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 miimon       = BOND_LINK_MON_INTERV;
  92 static int updelay      = 0;
  93 static int downdelay    = 0;
  94 static int use_carrier  = 1;
  95 static char *mode       = NULL;
  96 static char *primary    = NULL;
  97 static char *lacp_rate  = NULL;
  98 static char *xmit_hash_policy = NULL;
  99 static int arp_interval = BOND_LINK_ARP_INTERV;
 100 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 101 static char *arp_validate = NULL;
 102 static int fail_over_mac = 0;
 103 struct bond_params bonding_defaults;
 104
 105 module_param(max_bonds, int, 0);
 106 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 107 module_param(miimon, int, 0);
 108 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 109 module_param(updelay, int, 0);
 110 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 111 module_param(downdelay, int, 0);
 112 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 113                             "in milliseconds");
 114 module_param(use_carrier, int, 0);
 115 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 116                               "0 for off, 1 for on (default)");
 117 module_param(mode, charp, 0);
 118 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 119                        "1 for active-backup, 2 for balance-xor, "
 120                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 121                        "6 for balance-alb");
 122 module_param(primary, charp, 0);
 123 MODULE_PARM_DESC(primary, "Primary network device to use");
 124 module_param(lacp_rate, charp, 0);
 125 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 126                             "(slow/fast)");
 127 module_param(xmit_hash_policy, charp, 0);
 128 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 129                                    ", 1 for layer 3+4");
 130 module_param(arp_interval, int, 0);
 131 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 132 module_param_array(arp_ip_target, charp, NULL, 0);
 133 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 134 module_param(arp_validate, charp, 0);
 135 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 136 module_param(fail_over_mac, int, 0);
 137 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  0 of off (default), 1 for on.");
 138
 139 /*----------------------------- Global variables ----------------------------*/
 140
 141 static const char * const version =
 142         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 143
 144 LIST_HEAD(bond_dev_list);
 145
 146 #ifdef CONFIG_PROC_FS
 147 static struct proc_dir_entry *bond_proc_dir = NULL;
 148 #endif
 149
 150 extern struct rw_semaphore bonding_rwsem;
 151 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 152 static int arp_ip_count = 0;
 153 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 154 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 155 static int lacp_fast    = 0;
 156
 157
 158 struct bond_parm_tbl bond_lacp_tbl[] = {
 159 {       "slow",         AD_LACP_SLOW},
 160 {       "fast",         AD_LACP_FAST},
 161 {       NULL,           -1},
 162 };
 163
 164 struct bond_parm_tbl bond_mode_tbl[] = {
 165 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 166 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 167 {       "balance-xor",          BOND_MODE_XOR},
 168 {       "broadcast",            BOND_MODE_BROADCAST},
 169 {       "802.3ad",              BOND_MODE_8023AD},
 170 {       "balance-tlb",          BOND_MODE_TLB},
 171 {       "balance-alb",          BOND_MODE_ALB},
 172 {       NULL,                   -1},
 173 };
 174
 175 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 176 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 177 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 178 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 179 {       NULL,                   -1},
 180 };
 181
 182 struct bond_parm_tbl arp_validate_tbl[] = {
 183 {       "none",                 BOND_ARP_VALIDATE_NONE},
 184 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 185 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 186 {       "all",                  BOND_ARP_VALIDATE_ALL},
 187 {       NULL,                   -1},
 188 };
 189
 190 /*-------------------------- Forward declarations ---------------------------*/
 191
 192 static void bond_send_gratuitous_arp(struct bonding *bond);
 193 static void bond_deinit(struct net_device *bond_dev);
 194
 195 /*---------------------------- General routines -----------------------------*/
 196
 197 static const char *bond_mode_name(int mode)
 198 {
 199         switch (mode) {
 200         case BOND_MODE_ROUNDROBIN :
 201                 return "load balancing (round-robin)";
 202         case BOND_MODE_ACTIVEBACKUP :
 203                 return "fault-tolerance (active-backup)";
 204         case BOND_MODE_XOR :
 205                 return "load balancing (xor)";
 206         case BOND_MODE_BROADCAST :
 207                 return "fault-tolerance (broadcast)";
 208         case BOND_MODE_8023AD:
 209                 return "IEEE 802.3ad Dynamic link aggregation";
 210         case BOND_MODE_TLB:
 211                 return "transmit load balancing";
 212         case BOND_MODE_ALB:
 213                 return "adaptive load balancing";
 214         default:
 215                 return "unknown";
 216         }
 217 }
 218
 219 /*---------------------------------- VLAN -----------------------------------*/
 220
 221 /**
 222  * bond_add_vlan - add a new vlan id on bond
 223  * @bond: bond that got the notification
 224  * @vlan_id: the vlan id to add
 225  *
 226  * Returns -ENOMEM if allocation failed.
 227  */
 228 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 229 {
 230         struct vlan_entry *vlan;
 231
 232         dprintk("bond: %s, vlan id %d\n",
 233                 (bond ? bond->dev->name: "None"), vlan_id);
 234
 235         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 236         if (!vlan) {
 237                 return -ENOMEM;
 238         }
 239
 240         INIT_LIST_HEAD(&vlan->vlan_list);
 241         vlan->vlan_id = vlan_id;
 242         vlan->vlan_ip = 0;
 243
 244         write_lock_bh(&bond->lock);
 245
 246         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 247
 248         write_unlock_bh(&bond->lock);
 249
 250         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 251
 252         return 0;
 253 }
 254
 255 /**
 256  * bond_del_vlan - delete a vlan id from bond
 257  * @bond: bond that got the notification
 258  * @vlan_id: the vlan id to delete
 259  *
 260  * returns -ENODEV if @vlan_id was not found in @bond.
 261  */
 262 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 263 {
 264         struct vlan_entry *vlan, *next;
 265         int res = -ENODEV;
 266
 267         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 268
 269         write_lock_bh(&bond->lock);
 270
 271         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 272                 if (vlan->vlan_id == vlan_id) {
 273                         list_del(&vlan->vlan_list);
 274
 275                         if ((bond->params.mode == BOND_MODE_TLB) ||
 276                             (bond->params.mode == BOND_MODE_ALB)) {
 277                                 bond_alb_clear_vlan(bond, vlan_id);
 278                         }
 279
 280                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 281                                 bond->dev->name);
 282
 283                         kfree(vlan);
 284
 285                         if (list_empty(&bond->vlan_list) &&
 286                             (bond->slave_cnt == 0)) {
 287                                 /* Last VLAN removed and no slaves, so
 288                                  * restore block on adding VLANs. This will
 289                                  * be removed once new slaves that are not
 290                                  * VLAN challenged will be added.
 291                                  */
 292                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 293                         }
 294
 295                         res = 0;
 296                         goto out;
 297                 }
 298         }
 299
 300         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 301                 bond->dev->name);
 302
 303 out:
 304         write_unlock_bh(&bond->lock);
 305         return res;
 306 }
 307
 308 /**
 309  * bond_has_challenged_slaves
 310  * @bond: the bond we're working on
 311  *
 312  * Searches the slave list. Returns 1 if a vlan challenged slave
 313  * was found, 0 otherwise.
 314  *
 315  * Assumes bond->lock is held.
 316  */
 317 static int bond_has_challenged_slaves(struct bonding *bond)
 318 {
 319         struct slave *slave;
 320         int i;
 321
 322         bond_for_each_slave(bond, slave, i) {
 323                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 324                         dprintk("found VLAN challenged slave - %s\n",
 325                                 slave->dev->name);
 326                         return 1;
 327                 }
 328         }
 329
 330         dprintk("no VLAN challenged slaves found\n");
 331         return 0;
 332 }
 333
 334 /**
 335  * bond_next_vlan - safely skip to the next item in the vlans list.
 336  * @bond: the bond we're working on
 337  * @curr: item we're advancing from
 338  *
 339  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 340  * or @curr->next otherwise (even if it is @curr itself again).
 341  *
 342  * Caller must hold bond->lock
 343  */
 344 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 345 {
 346         struct vlan_entry *next, *last;
 347
 348         if (list_empty(&bond->vlan_list)) {
 349                 return NULL;
 350         }
 351
 352         if (!curr) {
 353                 next = list_entry(bond->vlan_list.next,
 354                                   struct vlan_entry, vlan_list);
 355         } else {
 356                 last = list_entry(bond->vlan_list.prev,
 357                                   struct vlan_entry, vlan_list);
 358                 if (last == curr) {
 359                         next = list_entry(bond->vlan_list.next,
 360                                           struct vlan_entry, vlan_list);
 361                 } else {
 362                         next = list_entry(curr->vlan_list.next,
 363                                           struct vlan_entry, vlan_list);
 364                 }
 365         }
 366
 367         return next;
 368 }
 369
 370 /**
 371  * bond_dev_queue_xmit - Prepare skb for xmit.
 372  *
 373  * @bond: bond device that got this skb for tx.
 374  * @skb: hw accel VLAN tagged skb to transmit
 375  * @slave_dev: slave that is supposed to xmit this skbuff
 376  *
 377  * When the bond gets an skb to transmit that is
 378  * already hardware accelerated VLAN tagged, and it
 379  * needs to relay this skb to a slave that is not
 380  * hw accel capable, the skb needs to be "unaccelerated",
 381  * i.e. strip the hwaccel tag and re-insert it as part
 382  * of the payload.
 383  */
 384 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 385 {
 386         unsigned short vlan_id;
 387
 388         if (!list_empty(&bond->vlan_list) &&
 389             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 390             vlan_get_tag(skb, &vlan_id) == 0) {
 391                 skb->dev = slave_dev;
 392                 skb = vlan_put_tag(skb, vlan_id);
 393                 if (!skb) {
 394                         /* vlan_put_tag() frees the skb in case of error,
 395                          * so return success here so the calling functions
 396                          * won't attempt to free is again.
 397                          */
 398                         return 0;
 399                 }
 400         } else {
 401                 skb->dev = slave_dev;
 402         }
 403
 404         skb->priority = 1;
 405         dev_queue_xmit(skb);
 406
 407         return 0;
 408 }
 409
 410 /*
 411  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 412  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 413  * lock because:
 414  * a. This operation is performed in IOCTL context,
 415  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 416  * c. Holding a lock with BH disabled while directly calling a base driver
 417  *    entry point is generally a BAD idea.
 418  *
 419  * The design of synchronization/protection for this operation in the 8021q
 420  * module is good for one or more VLAN devices over a single physical device
 421  * and cannot be extended for a teaming solution like bonding, so there is a
 422  * potential race condition here where a net device from the vlan group might
 423  * be referenced (either by a base driver or the 8021q code) while it is being
 424  * removed from the system. However, it turns out we're not making matters
 425  * worse, and if it works for regular VLAN usage it will work here too.
 426 */
 427
 428 /**
 429  * bond_vlan_rx_register - Propagates registration to slaves
 430  * @bond_dev: bonding net device that got called
 431  * @grp: vlan group being registered
 432  */
 433 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 434 {
 435         struct bonding *bond = bond_dev->priv;
 436         struct slave *slave;
 437         int i;
 438
 439         bond->vlgrp = grp;
 440
 441         bond_for_each_slave(bond, slave, i) {
 442                 struct net_device *slave_dev = slave->dev;
 443
 444                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 445                     slave_dev->vlan_rx_register) {
 446                         slave_dev->vlan_rx_register(slave_dev, grp);
 447                 }
 448         }
 449 }
 450
 451 /**
 452  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 453  * @bond_dev: bonding net device that got called
 454  * @vid: vlan id being added
 455  */
 456 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 457 {
 458         struct bonding *bond = bond_dev->priv;
 459         struct slave *slave;
 460         int i, res;
 461
 462         bond_for_each_slave(bond, slave, i) {
 463                 struct net_device *slave_dev = slave->dev;
 464
 465                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 466                     slave_dev->vlan_rx_add_vid) {
 467                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 468                 }
 469         }
 470
 471         res = bond_add_vlan(bond, vid);
 472         if (res) {
 473                 printk(KERN_ERR DRV_NAME
 474                        ": %s: Error: Failed to add vlan id %d\n",
 475                        bond_dev->name, vid);
 476         }
 477 }
 478
 479 /**
 480  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 481  * @bond_dev: bonding net device that got called
 482  * @vid: vlan id being removed
 483  */
 484 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 485 {
 486         struct bonding *bond = bond_dev->priv;
 487         struct slave *slave;
 488         struct net_device *vlan_dev;
 489         int i, res;
 490
 491         bond_for_each_slave(bond, slave, i) {
 492                 struct net_device *slave_dev = slave->dev;
 493
 494                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 495                     slave_dev->vlan_rx_kill_vid) {
 496                         /* Save and then restore vlan_dev in the grp array,
 497                          * since the slave's driver might clear it.
 498                          */
 499                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 500                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 501                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 502                 }
 503         }
 504
 505         res = bond_del_vlan(bond, vid);
 506         if (res) {
 507                 printk(KERN_ERR DRV_NAME
 508                        ": %s: Error: Failed to remove vlan id %d\n",
 509                        bond_dev->name, vid);
 510         }
 511 }
 512
 513 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 514 {
 515         struct vlan_entry *vlan;
 516
 517         write_lock_bh(&bond->lock);
 518
 519         if (list_empty(&bond->vlan_list)) {
 520                 goto out;
 521         }
 522
 523         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 524             slave_dev->vlan_rx_register) {
 525                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 526         }
 527
 528         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 529             !(slave_dev->vlan_rx_add_vid)) {
 530                 goto out;
 531         }
 532
 533         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 534                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 535         }
 536
 537 out:
 538         write_unlock_bh(&bond->lock);
 539 }
 540
 541 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 542 {
 543         struct vlan_entry *vlan;
 544         struct net_device *vlan_dev;
 545
 546         write_lock_bh(&bond->lock);
 547
 548         if (list_empty(&bond->vlan_list)) {
 549                 goto out;
 550         }
 551
 552         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 553             !(slave_dev->vlan_rx_kill_vid)) {
 554                 goto unreg;
 555         }
 556
 557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 558                 /* Save and then restore vlan_dev in the grp array,
 559                  * since the slave's driver might clear it.
 560                  */
 561                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 562                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 563                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 564         }
 565
 566 unreg:
 567         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 568             slave_dev->vlan_rx_register) {
 569                 slave_dev->vlan_rx_register(slave_dev, NULL);
 570         }
 571
 572 out:
 573         write_unlock_bh(&bond->lock);
 574 }
 575
 576 /*------------------------------- Link status -------------------------------*/
 577
 578 /*
 579  * Set the carrier state for the master according to the state of its
 580  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 581  * do special 802.3ad magic.
 582  *
 583  * Returns zero if carrier state does not change, nonzero if it does.
 584  */
 585 static int bond_set_carrier(struct bonding *bond)
 586 {
 587         struct slave *slave;
 588         int i;
 589
 590         if (bond->slave_cnt == 0)
 591                 goto down;
 592
 593         if (bond->params.mode == BOND_MODE_8023AD)
 594                 return bond_3ad_set_carrier(bond);
 595
 596         bond_for_each_slave(bond, slave, i) {
 597                 if (slave->link == BOND_LINK_UP) {
 598                         if (!netif_carrier_ok(bond->dev)) {
 599                                 netif_carrier_on(bond->dev);
 600                                 return 1;
 601                         }
 602                         return 0;
 603                 }
 604         }
 605
 606 down:
 607         if (netif_carrier_ok(bond->dev)) {
 608                 netif_carrier_off(bond->dev);
 609                 return 1;
 610         }
 611         return 0;
 612 }
 613
 614 /*
 615  * Get link speed and duplex from the slave's base driver
 616  * using ethtool. If for some reason the call fails or the
 617  * values are invalid, fake speed and duplex to 100/Full
 618  * and return error.
 619  */
 620 static int bond_update_speed_duplex(struct slave *slave)
 621 {
 622         struct net_device *slave_dev = slave->dev;
 623         struct ethtool_cmd etool;
 624         int res;
 625
 626         /* Fake speed and duplex */
 627         slave->speed = SPEED_100;
 628         slave->duplex = DUPLEX_FULL;
 629
 630         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 631                 return -1;
 632
 633         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 634         if (res < 0)
 635                 return -1;
 636
 637         switch (etool.speed) {
 638         case SPEED_10:
 639         case SPEED_100:
 640         case SPEED_1000:
 641         case SPEED_10000:
 642                 break;
 643         default:
 644                 return -1;
 645         }
 646
 647         switch (etool.duplex) {
 648         case DUPLEX_FULL:
 649         case DUPLEX_HALF:
 650                 break;
 651         default:
 652                 return -1;
 653         }
 654
 655         slave->speed = etool.speed;
 656         slave->duplex = etool.duplex;
 657
 658         return 0;
 659 }
 660
 661 /*
 662  * if <dev> supports MII link status reporting, check its link status.
 663  *
 664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 665  * depening upon the setting of the use_carrier parameter.
 666  *
 667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 668  * can't tell and just pretend it is), or 0, meaning that the link is
 669  * down.
