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

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 //#define BONDING_DEBUG 1
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <asm/system.h>
  57 #include <asm/io.h>
  58 #include <asm/dma.h>
  59 #include <asm/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <net/route.h>
  79 #include <net/net_namespace.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int num_grat_arp = 1;
  92 static int miimon       = BOND_LINK_MON_INTERV;
  93 static int updelay      = 0;
  94 static int downdelay    = 0;
  95 static int use_carrier  = 1;
  96 static char *mode       = NULL;
  97 static char *primary    = NULL;
  98 static char *lacp_rate  = 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(miimon, int, 0);
 111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 112 module_param(updelay, int, 0);
 113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 114 module_param(downdelay, int, 0);
 115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 116                             "in milliseconds");
 117 module_param(use_carrier, int, 0);
 118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 119                               "0 for off, 1 for on (default)");
 120 module_param(mode, charp, 0);
 121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 122                        "1 for active-backup, 2 for balance-xor, "
 123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 124                        "6 for balance-alb");
 125 module_param(primary, charp, 0);
 126 MODULE_PARM_DESC(primary, "Primary network device to use");
 127 module_param(lacp_rate, charp, 0);
 128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 129                             "(slow/fast)");
 130 module_param(xmit_hash_policy, charp, 0);
 131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 132                                    ", 1 for layer 3+4");
 133 module_param(arp_interval, int, 0);
 134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 135 module_param_array(arp_ip_target, charp, NULL, 0);
 136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 137 module_param(arp_validate, charp, 0);
 138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 139 module_param(fail_over_mac, charp, 0);
 140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 141
 142 /*----------------------------- Global variables ----------------------------*/
 143
 144 static const char * const version =
 145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 146
 147 LIST_HEAD(bond_dev_list);
 148
 149 #ifdef CONFIG_PROC_FS
 150 static struct proc_dir_entry *bond_proc_dir = NULL;
 151 #endif
 152
 153 extern struct rw_semaphore bonding_rwsem;
 154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 155 static int arp_ip_count = 0;
 156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 158 static int lacp_fast    = 0;
 159
 160
 161 struct bond_parm_tbl bond_lacp_tbl[] = {
 162 {       "slow",         AD_LACP_SLOW},
 163 {       "fast",         AD_LACP_FAST},
 164 {       NULL,           -1},
 165 };
 166
 167 struct bond_parm_tbl bond_mode_tbl[] = {
 168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 170 {       "balance-xor",          BOND_MODE_XOR},
 171 {       "broadcast",            BOND_MODE_BROADCAST},
 172 {       "802.3ad",              BOND_MODE_8023AD},
 173 {       "balance-tlb",          BOND_MODE_TLB},
 174 {       "balance-alb",          BOND_MODE_ALB},
 175 {       NULL,                   -1},
 176 };
 177
 178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 182 {       NULL,                   -1},
 183 };
 184
 185 struct bond_parm_tbl arp_validate_tbl[] = {
 186 {       "none",                 BOND_ARP_VALIDATE_NONE},
 187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 189 {       "all",                  BOND_ARP_VALIDATE_ALL},
 190 {       NULL,                   -1},
 191 };
 192
 193 struct bond_parm_tbl fail_over_mac_tbl[] = {
 194 {       "none",                 BOND_FOM_NONE},
 195 {       "active",               BOND_FOM_ACTIVE},
 196 {       "follow",               BOND_FOM_FOLLOW},
 197 {       NULL,                   -1},
 198 };
 199
 200 /*-------------------------- Forward declarations ---------------------------*/
 201
 202 static void bond_send_gratuitous_arp(struct bonding *bond);
 203 static void bond_deinit(struct net_device *bond_dev);
 204
 205 /*---------------------------- General routines -----------------------------*/
 206
 207 static const char *bond_mode_name(int mode)
 208 {
 209         switch (mode) {
 210         case BOND_MODE_ROUNDROBIN :
 211                 return "load balancing (round-robin)";
 212         case BOND_MODE_ACTIVEBACKUP :
 213                 return "fault-tolerance (active-backup)";
 214         case BOND_MODE_XOR :
 215                 return "load balancing (xor)";
 216         case BOND_MODE_BROADCAST :
 217                 return "fault-tolerance (broadcast)";
 218         case BOND_MODE_8023AD:
 219                 return "IEEE 802.3ad Dynamic link aggregation";
 220         case BOND_MODE_TLB:
 221                 return "transmit load balancing";
 222         case BOND_MODE_ALB:
 223                 return "adaptive load balancing";
 224         default:
 225                 return "unknown";
 226         }
 227 }
 228
 229 /*---------------------------------- VLAN -----------------------------------*/
 230
 231 /**
 232  * bond_add_vlan - add a new vlan id on bond
 233  * @bond: bond that got the notification
 234  * @vlan_id: the vlan id to add
 235  *
 236  * Returns -ENOMEM if allocation failed.
 237  */
 238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 239 {
 240         struct vlan_entry *vlan;
 241
 242         dprintk("bond: %s, vlan id %d\n",
 243                 (bond ? bond->dev->name: "None"), vlan_id);
 244
 245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 246         if (!vlan) {
 247                 return -ENOMEM;
 248         }
 249
 250         INIT_LIST_HEAD(&vlan->vlan_list);
 251         vlan->vlan_id = vlan_id;
 252         vlan->vlan_ip = 0;
 253
 254         write_lock_bh(&bond->lock);
 255
 256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 257
 258         write_unlock_bh(&bond->lock);
 259
 260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 261
 262         return 0;
 263 }
 264
 265 /**
 266  * bond_del_vlan - delete a vlan id from bond
 267  * @bond: bond that got the notification
 268  * @vlan_id: the vlan id to delete
 269  *
 270  * returns -ENODEV if @vlan_id was not found in @bond.
 271  */
 272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 273 {
 274         struct vlan_entry *vlan;
 275         int res = -ENODEV;
 276
 277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 278
 279         write_lock_bh(&bond->lock);
 280
 281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 282                 if (vlan->vlan_id == vlan_id) {
 283                         list_del(&vlan->vlan_list);
 284
 285                         if ((bond->params.mode == BOND_MODE_TLB) ||
 286                             (bond->params.mode == BOND_MODE_ALB)) {
 287                                 bond_alb_clear_vlan(bond, vlan_id);
 288                         }
 289
 290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 291                                 bond->dev->name);
 292
 293                         kfree(vlan);
 294
 295                         if (list_empty(&bond->vlan_list) &&
 296                             (bond->slave_cnt == 0)) {
 297                                 /* Last VLAN removed and no slaves, so
 298                                  * restore block on adding VLANs. This will
 299                                  * be removed once new slaves that are not
 300                                  * VLAN challenged will be added.
 301                                  */
 302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 303                         }
 304
 305                         res = 0;
 306                         goto out;
 307                 }
 308         }
 309
 310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 311                 bond->dev->name);
 312
 313 out:
 314         write_unlock_bh(&bond->lock);
 315         return res;
 316 }
 317
 318 /**
 319  * bond_has_challenged_slaves
 320  * @bond: the bond we're working on
 321  *
 322  * Searches the slave list. Returns 1 if a vlan challenged slave
 323  * was found, 0 otherwise.
 324  *
 325  * Assumes bond->lock is held.
 326  */
 327 static int bond_has_challenged_slaves(struct bonding *bond)
 328 {
 329         struct slave *slave;
 330         int i;
 331
 332         bond_for_each_slave(bond, slave, i) {
 333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 334                         dprintk("found VLAN challenged slave - %s\n",
 335                                 slave->dev->name);
 336                         return 1;
 337                 }
 338         }
 339
 340         dprintk("no VLAN challenged slaves found\n");
 341         return 0;
 342 }
 343
 344 /**
 345  * bond_next_vlan - safely skip to the next item in the vlans list.
 346  * @bond: the bond we're working on
 347  * @curr: item we're advancing from
 348  *
 349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 350  * or @curr->next otherwise (even if it is @curr itself again).
 351  *
 352  * Caller must hold bond->lock
 353  */
 354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 355 {
 356         struct vlan_entry *next, *last;
 357
 358         if (list_empty(&bond->vlan_list)) {
 359                 return NULL;
 360         }
 361
 362         if (!curr) {
 363                 next = list_entry(bond->vlan_list.next,
 364                                   struct vlan_entry, vlan_list);
 365         } else {
 366                 last = list_entry(bond->vlan_list.prev,
 367                                   struct vlan_entry, vlan_list);
 368                 if (last == curr) {
 369                         next = list_entry(bond->vlan_list.next,
 370                                           struct vlan_entry, vlan_list);
 371                 } else {
 372                         next = list_entry(curr->vlan_list.next,
 373                                           struct vlan_entry, vlan_list);
 374                 }
 375         }
 376
 377         return next;
 378 }
 379
 380 /**
 381  * bond_dev_queue_xmit - Prepare skb for xmit.
 382  *
 383  * @bond: bond device that got this skb for tx.
 384  * @skb: hw accel VLAN tagged skb to transmit
 385  * @slave_dev: slave that is supposed to xmit this skbuff
 386  *
 387  * When the bond gets an skb to transmit that is
 388  * already hardware accelerated VLAN tagged, and it
 389  * needs to relay this skb to a slave that is not
 390  * hw accel capable, the skb needs to be "unaccelerated",
 391  * i.e. strip the hwaccel tag and re-insert it as part
 392  * of the payload.
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 395 {
 396         unsigned short uninitialized_var(vlan_id);
 397
 398         if (!list_empty(&bond->vlan_list) &&
 399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 400             vlan_get_tag(skb, &vlan_id) == 0) {
 401                 skb->dev = slave_dev;
 402                 skb = vlan_put_tag(skb, vlan_id);
 403                 if (!skb) {
 404                         /* vlan_put_tag() frees the skb in case of error,
 405                          * so return success here so the calling functions
 406                          * won't attempt to free is again.
 407                          */
 408                         return 0;
 409                 }
 410         } else {
 411                 skb->dev = slave_dev;
 412         }
 413
 414         skb->priority = 1;
 415         dev_queue_xmit(skb);
 416
 417         return 0;
 418 }
 419
 420 /*
 421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 423  * lock because:
 424  * a. This operation is performed in IOCTL context,
 425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 426  * c. Holding a lock with BH disabled while directly calling a base driver
 427  *    entry point is generally a BAD idea.
 428  *
 429  * The design of synchronization/protection for this operation in the 8021q
 430  * module is good for one or more VLAN devices over a single physical device
 431  * and cannot be extended for a teaming solution like bonding, so there is a
 432  * potential race condition here where a net device from the vlan group might
 433  * be referenced (either by a base driver or the 8021q code) while it is being
 434  * removed from the system. However, it turns out we're not making matters
 435  * worse, and if it works for regular VLAN usage it will work here too.
 436 */
 437
 438 /**
 439  * bond_vlan_rx_register - Propagates registration to slaves
 440  * @bond_dev: bonding net device that got called
 441  * @grp: vlan group being registered
 442  */
 443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 444 {
 445         struct bonding *bond = bond_dev->priv;
 446         struct slave *slave;
 447         int i;
 448
 449         bond->vlgrp = grp;
 450
 451         bond_for_each_slave(bond, slave, i) {
 452                 struct net_device *slave_dev = slave->dev;
 453
 454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 455                     slave_dev->vlan_rx_register) {
 456                         slave_dev->vlan_rx_register(slave_dev, grp);
 457                 }
 458         }
 459 }
 460
 461 /**
 462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being added
 465  */
 466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = bond_dev->priv;
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i) {
 473                 struct net_device *slave_dev = slave->dev;
 474
 475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 476                     slave_dev->vlan_rx_add_vid) {
 477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 478                 }
 479         }
 480
 481         res = bond_add_vlan(bond, vid);
 482         if (res) {
 483                 printk(KERN_ERR DRV_NAME
 484                        ": %s: Error: Failed to add vlan id %d\n",
 485                        bond_dev->name, vid);
 486         }
 487 }
 488
 489 /**
 490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 491  * @bond_dev: bonding net device that got called
 492  * @vid: vlan id being removed
 493  */
 494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 495 {
 496         struct bonding *bond = bond_dev->priv;
 497         struct slave *slave;
 498         struct net_device *vlan_dev;
 499         int i, res;
 500
 501         bond_for_each_slave(bond, slave, i) {
 502                 struct net_device *slave_dev = slave->dev;
 503
 504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 505                     slave_dev->vlan_rx_kill_vid) {
 506                         /* Save and then restore vlan_dev in the grp array,
 507                          * since the slave's driver might clear it.
 508                          */
 509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 512                 }
 513         }
 514
 515         res = bond_del_vlan(bond, vid);
 516         if (res) {
 517                 printk(KERN_ERR DRV_NAME
 518                        ": %s: Error: Failed to remove vlan id %d\n",
 519                        bond_dev->name, vid);
 520         }
 521 }
 522
 523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 524 {
 525         struct vlan_entry *vlan;
 526
 527         write_lock_bh(&bond->lock);
 528
 529         if (list_empty(&bond->vlan_list)) {
 530                 goto out;
 531         }
 532
 533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 534             slave_dev->vlan_rx_register) {
 535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 536         }
 537
 538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 539             !(slave_dev->vlan_rx_add_vid)) {
 540                 goto out;
 541         }
 542
 543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 545         }
 546
 547 out:
 548         write_unlock_bh(&bond->lock);
 549 }
 550
 551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 552 {
 553         struct vlan_entry *vlan;
 554         struct net_device *vlan_dev;
 555
 556         write_lock_bh(&bond->lock);
 557
 558         if (list_empty(&bond->vlan_list)) {
 559                 goto out;
 560         }
 561
 562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 563             !(slave_dev->vlan_rx_kill_vid)) {
 564                 goto unreg;
 565         }
 566
 567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 568                 /* Save and then restore vlan_dev in the grp array,
 569                  * since the slave's driver might clear it.
 570                  */
 571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 574         }
 575
 576 unreg:
 577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 578             slave_dev->vlan_rx_register) {
 579                 slave_dev->vlan_rx_register(slave_dev, NULL);
 580         }
 581
 582 out:
 583         write_unlock_bh(&bond->lock);
 584 }
 585
 586 /*------------------------------- Link status -------------------------------*/
 587
 588 /*
 589  * Set the carrier state for the master according to the state of its
 590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 591  * do special 802.3ad magic.
 592  *
 593  * Returns zero if carrier state does not change, nonzero if it does.
 594  */
 595 static int bond_set_carrier(struct bonding *bond)
 596 {
 597         struct slave *slave;
 598         int i;
 599
 600         if (bond->slave_cnt == 0)
 601                 goto down;
 602
 603         if (bond->params.mode == BOND_MODE_8023AD)
 604                 return bond_3ad_set_carrier(bond);
 605
 606         bond_for_each_slave(bond, slave, i) {
 607                 if (slave->link == BOND_LINK_UP) {
 608                         if (!netif_carrier_ok(bond->dev)) {
 609                                 netif_carrier_on(bond->dev);
 610                                 return 1;
 611                         }
 612                         return 0;
 613                 }
 614         }
 615
 616 down:
 617         if (netif_carrier_ok(bond->dev)) {
 618                 netif_carrier_off(bond->dev);
 619                 return 1;
 620         }
 621         return 0;
 622 }
 623
 624 /*
 625  * Get link speed and duplex from the slave's base driver
 626  * using ethtool. If for some reason the call fails or the
 627  * values are invalid, fake speed and duplex to 100/Full
 628  * and return error.
 629  */
 630 static int bond_update_speed_duplex(struct slave *slave)
 631 {
 632         struct net_device *slave_dev = slave->dev;
 633         struct ethtool_cmd etool;
 634         int res;
 635
 636         /* Fake speed and duplex */
 637         slave->speed = SPEED_100;
 638         slave->duplex = DUPLEX_FULL;
 639
 640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 641                 return -1;
 642
 643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 644         if (res < 0)
 645                 return -1;
 646
 647         switch (etool.speed) {
 648         case SPEED_10:
 649         case SPEED_100:
 650         case SPEED_1000:
 651         case SPEED_10000:
 652                 break;
 653         default:
 654                 return -1;
 655         }
 656
 657         switch (etool.duplex) {
 658         case DUPLEX_FULL:
 659         case DUPLEX_HALF:
 660                 break;
 661         default:
 662                 return -1;
 663         }
 664
 665         slave->speed = etool.speed;
 666         slave->duplex = etool.duplex;
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * if <dev> supports MII link status reporting, check its link status.
 673  *
 674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 675  * depening upon the setting of the use_carrier parameter.
 676  *
 677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 678  * can't tell and just pretend it is), or 0, meaning that the link is
 679  * down.
 680  *
 681  * If reporting is non-zero, instead of faking link up, return -1 if
 682  * both ETHTOOL and MII ioctls fail (meaning the device does not
 683  * support them).  If use_carrier is set, return whatever it says.
 684  * It'd be nice if there was a good way to tell if a driver supports
 685  * netif_carrier, but there really isn't.
 686  */
 687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 688 {
 689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 690         struct ifreq ifr;
 691         struct mii_ioctl_data *mii;
 692
 693         if (bond->params.use_carrier) {
 694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 695         }
 696
 697         ioctl = slave_dev->do_ioctl;
 698         if (ioctl) {
 699                 /* TODO: set pointer to correct ioctl on a per team member */
 700                 /*       bases to make this more efficient. that is, once  */
 701                 /*       we determine the correct ioctl, we will always    */
 702                 /*       call it and not the others for that team          */
 703                 /*       member.                                           */
 704
 705                 /*
 706                  * We cannot assume that SIOCGMIIPHY will also read a
 707                  * register; not all network drivers (e.g., e100)
 708                  * support that.
 709                  */
 710
 711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 713                 mii = if_mii(&ifr);
 714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 715                         mii->reg_num = MII_BMSR;
 716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 717                                 return (mii->val_out & BMSR_LSTATUS);
 718                         }
 719                 }
 720         }
 721
 722         /*
 723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 724          * attempt to get link status from it if the above MII ioctls fail.
 725          */
 726         if (slave_dev->ethtool_ops) {
 727                 if (slave_dev->ethtool_ops->get_link) {
 728                         u32 link;
 729
 730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 731
 732                         return link ? BMSR_LSTATUS : 0;
 733                 }
 734         }
 735
 736         /*
 737          * If reporting, report that either there's no dev->do_ioctl,
 738          * or both SIOCGMIIREG and get_link failed (meaning that we
 739          * cannot report link status).  If not reporting, pretend
 740          * we're ok.
