < mintime) {
1119                                         mintime = new_active->delay;
1120                                         bestslave = new_active;
1121                                 }
1122                         }
1123                 }
1124         }
1125
1126         return bestslave;
1127 }
1128
1129 /**
1130  * change_active_interface - change the active slave into the specified one
1131  * @bond: our bonding struct
1132  * @new: the new slave to make the active one
1133  *
1134  * Set the new slave to the bond's settings and unset them on the old
1135  * curr_active_slave.
1136  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137  *
1138  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139  * because it is apparently the best available slave we have, even though its
1140  * updelay hasn't timed out yet.
1141  *
1142  * If new_active is not NULL, caller must hold bond->lock for read and
1143  * curr_slave_lock for write_bh.
1144  */
1145 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1146 {
1147         struct slave *old_active = bond->curr_active_slave;
1148
1149         if (old_active == new_active) {
1150                 return;
1151         }
1152
1153         if (new_active) {
1154                 new_active->jiffies = jiffies;
1155
1156                 if (new_active->link == BOND_LINK_BACK) {
1157                         if (USES_PRIMARY(bond->params.mode)) {
1158                                 printk(KERN_INFO DRV_NAME
1159                                        ": %s: making interface %s the new "
1160                                        "active one %d ms earlier.\n",
1161                                        bond->dev->name, new_active->dev->name,
1162                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1163                         }
1164
1165                         new_active->delay = 0;
1166                         new_active->link = BOND_LINK_UP;
1167
1168                         if (bond->params.mode == BOND_MODE_8023AD) {
1169                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1170                         }
1171
1172                         if (bond_is_lb(bond))
1173                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1174                 } else {
1175                         if (USES_PRIMARY(bond->params.mode)) {
1176                                 printk(KERN_INFO DRV_NAME
1177                                        ": %s: making interface %s the new "
1178                                        "active one.\n",
1179                                        bond->dev->name, new_active->dev->name);
1180                         }
1181                 }
1182         }
1183
1184         if (USES_PRIMARY(bond->params.mode)) {
1185                 bond_mc_swap(bond, new_active, old_active);
1186         }
1187
1188         if (bond_is_lb(bond)) {
1189                 bond_alb_handle_active_change(bond, new_active);
1190                 if (old_active)
1191                         bond_set_slave_inactive_flags(old_active);
1192                 if (new_active)
1193                         bond_set_slave_active_flags(new_active);
1194         } else {
1195                 bond->curr_active_slave = new_active;
1196         }
1197
1198         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1199                 if (old_active) {
1200                         bond_set_slave_inactive_flags(old_active);
1201                 }
1202
1203                 if (new_active) {
1204                         bond_set_slave_active_flags(new_active);
1205
1206                         if (bond->params.fail_over_mac)
1207                                 bond_do_fail_over_mac(bond, new_active,
1208                                                       old_active);
1209
1210                         bond->send_grat_arp = bond->params.num_grat_arp;
1211                         bond_send_gratuitous_arp(bond);
1212
1213                         bond->send_unsol_na = bond->params.num_unsol_na;
1214                         bond_send_unsolicited_na(bond);
1215
1216                         write_unlock_bh(&bond->curr_slave_lock);
1217                         read_unlock(&bond->lock);
1218
1219                         netdev_bonding_change(bond->dev);
1220
1221                         read_lock(&bond->lock);
1222                         write_lock_bh(&bond->curr_slave_lock);
1223                 }
1224         }
1225 }
1226
1227 /**
1228  * bond_select_active_slave - select a new active slave, if needed
1229  * @bond: our bonding struct
1230  *
1231  * This functions shoud be called when one of the following occurs:
1232  * - The old curr_active_slave has been released or lost its link.
1233  * - The primary_slave has got its link back.
1234  * - A slave has got its link back and there's no old curr_active_slave.
1235  *
1236  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1237  */
1238 void bond_select_active_slave(struct bonding *bond)
1239 {
1240         struct slave *best_slave;
1241         int rv;
1242
1243         best_slave = bond_find_best_slave(bond);
1244         if (best_slave != bond->curr_active_slave) {
1245                 bond_change_active_slave(bond, best_slave);
1246                 rv = bond_set_carrier(bond);
1247                 if (!rv)
1248                         return;
1249
1250                 if (netif_carrier_ok(bond->dev)) {
1251                         printk(KERN_INFO DRV_NAME
1252                                ": %s: first active interface up!\n",
1253                                bond->dev->name);
1254                 } else {
1255                         printk(KERN_INFO DRV_NAME ": %s: "
1256                                "now running without any active interface !\n",
1257                                bond->dev->name);
1258                 }
1259         }
1260 }
1261
1262 /*--------------------------- slave list handling ---------------------------*/
1263
1264 /*
1265  * This function attaches the slave to the end of list.
1266  *
1267  * bond->lock held for writing by caller.
1268  */
1269 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1270 {
1271         if (bond->first_slave == NULL) { /* attaching the first slave */
1272                 new_slave->next = new_slave;
1273                 new_slave->prev = new_slave;
1274                 bond->first_slave = new_slave;
1275         } else {
1276                 new_slave->next = bond->first_slave;
1277                 new_slave->prev = bond->first_slave->prev;
1278                 new_slave->next->prev = new_slave;
1279                 new_slave->prev->next = new_slave;
1280         }
1281
1282         bond->slave_cnt++;
1283 }
1284
1285 /*
1286  * This function detaches the slave from the list.
1287  * WARNING: no check is made to verify if the slave effectively
1288  * belongs to <bond>.
1289  * Nothing is freed on return, structures are just unchained.
1290  * If any slave pointer in bond was pointing to <slave>,
1291  * it should be changed by the calling function.
1292  *
1293  * bond->lock held for writing by caller.
1294  */
1295 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1296 {
1297         if (slave->next) {
1298                 slave->next->prev = slave->prev;
1299         }
1300
1301         if (slave->prev) {
1302                 slave->prev->next = slave->next;
1303         }
1304
1305         if (bond->first_slave == slave) { /* slave is the first slave */
1306                 if (bond->slave_cnt > 1) { /* there are more slave */
1307                         bond->first_slave = slave->next;
1308                 } else {
1309                         bond->first_slave = NULL; /* slave was the last one */
1310                 }
1311         }
1312
1313         slave->next = NULL;
1314         slave->prev = NULL;
1315         bond->slave_cnt--;
1316 }
1317
1318 /*---------------------------------- IOCTL ----------------------------------*/
1319
1320 static int bond_sethwaddr(struct net_device *bond_dev,
1321                           struct net_device *slave_dev)
1322 {
1323         pr_debug("bond_dev=%p\n", bond_dev);
1324         pr_debug("slave_dev=%p\n", slave_dev);
1325         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1326         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1327         return 0;
1328 }
1329
1330 #define BOND_VLAN_FEATURES \
1331         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1332          NETIF_F_HW_VLAN_FILTER)
1333
1334 /*
1335  * Compute the common dev->feature set available to all slaves.  Some
1336  * feature bits are managed elsewhere, so preserve those feature bits
1337  * on the master device.
1338  */
1339 static int bond_compute_features(struct bonding *bond)
1340 {
1341         struct slave *slave;
1342         struct net_device *bond_dev = bond->dev;
1343         unsigned long features = bond_dev->features;
1344         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1345                                                 bond_dev->hard_header_len);
1346         int i;
1347
1348         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1349         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1350
1351         if (!bond->first_slave)
1352                 goto done;
1353
1354         features &= ~NETIF_F_ONE_FOR_ALL;
1355
1356         bond_for_each_slave(bond, slave, i) {
1357                 features = netdev_increment_features(features,
1358                                                      slave->dev->features,
1359                                                      NETIF_F_ONE_FOR_ALL);
1360                 if (slave->dev->hard_header_len > max_hard_header_len)
1361                         max_hard_header_len = slave->dev->hard_header_len;
1362         }
1363
1364 done:
1365         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1366         bond_dev->features = netdev_fix_features(features, NULL);
1367         bond_dev->hard_header_len = max_hard_header_len;
1368
1369         return 0;
1370 }
1371
1372 static void bond_setup_by_slave(struct net_device *bond_dev,
1373                                 struct net_device *slave_dev)
1374 {
1375         struct bonding *bond = netdev_priv(bond_dev);
1376
1377         bond_dev->header_ops        = slave_dev->header_ops;
1378
1379         bond_dev->type              = slave_dev->type;
1380         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1381         bond_dev->addr_len          = slave_dev->addr_len;
1382
1383         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1384                 slave_dev->addr_len);
1385         bond->setup_by_slave = 1;
1386 }
1387
1388 /* enslave device <slave> to bond device <master> */
1389 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1390 {
1391         struct bonding *bond = netdev_priv(bond_dev);
1392         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1393         struct slave *new_slave = NULL;
1394         struct dev_mc_list *dmi;
1395         struct sockaddr addr;
1396         int link_reporting;
1397         int old_features = bond_dev->features;
1398         int res = 0;
1399
1400         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1401                 slave_ops->ndo_do_ioctl == NULL) {
1402                 printk(KERN_WARNING DRV_NAME
1403                        ": %s: Warning: no link monitoring support for %s\n",
1404                        bond_dev->name, slave_dev->name);
1405         }
1406
1407         /* bond must be initialized by bond_open() before enslaving */
1408         if (!(bond_dev->flags & IFF_UP)) {
1409                 printk(KERN_WARNING DRV_NAME
1410                         " %s: master_dev is not up in bond_enslave\n",
1411                         bond_dev->name);
1412         }
1413
1414         /* already enslaved */
1415         if (slave_dev->flags & IFF_SLAVE) {
1416                 pr_debug("Error, Device was already enslaved\n");
1417                 return -EBUSY;
1418         }
1419
1420         /* vlan challenged mutual exclusion */
1421         /* no need to lock since we're protected by rtnl_lock */
1422         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1423                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1424                 if (!list_empty(&bond->vlan_list)) {
1425                         printk(KERN_ERR DRV_NAME
1426                                ": %s: Error: cannot enslave VLAN "
1427                                "challenged slave %s on VLAN enabled "
1428                                "bond %s\n", bond_dev->name, slave_dev->name,
1429                                bond_dev->name);
1430                         return -EPERM;
1431                 } else {
1432                         printk(KERN_WARNING DRV_NAME
1433                                ": %s: Warning: enslaved VLAN challenged "
1434                                "slave %s. Adding VLANs will be blocked as "
1435                                "long as %s is part of bond %s\n",
1436                                bond_dev->name, slave_dev->name, slave_dev->name,
1437                                bond_dev->name);
1438                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1439                 }
1440         } else {
1441                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1442                 if (bond->slave_cnt == 0) {
1443                         /* First slave, and it is not VLAN challenged,
1444                          * so remove the block of adding VLANs over the bond.
1445                          */
1446                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1447                 }
1448         }
1449
1450         /*
1451          * Old ifenslave binaries are no longer supported.  These can
1452          * be identified with moderate accurary by the state of the slave:
1453          * the current ifenslave will set the interface down prior to
1454          * enslaving it; the old ifenslave will not.
1455          */
1456         if ((slave_dev->flags & IFF_UP)) {
1457                 printk(KERN_ERR DRV_NAME ": %s is up. "
1458                        "This may be due to an out of date ifenslave.\n",
1459                        slave_dev->name);
1460                 res = -EPERM;
1461                 goto err_undo_flags;
1462         }
1463
1464         /* set bonding device ether type by slave - bonding netdevices are
1465          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466          * there is a need to override some of the type dependent attribs/funcs.
1467          *
1468          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470          */
1471         if (bond->slave_cnt == 0) {
1472                 if (slave_dev->type != ARPHRD_ETHER)
1473                         bond_setup_by_slave(bond_dev, slave_dev);
1474         } else if (bond_dev->type != slave_dev->type) {
1475                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1476                         "from other slaves (%d), can not enslave it.\n",
1477                         slave_dev->name,
1478                         slave_dev->type, bond_dev->type);
1479                         res = -EINVAL;
1480                         goto err_undo_flags;
1481         }
1482
1483         if (slave_ops->ndo_set_mac_address == NULL) {
1484                 if (bond->slave_cnt == 0) {
1485                         printk(KERN_WARNING DRV_NAME
1486                                ": %s: Warning: The first slave device "
1487                                "specified does not support setting the MAC "
1488                                "address. Setting fail_over_mac to active.",
1489                                bond_dev->name);
1490                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1491                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1492                         printk(KERN_ERR DRV_NAME
1493                                 ": %s: Error: The slave device specified "
1494                                 "does not support setting the MAC address, "
1495                                 "but fail_over_mac is not set to active.\n"
1496                                 , bond_dev->name);
1497                         res = -EOPNOTSUPP;
1498                         goto err_undo_flags;
1499                 }
1500         }
1501
1502         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1503         if (!new_slave) {
1504                 res = -ENOMEM;
1505                 goto err_undo_flags;
1506         }
1507
1508         /* save slave's original flags before calling
1509          * netdev_set_master and dev_open
1510          */
1511         new_slave->original_flags = slave_dev->flags;
1512
1513         /*
1514          * Save slave's original ("permanent") mac address for modes
1515          * that need it, and for restoring it upon release, and then
1516          * set it to the master's address
1517          */
1518         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1519
1520         if (!bond->params.fail_over_mac) {
1521                 /*
1522                  * Set slave to master's mac address.  The application already
1523                  * set the master's mac address to that of the first slave
1524                  */
1525                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1526                 addr.sa_family = slave_dev->type;
1527                 res = dev_set_mac_address(slave_dev, &addr);
1528                 if (res) {
1529                         pr_debug("Error %d calling set_mac_address\n", res);
1530                         goto err_free;
1531                 }
1532         }
1533
1534         res = netdev_set_master(slave_dev, bond_dev);
1535         if (res) {
1536                 pr_debug("Error %d calling netdev_set_master\n", res);
1537                 goto err_restore_mac;
1538         }
1539         /* open the slave since the application closed it */
1540         res = dev_open(slave_dev);
1541         if (res) {
1542                 pr_debug("Openning slave %s failed\n", slave_dev->name);
1543                 goto err_unset_master;
1544         }
1545
1546         new_slave->dev = slave_dev;
1547         slave_dev->priv_flags |= IFF_BONDING;
1548
1549         if (bond_is_lb(bond)) {
1550                 /* bond_alb_init_slave() must be called before all other stages since
1551                  * it might fail and we do not want to have to undo everything
1552                  */
1553                 res = bond_alb_init_slave(bond, new_slave);
1554                 if (res) {
1555                         goto err_close;
1556                 }
1557         }
1558
1559         /* If the mode USES_PRIMARY, then the new slave gets the
1560          * master's promisc (and mc) settings only if it becomes the
1561          * curr_active_slave, and that is taken care of later when calling
1562          * bond_change_active()
1563          */
1564         if (!USES_PRIMARY(bond->params.mode)) {
1565                 /* set promiscuity level to new slave */
1566                 if (bond_dev->flags & IFF_PROMISC) {
1567                         res = dev_set_promiscuity(slave_dev, 1);
1568                         if (res)
1569                                 goto err_close;
1570                 }
1571
1572                 /* set allmulti level to new slave */
1573                 if (bond_dev->flags & IFF_ALLMULTI) {
1574                         res = dev_set_allmulti(slave_dev, 1);
1575                         if (res)
1576                                 goto err_close;
1577                 }
1578
1579                 netif_addr_lock_bh(bond_dev);
1580                 /* upload master's mc_list to new slave */
1581                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1582                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1583                 }
1584                 netif_addr_unlock_bh(bond_dev);
1585         }
1586
1587         if (bond->params.mode == BOND_MODE_8023AD) {
1588                 /* add lacpdu mc addr to mc list */
1589                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1590
1591                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1592         }
1593
1594         bond_add_vlans_on_slave(bond, slave_dev);
1595
1596         write_lock_bh(&bond->lock);
1597
1598         bond_attach_slave(bond, new_slave);
1599
1600         new_slave->delay = 0;
1601         new_slave->link_failure_count = 0;
1602
1603         bond_compute_features(bond);
1604
1605         write_unlock_bh(&bond->lock);
1606
1607         read_lock(&bond->lock);
1608
1609         new_slave->last_arp_rx = jiffies;
1610
1611         if (bond->params.miimon && !bond->params.use_carrier) {
1612                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1613
1614                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1615                         /*
1616                          * miimon is set but a bonded network driver
1617                          * does not support ETHTOOL/MII and
1618                          * arp_interval is not set.  Note: if
1619                          * use_carrier is enabled, we will never go
1620                          * here (because netif_carrier is always
1621                          * supported); thus, we don't need to change
1622                          * the messages for netif_carrier.