 670  *
 671  * If reporting is non-zero, instead of faking link up, return -1 if
 672  * both ETHTOOL and MII ioctls fail (meaning the device does not
 673  * support them).  If use_carrier is set, return whatever it says.
 674  * It'd be nice if there was a good way to tell if a driver supports
 675  * netif_carrier, but there really isn't.
 676  */
 677 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 678 {
 679         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 680         struct ifreq ifr;
 681         struct mii_ioctl_data *mii;
 682
 683         if (bond->params.use_carrier) {
 684                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 685         }
 686
 687         ioctl = slave_dev->do_ioctl;
 688         if (ioctl) {
 689                 /* TODO: set pointer to correct ioctl on a per team member */
 690                 /*       bases to make this more efficient. that is, once  */
 691                 /*       we determine the correct ioctl, we will always    */
 692                 /*       call it and not the others for that team          */
 693                 /*       member.                                           */
 694
 695                 /*
 696                  * We cannot assume that SIOCGMIIPHY will also read a
 697                  * register; not all network drivers (e.g., e100)
 698                  * support that.
 699                  */
 700
 701                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 702                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 703                 mii = if_mii(&ifr);
 704                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 705                         mii->reg_num = MII_BMSR;
 706                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 707                                 return (mii->val_out & BMSR_LSTATUS);
 708                         }
 709                 }
 710         }
 711
 712         /*
 713          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 714          * attempt to get link status from it if the above MII ioctls fail.
 715          */
 716         if (slave_dev->ethtool_ops) {
 717                 if (slave_dev->ethtool_ops->get_link) {
 718                         u32 link;
 719
 720                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 721
 722                         return link ? BMSR_LSTATUS : 0;
 723                 }
 724         }
 725
 726         /*
 727          * If reporting, report that either there's no dev->do_ioctl,
 728          * or both SIOCGMIIREG and get_link failed (meaning that we
 729          * cannot report link status).  If not reporting, pretend
 730          * we're ok.
 731          */
 732         return (reporting ? -1 : BMSR_LSTATUS);
 733 }
 734
 735 /*----------------------------- Multicast list ------------------------------*/
 736
 737 /*
 738  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 739  */
 740 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 741 {
 742         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 743                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 744 }
 745
 746 /*
 747  * returns dmi entry if found, NULL otherwise
 748  */
 749 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 750 {
 751         struct dev_mc_list *idmi;
 752
 753         for (idmi = mc_list; idmi; idmi = idmi->next) {
 754                 if (bond_is_dmi_same(dmi, idmi)) {
 755                         return idmi;
 756                 }
 757         }
 758
 759         return NULL;
 760 }
 761
 762 /*
 763  * Push the promiscuity flag down to appropriate slaves
 764  */
 765 static void bond_set_promiscuity(struct bonding *bond, int inc)
 766 {
 767         if (USES_PRIMARY(bond->params.mode)) {
 768                 /* write lock already acquired */
 769                 if (bond->curr_active_slave) {
 770                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 771                 }
 772         } else {
 773                 struct slave *slave;
 774                 int i;
 775                 bond_for_each_slave(bond, slave, i) {
 776                         dev_set_promiscuity(slave->dev, inc);
 777                 }
 778         }
 779 }
 780
 781 /*
 782  * Push the allmulti flag down to all slaves
 783  */
 784 static void bond_set_allmulti(struct bonding *bond, int inc)
 785 {
 786         if (USES_PRIMARY(bond->params.mode)) {
 787                 /* write lock already acquired */
 788                 if (bond->curr_active_slave) {
 789                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 790                 }
 791         } else {
 792                 struct slave *slave;
 793                 int i;
 794                 bond_for_each_slave(bond, slave, i) {
 795                         dev_set_allmulti(slave->dev, inc);
 796                 }
 797         }
 798 }
 799
 800 /*
 801  * Add a Multicast address to slaves
 802  * according to mode
 803  */
 804 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 805 {
 806         if (USES_PRIMARY(bond->params.mode)) {
 807                 /* write lock already acquired */
 808                 if (bond->curr_active_slave) {
 809                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 810                 }
 811         } else {
 812                 struct slave *slave;
 813                 int i;
 814                 bond_for_each_slave(bond, slave, i) {
 815                         dev_mc_add(slave->dev, addr, alen, 0);
 816                 }
 817         }
 818 }
 819
 820 /*
 821  * Remove a multicast address from slave
 822  * according to mode
 823  */
 824 static void bond_mc_delete(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_delete(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_delete(slave->dev, addr, alen, 0);
 836                 }
 837         }
 838 }
 839
 840
 841 /*
 842  * Retrieve the list of registered multicast addresses for the bonding
 843  * device and retransmit an IGMP JOIN request to the current active
 844  * slave.
 845  */
 846 static void bond_resend_igmp_join_requests(struct bonding *bond)
 847 {
 848         struct in_device *in_dev;
 849         struct ip_mc_list *im;
 850
 851         rcu_read_lock();
 852         in_dev = __in_dev_get_rcu(bond->dev);
 853         if (in_dev) {
 854                 for (im = in_dev->mc_list; im; im = im->next) {
 855                         ip_mc_rejoin_group(im);
 856                 }
 857         }
 858
 859         rcu_read_unlock();
 860 }
 861
 862 /*
 863  * Totally destroys the mc_list in bond
 864  */
 865 static void bond_mc_list_destroy(struct bonding *bond)
 866 {
 867         struct dev_mc_list *dmi;
 868
 869         dmi = bond->mc_list;
 870         while (dmi) {
 871                 bond->mc_list = dmi->next;
 872                 kfree(dmi);
 873                 dmi = bond->mc_list;
 874         }
 875         bond->mc_list = NULL;
 876 }
 877
 878 /*
 879  * Copy all the Multicast addresses from src to the bonding device dst
 880  */
 881 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 882                              gfp_t gfp_flag)
 883 {
 884         struct dev_mc_list *dmi, *new_dmi;
 885
 886         for (dmi = mc_list; dmi; dmi = dmi->next) {
 887                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 888
 889                 if (!new_dmi) {
 890                         /* FIXME: Potential memory leak !!! */
 891                         return -ENOMEM;
 892                 }
 893
 894                 new_dmi->next = bond->mc_list;
 895                 bond->mc_list = new_dmi;
 896                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 897                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 898                 new_dmi->dmi_users = dmi->dmi_users;
 899                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 900         }
 901
 902         return 0;
 903 }
 904
 905 /*
 906  * flush all members of flush->mc_list from device dev->mc_list
 907  */
 908 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 909 {
 910         struct bonding *bond = bond_dev->priv;
 911         struct dev_mc_list *dmi;
 912
 913         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 914                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 915         }
 916
 917         if (bond->params.mode == BOND_MODE_8023AD) {
 918                 /* del lacpdu mc addr from mc list */
 919                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 920
 921                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 922         }
 923 }
 924
 925 /*--------------------------- Active slave change ---------------------------*/
 926
 927 /*
 928  * Update the mc list and multicast-related flags for the new and
 929  * old active slaves (if any) according to the multicast mode, and
 930  * promiscuous flags unconditionally.
 931  */
 932 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 933 {
 934         struct dev_mc_list *dmi;
 935
 936         if (!USES_PRIMARY(bond->params.mode)) {
 937                 /* nothing to do -  mc list is already up-to-date on
 938                  * all slaves
 939                  */
 940                 return;
 941         }
 942
 943         if (old_active) {
 944                 if (bond->dev->flags & IFF_PROMISC) {
 945                         dev_set_promiscuity(old_active->dev, -1);
 946                 }
 947
 948                 if (bond->dev->flags & IFF_ALLMULTI) {
 949                         dev_set_allmulti(old_active->dev, -1);
 950                 }
 951
 952                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 953                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 954                 }
 955         }
 956
 957         if (new_active) {
 958                 if (bond->dev->flags & IFF_PROMISC) {
 959                         dev_set_promiscuity(new_active->dev, 1);
 960                 }
 961
 962                 if (bond->dev->flags & IFF_ALLMULTI) {
 963                         dev_set_allmulti(new_active->dev, 1);
 964                 }
 965
 966                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 967                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 968                 }
 969                 bond_resend_igmp_join_requests(bond);
 970         }
 971 }
 972
 973 /**
 974  * find_best_interface - select the best available slave to be the active one
 975  * @bond: our bonding struct
 976  *
 977  * Warning: Caller must hold curr_slave_lock for writing.
 978  */
 979 static struct slave *bond_find_best_slave(struct bonding *bond)
 980 {
 981         struct slave *new_active, *old_active;
 982         struct slave *bestslave = NULL;
 983         int mintime = bond->params.updelay;
 984         int i;
 985
 986         new_active = old_active = bond->curr_active_slave;
 987
 988         if (!new_active) { /* there were no active slaves left */
 989                 if (bond->slave_cnt > 0) {  /* found one slave */
 990                         new_active = bond->first_slave;
 991                 } else {
 992                         return NULL; /* still no slave, return NULL */
 993                 }
 994         }
 995
 996         /* first try the primary link; if arping, a link must tx/rx traffic
 997          * before it can be considered the curr_active_slave - also, we would skip
 998          * slaves between the curr_active_slave and primary_slave that may be up
 999          * and able to arp
1000          */
1001         if ((bond->primary_slave) &&
1002             (!bond->params.arp_interval) &&
1003             (IS_UP(bond->primary_slave->dev))) {
1004                 new_active = bond->primary_slave;
1005         }
1006
1007         /* remember where to stop iterating over the slaves */
1008         old_active = new_active;
1009
1010         bond_for_each_slave_from(bond, new_active, i, old_active) {
1011                 if (IS_UP(new_active->dev)) {
1012                         if (new_active->link == BOND_LINK_UP) {
1013                                 return new_active;
1014                         } else if (new_active->link == BOND_LINK_BACK) {
1015                                 /* link up, but waiting for stabilization */
1016                                 if (new_active->delay < mintime) {
1017                                         mintime = new_active->delay;
1018                                         bestslave = new_active;
1019                                 }
1020                         }
1021                 }
1022         }
1023
1024         return bestslave;
1025 }
1026
1027 /**
1028  * change_active_interface - change the active slave into the specified one
1029  * @bond: our bonding struct
1030  * @new: the new slave to make the active one
1031  *
1032  * Set the new slave to the bond's settings and unset them on the old
1033  * curr_active_slave.
1034  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035  *
1036  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037  * because it is apparently the best available slave we have, even though its
1038  * updelay hasn't timed out yet.
1039  *
1040  * Warning: Caller must hold curr_slave_lock for writing.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active) {
1047                 return;
1048         }
1049
1050         if (new_active) {
1051                 if (new_active->link == BOND_LINK_BACK) {
1052                         if (USES_PRIMARY(bond->params.mode)) {
1053                                 printk(KERN_INFO DRV_NAME
1054                                        ": %s: making interface %s the new "
1055                                        "active one %d ms earlier.\n",
1056                                        bond->dev->name, new_active->dev->name,
1057                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1058                         }
1059
1060                         new_active->delay = 0;
1061                         new_active->link = BOND_LINK_UP;
1062                         new_active->jiffies = jiffies;
1063
1064                         if (bond->params.mode == BOND_MODE_8023AD) {
1065                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1066                         }
1067
1068                         if ((bond->params.mode == BOND_MODE_TLB) ||
1069                             (bond->params.mode == BOND_MODE_ALB)) {
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                         }
1072                 } else {
1073                         if (USES_PRIMARY(bond->params.mode)) {
1074                                 printk(KERN_INFO DRV_NAME
1075                                        ": %s: making interface %s the new "
1076                                        "active one.\n",
1077                                        bond->dev->name, new_active->dev->name);
1078                         }
1079                 }
1080         }
1081
1082         if (USES_PRIMARY(bond->params.mode)) {
1083                 bond_mc_swap(bond, new_active, old_active);
1084         }
1085
1086         if ((bond->params.mode == BOND_MODE_TLB) ||
1087             (bond->params.mode == BOND_MODE_ALB)) {
1088                 bond_alb_handle_active_change(bond, new_active);
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091                 if (new_active)
1092                         bond_set_slave_active_flags(new_active);
1093         } else {
1094                 bond->curr_active_slave = new_active;
1095         }
1096
1097         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098                 if (old_active) {
1099                         bond_set_slave_inactive_flags(old_active);
1100                 }
1101
1102                 if (new_active) {
1103                         bond_set_slave_active_flags(new_active);
1104                 }
1105
1106                 /* when bonding does not set the slave MAC address, the bond MAC
1107                  * address is the one of the active slave.
1108                  */
1109                 if (new_active && bond->params.fail_over_mac)
1110                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1111                                 new_active->dev->addr_len);
1112                 if (bond->curr_active_slave &&
1113                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1114                                         &bond->curr_active_slave->dev->state)) {
1115                         dprintk("delaying gratuitous arp on %s\n",
1116                                 bond->curr_active_slave->dev->name);
1117                         bond->send_grat_arp = 1;
1118                 } else
1119                         bond_send_gratuitous_arp(bond);
1120         }
1121 }
1122
1123 /**
1124  * bond_select_active_slave - select a new active slave, if needed
1125  * @bond: our bonding struct
1126  *
1127  * This functions shoud be called when one of the following occurs:
1128  * - The old curr_active_slave has been released or lost its link.
1129  * - The primary_slave has got its link back.
1130  * - A slave has got its link back and there's no old curr_active_slave.
1131  *
1132  * Warning: Caller must hold curr_slave_lock for writing.
1133  */
1134 void bond_select_active_slave(struct bonding *bond)
1135 {
1136         struct slave *best_slave;
1137         int rv;
1138
1139         best_slave = bond_find_best_slave(bond);
1140         if (best_slave != bond->curr_active_slave) {
1141                 bond_change_active_slave(bond, best_slave);
1142                 rv = bond_set_carrier(bond);
1143                 if (!rv)
1144                         return;
1145
1146                 if (netif_carrier_ok(bond->dev)) {
1147                         printk(KERN_INFO DRV_NAME
1148                                ": %s: first active interface up!\n",
1149                                bond->dev->name);
1150                 } else {
1151                         printk(KERN_INFO DRV_NAME ": %s: "
1152                                "now running without any active interface !\n",
1153                                bond->dev->name);
1154                 }
1155         }
1156 }
1157
1158 /*--------------------------- slave list handling ---------------------------*/
1159
1160 /*
1161  * This function attaches the slave to the end of list.
1162  *
1163  * bond->lock held for writing by caller.
1164  */
1165 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1166 {
1167         if (bond->first_slave == NULL) { /* attaching the first slave */
1168                 new_slave->next = new_slave;
1169                 new_slave->prev = new_slave;
1170                 bond->first_slave = new_slave;
1171         } else {
1172                 new_slave->next = bond->first_slave;
1173                 new_slave->prev = bond->first_slave->prev;
1174                 new_slave->next->prev = new_slave;
1175                 new_slave->prev->next = new_slave;
1176         }
1177
1178         bond->slave_cnt++;
1179 }
1180
1181 /*
1182  * This function detaches the slave from the list.
1183  * WARNING: no check is made to verify if the slave effectively
1184  * belongs to <bond>.
1185  * Nothing is freed on return, structures are just unchained.
1186  * If any slave pointer in bond was pointing to <slave>,
1187  * it should be changed by the calling function.
1188  *
1189  * bond->lock held for writing by caller.
1190  */
1191 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1192 {
1193         if (slave->next) {
1194                 slave->next->prev = slave->prev;
1195         }
1196
1197         if (slave->prev) {
1198                 slave->prev->next = slave->next;
1199         }
1200
1201         if (bond->first_slave == slave) { /* slave is the first slave */
1202                 if (bond->slave_cnt > 1) { /* there are more slave */
1203                         bond->first_slave = slave->next;
1204                 } else {
1205                         bond->first_slave = NULL; /* slave was the last one */
1206                 }
1207         }
1208
1209         slave->next = NULL;
1210         slave->prev = NULL;
1211         bond->slave_cnt--;
1212 }
1213
1214 /*---------------------------------- IOCTL ----------------------------------*/
1215
1216 static int bond_sethwaddr(struct net_device *bond_dev,
1217                           struct net_device *slave_dev)
1218 {
1219         dprintk("bond_dev=%p\n", bond_dev);
1220         dprintk("slave_dev=%p\n", slave_dev);
1221         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1222         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1223         return 0;
1224 }
1225
1226 #define BOND_VLAN_FEATURES \
1227         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1228          NETIF_F_HW_VLAN_FILTER)
1229
1230 /*
1231  * Compute the common dev->feature set available to all slaves.  Some
1232  * feature bits are managed elsewhere, so preserve those feature bits
1233  * on the master device.