 741          */
 742         return (reporting ? -1 : BMSR_LSTATUS);
 743 }
 744
 745 /*----------------------------- Multicast list ------------------------------*/
 746
 747 /*
 748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 749  */
 750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 751 {
 752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 754 }
 755
 756 /*
 757  * returns dmi entry if found, NULL otherwise
 758  */
 759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 760 {
 761         struct dev_mc_list *idmi;
 762
 763         for (idmi = mc_list; idmi; idmi = idmi->next) {
 764                 if (bond_is_dmi_same(dmi, idmi)) {
 765                         return idmi;
 766                 }
 767         }
 768
 769         return NULL;
 770 }
 771
 772 /*
 773  * Push the promiscuity flag down to appropriate slaves
 774  */
 775 static int bond_set_promiscuity(struct bonding *bond, int inc)
 776 {
 777         int err = 0;
 778         if (USES_PRIMARY(bond->params.mode)) {
 779                 /* write lock already acquired */
 780                 if (bond->curr_active_slave) {
 781                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 782                                                   inc);
 783                 }
 784         } else {
 785                 struct slave *slave;
 786                 int i;
 787                 bond_for_each_slave(bond, slave, i) {
 788                         err = dev_set_promiscuity(slave->dev, inc);
 789                         if (err)
 790                                 return err;
 791                 }
 792         }
 793         return err;
 794 }
 795
 796 /*
 797  * Push the allmulti flag down to all slaves
 798  */
 799 static int bond_set_allmulti(struct bonding *bond, int inc)
 800 {
 801         int err = 0;
 802         if (USES_PRIMARY(bond->params.mode)) {
 803                 /* write lock already acquired */
 804                 if (bond->curr_active_slave) {
 805                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 806                                                inc);
 807                 }
 808         } else {
 809                 struct slave *slave;
 810                 int i;
 811                 bond_for_each_slave(bond, slave, i) {
 812                         err = dev_set_allmulti(slave->dev, inc);
 813                         if (err)
 814                                 return err;
 815                 }
 816         }
 817         return err;
 818 }
 819
 820 /*
 821  * Add a Multicast address to slaves
 822  * according to mode
 823  */
 824 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 825 {
 826         if (USES_PRIMARY(bond->params.mode)) {
 827                 /* write lock already acquired */
 828                 if (bond->curr_active_slave) {
 829                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 830                 }
 831         } else {
 832                 struct slave *slave;
 833                 int i;
 834                 bond_for_each_slave(bond, slave, i) {
 835                         dev_mc_add(slave->dev, addr, alen, 0);
 836                 }
 837         }
 838 }
 839
 840 /*
 841  * Remove a multicast address from slave
 842  * according to mode
 843  */
 844 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 845 {
 846         if (USES_PRIMARY(bond->params.mode)) {
 847                 /* write lock already acquired */
 848                 if (bond->curr_active_slave) {
 849                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 850                 }
 851         } else {
 852                 struct slave *slave;
 853                 int i;
 854                 bond_for_each_slave(bond, slave, i) {
 855                         dev_mc_delete(slave->dev, addr, alen, 0);
 856                 }
 857         }
 858 }
 859
 860
 861 /*
 862  * Retrieve the list of registered multicast addresses for the bonding
 863  * device and retransmit an IGMP JOIN request to the current active
 864  * slave.
 865  */
 866 static void bond_resend_igmp_join_requests(struct bonding *bond)
 867 {
 868         struct in_device *in_dev;
 869         struct ip_mc_list *im;
 870
 871         rcu_read_lock();
 872         in_dev = __in_dev_get_rcu(bond->dev);
 873         if (in_dev) {
 874                 for (im = in_dev->mc_list; im; im = im->next) {
 875                         ip_mc_rejoin_group(im);
 876                 }
 877         }
 878
 879         rcu_read_unlock();
 880 }
 881
 882 /*
 883  * Totally destroys the mc_list in bond
 884  */
 885 static void bond_mc_list_destroy(struct bonding *bond)
 886 {
 887         struct dev_mc_list *dmi;
 888
 889         dmi = bond->mc_list;
 890         while (dmi) {
 891                 bond->mc_list = dmi->next;
 892                 kfree(dmi);
 893                 dmi = bond->mc_list;
 894         }
 895         bond->mc_list = NULL;
 896 }
 897
 898 /*
 899  * Copy all the Multicast addresses from src to the bonding device dst
 900  */
 901 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 902                              gfp_t gfp_flag)
 903 {
 904         struct dev_mc_list *dmi, *new_dmi;
 905
 906         for (dmi = mc_list; dmi; dmi = dmi->next) {
 907                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 908
 909                 if (!new_dmi) {
 910                         /* FIXME: Potential memory leak !!! */
 911                         return -ENOMEM;
 912                 }
 913
 914                 new_dmi->next = bond->mc_list;
 915                 bond->mc_list = new_dmi;
 916                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 917                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 918                 new_dmi->dmi_users = dmi->dmi_users;
 919                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 920         }
 921
 922         return 0;
 923 }
 924
 925 /*
 926  * flush all members of flush->mc_list from device dev->mc_list
 927  */
 928 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 929 {
 930         struct bonding *bond = bond_dev->priv;
 931         struct dev_mc_list *dmi;
 932
 933         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 934                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 935         }
 936
 937         if (bond->params.mode == BOND_MODE_8023AD) {
 938                 /* del lacpdu mc addr from mc list */
 939                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 940
 941                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 942         }
 943 }
 944
 945 /*--------------------------- Active slave change ---------------------------*/
 946
 947 /*
 948  * Update the mc list and multicast-related flags for the new and
 949  * old active slaves (if any) according to the multicast mode, and
 950  * promiscuous flags unconditionally.
 951  */
 952 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 953 {
 954         struct dev_mc_list *dmi;
 955
 956         if (!USES_PRIMARY(bond->params.mode)) {
 957                 /* nothing to do -  mc list is already up-to-date on
 958                  * all slaves
 959                  */
 960                 return;
 961         }
 962
 963         if (old_active) {
 964                 if (bond->dev->flags & IFF_PROMISC) {
 965                         dev_set_promiscuity(old_active->dev, -1);
 966                 }
 967
 968                 if (bond->dev->flags & IFF_ALLMULTI) {
 969                         dev_set_allmulti(old_active->dev, -1);
 970                 }
 971
 972                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 973                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 974                 }
 975         }
 976
 977         if (new_active) {
 978                 /* FIXME: Signal errors upstream. */
 979                 if (bond->dev->flags & IFF_PROMISC) {
 980                         dev_set_promiscuity(new_active->dev, 1);
 981                 }
 982
 983                 if (bond->dev->flags & IFF_ALLMULTI) {
 984                         dev_set_allmulti(new_active->dev, 1);
 985                 }
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 988                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 989                 }
 990                 bond_resend_igmp_join_requests(bond);
 991         }
 992 }
 993
 994 /*
 995  * bond_do_fail_over_mac
 996  *
 997  * Perform special MAC address swapping for fail_over_mac settings
 998  *
 999  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000  */
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002                                   struct slave *new_active,
1003                                   struct slave *old_active)
1004 {
1005         u8 tmp_mac[ETH_ALEN];
1006         struct sockaddr saddr;
1007         int rv;
1008
1009         switch (bond->params.fail_over_mac) {
1010         case BOND_FOM_ACTIVE:
1011                 if (new_active)
1012                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1013                                new_active->dev->addr_len);
1014                 break;
1015         case BOND_FOM_FOLLOW:
1016                 /*
1017                  * if new_active && old_active, swap them
1018                  * if just old_active, do nothing (going to no active slave)
1019                  * if just new_active, set new_active to bond's MAC
1020                  */
1021                 if (!new_active)
1022                         return;
1023
1024                 write_unlock_bh(&bond->curr_slave_lock);
1025                 read_unlock(&bond->lock);
1026
1027                 if (old_active) {
1028                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030                                ETH_ALEN);
1031                         saddr.sa_family = new_active->dev->type;
1032                 } else {
1033                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034                         saddr.sa_family = bond->dev->type;
1035                 }
1036
1037                 rv = dev_set_mac_address(new_active->dev, &saddr);
1038                 if (rv) {
1039                         printk(KERN_ERR DRV_NAME
1040                                ": %s: Error %d setting MAC of slave %s\n",
1041                                bond->dev->name, -rv, new_active->dev->name);
1042                         goto out;
1043                 }
1044
1045                 if (!old_active)
1046                         goto out;
1047
1048                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049                 saddr.sa_family = old_active->dev->type;
1050
1051                 rv = dev_set_mac_address(old_active->dev, &saddr);
1052                 if (rv)
1053                         printk(KERN_ERR DRV_NAME
1054                                ": %s: Error %d setting MAC of slave %s\n",
1055                                bond->dev->name, -rv, new_active->dev->name);
1056 out:
1057                 read_lock(&bond->lock);
1058                 write_lock_bh(&bond->curr_slave_lock);
1059                 break;
1060         default:
1061                 printk(KERN_ERR DRV_NAME
1062                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063                        bond->dev->name, bond->params.fail_over_mac);
1064                 break;
1065         }
1066
1067 }
1068
1069
1070 /**
1071  * find_best_interface - select the best available slave to be the active one
1072  * @bond: our bonding struct
1073  *
1074  * Warning: Caller must hold curr_slave_lock for writing.
1075  */
1076 static struct slave *bond_find_best_slave(struct bonding *bond)
1077 {
1078         struct slave *new_active, *old_active;
1079         struct slave *bestslave = NULL;
1080         int mintime = bond->params.updelay;
1081         int i;
1082
1083         new_active = old_active = bond->curr_active_slave;
1084
1085         if (!new_active) { /* there were no active slaves left */
1086                 if (bond->slave_cnt > 0) {  /* found one slave */
1087                         new_active = bond->first_slave;
1088                 } else {
1089                         return NULL; /* still no slave, return NULL */
1090                 }
1091         }
1092
1093         /* first try the primary link; if arping, a link must tx/rx traffic
1094          * before it can be considered the curr_active_slave - also, we would skip
1095          * slaves between the curr_active_slave and primary_slave that may be up
1096          * and able to arp
1097          */
1098         if ((bond->primary_slave) &&
1099             (!bond->params.arp_interval) &&
1100             (IS_UP(bond->primary_slave->dev))) {
1101                 new_active = bond->primary_slave;
1102         }
1103
1104         /* remember where to stop iterating over the slaves */
1105         old_active = new_active;
1106
1107         bond_for_each_slave_from(bond, new_active, i, old_active) {
1108                 if (IS_UP(new_active->dev)) {
1109                         if (new_active->link == BOND_LINK_UP) {
1110                                 return new_active;
1111                         } else if (new_active->link == BOND_LINK_BACK) {
1112                                 /* link up, but waiting for stabilization */
1113                                 if (new_active->delay < mintime) {
1114                                         mintime = new_active->delay;
1115                                         bestslave = new_active;
1116                                 }
1117                         }
1118                 }
1119         }
1120
1121         return bestslave;
1122 }
1123
1124 /**
1125  * change_active_interface - change the active slave into the specified one
1126  * @bond: our bonding struct
1127  * @new: the new slave to make the active one
1128  *
1129  * Set the new slave to the bond's settings and unset them on the old
1130  * curr_active_slave.
1131  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132  *
1133  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134  * because it is apparently the best available slave we have, even though its
1135  * updelay hasn't timed out yet.
1136  *
1137  * If new_active is not NULL, caller must hold bond->lock for read and
1138  * curr_slave_lock for write_bh.
1139  */
1140 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 {
1142         struct slave *old_active = bond->curr_active_slave;
1143
1144         if (old_active == new_active) {
1145                 return;
1146         }
1147
1148         if (new_active) {
1149                 new_active->jiffies = jiffies;
1150
1151                 if (new_active->link == BOND_LINK_BACK) {
1152                         if (USES_PRIMARY(bond->params.mode)) {
1153                                 printk(KERN_INFO DRV_NAME
1154                                        ": %s: making interface %s the new "
1155                                        "active one %d ms earlier.\n",
1156                                        bond->dev->name, new_active->dev->name,
1157                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158                         }
1159
1160                         new_active->delay = 0;
1161                         new_active->link = BOND_LINK_UP;
1162
1163                         if (bond->params.mode == BOND_MODE_8023AD) {
1164                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165                         }
1166
1167                         if ((bond->params.mode == BOND_MODE_TLB) ||
1168                             (bond->params.mode == BOND_MODE_ALB)) {
1169                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170                         }
1171                 } else {
1172                         if (USES_PRIMARY(bond->params.mode)) {
1173                                 printk(KERN_INFO DRV_NAME
1174                                        ": %s: making interface %s the new "
1175                                        "active one.\n",
1176                                        bond->dev->name, new_active->dev->name);
1177                         }
1178                 }
1179         }
1180
1181         if (USES_PRIMARY(bond->params.mode)) {
1182                 bond_mc_swap(bond, new_active, old_active);
1183         }
1184
1185         if ((bond->params.mode == BOND_MODE_TLB) ||
1186             (bond->params.mode == BOND_MODE_ALB)) {
1187                 bond_alb_handle_active_change(bond, new_active);
1188                 if (old_active)
1189                         bond_set_slave_inactive_flags(old_active);
1190                 if (new_active)
1191                         bond_set_slave_active_flags(new_active);
1192         } else {
1193                 bond->curr_active_slave = new_active;
1194         }
1195
1196         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197                 if (old_active) {
1198                         bond_set_slave_inactive_flags(old_active);
1199                 }
1200
1201                 if (new_active) {
1202                         bond_set_slave_active_flags(new_active);
1203
1204                         if (bond->params.fail_over_mac)
1205                                 bond_do_fail_over_mac(bond, new_active,
1206                                                       old_active);
1207
1208                         bond->send_grat_arp = bond->params.num_grat_arp;
1209                         bond_send_gratuitous_arp(bond);
1210
1211                         write_unlock_bh(&bond->curr_slave_lock);
1212                         read_unlock(&bond->lock);
1213
1214                         netdev_bonding_change(bond->dev);
1215
1216                         read_lock(&bond->lock);
1217                         write_lock_bh(&bond->curr_slave_lock);
1218                 }
1219         }
1220 }
1221
1222 /**
1223  * bond_select_active_slave - select a new active slave, if needed
1224  * @bond: our bonding struct
1225  *
1226  * This functions shoud be called when one of the following occurs:
1227  * - The old curr_active_slave has been released or lost its link.
1228  * - The primary_slave has got its link back.
1229  * - A slave has got its link back and there's no old curr_active_slave.
1230  *
1231  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1232  */
1233 void bond_select_active_slave(struct bonding *bond)
1234 {
1235         struct slave *best_slave;
1236         int rv;
1237
1238         best_slave = bond_find_best_slave(bond);
1239         if (best_slave != bond->curr_active_slave) {
1240                 bond_change_active_slave(bond, best_slave);
1241                 rv = bond_set_carrier(bond);
1242                 if (!rv)
1243                         return;
1244
1245                 if (netif_carrier_ok(bond->dev)) {
1246                         printk(KERN_INFO DRV_NAME
1247                                ": %s: first active interface up!\n",
1248                                bond->dev->name);
1249                 } else {
1250                         printk(KERN_INFO DRV_NAME ": %s: "
1251                                "now running without any active interface !\n",
1252                                bond->dev->name);
1253                 }
1254         }
1255 }
1256
1257 /*--------------------------- slave list handling ---------------------------*/
1258
1259 /*
1260  * This function attaches the slave to the end of list.
1261  *
1262  * bond->lock held for writing by caller.
1263  */
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265 {
1266         if (bond->first_slave == NULL) { /* attaching the first slave */
1267                 new_slave->next = new_slave;
1268                 new_slave->prev = new_slave;
1269                 bond->first_slave = new_slave;
1270         } else {
1271                 new_slave->next = bond->first_slave;
1272                 new_slave->prev = bond->first_slave->prev;
1273                 new_slave->next->prev = new_slave;
1274                 new_slave->prev->next = new_slave;
1275         }
1276
1277         bond->slave_cnt++;
1278 }
1279
1280 /*
1281  * This function detaches the slave from the list.
1282  * WARNING: no check is made to verify if the slave effectively
1283  * belongs to <bond>.
1284  * Nothing is freed on return, structures are just unchained.
1285  * If any slave pointer in bond was pointing to <slave>,
1286  * it should be changed by the calling function.
1287  *
1288  * bond->lock held for writing by caller.
1289  */
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291 {
1292         if (slave->next) {
1293                 slave->next->prev = slave->prev;
1294         }
1295
1296         if (slave->prev) {
1297                 slave->prev->next = slave->next;
1298         }
1299
1300         if (bond->first_slave == slave) { /* slave is the first slave */
1301                 if (bond->slave_cnt > 1) { /* there are more slave */
1302                         bond->first_slave = slave->next;
1303                 } else {
1304                         bond->first_slave = NULL; /* slave was the last one */
1305                 }
1306         }
1307
1308         slave->next = NULL;
1309         slave->prev = NULL;
1310         bond->slave_cnt--;
1311 }
1312
1313 /*---------------------------------- IOCTL ----------------------------------*/
1314
1315 static int bond_sethwaddr(struct net_device *bond_dev,
1316                           struct net_device *slave_dev)
1317 {
1318         dprintk("bond_dev=%p\n", bond_dev);
1319         dprintk("slave_dev=%p\n", slave_dev);
1320         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1321         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1322         return 0;
1323 }
1324
1325 #define BOND_VLAN_FEATURES \
1326         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327          NETIF_F_HW_VLAN_FILTER)
1328
1329 /*
1330  * Compute the common dev->feature set available to all slaves.  Some
1331  * feature bits are managed elsewhere, so preserve those feature bits
1332  * on the master device.
1333  */
1334 static int bond_compute_features(struct bonding *bond)
1335 {
1336         struct slave *slave;
1337         struct net_device *bond_dev = bond->dev;
1338         unsigned long features = bond_dev->features;
1339         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1340                                                 bond_dev->hard_header_len);
1341         int i;
1342
1343         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1344         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1345
1346         if (!bond->first_slave)
1347                 goto done;
1348
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350
1351         bond_for_each_slave(bond, slave, i) {
1352                 features = netdev_increment_features(features,
1353                                                      slave->dev->features,
1354                                                      NETIF_F_ONE_FOR_ALL);
1355                 if (slave->dev->hard_header_len > max_hard_header_len)
1356                         max_hard_header_len = slave->dev->hard_header_len;
1357         }
1358
1359 done:
1360         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1361         bond_dev->features = netdev_fix_features(features, NULL);
1362         bond_dev->hard_header_len = max_hard_header_len;
1363
1364         return 0;
1365 }
1366
1367
1368 static void bond_setup_by_slave(struct net_device *bond_dev,
1369                                 struct net_device *slave_dev)
1370 {
1371         struct bonding *bond = bond_dev->priv;
1372
1373         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1374         bond_dev->header_ops            = slave_dev->header_ops;
1375
1376         bond_dev->type              = slave_dev->type;
1377         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1378         bond_dev->addr_len          = slave_dev->addr_len;
1379
1380         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1381                 slave_dev->addr_len);
1382         bond->setup_by_slave = 1;
1383 }
1384
1385 /* enslave device <slave> to bond device <master> */
1386 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1387 {
1388         struct bonding *bond = bond_dev->priv;
1389         struct slave *new_slave = NULL;
1390         struct dev_mc_list *dmi;
1391         struct sockaddr addr;
1392         int link_reporting;
1393         int old_features = bond_dev->features;
1394         int res = 0;
1395
1396         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397                 slave_dev->do_ioctl == NULL) {
1398                 printk(KERN_WARNING DRV_NAME
1399                        ": %s: Warning: no link monitoring support for %s\n",
1400                        bond_dev->name, slave_dev->name);
1401         }
1402
1403         /* bond must be initialized by bond_open() before enslaving */
1404         if (!(bond_dev->flags & IFF_UP)) {
1405                 printk(KERN_WARNING DRV_NAME
1406                         " %s: master_dev is not up in bond_enslave\n",
1407                         bond_dev->name);
1408         }
1409
1410         /* already enslaved */
1411         if (slave_dev->flags & IFF_SLAVE) {
1412                 dprintk("Error, Device was already enslaved\n");
1413                 return -EBUSY;
1414         }
1415
1416         /* vlan challenged mutual exclusion */
1417         /* no need to lock since we're protected by rtnl_lock */
1418         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420                 if (!list_empty(&bond->vlan_list)) {
1421                         printk(KERN_ERR DRV_NAME
1422                                ": %s: Error: cannot enslave VLAN "
1423                                "challenged slave %s on VLAN enabled "
1424                                "bond %s\n", bond_dev->name, slave_dev->name,
1425                                bond_dev->name);
1426                         return -EPERM;
1427                 } else {
1428                         printk(KERN_WARNING DRV_NAME
1429                                ": %s: Warning: enslaved VLAN challenged "
1430                                "slave %s. Adding VLANs will be blocked as "
1431                                "long as %s is part of bond %s\n",
1432                                bond_dev->name, slave_dev->name, slave_dev->name,
1433                                bond_dev->name);
1434                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1435                 }
1436         } else {
1437                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438                 if (bond->slave_cnt == 0) {
1439                         /* First slave, and it is not VLAN challenged,
1440                          * so remove the block of adding VLANs over the bond.