1623                          */
1624                         printk(KERN_WARNING DRV_NAME
1625                                ": %s: Warning: MII and ETHTOOL support not "
1626                                "available for interface %s, and "
1627                                "arp_interval/arp_ip_target module parameters "
1628                                "not specified, thus bonding will not detect "
1629                                "link failures! see bonding.txt for details.\n",
1630                                bond_dev->name, slave_dev->name);
1631                 } else if (link_reporting == -1) {
1632                         /* unable get link status using mii/ethtool */
1633                         printk(KERN_WARNING DRV_NAME
1634                                ": %s: Warning: can't get link status from "
1635                                "interface %s; the network driver associated "
1636                                "with this interface does not support MII or "
1637                                "ETHTOOL link status reporting, thus miimon "
1638                                "has no effect on this interface.\n",
1639                                bond_dev->name, slave_dev->name);
1640                 }
1641         }
1642
1643         /* check for initial state */
1644         if (!bond->params.miimon ||
1645             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1646                 if (bond->params.updelay) {
1647                         pr_debug("Initial state of slave_dev is "
1648                                 "BOND_LINK_BACK\n");
1649                         new_slave->link  = BOND_LINK_BACK;
1650                         new_slave->delay = bond->params.updelay;
1651                 } else {
1652                         pr_debug("Initial state of slave_dev is "
1653                                 "BOND_LINK_UP\n");
1654                         new_slave->link  = BOND_LINK_UP;
1655                 }
1656                 new_slave->jiffies = jiffies;
1657         } else {
1658                 pr_debug("Initial state of slave_dev is "
1659                         "BOND_LINK_DOWN\n");
1660                 new_slave->link  = BOND_LINK_DOWN;
1661         }
1662
1663         if (bond_update_speed_duplex(new_slave) &&
1664             (new_slave->link != BOND_LINK_DOWN)) {
1665                 printk(KERN_WARNING DRV_NAME
1666                        ": %s: Warning: failed to get speed and duplex from %s, "
1667                        "assumed to be 100Mb/sec and Full.\n",
1668                        bond_dev->name, new_slave->dev->name);
1669
1670                 if (bond->params.mode == BOND_MODE_8023AD) {
1671                         printk(KERN_WARNING DRV_NAME
1672                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1673                                "support in base driver for proper aggregator "
1674                                "selection.\n", bond_dev->name);
1675                 }
1676         }
1677
1678         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1679                 /* if there is a primary slave, remember it */
1680                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1681                         bond->primary_slave = new_slave;
1682                 }
1683         }
1684
1685         write_lock_bh(&bond->curr_slave_lock);
1686
1687         switch (bond->params.mode) {
1688         case BOND_MODE_ACTIVEBACKUP:
1689                 bond_set_slave_inactive_flags(new_slave);
1690                 bond_select_active_slave(bond);
1691                 break;
1692         case BOND_MODE_8023AD:
1693                 /* in 802.3ad mode, the internal mechanism
1694                  * will activate the slaves in the selected
1695                  * aggregator
1696                  */
1697                 bond_set_slave_inactive_flags(new_slave);
1698                 /* if this is the first slave */
1699                 if (bond->slave_cnt == 1) {
1700                         SLAVE_AD_INFO(new_slave).id = 1;
1701                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1702                          * can be called only after the mac address of the bond is set
1703                          */
1704                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1705                                             bond->params.lacp_fast);
1706                 } else {
1707                         SLAVE_AD_INFO(new_slave).id =
1708                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1709                 }
1710
1711                 bond_3ad_bind_slave(new_slave);
1712                 break;
1713         case BOND_MODE_TLB:
1714         case BOND_MODE_ALB:
1715                 new_slave->state = BOND_STATE_ACTIVE;
1716                 bond_set_slave_inactive_flags(new_slave);
1717                 bond_select_active_slave(bond);
1718                 break;
1719         default:
1720                 pr_debug("This slave is always active in trunk mode\n");
1721
1722                 /* always active in trunk mode */
1723                 new_slave->state = BOND_STATE_ACTIVE;
1724
1725                 /* In trunking mode there is little meaning to curr_active_slave
1726                  * anyway (it holds no special properties of the bond device),
1727                  * so we can change it without calling change_active_interface()
1728                  */
1729                 if (!bond->curr_active_slave) {
1730                         bond->curr_active_slave = new_slave;
1731                 }
1732                 break;
1733         } /* switch(bond_mode) */
1734
1735         write_unlock_bh(&bond->curr_slave_lock);
1736
1737         bond_set_carrier(bond);
1738
1739         read_unlock(&bond->lock);
1740
1741         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1742         if (res)
1743                 goto err_close;
1744
1745         printk(KERN_INFO DRV_NAME
1746                ": %s: enslaving %s as a%s interface with a%s link.\n",
1747                bond_dev->name, slave_dev->name,
1748                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1749                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1750
1751         /* enslave is successful */
1752         return 0;
1753
1754 /* Undo stages on error */
1755 err_close:
1756         dev_close(slave_dev);
1757
1758 err_unset_master:
1759         netdev_set_master(slave_dev, NULL);
1760
1761 err_restore_mac:
1762         if (!bond->params.fail_over_mac) {
1763                 /* XXX TODO - fom follow mode needs to change master's
1764                  * MAC if this slave's MAC is in use by the bond, or at
1765                  * least print a warning.
1766                  */
1767                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1768                 addr.sa_family = slave_dev->type;
1769                 dev_set_mac_address(slave_dev, &addr);
1770         }
1771
1772 err_free:
1773         kfree(new_slave);
1774
1775 err_undo_flags:
1776         bond_dev->features = old_features;
1777
1778         return res;
1779 }
1780
1781 /*
1782  * Try to release the slave device <slave> from the bond device <master>
1783  * It is legal to access curr_active_slave without a lock because all the function
1784  * is write-locked.
1785  *
1786  * The rules for slave state should be:
1787  *   for Active/Backup:
1788  *     Active stays on all backups go down
1789  *   for Bonded connections:
1790  *     The first up interface should be left on and all others downed.
1791  */
1792 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1793 {
1794         struct bonding *bond = netdev_priv(bond_dev);
1795         struct slave *slave, *oldcurrent;
1796         struct sockaddr addr;
1797         int mac_addr_differ;
1798
1799         /* slave is not a slave or master is not master of this slave */
1800         if (!(slave_dev->flags & IFF_SLAVE) ||
1801             (slave_dev->master != bond_dev)) {
1802                 printk(KERN_ERR DRV_NAME
1803                        ": %s: Error: cannot release %s.\n",
1804                        bond_dev->name, slave_dev->name);
1805                 return -EINVAL;
1806         }
1807
1808         write_lock_bh(&bond->lock);
1809
1810         slave = bond_get_slave_by_dev(bond, slave_dev);
1811         if (!slave) {
1812                 /* not a slave of this bond */
1813                 printk(KERN_INFO DRV_NAME
1814                        ": %s: %s not enslaved\n",
1815                        bond_dev->name, slave_dev->name);
1816                 write_unlock_bh(&bond->lock);
1817                 return -EINVAL;
1818         }
1819
1820         if (!bond->params.fail_over_mac) {
1821                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1822                                          ETH_ALEN);
1823                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1824                         printk(KERN_WARNING DRV_NAME
1825                                ": %s: Warning: the permanent HWaddr of %s - "
1826                                "%pM - is still in use by %s. "
1827                                "Set the HWaddr of %s to a different address "
1828                                "to avoid conflicts.\n",
1829                                bond_dev->name, slave_dev->name,
1830                                slave->perm_hwaddr,
1831                                bond_dev->name, slave_dev->name);
1832         }
1833
1834         /* Inform AD package of unbinding of slave. */
1835         if (bond->params.mode == BOND_MODE_8023AD) {
1836                 /* must be called before the slave is
1837                  * detached from the list
1838                  */
1839                 bond_3ad_unbind_slave(slave);
1840         }
1841
1842         printk(KERN_INFO DRV_NAME
1843                ": %s: releasing %s interface %s\n",
1844                bond_dev->name,
1845                (slave->state == BOND_STATE_ACTIVE)
1846                ? "active" : "backup",
1847                slave_dev->name);
1848
1849         oldcurrent = bond->curr_active_slave;
1850
1851         bond->current_arp_slave = NULL;
1852
1853         /* release the slave from its bond */
1854         bond_detach_slave(bond, slave);
1855
1856         bond_compute_features(bond);
1857
1858         if (bond->primary_slave == slave) {
1859                 bond->primary_slave = NULL;
1860         }
1861
1862         if (oldcurrent == slave) {
1863                 bond_change_active_slave(bond, NULL);
1864         }
1865
1866         if (bond_is_lb(bond)) {
1867                 /* Must be called only after the slave has been
1868                  * detached from the list and the curr_active_slave
1869                  * has been cleared (if our_slave == old_current),
1870                  * but before a new active slave is selected.
1871                  */
1872                 write_unlock_bh(&bond->lock);
1873                 bond_alb_deinit_slave(bond, slave);
1874                 write_lock_bh(&bond->lock);
1875         }
1876
1877         if (oldcurrent == slave) {
1878                 /*
1879                  * Note that we hold RTNL over this sequence, so there
1880                  * is no concern that another slave add/remove event
1881                  * will interfere.
1882                  */
1883                 write_unlock_bh(&bond->lock);
1884                 read_lock(&bond->lock);
1885                 write_lock_bh(&bond->curr_slave_lock);
1886
1887                 bond_select_active_slave(bond);
1888
1889                 write_unlock_bh(&bond->curr_slave_lock);
1890                 read_unlock(&bond->lock);
1891                 write_lock_bh(&bond->lock);
1892         }
1893
1894         if (bond->slave_cnt == 0) {
1895                 bond_set_carrier(bond);
1896
1897                 /* if the last slave was removed, zero the mac address
1898                  * of the master so it will be set by the application
1899                  * to the mac address of the first slave
1900                  */
1901                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1902
1903                 if (list_empty(&bond->vlan_list)) {
1904                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1905                 } else {
1906                         printk(KERN_WARNING DRV_NAME
1907                                ": %s: Warning: clearing HW address of %s while it "
1908                                "still has VLANs.\n",
1909                                bond_dev->name, bond_dev->name);
1910                         printk(KERN_WARNING DRV_NAME
1911                                ": %s: When re-adding slaves, make sure the bond's "
1912                                "HW address matches its VLANs'.\n",
1913                                bond_dev->name);
1914                 }
1915         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1916                    !bond_has_challenged_slaves(bond)) {
1917                 printk(KERN_INFO DRV_NAME
1918                        ": %s: last VLAN challenged slave %s "
1919                        "left bond %s. VLAN blocking is removed\n",
1920                        bond_dev->name, slave_dev->name, bond_dev->name);
1921                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1922         }
1923
1924         write_unlock_bh(&bond->lock);
1925
1926         /* must do this from outside any spinlocks */
1927         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1928
1929         bond_del_vlans_from_slave(bond, slave_dev);
1930
1931         /* If the mode USES_PRIMARY, then we should only remove its
1932          * promisc and mc settings if it was the curr_active_slave, but that was
1933          * already taken care of above when we detached the slave
1934          */
1935         if (!USES_PRIMARY(bond->params.mode)) {
1936                 /* unset promiscuity level from slave */
1937                 if (bond_dev->flags & IFF_PROMISC) {
1938                         dev_set_promiscuity(slave_dev, -1);
1939                 }
1940
1941                 /* unset allmulti level from slave */
1942                 if (bond_dev->flags & IFF_ALLMULTI) {
1943                         dev_set_allmulti(slave_dev, -1);
1944                 }
1945
1946                 /* flush master's mc_list from slave */
1947                 netif_addr_lock_bh(bond_dev);
1948                 bond_mc_list_flush(bond_dev, slave_dev);
1949                 netif_addr_unlock_bh(bond_dev);
1950         }
1951
1952         netdev_set_master(slave_dev, NULL);
1953
1954         /* close slave before restoring its mac address */
1955         dev_close(slave_dev);
1956
1957         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1958                 /* restore original ("permanent") mac address */
1959                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1960                 addr.sa_family = slave_dev->type;
1961                 dev_set_mac_address(slave_dev, &addr);
1962         }
1963
1964         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1965                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1966                                    IFF_SLAVE_NEEDARP);
1967
1968         kfree(slave);
1969
1970         return 0;  /* deletion OK */
1971 }
1972
1973 /*
1974 * Destroy a bonding device.
1975 * Must be under rtnl_lock when this function is called.
1976 */
1977 void bond_destroy(struct bonding *bond)
1978 {
1979         bond_deinit(bond->dev);
1980         bond_destroy_sysfs_entry(bond);
1981         unregister_netdevice(bond->dev);
1982 }
1983
1984 static void bond_destructor(struct net_device *bond_dev)
1985 {
1986         struct bonding *bond = netdev_priv(bond_dev);
1987
1988         if (bond->wq)
1989                 destroy_workqueue(bond->wq);
1990
1991         netif_addr_lock_bh(bond_dev);
1992         bond_mc_list_destroy(bond);
1993         netif_addr_unlock_bh(bond_dev);
1994
1995         free_netdev(bond_dev);
1996 }
1997
1998 /*
1999 * First release a slave and than destroy the bond if no more slaves iare left.
2000 * Must be under rtnl_lock when this function is called.
2001 */
2002 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2003 {
2004         struct bonding *bond = netdev_priv(bond_dev);
2005         int ret;
2006
2007         ret = bond_release(bond_dev, slave_dev);
2008         if ((ret == 0) && (bond->slave_cnt == 0)) {
2009                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2010                        bond_dev->name, bond_dev->name);
2011                 bond_destroy(bond);
2012         }
2013         return ret;
2014 }
2015
2016 /*
2017  * This function releases all slaves.
2018  */
2019 static int bond_release_all(struct net_device *bond_dev)
2020 {
2021         struct bonding *bond = netdev_priv(bond_dev);
2022         struct slave *slave;
2023         struct net_device *slave_dev;
2024         struct sockaddr addr;
2025
2026         write_lock_bh(&bond->lock);
2027
2028         netif_carrier_off(bond_dev);
2029
2030         if (bond->slave_cnt == 0) {
2031                 goto out;
2032         }
2033
2034         bond->current_arp_slave = NULL;
2035         bond->primary_slave = NULL;
2036         bond_change_active_slave(bond, NULL);
2037
2038         while ((slave = bond->first_slave) != NULL) {
2039                 /* Inform AD package of unbinding of slave
2040                  * before slave is detached from the list.
2041                  */
2042                 if (bond->params.mode == BOND_MODE_8023AD) {
2043                         bond_3ad_unbind_slave(slave);
2044                 }
2045
2046                 slave_dev = slave->dev;
2047                 bond_detach_slave(bond, slave);
2048
2049                 /* now that the slave is detached, unlock and perform
2050                  * all the undo steps that should not be called from
2051                  * within a lock.
2052                  */
2053                 write_unlock_bh(&bond->lock);
2054
2055                 if (bond_is_lb(bond)) {
2056                         /* must be called only after the slave
2057                          * has been detached from the list
2058                          */
2059                         bond_alb_deinit_slave(bond, slave);
2060                 }
2061
2062                 bond_compute_features(bond);
2063
2064                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2065                 bond_del_vlans_from_slave(bond, slave_dev);
2066
2067                 /* If the mode USES_PRIMARY, then we should only remove its
2068                  * promisc and mc settings if it was the curr_active_slave, but that was
2069                  * already taken care of above when we detached the slave
2070                  */
2071                 if (!USES_PRIMARY(bond->params.mode)) {
2072                         /* unset promiscuity level from slave */
2073                         if (bond_dev->flags & IFF_PROMISC) {
2074                                 dev_set_promiscuity(slave_dev, -1);
2075                         }
2076
2077                         /* unset allmulti level from slave */
2078                         if (bond_dev->flags & IFF_ALLMULTI) {
2079                                 dev_set_allmulti(slave_dev, -1);
2080                         }
2081
2082                         /* flush master's mc_list from slave */
2083                         netif_addr_lock_bh(bond_dev);
2084                         bond_mc_list_flush(bond_dev, slave_dev);
2085                         netif_addr_unlock_bh(bond_dev);
2086                 }
2087
2088                 netdev_set_master(slave_dev, NULL);
2089
2090                 /* close slave before restoring its mac address */
2091                 dev_close(slave_dev);
2092
2093                 if (!bond->params.fail_over_mac) {
2094                         /* restore original ("permanent") mac address*/
2095                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2096                         addr.sa_family = slave_dev->type;
2097                         dev_set_mac_address(slave_dev, &addr);
2098                 }
2099
2100                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2101                                            IFF_SLAVE_INACTIVE);
2102
2103                 kfree(slave);
2104
2105                 /* re-acquire the lock before getting the next slave */
2106                 write_lock_bh(&bond->lock);
2107         }
2108
2109         /* zero the mac address of the master so it will be
2110          * set by the application to the mac address of the
2111          * first slave
2112          */
2113         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2114
2115         if (list_empty(&bond->vlan_list)) {
2116                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2117         } else {
2118                 printk(KERN_WARNING DRV_NAME
2119                        ": %s: Warning: clearing HW address of %s while it "
2120                        "still has VLANs.\n",
2121                        bond_dev->name, bond_dev->name);
2122                 printk(KERN_WARNING DRV_NAME
2123                        ": %s: When re-adding slaves, make sure the bond's "
2124                        "HW address matches its VLANs'.\n",
2125                        bond_dev->name);
2126         }
2127
2128         printk(KERN_INFO DRV_NAME
2129                ": %s: released all slaves\n",
2130                bond_dev->name);
2131
2132 out:
2133         write_unlock_bh(&bond->lock);
2134
2135         return 0;
2136 }
2137
2138 /*
2139  * This function changes the active slave to slave <slave_dev>.