1234  */
1235 static int bond_compute_features(struct bonding *bond)
1236 {
1237         struct slave *slave;
1238         struct net_device *bond_dev = bond->dev;
1239         unsigned long features = bond_dev->features;
1240         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1241                                                 bond_dev->hard_header_len);
1242         int i;
1243
1244         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1245         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1246                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1247
1248         bond_for_each_slave(bond, slave, i) {
1249                 features = netdev_compute_features(features,
1250                                                    slave->dev->features);
1251                 if (slave->dev->hard_header_len > max_hard_header_len)
1252                         max_hard_header_len = slave->dev->hard_header_len;
1253         }
1254
1255         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1256         bond_dev->features = features;
1257         bond_dev->hard_header_len = max_hard_header_len;
1258
1259         return 0;
1260 }
1261
1262
1263 static void bond_setup_by_slave(struct net_device *bond_dev,
1264                                 struct net_device *slave_dev)
1265 {
1266         struct bonding *bond = bond_dev->priv;
1267
1268         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1269         bond_dev->header_ops            = slave_dev->header_ops;
1270
1271         bond_dev->type              = slave_dev->type;
1272         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1273         bond_dev->addr_len          = slave_dev->addr_len;
1274
1275         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1276                 slave_dev->addr_len);
1277         bond->setup_by_slave = 1;
1278 }
1279
1280 /* enslave device <slave> to bond device <master> */
1281 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1282 {
1283         struct bonding *bond = bond_dev->priv;
1284         struct slave *new_slave = NULL;
1285         struct dev_mc_list *dmi;
1286         struct sockaddr addr;
1287         int link_reporting;
1288         int old_features = bond_dev->features;
1289         int res = 0;
1290
1291         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1292                 slave_dev->do_ioctl == NULL) {
1293                 printk(KERN_WARNING DRV_NAME
1294                        ": %s: Warning: no link monitoring support for %s\n",
1295                        bond_dev->name, slave_dev->name);
1296         }
1297
1298         /* bond must be initialized by bond_open() before enslaving */
1299         if (!(bond_dev->flags & IFF_UP)) {
1300                 printk(KERN_WARNING DRV_NAME
1301                         " %s: master_dev is not up in bond_enslave\n",
1302                         bond_dev->name);
1303         }
1304
1305         /* already enslaved */
1306         if (slave_dev->flags & IFF_SLAVE) {
1307                 dprintk("Error, Device was already enslaved\n");
1308                 return -EBUSY;
1309         }
1310
1311         /* vlan challenged mutual exclusion */
1312         /* no need to lock since we're protected by rtnl_lock */
1313         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1315                 if (!list_empty(&bond->vlan_list)) {
1316                         printk(KERN_ERR DRV_NAME
1317                                ": %s: Error: cannot enslave VLAN "
1318                                "challenged slave %s on VLAN enabled "
1319                                "bond %s\n", bond_dev->name, slave_dev->name,
1320                                bond_dev->name);
1321                         return -EPERM;
1322                 } else {
1323                         printk(KERN_WARNING DRV_NAME
1324                                ": %s: Warning: enslaved VLAN challenged "
1325                                "slave %s. Adding VLANs will be blocked as "
1326                                "long as %s is part of bond %s\n",
1327                                bond_dev->name, slave_dev->name, slave_dev->name,
1328                                bond_dev->name);
1329                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1330                 }
1331         } else {
1332                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1333                 if (bond->slave_cnt == 0) {
1334                         /* First slave, and it is not VLAN challenged,
1335                          * so remove the block of adding VLANs over the bond.
1336                          */
1337                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1338                 }
1339         }
1340
1341         /*
1342          * Old ifenslave binaries are no longer supported.  These can
1343          * be identified with moderate accurary by the state of the slave:
1344          * the current ifenslave will set the interface down prior to
1345          * enslaving it; the old ifenslave will not.
1346          */
1347         if ((slave_dev->flags & IFF_UP)) {
1348                 printk(KERN_ERR DRV_NAME ": %s is up. "
1349                        "This may be due to an out of date ifenslave.\n",
1350                        slave_dev->name);
1351                 res = -EPERM;
1352                 goto err_undo_flags;
1353         }
1354
1355         /* set bonding device ether type by slave - bonding netdevices are
1356          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1357          * there is a need to override some of the type dependent attribs/funcs.
1358          *
1359          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1360          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1361          */
1362         if (bond->slave_cnt == 0) {
1363                 if (slave_dev->type != ARPHRD_ETHER)
1364                         bond_setup_by_slave(bond_dev, slave_dev);
1365         } else if (bond_dev->type != slave_dev->type) {
1366                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1367                         "from other slaves (%d), can not enslave it.\n",
1368                         slave_dev->name,
1369                         slave_dev->type, bond_dev->type);
1370                         res = -EINVAL;
1371                         goto err_undo_flags;
1372         }
1373
1374         if (slave_dev->set_mac_address == NULL) {
1375                 if (bond->slave_cnt == 0) {
1376                         printk(KERN_WARNING DRV_NAME
1377                                ": %s: Warning: The first slave device "
1378                                "specified does not support setting the MAC "
1379                                "address. Enabling the fail_over_mac option.",
1380                                bond_dev->name);
1381                         bond->params.fail_over_mac = 1;
1382                 } else if (!bond->params.fail_over_mac) {
1383                         printk(KERN_ERR DRV_NAME
1384                                 ": %s: Error: The slave device specified "
1385                                 "does not support setting the MAC address, "
1386                                 "but fail_over_mac is not enabled.\n"
1387                                 , bond_dev->name);
1388                         res = -EOPNOTSUPP;
1389                         goto err_undo_flags;
1390                 }
1391         }
1392
1393         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1394         if (!new_slave) {
1395                 res = -ENOMEM;
1396                 goto err_undo_flags;
1397         }
1398
1399         /* save slave's original flags before calling
1400          * netdev_set_master and dev_open
1401          */
1402         new_slave->original_flags = slave_dev->flags;
1403
1404         /*
1405          * Save slave's original ("permanent") mac address for modes
1406          * that need it, and for restoring it upon release, and then
1407          * set it to the master's address
1408          */
1409         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1410
1411         if (!bond->params.fail_over_mac) {
1412                 /*
1413                  * Set slave to master's mac address.  The application already
1414                  * set the master's mac address to that of the first slave
1415                  */
1416                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1417                 addr.sa_family = slave_dev->type;
1418                 res = dev_set_mac_address(slave_dev, &addr);
1419                 if (res) {
1420                         dprintk("Error %d calling set_mac_address\n", res);
1421                         goto err_free;
1422                 }
1423         }
1424
1425         res = netdev_set_master(slave_dev, bond_dev);
1426         if (res) {
1427                 dprintk("Error %d calling netdev_set_master\n", res);
1428                 goto err_close;
1429         }
1430         /* open the slave since the application closed it */
1431         res = dev_open(slave_dev);
1432         if (res) {
1433                 dprintk("Openning slave %s failed\n", slave_dev->name);
1434                 goto err_restore_mac;
1435         }
1436
1437         new_slave->dev = slave_dev;
1438         slave_dev->priv_flags |= IFF_BONDING;
1439
1440         if ((bond->params.mode == BOND_MODE_TLB) ||
1441             (bond->params.mode == BOND_MODE_ALB)) {
1442                 /* bond_alb_init_slave() must be called before all other stages since
1443                  * it might fail and we do not want to have to undo everything
1444                  */
1445                 res = bond_alb_init_slave(bond, new_slave);
1446                 if (res) {
1447                         goto err_unset_master;
1448                 }
1449         }
1450
1451         /* If the mode USES_PRIMARY, then the new slave gets the
1452          * master's promisc (and mc) settings only if it becomes the
1453          * curr_active_slave, and that is taken care of later when calling
1454          * bond_change_active()
1455          */
1456         if (!USES_PRIMARY(bond->params.mode)) {
1457                 /* set promiscuity level to new slave */
1458                 if (bond_dev->flags & IFF_PROMISC) {
1459                         dev_set_promiscuity(slave_dev, 1);
1460                 }
1461
1462                 /* set allmulti level to new slave */
1463                 if (bond_dev->flags & IFF_ALLMULTI) {
1464                         dev_set_allmulti(slave_dev, 1);
1465                 }
1466
1467                 /* upload master's mc_list to new slave */
1468                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1469                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1470                 }
1471         }
1472
1473         if (bond->params.mode == BOND_MODE_8023AD) {
1474                 /* add lacpdu mc addr to mc list */
1475                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1476
1477                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1478         }
1479
1480         bond_add_vlans_on_slave(bond, slave_dev);
1481
1482         write_lock_bh(&bond->lock);
1483
1484         bond_attach_slave(bond, new_slave);
1485
1486         new_slave->delay = 0;
1487         new_slave->link_failure_count = 0;
1488
1489         bond_compute_features(bond);
1490
1491         new_slave->last_arp_rx = jiffies;
1492
1493         if (bond->params.miimon && !bond->params.use_carrier) {
1494                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1495
1496                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1497                         /*
1498                          * miimon is set but a bonded network driver
1499                          * does not support ETHTOOL/MII and
1500                          * arp_interval is not set.  Note: if
1501                          * use_carrier is enabled, we will never go
1502                          * here (because netif_carrier is always
1503                          * supported); thus, we don't need to change
1504                          * the messages for netif_carrier.
1505                          */
1506                         printk(KERN_WARNING DRV_NAME
1507                                ": %s: Warning: MII and ETHTOOL support not "
1508                                "available for interface %s, and "
1509                                "arp_interval/arp_ip_target module parameters "
1510                                "not specified, thus bonding will not detect "
1511                                "link failures! see bonding.txt for details.\n",
1512                                bond_dev->name, slave_dev->name);
1513                 } else if (link_reporting == -1) {
1514                         /* unable get link status using mii/ethtool */
1515                         printk(KERN_WARNING DRV_NAME
1516                                ": %s: Warning: can't get link status from "
1517                                "interface %s; the network driver associated "
1518                                "with this interface does not support MII or "
1519                                "ETHTOOL link status reporting, thus miimon "
1520                                "has no effect on this interface.\n",
1521                                bond_dev->name, slave_dev->name);
1522                 }
1523         }
1524
1525         /* check for initial state */
1526         if (!bond->params.miimon ||
1527             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1528                 if (bond->params.updelay) {
1529                         dprintk("Initial state of slave_dev is "
1530                                 "BOND_LINK_BACK\n");
1531                         new_slave->link  = BOND_LINK_BACK;
1532                         new_slave->delay = bond->params.updelay;
1533                 } else {
1534                         dprintk("Initial state of slave_dev is "
1535                                 "BOND_LINK_UP\n");
1536                         new_slave->link  = BOND_LINK_UP;
1537                 }
1538                 new_slave->jiffies = jiffies;
1539         } else {
1540                 dprintk("Initial state of slave_dev is "
1541                         "BOND_LINK_DOWN\n");
1542                 new_slave->link  = BOND_LINK_DOWN;
1543         }
1544
1545         if (bond_update_speed_duplex(new_slave) &&
1546             (new_slave->link != BOND_LINK_DOWN)) {
1547                 printk(KERN_WARNING DRV_NAME
1548                        ": %s: Warning: failed to get speed and duplex from %s, "
1549                        "assumed to be 100Mb/sec and Full.\n",
1550                        bond_dev->name, new_slave->dev->name);
1551
1552                 if (bond->params.mode == BOND_MODE_8023AD) {
1553                         printk(KERN_WARNING DRV_NAME
1554                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1555                                "support in base driver for proper aggregator "
1556                                "selection.\n", bond_dev->name);
1557                 }
1558         }
1559
1560         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1561                 /* if there is a primary slave, remember it */
1562                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1563                         bond->primary_slave = new_slave;
1564                 }
1565         }
1566
1567         switch (bond->params.mode) {
1568         case BOND_MODE_ACTIVEBACKUP:
1569                 bond_set_slave_inactive_flags(new_slave);
1570                 bond_select_active_slave(bond);
1571                 break;
1572         case BOND_MODE_8023AD:
1573                 /* in 802.3ad mode, the internal mechanism
1574                  * will activate the slaves in the selected
1575                  * aggregator
1576                  */
1577                 bond_set_slave_inactive_flags(new_slave);
1578                 /* if this is the first slave */
1579                 if (bond->slave_cnt == 1) {
1580                         SLAVE_AD_INFO(new_slave).id = 1;
1581                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1582                          * can be called only after the mac address of the bond is set
1583                          */
1584                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1585                                             bond->params.lacp_fast);
1586                 } else {
1587                         SLAVE_AD_INFO(new_slave).id =
1588                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1589                 }
1590
1591                 bond_3ad_bind_slave(new_slave);
1592                 break;
1593         case BOND_MODE_TLB:
1594         case BOND_MODE_ALB:
1595                 new_slave->state = BOND_STATE_ACTIVE;
1596                 bond_set_slave_inactive_flags(new_slave);
1597                 break;
1598         default:
1599                 dprintk("This slave is always active in trunk mode\n");
1600
1601                 /* always active in trunk mode */
1602                 new_slave->state = BOND_STATE_ACTIVE;
1603
1604                 /* In trunking mode there is little meaning to curr_active_slave
1605                  * anyway (it holds no special properties of the bond device),
1606                  * so we can change it without calling change_active_interface()
1607                  */
1608                 if (!bond->curr_active_slave) {
1609                         bond->curr_active_slave = new_slave;
1610                 }
1611                 break;
1612         } /* switch(bond_mode) */
1613
1614         bond_set_carrier(bond);
1615
1616         write_unlock_bh(&bond->lock);
1617
1618         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1619         if (res)
1620                 goto err_unset_master;
1621
1622         printk(KERN_INFO DRV_NAME
1623                ": %s: enslaving %s as a%s interface with a%s link.\n",
1624                bond_dev->name, slave_dev->name,
1625                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1626                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1627
1628         /* enslave is successful */
1629         return 0;
1630
1631 /* Undo stages on error */
1632 err_unset_master:
1633         netdev_set_master(slave_dev, NULL);
1634
1635 err_close:
1636         dev_close(slave_dev);
1637
1638 err_restore_mac:
1639         if (!bond->params.fail_over_mac) {
1640                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1641                 addr.sa_family = slave_dev->type;
1642                 dev_set_mac_address(slave_dev, &addr);
1643         }
1644
1645 err_free:
1646         kfree(new_slave);
1647
1648 err_undo_flags:
1649         bond_dev->features = old_features;
1650
1651         return res;
1652 }
1653
1654 /*
1655  * Try to release the slave device <slave> from the bond device <master>
1656  * It is legal to access curr_active_slave without a lock because all the function
1657  * is write-locked.
1658  *
1659  * The rules for slave state should be:
1660  *   for Active/Backup:
1661  *     Active stays on all backups go down
1662  *   for Bonded connections:
1663  *     The first up interface should be left on and all others downed.
1664  */
1665 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1666 {
1667         struct bonding *bond = bond_dev->priv;
1668         struct slave *slave, *oldcurrent;
1669         struct sockaddr addr;
1670         int mac_addr_differ;
1671         DECLARE_MAC_BUF(mac);
1672
1673         /* slave is not a slave or master is not master of this slave */
1674         if (!(slave_dev->flags & IFF_SLAVE) ||
1675             (slave_dev->master != bond_dev)) {
1676                 printk(KERN_ERR DRV_NAME
1677                        ": %s: Error: cannot release %s.\n",
1678                        bond_dev->name, slave_dev->name);
1679                 return -EINVAL;
1680         }
1681
1682         write_lock_bh(&bond->lock);
1683
1684         slave = bond_get_slave_by_dev(bond, slave_dev);
1685         if (!slave) {
1686                 /* not a slave of this bond */
1687                 printk(KERN_INFO DRV_NAME
1688                        ": %s: %s not enslaved\n",
1689                        bond_dev->name, slave_dev->name);
1690                 write_unlock_bh(&bond->lock);
1691                 return -EINVAL;
1692         }
1693
1694         mac_addr_differ = memcmp(bond_dev->dev_addr,
1695                                  slave->perm_hwaddr,
1696                                  ETH_ALEN);
1697         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1698                 printk(KERN_WARNING DRV_NAME
1699                        ": %s: Warning: the permanent HWaddr of %s - "
1700                        "%s - is still in use by %s. "
1701                        "Set the HWaddr of %s to a different address "
1702                        "to avoid conflicts.\n",
1703                        bond_dev->name,
1704                        slave_dev->name,
1705                        print_mac(mac, slave->perm_hwaddr),
1706                        bond_dev->name,
1707                        slave_dev->name);
1708         }
1709
1710         /* Inform AD package of unbinding of slave. */
1711         if (bond->params.mode == BOND_MODE_8023AD) {
1712                 /* must be called before the slave is
1713                  * detached from the list
1714                  */
1715                 bond_3ad_unbind_slave(slave);
1716         }
1717
1718         printk(KERN_INFO DRV_NAME
1719                ": %s: releasing %s interface %s\n",
1720                bond_dev->name,
1721                (slave->state == BOND_STATE_ACTIVE)
1722                ? "active" : "backup",
1723                slave_dev->name);
1724
1725         oldcurrent = bond->curr_active_slave;
1726
1727         bond->current_arp_slave = NULL;
1728
1729         /* release the slave from its bond */
1730         bond_detach_slave(bond, slave);
1731
1732         bond_compute_features(bond);
1733
1734         if (bond->primary_slave == slave) {
1735                 bond->primary_slave = NULL;
1736         }
1737
1738         if (oldcurrent == slave) {
1739                 bond_change_active_slave(bond, NULL);
1740         }
1741
1742         if ((bond->params.mode == BOND_MODE_TLB) ||
1743             (bond->params.mode == BOND_MODE_ALB)) {
1744                 /* Must be called only after the slave has been
1745                  * detached from the list and the curr_active_slave
1746                  * has been cleared (if our_slave == old_current),
1747                  * but before a new active slave is selected.