1441                          */
1442                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1443                 }
1444         }
1445
1446         /*
1447          * Old ifenslave binaries are no longer supported.  These can
1448          * be identified with moderate accurary by the state of the slave:
1449          * the current ifenslave will set the interface down prior to
1450          * enslaving it; the old ifenslave will not.
1451          */
1452         if ((slave_dev->flags & IFF_UP)) {
1453                 printk(KERN_ERR DRV_NAME ": %s is up. "
1454                        "This may be due to an out of date ifenslave.\n",
1455                        slave_dev->name);
1456                 res = -EPERM;
1457                 goto err_undo_flags;
1458         }
1459
1460         /* set bonding device ether type by slave - bonding netdevices are
1461          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462          * there is a need to override some of the type dependent attribs/funcs.
1463          *
1464          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1466          */
1467         if (bond->slave_cnt == 0) {
1468                 if (slave_dev->type != ARPHRD_ETHER)
1469                         bond_setup_by_slave(bond_dev, slave_dev);
1470         } else if (bond_dev->type != slave_dev->type) {
1471                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1472                         "from other slaves (%d), can not enslave it.\n",
1473                         slave_dev->name,
1474                         slave_dev->type, bond_dev->type);
1475                         res = -EINVAL;
1476                         goto err_undo_flags;
1477         }
1478
1479         if (slave_dev->set_mac_address == NULL) {
1480                 if (bond->slave_cnt == 0) {
1481                         printk(KERN_WARNING DRV_NAME
1482                                ": %s: Warning: The first slave device "
1483                                "specified does not support setting the MAC "
1484                                "address. Setting fail_over_mac to active.",
1485                                bond_dev->name);
1486                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1488                         printk(KERN_ERR DRV_NAME
1489                                 ": %s: Error: The slave device specified "
1490                                 "does not support setting the MAC address, "
1491                                 "but fail_over_mac is not set to active.\n"
1492                                 , bond_dev->name);
1493                         res = -EOPNOTSUPP;
1494                         goto err_undo_flags;
1495                 }
1496         }
1497
1498         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1499         if (!new_slave) {
1500                 res = -ENOMEM;
1501                 goto err_undo_flags;
1502         }
1503
1504         /* save slave's original flags before calling
1505          * netdev_set_master and dev_open
1506          */
1507         new_slave->original_flags = slave_dev->flags;
1508
1509         /*
1510          * Save slave's original ("permanent") mac address for modes
1511          * that need it, and for restoring it upon release, and then
1512          * set it to the master's address
1513          */
1514         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1515
1516         if (!bond->params.fail_over_mac) {
1517                 /*
1518                  * Set slave to master's mac address.  The application already
1519                  * set the master's mac address to that of the first slave
1520                  */
1521                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1522                 addr.sa_family = slave_dev->type;
1523                 res = dev_set_mac_address(slave_dev, &addr);
1524                 if (res) {
1525                         dprintk("Error %d calling set_mac_address\n", res);
1526                         goto err_free;
1527                 }
1528         }
1529
1530         res = netdev_set_master(slave_dev, bond_dev);
1531         if (res) {
1532                 dprintk("Error %d calling netdev_set_master\n", res);
1533                 goto err_restore_mac;
1534         }
1535         /* open the slave since the application closed it */
1536         res = dev_open(slave_dev);
1537         if (res) {
1538                 dprintk("Openning slave %s failed\n", slave_dev->name);
1539                 goto err_unset_master;
1540         }
1541
1542         new_slave->dev = slave_dev;
1543         slave_dev->priv_flags |= IFF_BONDING;
1544
1545         if ((bond->params.mode == BOND_MODE_TLB) ||
1546             (bond->params.mode == BOND_MODE_ALB)) {
1547                 /* bond_alb_init_slave() must be called before all other stages since
1548                  * it might fail and we do not want to have to undo everything
1549                  */
1550                 res = bond_alb_init_slave(bond, new_slave);
1551                 if (res) {
1552                         goto err_close;
1553                 }
1554         }
1555
1556         /* If the mode USES_PRIMARY, then the new slave gets the
1557          * master's promisc (and mc) settings only if it becomes the
1558          * curr_active_slave, and that is taken care of later when calling
1559          * bond_change_active()
1560          */
1561         if (!USES_PRIMARY(bond->params.mode)) {
1562                 /* set promiscuity level to new slave */
1563                 if (bond_dev->flags & IFF_PROMISC) {
1564                         res = dev_set_promiscuity(slave_dev, 1);
1565                         if (res)
1566                                 goto err_close;
1567                 }
1568
1569                 /* set allmulti level to new slave */
1570                 if (bond_dev->flags & IFF_ALLMULTI) {
1571                         res = dev_set_allmulti(slave_dev, 1);
1572                         if (res)
1573                                 goto err_close;
1574                 }
1575
1576                 netif_addr_lock_bh(bond_dev);
1577                 /* upload master's mc_list to new slave */
1578                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1579                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1580                 }
1581                 netif_addr_unlock_bh(bond_dev);
1582         }
1583
1584         if (bond->params.mode == BOND_MODE_8023AD) {
1585                 /* add lacpdu mc addr to mc list */
1586                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1587
1588                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1589         }
1590
1591         bond_add_vlans_on_slave(bond, slave_dev);
1592
1593         write_lock_bh(&bond->lock);
1594
1595         bond_attach_slave(bond, new_slave);
1596
1597         new_slave->delay = 0;
1598         new_slave->link_failure_count = 0;
1599
1600         bond_compute_features(bond);
1601
1602         write_unlock_bh(&bond->lock);
1603
1604         read_lock(&bond->lock);
1605
1606         new_slave->last_arp_rx = jiffies;
1607
1608         if (bond->params.miimon && !bond->params.use_carrier) {
1609                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1610
1611                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1612                         /*
1613                          * miimon is set but a bonded network driver
1614                          * does not support ETHTOOL/MII and
1615                          * arp_interval is not set.  Note: if
1616                          * use_carrier is enabled, we will never go
1617                          * here (because netif_carrier is always
1618                          * supported); thus, we don't need to change
1619                          * the messages for netif_carrier.
1620                          */
1621                         printk(KERN_WARNING DRV_NAME
1622                                ": %s: Warning: MII and ETHTOOL support not "
1623                                "available for interface %s, and "
1624                                "arp_interval/arp_ip_target module parameters "
1625                                "not specified, thus bonding will not detect "
1626                                "link failures! see bonding.txt for details.\n",
1627                                bond_dev->name, slave_dev->name);
1628                 } else if (link_reporting == -1) {
1629                         /* unable get link status using mii/ethtool */
1630                         printk(KERN_WARNING DRV_NAME
1631                                ": %s: Warning: can't get link status from "
1632                                "interface %s; the network driver associated "
1633                                "with this interface does not support MII or "
1634                                "ETHTOOL link status reporting, thus miimon "
1635                                "has no effect on this interface.\n",
1636                                bond_dev->name, slave_dev->name);
1637                 }
1638         }
1639
1640         /* check for initial state */
1641         if (!bond->params.miimon ||
1642             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1643                 if (bond->params.updelay) {
1644                         dprintk("Initial state of slave_dev is "
1645                                 "BOND_LINK_BACK\n");
1646                         new_slave->link  = BOND_LINK_BACK;
1647                         new_slave->delay = bond->params.updelay;
1648                 } else {
1649                         dprintk("Initial state of slave_dev is "
1650                                 "BOND_LINK_UP\n");
1651                         new_slave->link  = BOND_LINK_UP;
1652                 }
1653                 new_slave->jiffies = jiffies;
1654         } else {
1655                 dprintk("Initial state of slave_dev is "
1656                         "BOND_LINK_DOWN\n");
1657                 new_slave->link  = BOND_LINK_DOWN;
1658         }
1659
1660         if (bond_update_speed_duplex(new_slave) &&
1661             (new_slave->link != BOND_LINK_DOWN)) {
1662                 printk(KERN_WARNING DRV_NAME
1663                        ": %s: Warning: failed to get speed and duplex from %s, "
1664                        "assumed to be 100Mb/sec and Full.\n",
1665                        bond_dev->name, new_slave->dev->name);
1666
1667                 if (bond->params.mode == BOND_MODE_8023AD) {
1668                         printk(KERN_WARNING DRV_NAME
1669                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1670                                "support in base driver for proper aggregator "
1671                                "selection.\n", bond_dev->name);
1672                 }
1673         }
1674
1675         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1676                 /* if there is a primary slave, remember it */
1677                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1678                         bond->primary_slave = new_slave;
1679                 }
1680         }
1681
1682         write_lock_bh(&bond->curr_slave_lock);
1683
1684         switch (bond->params.mode) {
1685         case BOND_MODE_ACTIVEBACKUP:
1686                 bond_set_slave_inactive_flags(new_slave);
1687                 bond_select_active_slave(bond);
1688                 break;
1689         case BOND_MODE_8023AD:
1690                 /* in 802.3ad mode, the internal mechanism
1691                  * will activate the slaves in the selected
1692                  * aggregator
1693                  */
1694                 bond_set_slave_inactive_flags(new_slave);
1695                 /* if this is the first slave */
1696                 if (bond->slave_cnt == 1) {
1697                         SLAVE_AD_INFO(new_slave).id = 1;
1698                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1699                          * can be called only after the mac address of the bond is set
1700                          */
1701                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1702                                             bond->params.lacp_fast);
1703                 } else {
1704                         SLAVE_AD_INFO(new_slave).id =
1705                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1706                 }
1707
1708                 bond_3ad_bind_slave(new_slave);
1709                 break;
1710         case BOND_MODE_TLB:
1711         case BOND_MODE_ALB:
1712                 new_slave->state = BOND_STATE_ACTIVE;
1713                 bond_set_slave_inactive_flags(new_slave);
1714                 break;
1715         default:
1716                 dprintk("This slave is always active in trunk mode\n");
1717
1718                 /* always active in trunk mode */
1719                 new_slave->state = BOND_STATE_ACTIVE;
1720
1721                 /* In trunking mode there is little meaning to curr_active_slave
1722                  * anyway (it holds no special properties of the bond device),
1723                  * so we can change it without calling change_active_interface()
1724                  */
1725                 if (!bond->curr_active_slave) {
1726                         bond->curr_active_slave = new_slave;
1727                 }
1728                 break;
1729         } /* switch(bond_mode) */
1730
1731         write_unlock_bh(&bond->curr_slave_lock);
1732
1733         bond_set_carrier(bond);
1734
1735         read_unlock(&bond->lock);
1736
1737         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738         if (res)
1739                 goto err_close;
1740
1741         printk(KERN_INFO DRV_NAME
1742                ": %s: enslaving %s as a%s interface with a%s link.\n",
1743                bond_dev->name, slave_dev->name,
1744                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1745                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1746
1747         /* enslave is successful */
1748         return 0;
1749
1750 /* Undo stages on error */
1751 err_close:
1752         dev_close(slave_dev);
1753
1754 err_unset_master:
1755         netdev_set_master(slave_dev, NULL);
1756
1757 err_restore_mac:
1758         if (!bond->params.fail_over_mac) {
1759                 /* XXX TODO - fom follow mode needs to change master's
1760                  * MAC if this slave's MAC is in use by the bond, or at
1761                  * least print a warning.
1762                  */
1763                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1764                 addr.sa_family = slave_dev->type;
1765                 dev_set_mac_address(slave_dev, &addr);
1766         }
1767
1768 err_free:
1769         kfree(new_slave);
1770
1771 err_undo_flags:
1772         bond_dev->features = old_features;
1773
1774         return res;
1775 }
1776
1777 /*
1778  * Try to release the slave device <slave> from the bond device <master>
1779  * It is legal to access curr_active_slave without a lock because all the function
1780  * is write-locked.
1781  *
1782  * The rules for slave state should be:
1783  *   for Active/Backup:
1784  *     Active stays on all backups go down
1785  *   for Bonded connections:
1786  *     The first up interface should be left on and all others downed.
1787  */
1788 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1789 {
1790         struct bonding *bond = bond_dev->priv;
1791         struct slave *slave, *oldcurrent;
1792         struct sockaddr addr;
1793         int mac_addr_differ;
1794         DECLARE_MAC_BUF(mac);
1795
1796         /* slave is not a slave or master is not master of this slave */
1797         if (!(slave_dev->flags & IFF_SLAVE) ||
1798             (slave_dev->master != bond_dev)) {
1799                 printk(KERN_ERR DRV_NAME
1800                        ": %s: Error: cannot release %s.\n",
1801                        bond_dev->name, slave_dev->name);
1802                 return -EINVAL;
1803         }
1804
1805         write_lock_bh(&bond->lock);
1806
1807         slave = bond_get_slave_by_dev(bond, slave_dev);
1808         if (!slave) {
1809                 /* not a slave of this bond */
1810                 printk(KERN_INFO DRV_NAME
1811                        ": %s: %s not enslaved\n",
1812                        bond_dev->name, slave_dev->name);
1813                 write_unlock_bh(&bond->lock);
1814                 return -EINVAL;
1815         }
1816
1817         if (!bond->params.fail_over_mac) {
1818                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1819                                          ETH_ALEN);
1820                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1821                         printk(KERN_WARNING DRV_NAME
1822                                ": %s: Warning: the permanent HWaddr of %s - "
1823                                "%s - is still in use by %s. "
1824                                "Set the HWaddr of %s to a different address "
1825                                "to avoid conflicts.\n",
1826                                bond_dev->name, slave_dev->name,
1827                                print_mac(mac, slave->perm_hwaddr),
1828                                bond_dev->name, slave_dev->name);
1829         }
1830
1831         /* Inform AD package of unbinding of slave. */
1832         if (bond->params.mode == BOND_MODE_8023AD) {
1833                 /* must be called before the slave is
1834                  * detached from the list
1835                  */
1836                 bond_3ad_unbind_slave(slave);
1837         }
1838
1839         printk(KERN_INFO DRV_NAME
1840                ": %s: releasing %s interface %s\n",
1841                bond_dev->name,
1842                (slave->state == BOND_STATE_ACTIVE)
1843                ? "active" : "backup",
1844                slave_dev->name);
1845
1846         oldcurrent = bond->curr_active_slave;
1847
1848         bond->current_arp_slave = NULL;
1849
1850         /* release the slave from its bond */
1851         bond_detach_slave(bond, slave);
1852
1853         bond_compute_features(bond);
1854
1855         if (bond->primary_slave == slave) {
1856                 bond->primary_slave = NULL;
1857         }
1858
1859         if (oldcurrent == slave) {
1860                 bond_change_active_slave(bond, NULL);
1861         }
1862
1863         if ((bond->params.mode == BOND_MODE_TLB) ||
1864             (bond->params.mode == BOND_MODE_ALB)) {
1865                 /* Must be called only after the slave has been
1866                  * detached from the list and the curr_active_slave
1867                  * has been cleared (if our_slave == old_current),
1868                  * but before a new active slave is selected.
1869                  */
1870                 write_unlock_bh(&bond->lock);
1871                 bond_alb_deinit_slave(bond, slave);
1872                 write_lock_bh(&bond->lock);
1873         }
1874
1875         if (oldcurrent == slave) {
1876                 /*
1877                  * Note that we hold RTNL over this sequence, so there
1878                  * is no concern that another slave add/remove event
1879                  * will interfere.
1880                  */
1881                 write_unlock_bh(&bond->lock);
1882                 read_lock(&bond->lock);
1883                 write_lock_bh(&bond->curr_slave_lock);
1884
1885                 bond_select_active_slave(bond);
1886
1887                 write_unlock_bh(&bond->curr_slave_lock);
1888                 read_unlock(&bond->lock);
1889                 write_lock_bh(&bond->lock);
1890         }
1891
1892         if (bond->slave_cnt == 0) {
1893                 bond_set_carrier(bond);
1894
1895                 /* if the last slave was removed, zero the mac address
1896                  * of the master so it will be set by the application
1897                  * to the mac address of the first slave
1898                  */
1899                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901                 if (list_empty(&bond->vlan_list)) {
1902                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903                 } else {
1904                         printk(KERN_WARNING DRV_NAME
1905                                ": %s: Warning: clearing HW address of %s while it "
1906                                "still has VLANs.\n",
1907                                bond_dev->name, bond_dev->name);
1908                         printk(KERN_WARNING DRV_NAME
1909                                ": %s: When re-adding slaves, make sure the bond's "
1910                                "HW address matches its VLANs'.\n",
1911                                bond_dev->name);
1912                 }
1913         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914                    !bond_has_challenged_slaves(bond)) {
1915                 printk(KERN_INFO DRV_NAME
1916                        ": %s: last VLAN challenged slave %s "
1917                        "left bond %s. VLAN blocking is removed\n",
1918                        bond_dev->name, slave_dev->name, bond_dev->name);
1919                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920         }
1921
1922         write_unlock_bh(&bond->lock);
1923
1924         /* must do this from outside any spinlocks */
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927         bond_del_vlans_from_slave(bond, slave_dev);
1928
1929         /* If the mode USES_PRIMARY, then we should only remove its
1930          * promisc and mc settings if it was the curr_active_slave, but that was
1931          * already taken care of above when we detached the slave
1932          */
1933         if (!USES_PRIMARY(bond->params.mode)) {
1934                 /* unset promiscuity level from slave */
1935                 if (bond_dev->flags & IFF_PROMISC) {
1936                         dev_set_promiscuity(slave_dev, -1);
1937                 }
1938
1939                 /* unset allmulti level from slave */
1940                 if (bond_dev->flags & IFF_ALLMULTI) {
1941                         dev_set_allmulti(slave_dev, -1);
1942                 }
1943
1944                 /* flush master's mc_list from slave */
1945                 netif_addr_lock_bh(bond_dev);
1946                 bond_mc_list_flush(bond_dev, slave_dev);
1947                 netif_addr_unlock_bh(bond_dev);
1948         }
1949
1950         netdev_set_master(slave_dev, NULL);
1951
1952         /* close slave before restoring its mac address */
1953         dev_close(slave_dev);
1954
1955         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1956                 /* restore original ("permanent") mac address */
1957                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958                 addr.sa_family = slave_dev->type;
1959                 dev_set_mac_address(slave_dev, &addr);
1960         }
1961
1962         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1963                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1964                                    IFF_SLAVE_NEEDARP);
1965
1966         kfree(slave);
1967
1968         return 0;  /* deletion OK */
1969 }
1970
1971 /*
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1974 */
1975 void bond_destroy(struct bonding *bond)
1976 {
1977         bond_deinit(bond->dev);
1978         bond_destroy_sysfs_entry(bond);
1979         unregister_netdevice(bond->dev);
1980 }
1981
1982 static void bond_destructor(struct net_device *bond_dev)
1983 {
1984         struct bonding *bond = bond_dev->priv;
1985
1986         if (bond->wq)
1987                 destroy_workqueue(bond->wq);
1988
1989         netif_addr_lock_bh(bond_dev);
1990         bond_mc_list_destroy(bond);
1991         netif_addr_unlock_bh(bond_dev);
1992
1993         free_netdev(bond_dev);
1994 }
1995
1996 /*
1997 * First release a slave and than destroy the bond if no more slaves iare left.