2140  * It returns -EINVAL in the following cases.
2141  *  - <slave_dev> is not found in the list.
2142  *  - There is not active slave now.
2143  *  - <slave_dev> is already active.
2144  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2145  *  - <slave_dev> is not running.
2146  * In these cases, this fuction does nothing.
2147  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2148  */
2149 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2150 {
2151         struct bonding *bond = netdev_priv(bond_dev);
2152         struct slave *old_active = NULL;
2153         struct slave *new_active = NULL;
2154         int res = 0;
2155
2156         if (!USES_PRIMARY(bond->params.mode)) {
2157                 return -EINVAL;
2158         }
2159
2160         /* Verify that master_dev is indeed the master of slave_dev */
2161         if (!(slave_dev->flags & IFF_SLAVE) ||
2162             (slave_dev->master != bond_dev)) {
2163                 return -EINVAL;
2164         }
2165
2166         read_lock(&bond->lock);
2167
2168         read_lock(&bond->curr_slave_lock);
2169         old_active = bond->curr_active_slave;
2170         read_unlock(&bond->curr_slave_lock);
2171
2172         new_active = bond_get_slave_by_dev(bond, slave_dev);
2173
2174         /*
2175          * Changing to the current active: do nothing; return success.
2176          */
2177         if (new_active && (new_active == old_active)) {
2178                 read_unlock(&bond->lock);
2179                 return 0;
2180         }
2181
2182         if ((new_active) &&
2183             (old_active) &&
2184             (new_active->link == BOND_LINK_UP) &&
2185             IS_UP(new_active->dev)) {
2186                 write_lock_bh(&bond->curr_slave_lock);
2187                 bond_change_active_slave(bond, new_active);
2188                 write_unlock_bh(&bond->curr_slave_lock);
2189         } else {
2190                 res = -EINVAL;
2191         }
2192
2193         read_unlock(&bond->lock);
2194
2195         return res;
2196 }
2197
2198 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2199 {
2200         struct bonding *bond = netdev_priv(bond_dev);
2201
2202         info->bond_mode = bond->params.mode;
2203         info->miimon = bond->params.miimon;
2204
2205         read_lock(&bond->lock);
2206         info->num_slaves = bond->slave_cnt;
2207         read_unlock(&bond->lock);
2208
2209         return 0;
2210 }
2211
2212 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2213 {
2214         struct bonding *bond = netdev_priv(bond_dev);
2215         struct slave *slave;
2216         int i, res = -ENODEV;
2217
2218         read_lock(&bond->lock);
2219
2220         bond_for_each_slave(bond, slave, i) {
2221                 if (i == (int)info->slave_id) {
2222                         res = 0;
2223                         strcpy(info->slave_name, slave->dev->name);
2224                         info->link = slave->link;
2225                         info->state = slave->state;
2226                         info->link_failure_count = slave->link_failure_count;
2227                         break;
2228                 }
2229         }
2230
2231         read_unlock(&bond->lock);
2232
2233         return res;
2234 }
2235
2236 /*-------------------------------- Monitoring -------------------------------*/
2237
2238
2239 static int bond_miimon_inspect(struct bonding *bond)
2240 {
2241         struct slave *slave;
2242         int i, link_state, commit = 0;
2243
2244         bond_for_each_slave(bond, slave, i) {
2245                 slave->new_link = BOND_LINK_NOCHANGE;
2246
2247                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2248
2249                 switch (slave->link) {
2250                 case BOND_LINK_UP:
2251                         if (link_state)
2252                                 continue;
2253
2254                         slave->link = BOND_LINK_FAIL;
2255                         slave->delay = bond->params.downdelay;
2256                         if (slave->delay) {
2257                                 printk(KERN_INFO DRV_NAME
2258                                        ": %s: link status down for %s"
2259                                        "interface %s, disabling it in %d ms.\n",
2260                                        bond->dev->name,
2261                                        (bond->params.mode ==
2262                                         BOND_MODE_ACTIVEBACKUP) ?
2263                                        ((slave->state == BOND_STATE_ACTIVE) ?
2264                                         "active " : "backup ") : "",
2265                                        slave->dev->name,
2266                                        bond->params.downdelay * bond->params.miimon);
2267                         }
2268                         /*FALLTHRU*/
2269                 case BOND_LINK_FAIL:
2270                         if (link_state) {
2271                                 /*
2272                                  * recovered before downdelay expired
2273                                  */
2274                                 slave->link = BOND_LINK_UP;
2275                                 slave->jiffies = jiffies;
2276                                 printk(KERN_INFO DRV_NAME
2277                                        ": %s: link status up again after %d "
2278                                        "ms for interface %s.\n",
2279                                        bond->dev->name,
2280                                        (bond->params.downdelay - slave->delay) *
2281                                        bond->params.miimon,
2282                                        slave->dev->name);
2283                                 continue;
2284                         }
2285
2286                         if (slave->delay <= 0) {
2287                                 slave->new_link = BOND_LINK_DOWN;
2288                                 commit++;
2289                                 continue;
2290                         }
2291
2292                         slave->delay--;
2293                         break;
2294
2295                 case BOND_LINK_DOWN:
2296                         if (!link_state)
2297                                 continue;
2298
2299                         slave->link = BOND_LINK_BACK;
2300                         slave->delay = bond->params.updelay;
2301
2302                         if (slave->delay) {
2303                                 printk(KERN_INFO DRV_NAME
2304                                        ": %s: link status up for "
2305                                        "interface %s, enabling it in %d ms.\n",
2306                                        bond->dev->name, slave->dev->name,
2307                                        bond->params.updelay *
2308                                        bond->params.miimon);
2309                         }
2310                         /*FALLTHRU*/
2311                 case BOND_LINK_BACK:
2312                         if (!link_state) {
2313                                 slave->link = BOND_LINK_DOWN;
2314                                 printk(KERN_INFO DRV_NAME
2315                                        ": %s: link status down again after %d "
2316                                        "ms for interface %s.\n",
2317                                        bond->dev->name,
2318                                        (bond->params.updelay - slave->delay) *
2319                                        bond->params.miimon,
2320                                        slave->dev->name);
2321
2322                                 continue;
2323                         }
2324
2325                         if (slave->delay <= 0) {
2326                                 slave->new_link = BOND_LINK_UP;
2327                                 commit++;
2328                                 continue;
2329                         }
2330
2331                         slave->delay--;
2332                         break;
2333                 }
2334         }
2335
2336         return commit;
2337 }
2338
2339 static void bond_miimon_commit(struct bonding *bond)
2340 {
2341         struct slave *slave;
2342         int i;
2343
2344         bond_for_each_slave(bond, slave, i) {
2345                 switch (slave->new_link) {
2346                 case BOND_LINK_NOCHANGE:
2347                         continue;
2348
2349                 case BOND_LINK_UP:
2350                         slave->link = BOND_LINK_UP;
2351                         slave->jiffies = jiffies;
2352
2353                         if (bond->params.mode == BOND_MODE_8023AD) {
2354                                 /* prevent it from being the active one */
2355                                 slave->state = BOND_STATE_BACKUP;
2356                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2357                                 /* make it immediately active */
2358                                 slave->state = BOND_STATE_ACTIVE;
2359                         } else if (slave != bond->primary_slave) {
2360                                 /* prevent it from being the active one */
2361                                 slave->state = BOND_STATE_BACKUP;
2362                         }
2363
2364                         printk(KERN_INFO DRV_NAME
2365                                ": %s: link status definitely "
2366                                "up for interface %s.\n",
2367                                bond->dev->name, slave->dev->name);
2368
2369                         /* notify ad that the link status has changed */
2370                         if (bond->params.mode == BOND_MODE_8023AD)
2371                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2372
2373                         if (bond_is_lb(bond))
2374                                 bond_alb_handle_link_change(bond, slave,
2375                                                             BOND_LINK_UP);
2376
2377                         if (!bond->curr_active_slave ||
2378                             (slave == bond->primary_slave))
2379                                 goto do_failover;
2380
2381                         continue;
2382
2383                 case BOND_LINK_DOWN:
2384                         if (slave->link_failure_count < UINT_MAX)
2385                                 slave->link_failure_count++;
2386
2387                         slave->link = BOND_LINK_DOWN;
2388
2389                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2390                             bond->params.mode == BOND_MODE_8023AD)
2391                                 bond_set_slave_inactive_flags(slave);
2392
2393                         printk(KERN_INFO DRV_NAME
2394                                ": %s: link status definitely down for "
2395                                "interface %s, disabling it\n",
2396                                bond->dev->name, slave->dev->name);
2397
2398                         if (bond->params.mode == BOND_MODE_8023AD)
2399                                 bond_3ad_handle_link_change(slave,
2400                                                             BOND_LINK_DOWN);
2401
2402                         if (bond->params.mode == BOND_MODE_TLB ||
2403                             bond->params.mode == BOND_MODE_ALB)
2404                                 bond_alb_handle_link_change(bond, slave,
2405                                                             BOND_LINK_DOWN);
2406
2407                         if (slave == bond->curr_active_slave)
2408                                 goto do_failover;
2409
2410                         continue;
2411
2412                 default:
2413                         printk(KERN_ERR DRV_NAME
2414                                ": %s: invalid new link %d on slave %s\n",
2415                                bond->dev->name, slave->new_link,
2416                                slave->dev->name);
2417                         slave->new_link = BOND_LINK_NOCHANGE;
2418
2419                         continue;
2420                 }
2421
2422 do_failover:
2423                 ASSERT_RTNL();
2424                 write_lock_bh(&bond->curr_slave_lock);
2425                 bond_select_active_slave(bond);
2426                 write_unlock_bh(&bond->curr_slave_lock);
2427         }
2428
2429         bond_set_carrier(bond);
2430 }
2431
2432 /*
2433  * bond_mii_monitor
2434  *
2435  * Really a wrapper that splits the mii monitor into two phases: an
2436  * inspection, then (if inspection indicates something needs to be done)
2437  * an acquisition of appropriate locks followed by a commit phase to
2438  * implement whatever link state changes are indicated.
2439  */
2440 void bond_mii_monitor(struct work_struct *work)
2441 {
2442         struct bonding *bond = container_of(work, struct bonding,
2443                                             mii_work.work);
2444
2445         read_lock(&bond->lock);
2446         if (bond->kill_timers)
2447                 goto out;
2448
2449         if (bond->slave_cnt == 0)
2450                 goto re_arm;
2451
2452         if (bond->send_grat_arp) {
2453                 read_lock(&bond->curr_slave_lock);
2454                 bond_send_gratuitous_arp(bond);
2455                 read_unlock(&bond->curr_slave_lock);
2456         }
2457
2458         if (bond->send_unsol_na) {
2459                 read_lock(&bond->curr_slave_lock);
2460                 bond_send_unsolicited_na(bond);
2461                 read_unlock(&bond->curr_slave_lock);
2462         }
2463
2464         if (bond_miimon_inspect(bond)) {
2465                 read_unlock(&bond->lock);
2466                 rtnl_lock();
2467                 read_lock(&bond->lock);
2468
2469                 bond_miimon_commit(bond);
2470
2471                 read_unlock(&bond->lock);
2472                 rtnl_unlock();  /* might sleep, hold no other locks */
2473                 read_lock(&bond->lock);
2474         }
2475
2476 re_arm:
2477         if (bond->params.miimon)
2478                 queue_delayed_work(bond->wq, &bond->mii_work,
2479                                    msecs_to_jiffies(bond->params.miimon));
2480 out:
2481         read_unlock(&bond->lock);
2482 }
2483
2484 static __be32 bond_glean_dev_ip(struct net_device *dev)
2485 {
2486         struct in_device *idev;
2487         struct in_ifaddr *ifa;
2488         __be32 addr = 0;
2489
2490         if (!dev)
2491                 return 0;
2492
2493         rcu_read_lock();
2494         idev = __in_dev_get_rcu(dev);
2495         if (!idev)
2496                 goto out;
2497
2498         ifa = idev->ifa_list;
2499         if (!ifa)
2500                 goto out;
2501
2502         addr = ifa->ifa_local;
2503 out:
2504         rcu_read_unlock();
2505         return addr;
2506 }
2507
2508 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2509 {
2510         struct vlan_entry *vlan;
2511
2512         if (ip == bond->master_ip)
2513                 return 1;
2514
2515         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2516                 if (ip == vlan->vlan_ip)
2517                         return 1;
2518         }
2519
2520         return 0;
2521 }
2522
2523 /*
2524  * We go to the (large) trouble of VLAN tagging ARP frames because
2525  * switches in VLAN mode (especially if ports are configured as
2526  * "native" to a VLAN) might not pass non-tagged frames.
2527  */
2528 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2529 {
2530         struct sk_buff *skb;
2531
2532         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2533                slave_dev->name, dest_ip, src_ip, vlan_id);
2534
2535         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2536                          NULL, slave_dev->dev_addr, NULL);
2537
2538         if (!skb) {
2539                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2540                 return;
2541         }
2542         if (vlan_id) {
2543                 skb = vlan_put_tag(skb, vlan_id);
2544                 if (!skb) {
2545                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2546                         return;
2547                 }
2548         }
2549         arp_xmit(skb);
2550 }
2551
2552
2553 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2554 {
2555         int i, vlan_id, rv;
2556         __be32 *targets = bond->params.arp_targets;
2557         struct vlan_entry *vlan;
2558         struct net_device *vlan_dev;
2559         struct flowi fl;
2560         struct rtable *rt;
2561
2562         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2563                 if (!targets[i])
2564                         break;
2565                 pr_debug("basa: target %x\n", targets[i]);
2566                 if (list_empty(&bond->vlan_list)) {
2567                         pr_debug("basa: empty vlan: arp_send\n");
2568                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2569                                       bond->master_ip, 0);
2570                         continue;
2571                 }
2572
2573                 /*
2574                  * If VLANs are configured, we do a route lookup to
2575                  * determine which VLAN interface would be used, so we
2576                  * can tag the ARP with the proper VLAN tag.
2577                  */
2578                 memset(&fl, 0, sizeof(fl));
2579                 fl.fl4_dst = targets[i];
2580                 fl.fl4_tos = RTO_ONLINK;
2581
2582                 rv = ip_route_output_key(&init_net, &rt, &fl);
2583                 if (rv) {
2584                         if (net_ratelimit()) {
2585                                 printk(KERN_WARNING DRV_NAME
2586                              ": %s: no route to arp_ip_target %pI4\n",
2587                                        bond->dev->name, &fl.fl4_dst);
2588                         }
2589                         continue;
2590                 }
2591
2592                 /*
2593                  * This target is not on a VLAN
2594                  */
2595                 if (rt->u.dst.dev == bond->dev) {
2596                         ip_rt_put(rt);
2597                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2598                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2599                                       bond->master_ip, 0);
2600                         continue;
2601                 }
2602
2603                 vlan_id = 0;
2604                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2605                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2606                         if (vlan_dev == rt->u.dst.dev) {
2607                                 vlan_id = vlan->vlan_id;
2608                                 pr_debug("basa: vlan match on %s %d\n",
2609                                        vlan_dev->name, vlan_id);
2610                                 break;
2611                         }
2612                 }
2613
2614                 if (vlan_id) {
2615                         ip_rt_put(rt);
2616                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617                                       vlan->vlan_ip, vlan_id);
2618                         continue;
2619                 }
2620
2621                 if (net_ratelimit()) {
2622                         printk(KERN_WARNING DRV_NAME
2623                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2624                                bond->dev->name, &fl.fl4_dst,
2625                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2626                 }
2627                 ip_rt_put(rt);
2628         }
2629 }
2630
2631 /*
2632  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2633  * for each VLAN above us.