1748                  */
1749                 write_unlock_bh(&bond->lock);
1750                 bond_alb_deinit_slave(bond, slave);
1751                 write_lock_bh(&bond->lock);
1752         }
1753
1754         if (oldcurrent == slave) {
1755                 /*
1756                  * Note that we hold RTNL over this sequence, so there
1757                  * is no concern that another slave add/remove event
1758                  * will interfere.
1759                  */
1760                 write_unlock_bh(&bond->lock);
1761                 read_lock(&bond->lock);
1762                 write_lock_bh(&bond->curr_slave_lock);
1763
1764                 bond_select_active_slave(bond);
1765
1766                 write_unlock_bh(&bond->curr_slave_lock);
1767                 read_unlock(&bond->lock);
1768                 write_lock_bh(&bond->lock);
1769         }
1770
1771         if (bond->slave_cnt == 0) {
1772                 bond_set_carrier(bond);
1773
1774                 /* if the last slave was removed, zero the mac address
1775                  * of the master so it will be set by the application
1776                  * to the mac address of the first slave
1777                  */
1778                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1779
1780                 if (list_empty(&bond->vlan_list)) {
1781                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1782                 } else {
1783                         printk(KERN_WARNING DRV_NAME
1784                                ": %s: Warning: clearing HW address of %s while it "
1785                                "still has VLANs.\n",
1786                                bond_dev->name, bond_dev->name);
1787                         printk(KERN_WARNING DRV_NAME
1788                                ": %s: When re-adding slaves, make sure the bond's "
1789                                "HW address matches its VLANs'.\n",
1790                                bond_dev->name);
1791                 }
1792         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1793                    !bond_has_challenged_slaves(bond)) {
1794                 printk(KERN_INFO DRV_NAME
1795                        ": %s: last VLAN challenged slave %s "
1796                        "left bond %s. VLAN blocking is removed\n",
1797                        bond_dev->name, slave_dev->name, bond_dev->name);
1798                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1799         }
1800
1801         write_unlock_bh(&bond->lock);
1802
1803         /* must do this from outside any spinlocks */
1804         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1805
1806         bond_del_vlans_from_slave(bond, slave_dev);
1807
1808         /* If the mode USES_PRIMARY, then we should only remove its
1809          * promisc and mc settings if it was the curr_active_slave, but that was
1810          * already taken care of above when we detached the slave
1811          */
1812         if (!USES_PRIMARY(bond->params.mode)) {
1813                 /* unset promiscuity level from slave */
1814                 if (bond_dev->flags & IFF_PROMISC) {
1815                         dev_set_promiscuity(slave_dev, -1);
1816                 }
1817
1818                 /* unset allmulti level from slave */
1819                 if (bond_dev->flags & IFF_ALLMULTI) {
1820                         dev_set_allmulti(slave_dev, -1);
1821                 }
1822
1823                 /* flush master's mc_list from slave */
1824                 bond_mc_list_flush(bond_dev, slave_dev);
1825         }
1826
1827         netdev_set_master(slave_dev, NULL);
1828
1829         /* close slave before restoring its mac address */
1830         dev_close(slave_dev);
1831
1832         if (!bond->params.fail_over_mac) {
1833                 /* restore original ("permanent") mac address */
1834                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1835                 addr.sa_family = slave_dev->type;
1836                 dev_set_mac_address(slave_dev, &addr);
1837         }
1838
1839         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1840                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1841                                    IFF_SLAVE_NEEDARP);
1842
1843         kfree(slave);
1844
1845         return 0;  /* deletion OK */
1846 }
1847
1848 /*
1849 * Destroy a bonding device.
1850 * Must be under rtnl_lock when this function is called.
1851 */
1852 void bond_destroy(struct bonding *bond)
1853 {
1854         bond_deinit(bond->dev);
1855         bond_destroy_sysfs_entry(bond);
1856         unregister_netdevice(bond->dev);
1857 }
1858
1859 /*
1860 * First release a slave and than destroy the bond if no more slaves iare left.
1861 * Must be under rtnl_lock when this function is called.
1862 */
1863 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1864 {
1865         struct bonding *bond = bond_dev->priv;
1866         int ret;
1867
1868         ret = bond_release(bond_dev, slave_dev);
1869         if ((ret == 0) && (bond->slave_cnt == 0)) {
1870                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1871                        bond_dev->name, bond_dev->name);
1872                 bond_destroy(bond);
1873         }
1874         return ret;
1875 }
1876
1877 /*
1878  * This function releases all slaves.
1879  */
1880 static int bond_release_all(struct net_device *bond_dev)
1881 {
1882         struct bonding *bond = bond_dev->priv;
1883         struct slave *slave;
1884         struct net_device *slave_dev;
1885         struct sockaddr addr;
1886
1887         write_lock_bh(&bond->lock);
1888
1889         netif_carrier_off(bond_dev);
1890
1891         if (bond->slave_cnt == 0) {
1892                 goto out;
1893         }
1894
1895         bond->current_arp_slave = NULL;
1896         bond->primary_slave = NULL;
1897         bond_change_active_slave(bond, NULL);
1898
1899         while ((slave = bond->first_slave) != NULL) {
1900                 /* Inform AD package of unbinding of slave
1901                  * before slave is detached from the list.
1902                  */
1903                 if (bond->params.mode == BOND_MODE_8023AD) {
1904                         bond_3ad_unbind_slave(slave);
1905                 }
1906
1907                 slave_dev = slave->dev;
1908                 bond_detach_slave(bond, slave);
1909
1910                 /* now that the slave is detached, unlock and perform
1911                  * all the undo steps that should not be called from
1912                  * within a lock.
1913                  */
1914                 write_unlock_bh(&bond->lock);
1915
1916                 if ((bond->params.mode == BOND_MODE_TLB) ||
1917                     (bond->params.mode == BOND_MODE_ALB)) {
1918                         /* must be called only after the slave
1919                          * has been detached from the list
1920                          */
1921                         bond_alb_deinit_slave(bond, slave);
1922                 }
1923
1924                 bond_compute_features(bond);
1925
1926                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
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                         bond_mc_list_flush(bond_dev, slave_dev);
1946                 }
1947
1948                 netdev_set_master(slave_dev, NULL);
1949
1950                 /* close slave before restoring its mac address */
1951                 dev_close(slave_dev);
1952
1953                 if (!bond->params.fail_over_mac) {
1954                         /* restore original ("permanent") mac address*/
1955                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1956                         addr.sa_family = slave_dev->type;
1957                         dev_set_mac_address(slave_dev, &addr);
1958                 }
1959
1960                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1961                                            IFF_SLAVE_INACTIVE);
1962
1963                 kfree(slave);
1964
1965                 /* re-acquire the lock before getting the next slave */
1966                 write_lock_bh(&bond->lock);
1967         }
1968
1969         /* zero the mac address of the master so it will be
1970          * set by the application to the mac address of the
1971          * first slave
1972          */
1973         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1974
1975         if (list_empty(&bond->vlan_list)) {
1976                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1977         } else {
1978                 printk(KERN_WARNING DRV_NAME
1979                        ": %s: Warning: clearing HW address of %s while it "
1980                        "still has VLANs.\n",
1981                        bond_dev->name, bond_dev->name);
1982                 printk(KERN_WARNING DRV_NAME
1983                        ": %s: When re-adding slaves, make sure the bond's "
1984                        "HW address matches its VLANs'.\n",
1985                        bond_dev->name);
1986         }
1987
1988         printk(KERN_INFO DRV_NAME
1989                ": %s: released all slaves\n",
1990                bond_dev->name);
1991
1992 out:
1993         write_unlock_bh(&bond->lock);
1994
1995         return 0;
1996 }
1997
1998 /*
1999  * This function changes the active slave to slave <slave_dev>.
2000  * It returns -EINVAL in the following cases.
2001  *  - <slave_dev> is not found in the list.
2002  *  - There is not active slave now.
2003  *  - <slave_dev> is already active.
2004  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2005  *  - <slave_dev> is not running.
2006  * In these cases, this fuction does nothing.
2007  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2008  */
2009 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2010 {
2011         struct bonding *bond = bond_dev->priv;
2012         struct slave *old_active = NULL;
2013         struct slave *new_active = NULL;
2014         int res = 0;
2015
2016         if (!USES_PRIMARY(bond->params.mode)) {
2017                 return -EINVAL;
2018         }
2019
2020         /* Verify that master_dev is indeed the master of slave_dev */
2021         if (!(slave_dev->flags & IFF_SLAVE) ||
2022             (slave_dev->master != bond_dev)) {
2023                 return -EINVAL;
2024         }
2025
2026         read_lock(&bond->lock);
2027
2028         read_lock(&bond->curr_slave_lock);
2029         old_active = bond->curr_active_slave;
2030         read_unlock(&bond->curr_slave_lock);
2031
2032         new_active = bond_get_slave_by_dev(bond, slave_dev);
2033
2034         /*
2035          * Changing to the current active: do nothing; return success.
2036          */
2037         if (new_active && (new_active == old_active)) {
2038                 read_unlock(&bond->lock);
2039                 return 0;
2040         }
2041
2042         if ((new_active) &&
2043             (old_active) &&
2044             (new_active->link == BOND_LINK_UP) &&
2045             IS_UP(new_active->dev)) {
2046                 write_lock_bh(&bond->curr_slave_lock);
2047                 bond_change_active_slave(bond, new_active);
2048                 write_unlock_bh(&bond->curr_slave_lock);
2049         } else {
2050                 res = -EINVAL;
2051         }
2052
2053         read_unlock(&bond->lock);
2054
2055         return res;
2056 }
2057
2058 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2059 {
2060         struct bonding *bond = bond_dev->priv;
2061
2062         info->bond_mode = bond->params.mode;
2063         info->miimon = bond->params.miimon;
2064
2065         read_lock(&bond->lock);
2066         info->num_slaves = bond->slave_cnt;
2067         read_unlock(&bond->lock);
2068
2069         return 0;
2070 }
2071
2072 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2073 {
2074         struct bonding *bond = bond_dev->priv;
2075         struct slave *slave;
2076         int i, found = 0;
2077
2078         if (info->slave_id < 0) {
2079                 return -ENODEV;
2080         }
2081
2082         read_lock(&bond->lock);
2083
2084         bond_for_each_slave(bond, slave, i) {
2085                 if (i == (int)info->slave_id) {
2086                         found = 1;
2087                         break;
2088                 }
2089         }
2090
2091         read_unlock(&bond->lock);
2092
2093         if (found) {
2094                 strcpy(info->slave_name, slave->dev->name);
2095                 info->link = slave->link;
2096                 info->state = slave->state;
2097                 info->link_failure_count = slave->link_failure_count;
2098         } else {
2099                 return -ENODEV;
2100         }
2101
2102         return 0;
2103 }
2104
2105 /*-------------------------------- Monitoring -------------------------------*/
2106
2107 /*
2108  * if !have_locks, return nonzero if a failover is necessary.  if
2109  * have_locks, do whatever failover activities are needed.
2110  *
2111  * This is to separate the inspection and failover steps for locking
2112  * purposes; failover requires rtnl, but acquiring it for every
2113  * inspection is undesirable, so a wrapper first does inspection, and
2114  * the acquires the necessary locks and calls again to perform
2115  * failover if needed.  Since all locks are dropped, a complete
2116  * restart is needed between calls.
2117  */
2118 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2119 {
2120         struct slave *slave, *oldcurrent;
2121         int do_failover = 0;
2122         int i;
2123
2124         if (bond->slave_cnt == 0)
2125                 goto out;
2126
2127         /* we will try to read the link status of each of our slaves, and
2128          * set their IFF_RUNNING flag appropriately. For each slave not
2129          * supporting MII status, we won't do anything so that a user-space
2130          * program could monitor the link itself if needed.
2131          */
2132
2133         if (bond->send_grat_arp) {
2134                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2135                                 &bond->curr_active_slave->dev->state))
2136                         dprintk("Needs to send gratuitous arp but not yet\n");
2137                 else {
2138                         dprintk("sending delayed gratuitous arp on on %s\n",
2139                                 bond->curr_active_slave->dev->name);
2140                         bond_send_gratuitous_arp(bond);
2141                         bond->send_grat_arp = 0;
2142                 }
2143         }
2144         read_lock(&bond->curr_slave_lock);
2145         oldcurrent = bond->curr_active_slave;
2146         read_unlock(&bond->curr_slave_lock);
2147
2148         bond_for_each_slave(bond, slave, i) {
2149                 struct net_device *slave_dev = slave->dev;
2150                 int link_state;
2151                 u16 old_speed = slave->speed;
2152                 u8 old_duplex = slave->duplex;
2153
2154                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2155
2156                 switch (slave->link) {
2157                 case BOND_LINK_UP:      /* the link was up */
2158                         if (link_state == BMSR_LSTATUS) {
2159                                 if (!oldcurrent) {
2160                                         if (!have_locks)
2161                                                 return 1;
2162                                         do_failover = 1;
2163                                 }
2164                                 break;
2165                         } else { /* link going down */
2166                                 slave->link  = BOND_LINK_FAIL;
2167                                 slave->delay = bond->params.downdelay;
2168
2169                                 if (slave->link_failure_count < UINT_MAX) {
2170                                         slave->link_failure_count++;
2171                                 }
2172
2173                                 if (bond->params.downdelay) {
2174                                         printk(KERN_INFO DRV_NAME
2175                                                ": %s: link status down for %s "
2176                                                "interface %s, disabling it in "
2177                                                "%d ms.\n",
2178                                                bond->dev->name,
2179                                                IS_UP(slave_dev)
2180                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2181                                                   ? ((slave == oldcurrent)
2182                                                      ? "active " : "backup ")
2183                                                   : "")
2184                                                : "idle ",
2185                                                slave_dev->name,
2186                                                bond->params.downdelay * bond->params.miimon);
2187                                 }
2188                         }
2189                         /* no break ! fall through the BOND_LINK_FAIL test to
2190                            ensure proper action to be taken
2191                         */
2192                 case BOND_LINK_FAIL:    /* the link has just gone down */
2193                         if (link_state != BMSR_LSTATUS) {
2194                                 /* link stays down */
2195                                 if (slave->delay <= 0) {
2196                                         if (!have_locks)
2197                                                 return 1;
2198
2199                                         /* link down for too long time */
2200                                         slave->link = BOND_LINK_DOWN;
2201
2202                                         /* in active/backup mode, we must
2203                                          * completely disable this interface
2204                                          */
2205                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2206                                             (bond->params.mode == BOND_MODE_8023AD)) {
2207                                                 bond_set_slave_inactive_flags(slave);
2208                                         }
2209
2210                                         printk(KERN_INFO DRV_NAME
2211                                                ": %s: link status definitely "
2212                                                "down for interface %s, "
2213                                                "disabling it\n",
2214                                                bond->dev->name,
2215                                                slave_dev->name);
2216
2217                                         /* notify ad that the link status has changed */
2218                                         if (bond->params.mode == BOND_MODE_8023AD) {
2219                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2220                                         }
2221
2222                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2223                                             (bond->params.mode == BOND_MODE_ALB)) {
2224                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2225                                         }
2226
2227                                         if (slave == oldcurrent) {
2228                                                 do_failover = 1;
2229                                         }
2230                                 } else {
2231                                         slave->delay--;
2232                                 }
2233                         } else {
2234                                 /* link up again */
2235                                 slave->link  = BOND_LINK_UP;
2236                                 slave->jiffies = jiffies;
2237                                 printk(KERN_INFO DRV_NAME
2238                                        ": %s: link status up again after %d "
2239                                        "ms for interface %s.\n",
2240                                        bond->dev->name,
2241                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2242                                        slave_dev->name);
2243                         }
2244                         break;
2245                 case BOND_LINK_DOWN:    /* the link was down */
2246                         if (link_state != BMSR_LSTATUS) {
2247                                 /* the link stays down, nothing more to do */
2248                                 break;
2249                         } else {        /* link going up */
2250                                 slave->link  = BOND_LINK_BACK;
2251                                 slave->delay = bond->params.updelay;
2252
2253                                 if (bond->params.updelay) {
2254                                         /* if updelay == 0, no need to
2255                                            advertise about a 0 ms delay */
2256                                         printk(KERN_INFO DRV_NAME
2257                                                ": %s: link status up for "
2258                                                "interface %s, enabling it "
2259                                                "in %d ms.\n",
2260                                                bond->dev->name,
2261                                                slave_dev->name,
2262                                                bond->params.updelay * bond->params.miimon);
2263                                 }
2264                         }
2265                         /* no break ! fall through the BOND_LINK_BACK state in
2266                            case there's something to do.