1998 * Must be under rtnl_lock when this function is called.
1999 */
2000 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2001 {
2002         struct bonding *bond = bond_dev->priv;
2003         int ret;
2004
2005         ret = bond_release(bond_dev, slave_dev);
2006         if ((ret == 0) && (bond->slave_cnt == 0)) {
2007                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2008                        bond_dev->name, bond_dev->name);
2009                 bond_destroy(bond);
2010         }
2011         return ret;
2012 }
2013
2014 /*
2015  * This function releases all slaves.
2016  */
2017 static int bond_release_all(struct net_device *bond_dev)
2018 {
2019         struct bonding *bond = bond_dev->priv;
2020         struct slave *slave;
2021         struct net_device *slave_dev;
2022         struct sockaddr addr;
2023
2024         write_lock_bh(&bond->lock);
2025
2026         netif_carrier_off(bond_dev);
2027
2028         if (bond->slave_cnt == 0) {
2029                 goto out;
2030         }
2031
2032         bond->current_arp_slave = NULL;
2033         bond->primary_slave = NULL;
2034         bond_change_active_slave(bond, NULL);
2035
2036         while ((slave = bond->first_slave) != NULL) {
2037                 /* Inform AD package of unbinding of slave
2038                  * before slave is detached from the list.
2039                  */
2040                 if (bond->params.mode == BOND_MODE_8023AD) {
2041                         bond_3ad_unbind_slave(slave);
2042                 }
2043
2044                 slave_dev = slave->dev;
2045                 bond_detach_slave(bond, slave);
2046
2047                 /* now that the slave is detached, unlock and perform
2048                  * all the undo steps that should not be called from
2049                  * within a lock.
2050                  */
2051                 write_unlock_bh(&bond->lock);
2052
2053                 if ((bond->params.mode == BOND_MODE_TLB) ||
2054                     (bond->params.mode == BOND_MODE_ALB)) {
2055                         /* must be called only after the slave
2056                          * has been detached from the list
2057                          */
2058                         bond_alb_deinit_slave(bond, slave);
2059                 }
2060
2061                 bond_compute_features(bond);
2062
2063                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2064                 bond_del_vlans_from_slave(bond, slave_dev);
2065
2066                 /* If the mode USES_PRIMARY, then we should only remove its
2067                  * promisc and mc settings if it was the curr_active_slave, but that was
2068                  * already taken care of above when we detached the slave
2069                  */
2070                 if (!USES_PRIMARY(bond->params.mode)) {
2071                         /* unset promiscuity level from slave */
2072                         if (bond_dev->flags & IFF_PROMISC) {
2073                                 dev_set_promiscuity(slave_dev, -1);
2074                         }
2075
2076                         /* unset allmulti level from slave */
2077                         if (bond_dev->flags & IFF_ALLMULTI) {
2078                                 dev_set_allmulti(slave_dev, -1);
2079                         }
2080
2081                         /* flush master's mc_list from slave */
2082                         netif_addr_lock_bh(bond_dev);
2083                         bond_mc_list_flush(bond_dev, slave_dev);
2084                         netif_addr_unlock_bh(bond_dev);
2085                 }
2086
2087                 netdev_set_master(slave_dev, NULL);
2088
2089                 /* close slave before restoring its mac address */
2090                 dev_close(slave_dev);
2091
2092                 if (!bond->params.fail_over_mac) {
2093                         /* restore original ("permanent") mac address*/
2094                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2095                         addr.sa_family = slave_dev->type;
2096                         dev_set_mac_address(slave_dev, &addr);
2097                 }
2098
2099                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2100                                            IFF_SLAVE_INACTIVE);
2101
2102                 kfree(slave);
2103
2104                 /* re-acquire the lock before getting the next slave */
2105                 write_lock_bh(&bond->lock);
2106         }
2107
2108         /* zero the mac address of the master so it will be
2109          * set by the application to the mac address of the
2110          * first slave
2111          */
2112         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2113
2114         if (list_empty(&bond->vlan_list)) {
2115                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2116         } else {
2117                 printk(KERN_WARNING DRV_NAME
2118                        ": %s: Warning: clearing HW address of %s while it "
2119                        "still has VLANs.\n",
2120                        bond_dev->name, bond_dev->name);
2121                 printk(KERN_WARNING DRV_NAME
2122                        ": %s: When re-adding slaves, make sure the bond's "
2123                        "HW address matches its VLANs'.\n",
2124                        bond_dev->name);
2125         }
2126
2127         printk(KERN_INFO DRV_NAME
2128                ": %s: released all slaves\n",
2129                bond_dev->name);
2130
2131 out:
2132         write_unlock_bh(&bond->lock);
2133
2134         return 0;
2135 }
2136
2137 /*
2138  * This function changes the active slave to slave <slave_dev>.
2139  * It returns -EINVAL in the following cases.
2140  *  - <slave_dev> is not found in the list.
2141  *  - There is not active slave now.
2142  *  - <slave_dev> is already active.
2143  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2144  *  - <slave_dev> is not running.
2145  * In these cases, this fuction does nothing.
2146  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2147  */
2148 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2149 {
2150         struct bonding *bond = bond_dev->priv;
2151         struct slave *old_active = NULL;
2152         struct slave *new_active = NULL;
2153         int res = 0;
2154
2155         if (!USES_PRIMARY(bond->params.mode)) {
2156                 return -EINVAL;
2157         }
2158
2159         /* Verify that master_dev is indeed the master of slave_dev */
2160         if (!(slave_dev->flags & IFF_SLAVE) ||
2161             (slave_dev->master != bond_dev)) {
2162                 return -EINVAL;
2163         }
2164
2165         read_lock(&bond->lock);
2166
2167         read_lock(&bond->curr_slave_lock);
2168         old_active = bond->curr_active_slave;
2169         read_unlock(&bond->curr_slave_lock);
2170
2171         new_active = bond_get_slave_by_dev(bond, slave_dev);
2172
2173         /*
2174          * Changing to the current active: do nothing; return success.
2175          */
2176         if (new_active && (new_active == old_active)) {
2177                 read_unlock(&bond->lock);
2178                 return 0;
2179         }
2180
2181         if ((new_active) &&
2182             (old_active) &&
2183             (new_active->link == BOND_LINK_UP) &&
2184             IS_UP(new_active->dev)) {
2185                 write_lock_bh(&bond->curr_slave_lock);
2186                 bond_change_active_slave(bond, new_active);
2187                 write_unlock_bh(&bond->curr_slave_lock);
2188         } else {
2189                 res = -EINVAL;
2190         }
2191
2192         read_unlock(&bond->lock);
2193
2194         return res;
2195 }
2196
2197 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2198 {
2199         struct bonding *bond = bond_dev->priv;
2200
2201         info->bond_mode = bond->params.mode;
2202         info->miimon = bond->params.miimon;
2203
2204         read_lock(&bond->lock);
2205         info->num_slaves = bond->slave_cnt;
2206         read_unlock(&bond->lock);
2207
2208         return 0;
2209 }
2210
2211 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2212 {
2213         struct bonding *bond = bond_dev->priv;
2214         struct slave *slave;
2215         int i, found = 0;
2216
2217         if (info->slave_id < 0) {
2218                 return -ENODEV;
2219         }
2220
2221         read_lock(&bond->lock);
2222
2223         bond_for_each_slave(bond, slave, i) {
2224                 if (i == (int)info->slave_id) {
2225                         found = 1;
2226                         break;
2227                 }
2228         }
2229
2230         read_unlock(&bond->lock);
2231
2232         if (found) {
2233                 strcpy(info->slave_name, slave->dev->name);
2234                 info->link = slave->link;
2235                 info->state = slave->state;
2236                 info->link_failure_count = slave->link_failure_count;
2237         } else {
2238                 return -ENODEV;
2239         }
2240
2241         return 0;
2242 }
2243
2244 /*-------------------------------- Monitoring -------------------------------*/
2245
2246
2247 static int bond_miimon_inspect(struct bonding *bond)
2248 {
2249         struct slave *slave;
2250         int i, link_state, commit = 0;
2251
2252         bond_for_each_slave(bond, slave, i) {
2253                 slave->new_link = BOND_LINK_NOCHANGE;
2254
2255                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2256
2257                 switch (slave->link) {
2258                 case BOND_LINK_UP:
2259                         if (link_state)
2260                                 continue;
2261
2262                         slave->link = BOND_LINK_FAIL;
2263                         slave->delay = bond->params.downdelay;
2264                         if (slave->delay) {
2265                                 printk(KERN_INFO DRV_NAME
2266                                        ": %s: link status down for %s"
2267                                        "interface %s, disabling it in %d ms.\n",
2268                                        bond->dev->name,
2269                                        (bond->params.mode ==
2270                                         BOND_MODE_ACTIVEBACKUP) ?
2271                                        ((slave->state == BOND_STATE_ACTIVE) ?
2272                                         "active " : "backup ") : "",
2273                                        slave->dev->name,
2274                                        bond->params.downdelay * bond->params.miimon);
2275                         }
2276                         /*FALLTHRU*/
2277                 case BOND_LINK_FAIL:
2278                         if (link_state) {
2279                                 /*
2280                                  * recovered before downdelay expired
2281                                  */
2282                                 slave->link = BOND_LINK_UP;
2283                                 slave->jiffies = jiffies;
2284                                 printk(KERN_INFO DRV_NAME
2285                                        ": %s: link status up again after %d "
2286                                        "ms for interface %s.\n",
2287                                        bond->dev->name,
2288                                        (bond->params.downdelay - slave->delay) *
2289                                        bond->params.miimon,
2290                                        slave->dev->name);
2291                                 continue;
2292                         }
2293
2294                         if (slave->delay <= 0) {
2295                                 slave->new_link = BOND_LINK_DOWN;
2296                                 commit++;
2297                                 continue;
2298                         }
2299
2300                         slave->delay--;
2301                         break;
2302
2303                 case BOND_LINK_DOWN:
2304                         if (!link_state)
2305                                 continue;
2306
2307                         slave->link = BOND_LINK_BACK;
2308                         slave->delay = bond->params.updelay;
2309
2310                         if (slave->delay) {
2311                                 printk(KERN_INFO DRV_NAME
2312                                        ": %s: link status up for "
2313                                        "interface %s, enabling it in %d ms.\n",
2314                                        bond->dev->name, slave->dev->name,
2315                                        bond->params.updelay *
2316                                        bond->params.miimon);
2317                         }
2318                         /*FALLTHRU*/
2319                 case BOND_LINK_BACK:
2320                         if (!link_state) {
2321                                 slave->link = BOND_LINK_DOWN;
2322                                 printk(KERN_INFO DRV_NAME
2323                                        ": %s: link status down again after %d "
2324                                        "ms for interface %s.\n",
2325                                        bond->dev->name,
2326                                        (bond->params.updelay - slave->delay) *
2327                                        bond->params.miimon,
2328                                        slave->dev->name);
2329
2330                                 continue;
2331                         }
2332
2333                         if (slave->delay <= 0) {
2334                                 slave->new_link = BOND_LINK_UP;
2335                                 commit++;
2336                                 continue;
2337                         }
2338
2339                         slave->delay--;
2340                         break;
2341                 }
2342         }
2343
2344         return commit;
2345 }
2346
2347 static void bond_miimon_commit(struct bonding *bond)
2348 {
2349         struct slave *slave;
2350         int i;
2351
2352         bond_for_each_slave(bond, slave, i) {
2353                 switch (slave->new_link) {
2354                 case BOND_LINK_NOCHANGE:
2355                         continue;
2356
2357                 case BOND_LINK_UP:
2358                         slave->link = BOND_LINK_UP;
2359                         slave->jiffies = jiffies;
2360
2361                         if (bond->params.mode == BOND_MODE_8023AD) {
2362                                 /* prevent it from being the active one */
2363                                 slave->state = BOND_STATE_BACKUP;
2364                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2365                                 /* make it immediately active */
2366                                 slave->state = BOND_STATE_ACTIVE;
2367                         } else if (slave != bond->primary_slave) {
2368                                 /* prevent it from being the active one */
2369                                 slave->state = BOND_STATE_BACKUP;
2370                         }
2371
2372                         printk(KERN_INFO DRV_NAME
2373                                ": %s: link status definitely "
2374                                "up for interface %s.\n",
2375                                bond->dev->name, slave->dev->name);
2376
2377                         /* notify ad that the link status has changed */
2378                         if (bond->params.mode == BOND_MODE_8023AD)
2379                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2380
2381                         if ((bond->params.mode == BOND_MODE_TLB) ||
2382                             (bond->params.mode == BOND_MODE_ALB))
2383                                 bond_alb_handle_link_change(bond, slave,
2384                                                             BOND_LINK_UP);
2385
2386                         if (!bond->curr_active_slave ||
2387                             (slave == bond->primary_slave))
2388                                 goto do_failover;
2389
2390                         continue;
2391
2392                 case BOND_LINK_DOWN:
2393                         if (slave->link_failure_count < UINT_MAX)
2394                                 slave->link_failure_count++;
2395
2396                         slave->link = BOND_LINK_DOWN;
2397
2398                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2399                             bond->params.mode == BOND_MODE_8023AD)
2400                                 bond_set_slave_inactive_flags(slave);
2401
2402                         printk(KERN_INFO DRV_NAME
2403                                ": %s: link status definitely down for "
2404                                "interface %s, disabling it\n",
2405                                bond->dev->name, slave->dev->name);
2406
2407                         if (bond->params.mode == BOND_MODE_8023AD)
2408                                 bond_3ad_handle_link_change(slave,
2409                                                             BOND_LINK_DOWN);
2410
2411                         if (bond->params.mode == BOND_MODE_TLB ||
2412                             bond->params.mode == BOND_MODE_ALB)
2413                                 bond_alb_handle_link_change(bond, slave,
2414                                                             BOND_LINK_DOWN);
2415
2416                         if (slave == bond->curr_active_slave)
2417                                 goto do_failover;
2418
2419                         continue;
2420
2421                 default:
2422                         printk(KERN_ERR DRV_NAME
2423                                ": %s: invalid new link %d on slave %s\n",
2424                                bond->dev->name, slave->new_link,
2425                                slave->dev->name);
2426                         slave->new_link = BOND_LINK_NOCHANGE;
2427
2428                         continue;
2429                 }
2430
2431 do_failover:
2432                 ASSERT_RTNL();
2433                 write_lock_bh(&bond->curr_slave_lock);
2434                 bond_select_active_slave(bond);
2435                 write_unlock_bh(&bond->curr_slave_lock);
2436         }
2437
2438         bond_set_carrier(bond);
2439 }
2440
2441 /*
2442  * bond_mii_monitor
2443  *
2444  * Really a wrapper that splits the mii monitor into two phases: an
2445  * inspection, then (if inspection indicates something needs to be done)
2446  * an acquisition of appropriate locks followed by a commit phase to
2447  * implement whatever link state changes are indicated.
2448  */
2449 void bond_mii_monitor(struct work_struct *work)
2450 {
2451         struct bonding *bond = container_of(work, struct bonding,
2452                                             mii_work.work);
2453
2454         read_lock(&bond->lock);
2455         if (bond->kill_timers)
2456                 goto out;
2457
2458         if (bond->slave_cnt == 0)
2459                 goto re_arm;
2460
2461         if (bond->send_grat_arp) {
2462                 read_lock(&bond->curr_slave_lock);
2463                 bond_send_gratuitous_arp(bond);
2464                 read_unlock(&bond->curr_slave_lock);
2465         }
2466
2467         if (bond_miimon_inspect(bond)) {
2468                 read_unlock(&bond->lock);
2469                 rtnl_lock();
2470                 read_lock(&bond->lock);
2471
2472                 bond_miimon_commit(bond);
2473
2474                 read_unlock(&bond->lock);
2475                 rtnl_unlock();  /* might sleep, hold no other locks */
2476                 read_lock(&bond->lock);
2477         }
2478
2479 re_arm:
2480         if (bond->params.miimon)
2481                 queue_delayed_work(bond->wq, &bond->mii_work,
2482                                    msecs_to_jiffies(bond->params.miimon));
2483 out:
2484         read_unlock(&bond->lock);
2485 }
2486
2487 static __be32 bond_glean_dev_ip(struct net_device *dev)
2488 {
2489         struct in_device *idev;
2490         struct in_ifaddr *ifa;
2491         __be32 addr = 0;
2492
2493         if (!dev)
2494                 return 0;
2495
2496         rcu_read_lock();
2497         idev = __in_dev_get_rcu(dev);
2498         if (!idev)
2499                 goto out;
2500
2501         ifa = idev->ifa_list;
2502         if (!ifa)
2503                 goto out;
2504
2505         addr = ifa->ifa_local;
2506 out:
2507         rcu_read_unlock();
2508         return addr;
2509 }
2510
2511 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2512 {
2513         struct vlan_entry *vlan;
2514
2515         if (ip == bond->master_ip)
2516                 return 1;
2517
2518         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2519                 if (ip == vlan->vlan_ip)
2520                         return 1;
2521         }
2522
2523         return 0;
2524 }
2525
2526 /*
2527  * We go to the (large) trouble of VLAN tagging ARP frames because
2528  * switches in VLAN mode (especially if ports are configured as
2529  * "native" to a VLAN) might not pass non-tagged frames.
2530  */
2531 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2532 {
2533         struct sk_buff *skb;
2534
2535         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2536                slave_dev->name, dest_ip, src_ip, vlan_id);
2537
2538         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2539                          NULL, slave_dev->dev_addr, NULL);
2540
2541         if (!skb) {
2542                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2543                 return;
2544         }
2545         if (vlan_id) {
2546                 skb = vlan_put_tag(skb, vlan_id);
2547                 if (!skb) {
2548                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2549                         return;
2550                 }
2551         }
2552         arp_xmit(skb);
2553 }
2554
2555
2556 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2557 {
2558         int i, vlan_id, rv;
2559         __be32 *targets = bond->params.arp_targets;
2560         struct vlan_entry *vlan;
2561         struct net_device *vlan_dev;
2562         struct flowi fl;
2563         struct rtable *rt;
2564
2565         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2566                 if (!targets[i])
2567                         continue;
2568                 dprintk("basa: target %x\n", targets[i]);
2569                 if (list_empty(&bond->vlan_list)) {
2570                         dprintk("basa: empty vlan: arp_send\n");
2571                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2572                                       bond->master_ip, 0);
2573                         continue;
2574                 }
2575
2576                 /*
2577                  * If VLANs are configured, we do a route lookup to
2578                  * determine which VLAN interface would be used, so we
2579                  * can tag the ARP with the proper VLAN tag.