2634  *
2635  * Caller must hold curr_slave_lock for read or better
2636  */
2637 static void bond_send_gratuitous_arp(struct bonding *bond)
2638 {
2639         struct slave *slave = bond->curr_active_slave;
2640         struct vlan_entry *vlan;
2641         struct net_device *vlan_dev;
2642
2643         pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2644                                 slave ? slave->dev->name : "NULL");
2645
2646         if (!slave || !bond->send_grat_arp ||
2647             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2648                 return;
2649
2650         bond->send_grat_arp--;
2651
2652         if (bond->master_ip) {
2653                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2654                                 bond->master_ip, 0);
2655         }
2656
2657         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2658                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2659                 if (vlan->vlan_ip) {
2660                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2661                                       vlan->vlan_ip, vlan->vlan_id);
2662                 }
2663         }
2664 }
2665
2666 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2667 {
2668         int i;
2669         __be32 *targets = bond->params.arp_targets;
2670
2671         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2672                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2673                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2674                 if (sip == targets[i]) {
2675                         if (bond_has_this_ip(bond, tip))
2676                                 slave->last_arp_rx = jiffies;
2677                         return;
2678                 }
2679         }
2680 }
2681
2682 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2683 {
2684         struct arphdr *arp;
2685         struct slave *slave;
2686         struct bonding *bond;
2687         unsigned char *arp_ptr;
2688         __be32 sip, tip;
2689
2690         if (dev_net(dev) != &init_net)
2691                 goto out;
2692
2693         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2694                 goto out;
2695
2696         bond = netdev_priv(dev);
2697         read_lock(&bond->lock);
2698
2699         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2700                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2701                 orig_dev ? orig_dev->name : "NULL");
2702
2703         slave = bond_get_slave_by_dev(bond, orig_dev);
2704         if (!slave || !slave_do_arp_validate(bond, slave))
2705                 goto out_unlock;
2706
2707         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2708                 goto out_unlock;
2709
2710         arp = arp_hdr(skb);
2711         if (arp->ar_hln != dev->addr_len ||
2712             skb->pkt_type == PACKET_OTHERHOST ||
2713             skb->pkt_type == PACKET_LOOPBACK ||
2714             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2715             arp->ar_pro != htons(ETH_P_IP) ||
2716             arp->ar_pln != 4)
2717                 goto out_unlock;
2718
2719         arp_ptr = (unsigned char *)(arp + 1);
2720         arp_ptr += dev->addr_len;
2721         memcpy(&sip, arp_ptr, 4);
2722         arp_ptr += 4 + dev->addr_len;
2723         memcpy(&tip, arp_ptr, 4);
2724
2725         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2726                 bond->dev->name, slave->dev->name, slave->state,
2727                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2728                 &sip, &tip);
2729
2730         /*
2731          * Backup slaves won't see the ARP reply, but do come through
2732          * here for each ARP probe (so we swap the sip/tip to validate
2733          * the probe).  In a "redundant switch, common router" type of
2734          * configuration, the ARP probe will (hopefully) travel from
2735          * the active, through one switch, the router, then the other
2736          * switch before reaching the backup.
2737          */
2738         if (slave->state == BOND_STATE_ACTIVE)
2739                 bond_validate_arp(bond, slave, sip, tip);
2740         else
2741                 bond_validate_arp(bond, slave, tip, sip);
2742
2743 out_unlock:
2744         read_unlock(&bond->lock);
2745 out:
2746         dev_kfree_skb(skb);
2747         return NET_RX_SUCCESS;
2748 }
2749
2750 /*
2751  * this function is called regularly to monitor each slave's link
2752  * ensuring that traffic is being sent and received when arp monitoring
2753  * is used in load-balancing mode. if the adapter has been dormant, then an
2754  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2755  * arp monitoring in active backup mode.
2756  */
2757 void bond_loadbalance_arp_mon(struct work_struct *work)
2758 {
2759         struct bonding *bond = container_of(work, struct bonding,
2760                                             arp_work.work);
2761         struct slave *slave, *oldcurrent;
2762         int do_failover = 0;
2763         int delta_in_ticks;
2764         int i;
2765
2766         read_lock(&bond->lock);
2767
2768         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2769
2770         if (bond->kill_timers) {
2771                 goto out;
2772         }
2773
2774         if (bond->slave_cnt == 0) {
2775                 goto re_arm;
2776         }
2777
2778         read_lock(&bond->curr_slave_lock);
2779         oldcurrent = bond->curr_active_slave;
2780         read_unlock(&bond->curr_slave_lock);
2781
2782         /* see if any of the previous devices are up now (i.e. they have
2783          * xmt and rcv traffic). the curr_active_slave does not come into
2784          * the picture unless it is null. also, slave->jiffies is not needed
2785          * here because we send an arp on each slave and give a slave as
2786          * long as it needs to get the tx/rx within the delta.
2787          * TODO: what about up/down delay in arp mode? it wasn't here before
2788          *       so it can wait
2789          */
2790         bond_for_each_slave(bond, slave, i) {
2791                 if (slave->link != BOND_LINK_UP) {
2792                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2793                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2794
2795                                 slave->link  = BOND_LINK_UP;
2796                                 slave->state = BOND_STATE_ACTIVE;
2797
2798                                 /* primary_slave has no meaning in round-robin
2799                                  * mode. the window of a slave being up and
2800                                  * curr_active_slave being null after enslaving
2801                                  * is closed.
2802                                  */
2803                                 if (!oldcurrent) {
2804                                         printk(KERN_INFO DRV_NAME
2805                                                ": %s: link status definitely "
2806                                                "up for interface %s, ",
2807                                                bond->dev->name,
2808                                                slave->dev->name);
2809                                         do_failover = 1;
2810                                 } else {
2811                                         printk(KERN_INFO DRV_NAME
2812                                                ": %s: interface %s is now up\n",
2813                                                bond->dev->name,
2814                                                slave->dev->name);
2815                                 }
2816                         }
2817                 } else {
2818                         /* slave->link == BOND_LINK_UP */
2819
2820                         /* not all switches will respond to an arp request
2821                          * when the source ip is 0, so don't take the link down
2822                          * if we don't know our ip yet
2823                          */
2824                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2825                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2826
2827                                 slave->link  = BOND_LINK_DOWN;
2828                                 slave->state = BOND_STATE_BACKUP;
2829
2830                                 if (slave->link_failure_count < UINT_MAX) {
2831                                         slave->link_failure_count++;
2832                                 }
2833
2834                                 printk(KERN_INFO DRV_NAME
2835                                        ": %s: interface %s is now down.\n",
2836                                        bond->dev->name,
2837                                        slave->dev->name);
2838
2839                                 if (slave == oldcurrent) {
2840                                         do_failover = 1;
2841                                 }
2842                         }
2843                 }
2844
2845                 /* note: if switch is in round-robin mode, all links
2846                  * must tx arp to ensure all links rx an arp - otherwise
2847                  * links may oscillate or not come up at all; if switch is
2848                  * in something like xor mode, there is nothing we can
2849                  * do - all replies will be rx'ed on same link causing slaves
2850                  * to be unstable during low/no traffic periods
2851                  */
2852                 if (IS_UP(slave->dev)) {
2853                         bond_arp_send_all(bond, slave);
2854                 }
2855         }
2856
2857         if (do_failover) {
2858                 write_lock_bh(&bond->curr_slave_lock);
2859
2860                 bond_select_active_slave(bond);
2861
2862                 write_unlock_bh(&bond->curr_slave_lock);
2863         }
2864
2865 re_arm:
2866         if (bond->params.arp_interval)
2867                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2868 out:
2869         read_unlock(&bond->lock);
2870 }
2871
2872 /*
2873  * Called to inspect slaves for active-backup mode ARP monitor link state
2874  * changes.  Sets new_link in slaves to specify what action should take
2875  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2876  * to link states must be committed.
2877  *
2878  * Called with bond->lock held for read.
2879  */
2880 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2881 {
2882         struct slave *slave;
2883         int i, commit = 0;
2884
2885         bond_for_each_slave(bond, slave, i) {
2886                 slave->new_link = BOND_LINK_NOCHANGE;
2887
2888                 if (slave->link != BOND_LINK_UP) {
2889                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2890                                            delta_in_ticks)) {
2891                                 slave->new_link = BOND_LINK_UP;
2892                                 commit++;
2893                         }
2894
2895                         continue;
2896                 }
2897
2898                 /*
2899                  * Give slaves 2*delta after being enslaved or made
2900                  * active.  This avoids bouncing, as the last receive
2901                  * times need a full ARP monitor cycle to be updated.
2902                  */
2903                 if (!time_after_eq(jiffies, slave->jiffies +
2904                                    2 * delta_in_ticks))
2905                         continue;
2906
2907                 /*
2908                  * Backup slave is down if:
2909                  * - No current_arp_slave AND
2910                  * - more than 3*delta since last receive AND
2911                  * - the bond has an IP address
2912                  *
2913                  * Note: a non-null current_arp_slave indicates
2914                  * the curr_active_slave went down and we are
2915                  * searching for a new one; under this condition
2916                  * we only take the curr_active_slave down - this
2917                  * gives each slave a chance to tx/rx traffic
2918                  * before being taken out
2919                  */
2920                 if (slave->state == BOND_STATE_BACKUP &&
2921                     !bond->current_arp_slave &&
2922                     time_after(jiffies, slave_last_rx(bond, slave) +
2923                                3 * delta_in_ticks)) {
2924                         slave->new_link = BOND_LINK_DOWN;
2925                         commit++;
2926                 }
2927
2928                 /*
2929                  * Active slave is down if:
2930                  * - more than 2*delta since transmitting OR
2931                  * - (more than 2*delta since receive AND
2932                  *    the bond has an IP address)
2933                  */
2934                 if ((slave->state == BOND_STATE_ACTIVE) &&
2935                     (time_after_eq(jiffies, slave->dev->trans_start +
2936                                     2 * delta_in_ticks) ||
2937                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2938                                      + 2 * delta_in_ticks)))) {
2939                         slave->new_link = BOND_LINK_DOWN;
2940                         commit++;
2941                 }
2942         }
2943
2944         read_lock(&bond->curr_slave_lock);
2945
2946         /*
2947          * Trigger a commit if the primary option setting has changed.
2948          */
2949         if (bond->primary_slave &&
2950             (bond->primary_slave != bond->curr_active_slave) &&
2951             (bond->primary_slave->link == BOND_LINK_UP))
2952                 commit++;
2953
2954         read_unlock(&bond->curr_slave_lock);
2955
2956         return commit;
2957 }
2958
2959 /*
2960  * Called to commit link state changes noted by inspection step of
2961  * active-backup mode ARP monitor.
2962  *
2963  * Called with RTNL and bond->lock for read.
2964  */
2965 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2966 {
2967         struct slave *slave;
2968         int i;
2969
2970         bond_for_each_slave(bond, slave, i) {
2971                 switch (slave->new_link) {
2972                 case BOND_LINK_NOCHANGE:
2973                         continue;
2974
2975                 case BOND_LINK_UP:
2976                         write_lock_bh(&bond->curr_slave_lock);
2977
2978                         if (!bond->curr_active_slave &&
2979                             time_before_eq(jiffies, slave->dev->trans_start +
2980                                            delta_in_ticks)) {
2981                                 slave->link = BOND_LINK_UP;
2982                                 bond_change_active_slave(bond, slave);
2983                                 bond->current_arp_slave = NULL;
2984
2985                                 printk(KERN_INFO DRV_NAME
2986                                        ": %s: %s is up and now the "
2987                                        "active interface\n",
2988                                        bond->dev->name, slave->dev->name);
2989
2990                         } else if (bond->curr_active_slave != slave) {
2991                                 /* this slave has just come up but we
2992                                  * already have a current slave; this can
2993                                  * also happen if bond_enslave adds a new
2994                                  * slave that is up while we are searching
2995                                  * for a new slave
2996                                  */
2997                                 slave->link = BOND_LINK_UP;
2998                                 bond_set_slave_inactive_flags(slave);
2999                                 bond->current_arp_slave = NULL;
3000
3001                                 printk(KERN_INFO DRV_NAME
3002                                        ": %s: backup interface %s is now up\n",
3003                                        bond->dev->name, slave->dev->name);
3004                         }
3005
3006                         write_unlock_bh(&bond->curr_slave_lock);
3007
3008                         break;
3009
3010                 case BOND_LINK_DOWN:
3011                         if (slave->link_failure_count < UINT_MAX)
3012                                 slave->link_failure_count++;
3013
3014                         slave->link = BOND_LINK_DOWN;
3015
3016                         if (slave == bond->curr_active_slave) {
3017                                 printk(KERN_INFO DRV_NAME
3018                                        ": %s: link status down for active "
3019                                        "interface %s, disabling it\n",
3020                                        bond->dev->name, slave->dev->name);
3021
3022                                 bond_set_slave_inactive_flags(slave);
3023
3024                                 write_lock_bh(&bond->curr_slave_lock);
3025
3026                                 bond_select_active_slave(bond);
3027                                 if (bond->curr_active_slave)
3028                                         bond->curr_active_slave->jiffies =
3029                                                 jiffies;
3030
3031                                 write_unlock_bh(&bond->curr_slave_lock);
3032
3033                                 bond->current_arp_slave = NULL;
3034
3035                         } else if (slave->state == BOND_STATE_BACKUP) {
3036                                 printk(KERN_INFO DRV_NAME
3037                                        ": %s: backup interface %s is now down\n",
3038                                        bond->dev->name, slave->dev->name);
3039
3040                                 bond_set_slave_inactive_flags(slave);
3041                         }
3042                         break;
3043
3044                 default:
3045                         printk(KERN_ERR DRV_NAME
3046                                ": %s: impossible: new_link %d on slave %s\n",
3047                                bond->dev->name, slave->new_link,
3048                                slave->dev->name);
3049                 }
3050         }
3051
3052         /*
3053          * No race with changes to primary via sysfs, as we hold rtnl.
3054          */
3055         if (bond->primary_slave &&
3056             (bond->primary_slave != bond->curr_active_slave) &&
3057             (bond->primary_slave->link == BOND_LINK_UP)) {
3058                 write_lock_bh(&bond->curr_slave_lock);
3059                 bond_change_active_slave(bond, bond->primary_slave);
3060                 write_unlock_bh(&bond->curr_slave_lock);
3061         }
3062
3063         bond_set_carrier(bond);
3064 }
3065
3066 /*
3067  * Send ARP probes for active-backup mode ARP monitor.
3068  *
3069  * Called with bond->lock held for read.