2267                         */
2268                 case BOND_LINK_BACK:    /* the link has just come back */
2269                         if (link_state != BMSR_LSTATUS) {
2270                                 /* link down again */
2271                                 slave->link  = BOND_LINK_DOWN;
2272
2273                                 printk(KERN_INFO DRV_NAME
2274                                        ": %s: link status down again after %d "
2275                                        "ms for interface %s.\n",
2276                                        bond->dev->name,
2277                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2278                                        slave_dev->name);
2279                         } else {
2280                                 /* link stays up */
2281                                 if (slave->delay == 0) {
2282                                         if (!have_locks)
2283                                                 return 1;
2284
2285                                         /* now the link has been up for long time enough */
2286                                         slave->link = BOND_LINK_UP;
2287                                         slave->jiffies = jiffies;
2288
2289                                         if (bond->params.mode == BOND_MODE_8023AD) {
2290                                                 /* prevent it from being the active one */
2291                                                 slave->state = BOND_STATE_BACKUP;
2292                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2293                                                 /* make it immediately active */
2294                                                 slave->state = BOND_STATE_ACTIVE;
2295                                         } else if (slave != bond->primary_slave) {
2296                                                 /* prevent it from being the active one */
2297                                                 slave->state = BOND_STATE_BACKUP;
2298                                         }
2299
2300                                         printk(KERN_INFO DRV_NAME
2301                                                ": %s: link status definitely "
2302                                                "up for interface %s.\n",
2303                                                bond->dev->name,
2304                                                slave_dev->name);
2305
2306                                         /* notify ad that the link status has changed */
2307                                         if (bond->params.mode == BOND_MODE_8023AD) {
2308                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2309                                         }
2310
2311                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2312                                             (bond->params.mode == BOND_MODE_ALB)) {
2313                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2314                                         }
2315
2316                                         if ((!oldcurrent) ||
2317                                             (slave == bond->primary_slave)) {
2318                                                 do_failover = 1;
2319                                         }
2320                                 } else {
2321                                         slave->delay--;
2322                                 }
2323                         }
2324                         break;
2325                 default:
2326                         /* Should not happen */
2327                         printk(KERN_ERR DRV_NAME
2328                                ": %s: Error: %s Illegal value (link=%d)\n",
2329                                bond->dev->name,
2330                                slave->dev->name,
2331                                slave->link);
2332                         goto out;
2333                 } /* end of switch (slave->link) */
2334
2335                 bond_update_speed_duplex(slave);
2336
2337                 if (bond->params.mode == BOND_MODE_8023AD) {
2338                         if (old_speed != slave->speed) {
2339                                 bond_3ad_adapter_speed_changed(slave);
2340                         }
2341
2342                         if (old_duplex != slave->duplex) {
2343                                 bond_3ad_adapter_duplex_changed(slave);
2344                         }
2345                 }
2346
2347         } /* end of for */
2348
2349         if (do_failover) {
2350                 ASSERT_RTNL();
2351
2352                 write_lock_bh(&bond->curr_slave_lock);
2353
2354                 bond_select_active_slave(bond);
2355
2356                 write_unlock_bh(&bond->curr_slave_lock);
2357
2358         } else
2359                 bond_set_carrier(bond);
2360
2361 out:
2362         return 0;
2363 }
2364
2365 /*
2366  * bond_mii_monitor
2367  *
2368  * Really a wrapper that splits the mii monitor into two phases: an
2369  * inspection, then (if inspection indicates something needs to be
2370  * done) an acquisition of appropriate locks followed by another pass
2371  * to implement whatever link state changes are indicated.
2372  */
2373 void bond_mii_monitor(struct work_struct *work)
2374 {
2375         struct bonding *bond = container_of(work, struct bonding,
2376                                             mii_work.work);
2377         unsigned long delay;
2378
2379         read_lock(&bond->lock);
2380         if (bond->kill_timers) {
2381                 read_unlock(&bond->lock);
2382                 return;
2383         }
2384         if (__bond_mii_monitor(bond, 0)) {
2385                 read_unlock(&bond->lock);
2386                 rtnl_lock();
2387                 read_lock(&bond->lock);
2388                 __bond_mii_monitor(bond, 1);
2389                 read_unlock(&bond->lock);
2390                 rtnl_unlock();  /* might sleep, hold no other locks */
2391                 read_lock(&bond->lock);
2392         }
2393
2394         delay = ((bond->params.miimon * HZ) / 1000) ? : 1;
2395         read_unlock(&bond->lock);
2396         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2397 }
2398
2399 static __be32 bond_glean_dev_ip(struct net_device *dev)
2400 {
2401         struct in_device *idev;
2402         struct in_ifaddr *ifa;
2403         __be32 addr = 0;
2404
2405         if (!dev)
2406                 return 0;
2407
2408         rcu_read_lock();
2409         idev = __in_dev_get_rcu(dev);
2410         if (!idev)
2411                 goto out;
2412
2413         ifa = idev->ifa_list;
2414         if (!ifa)
2415                 goto out;
2416
2417         addr = ifa->ifa_local;
2418 out:
2419         rcu_read_unlock();
2420         return addr;
2421 }
2422
2423 static int bond_has_ip(struct bonding *bond)
2424 {
2425         struct vlan_entry *vlan, *vlan_next;
2426
2427         if (bond->master_ip)
2428                 return 1;
2429
2430         if (list_empty(&bond->vlan_list))
2431                 return 0;
2432
2433         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2434                                  vlan_list) {
2435                 if (vlan->vlan_ip)
2436                         return 1;
2437         }
2438
2439         return 0;
2440 }
2441
2442 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2443 {
2444         struct vlan_entry *vlan, *vlan_next;
2445
2446         if (ip == bond->master_ip)
2447                 return 1;
2448
2449         if (list_empty(&bond->vlan_list))
2450                 return 0;
2451
2452         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2453                                  vlan_list) {
2454                 if (ip == vlan->vlan_ip)
2455                         return 1;
2456         }
2457
2458         return 0;
2459 }
2460
2461 /*
2462  * We go to the (large) trouble of VLAN tagging ARP frames because
2463  * switches in VLAN mode (especially if ports are configured as
2464  * "native" to a VLAN) might not pass non-tagged frames.
2465  */
2466 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2467 {
2468         struct sk_buff *skb;
2469
2470         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2471                slave_dev->name, dest_ip, src_ip, vlan_id);
2472
2473         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2474                          NULL, slave_dev->dev_addr, NULL);
2475
2476         if (!skb) {
2477                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2478                 return;
2479         }
2480         if (vlan_id) {
2481                 skb = vlan_put_tag(skb, vlan_id);
2482                 if (!skb) {
2483                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2484                         return;
2485                 }
2486         }
2487         arp_xmit(skb);
2488 }
2489
2490
2491 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2492 {
2493         int i, vlan_id, rv;
2494         __be32 *targets = bond->params.arp_targets;
2495         struct vlan_entry *vlan, *vlan_next;
2496         struct net_device *vlan_dev;
2497         struct flowi fl;
2498         struct rtable *rt;
2499
2500         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2501                 if (!targets[i])
2502                         continue;
2503                 dprintk("basa: target %x\n", targets[i]);
2504                 if (list_empty(&bond->vlan_list)) {
2505                         dprintk("basa: empty vlan: arp_send\n");
2506                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2507                                       bond->master_ip, 0);
2508                         continue;
2509                 }
2510
2511                 /*
2512                  * If VLANs are configured, we do a route lookup to
2513                  * determine which VLAN interface would be used, so we
2514                  * can tag the ARP with the proper VLAN tag.
2515                  */
2516                 memset(&fl, 0, sizeof(fl));
2517                 fl.fl4_dst = targets[i];
2518                 fl.fl4_tos = RTO_ONLINK;
2519
2520                 rv = ip_route_output_key(&init_net, &rt, &fl);
2521                 if (rv) {
2522                         if (net_ratelimit()) {
2523                                 printk(KERN_WARNING DRV_NAME
2524                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2525                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2526                         }
2527                         continue;
2528                 }
2529
2530                 /*
2531                  * This target is not on a VLAN
2532                  */
2533                 if (rt->u.dst.dev == bond->dev) {
2534                         ip_rt_put(rt);
2535                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2536                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2537                                       bond->master_ip, 0);
2538                         continue;
2539                 }
2540
2541                 vlan_id = 0;
2542                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2543                                          vlan_list) {
2544                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2545                         if (vlan_dev == rt->u.dst.dev) {
2546                                 vlan_id = vlan->vlan_id;
2547                                 dprintk("basa: vlan match on %s %d\n",
2548                                        vlan_dev->name, vlan_id);
2549                                 break;
2550                         }
2551                 }
2552
2553                 if (vlan_id) {
2554                         ip_rt_put(rt);
2555                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2556                                       vlan->vlan_ip, vlan_id);
2557                         continue;
2558                 }
2559
2560                 if (net_ratelimit()) {
2561                         printk(KERN_WARNING DRV_NAME
2562                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2563                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2564                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2565                 }
2566                 ip_rt_put(rt);
2567         }
2568 }
2569
2570 /*
2571  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2572  * for each VLAN above us.
2573  */
2574 static void bond_send_gratuitous_arp(struct bonding *bond)
2575 {
2576         struct slave *slave = bond->curr_active_slave;
2577         struct vlan_entry *vlan;
2578         struct net_device *vlan_dev;
2579
2580         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2581                                 slave ? slave->dev->name : "NULL");
2582         if (!slave)
2583                 return;
2584
2585         if (bond->master_ip) {
2586                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2587                                 bond->master_ip, 0);
2588         }
2589
2590         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2591                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2592                 if (vlan->vlan_ip) {
2593                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2594                                       vlan->vlan_ip, vlan->vlan_id);
2595                 }
2596         }
2597 }
2598
2599 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2600 {
2601         int i;
2602         __be32 *targets = bond->params.arp_targets;
2603
2604         targets = bond->params.arp_targets;
2605         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2606                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2607                         "%u.%u.%u.%u bhti(tip) %d\n",
2608                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2609                        bond_has_this_ip(bond, tip));
2610                 if (sip == targets[i]) {
2611                         if (bond_has_this_ip(bond, tip))
2612                                 slave->last_arp_rx = jiffies;
2613                         return;
2614                 }
2615         }
2616 }
2617
2618 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2619 {
2620         struct arphdr *arp;
2621         struct slave *slave;
2622         struct bonding *bond;
2623         unsigned char *arp_ptr;
2624         __be32 sip, tip;
2625
2626         if (dev->nd_net != &init_net)
2627                 goto out;
2628
2629         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2630                 goto out;
2631
2632         bond = dev->priv;
2633         read_lock(&bond->lock);
2634
2635         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2636                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2637                 orig_dev ? orig_dev->name : "NULL");
2638
2639         slave = bond_get_slave_by_dev(bond, orig_dev);
2640         if (!slave || !slave_do_arp_validate(bond, slave))
2641                 goto out_unlock;
2642
2643         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2644         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2645                                  (2 * dev->addr_len) +
2646                                  (2 * sizeof(u32)))))
2647                 goto out_unlock;
2648
2649         arp = arp_hdr(skb);
2650         if (arp->ar_hln != dev->addr_len ||
2651             skb->pkt_type == PACKET_OTHERHOST ||
2652             skb->pkt_type == PACKET_LOOPBACK ||
2653             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2654             arp->ar_pro != htons(ETH_P_IP) ||
2655             arp->ar_pln != 4)
2656                 goto out_unlock;
2657
2658         arp_ptr = (unsigned char *)(arp + 1);
2659         arp_ptr += dev->addr_len;
2660         memcpy(&sip, arp_ptr, 4);
2661         arp_ptr += 4 + dev->addr_len;
2662         memcpy(&tip, arp_ptr, 4);
2663
2664         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2665                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2666                 slave->state, bond->params.arp_validate,
2667                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2668
2669         /*
2670          * Backup slaves won't see the ARP reply, but do come through
2671          * here for each ARP probe (so we swap the sip/tip to validate
2672          * the probe).  In a "redundant switch, common router" type of
2673          * configuration, the ARP probe will (hopefully) travel from
2674          * the active, through one switch, the router, then the other
2675          * switch before reaching the backup.
2676          */
2677         if (slave->state == BOND_STATE_ACTIVE)
2678                 bond_validate_arp(bond, slave, sip, tip);
2679         else
2680                 bond_validate_arp(bond, slave, tip, sip);
2681
2682 out_unlock:
2683         read_unlock(&bond->lock);
2684 out:
2685         dev_kfree_skb(skb);
2686         return NET_RX_SUCCESS;
2687 }
2688
2689 /*
2690  * this function is called regularly to monitor each slave's link
2691  * ensuring that traffic is being sent and received when arp monitoring
2692  * is used in load-balancing mode. if the adapter has been dormant, then an
2693  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2694  * arp monitoring in active backup mode.
2695  */
2696 void bond_loadbalance_arp_mon(struct work_struct *work)
2697 {
2698         struct bonding *bond = container_of(work, struct bonding,
2699                                             arp_work.work);
2700         struct slave *slave, *oldcurrent;
2701         int do_failover = 0;
2702         int delta_in_ticks;
2703         int i;
2704
2705         read_lock(&bond->lock);
2706
2707         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2708
2709         if (bond->kill_timers) {
2710                 goto out;
2711         }
2712
2713         if (bond->slave_cnt == 0) {
2714                 goto re_arm;
2715         }
2716
2717         read_lock(&bond->curr_slave_lock);
2718         oldcurrent = bond->curr_active_slave;
2719         read_unlock(&bond->curr_slave_lock);
2720
2721         /* see if any of the previous devices are up now (i.e. they have
2722          * xmt and rcv traffic). the curr_active_slave does not come into
2723          * the picture unless it is null. also, slave->jiffies is not needed
2724          * here because we send an arp on each slave and give a slave as
2725          * long as it needs to get the tx/rx within the delta.
2726          * TODO: what about up/down delay in arp mode? it wasn't here before
2727          *       so it can wait
2728          */
2729         bond_for_each_slave(bond, slave, i) {
2730                 if (slave->link != BOND_LINK_UP) {
2731                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2732                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2733
2734                                 slave->link  = BOND_LINK_UP;
2735                                 slave->state = BOND_STATE_ACTIVE;
2736
2737                                 /* primary_slave has no meaning in round-robin
2738                                  * mode. the window of a slave being up and
2739                                  * curr_active_slave being null after enslaving
2740                                  * is closed.
2741                                  */
2742                                 if (!oldcurrent) {
2743                                         printk(KERN_INFO DRV_NAME
2744                                                ": %s: link status definitely "
2745                                                "up for interface %s, ",
2746                                                bond->dev->name,
2747                                                slave->dev->name);
2748                                         do_failover = 1;
2749                                 } else {
2750                                         printk(KERN_INFO DRV_NAME
2751                                                ": %s: interface %s is now up\n",
2752                                                bond->dev->name,
2753                                                slave->dev->name);
2754                                 }
2755                         }
2756                 } else {
2757                         /* slave->link == BOND_LINK_UP */
2758
2759                         /* not all switches will respond to an arp request
2760                          * when the source ip is 0, so don't take the link down
2761                          * if we don't know our ip yet
2762                          */
2763                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2764                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks) &&
2765                              bond_has_ip(bond))) {
2766
2767                                 slave->link  = BOND_LINK_DOWN;
2768                                 slave->state = BOND_STATE_BACKUP;
2769
2770                                 if (slave->link_failure_count < UINT_MAX) {
2771                                         slave->link_failure_count++;
2772                                 }
2773
2774                                 printk(KERN_INFO DRV_NAME
2775                                        ": %s: interface %s is now down.\n",
2776                                        bond->dev->name,
2777                                        slave->dev->name);
2778
2779                                 if (slave == oldcurrent) {
2780                                         do_failover = 1;
2781                                 }
2782                         }
2783                 }
2784
2785                 /* note: if switch is in round-robin mode, all links
2786                  * must tx arp to ensure all links rx an arp - otherwise
2787                  * links may oscillate or not come up at all; if switch is
2788                  * in something like xor mode, there is nothing we can
2789                  * do - all replies will be rx'ed on same link causing slaves
2790                  * to be unstable during low/no traffic periods
2791                  */
2792                 if (IS_UP(slave->dev)) {
2793                         bond_arp_send_all(bond, slave);
2794                 }
2795         }
2796
2797         if (do_failover) {
2798                 rtnl_lock();
2799                 write_lock_bh(&bond->curr_slave_lock);
2800
2801                 bond_select_active_slave(bond);
2802
2803                 write_unlock_bh(&bond->curr_slave_lock);
2804                 rtnl_unlock();
2805
2806         }
2807
2808 re_arm:
2809         if (bond->params.arp_interval)
2810                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2811 out:
2812         read_unlock(&bond->lock);
2813 }
2814
2815 /*
2816  * When using arp monitoring in active-backup mode, this function is
2817  * called to determine if any backup slaves have went down or a new
2818  * current slave needs to be found.