2580                  */
2581                 memset(&fl, 0, sizeof(fl));
2582                 fl.fl4_dst = targets[i];
2583                 fl.fl4_tos = RTO_ONLINK;
2584
2585                 rv = ip_route_output_key(&init_net, &rt, &fl);
2586                 if (rv) {
2587                         if (net_ratelimit()) {
2588                                 printk(KERN_WARNING DRV_NAME
2589                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2590                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2591                         }
2592                         continue;
2593                 }
2594
2595                 /*
2596                  * This target is not on a VLAN
2597                  */
2598                 if (rt->u.dst.dev == bond->dev) {
2599                         ip_rt_put(rt);
2600                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2601                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2602                                       bond->master_ip, 0);
2603                         continue;
2604                 }
2605
2606                 vlan_id = 0;
2607                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2608                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2609                         if (vlan_dev == rt->u.dst.dev) {
2610                                 vlan_id = vlan->vlan_id;
2611                                 dprintk("basa: vlan match on %s %d\n",
2612                                        vlan_dev->name, vlan_id);
2613                                 break;
2614                         }
2615                 }
2616
2617                 if (vlan_id) {
2618                         ip_rt_put(rt);
2619                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2620                                       vlan->vlan_ip, vlan_id);
2621                         continue;
2622                 }
2623
2624                 if (net_ratelimit()) {
2625                         printk(KERN_WARNING DRV_NAME
2626                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2627                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2628                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2629                 }
2630                 ip_rt_put(rt);
2631         }
2632 }
2633
2634 /*
2635  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2636  * for each VLAN above us.
2637  *
2638  * Caller must hold curr_slave_lock for read or better
2639  */
2640 static void bond_send_gratuitous_arp(struct bonding *bond)
2641 {
2642         struct slave *slave = bond->curr_active_slave;
2643         struct vlan_entry *vlan;
2644         struct net_device *vlan_dev;
2645
2646         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2647                                 slave ? slave->dev->name : "NULL");
2648
2649         if (!slave || !bond->send_grat_arp ||
2650             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2651                 return;
2652
2653         bond->send_grat_arp--;
2654
2655         if (bond->master_ip) {
2656                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2657                                 bond->master_ip, 0);
2658         }
2659
2660         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2661                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2662                 if (vlan->vlan_ip) {
2663                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2664                                       vlan->vlan_ip, vlan->vlan_id);
2665                 }
2666         }
2667 }
2668
2669 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2670 {
2671         int i;
2672         __be32 *targets = bond->params.arp_targets;
2673
2674         targets = bond->params.arp_targets;
2675         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2676                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2677                         "%u.%u.%u.%u bhti(tip) %d\n",
2678                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2679                        bond_has_this_ip(bond, tip));
2680                 if (sip == targets[i]) {
2681                         if (bond_has_this_ip(bond, tip))
2682                                 slave->last_arp_rx = jiffies;
2683                         return;
2684                 }
2685         }
2686 }
2687
2688 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2689 {
2690         struct arphdr *arp;
2691         struct slave *slave;
2692         struct bonding *bond;
2693         unsigned char *arp_ptr;
2694         __be32 sip, tip;
2695
2696         if (dev_net(dev) != &init_net)
2697                 goto out;
2698
2699         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2700                 goto out;
2701
2702         bond = dev->priv;
2703         read_lock(&bond->lock);
2704
2705         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2706                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2707                 orig_dev ? orig_dev->name : "NULL");
2708
2709         slave = bond_get_slave_by_dev(bond, orig_dev);
2710         if (!slave || !slave_do_arp_validate(bond, slave))
2711                 goto out_unlock;
2712
2713         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2714                 goto out_unlock;
2715
2716         arp = arp_hdr(skb);
2717         if (arp->ar_hln != dev->addr_len ||
2718             skb->pkt_type == PACKET_OTHERHOST ||
2719             skb->pkt_type == PACKET_LOOPBACK ||
2720             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2721             arp->ar_pro != htons(ETH_P_IP) ||
2722             arp->ar_pln != 4)
2723                 goto out_unlock;
2724
2725         arp_ptr = (unsigned char *)(arp + 1);
2726         arp_ptr += dev->addr_len;
2727         memcpy(&sip, arp_ptr, 4);
2728         arp_ptr += 4 + dev->addr_len;
2729         memcpy(&tip, arp_ptr, 4);
2730
2731         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2732                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2733                 slave->state, bond->params.arp_validate,
2734                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2735
2736         /*
2737          * Backup slaves won't see the ARP reply, but do come through
2738          * here for each ARP probe (so we swap the sip/tip to validate
2739          * the probe).  In a "redundant switch, common router" type of
2740          * configuration, the ARP probe will (hopefully) travel from
2741          * the active, through one switch, the router, then the other
2742          * switch before reaching the backup.
2743          */
2744         if (slave->state == BOND_STATE_ACTIVE)
2745                 bond_validate_arp(bond, slave, sip, tip);
2746         else
2747                 bond_validate_arp(bond, slave, tip, sip);
2748
2749 out_unlock:
2750         read_unlock(&bond->lock);
2751 out:
2752         dev_kfree_skb(skb);
2753         return NET_RX_SUCCESS;
2754 }
2755
2756 /*
2757  * this function is called regularly to monitor each slave's link
2758  * ensuring that traffic is being sent and received when arp monitoring
2759  * is used in load-balancing mode. if the adapter has been dormant, then an
2760  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2761  * arp monitoring in active backup mode.
2762  */
2763 void bond_loadbalance_arp_mon(struct work_struct *work)
2764 {
2765         struct bonding *bond = container_of(work, struct bonding,
2766                                             arp_work.work);
2767         struct slave *slave, *oldcurrent;
2768         int do_failover = 0;
2769         int delta_in_ticks;
2770         int i;
2771
2772         read_lock(&bond->lock);
2773
2774         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2775
2776         if (bond->kill_timers) {
2777                 goto out;
2778         }
2779
2780         if (bond->slave_cnt == 0) {
2781                 goto re_arm;
2782         }
2783
2784         read_lock(&bond->curr_slave_lock);
2785         oldcurrent = bond->curr_active_slave;
2786         read_unlock(&bond->curr_slave_lock);
2787
2788         /* see if any of the previous devices are up now (i.e. they have
2789          * xmt and rcv traffic). the curr_active_slave does not come into
2790          * the picture unless it is null. also, slave->jiffies is not needed
2791          * here because we send an arp on each slave and give a slave as
2792          * long as it needs to get the tx/rx within the delta.
2793          * TODO: what about up/down delay in arp mode? it wasn't here before
2794          *       so it can wait
2795          */
2796         bond_for_each_slave(bond, slave, i) {
2797                 if (slave->link != BOND_LINK_UP) {
2798                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2799                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2800
2801                                 slave->link  = BOND_LINK_UP;
2802                                 slave->state = BOND_STATE_ACTIVE;
2803
2804                                 /* primary_slave has no meaning in round-robin
2805                                  * mode. the window of a slave being up and
2806                                  * curr_active_slave being null after enslaving
2807                                  * is closed.
2808                                  */
2809                                 if (!oldcurrent) {
2810                                         printk(KERN_INFO DRV_NAME
2811                                                ": %s: link status definitely "
2812                                                "up for interface %s, ",
2813                                                bond->dev->name,
2814                                                slave->dev->name);
2815                                         do_failover = 1;
2816                                 } else {
2817                                         printk(KERN_INFO DRV_NAME
2818                                                ": %s: interface %s is now up\n",
2819                                                bond->dev->name,
2820                                                slave->dev->name);
2821                                 }
2822                         }
2823                 } else {
2824                         /* slave->link == BOND_LINK_UP */
2825
2826                         /* not all switches will respond to an arp request
2827                          * when the source ip is 0, so don't take the link down
2828                          * if we don't know our ip yet
2829                          */
2830                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2831                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2832
2833                                 slave->link  = BOND_LINK_DOWN;
2834                                 slave->state = BOND_STATE_BACKUP;
2835
2836                                 if (slave->link_failure_count < UINT_MAX) {
2837                                         slave->link_failure_count++;
2838                                 }
2839
2840                                 printk(KERN_INFO DRV_NAME
2841                                        ": %s: interface %s is now down.\n",
2842                                        bond->dev->name,
2843                                        slave->dev->name);
2844
2845                                 if (slave == oldcurrent) {
2846                                         do_failover = 1;
2847                                 }
2848                         }
2849                 }
2850
2851                 /* note: if switch is in round-robin mode, all links
2852                  * must tx arp to ensure all links rx an arp - otherwise
2853                  * links may oscillate or not come up at all; if switch is
2854                  * in something like xor mode, there is nothing we can
2855                  * do - all replies will be rx'ed on same link causing slaves
2856                  * to be unstable during low/no traffic periods
2857                  */
2858                 if (IS_UP(slave->dev)) {
2859                         bond_arp_send_all(bond, slave);
2860                 }
2861         }
2862
2863         if (do_failover) {
2864                 write_lock_bh(&bond->curr_slave_lock);
2865
2866                 bond_select_active_slave(bond);
2867
2868                 write_unlock_bh(&bond->curr_slave_lock);
2869         }
2870
2871 re_arm:
2872         if (bond->params.arp_interval)
2873                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2874 out:
2875         read_unlock(&bond->lock);
2876 }
2877
2878 /*
2879  * Called to inspect slaves for active-backup mode ARP monitor link state
2880  * changes.  Sets new_link in slaves to specify what action should take
2881  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2882  * to link states must be committed.
2883  *
2884  * Called with bond->lock held for read.
2885  */
2886 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2887 {
2888         struct slave *slave;
2889         int i, commit = 0;
2890
2891         bond_for_each_slave(bond, slave, i) {
2892                 slave->new_link = BOND_LINK_NOCHANGE;
2893
2894                 if (slave->link != BOND_LINK_UP) {
2895                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2896                                            delta_in_ticks)) {
2897                                 slave->new_link = BOND_LINK_UP;
2898                                 commit++;
2899                         }
2900
2901                         continue;
2902                 }
2903
2904                 /*
2905                  * Give slaves 2*delta after being enslaved or made
2906                  * active.  This avoids bouncing, as the last receive
2907                  * times need a full ARP monitor cycle to be updated.
2908                  */
2909                 if (!time_after_eq(jiffies, slave->jiffies +
2910                                    2 * delta_in_ticks))
2911                         continue;
2912
2913                 /*
2914                  * Backup slave is down if:
2915                  * - No current_arp_slave AND
2916                  * - more than 3*delta since last receive AND
2917                  * - the bond has an IP address
2918                  *
2919                  * Note: a non-null current_arp_slave indicates
2920                  * the curr_active_slave went down and we are
2921                  * searching for a new one; under this condition
2922                  * we only take the curr_active_slave down - this
2923                  * gives each slave a chance to tx/rx traffic
2924                  * before being taken out
2925                  */
2926                 if (slave->state == BOND_STATE_BACKUP &&
2927                     !bond->current_arp_slave &&
2928                     time_after(jiffies, slave_last_rx(bond, slave) +
2929                                3 * delta_in_ticks)) {
2930                         slave->new_link = BOND_LINK_DOWN;
2931                         commit++;
2932                 }
2933
2934                 /*
2935                  * Active slave is down if:
2936                  * - more than 2*delta since transmitting OR
2937                  * - (more than 2*delta since receive AND
2938                  *    the bond has an IP address)
2939                  */
2940                 if ((slave->state == BOND_STATE_ACTIVE) &&
2941                     (time_after_eq(jiffies, slave->dev->trans_start +
2942                                     2 * delta_in_ticks) ||
2943                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2944                                      + 2 * delta_in_ticks)))) {
2945                         slave->new_link = BOND_LINK_DOWN;
2946                         commit++;
2947                 }
2948         }
2949
2950         read_lock(&bond->curr_slave_lock);
2951
2952         /*
2953          * Trigger a commit if the primary option setting has changed.
2954          */
2955         if (bond->primary_slave &&
2956             (bond->primary_slave != bond->curr_active_slave) &&
2957             (bond->primary_slave->link == BOND_LINK_UP))
2958                 commit++;
2959
2960         read_unlock(&bond->curr_slave_lock);
2961
2962         return commit;
2963 }
2964
2965 /*
2966  * Called to commit link state changes noted by inspection step of
2967  * active-backup mode ARP monitor.
2968  *
2969  * Called with RTNL and bond->lock for read.
2970  */
2971 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2972 {
2973         struct slave *slave;
2974         int i;
2975
2976         bond_for_each_slave(bond, slave, i) {
2977                 switch (slave->new_link) {
2978                 case BOND_LINK_NOCHANGE:
2979                         continue;
2980
2981                 case BOND_LINK_UP:
2982                         write_lock_bh(&bond->curr_slave_lock);
2983
2984                         if (!bond->curr_active_slave &&
2985                             time_before_eq(jiffies, slave->dev->trans_start +
2986                                            delta_in_ticks)) {
2987                                 slave->link = BOND_LINK_UP;
2988                                 bond_change_active_slave(bond, slave);
2989                                 bond->current_arp_slave = NULL;
2990
2991                                 printk(KERN_INFO DRV_NAME
2992                                        ": %s: %s is up and now the "
2993                                        "active interface\n",
2994                                        bond->dev->name, slave->dev->name);
2995
2996                         } else if (bond->curr_active_slave != slave) {
2997                                 /* this slave has just come up but we
2998                                  * already have a current slave; this can
2999                                  * also happen if bond_enslave adds a new
3000                                  * slave that is up while we are searching
3001                                  * for a new slave
3002                                  */
3003                                 slave->link = BOND_LINK_UP;
3004                                 bond_set_slave_inactive_flags(slave);
3005                                 bond->current_arp_slave = NULL;
3006
3007                                 printk(KERN_INFO DRV_NAME
3008                                        ": %s: backup interface %s is now up\n",
3009                                        bond->dev->name, slave->dev->name);
3010                         }
3011
3012                         write_unlock_bh(&bond->curr_slave_lock);
3013
3014                         break;
3015
3016                 case BOND_LINK_DOWN:
3017                         if (slave->link_failure_count < UINT_MAX)
3018                                 slave->link_failure_count++;
3019
3020                         slave->link = BOND_LINK_DOWN;
3021
3022                         if (slave == bond->curr_active_slave) {
3023                                 printk(KERN_INFO DRV_NAME
3024                                        ": %s: link status down for active "
3025                                        "interface %s, disabling it\n",
3026                                        bond->dev->name, slave->dev->name);
3027
3028                                 bond_set_slave_inactive_flags(slave);
3029
3030                                 write_lock_bh(&bond->curr_slave_lock);
3031
3032                                 bond_select_active_slave(bond);
3033                                 if (bond->curr_active_slave)
3034                                         bond->curr_active_slave->jiffies =
3035                                                 jiffies;
3036
3037                                 write_unlock_bh(&bond->curr_slave_lock);
3038
3039                                 bond->current_arp_slave = NULL;
3040
3041                         } else if (slave->state == BOND_STATE_BACKUP) {
3042                                 printk(KERN_INFO DRV_NAME
3043                                        ": %s: backup interface %s is now down\n",
3044                                        bond->dev->name, slave->dev->name);
3045
3046                                 bond_set_slave_inactive_flags(slave);
3047                         }
3048                         break;
3049
3050                 default:
3051                         printk(KERN_ERR DRV_NAME
3052                                ": %s: impossible: new_link %d on slave %s\n",
3053                                bond->dev->name, slave->new_link,
3054                                slave->dev->name);
3055                 }
3056         }
3057
3058         /*
3059          * No race with changes to primary via sysfs, as we hold rtnl.
3060          */
3061         if (bond->primary_slave &&
3062             (bond->primary_slave != bond->curr_active_slave) &&
3063             (bond->primary_slave->link == BOND_LINK_UP)) {
3064                 write_lock_bh(&bond->curr_slave_lock);
3065                 bond_change_active_slave(bond, bond->primary_slave);
3066                 write_unlock_bh(&bond->curr_slave_lock);
3067         }
3068
3069         bond_set_carrier(bond);
3070 }
3071
3072 /*
3073  * Send ARP probes for active-backup mode ARP monitor.
3074  *
3075  * Called with bond->lock held for read.
3076  */
3077 static void bond_ab_arp_probe(struct bonding *bond)
3078 {
3079         struct slave *slave;
3080         int i;
3081
3082         read_lock(&bond->curr_slave_lock);
3083
3084         if (bond->current_arp_slave && bond->curr_active_slave)
3085                 printk("PROBE: c_arp %s && cas %s BAD\n",
3086                        bond->current_arp_slave->dev->name,
3087                        bond->curr_active_slave->dev->name);
3088
3089         if (bond->curr_active_slave) {
3090                 bond_arp_send_all(bond, bond->curr_active_slave);
3091                 read_unlock(&bond->curr_slave_lock);
3092                 return;
3093         }
3094
3095         read_unlock(&bond->curr_slave_lock);
3096
3097         /* if we don't have a curr_active_slave, search for the next available
3098          * backup slave from the current_arp_slave and make it the candidate
3099          * for becoming the curr_active_slave
3100          */
3101
3102         if (!bond->current_arp_slave) {
3103                 bond->current_arp_slave = bond->first_slave;
3104                 if (!bond->current_arp_slave)
3105                         return;
3106         }
3107
3108         bond_set_slave_inactive_flags(bond->current_arp_slave);
3109
3110         /* search for next candidate */
3111         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3112                 if (IS_UP(slave->dev)) {
3113                         slave->link = BOND_LINK_BACK;
3114                         bond_set_slave_active_flags(slave);
3115                         bond_arp_send_all(bond, slave);
3116                         slave->jiffies = jiffies;
3117                         bond->current_arp_slave = slave;
3118                         break;
3119                 }
3120
3121                 /* if the link state is up at this point, we
3122                  * mark it down - this can happen if we have
3123                  * simultaneous link failures and
3124                  * reselect_active_interface doesn't make this
3125                  * one the current slave so it is still marked
3126                  * up when it is actually down
3127                  */
3128                 if (slave->link == BOND_LINK_UP) {
3129                         slave->link = BOND_LINK_DOWN;
3130                         if (slave->link_failure_count < UINT_MAX)
3131                                 slave->link_failure_count++;
3132
3133                         bond_set_slave_inactive_flags(slave);
3134
3135                         printk(KERN_INFO DRV_NAME
3136                                ": %s: backup interface %s is now down.\n",
3137                                bond->dev->name, slave->dev->name);
3138                 }
3139         }
3140 }
3141
3142 void bond_activebackup_arp_mon(struct work_struct *work)
3143 {
3144         struct bonding *bond = container_of(work, struct bonding,
3145                                             arp_work.work);
3146         int delta_in_ticks;
3147
3148         read_lock(&bond->lock);
3149
3150         if (bond->kill_timers)
3151                 goto out;
3152
3153         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3154
3155         if (bond->slave_cnt == 0)
3156                 goto re_arm;
3157
3158         if (bond->send_grat_arp) {
3159                 read_lock(&bond->curr_slave_lock);
3160                 bond_send_gratuitous_arp(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 {
3192         struct bonding *bond = seq->private;
3193         loff_t off = 0;
3194         struct slave *slave;
3195         int i;
3196
3197         /* make sure the bond won't be taken away */
3198         read_lock(&dev_base_lock);
3199         read_lock(&bond->lock);
3200
3201         if (*pos == 0) {
3202                 return SEQ_START_TOKEN;
3203         }
3204
3205         bond_for_each_slave(bond, slave, i) {
3206                 if (++off == *pos) {
3207                         return slave;
3208                 }
3209         }
3210
3211         return NULL;
3212 }
3213
3214 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3215 {
3216         struct bonding *bond = seq->private;
3217         struct slave *slave = v;
3218
3219         ++*pos;
3220         if (v == SEQ_START_TOKEN) {
3221                 return bond->first_slave;
3222         }
3223
3224         slave = slave->next;
3225
3226         return (slave == bond->first_slave) ? NULL : slave;
3227 }
3228
3229 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3230 {
3231         struct bonding *bond = seq->private;
3232
3233         read_unlock(&bond->lock);
3234         read_unlock(&dev_base_lock);
3235 }
3236
3237 static void bond_info_show_master(struct seq_file *seq)
3238 {
3239         struct bonding *bond = seq->private;
3240         struct slave *curr;
3241         int i;
3242         u32 target;
3243
3244         read_lock(&bond->curr_slave_lock);
3245         curr = bond->curr_active_slave;
3246         read_unlock(&bond->curr_slave_lock);
3247
3248         seq_printf(seq, "Bonding Mode: %s",
3249                    bond_mode_name(bond->params.mode));
3250
3251         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3252             bond->params.fail_over_mac)
3253                 seq_printf(seq, " (fail_over_mac %s)",
3254                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3255
3256         seq_printf(seq, "\n");
3257
3258         if (bond->params.mode == BOND_MODE_XOR ||
3259                 bond->params.mode == BOND_MODE_8023AD) {
3260                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3261                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3262                         bond->params.xmit_policy);
3263         }
3264
3265         if (USES_PRIMARY(bond->params.mode)) {
3266                 seq_printf(seq, "Primary Slave: %s\n",
3267                            (bond->primary_slave) ?