3070  */
3071 static void bond_ab_arp_probe(struct bonding *bond)
3072 {
3073         struct slave *slave;
3074         int i;
3075
3076         read_lock(&bond->curr_slave_lock);
3077
3078         if (bond->current_arp_slave && bond->curr_active_slave)
3079                 printk("PROBE: c_arp %s && cas %s BAD\n",
3080                        bond->current_arp_slave->dev->name,
3081                        bond->curr_active_slave->dev->name);
3082
3083         if (bond->curr_active_slave) {
3084                 bond_arp_send_all(bond, bond->curr_active_slave);
3085                 read_unlock(&bond->curr_slave_lock);
3086                 return;
3087         }
3088
3089         read_unlock(&bond->curr_slave_lock);
3090
3091         /* if we don't have a curr_active_slave, search for the next available
3092          * backup slave from the current_arp_slave and make it the candidate
3093          * for becoming the curr_active_slave
3094          */
3095
3096         if (!bond->current_arp_slave) {
3097                 bond->current_arp_slave = bond->first_slave;
3098                 if (!bond->current_arp_slave)
3099                         return;
3100         }
3101
3102         bond_set_slave_inactive_flags(bond->current_arp_slave);
3103
3104         /* search for next candidate */
3105         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3106                 if (IS_UP(slave->dev)) {
3107                         slave->link = BOND_LINK_BACK;
3108                         bond_set_slave_active_flags(slave);
3109                         bond_arp_send_all(bond, slave);
3110                         slave->jiffies = jiffies;
3111                         bond->current_arp_slave = slave;
3112                         break;
3113                 }
3114
3115                 /* if the link state is up at this point, we
3116                  * mark it down - this can happen if we have
3117                  * simultaneous link failures and
3118                  * reselect_active_interface doesn't make this
3119                  * one the current slave so it is still marked
3120                  * up when it is actually down
3121                  */
3122                 if (slave->link == BOND_LINK_UP) {
3123                         slave->link = BOND_LINK_DOWN;
3124                         if (slave->link_failure_count < UINT_MAX)
3125                                 slave->link_failure_count++;
3126
3127                         bond_set_slave_inactive_flags(slave);
3128
3129                         printk(KERN_INFO DRV_NAME
3130                                ": %s: backup interface %s is now down.\n",
3131                                bond->dev->name, slave->dev->name);
3132                 }
3133         }
3134 }
3135
3136 void bond_activebackup_arp_mon(struct work_struct *work)
3137 {
3138         struct bonding *bond = container_of(work, struct bonding,
3139                                             arp_work.work);
3140         int delta_in_ticks;
3141
3142         read_lock(&bond->lock);
3143
3144         if (bond->kill_timers)
3145                 goto out;
3146
3147         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3148
3149         if (bond->slave_cnt == 0)
3150                 goto re_arm;
3151
3152         if (bond->send_grat_arp) {
3153                 read_lock(&bond->curr_slave_lock);
3154                 bond_send_gratuitous_arp(bond);
3155                 read_unlock(&bond->curr_slave_lock);
3156         }
3157
3158         if (bond->send_unsol_na) {
3159                 read_lock(&bond->curr_slave_lock);
3160                 bond_send_unsolicited_na(bond);
3161                 read_unlock(&bond->curr_slave_lock);
3162         }
3163
3164         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3165                 read_unlock(&bond->lock);
3166                 rtnl_lock();
3167                 read_lock(&bond->lock);
3168
3169                 bond_ab_arp_commit(bond, delta_in_ticks);
3170
3171                 read_unlock(&bond->lock);
3172                 rtnl_unlock();
3173                 read_lock(&bond->lock);
3174         }
3175
3176         bond_ab_arp_probe(bond);
3177
3178 re_arm:
3179         if (bond->params.arp_interval) {
3180                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3181         }
3182 out:
3183         read_unlock(&bond->lock);
3184 }
3185
3186 /*------------------------------ proc/seq_file-------------------------------*/
3187
3188 #ifdef CONFIG_PROC_FS
3189
3190 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3191         __acquires(&dev_base_lock)
3192         __acquires(&bond->lock)
3193 {
3194         struct bonding *bond = seq->private;
3195         loff_t off = 0;
3196         struct slave *slave;
3197         int i;
3198
3199         /* make sure the bond won't be taken away */
3200         read_lock(&dev_base_lock);
3201         read_lock(&bond->lock);
3202
3203         if (*pos == 0) {
3204                 return SEQ_START_TOKEN;
3205         }
3206
3207         bond_for_each_slave(bond, slave, i) {
3208                 if (++off == *pos) {
3209                         return slave;
3210                 }
3211         }
3212
3213         return NULL;
3214 }
3215
3216 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3217 {
3218         struct bonding *bond = seq->private;
3219         struct slave *slave = v;
3220
3221         ++*pos;
3222         if (v == SEQ_START_TOKEN) {
3223                 return bond->first_slave;
3224         }
3225
3226         slave = slave->next;
3227
3228         return (slave == bond->first_slave) ? NULL : slave;
3229 }
3230
3231 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3232         __releases(&bond->lock)
3233         __releases(&dev_base_lock)
3234 {
3235         struct bonding *bond = seq->private;
3236
3237         read_unlock(&bond->lock);
3238         read_unlock(&dev_base_lock);
3239 }
3240
3241 static void bond_info_show_master(struct seq_file *seq)
3242 {
3243         struct bonding *bond = seq->private;
3244         struct slave *curr;
3245         int i;
3246
3247         read_lock(&bond->curr_slave_lock);
3248         curr = bond->curr_active_slave;
3249         read_unlock(&bond->curr_slave_lock);
3250
3251         seq_printf(seq, "Bonding Mode: %s",
3252                    bond_mode_name(bond->params.mode));
3253
3254         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3255             bond->params.fail_over_mac)
3256                 seq_printf(seq, " (fail_over_mac %s)",
3257                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3258
3259         seq_printf(seq, "\n");
3260
3261         if (bond->params.mode == BOND_MODE_XOR ||
3262                 bond->params.mode == BOND_MODE_8023AD) {
3263                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3264                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3265                         bond->params.xmit_policy);
3266         }
3267
3268         if (USES_PRIMARY(bond->params.mode)) {
3269                 seq_printf(seq, "Primary Slave: %s\n",
3270                            (bond->primary_slave) ?
3271                            bond->primary_slave->dev->name : "None");
3272
3273                 seq_printf(seq, "Currently Active Slave: %s\n",
3274                            (curr) ? curr->dev->name : "None");
3275         }
3276
3277         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3278                    "up" : "down");
3279         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3280         seq_printf(seq, "Up Delay (ms): %d\n",
3281                    bond->params.updelay * bond->params.miimon);
3282         seq_printf(seq, "Down Delay (ms): %d\n",
3283                    bond->params.downdelay * bond->params.miimon);
3284
3285
3286         /* ARP information */
3287         if(bond->params.arp_interval > 0) {
3288                 int printed=0;
3289                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3290                                 bond->params.arp_interval);
3291
3292                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3293
3294                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3295                         if (!bond->params.arp_targets[i])
3296                                 break;
3297                         if (printed)
3298                                 seq_printf(seq, ",");
3299                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3300                         printed = 1;
3301                 }
3302                 seq_printf(seq, "\n");
3303         }
3304
3305         if (bond->params.mode == BOND_MODE_8023AD) {
3306                 struct ad_info ad_info;
3307
3308                 seq_puts(seq, "\n802.3ad info\n");
3309                 seq_printf(seq, "LACP rate: %s\n",
3310                            (bond->params.lacp_fast) ? "fast" : "slow");
3311                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3312                            ad_select_tbl[bond->params.ad_select].modename);
3313
3314                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3315                         seq_printf(seq, "bond %s has no active aggregator\n",
3316                                    bond->dev->name);
3317                 } else {
3318                         seq_printf(seq, "Active Aggregator Info:\n");
3319
3320                         seq_printf(seq, "\tAggregator ID: %d\n",
3321                                    ad_info.aggregator_id);
3322                         seq_printf(seq, "\tNumber of ports: %d\n",
3323                                    ad_info.ports);
3324                         seq_printf(seq, "\tActor Key: %d\n",
3325                                    ad_info.actor_key);
3326                         seq_printf(seq, "\tPartner Key: %d\n",
3327                                    ad_info.partner_key);
3328                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3329                                    ad_info.partner_system);
3330                 }
3331         }
3332 }
3333
3334 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3335 {
3336         struct bonding *bond = seq->private;
3337
3338         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3339         seq_printf(seq, "MII Status: %s\n",
3340                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3341         seq_printf(seq, "Link Failure Count: %u\n",
3342                    slave->link_failure_count);
3343
3344         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3345
3346         if (bond->params.mode == BOND_MODE_8023AD) {
3347                 const struct aggregator *agg
3348                         = SLAVE_AD_INFO(slave).port.aggregator;
3349
3350                 if (agg) {
3351                         seq_printf(seq, "Aggregator ID: %d\n",
3352                                    agg->aggregator_identifier);
3353                 } else {
3354                         seq_puts(seq, "Aggregator ID: N/A\n");
3355                 }
3356         }
3357 }
3358
3359 static int bond_info_seq_show(struct seq_file *seq, void *v)
3360 {
3361         if (v == SEQ_START_TOKEN) {
3362                 seq_printf(seq, "%s\n", version);
3363                 bond_info_show_master(seq);
3364         } else {
3365                 bond_info_show_slave(seq, v);
3366         }
3367
3368         return 0;
3369 }
3370
3371 static const struct seq_operations bond_info_seq_ops = {
3372         .start = bond_info_seq_start,
3373         .next  = bond_info_seq_next,
3374         .stop  = bond_info_seq_stop,
3375         .show  = bond_info_seq_show,
3376 };
3377
3378 static int bond_info_open(struct inode *inode, struct file *file)
3379 {
3380         struct seq_file *seq;
3381         struct proc_dir_entry *proc;
3382         int res;
3383
3384         res = seq_open(file, &bond_info_seq_ops);
3385         if (!res) {
3386                 /* recover the pointer buried in proc_dir_entry data */
3387                 seq = file->private_data;
3388                 proc = PDE(inode);
3389                 seq->private = proc->data;
3390         }
3391
3392         return res;
3393 }
3394
3395 static const struct file_operations bond_info_fops = {
3396         .owner   = THIS_MODULE,
3397         .open    = bond_info_open,
3398         .read    = seq_read,
3399         .llseek  = seq_lseek,
3400         .release = seq_release,
3401 };
3402
3403 static int bond_create_proc_entry(struct bonding *bond)
3404 {
3405         struct net_device *bond_dev = bond->dev;
3406
3407         if (bond_proc_dir) {
3408                 bond->proc_entry = proc_create_data(bond_dev->name,
3409                                                     S_IRUGO, bond_proc_dir,
3410                                                     &bond_info_fops, bond);
3411                 if (bond->proc_entry == NULL) {
3412                         printk(KERN_WARNING DRV_NAME
3413                                ": Warning: Cannot create /proc/net/%s/%s\n",
3414                                DRV_NAME, bond_dev->name);
3415                 } else {
3416                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3417                 }
3418         }
3419
3420         return 0;
3421 }
3422
3423 static void bond_remove_proc_entry(struct bonding *bond)
3424 {
3425         if (bond_proc_dir && bond->proc_entry) {
3426                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3427                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3428                 bond->proc_entry = NULL;
3429         }
3430 }
3431
3432 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3433  * Caller must hold rtnl_lock.
3434  */
3435 static void bond_create_proc_dir(void)
3436 {
3437         if (!bond_proc_dir) {
3438                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3439                 if (!bond_proc_dir)
3440                         printk(KERN_WARNING DRV_NAME
3441                                 ": Warning: cannot create /proc/net/%s\n",
3442                                 DRV_NAME);
3443         }
3444 }
3445
3446 /* Destroy the bonding directory under /proc/net, if empty.
3447  * Caller must hold rtnl_lock.
3448  */
3449 static void bond_destroy_proc_dir(void)
3450 {
3451         if (bond_proc_dir) {
3452                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3453                 bond_proc_dir = NULL;
3454         }
3455 }
3456 #endif /* CONFIG_PROC_FS */
3457
3458 /*-------------------------- netdev event handling --------------------------*/
3459
3460 /*
3461  * Change device name
3462  */
3463 static int bond_event_changename(struct bonding *bond)
3464 {
3465 #ifdef CONFIG_PROC_FS
3466         bond_remove_proc_entry(bond);
3467         bond_create_proc_entry(bond);
3468 #endif
3469         down_write(&(bonding_rwsem));
3470         bond_destroy_sysfs_entry(bond);
3471         bond_create_sysfs_entry(bond);
3472         up_write(&(bonding_rwsem));
3473         return NOTIFY_DONE;
3474 }
3475
3476 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3477 {
3478         struct bonding *event_bond = netdev_priv(bond_dev);
3479
3480         switch (event) {
3481         case NETDEV_CHANGENAME:
3482                 return bond_event_changename(event_bond);
3483         case NETDEV_UNREGISTER:
3484                 bond_release_all(event_bond->dev);
3485                 break;
3486         default:
3487                 break;
3488         }
3489
3490         return NOTIFY_DONE;
3491 }
3492
3493 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3494 {
3495         struct net_device *bond_dev = slave_dev->master;
3496         struct bonding *bond = netdev_priv(bond_dev);
3497
3498         switch (event) {
3499         case NETDEV_UNREGISTER:
3500                 if (bond_dev) {
3501                         if (bond->setup_by_slave)
3502                                 bond_release_and_destroy(bond_dev, slave_dev);
3503                         else
3504                                 bond_release(bond_dev, slave_dev);
3505                 }
3506                 break;
3507         case NETDEV_CHANGE:
3508                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3509                         struct slave *slave;
3510
3511                         slave = bond_get_slave_by_dev(bond, slave_dev);
3512                         if (slave) {
3513                                 u16 old_speed = slave->speed;
3514                                 u16 old_duplex = slave->duplex;
3515
3516                                 bond_update_speed_duplex(slave);
3517
3518                                 if (bond_is_lb(bond))
3519                                         break;
3520
3521                                 if (old_speed != slave->speed)
3522                                         bond_3ad_adapter_speed_changed(slave);
3523                                 if (old_duplex != slave->duplex)
3524                                         bond_3ad_adapter_duplex_changed(slave);
3525                         }
3526                 }
3527
3528                 break;
3529         case NETDEV_DOWN:
3530                 /*
3531                  * ... Or is it this?
3532                  */
3533                 break;
3534         case NETDEV_CHANGEMTU:
3535                 /*
3536                  * TODO: Should slaves be allowed to
3537                  * independently alter their MTU?  For
3538                  * an active-backup bond, slaves need
3539                  * not be the same type of device, so
3540                  * MTUs may vary.  For other modes,
3541                  * slaves arguably should have the
3542                  * same MTUs. To do this, we'd need to
3543                  * take over the slave's change_mtu
3544                  * function for the duration of their
3545                  * servitude.
3546                  */
3547                 break;
3548         case NETDEV_CHANGENAME:
3549                 /*
3550                  * TODO: handle changing the primary's name
3551                  */
3552                 break;
3553         case NETDEV_FEAT_CHANGE:
3554                 bond_compute_features(bond);
3555                 break;
3556         default:
3557                 break;
3558         }
3559
3560         return NOTIFY_DONE;
3561 }
3562
3563 /*
3564  * bond_netdev_event: handle netdev notifier chain events.
3565  *
3566  * This function receives events for the netdev chain.  The caller (an
3567  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3568  * locks for us to safely manipulate the slave devices (RTNL lock,
3569  * dev_probe_lock).
3570  */
3571 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3572 {
3573         struct net_device *event_dev = (struct net_device *)ptr;
3574
3575         if (dev_net(event_dev) != &init_net)
3576                 return NOTIFY_DONE;
3577
3578         pr_debug("event_dev: %s, event: %lx\n",
3579                 (event_dev ? event_dev->name : "None"),
3580                 event);
3581
3582         if (!(event_dev->priv_flags & IFF_BONDING))
3583                 return NOTIFY_DONE;
3584
3585         if (event_dev->flags & IFF_MASTER) {
3586                 pr_debug("IFF_MASTER\n");
3587                 return bond_master_netdev_event(event, event_dev);
3588         }
3589
3590         if (event_dev->flags & IFF_SLAVE) {
3591                 pr_debug("IFF_SLAVE\n");
3592                 return bond_slave_netdev_event(event, event_dev);
3593         }
3594
3595         return NOTIFY_DONE;
3596 }
3597
3598 /*
3599  * bond_inetaddr_event: handle inetaddr notifier chain events.
3600  *
3601  * We keep track of device IPs primarily to use as source addresses in
3602  * ARP monitor probes (rather than spewing out broadcasts all the time).
3603  *
3604  * We track one IP for the main device (if it has one), plus one per VLAN.