2819  * The backup slaves never generate traffic, they are considered up by merely
2820  * receiving traffic. If the current slave goes down, each backup slave will
2821  * be given the opportunity to tx/rx an arp before being taken down - this
2822  * prevents all slaves from being taken down due to the current slave not
2823  * sending any traffic for the backups to receive. The arps are not necessarily
2824  * necessary, any tx and rx traffic will keep the current slave up. While any
2825  * rx traffic will keep the backup slaves up, the current slave is responsible
2826  * for generating traffic to keep them up regardless of any other traffic they
2827  * may have received.
2828  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2829  */
2830 void bond_activebackup_arp_mon(struct work_struct *work)
2831 {
2832         struct bonding *bond = container_of(work, struct bonding,
2833                                             arp_work.work);
2834         struct slave *slave;
2835         int delta_in_ticks;
2836         int i;
2837
2838         read_lock(&bond->lock);
2839
2840         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2841
2842         if (bond->kill_timers) {
2843                 goto out;
2844         }
2845
2846         if (bond->slave_cnt == 0) {
2847                 goto re_arm;
2848         }
2849
2850         /* determine if any slave has come up or any backup slave has
2851          * gone down
2852          * TODO: what about up/down delay in arp mode? it wasn't here before
2853          *       so it can wait
2854          */
2855         bond_for_each_slave(bond, slave, i) {
2856                 if (slave->link != BOND_LINK_UP) {
2857                         if (time_before_eq(jiffies,
2858                             slave_last_rx(bond, slave) + delta_in_ticks)) {
2859
2860                                 slave->link = BOND_LINK_UP;
2861
2862                                 rtnl_lock();
2863
2864                                 write_lock_bh(&bond->curr_slave_lock);
2865
2866                                 if ((!bond->curr_active_slave) &&
2867                                     time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks)) {
2868                                         bond_change_active_slave(bond, slave);
2869                                         bond->current_arp_slave = NULL;
2870                                 } else if (bond->curr_active_slave != slave) {
2871                                         /* this slave has just come up but we
2872                                          * already have a current slave; this
2873                                          * can also happen if bond_enslave adds
2874                                          * a new slave that is up while we are
2875                                          * searching for a new slave
2876                                          */
2877                                         bond_set_slave_inactive_flags(slave);
2878                                         bond->current_arp_slave = NULL;
2879                                 }
2880
2881                                 bond_set_carrier(bond);
2882
2883                                 if (slave == bond->curr_active_slave) {
2884                                         printk(KERN_INFO DRV_NAME
2885                                                ": %s: %s is up and now the "
2886                                                "active interface\n",
2887                                                bond->dev->name,
2888                                                slave->dev->name);
2889                                         netif_carrier_on(bond->dev);
2890                                 } else {
2891                                         printk(KERN_INFO DRV_NAME
2892                                                ": %s: backup interface %s is "
2893                                                "now up\n",
2894                                                bond->dev->name,
2895                                                slave->dev->name);
2896                                 }
2897
2898                                 write_unlock_bh(&bond->curr_slave_lock);
2899                                 rtnl_unlock();
2900                         }
2901                 } else {
2902                         read_lock(&bond->curr_slave_lock);
2903
2904                         if ((slave != bond->curr_active_slave) &&
2905                             (!bond->current_arp_slave) &&
2906                             (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks) &&
2907                              bond_has_ip(bond))) {
2908                                 /* a backup slave has gone down; three times
2909                                  * the delta allows the current slave to be
2910                                  * taken out before the backup slave.
2911                                  * note: a non-null current_arp_slave indicates
2912                                  * the curr_active_slave went down and we are
2913                                  * searching for a new one; under this
2914                                  * condition we only take the curr_active_slave
2915                                  * down - this gives each slave a chance to
2916                                  * tx/rx traffic before being taken out
2917                                  */
2918
2919                                 read_unlock(&bond->curr_slave_lock);
2920
2921                                 slave->link  = BOND_LINK_DOWN;
2922
2923                                 if (slave->link_failure_count < UINT_MAX) {
2924                                         slave->link_failure_count++;
2925                                 }
2926
2927                                 bond_set_slave_inactive_flags(slave);
2928
2929                                 printk(KERN_INFO DRV_NAME
2930                                        ": %s: backup interface %s is now down\n",
2931                                        bond->dev->name,
2932                                        slave->dev->name);
2933                         } else {
2934                                 read_unlock(&bond->curr_slave_lock);
2935                         }
2936                 }
2937         }
2938
2939         read_lock(&bond->curr_slave_lock);
2940         slave = bond->curr_active_slave;
2941         read_unlock(&bond->curr_slave_lock);
2942
2943         if (slave) {
2944                 /* if we have sent traffic in the past 2*arp_intervals but
2945                  * haven't xmit and rx traffic in that time interval, select
2946                  * a different slave. slave->jiffies is only updated when
2947                  * a slave first becomes the curr_active_slave - not necessarily
2948                  * after every arp; this ensures the slave has a full 2*delta
2949                  * before being taken out. if a primary is being used, check
2950                  * if it is up and needs to take over as the curr_active_slave
2951                  */
2952                 if ((time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2953                         (time_after_eq(jiffies, slave_last_rx(bond, slave) + 2*delta_in_ticks) &&
2954                          bond_has_ip(bond))) &&
2955                         time_after_eq(jiffies, slave->jiffies + 2*delta_in_ticks)) {
2956
2957                         slave->link  = BOND_LINK_DOWN;
2958
2959                         if (slave->link_failure_count < UINT_MAX) {
2960                                 slave->link_failure_count++;
2961                         }
2962
2963                         printk(KERN_INFO DRV_NAME
2964                                ": %s: link status down for active interface "
2965                                "%s, disabling it\n",
2966                                bond->dev->name,
2967                                slave->dev->name);
2968
2969                         rtnl_lock();
2970                         write_lock_bh(&bond->curr_slave_lock);
2971
2972                         bond_select_active_slave(bond);
2973                         slave = bond->curr_active_slave;
2974
2975                         write_unlock_bh(&bond->curr_slave_lock);
2976
2977                         rtnl_unlock();
2978
2979                         bond->current_arp_slave = slave;
2980
2981                         if (slave) {
2982                                 slave->jiffies = jiffies;
2983                         }
2984                 } else if ((bond->primary_slave) &&
2985                            (bond->primary_slave != slave) &&
2986                            (bond->primary_slave->link == BOND_LINK_UP)) {
2987                         /* at this point, slave is the curr_active_slave */
2988                         printk(KERN_INFO DRV_NAME
2989                                ": %s: changing from interface %s to primary "
2990                                "interface %s\n",
2991                                bond->dev->name,
2992                                slave->dev->name,
2993                                bond->primary_slave->dev->name);
2994
2995                         /* primary is up so switch to it */
2996                         rtnl_lock();
2997                         write_lock_bh(&bond->curr_slave_lock);
2998                         bond_change_active_slave(bond, bond->primary_slave);
2999                         write_unlock_bh(&bond->curr_slave_lock);
3000
3001                         rtnl_unlock();
3002
3003                         slave = bond->primary_slave;
3004                         slave->jiffies = jiffies;
3005                 } else {
3006                         bond->current_arp_slave = NULL;
3007                 }
3008
3009                 /* the current slave must tx an arp to ensure backup slaves
3010                  * rx traffic
3011                  */
3012                 if (slave && bond_has_ip(bond)) {
3013                         bond_arp_send_all(bond, slave);
3014                 }
3015         }
3016
3017         /* if we don't have a curr_active_slave, search for the next available
3018          * backup slave from the current_arp_slave and make it the candidate
3019          * for becoming the curr_active_slave
3020          */
3021         if (!slave) {
3022                 if (!bond->current_arp_slave) {
3023                         bond->current_arp_slave = bond->first_slave;
3024                 }
3025
3026                 if (bond->current_arp_slave) {
3027                         bond_set_slave_inactive_flags(bond->current_arp_slave);
3028
3029                         /* search for next candidate */
3030                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3031                                 if (IS_UP(slave->dev)) {
3032                                         slave->link = BOND_LINK_BACK;
3033                                         bond_set_slave_active_flags(slave);
3034                                         bond_arp_send_all(bond, slave);
3035                                         slave->jiffies = jiffies;
3036                                         bond->current_arp_slave = slave;
3037                                         break;
3038                                 }
3039
3040                                 /* if the link state is up at this point, we
3041                                  * mark it down - this can happen if we have
3042                                  * simultaneous link failures and
3043                                  * reselect_active_interface doesn't make this
3044                                  * one the current slave so it is still marked
3045                                  * up when it is actually down
3046                                  */
3047                                 if (slave->link == BOND_LINK_UP) {
3048                                         slave->link  = BOND_LINK_DOWN;
3049                                         if (slave->link_failure_count < UINT_MAX) {
3050                                                 slave->link_failure_count++;
3051                                         }
3052
3053                                         bond_set_slave_inactive_flags(slave);
3054
3055                                         printk(KERN_INFO DRV_NAME
3056                                                ": %s: backup interface %s is "
3057                                                "now down.\n",
3058                                                bond->dev->name,
3059                                                slave->dev->name);
3060                                 }
3061                         }
3062                 }
3063         }
3064
3065 re_arm:
3066         if (bond->params.arp_interval) {
3067                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3068         }
3069 out:
3070         read_unlock(&bond->lock);
3071 }
3072
3073 /*------------------------------ proc/seq_file-------------------------------*/
3074
3075 #ifdef CONFIG_PROC_FS
3076
3077 #define SEQ_START_TOKEN ((void *)1)
3078
3079 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3080 {
3081         struct bonding *bond = seq->private;
3082         loff_t off = 0;
3083         struct slave *slave;
3084         int i;
3085
3086         /* make sure the bond won't be taken away */
3087         read_lock(&dev_base_lock);
3088         read_lock(&bond->lock);
3089
3090         if (*pos == 0) {
3091                 return SEQ_START_TOKEN;
3092         }
3093
3094         bond_for_each_slave(bond, slave, i) {
3095                 if (++off == *pos) {
3096                         return slave;
3097                 }
3098         }
3099
3100         return NULL;
3101 }
3102
3103 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3104 {
3105         struct bonding *bond = seq->private;
3106         struct slave *slave = v;
3107
3108         ++*pos;
3109         if (v == SEQ_START_TOKEN) {
3110                 return bond->first_slave;
3111         }
3112
3113         slave = slave->next;
3114
3115         return (slave == bond->first_slave) ? NULL : slave;
3116 }
3117
3118 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3119 {
3120         struct bonding *bond = seq->private;
3121
3122         read_unlock(&bond->lock);
3123         read_unlock(&dev_base_lock);
3124 }
3125
3126 static void bond_info_show_master(struct seq_file *seq)
3127 {
3128         struct bonding *bond = seq->private;
3129         struct slave *curr;
3130         int i;
3131         u32 target;
3132
3133         read_lock(&bond->curr_slave_lock);
3134         curr = bond->curr_active_slave;
3135         read_unlock(&bond->curr_slave_lock);
3136
3137         seq_printf(seq, "Bonding Mode: %s",
3138                    bond_mode_name(bond->params.mode));
3139
3140         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3141             bond->params.fail_over_mac)
3142                 seq_printf(seq, " (fail_over_mac)");
3143
3144         seq_printf(seq, "\n");
3145
3146         if (bond->params.mode == BOND_MODE_XOR ||
3147                 bond->params.mode == BOND_MODE_8023AD) {
3148                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3149                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3150                         bond->params.xmit_policy);
3151         }
3152
3153         if (USES_PRIMARY(bond->params.mode)) {
3154                 seq_printf(seq, "Primary Slave: %s\n",
3155                            (bond->primary_slave) ?
3156                            bond->primary_slave->dev->name : "None");
3157
3158                 seq_printf(seq, "Currently Active Slave: %s\n",
3159                            (curr) ? curr->dev->name : "None");
3160         }
3161
3162         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3163                    "up" : "down");
3164         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3165         seq_printf(seq, "Up Delay (ms): %d\n",
3166                    bond->params.updelay * bond->params.miimon);
3167         seq_printf(seq, "Down Delay (ms): %d\n",
3168                    bond->params.downdelay * bond->params.miimon);
3169
3170
3171         /* ARP information */
3172         if(bond->params.arp_interval > 0) {
3173                 int printed=0;
3174                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3175                                 bond->params.arp_interval);
3176
3177                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3178
3179                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3180                         if (!bond->params.arp_targets[i])
3181                                 continue;
3182                         if (printed)
3183                                 seq_printf(seq, ",");
3184                         target = ntohl(bond->params.arp_targets[i]);
3185                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3186                         printed = 1;
3187                 }
3188                 seq_printf(seq, "\n");
3189         }
3190
3191         if (bond->params.mode == BOND_MODE_8023AD) {
3192                 struct ad_info ad_info;
3193                 DECLARE_MAC_BUF(mac);
3194
3195                 seq_puts(seq, "\n802.3ad info\n");
3196                 seq_printf(seq, "LACP rate: %s\n",
3197                            (bond->params.lacp_fast) ? "fast" : "slow");
3198
3199                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3200                         seq_printf(seq, "bond %s has no active aggregator\n",
3201                                    bond->dev->name);
3202                 } else {
3203                         seq_printf(seq, "Active Aggregator Info:\n");
3204
3205                         seq_printf(seq, "\tAggregator ID: %d\n",
3206                                    ad_info.aggregator_id);
3207                         seq_printf(seq, "\tNumber of ports: %d\n",
3208                                    ad_info.ports);
3209                         seq_printf(seq, "\tActor Key: %d\n",
3210                                    ad_info.actor_key);
3211                         seq_printf(seq, "\tPartner Key: %d\n",
3212                                    ad_info.partner_key);
3213                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3214                                    print_mac(mac, ad_info.partner_system));
3215                 }
3216         }
3217 }
3218
3219 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3220 {
3221         struct bonding *bond = seq->private;
3222         DECLARE_MAC_BUF(mac);
3223
3224         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3225         seq_printf(seq, "MII Status: %s\n",
3226                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3227         seq_printf(seq, "Link Failure Count: %u\n",
3228                    slave->link_failure_count);
3229
3230         seq_printf(seq,
3231                    "Permanent HW addr: %s\n",
3232                    print_mac(mac, slave->perm_hwaddr));
3233
3234         if (bond->params.mode == BOND_MODE_8023AD) {
3235                 const struct aggregator *agg
3236                         = SLAVE_AD_INFO(slave).port.aggregator;
3237
3238                 if (agg) {
3239                         seq_printf(seq, "Aggregator ID: %d\n",
3240                                    agg->aggregator_identifier);
3241                 } else {
3242                         seq_puts(seq, "Aggregator ID: N/A\n");
3243                 }
3244         }
3245 }
3246
3247 static int bond_info_seq_show(struct seq_file *seq, void *v)
3248 {
3249         if (v == SEQ_START_TOKEN) {
3250                 seq_printf(seq, "%s\n", version);
3251                 bond_info_show_master(seq);
3252         } else {
3253                 bond_info_show_slave(seq, v);
3254         }
3255
3256         return 0;
3257 }
3258
3259 static struct seq_operations bond_info_seq_ops = {
3260         .start = bond_info_seq_start,
3261         .next  = bond_info_seq_next,
3262         .stop  = bond_info_seq_stop,
3263         .show  = bond_info_seq_show,
3264 };
3265
3266 static int bond_info_open(struct inode *inode, struct file *file)
3267 {
3268         struct seq_file *seq;
3269         struct proc_dir_entry *proc;
3270         int res;
3271
3272         res = seq_open(file, &bond_info_seq_ops);
3273         if (!res) {
3274                 /* recover the pointer buried in proc_dir_entry data */
3275                 seq = file->private_data;
3276                 proc = PDE(inode);
3277                 seq->private = proc->data;
3278         }
3279
3280         return res;
3281 }
3282
3283 static const struct file_operations bond_info_fops = {
3284         .owner   = THIS_MODULE,
3285         .open    = bond_info_open,
3286         .read    = seq_read,
3287         .llseek  = seq_lseek,
3288         .release = seq_release,
3289 };
3290
3291 static int bond_create_proc_entry(struct bonding *bond)
3292 {
3293         struct net_device *bond_dev = bond->dev;
3294
3295         if (bond_proc_dir) {
3296                 bond->proc_entry = create_proc_entry(bond_dev->name,
3297                                                      S_IRUGO,
3298                                                      bond_proc_dir);
3299                 if (bond->proc_entry == NULL) {
3300                         printk(KERN_WARNING DRV_NAME
3301                                ": Warning: Cannot create /proc/net/%s/%s\n",
3302                                DRV_NAME, bond_dev->name);
3303                 } else {
3304                         bond->proc_entry->data = bond;
3305                         bond->proc_entry->proc_fops = &bond_info_fops;
3306                         bond->proc_entry->owner = THIS_MODULE;
3307                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3308                 }
3309         }
3310
3311         return 0;
3312 }
3313
3314 static void bond_remove_proc_entry(struct bonding *bond)
3315 {
3316         if (bond_proc_dir && bond->proc_entry) {
3317                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3318                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3319                 bond->proc_entry = NULL;
3320         }
3321 }
3322
3323 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3324  * Caller must hold rtnl_lock.