3268                            bond->primary_slave->dev->name : "None");
3269
3270                 seq_printf(seq, "Currently Active Slave: %s\n",
3271                            (curr) ? curr->dev->name : "None");
3272         }
3273
3274         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3275                    "up" : "down");
3276         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3277         seq_printf(seq, "Up Delay (ms): %d\n",
3278                    bond->params.updelay * bond->params.miimon);
3279         seq_printf(seq, "Down Delay (ms): %d\n",
3280                    bond->params.downdelay * bond->params.miimon);
3281
3282
3283         /* ARP information */
3284         if(bond->params.arp_interval > 0) {
3285                 int printed=0;
3286                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3287                                 bond->params.arp_interval);
3288
3289                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3290
3291                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3292                         if (!bond->params.arp_targets[i])
3293                                 continue;
3294                         if (printed)
3295                                 seq_printf(seq, ",");
3296                         target = ntohl(bond->params.arp_targets[i]);
3297                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3298                         printed = 1;
3299                 }
3300                 seq_printf(seq, "\n");
3301         }
3302
3303         if (bond->params.mode == BOND_MODE_8023AD) {
3304                 struct ad_info ad_info;
3305                 DECLARE_MAC_BUF(mac);
3306
3307                 seq_puts(seq, "\n802.3ad info\n");
3308                 seq_printf(seq, "LACP rate: %s\n",
3309                            (bond->params.lacp_fast) ? "fast" : "slow");
3310
3311                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3312                         seq_printf(seq, "bond %s has no active aggregator\n",
3313                                    bond->dev->name);
3314                 } else {
3315                         seq_printf(seq, "Active Aggregator Info:\n");
3316
3317                         seq_printf(seq, "\tAggregator ID: %d\n",
3318                                    ad_info.aggregator_id);
3319                         seq_printf(seq, "\tNumber of ports: %d\n",
3320                                    ad_info.ports);
3321                         seq_printf(seq, "\tActor Key: %d\n",
3322                                    ad_info.actor_key);
3323                         seq_printf(seq, "\tPartner Key: %d\n",
3324                                    ad_info.partner_key);
3325                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3326                                    print_mac(mac, ad_info.partner_system));
3327                 }
3328         }
3329 }
3330
3331 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3332 {
3333         struct bonding *bond = seq->private;
3334         DECLARE_MAC_BUF(mac);
3335
3336         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3337         seq_printf(seq, "MII Status: %s\n",
3338                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3339         seq_printf(seq, "Link Failure Count: %u\n",
3340                    slave->link_failure_count);
3341
3342         seq_printf(seq,
3343                    "Permanent HW addr: %s\n",
3344                    print_mac(mac, 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 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         int len = strlen(DRV_NAME);
3438
3439         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3440              bond_proc_dir = bond_proc_dir->next) {
3441                 if ((bond_proc_dir->namelen == len) &&
3442                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3443                         break;
3444                 }
3445         }
3446
3447         if (!bond_proc_dir) {
3448                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3449                 if (bond_proc_dir) {
3450                         bond_proc_dir->owner = THIS_MODULE;
3451                 } else {
3452                         printk(KERN_WARNING DRV_NAME
3453                                 ": Warning: cannot create /proc/net/%s\n",
3454                                 DRV_NAME);
3455                 }
3456         }
3457 }
3458
3459 /* Destroy the bonding directory under /proc/net, if empty.
3460  * Caller must hold rtnl_lock.
3461  */
3462 static void bond_destroy_proc_dir(void)
3463 {
3464         struct proc_dir_entry *de;
3465
3466         if (!bond_proc_dir) {
3467                 return;
3468         }
3469
3470         /* verify that the /proc dir is empty */
3471         for (de = bond_proc_dir->subdir; de; de = de->next) {
3472                 /* ignore . and .. */
3473                 if (*(de->name) != '.') {
3474                         break;
3475                 }
3476         }
3477
3478         if (de) {
3479                 if (bond_proc_dir->owner == THIS_MODULE) {
3480                         bond_proc_dir->owner = NULL;
3481                 }
3482         } else {
3483                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3484                 bond_proc_dir = NULL;
3485         }
3486 }
3487 #endif /* CONFIG_PROC_FS */
3488
3489 /*-------------------------- netdev event handling --------------------------*/
3490
3491 /*
3492  * Change device name
3493  */
3494 static int bond_event_changename(struct bonding *bond)
3495 {
3496 #ifdef CONFIG_PROC_FS
3497         bond_remove_proc_entry(bond);
3498         bond_create_proc_entry(bond);
3499 #endif
3500         down_write(&(bonding_rwsem));
3501         bond_destroy_sysfs_entry(bond);
3502         bond_create_sysfs_entry(bond);
3503         up_write(&(bonding_rwsem));
3504         return NOTIFY_DONE;
3505 }
3506
3507 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3508 {
3509         struct bonding *event_bond = bond_dev->priv;
3510
3511         switch (event) {
3512         case NETDEV_CHANGENAME:
3513                 return bond_event_changename(event_bond);
3514         case NETDEV_UNREGISTER:
3515                 bond_release_all(event_bond->dev);
3516                 break;
3517         default:
3518                 break;
3519         }
3520
3521         return NOTIFY_DONE;
3522 }
3523
3524 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3525 {
3526         struct net_device *bond_dev = slave_dev->master;
3527         struct bonding *bond = bond_dev->priv;
3528
3529         switch (event) {
3530         case NETDEV_UNREGISTER:
3531                 if (bond_dev) {
3532                         if (bond->setup_by_slave)
3533                                 bond_release_and_destroy(bond_dev, slave_dev);
3534                         else
3535                                 bond_release(bond_dev, slave_dev);
3536                 }
3537                 break;
3538         case NETDEV_CHANGE:
3539                 /*
3540                  * TODO: is this what we get if somebody
3541                  * sets up a hierarchical bond, then rmmod's
3542                  * one of the slave bonding devices?
3543                  */
3544                 break;
3545         case NETDEV_DOWN:
3546                 /*
3547                  * ... Or is it this?
3548                  */
3549                 break;
3550         case NETDEV_CHANGEMTU:
3551                 /*
3552                  * TODO: Should slaves be allowed to
3553                  * independently alter their MTU?  For
3554                  * an active-backup bond, slaves need
3555                  * not be the same type of device, so
3556                  * MTUs may vary.  For other modes,
3557                  * slaves arguably should have the
3558                  * same MTUs. To do this, we'd need to
3559                  * take over the slave's change_mtu
3560                  * function for the duration of their
3561                  * servitude.
3562                  */
3563                 break;
3564         case NETDEV_CHANGENAME:
3565                 /*
3566                  * TODO: handle changing the primary's name
3567                  */
3568                 break;
3569         case NETDEV_FEAT_CHANGE:
3570                 bond_compute_features(bond);
3571                 break;
3572         default:
3573                 break;
3574         }
3575
3576         return NOTIFY_DONE;
3577 }
3578
3579 /*
3580  * bond_netdev_event: handle netdev notifier chain events.
3581  *
3582  * This function receives events for the netdev chain.  The caller (an
3583  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3584  * locks for us to safely manipulate the slave devices (RTNL lock,
3585  * dev_probe_lock).
3586  */
3587 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3588 {
3589         struct net_device *event_dev = (struct net_device *)ptr;
3590
3591         if (dev_net(event_dev) != &init_net)
3592                 return NOTIFY_DONE;
3593
3594         dprintk("event_dev: %s, event: %lx\n",
3595                 (event_dev ? event_dev->name : "None"),
3596                 event);
3597
3598         if (!(event_dev->priv_flags & IFF_BONDING))
3599                 return NOTIFY_DONE;
3600
3601         if (event_dev->flags & IFF_MASTER) {
3602                 dprintk("IFF_MASTER\n");
3603                 return bond_master_netdev_event(event, event_dev);
3604         }
3605
3606         if (event_dev->flags & IFF_SLAVE) {
3607                 dprintk("IFF_SLAVE\n");
3608                 return bond_slave_netdev_event(event, event_dev);
3609         }
3610
3611         return NOTIFY_DONE;
3612 }
3613
3614 /*
3615  * bond_inetaddr_event: handle inetaddr notifier chain events.
3616  *
3617  * We keep track of device IPs primarily to use as source addresses in
3618  * ARP monitor probes (rather than spewing out broadcasts all the time).
3619  *
3620  * We track one IP for the main device (if it has one), plus one per VLAN.
3621  */
3622 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3623 {
3624         struct in_ifaddr *ifa = ptr;
3625         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3626         struct bonding *bond;
3627         struct vlan_entry *vlan;
3628
3629         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3630                 return NOTIFY_DONE;
3631
3632         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3633                 if (bond->dev == event_dev) {
3634                         switch (event) {
3635                         case NETDEV_UP:
3636                                 bond->master_ip = ifa->ifa_local;
3637                                 return NOTIFY_OK;
3638                         case NETDEV_DOWN:
3639                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3640                                 return NOTIFY_OK;
3641                         default:
3642                                 return NOTIFY_DONE;
3643                         }
3644                 }
3645
3646                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3647                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3648                         if (vlan_dev == event_dev) {
3649                                 switch (event) {
3650                                 case NETDEV_UP:
3651                                         vlan->vlan_ip = ifa->ifa_local;
3652                                         return NOTIFY_OK;
3653                                 case NETDEV_DOWN:
3654                                         vlan->vlan_ip =
3655                                                 bond_glean_dev_ip(vlan_dev);
3656                                         return NOTIFY_OK;
3657                                 default:
3658                                         return NOTIFY_DONE;
3659                                 }
3660                         }
3661                 }
3662         }
3663         return NOTIFY_DONE;
3664 }
3665
3666 static struct notifier_block bond_netdev_notifier = {
3667         .notifier_call = bond_netdev_event,
3668 };
3669
3670 static struct notifier_block bond_inetaddr_notifier = {
3671         .notifier_call = bond_inetaddr_event,
3672 };
3673
3674 /*-------------------------- Packet type handling ---------------------------*/
3675
3676 /* register to receive lacpdus on a bond */
3677 static void bond_register_lacpdu(struct bonding *bond)
3678 {
3679         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3680
3681         /* initialize packet type */
3682         pk_type->type = PKT_TYPE_LACPDU;
3683         pk_type->dev = bond->dev;
3684         pk_type->func = bond_3ad_lacpdu_recv;
3685
3686         dev_add_pack(pk_type);
3687 }
3688
3689 /* unregister to receive lacpdus on a bond */
3690 static void bond_unregister_lacpdu(struct bonding *bond)
3691 {
3692         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3693 }
3694
3695 void bond_register_arp(struct bonding *bond)
3696 {
3697         struct packet_type *pt = &bond->arp_mon_pt;
3698
3699         if (pt->type)
3700                 return;
3701
3702         pt->type = htons(ETH_P_ARP);
3703         pt->dev = bond->dev;
3704         pt->func = bond_arp_rcv;
3705         dev_add_pack(pt);
3706 }
3707
3708 void bond_unregister_arp(struct bonding *bond)
3709 {
3710         struct packet_type *pt = &bond->arp_mon_pt;
3711
3712         dev_remove_pack(pt);
3713         pt->type = 0;
3714 }
3715
3716 /*---------------------------- Hashing Policies -----------------------------*/
3717
3718 /*
3719  * Hash for the output device based upon layer 2 and layer 3 data. If
3720  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3721  */
3722 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3723                                      struct net_device *bond_dev, int count)
3724 {
3725         struct ethhdr *data = (struct ethhdr *)skb->data;
3726         struct iphdr *iph = ip_hdr(skb);
3727
3728         if (skb->protocol == htons(ETH_P_IP)) {
3729                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3730                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3731         }
3732
3733         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3734 }
3735
3736 /*
3737  * Hash for the output device based upon layer 3 and layer 4 data. If
3738  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3739  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3740  */
3741 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3742                                     struct net_device *bond_dev, int count)
3743 {
3744         struct ethhdr *data = (struct ethhdr *)skb->data;
3745         struct iphdr *iph = ip_hdr(skb);
3746         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3747         int layer4_xor = 0;
3748
3749         if (skb->protocol == htons(ETH_P_IP)) {
3750                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3751                     (iph->protocol == IPPROTO_TCP ||
3752                      iph->protocol == IPPROTO_UDP)) {
3753                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3754                 }
3755                 return (layer4_xor ^
3756                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3757
3758         }
3759
3760         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3761 }
3762
3763 /*
3764  * Hash for the output device based upon layer 2 data
3765  */
3766 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3767                                    struct net_device *bond_dev, int count)
3768 {
3769         struct ethhdr *data = (struct ethhdr *)skb->data;
3770
3771         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3772 }
3773
3774 /*-------------------------- Device entry points ----------------------------*/
3775
3776 static int bond_open(struct net_device *bond_dev)
3777 {
3778         struct bonding *bond = bond_dev->priv;
3779
3780         bond->kill_timers = 0;
3781
3782         if ((bond->params.mode == BOND_MODE_TLB) ||
3783             (bond->params.mode == BOND_MODE_ALB)) {
3784                 /* bond_alb_initialize must be called before the timer
3785                  * is started.
3786                  */
3787                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3788                         /* something went wrong - fail the open operation */
3789                         return -1;
3790                 }
3791
3792                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3793                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3794         }
3795
3796         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3797                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3798                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3799         }
3800
3801         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3802                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3803                         INIT_DELAYED_WORK(&bond->arp_work,
3804                                           bond_activebackup_arp_mon);
3805                 else
3806                         INIT_DELAYED_WORK(&bond->arp_work,
3807                                           bond_loadbalance_arp_mon);
3808
3809                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3810                 if (bond->params.arp_validate)
3811                         bond_register_arp(bond);
3812         }
3813
3814         if (bond->params.mode == BOND_MODE_8023AD) {
3815                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3816                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3817                 /* register to receive LACPDUs */
3818                 bond_register_lacpdu(bond);
3819         }
3820
3821         return 0;
3822 }
3823
3824 static int bond_close(struct net_device *bond_dev)
3825 {
3826         struct bonding *bond = bond_dev->priv;
3827
3828         if (bond->params.mode == BOND_MODE_8023AD) {
3829                 /* Unregister the receive of LACPDUs */
3830                 bond_unregister_lacpdu(bond);
3831         }
3832
3833         if (bond->params.arp_validate)
3834                 bond_unregister_arp(bond);
3835
3836         write_lock_bh(&bond->lock);
3837
3838         bond->send_grat_arp = 0;
3839
3840         /* signal timers not to re-arm */
3841         bond->kill_timers = 1;
3842
3843         write_unlock_bh(&bond->lock);
3844
3845         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3846                 cancel_delayed_work(&bond->mii_work);
3847         }
3848
3849         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3850                 cancel_delayed_work(&bond->arp_work);
3851         }
3852
3853         switch (bond->params.mode) {
3854         case BOND_MODE_8023AD:
3855                 cancel_delayed_work(&bond->ad_work);
3856                 break;
3857         case BOND_MODE_TLB:
3858         case BOND_MODE_ALB:
3859                 cancel_delayed_work(&bond->alb_work);
3860                 break;
3861         default:
3862                 break;
3863         }
3864
3865
3866         if ((bond->params.mode == BOND_MODE_TLB) ||
3867             (bond->params.mode == BOND_MODE_ALB)) {
3868                 /* Must be called only after all
3869                  * slaves have been released
3870                  */
3871                 bond_alb_deinitialize(bond);
3872         }
3873
3874         return 0;
3875 }
3876
3877 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3878 {
3879         struct bonding *bond = bond_dev->priv;
3880         struct net_device_stats *stats = &(bond->stats), *sstats;
3881         struct net_device_stats local_stats;
3882         struct slave *slave;
3883         int i;
3884
3885         memset(&local_stats, 0, sizeof(struct net_device_stats));
3886
3887         read_lock_bh(&bond->lock);
3888
3889         bond_for_each_slave(bond, slave, i) {
3890                 sstats = slave->dev->get_stats(slave->dev);
3891                 local_stats.rx_packets += sstats->rx_packets;
3892                 local_stats.rx_bytes += sstats->rx_bytes;
3893                 local_stats.rx_errors += sstats->rx_errors;
3894                 local_stats.rx_dropped += sstats->rx_dropped;
3895
3896                 local_stats.tx_packets += sstats->tx_packets;
3897                 local_stats.tx_bytes += sstats->tx_bytes;
3898                 local_stats.tx_errors += sstats->tx_errors;
3899                 local_stats.tx_dropped += sstats->tx_dropped;
3900
3901                 local_stats.multicast += sstats->multicast;
3902                 local_stats.collisions += sstats->collisions;
3903
3904                 local_stats.rx_length_errors += sstats->rx_length_errors;
3905                 local_stats.rx_over_errors += sstats->rx_over_errors;
3906                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3907                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3908                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3909                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3910
3911                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3912                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3913                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3914                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3915                 local_stats.tx_window_errors += sstats->tx_window_errors;
3916         }
3917
3918         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3919
3920         read_unlock_bh(&bond->lock);
3921
3922         return stats;
3923 }
3924
3925 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3926 {
3927         struct net_device *slave_dev = NULL;
3928         struct ifbond k_binfo;
3929         struct ifbond __user *u_binfo = NULL;
3930         struct ifslave k_sinfo;
3931         struct ifslave __user *u_sinfo = NULL;
3932         struct mii_ioctl_data *mii = NULL;
3933         int res = 0;
3934
3935         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3936                 bond_dev->name, cmd);
3937
3938         switch (cmd) {
3939         case SIOCGMIIPHY:
3940                 mii = if_mii(ifr);
3941                 if (!mii) {
3942                         return -EINVAL;
3943                 }
3944                 mii->phy_id = 0;
3945                 /* Fall Through */
3946         case SIOCGMIIREG:
3947                 /*
3948                  * We do this again just in case we were called by SIOCGMIIREG
3949                  * instead of SIOCGMIIPHY.