3605  */
3606 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3607 {
3608         struct in_ifaddr *ifa = ptr;
3609         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3610         struct bonding *bond;
3611         struct vlan_entry *vlan;
3612
3613         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3614                 return NOTIFY_DONE;
3615
3616         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3617                 if (bond->dev == event_dev) {
3618                         switch (event) {
3619                         case NETDEV_UP:
3620                                 bond->master_ip = ifa->ifa_local;
3621                                 return NOTIFY_OK;
3622                         case NETDEV_DOWN:
3623                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3624                                 return NOTIFY_OK;
3625                         default:
3626                                 return NOTIFY_DONE;
3627                         }
3628                 }
3629
3630                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3631                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3632                         if (vlan_dev == event_dev) {
3633                                 switch (event) {
3634                                 case NETDEV_UP:
3635                                         vlan->vlan_ip = ifa->ifa_local;
3636                                         return NOTIFY_OK;
3637                                 case NETDEV_DOWN:
3638                                         vlan->vlan_ip =
3639                                                 bond_glean_dev_ip(vlan_dev);
3640                                         return NOTIFY_OK;
3641                                 default:
3642                                         return NOTIFY_DONE;
3643                                 }
3644                         }
3645                 }
3646         }
3647         return NOTIFY_DONE;
3648 }
3649
3650 static struct notifier_block bond_netdev_notifier = {
3651         .notifier_call = bond_netdev_event,
3652 };
3653
3654 static struct notifier_block bond_inetaddr_notifier = {
3655         .notifier_call = bond_inetaddr_event,
3656 };
3657
3658 /*-------------------------- Packet type handling ---------------------------*/
3659
3660 /* register to receive lacpdus on a bond */
3661 static void bond_register_lacpdu(struct bonding *bond)
3662 {
3663         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3664
3665         /* initialize packet type */
3666         pk_type->type = PKT_TYPE_LACPDU;
3667         pk_type->dev = bond->dev;
3668         pk_type->func = bond_3ad_lacpdu_recv;
3669
3670         dev_add_pack(pk_type);
3671 }
3672
3673 /* unregister to receive lacpdus on a bond */
3674 static void bond_unregister_lacpdu(struct bonding *bond)
3675 {
3676         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3677 }
3678
3679 void bond_register_arp(struct bonding *bond)
3680 {
3681         struct packet_type *pt = &bond->arp_mon_pt;
3682
3683         if (pt->type)
3684                 return;
3685
3686         pt->type = htons(ETH_P_ARP);
3687         pt->dev = bond->dev;
3688         pt->func = bond_arp_rcv;
3689         dev_add_pack(pt);
3690 }
3691
3692 void bond_unregister_arp(struct bonding *bond)
3693 {
3694         struct packet_type *pt = &bond->arp_mon_pt;
3695
3696         dev_remove_pack(pt);
3697         pt->type = 0;
3698 }
3699
3700 /*---------------------------- Hashing Policies -----------------------------*/
3701
3702 /*
3703  * Hash for the output device based upon layer 2 and layer 3 data. If
3704  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3705  */
3706 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3707                                      struct net_device *bond_dev, int count)
3708 {
3709         struct ethhdr *data = (struct ethhdr *)skb->data;
3710         struct iphdr *iph = ip_hdr(skb);
3711
3712         if (skb->protocol == htons(ETH_P_IP)) {
3713                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3714                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3715         }
3716
3717         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3718 }
3719
3720 /*
3721  * Hash for the output device based upon layer 3 and layer 4 data. If
3722  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3723  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3724  */
3725 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3726                                     struct net_device *bond_dev, int count)
3727 {
3728         struct ethhdr *data = (struct ethhdr *)skb->data;
3729         struct iphdr *iph = ip_hdr(skb);
3730         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3731         int layer4_xor = 0;
3732
3733         if (skb->protocol == htons(ETH_P_IP)) {
3734                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3735                     (iph->protocol == IPPROTO_TCP ||
3736                      iph->protocol == IPPROTO_UDP)) {
3737                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3738                 }
3739                 return (layer4_xor ^
3740                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3741
3742         }
3743
3744         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3745 }
3746
3747 /*
3748  * Hash for the output device based upon layer 2 data
3749  */
3750 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3751                                    struct net_device *bond_dev, int count)
3752 {
3753         struct ethhdr *data = (struct ethhdr *)skb->data;
3754
3755         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3756 }
3757
3758 /*-------------------------- Device entry points ----------------------------*/
3759
3760 static int bond_open(struct net_device *bond_dev)
3761 {
3762         struct bonding *bond = netdev_priv(bond_dev);
3763
3764         bond->kill_timers = 0;
3765
3766         if (bond_is_lb(bond)) {
3767                 /* bond_alb_initialize must be called before the timer
3768                  * is started.
3769                  */
3770                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3771                         /* something went wrong - fail the open operation */
3772                         return -1;
3773                 }
3774
3775                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3776                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3777         }
3778
3779         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3780                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3781                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3782         }
3783
3784         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3785                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3786                         INIT_DELAYED_WORK(&bond->arp_work,
3787                                           bond_activebackup_arp_mon);
3788                 else
3789                         INIT_DELAYED_WORK(&bond->arp_work,
3790                                           bond_loadbalance_arp_mon);
3791
3792                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3793                 if (bond->params.arp_validate)
3794                         bond_register_arp(bond);
3795         }
3796
3797         if (bond->params.mode == BOND_MODE_8023AD) {
3798                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3799                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3800                 /* register to receive LACPDUs */
3801                 bond_register_lacpdu(bond);
3802                 bond_3ad_initiate_agg_selection(bond, 1);
3803         }
3804
3805         return 0;
3806 }
3807
3808 static int bond_close(struct net_device *bond_dev)
3809 {
3810         struct bonding *bond = netdev_priv(bond_dev);
3811
3812         if (bond->params.mode == BOND_MODE_8023AD) {
3813                 /* Unregister the receive of LACPDUs */
3814                 bond_unregister_lacpdu(bond);
3815         }
3816
3817         if (bond->params.arp_validate)
3818                 bond_unregister_arp(bond);
3819
3820         write_lock_bh(&bond->lock);
3821
3822         bond->send_grat_arp = 0;
3823         bond->send_unsol_na = 0;
3824
3825         /* signal timers not to re-arm */
3826         bond->kill_timers = 1;
3827
3828         write_unlock_bh(&bond->lock);
3829
3830         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3831                 cancel_delayed_work(&bond->mii_work);
3832         }
3833
3834         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3835                 cancel_delayed_work(&bond->arp_work);
3836         }
3837
3838         switch (bond->params.mode) {
3839         case BOND_MODE_8023AD:
3840                 cancel_delayed_work(&bond->ad_work);
3841                 break;
3842         case BOND_MODE_TLB:
3843         case BOND_MODE_ALB:
3844                 cancel_delayed_work(&bond->alb_work);
3845                 break;
3846         default:
3847                 break;
3848         }
3849
3850
3851         if (bond_is_lb(bond)) {
3852                 /* Must be called only after all
3853                  * slaves have been released
3854                  */
3855                 bond_alb_deinitialize(bond);
3856         }
3857
3858         return 0;
3859 }
3860
3861 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3862 {
3863         struct bonding *bond = netdev_priv(bond_dev);
3864         struct net_device_stats *stats = &bond->stats;
3865         struct net_device_stats local_stats;
3866         struct slave *slave;
3867         int i;
3868
3869         memset(&local_stats, 0, sizeof(struct net_device_stats));
3870
3871         read_lock_bh(&bond->lock);
3872
3873         bond_for_each_slave(bond, slave, i) {
3874                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3875
3876                 local_stats.rx_packets += sstats->rx_packets;
3877                 local_stats.rx_bytes += sstats->rx_bytes;
3878                 local_stats.rx_errors += sstats->rx_errors;
3879                 local_stats.rx_dropped += sstats->rx_dropped;
3880
3881                 local_stats.tx_packets += sstats->tx_packets;
3882                 local_stats.tx_bytes += sstats->tx_bytes;
3883                 local_stats.tx_errors += sstats->tx_errors;
3884                 local_stats.tx_dropped += sstats->tx_dropped;
3885
3886                 local_stats.multicast += sstats->multicast;
3887                 local_stats.collisions += sstats->collisions;
3888
3889                 local_stats.rx_length_errors += sstats->rx_length_errors;
3890                 local_stats.rx_over_errors += sstats->rx_over_errors;
3891                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3892                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3893                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3894                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3895
3896                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3897                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3898                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3899                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3900                 local_stats.tx_window_errors += sstats->tx_window_errors;
3901         }
3902
3903         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3904
3905         read_unlock_bh(&bond->lock);
3906
3907         return stats;
3908 }
3909
3910 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3911 {
3912         struct net_device *slave_dev = NULL;
3913         struct ifbond k_binfo;
3914         struct ifbond __user *u_binfo = NULL;
3915         struct ifslave k_sinfo;
3916         struct ifslave __user *u_sinfo = NULL;
3917         struct mii_ioctl_data *mii = NULL;
3918         int res = 0;
3919
3920         pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3921                 bond_dev->name, cmd);
3922
3923         switch (cmd) {
3924         case SIOCGMIIPHY:
3925                 mii = if_mii(ifr);
3926                 if (!mii) {
3927                         return -EINVAL;
3928                 }
3929                 mii->phy_id = 0;
3930                 /* Fall Through */
3931         case SIOCGMIIREG:
3932                 /*
3933                  * We do this again just in case we were called by SIOCGMIIREG
3934                  * instead of SIOCGMIIPHY.
3935                  */
3936                 mii = if_mii(ifr);
3937                 if (!mii) {
3938                         return -EINVAL;
3939                 }
3940
3941                 if (mii->reg_num == 1) {
3942                         struct bonding *bond = netdev_priv(bond_dev);
3943                         mii->val_out = 0;
3944                         read_lock(&bond->lock);
3945                         read_lock(&bond->curr_slave_lock);
3946                         if (netif_carrier_ok(bond->dev)) {
3947                                 mii->val_out = BMSR_LSTATUS;
3948                         }
3949                         read_unlock(&bond->curr_slave_lock);
3950                         read_unlock(&bond->lock);
3951                 }
3952
3953                 return 0;
3954         case BOND_INFO_QUERY_OLD:
3955         case SIOCBONDINFOQUERY:
3956                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3957
3958                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3959                         return -EFAULT;
3960                 }
3961
3962                 res = bond_info_query(bond_dev, &k_binfo);
3963                 if (res == 0) {
3964                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3965                                 return -EFAULT;
3966                         }
3967                 }
3968
3969                 return res;
3970         case BOND_SLAVE_INFO_QUERY_OLD:
3971         case SIOCBONDSLAVEINFOQUERY:
3972                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3973
3974                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3975                         return -EFAULT;
3976                 }
3977
3978                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3979                 if (res == 0) {
3980                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3981                                 return -EFAULT;
3982                         }
3983                 }
3984
3985                 return res;
3986         default:
3987                 /* Go on */
3988                 break;
3989         }
3990
3991         if (!capable(CAP_NET_ADMIN)) {
3992                 return -EPERM;
3993         }
3994
3995         down_write(&(bonding_rwsem));
3996         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3997
3998         pr_debug("slave_dev=%p: \n", slave_dev);
3999
4000         if (!slave_dev) {
4001                 res = -ENODEV;
4002         } else {
4003                 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
4004                 switch (cmd) {
4005                 case BOND_ENSLAVE_OLD:
4006                 case SIOCBONDENSLAVE:
4007                         res = bond_enslave(bond_dev, slave_dev);
4008                         break;
4009                 case BOND_RELEASE_OLD:
4010                 case SIOCBONDRELEASE:
4011                         res = bond_release(bond_dev, slave_dev);
4012                         break;
4013                 case BOND_SETHWADDR_OLD:
4014                 case SIOCBONDSETHWADDR:
4015                         res = bond_sethwaddr(bond_dev, slave_dev);
4016                         break;
4017                 case BOND_CHANGE_ACTIVE_OLD:
4018                 case SIOCBONDCHANGEACTIVE:
4019                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4020                         break;
4021                 default:
4022                         res = -EOPNOTSUPP;
4023                 }
4024
4025                 dev_put(slave_dev);
4026         }
4027
4028         up_write(&(bonding_rwsem));
4029         return res;
4030 }
4031
4032 static void bond_set_multicast_list(struct net_device *bond_dev)
4033 {
4034         struct bonding *bond = netdev_priv(bond_dev);
4035         struct dev_mc_list *dmi;
4036
4037         /*
4038          * Do promisc before checking multicast_mode
4039          */
4040         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4041                 /*
4042                  * FIXME: Need to handle the error when one of the multi-slaves
4043                  * encounters error.
4044                  */
4045                 bond_set_promiscuity(bond, 1);
4046         }
4047
4048         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4049                 bond_set_promiscuity(bond, -1);
4050         }
4051
4052         /* set allmulti flag to slaves */
4053         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4054                 /*
4055                  * FIXME: Need to handle the error when one of the multi-slaves
4056                  * encounters error.
4057                  */
4058                 bond_set_allmulti(bond, 1);
4059         }
4060
4061         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4062                 bond_set_allmulti(bond, -1);
4063         }
4064
4065         read_lock(&bond->lock);
4066
4067         bond->flags = bond_dev->flags;
4068
4069         /* looking for addresses to add to slaves' mc list */
4070         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4071                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4072                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4073                 }
4074         }
4075
4076         /* looking for addresses to delete from slaves' list */
4077         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4078                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4079                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4080                 }
4081         }
4082
4083         /* save master's multicast list */
4084         bond_mc_list_destroy(bond);
4085         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4086
4087         read_unlock(&bond->lock);
4088 }
4089
4090 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4091 {
4092         struct bonding *bond = netdev_priv(dev);
4093         struct slave *slave = bond->first_slave;
4094
4095         if (slave) {
4096                 const struct net_device_ops *slave_ops
4097                         = slave->dev->netdev_ops;
4098                 if (slave_ops->ndo_neigh_setup)
4099                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4100         }
4101         return 0;
4102 }
4103
4104 /*
4105  * Change the MTU of all of a master's slaves to match the master
4106  */
4107 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4108 {
4109         struct bonding *bond = netdev_priv(bond_dev);
4110         struct slave *slave, *stop_at;
4111         int res = 0;
4112         int i;
4113
4114         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4115                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4116
4117         /* Can't hold bond->lock with bh disabled here since
4118          * some base drivers panic. On the other hand we can't
4119          * hold bond->lock without bh disabled because we'll
4120          * deadlock. The only solution is to rely on the fact
4121          * that we're under rtnl_lock here, and the slaves
4122          * list won't change. This doesn't solve the problem
4123          * of setting the slave's MTU while it is
4124          * transmitting, but the assumption is that the base
4125          * driver can handle that.
4126          *
4127          * TODO: figure out a way to safely iterate the slaves
4128          * list, but without holding a lock around the actual
4129          * call to the base driver.
4130          */
4131
4132         bond_for_each_slave(bond, slave, i) {
4133                 pr_debug("s %p s->p %p c_m %p\n", slave,
4134                         slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4135
4136                 res = dev_set_mtu(slave->dev, new_mtu);
4137
4138                 if (res) {
4139                         /* If we failed to set the slave's mtu to the new value
4140                          * we must abort the operation even in ACTIVE_BACKUP
4141                          * mode, because if we allow the backup slaves to have
4142                          * different mtu values than the active slave we'll
4143                          * need to change their mtu when doing a failover. That
4144                          * means changing their mtu from timer context, which
4145                          * is probably not a good idea.
4146                          */
4147                         pr_debug("err %d %s\n", res, slave->dev->name);
4148                         goto unwind;
4149                 }
4150         }
4151
4152         bond_dev->mtu = new_mtu;
4153
4154         return 0;
4155
4156 unwind:
4157         /* unwind from head to the slave that failed */
4158         stop_at = slave;
4159         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4160                 int tmp_res;
4161
4162                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4163                 if (tmp_res) {
4164                         pr_debug("unwind err %d dev %s\n", tmp_res,
4165                                 slave->dev->name);
4166                 }
4167         }
4168
4169         return res;
4170 }
4171
4172 /*
4173  * Change HW address
4174  *
4175  * Note that many devices must be down to change the HW address, and
4176  * downing the master releases all slaves.  We can make bonds full of
4177  * bonding devices to test this, however.
4178  */
4179 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4180 {
4181         struct bonding *bond = netdev_priv(bond_dev);
4182         struct sockaddr *sa = addr, tmp_sa;
4183         struct slave *slave, *stop_at;
4184         int res = 0;
4185         int i;
4186
4187         if (bond->params.mode == BOND_MODE_ALB)
4188                 return bond_alb_set_mac_address(bond_dev, addr);
4189
4190
4191         pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4192
4193         /*
4194          * If fail_over_mac is set to active, do nothing and return
4195          * success.  Returning an error causes ifenslave to fail.
4196          */
4197         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4198                 return 0;
4199
4200         if (!is_valid_ether_addr(sa->sa_data)) {
4201                 return -EADDRNOTAVAIL;
4202         }
4203
4204         /* Can't hold bond->lock with bh disabled here since
4205          * some base drivers panic. On the other hand we can't
4206          * hold bond->lock without bh disabled because we'll
4207          * deadlock. The only solution is to rely on the fact
4208          * that we're under rtnl_lock here, and the slaves
4209          * list won't change. This doesn't solve the problem
4210          * of setting the slave's hw address while it is
4211          * transmitting, but the assumption is that the base
4212          * driver can handle that.
4213          *
4214          * TODO: figure out a way to safely iterate the slaves
4215          * list, but without holding a lock around the actual
4216          * call to the base driver.
4217          */
4218
4219         bond_for_each_slave(bond, slave, i) {
4220                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4221                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4222
4223                 if (slave_ops->ndo_set_mac_address == NULL) {
4224                         res = -EOPNOTSUPP;
4225                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4226                         goto unwind;
4227                 }
4228
4229                 res = dev_set_mac_address(slave->dev, addr);
4230                 if (res) {
4231                         /* TODO: consider downing the slave
4232                          * and retry ?