3325  */
3326 static void bond_create_proc_dir(void)
3327 {
3328         int len = strlen(DRV_NAME);
3329
3330         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3331              bond_proc_dir = bond_proc_dir->next) {
3332                 if ((bond_proc_dir->namelen == len) &&
3333                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3334                         break;
3335                 }
3336         }
3337
3338         if (!bond_proc_dir) {
3339                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3340                 if (bond_proc_dir) {
3341                         bond_proc_dir->owner = THIS_MODULE;
3342                 } else {
3343                         printk(KERN_WARNING DRV_NAME
3344                                 ": Warning: cannot create /proc/net/%s\n",
3345                                 DRV_NAME);
3346                 }
3347         }
3348 }
3349
3350 /* Destroy the bonding directory under /proc/net, if empty.
3351  * Caller must hold rtnl_lock.
3352  */
3353 static void bond_destroy_proc_dir(void)
3354 {
3355         struct proc_dir_entry *de;
3356
3357         if (!bond_proc_dir) {
3358                 return;
3359         }
3360
3361         /* verify that the /proc dir is empty */
3362         for (de = bond_proc_dir->subdir; de; de = de->next) {
3363                 /* ignore . and .. */
3364                 if (*(de->name) != '.') {
3365                         break;
3366                 }
3367         }
3368
3369         if (de) {
3370                 if (bond_proc_dir->owner == THIS_MODULE) {
3371                         bond_proc_dir->owner = NULL;
3372                 }
3373         } else {
3374                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3375                 bond_proc_dir = NULL;
3376         }
3377 }
3378 #endif /* CONFIG_PROC_FS */
3379
3380 /*-------------------------- netdev event handling --------------------------*/
3381
3382 /*
3383  * Change device name
3384  */
3385 static int bond_event_changename(struct bonding *bond)
3386 {
3387 #ifdef CONFIG_PROC_FS
3388         bond_remove_proc_entry(bond);
3389         bond_create_proc_entry(bond);
3390 #endif
3391         down_write(&(bonding_rwsem));
3392         bond_destroy_sysfs_entry(bond);
3393         bond_create_sysfs_entry(bond);
3394         up_write(&(bonding_rwsem));
3395         return NOTIFY_DONE;
3396 }
3397
3398 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3399 {
3400         struct bonding *event_bond = bond_dev->priv;
3401
3402         switch (event) {
3403         case NETDEV_CHANGENAME:
3404                 return bond_event_changename(event_bond);
3405         case NETDEV_UNREGISTER:
3406                 bond_release_all(event_bond->dev);
3407                 break;
3408         default:
3409                 break;
3410         }
3411
3412         return NOTIFY_DONE;
3413 }
3414
3415 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3416 {
3417         struct net_device *bond_dev = slave_dev->master;
3418         struct bonding *bond = bond_dev->priv;
3419
3420         switch (event) {
3421         case NETDEV_UNREGISTER:
3422                 if (bond_dev) {
3423                         if (bond->setup_by_slave)
3424                                 bond_release_and_destroy(bond_dev, slave_dev);
3425                         else
3426                                 bond_release(bond_dev, slave_dev);
3427                 }
3428                 break;
3429         case NETDEV_CHANGE:
3430                 /*
3431                  * TODO: is this what we get if somebody
3432                  * sets up a hierarchical bond, then rmmod's
3433                  * one of the slave bonding devices?
3434                  */
3435                 break;
3436         case NETDEV_DOWN:
3437                 /*
3438                  * ... Or is it this?
3439                  */
3440                 break;
3441         case NETDEV_CHANGEMTU:
3442                 /*
3443                  * TODO: Should slaves be allowed to
3444                  * independently alter their MTU?  For
3445                  * an active-backup bond, slaves need
3446                  * not be the same type of device, so
3447                  * MTUs may vary.  For other modes,
3448                  * slaves arguably should have the
3449                  * same MTUs. To do this, we'd need to
3450                  * take over the slave's change_mtu
3451                  * function for the duration of their
3452                  * servitude.
3453                  */
3454                 break;
3455         case NETDEV_CHANGENAME:
3456                 /*
3457                  * TODO: handle changing the primary's name
3458                  */
3459                 break;
3460         case NETDEV_FEAT_CHANGE:
3461                 bond_compute_features(bond);
3462                 break;
3463         default:
3464                 break;
3465         }
3466
3467         return NOTIFY_DONE;
3468 }
3469
3470 /*
3471  * bond_netdev_event: handle netdev notifier chain events.
3472  *
3473  * This function receives events for the netdev chain.  The caller (an
3474  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3475  * locks for us to safely manipulate the slave devices (RTNL lock,
3476  * dev_probe_lock).
3477  */
3478 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3479 {
3480         struct net_device *event_dev = (struct net_device *)ptr;
3481
3482         if (event_dev->nd_net != &init_net)
3483                 return NOTIFY_DONE;
3484
3485         dprintk("event_dev: %s, event: %lx\n",
3486                 (event_dev ? event_dev->name : "None"),
3487                 event);
3488
3489         if (!(event_dev->priv_flags & IFF_BONDING))
3490                 return NOTIFY_DONE;
3491
3492         if (event_dev->flags & IFF_MASTER) {
3493                 dprintk("IFF_MASTER\n");
3494                 return bond_master_netdev_event(event, event_dev);
3495         }
3496
3497         if (event_dev->flags & IFF_SLAVE) {
3498                 dprintk("IFF_SLAVE\n");
3499                 return bond_slave_netdev_event(event, event_dev);
3500         }
3501
3502         return NOTIFY_DONE;
3503 }
3504
3505 /*
3506  * bond_inetaddr_event: handle inetaddr notifier chain events.
3507  *
3508  * We keep track of device IPs primarily to use as source addresses in
3509  * ARP monitor probes (rather than spewing out broadcasts all the time).
3510  *
3511  * We track one IP for the main device (if it has one), plus one per VLAN.
3512  */
3513 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3514 {
3515         struct in_ifaddr *ifa = ptr;
3516         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3517         struct bonding *bond, *bond_next;
3518         struct vlan_entry *vlan, *vlan_next;
3519
3520         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3521                 if (bond->dev == event_dev) {
3522                         switch (event) {
3523                         case NETDEV_UP:
3524                                 bond->master_ip = ifa->ifa_local;
3525                                 return NOTIFY_OK;
3526                         case NETDEV_DOWN:
3527                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3528                                 return NOTIFY_OK;
3529                         default:
3530                                 return NOTIFY_DONE;
3531                         }
3532                 }
3533
3534                 if (list_empty(&bond->vlan_list))
3535                         continue;
3536
3537                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3538                                          vlan_list) {
3539                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3540                         if (vlan_dev == event_dev) {
3541                                 switch (event) {
3542                                 case NETDEV_UP:
3543                                         vlan->vlan_ip = ifa->ifa_local;
3544                                         return NOTIFY_OK;
3545                                 case NETDEV_DOWN:
3546                                         vlan->vlan_ip =
3547                                                 bond_glean_dev_ip(vlan_dev);
3548                                         return NOTIFY_OK;
3549                                 default:
3550                                         return NOTIFY_DONE;
3551                                 }
3552                         }
3553                 }
3554         }
3555         return NOTIFY_DONE;
3556 }
3557
3558 static struct notifier_block bond_netdev_notifier = {
3559         .notifier_call = bond_netdev_event,
3560 };
3561
3562 static struct notifier_block bond_inetaddr_notifier = {
3563         .notifier_call = bond_inetaddr_event,
3564 };
3565
3566 /*-------------------------- Packet type handling ---------------------------*/
3567
3568 /* register to receive lacpdus on a bond */
3569 static void bond_register_lacpdu(struct bonding *bond)
3570 {
3571         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3572
3573         /* initialize packet type */
3574         pk_type->type = PKT_TYPE_LACPDU;
3575         pk_type->dev = bond->dev;
3576         pk_type->func = bond_3ad_lacpdu_recv;
3577
3578         dev_add_pack(pk_type);
3579 }
3580
3581 /* unregister to receive lacpdus on a bond */
3582 static void bond_unregister_lacpdu(struct bonding *bond)
3583 {
3584         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3585 }
3586
3587 void bond_register_arp(struct bonding *bond)
3588 {
3589         struct packet_type *pt = &bond->arp_mon_pt;
3590
3591         if (pt->type)
3592                 return;
3593
3594         pt->type = htons(ETH_P_ARP);
3595         pt->dev = bond->dev;
3596         pt->func = bond_arp_rcv;
3597         dev_add_pack(pt);
3598 }
3599
3600 void bond_unregister_arp(struct bonding *bond)
3601 {
3602         struct packet_type *pt = &bond->arp_mon_pt;
3603
3604         dev_remove_pack(pt);
3605         pt->type = 0;
3606 }
3607
3608 /*---------------------------- Hashing Policies -----------------------------*/
3609
3610 /*
3611  * Hash for the output device based upon layer 2 and layer 3 data. If
3612  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3613  */
3614 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3615                                      struct net_device *bond_dev, int count)
3616 {
3617         struct ethhdr *data = (struct ethhdr *)skb->data;
3618         struct iphdr *iph = ip_hdr(skb);
3619
3620         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3621                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3622                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3623         }
3624
3625         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3626 }
3627
3628 /*
3629  * Hash for the output device based upon layer 3 and layer 4 data. If
3630  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3631  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3632  */
3633 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3634                                     struct net_device *bond_dev, int count)
3635 {
3636         struct ethhdr *data = (struct ethhdr *)skb->data;
3637         struct iphdr *iph = ip_hdr(skb);
3638         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3639         int layer4_xor = 0;
3640
3641         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3642                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3643                     (iph->protocol == IPPROTO_TCP ||
3644                      iph->protocol == IPPROTO_UDP)) {
3645                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3646                 }
3647                 return (layer4_xor ^
3648                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3649
3650         }
3651
3652         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3653 }
3654
3655 /*
3656  * Hash for the output device based upon layer 2 data
3657  */
3658 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3659                                    struct net_device *bond_dev, int count)
3660 {
3661         struct ethhdr *data = (struct ethhdr *)skb->data;
3662
3663         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3664 }
3665
3666 /*-------------------------- Device entry points ----------------------------*/
3667
3668 static int bond_open(struct net_device *bond_dev)
3669 {
3670         struct bonding *bond = bond_dev->priv;
3671
3672         bond->kill_timers = 0;
3673
3674         if ((bond->params.mode == BOND_MODE_TLB) ||
3675             (bond->params.mode == BOND_MODE_ALB)) {
3676                 /* bond_alb_initialize must be called before the timer
3677                  * is started.
3678                  */
3679                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3680                         /* something went wrong - fail the open operation */
3681                         return -1;
3682                 }
3683
3684                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3685                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3686         }
3687
3688         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3689                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3690                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3691         }
3692
3693         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3694                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3695                         INIT_DELAYED_WORK(&bond->arp_work,
3696                                           bond_activebackup_arp_mon);
3697                 else
3698                         INIT_DELAYED_WORK(&bond->arp_work,
3699                                           bond_loadbalance_arp_mon);
3700
3701                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3702                 if (bond->params.arp_validate)
3703                         bond_register_arp(bond);
3704         }
3705
3706         if (bond->params.mode == BOND_MODE_8023AD) {
3707                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3708                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3709                 /* register to receive LACPDUs */
3710                 bond_register_lacpdu(bond);
3711         }
3712
3713         return 0;
3714 }
3715
3716 static int bond_close(struct net_device *bond_dev)
3717 {
3718         struct bonding *bond = bond_dev->priv;
3719
3720         if (bond->params.mode == BOND_MODE_8023AD) {
3721                 /* Unregister the receive of LACPDUs */
3722                 bond_unregister_lacpdu(bond);
3723         }
3724
3725         if (bond->params.arp_validate)
3726                 bond_unregister_arp(bond);
3727
3728         write_lock_bh(&bond->lock);
3729
3730
3731         /* signal timers not to re-arm */
3732         bond->kill_timers = 1;
3733
3734         write_unlock_bh(&bond->lock);
3735
3736         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3737                 cancel_delayed_work(&bond->mii_work);
3738         }
3739
3740         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3741                 cancel_delayed_work(&bond->arp_work);
3742         }
3743
3744         switch (bond->params.mode) {
3745         case BOND_MODE_8023AD:
3746                 cancel_delayed_work(&bond->ad_work);
3747                 break;
3748         case BOND_MODE_TLB:
3749         case BOND_MODE_ALB:
3750                 cancel_delayed_work(&bond->alb_work);
3751                 break;
3752         default:
3753                 break;
3754         }
3755
3756
3757         if ((bond->params.mode == BOND_MODE_TLB) ||
3758             (bond->params.mode == BOND_MODE_ALB)) {
3759                 /* Must be called only after all
3760                  * slaves have been released
3761                  */
3762                 bond_alb_deinitialize(bond);
3763         }
3764
3765         return 0;
3766 }
3767
3768 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3769 {
3770         struct bonding *bond = bond_dev->priv;
3771         struct net_device_stats *stats = &(bond->stats), *sstats;
3772         struct slave *slave;
3773         int i;
3774
3775         memset(stats, 0, sizeof(struct net_device_stats));
3776
3777         read_lock_bh(&bond->lock);
3778
3779         bond_for_each_slave(bond, slave, i) {
3780                 sstats = slave->dev->get_stats(slave->dev);
3781                 stats->rx_packets += sstats->rx_packets;
3782                 stats->rx_bytes += sstats->rx_bytes;
3783                 stats->rx_errors += sstats->rx_errors;
3784                 stats->rx_dropped += sstats->rx_dropped;
3785
3786                 stats->tx_packets += sstats->tx_packets;
3787                 stats->tx_bytes += sstats->tx_bytes;
3788                 stats->tx_errors += sstats->tx_errors;
3789                 stats->tx_dropped += sstats->tx_dropped;
3790
3791                 stats->multicast += sstats->multicast;
3792                 stats->collisions += sstats->collisions;
3793
3794                 stats->rx_length_errors += sstats->rx_length_errors;
3795                 stats->rx_over_errors += sstats->rx_over_errors;
3796                 stats->rx_crc_errors += sstats->rx_crc_errors;
3797                 stats->rx_frame_errors += sstats->rx_frame_errors;
3798                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3799                 stats->rx_missed_errors += sstats->rx_missed_errors;
3800
3801                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3802                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3803                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3804                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3805                 stats->tx_window_errors += sstats->tx_window_errors;
3806         }
3807
3808         read_unlock_bh(&bond->lock);
3809
3810         return stats;
3811 }
3812
3813 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3814 {
3815         struct net_device *slave_dev = NULL;
3816         struct ifbond k_binfo;
3817         struct ifbond __user *u_binfo = NULL;
3818         struct ifslave k_sinfo;
3819         struct ifslave __user *u_sinfo = NULL;
3820         struct mii_ioctl_data *mii = NULL;
3821         int res = 0;
3822
3823         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3824                 bond_dev->name, cmd);
3825
3826         switch (cmd) {
3827         case SIOCGMIIPHY:
3828                 mii = if_mii(ifr);
3829                 if (!mii) {
3830                         return -EINVAL;
3831                 }
3832                 mii->phy_id = 0;
3833                 /* Fall Through */
3834         case SIOCGMIIREG:
3835                 /*
3836                  * We do this again just in case we were called by SIOCGMIIREG
3837                  * instead of SIOCGMIIPHY.