3950                  */
3951                 mii = if_mii(ifr);
3952                 if (!mii) {
3953                         return -EINVAL;
3954                 }
3955
3956                 if (mii->reg_num == 1) {
3957                         struct bonding *bond = bond_dev->priv;
3958                         mii->val_out = 0;
3959                         read_lock(&bond->lock);
3960                         read_lock(&bond->curr_slave_lock);
3961                         if (netif_carrier_ok(bond->dev)) {
3962                                 mii->val_out = BMSR_LSTATUS;
3963                         }
3964                         read_unlock(&bond->curr_slave_lock);
3965                         read_unlock(&bond->lock);
3966                 }
3967
3968                 return 0;
3969         case BOND_INFO_QUERY_OLD:
3970         case SIOCBONDINFOQUERY:
3971                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3972
3973                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3974                         return -EFAULT;
3975                 }
3976
3977                 res = bond_info_query(bond_dev, &k_binfo);
3978                 if (res == 0) {
3979                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3980                                 return -EFAULT;
3981                         }
3982                 }
3983
3984                 return res;
3985         case BOND_SLAVE_INFO_QUERY_OLD:
3986         case SIOCBONDSLAVEINFOQUERY:
3987                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3988
3989                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3990                         return -EFAULT;
3991                 }
3992
3993                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3994                 if (res == 0) {
3995                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3996                                 return -EFAULT;
3997                         }
3998                 }
3999
4000                 return res;
4001         default:
4002                 /* Go on */
4003                 break;
4004         }
4005
4006         if (!capable(CAP_NET_ADMIN)) {
4007                 return -EPERM;
4008         }
4009
4010         down_write(&(bonding_rwsem));
4011         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4012
4013         dprintk("slave_dev=%p: \n", slave_dev);
4014
4015         if (!slave_dev) {
4016                 res = -ENODEV;
4017         } else {
4018                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4019                 switch (cmd) {
4020                 case BOND_ENSLAVE_OLD:
4021                 case SIOCBONDENSLAVE:
4022                         res = bond_enslave(bond_dev, slave_dev);
4023                         break;
4024                 case BOND_RELEASE_OLD:
4025                 case SIOCBONDRELEASE:
4026                         res = bond_release(bond_dev, slave_dev);
4027                         break;
4028                 case BOND_SETHWADDR_OLD:
4029                 case SIOCBONDSETHWADDR:
4030                         res = bond_sethwaddr(bond_dev, slave_dev);
4031                         break;
4032                 case BOND_CHANGE_ACTIVE_OLD:
4033                 case SIOCBONDCHANGEACTIVE:
4034                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4035                         break;
4036                 default:
4037                         res = -EOPNOTSUPP;
4038                 }
4039
4040                 dev_put(slave_dev);
4041         }
4042
4043         up_write(&(bonding_rwsem));
4044         return res;
4045 }
4046
4047 static void bond_set_multicast_list(struct net_device *bond_dev)
4048 {
4049         struct bonding *bond = bond_dev->priv;
4050         struct dev_mc_list *dmi;
4051
4052         /*
4053          * Do promisc before checking multicast_mode
4054          */
4055         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4056                 /*
4057                  * FIXME: Need to handle the error when one of the multi-slaves
4058                  * encounters error.
4059                  */
4060                 bond_set_promiscuity(bond, 1);
4061         }
4062
4063         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4064                 bond_set_promiscuity(bond, -1);
4065         }
4066
4067         /* set allmulti flag to slaves */
4068         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4069                 /*
4070                  * FIXME: Need to handle the error when one of the multi-slaves
4071                  * encounters error.
4072                  */
4073                 bond_set_allmulti(bond, 1);
4074         }
4075
4076         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4077                 bond_set_allmulti(bond, -1);
4078         }
4079
4080         read_lock(&bond->lock);
4081
4082         bond->flags = bond_dev->flags;
4083
4084         /* looking for addresses to add to slaves' mc list */
4085         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4086                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4087                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4088                 }
4089         }
4090
4091         /* looking for addresses to delete from slaves' list */
4092         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4093                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4094                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4095                 }
4096         }
4097
4098         /* save master's multicast list */
4099         bond_mc_list_destroy(bond);
4100         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4101
4102         read_unlock(&bond->lock);
4103 }
4104
4105 /*
4106  * Change the MTU of all of a master's slaves to match the master
4107  */
4108 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4109 {
4110         struct bonding *bond = bond_dev->priv;
4111         struct slave *slave, *stop_at;
4112         int res = 0;
4113         int i;
4114
4115         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4116                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4117
4118         /* Can't hold bond->lock with bh disabled here since
4119          * some base drivers panic. On the other hand we can't
4120          * hold bond->lock without bh disabled because we'll
4121          * deadlock. The only solution is to rely on the fact
4122          * that we're under rtnl_lock here, and the slaves
4123          * list won't change. This doesn't solve the problem
4124          * of setting the slave's MTU while it is
4125          * transmitting, but the assumption is that the base
4126          * driver can handle that.
4127          *
4128          * TODO: figure out a way to safely iterate the slaves
4129          * list, but without holding a lock around the actual
4130          * call to the base driver.
4131          */
4132
4133         bond_for_each_slave(bond, slave, i) {
4134                 dprintk("s %p s->p %p c_m %p\n", slave,
4135                         slave->prev, slave->dev->change_mtu);
4136
4137                 res = dev_set_mtu(slave->dev, new_mtu);
4138
4139                 if (res) {
4140                         /* If we failed to set the slave's mtu to the new value
4141                          * we must abort the operation even in ACTIVE_BACKUP
4142                          * mode, because if we allow the backup slaves to have
4143                          * different mtu values than the active slave we'll
4144                          * need to change their mtu when doing a failover. That
4145                          * means changing their mtu from timer context, which
4146                          * is probably not a good idea.
4147                          */
4148                         dprintk("err %d %s\n", res, slave->dev->name);
4149                         goto unwind;
4150                 }
4151         }
4152
4153         bond_dev->mtu = new_mtu;
4154
4155         return 0;
4156
4157 unwind:
4158         /* unwind from head to the slave that failed */
4159         stop_at = slave;
4160         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4161                 int tmp_res;
4162
4163                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4164                 if (tmp_res) {
4165                         dprintk("unwind err %d dev %s\n", tmp_res,
4166                                 slave->dev->name);
4167                 }
4168         }
4169
4170         return res;
4171 }
4172
4173 /*
4174  * Change HW address
4175  *
4176  * Note that many devices must be down to change the HW address, and
4177  * downing the master releases all slaves.  We can make bonds full of
4178  * bonding devices to test this, however.
4179  */
4180 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4181 {
4182         struct bonding *bond = bond_dev->priv;
4183         struct sockaddr *sa = addr, tmp_sa;
4184         struct slave *slave, *stop_at;
4185         int res = 0;
4186         int i;
4187
4188         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4189
4190         /*
4191          * If fail_over_mac is set to active, do nothing and return
4192          * success.  Returning an error causes ifenslave to fail.
4193          */
4194         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4195                 return 0;
4196
4197         if (!is_valid_ether_addr(sa->sa_data)) {
4198                 return -EADDRNOTAVAIL;
4199         }
4200
4201         /* Can't hold bond->lock with bh disabled here since
4202          * some base drivers panic. On the other hand we can't
4203          * hold bond->lock without bh disabled because we'll
4204          * deadlock. The only solution is to rely on the fact
4205          * that we're under rtnl_lock here, and the slaves
4206          * list won't change. This doesn't solve the problem
4207          * of setting the slave's hw address while it is
4208          * transmitting, but the assumption is that the base
4209          * driver can handle that.
4210          *
4211          * TODO: figure out a way to safely iterate the slaves
4212          * list, but without holding a lock around the actual
4213          * call to the base driver.
4214          */
4215
4216         bond_for_each_slave(bond, slave, i) {
4217                 dprintk("slave %p %s\n", slave, slave->dev->name);
4218
4219                 if (slave->dev->set_mac_address == NULL) {
4220                         res = -EOPNOTSUPP;
4221                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4222                         goto unwind;
4223                 }
4224
4225                 res = dev_set_mac_address(slave->dev, addr);
4226                 if (res) {
4227                         /* TODO: consider downing the slave
4228                          * and retry ?
4229                          * User should expect communications
4230                          * breakage anyway until ARP finish
4231                          * updating, so...
4232                          */
4233                         dprintk("err %d %s\n", res, slave->dev->name);
4234                         goto unwind;
4235                 }
4236         }
4237
4238         /* success */
4239         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4240         return 0;
4241
4242 unwind:
4243         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4244         tmp_sa.sa_family = bond_dev->type;
4245
4246         /* unwind from head to the slave that failed */
4247         stop_at = slave;
4248         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4249                 int tmp_res;
4250
4251                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4252                 if (tmp_res) {
4253                         dprintk("unwind err %d dev %s\n", tmp_res,
4254                                 slave->dev->name);
4255                 }
4256         }
4257
4258         return res;
4259 }
4260
4261 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4262 {
4263         struct bonding *bond = bond_dev->priv;
4264         struct slave *slave, *start_at;
4265         int i, slave_no, res = 1;
4266
4267         read_lock(&bond->lock);
4268
4269         if (!BOND_IS_OK(bond)) {
4270                 goto out;
4271         }
4272
4273         /*
4274          * Concurrent TX may collide on rr_tx_counter; we accept that
4275          * as being rare enough not to justify using an atomic op here
4276          */
4277         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4278
4279         bond_for_each_slave(bond, slave, i) {
4280                 slave_no--;
4281                 if (slave_no < 0) {
4282                         break;
4283                 }
4284         }
4285
4286         start_at = slave;
4287         bond_for_each_slave_from(bond, slave, i, start_at) {
4288                 if (IS_UP(slave->dev) &&
4289                     (slave->link == BOND_LINK_UP) &&
4290                     (slave->state == BOND_STATE_ACTIVE)) {
4291                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4292                         break;
4293                 }
4294         }
4295
4296 out:
4297         if (res) {
4298                 /* no suitable interface, frame not sent */
4299                 dev_kfree_skb(skb);
4300         }
4301         read_unlock(&bond->lock);
4302         return 0;
4303 }
4304
4305
4306 /*
4307  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4308  * the bond has a usable interface.
4309  */
4310 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4311 {
4312         struct bonding *bond = bond_dev->priv;
4313         int res = 1;
4314
4315         read_lock(&bond->lock);
4316         read_lock(&bond->curr_slave_lock);
4317
4318         if (!BOND_IS_OK(bond)) {
4319                 goto out;
4320         }
4321
4322         if (!bond->curr_active_slave)
4323                 goto out;
4324
4325         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4326
4327 out:
4328         if (res) {
4329                 /* no suitable interface, frame not sent */
4330                 dev_kfree_skb(skb);
4331         }
4332         read_unlock(&bond->curr_slave_lock);
4333         read_unlock(&bond->lock);
4334         return 0;
4335 }
4336
4337 /*
4338  * In bond_xmit_xor() , we determine the output device by using a pre-
4339  * determined xmit_hash_policy(), If the selected device is not enabled,
4340  * find the next active slave.
4341  */
4342 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4343 {
4344         struct bonding *bond = bond_dev->priv;
4345         struct slave *slave, *start_at;
4346         int slave_no;
4347         int i;
4348         int res = 1;
4349
4350         read_lock(&bond->lock);
4351
4352         if (!BOND_IS_OK(bond)) {
4353                 goto out;
4354         }
4355
4356         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4357
4358         bond_for_each_slave(bond, slave, i) {
4359                 slave_no--;
4360                 if (slave_no < 0) {
4361                         break;
4362                 }
4363         }
4364
4365         start_at = slave;
4366
4367         bond_for_each_slave_from(bond, slave, i, start_at) {
4368                 if (IS_UP(slave->dev) &&
4369                     (slave->link == BOND_LINK_UP) &&
4370                     (slave->state == BOND_STATE_ACTIVE)) {
4371                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4372                         break;
4373                 }
4374         }
4375
4376 out:
4377         if (res) {
4378                 /* no suitable interface, frame not sent */
4379                 dev_kfree_skb(skb);
4380         }
4381         read_unlock(&bond->lock);
4382         return 0;
4383 }
4384
4385 /*
4386  * in broadcast mode, we send everything to all usable interfaces.
4387  */
4388 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4389 {
4390         struct bonding *bond = bond_dev->priv;
4391         struct slave *slave, *start_at;
4392         struct net_device *tx_dev = NULL;
4393         int i;
4394         int res = 1;
4395
4396         read_lock(&bond->lock);
4397
4398         if (!BOND_IS_OK(bond)) {
4399                 goto out;
4400         }
4401
4402         read_lock(&bond->curr_slave_lock);
4403         start_at = bond->curr_active_slave;
4404         read_unlock(&bond->curr_slave_lock);
4405
4406         if (!start_at) {
4407                 goto out;
4408         }
4409
4410         bond_for_each_slave_from(bond, slave, i, start_at) {
4411                 if (IS_UP(slave->dev) &&
4412                     (slave->link == BOND_LINK_UP) &&
4413                     (slave->state == BOND_STATE_ACTIVE)) {
4414                         if (tx_dev) {
4415                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4416                                 if (!skb2) {
4417                                         printk(KERN_ERR DRV_NAME
4418                                                ": %s: Error: bond_xmit_broadcast(): "
4419                                                "skb_clone() failed\n",
4420                                                bond_dev->name);
4421                                         continue;
4422                                 }
4423
4424                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4425                                 if (res) {
4426                                         dev_kfree_skb(skb2);
4427                                         continue;
4428                                 }
4429                         }
4430                         tx_dev = slave->dev;
4431                 }
4432         }
4433
4434         if (tx_dev) {
4435                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4436         }
4437
4438 out:
4439         if (res) {
4440                 /* no suitable interface, frame not sent */
4441                 dev_kfree_skb(skb);
4442         }
4443         /* frame sent to all suitable interfaces */
4444         read_unlock(&bond->lock);
4445         return 0;
4446 }
4447
4448 /*------------------------- Device initialization ---------------------------*/
4449
4450 static void bond_set_xmit_hash_policy(struct bonding *bond)
4451 {
4452         switch (bond->params.xmit_policy) {
4453         case BOND_XMIT_POLICY_LAYER23:
4454                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4455                 break;
4456         case BOND_XMIT_POLICY_LAYER34:
4457                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4458                 break;
4459         case BOND_XMIT_POLICY_LAYER2:
4460         default:
4461                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4462                 break;
4463         }
4464 }
4465
4466 /*
4467  * set bond mode specific net device operations
4468  */
4469 void bond_set_mode_ops(struct bonding *bond, int mode)
4470 {
4471         struct net_device *bond_dev = bond->dev;
4472
4473         switch (mode) {
4474         case BOND_MODE_ROUNDROBIN:
4475                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4476                 break;
4477         case BOND_MODE_ACTIVEBACKUP:
4478                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4479                 break;
4480         case BOND_MODE_XOR:
4481                 bond_dev->hard_start_xmit = bond_xmit_xor;
4482                 bond_set_xmit_hash_policy(bond);
4483                 break;
4484         case BOND_MODE_BROADCAST:
4485                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4486                 break;
4487         case BOND_MODE_8023AD:
4488                 bond_set_master_3ad_flags(bond);
4489                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4490                 bond_set_xmit_hash_policy(bond);
4491                 break;
4492         case BOND_MODE_ALB:
4493                 bond_set_master_alb_flags(bond);
4494                 /* FALLTHRU */
4495         case BOND_MODE_TLB:
4496                 bond_dev->hard_start_xmit = bond_alb_xmit;
4497                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4498                 break;
4499         default:
4500                 /* Should never happen, mode already checked */
4501                 printk(KERN_ERR DRV_NAME
4502                        ": %s: Error: Unknown bonding mode %d\n",
4503                        bond_dev->name,
4504                        mode);
4505                 break;
4506         }
4507 }
4508
4509 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4510                                     struct ethtool_drvinfo *drvinfo)
4511 {
4512         strncpy(drvinfo->driver, DRV_NAME, 32);
4513         strncpy(drvinfo->version, DRV_VERSION, 32);
4514         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4515 }
4516
4517 static const struct ethtool_ops bond_ethtool_ops = {
4518         .get_drvinfo            = bond_ethtool_get_drvinfo,
4519         .get_link               = ethtool_op_get_link,
4520         .get_tx_csum            = ethtool_op_get_tx_csum,
4521         .get_sg                 = ethtool_op_get_sg,
4522         .get_tso                = ethtool_op_get_tso,
4523         .get_ufo                = ethtool_op_get_ufo,
4524         .get_flags              = ethtool_op_get_flags,
4525 };
4526
4527 /*
4528  * Does not allocate but creates a /proc entry.
4529  * Allowed to fail.
4530  */
4531 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4532 {
4533         struct bonding *bond = bond_dev->priv;
4534
4535         dprintk("Begin bond_init for %s\n", bond_dev->name);
4536
4537         /* initialize rwlocks */
4538         rwlock_init(&bond->lock);
4539         rwlock_init(&bond->curr_slave_lock);
4540
4541         bond->params = *params; /* copy params struct */
4542
4543         bond->wq = create_singlethread_workqueue(bond_dev->name);
4544         if (!bond->wq)
4545                 return -ENOMEM;
4546
4547         /* Initialize pointers */
4548         bond->first_slave = NULL;
4549         bond->curr_active_slave = NULL;
4550         bond->current_arp_slave = NULL;
4551         bond->primary_slave = NULL;
4552         bond->dev = bond_dev;
4553         bond->send_grat_arp = 0;
4554         bond->setup_by_slave = 0;
4555         INIT_LIST_HEAD(&bond->vlan_list);
4556
4557         /* Initialize the device entry points */
4558         bond_dev->open = bond_open;
4559         bond_dev->stop = bond_close;
4560         bond_dev->get_stats = bond_get_stats;
4561         bond_dev->do_ioctl = bond_do_ioctl;
4562         bond_dev->ethtool_ops = &bond_ethtool_ops;
4563         bond_dev->set_multicast_list = bond_set_multicast_list;
4564         bond_dev->change_mtu = bond_change_mtu;
4565         bond_dev->set_mac_address = bond_set_mac_address;
4566         bond_dev->validate_addr = NULL;
4567
4568         bond_set_mode_ops(bond, bond->params.mode);
4569
4570         bond_dev->destructor = bond_destructor;
4571
4572         /* Initialize the device options */
4573         bond_dev->tx_queue_len = 0;
4574         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4575         bond_dev->priv_flags |= IFF_BONDING;
4576
4577         /* At first, we block adding VLANs. That's the only way to
4578          * prevent problems that occur when adding VLANs over an
4579          * empty bond. The block will be removed once non-challenged
4580          * slaves are enslaved.
4581          */
4582         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4583
4584         /* don't acquire bond device's netif_tx_lock when
4585          * transmitting */
4586         bond_dev->features |= NETIF_F_LLTX;
4587
4588         /* By default, we declare the bond to be fully
4589          * VLAN hardware accelerated capable. Special
4590          * care is taken in the various xmit functions
4591          * when there are slaves that are not hw accel
4592          * capable
4593          */
4594         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4595         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4596         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4597         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4598                                NETIF_F_HW_VLAN_RX |
4599                                NETIF_F_HW_VLAN_FILTER);
4600
4601 #ifdef CONFIG_PROC_FS
4602         bond_create_proc_entry(bond);
4603 #endif
4604         list_add_tail(&bond->bond_list, &bond_dev_list);
4605
4606         return 0;
4607 }
4608
4609 static void bond_work_cancel_all(struct bonding *bond)
4610 {
4611         write_lock_bh(&bond->lock);
4612         bond->kill_timers = 1;
4613         write_unlock_bh(&bond->lock);
4614
4615         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4616                 cancel_delayed_work(&bond->mii_work);
4617
4618         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4619                 cancel_delayed_work(&bond->arp_work);
4620
4621         if (bond->params.mode == BOND_MODE_ALB &&
4622             delayed_work_pending(&bond->alb_work))
4623                 cancel_delayed_work(&bond->alb_work);
4624
4625         if (bond->params.mode == BOND_MODE_8023AD &&
4626             delayed_work_pending(&bond->ad_work))
4627                 cancel_delayed_work(&bond->ad_work);
4628 }
4629
4630 /* De-initialize device specific data.