4233                          * User should expect communications
4234                          * breakage anyway until ARP finish
4235                          * updating, so...
4236                          */
4237                         pr_debug("err %d %s\n", res, slave->dev->name);
4238                         goto unwind;
4239                 }
4240         }
4241
4242         /* success */
4243         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4244         return 0;
4245
4246 unwind:
4247         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4248         tmp_sa.sa_family = bond_dev->type;
4249
4250         /* unwind from head to the slave that failed */
4251         stop_at = slave;
4252         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4253                 int tmp_res;
4254
4255                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4256                 if (tmp_res) {
4257                         pr_debug("unwind err %d dev %s\n", tmp_res,
4258                                 slave->dev->name);
4259                 }
4260         }
4261
4262         return res;
4263 }
4264
4265 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4266 {
4267         struct bonding *bond = netdev_priv(bond_dev);
4268         struct slave *slave, *start_at;
4269         int i, slave_no, res = 1;
4270
4271         read_lock(&bond->lock);
4272
4273         if (!BOND_IS_OK(bond)) {
4274                 goto out;
4275         }
4276
4277         /*
4278          * Concurrent TX may collide on rr_tx_counter; we accept that
4279          * as being rare enough not to justify using an atomic op here
4280          */
4281         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4282
4283         bond_for_each_slave(bond, slave, i) {
4284                 slave_no--;
4285                 if (slave_no < 0) {
4286                         break;
4287                 }
4288         }
4289
4290         start_at = slave;
4291         bond_for_each_slave_from(bond, slave, i, start_at) {
4292                 if (IS_UP(slave->dev) &&
4293                     (slave->link == BOND_LINK_UP) &&
4294                     (slave->state == BOND_STATE_ACTIVE)) {
4295                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4296                         break;
4297                 }
4298         }
4299
4300 out:
4301         if (res) {
4302                 /* no suitable interface, frame not sent */
4303                 dev_kfree_skb(skb);
4304         }
4305         read_unlock(&bond->lock);
4306         return 0;
4307 }
4308
4309
4310 /*
4311  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4312  * the bond has a usable interface.
4313  */
4314 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4315 {
4316         struct bonding *bond = netdev_priv(bond_dev);
4317         int res = 1;
4318
4319         read_lock(&bond->lock);
4320         read_lock(&bond->curr_slave_lock);
4321
4322         if (!BOND_IS_OK(bond)) {
4323                 goto out;
4324         }
4325
4326         if (!bond->curr_active_slave)
4327                 goto out;
4328
4329         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4330
4331 out:
4332         if (res) {
4333                 /* no suitable interface, frame not sent */
4334                 dev_kfree_skb(skb);
4335         }
4336         read_unlock(&bond->curr_slave_lock);
4337         read_unlock(&bond->lock);
4338         return 0;
4339 }
4340
4341 /*
4342  * In bond_xmit_xor() , we determine the output device by using a pre-
4343  * determined xmit_hash_policy(), If the selected device is not enabled,
4344  * find the next active slave.
4345  */
4346 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4347 {
4348         struct bonding *bond = netdev_priv(bond_dev);
4349         struct slave *slave, *start_at;
4350         int slave_no;
4351         int i;
4352         int res = 1;
4353
4354         read_lock(&bond->lock);
4355
4356         if (!BOND_IS_OK(bond)) {
4357                 goto out;
4358         }
4359
4360         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4361
4362         bond_for_each_slave(bond, slave, i) {
4363                 slave_no--;
4364                 if (slave_no < 0) {
4365                         break;
4366                 }
4367         }
4368
4369         start_at = slave;
4370
4371         bond_for_each_slave_from(bond, slave, i, start_at) {
4372                 if (IS_UP(slave->dev) &&
4373                     (slave->link == BOND_LINK_UP) &&
4374                     (slave->state == BOND_STATE_ACTIVE)) {
4375                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4376                         break;
4377                 }
4378         }
4379
4380 out:
4381         if (res) {
4382                 /* no suitable interface, frame not sent */
4383                 dev_kfree_skb(skb);
4384         }
4385         read_unlock(&bond->lock);
4386         return 0;
4387 }
4388
4389 /*
4390  * in broadcast mode, we send everything to all usable interfaces.
4391  */
4392 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4393 {
4394         struct bonding *bond = netdev_priv(bond_dev);
4395         struct slave *slave, *start_at;
4396         struct net_device *tx_dev = NULL;
4397         int i;
4398         int res = 1;
4399
4400         read_lock(&bond->lock);
4401
4402         if (!BOND_IS_OK(bond)) {
4403                 goto out;
4404         }
4405
4406         read_lock(&bond->curr_slave_lock);
4407         start_at = bond->curr_active_slave;
4408         read_unlock(&bond->curr_slave_lock);
4409
4410         if (!start_at) {
4411                 goto out;
4412         }
4413
4414         bond_for_each_slave_from(bond, slave, i, start_at) {
4415                 if (IS_UP(slave->dev) &&
4416                     (slave->link == BOND_LINK_UP) &&
4417                     (slave->state == BOND_STATE_ACTIVE)) {
4418                         if (tx_dev) {
4419                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4420                                 if (!skb2) {
4421                                         printk(KERN_ERR DRV_NAME
4422                                                ": %s: Error: bond_xmit_broadcast(): "
4423                                                "skb_clone() failed\n",
4424                                                bond_dev->name);
4425                                         continue;
4426                                 }
4427
4428                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4429                                 if (res) {
4430                                         dev_kfree_skb(skb2);
4431                                         continue;
4432                                 }
4433                         }
4434                         tx_dev = slave->dev;
4435                 }
4436         }
4437
4438         if (tx_dev) {
4439                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4440         }
4441
4442 out:
4443         if (res) {
4444                 /* no suitable interface, frame not sent */
4445                 dev_kfree_skb(skb);
4446         }
4447         /* frame sent to all suitable interfaces */
4448         read_unlock(&bond->lock);
4449         return 0;
4450 }
4451
4452 /*------------------------- Device initialization ---------------------------*/
4453
4454 static void bond_set_xmit_hash_policy(struct bonding *bond)
4455 {
4456         switch (bond->params.xmit_policy) {
4457         case BOND_XMIT_POLICY_LAYER23:
4458                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4459                 break;
4460         case BOND_XMIT_POLICY_LAYER34:
4461                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4462                 break;
4463         case BOND_XMIT_POLICY_LAYER2:
4464         default:
4465                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4466                 break;
4467         }
4468 }
4469
4470 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4471 {
4472         const struct bonding *bond = netdev_priv(dev);
4473
4474         switch (bond->params.mode) {
4475         case BOND_MODE_ROUNDROBIN:
4476                 return bond_xmit_roundrobin(skb, dev);
4477         case BOND_MODE_ACTIVEBACKUP:
4478                 return bond_xmit_activebackup(skb, dev);
4479         case BOND_MODE_XOR:
4480                 return bond_xmit_xor(skb, dev);
4481         case BOND_MODE_BROADCAST:
4482                 return bond_xmit_broadcast(skb, dev);
4483         case BOND_MODE_8023AD:
4484                 return bond_3ad_xmit_xor(skb, dev);
4485         case BOND_MODE_ALB:
4486         case BOND_MODE_TLB:
4487                 return bond_alb_xmit(skb, dev);
4488         default:
4489                 /* Should never happen, mode already checked */
4490                 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4491                      dev->name, bond->params.mode);
4492                 WARN_ON_ONCE(1);
4493                 dev_kfree_skb(skb);
4494                 return NETDEV_TX_OK;
4495         }
4496 }
4497
4498
4499 /*
4500  * set bond mode specific net device operations
4501  */
4502 void bond_set_mode_ops(struct bonding *bond, int mode)
4503 {
4504         struct net_device *bond_dev = bond->dev;
4505
4506         switch (mode) {
4507         case BOND_MODE_ROUNDROBIN:
4508                 break;
4509         case BOND_MODE_ACTIVEBACKUP:
4510                 break;
4511         case BOND_MODE_XOR:
4512                 bond_set_xmit_hash_policy(bond);
4513                 break;
4514         case BOND_MODE_BROADCAST:
4515                 break;
4516         case BOND_MODE_8023AD:
4517                 bond_set_master_3ad_flags(bond);
4518                 bond_set_xmit_hash_policy(bond);
4519                 break;
4520         case BOND_MODE_ALB:
4521                 bond_set_master_alb_flags(bond);
4522                 /* FALLTHRU */
4523         case BOND_MODE_TLB:
4524                 break;
4525         default:
4526                 /* Should never happen, mode already checked */
4527                 printk(KERN_ERR DRV_NAME
4528                        ": %s: Error: Unknown bonding mode %d\n",
4529                        bond_dev->name,
4530                        mode);
4531                 break;
4532         }
4533 }
4534
4535 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4536                                     struct ethtool_drvinfo *drvinfo)
4537 {
4538         strncpy(drvinfo->driver, DRV_NAME, 32);
4539         strncpy(drvinfo->version, DRV_VERSION, 32);
4540         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4541 }
4542
4543 static const struct ethtool_ops bond_ethtool_ops = {
4544         .get_drvinfo            = bond_ethtool_get_drvinfo,
4545         .get_link               = ethtool_op_get_link,
4546         .get_tx_csum            = ethtool_op_get_tx_csum,
4547         .get_sg                 = ethtool_op_get_sg,
4548         .get_tso                = ethtool_op_get_tso,
4549         .get_ufo                = ethtool_op_get_ufo,
4550         .get_flags              = ethtool_op_get_flags,
4551 };
4552
4553 static const struct net_device_ops bond_netdev_ops = {
4554         .ndo_open               = bond_open,
4555         .ndo_stop               = bond_close,
4556         .ndo_start_xmit         = bond_start_xmit,
4557         .ndo_get_stats          = bond_get_stats,
4558         .ndo_do_ioctl           = bond_do_ioctl,
4559         .ndo_set_multicast_list = bond_set_multicast_list,
4560         .ndo_change_mtu         = bond_change_mtu,
4561         .ndo_set_mac_address    = bond_set_mac_address,
4562         .ndo_neigh_setup        = bond_neigh_setup,
4563         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4564         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4565         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4566 };
4567
4568 /*
4569  * Does not allocate but creates a /proc entry.
4570  * Allowed to fail.
4571  */
4572 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4573 {
4574         struct bonding *bond = netdev_priv(bond_dev);
4575
4576         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4577
4578         /* initialize rwlocks */
4579         rwlock_init(&bond->lock);
4580         rwlock_init(&bond->curr_slave_lock);
4581
4582         bond->params = *params; /* copy params struct */
4583
4584         bond->wq = create_singlethread_workqueue(bond_dev->name);
4585         if (!bond->wq)
4586                 return -ENOMEM;
4587
4588         /* Initialize pointers */
4589         bond->first_slave = NULL;
4590         bond->curr_active_slave = NULL;
4591         bond->current_arp_slave = NULL;
4592         bond->primary_slave = NULL;
4593         bond->dev = bond_dev;
4594         bond->send_grat_arp = 0;
4595         bond->send_unsol_na = 0;
4596         bond->setup_by_slave = 0;
4597         INIT_LIST_HEAD(&bond->vlan_list);
4598
4599         /* Initialize the device entry points */
4600         bond_dev->netdev_ops = &bond_netdev_ops;
4601         bond_dev->ethtool_ops = &bond_ethtool_ops;
4602         bond_set_mode_ops(bond, bond->params.mode);
4603
4604         bond_dev->destructor = bond_destructor;
4605
4606         /* Initialize the device options */
4607         bond_dev->tx_queue_len = 0;
4608         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4609         bond_dev->priv_flags |= IFF_BONDING;
4610         if (bond->params.arp_interval)
4611                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4612
4613         /* At first, we block adding VLANs. That's the only way to
4614          * prevent problems that occur when adding VLANs over an
4615          * empty bond. The block will be removed once non-challenged
4616          * slaves are enslaved.
4617          */
4618         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4619
4620         /* don't acquire bond device's netif_tx_lock when
4621          * transmitting */
4622         bond_dev->features |= NETIF_F_LLTX;
4623
4624         /* By default, we declare the bond to be fully
4625          * VLAN hardware accelerated capable. Special
4626          * care is taken in the various xmit functions
4627          * when there are slaves that are not hw accel
4628          * capable
4629          */
4630         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4631                                NETIF_F_HW_VLAN_RX |
4632                                NETIF_F_HW_VLAN_FILTER);
4633
4634 #ifdef CONFIG_PROC_FS
4635         bond_create_proc_entry(bond);
4636 #endif
4637         list_add_tail(&bond->bond_list, &bond_dev_list);
4638
4639         return 0;
4640 }
4641
4642 static void bond_work_cancel_all(struct bonding *bond)
4643 {
4644         write_lock_bh(&bond->lock);
4645         bond->kill_timers = 1;
4646         write_unlock_bh(&bond->lock);
4647
4648         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4649                 cancel_delayed_work(&bond->mii_work);
4650
4651         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4652                 cancel_delayed_work(&bond->arp_work);
4653
4654         if (bond->params.mode == BOND_MODE_ALB &&
4655             delayed_work_pending(&bond->alb_work))
4656                 cancel_delayed_work(&bond->alb_work);
4657
4658         if (bond->params.mode == BOND_MODE_8023AD &&
4659             delayed_work_pending(&bond->ad_work))
4660                 cancel_delayed_work(&bond->ad_work);
4661 }
4662
4663 /* De-initialize device specific data.
4664  * Caller must hold rtnl_lock.
4665  */
4666 static void bond_deinit(struct net_device *bond_dev)
4667 {
4668         struct bonding *bond = netdev_priv(bond_dev);
4669
4670         list_del(&bond->bond_list);
4671
4672         bond_work_cancel_all(bond);
4673
4674 #ifdef CONFIG_PROC_FS
4675         bond_remove_proc_entry(bond);
4676 #endif
4677 }
4678
4679 /* Unregister and free all bond devices.
4680  * Caller must hold rtnl_lock.
4681  */
4682 static void bond_free_all(void)
4683 {
4684         struct bonding *bond, *nxt;
4685
4686         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4687                 struct net_device *bond_dev = bond->dev;
4688
4689                 bond_work_cancel_all(bond);
4690                 /* Release the bonded slaves */
4691                 bond_release_all(bond_dev);
4692                 bond_destroy(bond);
4693         }
4694
4695 #ifdef CONFIG_PROC_FS
4696         bond_destroy_proc_dir();
4697 #endif
4698 }
4699
4700 /*------------------------- Module initialization ---------------------------*/
4701
4702 /*
4703  * Convert string input module parms.  Accept either the
4704  * number of the mode or its string name.  A bit complicated because
4705  * some mode names are substrings of other names, and calls from sysfs
4706  * may have whitespace in the name (trailing newlines, for example).
4707  */
4708 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4709 {
4710         int modeint = -1, i, rv;
4711         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4712
4713         for (p = (char *)buf; *p; p++)
4714                 if (!(isdigit(*p) || isspace(*p)))
4715                         break;
4716
4717         if (*p)
4718                 rv = sscanf(buf, "%20s", modestr);
4719         else
4720                 rv = sscanf(buf, "%d", &modeint);
4721
4722         if (!rv)
4723                 return -1;
4724
4725         for (i = 0; tbl[i].modename; i++) {
4726                 if (modeint == tbl[i].mode)
4727                         return tbl[i].mode;
4728                 if (strcmp(modestr, tbl[i].modename) == 0)
4729                         return tbl[i].mode;
4730         }
4731
4732         return -1;
4733 }
4734
4735 static int bond_check_params(struct bond_params *params)
4736 {
4737         int arp_validate_value, fail_over_mac_value;
4738
4739         /*
4740          * Convert string parameters.
4741          */
4742         if (mode) {
4743                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4744                 if (bond_mode == -1) {
4745                         printk(KERN_ERR DRV_NAME
4746                                ": Error: Invalid bonding mode \"%s\"\n",
4747                                mode == NULL ? "NULL" : mode);
4748                         return -EINVAL;
4749                 }
4750         }
4751
4752         if (xmit_hash_policy) {
4753                 if ((bond_mode != BOND_MODE_XOR) &&
4754                     (bond_mode != BOND_MODE_8023AD)) {
4755                         printk(KERN_INFO DRV_NAME
4756                                ": xor_mode param is irrelevant in mode %s\n",
4757                                bond_mode_name(bond_mode));
4758                 } else {
4759                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4760                                                         xmit_hashtype_tbl);
4761                         if (xmit_hashtype == -1) {
4762                                 printk(KERN_ERR DRV_NAME
4763                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4764                                 xmit_hash_policy == NULL ? "NULL" :
4765                                        xmit_hash_policy);
4766                                 return -EINVAL;
4767                         }
4768                 }
4769         }
4770
4771         if (lacp_rate) {
4772                 if (bond_mode != BOND_MODE_8023AD) {
4773                         printk(KERN_INFO DRV_NAME
4774                                ": lacp_rate param is irrelevant in mode %s\n",
4775                                bond_mode_name(bond_mode));
4776                 } else {
4777                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4778                         if (lacp_fast == -1) {
4779                                 printk(KERN_ERR DRV_NAME
4780                                        ": Error: Invalid lacp rate \"%s\"\n",
4781                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4782                                 return -EINVAL;
4783                         }
4784                 }
4785         }
4786
4787         if (ad_select) {
4788                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4789                 if (params->ad_select == -1) {
4790                         printk(KERN_ERR DRV_NAME
4791                                ": Error: Invalid ad_select \"%s\"\n",
4792                                ad_select == NULL ? "NULL" : ad_select);
4793                         return -EINVAL;
4794                 }
4795
4796                 if (bond_mode != BOND_MODE_8023AD) {
4797                         printk(KERN_WARNING DRV_NAME
4798                                ": ad_select param only affects 802.3ad mode\n");
4799                 }
4800         } else {
4801                 params->ad_select = BOND_AD_STABLE;
4802         }
4803
4804         if (max_bonds < 0 || max_bonds > INT_MAX) {
4805                 printk(KERN_WARNING DRV_NAME
4806                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4807                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4808                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4809                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4810         }
4811
4812         if (miimon < 0) {
4813                 printk(KERN_WARNING DRV_NAME
4814                        ": Warning: miimon module parameter (%d), "
4815                        "not in range 0-%d, so it was reset to %d\n",
4816                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4817                 miimon = BOND_LINK_MON_INTERV;
4818         }
4819
4820         if (updelay < 0) {
4821                 printk(KERN_WARNING DRV_NAME
4822                        ": Warning: updelay module parameter (%d), "
4823                        "not in range 0-%d, so it was reset to 0\n",
4824                        updelay, INT_MAX);
4825                 updelay = 0;
4826         }
4827
4828         if (downdelay < 0) {
4829                 printk(KERN_WARNING DRV_NAME
4830                        ": Warning: downdelay module parameter (%d), "
4831                        "not in range 0-%d, so it was reset to 0\n",
4832                        downdelay, INT_MAX);
4833                 downdelay = 0;
4834         }
4835
4836         if ((use_carrier != 0) && (use_carrier != 1)) {
4837                 printk(KERN_WARNING DRV_NAME
4838                        ": Warning: use_carrier module parameter (%d), "
4839                        "not of valid value (0/1), so it was set to 1\n",
4840                        use_carrier);
4841                 use_carrier = 1;
4842         }
4843
4844         if (num_grat_arp < 0 || num_grat_arp > 255) {
4845                 printk(KERN_WARNING DRV_NAME
4846                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4847                        "was reset to 1 \n", num_grat_arp);
4848                 num_grat_arp = 1;
4849         }
4850
4851         if (num_unsol_na < 0 || num_unsol_na > 255) {
4852                 printk(KERN_WARNING DRV_NAME
4853                        ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4854                        "was reset to 1 \n", num_unsol_na);
4855                 num_unsol_na = 1;
4856         }
4857
4858         /* reset values for 802.3ad */
4859         if (bond_mode == BOND_MODE_8023AD) {
4860                 if (!miimon) {
4861                         printk(KERN_WARNING DRV_NAME
4862                                ": Warning: miimon must be specified, "
4863                                "otherwise bonding will not detect link "
4864                                "failure, speed and duplex which are "
4865                                "essential for 802.3ad operation\n");
4866                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4867                         miimon = 100;
4868                 }
4869         }
4870
4871         /* reset values for TLB/ALB */
4872         if ((bond_mode == BOND_MODE_TLB) ||
4873             (bond_mode == BOND_MODE_ALB)) {
4874                 if (!miimon) {
4875                         printk(KERN_WARNING DRV_NAME
4876                                ": Warning: miimon must be specified, "
4877                                "otherwise bonding will not detect link "
4878                                "failure and link speed which are essential "
4879                                "for TLB/ALB load balancing\n");
4880                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4881                         miimon = 100;
4882                 }
4883         }
4884
4885         if (bond_mode == BOND_MODE_ALB) {
4886                 printk(KERN_NOTICE DRV_NAME
4887                        ": In ALB mode you might experience client "
4888                        "disconnections upon reconnection of a link if the "
4889                        "bonding module updelay parameter (%d msec) is "
4890                        "incompatible with the forwarding delay time of the "
4891                        "switch\n",
4892                        updelay);
4893         }
4894
4895         if (!miimon) {
4896                 if (updelay || downdelay) {
4897                         /* just warn the user the up/down delay will have
4898                          * no effect since miimon is zero...
4899                          */
4900                         printk(KERN_WARNING DRV_NAME
4901                                ": Warning: miimon module parameter not set "
4902                                "and updelay (%d) or downdelay (%d) module "
4903                                "parameter is set; updelay and downdelay have "
4904                                "no effect unless miimon is set\n",
4905                                updelay, downdelay);
4906                 }
4907         } else {
4908                 /* don't allow arp monitoring */
4909                 if (arp_interval) {
4910                         printk(KERN_WARNING DRV_NAME
4911                                ": Warning: miimon (%d) and arp_interval (%d) "
4912                                "can't be used simultaneously, disabling ARP "
4913                                "monitoring\n",
4914                                miimon, arp_interval);
4915                         arp_interval = 0;
4916                 }
4917
4918                 if ((updelay % miimon) != 0) {
4919                         printk(KERN_WARNING DRV_NAME
4920                                ": Warning: updelay (%d) is not a multiple "
4921                                "of miimon (%d), updelay rounded to %d ms\n",
4922                                updelay, miimon, (updelay / miimon) * miimon);
4923                 }
4924
4925                 updelay /= miimon;
4926
4927                 if ((downdelay % miimon) != 0) {
4928                         printk(KERN_WARNING DRV_NAME
4929                                ": Warning: downdelay (%d) is not a multiple "
4930                                "of miimon (%d), downdelay rounded to %d ms\n",
4931                                downdelay, miimon,
4932                                (downdelay / miimon) * miimon);
4933                 }
4934
4935                 downdelay /= miimon;
4936         }
4937
4938         if (arp_interval < 0) {
4939                 printk(KERN_WARNING DRV_NAME
4940                        ": Warning: arp_interval module parameter (%d) "
4941                        ", not in range 0-%d, so it was reset to %d\n",
4942                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4943                 arp_interval = BOND_LINK_ARP_INTERV;
4944         }
4945
4946         for (arp_ip_count = 0;
4947              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4948              arp_ip_count++) {
4949                 /* not complete check, but should be good enough to
4950                    catch mistakes */
4951                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4952                         printk(KERN_WARNING DRV_NAME
4953                                ": Warning: bad arp_ip_target module parameter "
4954                                "(%s), ARP monitoring will not be performed\n",
4955                                arp_ip_target[arp_ip_count]);
4956                         arp_interval = 0;
4957                 } else {
4958                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4959                         arp_target[arp_ip_count] = ip;
4960                 }
4961         }
4962
4963         if (arp_interval && !arp_ip_count) {
4964                 /* don't allow arping if no arp_ip_target given... */
4965                 printk(KERN_WARNING DRV_NAME
4966                        ": Warning: arp_interval module parameter (%d) "
4967                        "specified without providing an arp_ip_target "
4968                        "parameter, arp_interval was reset to 0\n",
4969                        arp_interval);
4970                 arp_interval = 0;
4971         }
4972
4973         if (arp_validate) {
4974                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4975                         printk(KERN_ERR DRV_NAME
4976                ": arp_validate only supported in active-backup mode\n");
4977                         return -EINVAL;
4978                 }
4979                 if (!arp_interval) {
4980                         printk(KERN_ERR DRV_NAME
4981                                ": arp_validate requires arp_interval\n");
4982                         return -EINVAL;
4983                 }
4984
4985                 arp_validate_value = bond_parse_parm(arp_validate,
4986                                                      arp_validate_tbl);
4987                 if (arp_validate_value == -1) {
4988                         printk(KERN_ERR DRV_NAME
4989                                ": Error: invalid arp_validate \"%s\"\n",
4990                                arp_validate == NULL ? "NULL" : arp_validate);
4991                         return -EINVAL;
4992                 }
4993         } else
4994                 arp_validate_value = 0;
4995
4996         if (miimon) {
4997                 printk(KERN_INFO DRV_NAME
4998                        ": MII link monitoring set to %d ms\n",
4999                        miimon);
5000         } else if (arp_interval) {
5001                 int i;
5002
5003                 printk(KERN_INFO DRV_NAME
5004                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5005                        arp_interval,
5006                        arp_validate_tbl[arp_validate_value].modename,
5007                        arp_ip_count);
5008
5009                 for (i = 0; i < arp_ip_count; i++)
5010                         printk (" %s", arp_ip_target[i]);
5011
5012                 printk("\n");
5013
5014         } else if (max_bonds) {
5015                 /* miimon and arp_interval not set, we need one so things
5016                  * work as expected, see bonding.txt for details
5017                  */
5018                 printk(KERN_WARNING DRV_NAME
5019                        ": Warning: either miimon or arp_interval and "
5020                        "arp_ip_target module parameters must be specified, "
5021                        "otherwise bonding will not detect link failures! see "
5022                        "bonding.txt for details.\n");
5023         }
5024
5025         if (primary && !USES_PRIMARY(bond_mode)) {
5026                 /* currently, using a primary only makes sense
5027                  * in active backup, TLB or ALB modes
5028                  */
5029                 printk(KERN_WARNING DRV_NAME
5030                        ": Warning: %s primary device specified but has no "
5031                        "effect in %s mode\n",
5032                        primary, bond_mode_name(bond_mode));
5033                 primary = NULL;
5034         }
5035
5036         if (fail_over_mac) {
5037                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5038                                                       fail_over_mac_tbl);
5039                 if (fail_over_mac_value == -1) {
5040                         printk(KERN_ERR DRV_NAME
5041                                ": Error: invalid fail_over_mac \"%s\"\n",
5042                                arp_validate == NULL ? "NULL" : arp_validate);
5043                         return -EINVAL;
5044                 }
5045
5046                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5047                         printk(KERN_WARNING DRV_NAME
5048                                ": Warning: fail_over_mac only affects "
5049                                "active-backup mode.\n");
5050         } else {
5051                 fail_over_mac_value = BOND_FOM_NONE;
5052         }
5053
5054         /* fill params struct with the proper values */
5055         params->mode = bond_mode;
5056         params->xmit_policy = xmit_hashtype;
5057         params->miimon = miimon;
5058         params->num_grat_arp = num_grat_arp;
5059         params->num_unsol_na = num_unsol_na;
5060         params->arp_interval = arp_interval;
5061         params->arp_validate = arp_validate_value;
5062         params->updelay = updelay;
5063         params->downdelay = downdelay;
5064         params->use_carrier = use_carrier;
5065         params->lacp_fast = lacp_fast;
5066         params->primary[0] = 0;
5067         params->fail_over_mac = fail_over_mac_value;
5068
5069         if (primary) {
5070                 strncpy(params->primary, primary, IFNAMSIZ);
5071                 params->primary[IFNAMSIZ - 1] = 0;
5072         }
5073
5074         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5075
5076         return 0;
5077 }
5078
5079 static struct lock_class_key bonding_netdev_xmit_lock_key;
5080 static struct lock_class_key bonding_netdev_addr_lock_key;
5081
5082 static void bond_set_lockdep_class_one(struct net_device *dev,
5083                                        struct netdev_queue *txq,
5084                                        void *_unused)
5085 {
5086         lockdep_set_class(&txq->_xmit_lock,
5087                           &bonding_netdev_xmit_lock_key);
5088 }
5089
5090 static void bond_set_lockdep_class(struct net_device *dev)
5091 {
5092         lockdep_set_class(&dev->addr_list_lock,
5093                           &bonding_netdev_addr_lock_key);
5094         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5095 }
5096
5097 /* Create a new bond based on the specified name and bonding parameters.
5098  * If name is NULL, obtain a suitable "bond%d" name for us.
5099  * Caller must NOT hold rtnl_lock; we need to release it here before we
5100  * set up our sysfs entries.
5101  */
5102 int bond_create(char *name, struct bond_params *params)
5103 {
5104         struct net_device *bond_dev;
5105         struct bonding *bond;
5106         int res;
5107
5108         rtnl_lock();
5109         down_write(&bonding_rwsem);
5110
5111         /* Check to see if the bond already exists. */
5112         if (name) {
5113                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5114                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5115                                 printk(KERN_ERR DRV_NAME
5116                                ": cannot add bond %s; it already exists\n",
5117                                        name);
5118                                 res = -EPERM;
5119                                 goto out_rtnl;
5120                         }
5121         }
5122
5123         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5124                                 ether_setup);
5125         if (!bond_dev) {
5126                 printk(KERN_ERR DRV_NAME
5127                        ": %s: eek! can't alloc netdev!\n",
5128                        name);
5129                 res = -ENOMEM;
5130                 goto out_rtnl;
5131         }
5132
5133         if (!name) {
5134                 res = dev_alloc_name(bond_dev, "bond%d");
5135                 if (res < 0)
5136                         goto out_netdev;
5137         }
5138
5139         /* bond_init() must be called after dev_alloc_name() (for the
5140          * /proc files), but before register_netdevice(), because we
5141          * need to set function pointers.
5142          */
5143
5144         res = bond_init(bond_dev, params);
5145         if (res < 0) {
5146                 goto out_netdev;
5147         }
5148
5149         res = register_netdevice(bond_dev);
5150         if (res < 0) {
5151                 goto out_bond;
5152         }
5153
5154         bond_set_lockdep_class(bond_dev);
5155
5156         netif_carrier_off(bond_dev);
5157
5158         up_write(&bonding_rwsem);
5159         rtnl_unlock(); /* allows sysfs registration of net device */
5160         res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5161         if (res < 0)
5162                 goto out_unreg;
5163
5164         return 0;
5165
5166 out_unreg:
5167         rtnl_lock();
5168         down_write(&bonding_rwsem);
5169         unregister_netdevice(bond_dev);
5170 out_bond:
5171         bond_deinit(bond_dev);
5172 out_netdev:
5173         free_netdev(bond_dev);
5174 out_rtnl:
5175         up_write(&bonding_rwsem);
5176         rtnl_unlock();
5177         return res;
5178 }
5179
5180 static int __init bonding_init(void)
5181 {
5182         int i;
5183         int res;
5184
5185         printk(KERN_INFO "%s", version);
5186
5187         res = bond_check_params(&bonding_defaults);
5188         if (res) {
5189                 goto out;
5190         }
5191
5192 #ifdef CONFIG_PROC_FS
5193         bond_create_proc_dir();
5194 #endif
5195
5196         init_rwsem(&bonding_rwsem);
5197
5198         for (i = 0; i < max_bonds; i++) {
5199                 res = bond_create(NULL, &bonding_defaults);
5200                 if (res)
5201                         goto err;
5202         }
5203
5204         res = bond_create_sysfs();
5205         if (res)
5206                 goto err;
5207
5208         register_netdevice_notifier(&bond_netdev_notifier);
5209         register_inetaddr_notifier(&bond_inetaddr_notifier);
5210         bond_register_ipv6_notifier();
5211
5212         goto out;
5213 err:
5214         rtnl_lock();
5215         bond_free_all();
5216         rtnl_unlock();
5217 out:
5218         return res;
5219
5220 }
5221
5222 static void __exit bonding_exit(void)
5223 {
5224         unregister_netdevice_notifier(&bond_netdev_notifier);
5225         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5226         bond_unregister_ipv6_notifier();
5227
5228         bond_destroy_sysfs();
5229
5230         rtnl_lock();
5231         bond_free_all();
5232         rtnl_unlock();
5233 }
5234
5235 module_init(bonding_init);
5236 module_exit(bonding_exit);
5237 MODULE_LICENSE("GPL");
5238 MODULE_VERSION(DRV_VERSION);
5239 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5240 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5241 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5242
5243 /*
5244  * Local variables:
5245  *  c-indent-level: 8
5246  *  c-basic-offset: 8
5247  *  tab-width: 8
5248  * End:
5249  */
5250