4631  * Caller must hold rtnl_lock.
4632  */
4633 static void bond_deinit(struct net_device *bond_dev)
4634 {
4635         struct bonding *bond = bond_dev->priv;
4636
4637         list_del(&bond->bond_list);
4638
4639         bond_work_cancel_all(bond);
4640
4641 #ifdef CONFIG_PROC_FS
4642         bond_remove_proc_entry(bond);
4643 #endif
4644 }
4645
4646 /* Unregister and free all bond devices.
4647  * Caller must hold rtnl_lock.
4648  */
4649 static void bond_free_all(void)
4650 {
4651         struct bonding *bond, *nxt;
4652
4653         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4654                 struct net_device *bond_dev = bond->dev;
4655
4656                 bond_work_cancel_all(bond);
4657                 /* Release the bonded slaves */
4658                 bond_release_all(bond_dev);
4659                 bond_destroy(bond);
4660         }
4661
4662 #ifdef CONFIG_PROC_FS
4663         bond_destroy_proc_dir();
4664 #endif
4665 }
4666
4667 /*------------------------- Module initialization ---------------------------*/
4668
4669 /*
4670  * Convert string input module parms.  Accept either the
4671  * number of the mode or its string name.  A bit complicated because
4672  * some mode names are substrings of other names, and calls from sysfs
4673  * may have whitespace in the name (trailing newlines, for example).
4674  */
4675 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4676 {
4677         int mode = -1, i, rv;
4678         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4679
4680         for (p = (char *)buf; *p; p++)
4681                 if (!(isdigit(*p) || isspace(*p)))
4682                         break;
4683
4684         if (*p)
4685                 rv = sscanf(buf, "%20s", modestr);
4686         else
4687                 rv = sscanf(buf, "%d", &mode);
4688
4689         if (!rv)
4690                 return -1;
4691
4692         for (i = 0; tbl[i].modename; i++) {
4693                 if (mode == tbl[i].mode)
4694                         return tbl[i].mode;
4695                 if (strcmp(modestr, tbl[i].modename) == 0)
4696                         return tbl[i].mode;
4697         }
4698
4699         return -1;
4700 }
4701
4702 static int bond_check_params(struct bond_params *params)
4703 {
4704         int arp_validate_value, fail_over_mac_value;
4705
4706         /*
4707          * Convert string parameters.
4708          */
4709         if (mode) {
4710                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4711                 if (bond_mode == -1) {
4712                         printk(KERN_ERR DRV_NAME
4713                                ": Error: Invalid bonding mode \"%s\"\n",
4714                                mode == NULL ? "NULL" : mode);
4715                         return -EINVAL;
4716                 }
4717         }
4718
4719         if (xmit_hash_policy) {
4720                 if ((bond_mode != BOND_MODE_XOR) &&
4721                     (bond_mode != BOND_MODE_8023AD)) {
4722                         printk(KERN_INFO DRV_NAME
4723                                ": xor_mode param is irrelevant in mode %s\n",
4724                                bond_mode_name(bond_mode));
4725                 } else {
4726                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4727                                                         xmit_hashtype_tbl);
4728                         if (xmit_hashtype == -1) {
4729                                 printk(KERN_ERR DRV_NAME
4730                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4731                                 xmit_hash_policy == NULL ? "NULL" :
4732                                        xmit_hash_policy);
4733                                 return -EINVAL;
4734                         }
4735                 }
4736         }
4737
4738         if (lacp_rate) {
4739                 if (bond_mode != BOND_MODE_8023AD) {
4740                         printk(KERN_INFO DRV_NAME
4741                                ": lacp_rate param is irrelevant in mode %s\n",
4742                                bond_mode_name(bond_mode));
4743                 } else {
4744                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4745                         if (lacp_fast == -1) {
4746                                 printk(KERN_ERR DRV_NAME
4747                                        ": Error: Invalid lacp rate \"%s\"\n",
4748                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4749                                 return -EINVAL;
4750                         }
4751                 }
4752         }
4753
4754         if (max_bonds < 0 || max_bonds > INT_MAX) {
4755                 printk(KERN_WARNING DRV_NAME
4756                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4757                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4758                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4759                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4760         }
4761
4762         if (miimon < 0) {
4763                 printk(KERN_WARNING DRV_NAME
4764                        ": Warning: miimon module parameter (%d), "
4765                        "not in range 0-%d, so it was reset to %d\n",
4766                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4767                 miimon = BOND_LINK_MON_INTERV;
4768         }
4769
4770         if (updelay < 0) {
4771                 printk(KERN_WARNING DRV_NAME
4772                        ": Warning: updelay module parameter (%d), "
4773                        "not in range 0-%d, so it was reset to 0\n",
4774                        updelay, INT_MAX);
4775                 updelay = 0;
4776         }
4777
4778         if (downdelay < 0) {
4779                 printk(KERN_WARNING DRV_NAME
4780                        ": Warning: downdelay module parameter (%d), "
4781                        "not in range 0-%d, so it was reset to 0\n",
4782                        downdelay, INT_MAX);
4783                 downdelay = 0;
4784         }
4785
4786         if ((use_carrier != 0) && (use_carrier != 1)) {
4787                 printk(KERN_WARNING DRV_NAME
4788                        ": Warning: use_carrier module parameter (%d), "
4789                        "not of valid value (0/1), so it was set to 1\n",
4790                        use_carrier);
4791                 use_carrier = 1;
4792         }
4793
4794         if (num_grat_arp < 0 || num_grat_arp > 255) {
4795                 printk(KERN_WARNING DRV_NAME
4796                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4797                        "was reset to 1 \n", num_grat_arp);
4798                 num_grat_arp = 1;
4799         }
4800
4801         /* reset values for 802.3ad */
4802         if (bond_mode == BOND_MODE_8023AD) {
4803                 if (!miimon) {
4804                         printk(KERN_WARNING DRV_NAME
4805                                ": Warning: miimon must be specified, "
4806                                "otherwise bonding will not detect link "
4807                                "failure, speed and duplex which are "
4808                                "essential for 802.3ad operation\n");
4809                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4810                         miimon = 100;
4811                 }
4812         }
4813
4814         /* reset values for TLB/ALB */
4815         if ((bond_mode == BOND_MODE_TLB) ||
4816             (bond_mode == BOND_MODE_ALB)) {
4817                 if (!miimon) {
4818                         printk(KERN_WARNING DRV_NAME
4819                                ": Warning: miimon must be specified, "
4820                                "otherwise bonding will not detect link "
4821                                "failure and link speed which are essential "
4822                                "for TLB/ALB load balancing\n");
4823                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4824                         miimon = 100;
4825                 }
4826         }
4827
4828         if (bond_mode == BOND_MODE_ALB) {
4829                 printk(KERN_NOTICE DRV_NAME
4830                        ": In ALB mode you might experience client "
4831                        "disconnections upon reconnection of a link if the "
4832                        "bonding module updelay parameter (%d msec) is "
4833                        "incompatible with the forwarding delay time of the "
4834                        "switch\n",
4835                        updelay);
4836         }
4837
4838         if (!miimon) {
4839                 if (updelay || downdelay) {
4840                         /* just warn the user the up/down delay will have
4841                          * no effect since miimon is zero...
4842                          */
4843                         printk(KERN_WARNING DRV_NAME
4844                                ": Warning: miimon module parameter not set "
4845                                "and updelay (%d) or downdelay (%d) module "
4846                                "parameter is set; updelay and downdelay have "
4847                                "no effect unless miimon is set\n",
4848                                updelay, downdelay);
4849                 }
4850         } else {
4851                 /* don't allow arp monitoring */
4852                 if (arp_interval) {
4853                         printk(KERN_WARNING DRV_NAME
4854                                ": Warning: miimon (%d) and arp_interval (%d) "
4855                                "can't be used simultaneously, disabling ARP "
4856                                "monitoring\n",
4857                                miimon, arp_interval);
4858                         arp_interval = 0;
4859                 }
4860
4861                 if ((updelay % miimon) != 0) {
4862                         printk(KERN_WARNING DRV_NAME
4863                                ": Warning: updelay (%d) is not a multiple "
4864                                "of miimon (%d), updelay rounded to %d ms\n",
4865                                updelay, miimon, (updelay / miimon) * miimon);
4866                 }
4867
4868                 updelay /= miimon;
4869
4870                 if ((downdelay % miimon) != 0) {
4871                         printk(KERN_WARNING DRV_NAME
4872                                ": Warning: downdelay (%d) is not a multiple "
4873                                "of miimon (%d), downdelay rounded to %d ms\n",
4874                                downdelay, miimon,
4875                                (downdelay / miimon) * miimon);
4876                 }
4877
4878                 downdelay /= miimon;
4879         }
4880
4881         if (arp_interval < 0) {
4882                 printk(KERN_WARNING DRV_NAME
4883                        ": Warning: arp_interval module parameter (%d) "
4884                        ", not in range 0-%d, so it was reset to %d\n",
4885                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4886                 arp_interval = BOND_LINK_ARP_INTERV;
4887         }
4888
4889         for (arp_ip_count = 0;
4890              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4891              arp_ip_count++) {
4892                 /* not complete check, but should be good enough to
4893                    catch mistakes */
4894                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4895                         printk(KERN_WARNING DRV_NAME
4896                                ": Warning: bad arp_ip_target module parameter "
4897                                "(%s), ARP monitoring will not be performed\n",
4898                                arp_ip_target[arp_ip_count]);
4899                         arp_interval = 0;
4900                 } else {
4901                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4902                         arp_target[arp_ip_count] = ip;
4903                 }
4904         }
4905
4906         if (arp_interval && !arp_ip_count) {
4907                 /* don't allow arping if no arp_ip_target given... */
4908                 printk(KERN_WARNING DRV_NAME
4909                        ": Warning: arp_interval module parameter (%d) "
4910                        "specified without providing an arp_ip_target "
4911                        "parameter, arp_interval was reset to 0\n",
4912                        arp_interval);
4913                 arp_interval = 0;
4914         }
4915
4916         if (arp_validate) {
4917                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4918                         printk(KERN_ERR DRV_NAME
4919                ": arp_validate only supported in active-backup mode\n");
4920                         return -EINVAL;
4921                 }
4922                 if (!arp_interval) {
4923                         printk(KERN_ERR DRV_NAME
4924                                ": arp_validate requires arp_interval\n");
4925                         return -EINVAL;
4926                 }
4927
4928                 arp_validate_value = bond_parse_parm(arp_validate,
4929                                                      arp_validate_tbl);
4930                 if (arp_validate_value == -1) {
4931                         printk(KERN_ERR DRV_NAME
4932                                ": Error: invalid arp_validate \"%s\"\n",
4933                                arp_validate == NULL ? "NULL" : arp_validate);
4934                         return -EINVAL;
4935                 }
4936         } else
4937                 arp_validate_value = 0;
4938
4939         if (miimon) {
4940                 printk(KERN_INFO DRV_NAME
4941                        ": MII link monitoring set to %d ms\n",
4942                        miimon);
4943         } else if (arp_interval) {
4944                 int i;
4945
4946                 printk(KERN_INFO DRV_NAME
4947                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4948                        arp_interval,
4949                        arp_validate_tbl[arp_validate_value].modename,
4950                        arp_ip_count);
4951
4952                 for (i = 0; i < arp_ip_count; i++)
4953                         printk (" %s", arp_ip_target[i]);
4954
4955                 printk("\n");
4956
4957         } else if (max_bonds) {
4958                 /* miimon and arp_interval not set, we need one so things
4959                  * work as expected, see bonding.txt for details
4960                  */
4961                 printk(KERN_WARNING DRV_NAME
4962                        ": Warning: either miimon or arp_interval and "
4963                        "arp_ip_target module parameters must be specified, "
4964                        "otherwise bonding will not detect link failures! see "
4965                        "bonding.txt for details.\n");
4966         }
4967
4968         if (primary && !USES_PRIMARY(bond_mode)) {
4969                 /* currently, using a primary only makes sense
4970                  * in active backup, TLB or ALB modes
4971                  */
4972                 printk(KERN_WARNING DRV_NAME
4973                        ": Warning: %s primary device specified but has no "
4974                        "effect in %s mode\n",
4975                        primary, bond_mode_name(bond_mode));
4976                 primary = NULL;
4977         }
4978
4979         if (fail_over_mac) {
4980                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4981                                                       fail_over_mac_tbl);
4982                 if (fail_over_mac_value == -1) {
4983                         printk(KERN_ERR DRV_NAME
4984                                ": Error: invalid fail_over_mac \"%s\"\n",
4985                                arp_validate == NULL ? "NULL" : arp_validate);
4986                         return -EINVAL;
4987                 }
4988
4989                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4990                         printk(KERN_WARNING DRV_NAME
4991                                ": Warning: fail_over_mac only affects "
4992                                "active-backup mode.\n");
4993         } else {
4994                 fail_over_mac_value = BOND_FOM_NONE;
4995         }
4996
4997         /* fill params struct with the proper values */
4998         params->mode = bond_mode;
4999         params->xmit_policy = xmit_hashtype;
5000         params->miimon = miimon;
5001         params->num_grat_arp = num_grat_arp;
5002         params->arp_interval = arp_interval;
5003         params->arp_validate = arp_validate_value;
5004         params->updelay = updelay;
5005         params->downdelay = downdelay;
5006         params->use_carrier = use_carrier;
5007         params->lacp_fast = lacp_fast;
5008         params->primary[0] = 0;
5009         params->fail_over_mac = fail_over_mac_value;
5010
5011         if (primary) {
5012                 strncpy(params->primary, primary, IFNAMSIZ);
5013                 params->primary[IFNAMSIZ - 1] = 0;
5014         }
5015
5016         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5017
5018         return 0;
5019 }
5020
5021 static struct lock_class_key bonding_netdev_xmit_lock_key;
5022 static struct lock_class_key bonding_netdev_addr_lock_key;
5023
5024 static void bond_set_lockdep_class_one(struct net_device *dev,
5025                                        struct netdev_queue *txq,
5026                                        void *_unused)
5027 {
5028         lockdep_set_class(&txq->_xmit_lock,
5029                           &bonding_netdev_xmit_lock_key);
5030 }
5031
5032 static void bond_set_lockdep_class(struct net_device *dev)
5033 {
5034         lockdep_set_class(&dev->addr_list_lock,
5035                           &bonding_netdev_addr_lock_key);
5036         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5037 }
5038
5039 /* Create a new bond based on the specified name and bonding parameters.
5040  * If name is NULL, obtain a suitable "bond%d" name for us.
5041  * Caller must NOT hold rtnl_lock; we need to release it here before we
5042  * set up our sysfs entries.
5043  */
5044 int bond_create(char *name, struct bond_params *params)
5045 {
5046         struct net_device *bond_dev;
5047         struct bonding *bond;
5048         int res;
5049
5050         rtnl_lock();
5051         down_write(&bonding_rwsem);
5052
5053         /* Check to see if the bond already exists. */
5054         if (name) {
5055                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5056                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5057                                 printk(KERN_ERR DRV_NAME
5058                                ": cannot add bond %s; it already exists\n",
5059                                        name);
5060                                 res = -EPERM;
5061                                 goto out_rtnl;
5062                         }
5063         }
5064
5065         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5066                                 ether_setup);
5067         if (!bond_dev) {
5068                 printk(KERN_ERR DRV_NAME
5069                        ": %s: eek! can't alloc netdev!\n",
5070                        name);
5071                 res = -ENOMEM;
5072                 goto out_rtnl;
5073         }
5074
5075         if (!name) {
5076                 res = dev_alloc_name(bond_dev, "bond%d");
5077                 if (res < 0)
5078                         goto out_netdev;
5079         }
5080
5081         /* bond_init() must be called after dev_alloc_name() (for the
5082          * /proc files), but before register_netdevice(), because we
5083          * need to set function pointers.
5084          */
5085
5086         res = bond_init(bond_dev, params);
5087         if (res < 0) {
5088                 goto out_netdev;
5089         }
5090
5091         res = register_netdevice(bond_dev);
5092         if (res < 0) {
5093                 goto out_bond;
5094         }
5095
5096         bond_set_lockdep_class(bond_dev);
5097
5098         netif_carrier_off(bond_dev);
5099
5100         up_write(&bonding_rwsem);
5101         rtnl_unlock(); /* allows sysfs registration of net device */
5102         res = bond_create_sysfs_entry(bond_dev->priv);
5103         if (res < 0) {
5104                 rtnl_lock();
5105                 down_write(&bonding_rwsem);
5106                 bond_deinit(bond_dev);
5107                 unregister_netdevice(bond_dev);
5108                 goto out_rtnl;
5109         }
5110
5111         return 0;
5112
5113 out_bond:
5114         bond_deinit(bond_dev);
5115 out_netdev:
5116         free_netdev(bond_dev);
5117 out_rtnl:
5118         up_write(&bonding_rwsem);
5119         rtnl_unlock();
5120         return res;
5121 }
5122
5123 static int __init bonding_init(void)
5124 {
5125         int i;
5126         int res;
5127         struct bonding *bond;
5128
5129         printk(KERN_INFO "%s", version);
5130
5131         res = bond_check_params(&bonding_defaults);
5132         if (res) {
5133                 goto out;
5134         }
5135
5136 #ifdef CONFIG_PROC_FS
5137         bond_create_proc_dir();
5138 #endif
5139
5140         init_rwsem(&bonding_rwsem);
5141
5142         for (i = 0; i < max_bonds; i++) {
5143                 res = bond_create(NULL, &bonding_defaults);
5144                 if (res)
5145                         goto err;
5146         }
5147
5148         res = bond_create_sysfs();
5149         if (res)
5150                 goto err;
5151
5152         register_netdevice_notifier(&bond_netdev_notifier);
5153         register_inetaddr_notifier(&bond_inetaddr_notifier);
5154
5155         goto out;
5156 err:
5157         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5158                 bond_work_cancel_all(bond);
5159                 destroy_workqueue(bond->wq);
5160         }
5161
5162         bond_destroy_sysfs();
5163
5164         rtnl_lock();
5165         bond_free_all();
5166         rtnl_unlock();
5167 out:
5168         return res;
5169
5170 }
5171
5172 static void __exit bonding_exit(void)
5173 {
5174         unregister_netdevice_notifier(&bond_netdev_notifier);
5175         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5176
5177         bond_destroy_sysfs();
5178
5179         rtnl_lock();
5180         bond_free_all();
5181         rtnl_unlock();
5182 }
5183
5184 module_init(bonding_init);
5185 module_exit(bonding_exit);
5186 MODULE_LICENSE("GPL");
5187 MODULE_VERSION(DRV_VERSION);
5188 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5189 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5190 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5191
5192 /*
5193  * Local variables:
5194  *  c-indent-level: 8
5195  *  c-basic-offset: 8
5196  *  tab-width: 8
5197  * End:
5198  */
5199