3838                  */
3839                 mii = if_mii(ifr);
3840                 if (!mii) {
3841                         return -EINVAL;
3842                 }
3843
3844                 if (mii->reg_num == 1) {
3845                         struct bonding *bond = bond_dev->priv;
3846                         mii->val_out = 0;
3847                         read_lock(&bond->lock);
3848                         read_lock(&bond->curr_slave_lock);
3849                         if (netif_carrier_ok(bond->dev)) {
3850                                 mii->val_out = BMSR_LSTATUS;
3851                         }
3852                         read_unlock(&bond->curr_slave_lock);
3853                         read_unlock(&bond->lock);
3854                 }
3855
3856                 return 0;
3857         case BOND_INFO_QUERY_OLD:
3858         case SIOCBONDINFOQUERY:
3859                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3860
3861                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3862                         return -EFAULT;
3863                 }
3864
3865                 res = bond_info_query(bond_dev, &k_binfo);
3866                 if (res == 0) {
3867                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3868                                 return -EFAULT;
3869                         }
3870                 }
3871
3872                 return res;
3873         case BOND_SLAVE_INFO_QUERY_OLD:
3874         case SIOCBONDSLAVEINFOQUERY:
3875                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3876
3877                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3878                         return -EFAULT;
3879                 }
3880
3881                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3882                 if (res == 0) {
3883                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3884                                 return -EFAULT;
3885                         }
3886                 }
3887
3888                 return res;
3889         default:
3890                 /* Go on */
3891                 break;
3892         }
3893
3894         if (!capable(CAP_NET_ADMIN)) {
3895                 return -EPERM;
3896         }
3897
3898         down_write(&(bonding_rwsem));
3899         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3900
3901         dprintk("slave_dev=%p: \n", slave_dev);
3902
3903         if (!slave_dev) {
3904                 res = -ENODEV;
3905         } else {
3906                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3907                 switch (cmd) {
3908                 case BOND_ENSLAVE_OLD:
3909                 case SIOCBONDENSLAVE:
3910                         res = bond_enslave(bond_dev, slave_dev);
3911                         break;
3912                 case BOND_RELEASE_OLD:
3913                 case SIOCBONDRELEASE:
3914                         res = bond_release(bond_dev, slave_dev);
3915                         break;
3916                 case BOND_SETHWADDR_OLD:
3917                 case SIOCBONDSETHWADDR:
3918                         res = bond_sethwaddr(bond_dev, slave_dev);
3919                         break;
3920                 case BOND_CHANGE_ACTIVE_OLD:
3921                 case SIOCBONDCHANGEACTIVE:
3922                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3923                         break;
3924                 default:
3925                         res = -EOPNOTSUPP;
3926                 }
3927
3928                 dev_put(slave_dev);
3929         }
3930
3931         up_write(&(bonding_rwsem));
3932         return res;
3933 }
3934
3935 static void bond_set_multicast_list(struct net_device *bond_dev)
3936 {
3937         struct bonding *bond = bond_dev->priv;
3938         struct dev_mc_list *dmi;
3939
3940         write_lock_bh(&bond->lock);
3941
3942         /*
3943          * Do promisc before checking multicast_mode
3944          */
3945         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3946                 bond_set_promiscuity(bond, 1);
3947         }
3948
3949         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3950                 bond_set_promiscuity(bond, -1);
3951         }
3952
3953         /* set allmulti flag to slaves */
3954         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3955                 bond_set_allmulti(bond, 1);
3956         }
3957
3958         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3959                 bond_set_allmulti(bond, -1);
3960         }
3961
3962         bond->flags = bond_dev->flags;
3963
3964         /* looking for addresses to add to slaves' mc list */
3965         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3966                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3967                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3968                 }
3969         }
3970
3971         /* looking for addresses to delete from slaves' list */
3972         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3973                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3974                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3975                 }
3976         }
3977
3978         /* save master's multicast list */
3979         bond_mc_list_destroy(bond);
3980         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3981
3982         write_unlock_bh(&bond->lock);
3983 }
3984
3985 /*
3986  * Change the MTU of all of a master's slaves to match the master
3987  */
3988 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3989 {
3990         struct bonding *bond = bond_dev->priv;
3991         struct slave *slave, *stop_at;
3992         int res = 0;
3993         int i;
3994
3995         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3996                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3997
3998         /* Can't hold bond->lock with bh disabled here since
3999          * some base drivers panic. On the other hand we can't
4000          * hold bond->lock without bh disabled because we'll
4001          * deadlock. The only solution is to rely on the fact
4002          * that we're under rtnl_lock here, and the slaves
4003          * list won't change. This doesn't solve the problem
4004          * of setting the slave's MTU while it is
4005          * transmitting, but the assumption is that the base
4006          * driver can handle that.
4007          *
4008          * TODO: figure out a way to safely iterate the slaves
4009          * list, but without holding a lock around the actual
4010          * call to the base driver.
4011          */
4012
4013         bond_for_each_slave(bond, slave, i) {
4014                 dprintk("s %p s->p %p c_m %p\n", slave,
4015                         slave->prev, slave->dev->change_mtu);
4016
4017                 res = dev_set_mtu(slave->dev, new_mtu);
4018
4019                 if (res) {
4020                         /* If we failed to set the slave's mtu to the new value
4021                          * we must abort the operation even in ACTIVE_BACKUP
4022                          * mode, because if we allow the backup slaves to have
4023                          * different mtu values than the active slave we'll
4024                          * need to change their mtu when doing a failover. That
4025                          * means changing their mtu from timer context, which
4026                          * is probably not a good idea.
4027                          */
4028                         dprintk("err %d %s\n", res, slave->dev->name);
4029                         goto unwind;
4030                 }
4031         }
4032
4033         bond_dev->mtu = new_mtu;
4034
4035         return 0;
4036
4037 unwind:
4038         /* unwind from head to the slave that failed */
4039         stop_at = slave;
4040         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4041                 int tmp_res;
4042
4043                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4044                 if (tmp_res) {
4045                         dprintk("unwind err %d dev %s\n", tmp_res,
4046                                 slave->dev->name);
4047                 }
4048         }
4049
4050         return res;
4051 }
4052
4053 /*
4054  * Change HW address
4055  *
4056  * Note that many devices must be down to change the HW address, and
4057  * downing the master releases all slaves.  We can make bonds full of
4058  * bonding devices to test this, however.
4059  */
4060 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4061 {
4062         struct bonding *bond = bond_dev->priv;
4063         struct sockaddr *sa = addr, tmp_sa;
4064         struct slave *slave, *stop_at;
4065         int res = 0;
4066         int i;
4067
4068         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4069
4070         /*
4071          * If fail_over_mac is enabled, do nothing and return success.
4072          * Returning an error causes ifenslave to fail.
4073          */
4074         if (bond->params.fail_over_mac)
4075                 return 0;
4076
4077         if (!is_valid_ether_addr(sa->sa_data)) {
4078                 return -EADDRNOTAVAIL;
4079         }
4080
4081         /* Can't hold bond->lock with bh disabled here since
4082          * some base drivers panic. On the other hand we can't
4083          * hold bond->lock without bh disabled because we'll
4084          * deadlock. The only solution is to rely on the fact
4085          * that we're under rtnl_lock here, and the slaves
4086          * list won't change. This doesn't solve the problem
4087          * of setting the slave's hw address while it is
4088          * transmitting, but the assumption is that the base
4089          * driver can handle that.
4090          *
4091          * TODO: figure out a way to safely iterate the slaves
4092          * list, but without holding a lock around the actual
4093          * call to the base driver.
4094          */
4095
4096         bond_for_each_slave(bond, slave, i) {
4097                 dprintk("slave %p %s\n", slave, slave->dev->name);
4098
4099                 if (slave->dev->set_mac_address == NULL) {
4100                         res = -EOPNOTSUPP;
4101                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4102                         goto unwind;
4103                 }
4104
4105                 res = dev_set_mac_address(slave->dev, addr);
4106                 if (res) {
4107                         /* TODO: consider downing the slave
4108                          * and retry ?
4109                          * User should expect communications
4110                          * breakage anyway until ARP finish
4111                          * updating, so...
4112                          */
4113                         dprintk("err %d %s\n", res, slave->dev->name);
4114                         goto unwind;
4115                 }
4116         }
4117
4118         /* success */
4119         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4120         return 0;
4121
4122 unwind:
4123         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4124         tmp_sa.sa_family = bond_dev->type;
4125
4126         /* unwind from head to the slave that failed */
4127         stop_at = slave;
4128         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4129                 int tmp_res;
4130
4131                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4132                 if (tmp_res) {
4133                         dprintk("unwind err %d dev %s\n", tmp_res,
4134                                 slave->dev->name);
4135                 }
4136         }
4137
4138         return res;
4139 }
4140
4141 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4142 {
4143         struct bonding *bond = bond_dev->priv;
4144         struct slave *slave, *start_at;
4145         int i, slave_no, res = 1;
4146
4147         read_lock(&bond->lock);
4148
4149         if (!BOND_IS_OK(bond)) {
4150                 goto out;
4151         }
4152
4153         /*
4154          * Concurrent TX may collide on rr_tx_counter; we accept that
4155          * as being rare enough not to justify using an atomic op here
4156          */
4157         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4158
4159         bond_for_each_slave(bond, slave, i) {
4160                 slave_no--;
4161                 if (slave_no < 0) {
4162                         break;
4163                 }
4164         }
4165
4166         start_at = slave;
4167         bond_for_each_slave_from(bond, slave, i, start_at) {
4168                 if (IS_UP(slave->dev) &&
4169                     (slave->link == BOND_LINK_UP) &&
4170                     (slave->state == BOND_STATE_ACTIVE)) {
4171                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4172                         break;
4173                 }
4174         }
4175
4176 out:
4177         if (res) {
4178                 /* no suitable interface, frame not sent */
4179                 dev_kfree_skb(skb);
4180         }
4181         read_unlock(&bond->lock);
4182         return 0;
4183 }
4184
4185
4186 /*
4187  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4188  * the bond has a usable interface.
4189  */
4190 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4191 {
4192         struct bonding *bond = bond_dev->priv;
4193         int res = 1;
4194
4195         read_lock(&bond->lock);
4196         read_lock(&bond->curr_slave_lock);
4197
4198         if (!BOND_IS_OK(bond)) {
4199                 goto out;
4200         }
4201
4202         if (!bond->curr_active_slave)
4203                 goto out;
4204
4205         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4206
4207 out:
4208         if (res) {
4209                 /* no suitable interface, frame not sent */
4210                 dev_kfree_skb(skb);
4211         }
4212         read_unlock(&bond->curr_slave_lock);
4213         read_unlock(&bond->lock);
4214         return 0;
4215 }
4216
4217 /*
4218  * In bond_xmit_xor() , we determine the output device by using a pre-
4219  * determined xmit_hash_policy(), If the selected device is not enabled,
4220  * find the next active slave.
4221  */
4222 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4223 {
4224         struct bonding *bond = bond_dev->priv;
4225         struct slave *slave, *start_at;
4226         int slave_no;
4227         int i;
4228         int res = 1;
4229
4230         read_lock(&bond->lock);
4231
4232         if (!BOND_IS_OK(bond)) {
4233                 goto out;
4234         }
4235
4236         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4237
4238         bond_for_each_slave(bond, slave, i) {
4239                 slave_no--;
4240                 if (slave_no < 0) {
4241                         break;
4242                 }
4243         }
4244
4245         start_at = slave;
4246
4247         bond_for_each_slave_from(bond, slave, i, start_at) {
4248                 if (IS_UP(slave->dev) &&
4249                     (slave->link == BOND_LINK_UP) &&
4250                     (slave->state == BOND_STATE_ACTIVE)) {
4251                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4252                         break;
4253                 }
4254         }
4255
4256 out:
4257         if (res) {
4258                 /* no suitable interface, frame not sent */
4259                 dev_kfree_skb(skb);
4260         }
4261         read_unlock(&bond->lock);
4262         return 0;
4263 }
4264
4265 /*
4266  * in broadcast mode, we send everything to all usable interfaces.
4267  */
4268 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4269 {
4270         struct bonding *bond = bond_dev->priv;
4271         struct slave *slave, *start_at;
4272         struct net_device *tx_dev = NULL;
4273         int i;
4274         int res = 1;
4275
4276         read_lock(&bond->lock);
4277
4278         if (!BOND_IS_OK(bond)) {
4279                 goto out;
4280         }
4281
4282         read_lock(&bond->curr_slave_lock);
4283         start_at = bond->curr_active_slave;
4284         read_unlock(&bond->curr_slave_lock);
4285
4286         if (!start_at) {
4287                 goto out;
4288         }
4289
4290         bond_for_each_slave_from(bond, slave, i, start_at) {
4291                 if (IS_UP(slave->dev) &&
4292                     (slave->link == BOND_LINK_UP) &&
4293                     (slave->state == BOND_STATE_ACTIVE)) {
4294                         if (tx_dev) {
4295                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4296                                 if (!skb2) {
4297                                         printk(KERN_ERR DRV_NAME
4298                                                ": %s: Error: bond_xmit_broadcast(): "
4299                                                "skb_clone() failed\n",
4300                                                bond_dev->name);
4301                                         continue;
4302                                 }
4303
4304                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4305                                 if (res) {
4306                                         dev_kfree_skb(skb2);
4307                                         continue;
4308                                 }
4309                         }
4310                         tx_dev = slave->dev;
4311                 }
4312         }
4313
4314         if (tx_dev) {
4315                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4316         }
4317
4318 out:
4319         if (res) {
4320                 /* no suitable interface, frame not sent */
4321                 dev_kfree_skb(skb);
4322         }
4323         /* frame sent to all suitable interfaces */
4324         read_unlock(&bond->lock);
4325         return 0;
4326 }
4327
4328 /*------------------------- Device initialization ---------------------------*/
4329
4330 static void bond_set_xmit_hash_policy(struct bonding *bond)
4331 {
4332         switch (bond->params.xmit_policy) {
4333         case BOND_XMIT_POLICY_LAYER23:
4334                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4335                 break;
4336         case BOND_XMIT_POLICY_LAYER34:
4337                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4338                 break;
4339         case BOND_XMIT_POLICY_LAYER2:
4340         default:
4341                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4342                 break;
4343         }
4344 }
4345
4346 /*
4347  * set bond mode specific net device operations
4348  */
4349 void bond_set_mode_ops(struct bonding *bond, int mode)
4350 {
4351         struct net_device *bond_dev = bond->dev;
4352
4353         switch (mode) {
4354         case BOND_MODE_ROUNDROBIN:
4355                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4356                 break;
4357         case BOND_MODE_ACTIVEBACKUP:
4358                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4359                 break;
4360         case BOND_MODE_XOR:
4361                 bond_dev->hard_start_xmit = bond_xmit_xor;
4362                 bond_set_xmit_hash_policy(bond);
4363                 break;
4364         case BOND_MODE_BROADCAST:
4365                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4366                 break;
4367         case BOND_MODE_8023AD:
4368                 bond_set_master_3ad_flags(bond);
4369                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4370                 bond_set_xmit_hash_policy(bond);
4371                 break;
4372         case BOND_MODE_ALB:
4373                 bond_set_master_alb_flags(bond);
4374                 /* FALLTHRU */
4375         case BOND_MODE_TLB:
4376                 bond_dev->hard_start_xmit = bond_alb_xmit;
4377                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4378                 break;
4379         default:
4380                 /* Should never happen, mode already checked */
4381                 printk(KERN_ERR DRV_NAME
4382                        ": %s: Error: Unknown bonding mode %d\n",
4383                        bond_dev->name,
4384                        mode);
4385                 break;
4386         }
4387 }
4388
4389 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4390                                     struct ethtool_drvinfo *drvinfo)
4391 {
4392         strncpy(drvinfo->driver, DRV_NAME, 32);
4393         strncpy(drvinfo->version, DRV_VERSION, 32);
4394         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4395 }
4396
4397 static const struct ethtool_ops bond_ethtool_ops = {
4398         .get_drvinfo            = bond_ethtool_get_drvinfo,
4399 };
4400
4401 /*
4402  * Does not allocate but creates a /proc entry.
4403  * Allowed to fail.
4404  */
4405 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4406 {
4407         struct bonding *bond = bond_dev->priv;
4408
4409         dprintk("Begin bond_init for %s\n", bond_dev->name);
4410
4411         /* initialize rwlocks */
4412         rwlock_init(&bond->lock);
4413         rwlock_init(&bond->curr_slave_lock);
4414
4415         bond->params = *params; /* copy params struct */
4416
4417         bond->wq = create_singlethread_workqueue(bond_dev->name);
4418         if (!bond->wq)
4419                 return -ENOMEM;
4420
4421         /* Initialize pointers */
4422         bond->first_slave = NULL;
4423         bond->curr_active_slave = NULL;
4424         bond->current_arp_slave = NULL;
4425         bond->primary_slave = NULL;
4426         bond->dev = bond_dev;
4427         bond->send_grat_arp = 0;
4428         bond->setup_by_slave = 0;
4429         INIT_LIST_HEAD(&bond->vlan_list);
4430
4431         /* Initialize the device entry points */
4432         bond_dev->open = bond_open;
4433         bond_dev->stop = bond_close;
4434         bond_dev->get_stats = bond_get_stats;
4435         bond_dev->do_ioctl = bond_do_ioctl;
4436         bond_dev->ethtool_ops = &bond_ethtool_ops;
4437         bond_dev->set_multicast_list = bond_set_multicast_list;
4438         bond_dev->change_mtu = bond_change_mtu;
4439         bond_dev->set_mac_address = bond_set_mac_address;
4440         bond_dev->validate_addr = NULL;
4441
4442         bond_set_mode_ops(bond, bond->params.mode);
4443
4444         bond_dev->destructor = free_netdev;
4445
4446         /* Initialize the device options */
4447         bond_dev->tx_queue_len = 0;
4448         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4449         bond_dev->priv_flags |= IFF_BONDING;
4450
4451         /* At first, we block adding VLANs. That's the only way to
4452          * prevent problems that occur when adding VLANs over an
4453          * empty bond. The block will be removed once non-challenged
4454          * slaves are enslaved.
4455          */
4456         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4457
4458         /* don't acquire bond device's netif_tx_lock when
4459          * transmitting */
4460         bond_dev->features |= NETIF_F_LLTX;
4461
4462         /* By default, we declare the bond to be fully
4463          * VLAN hardware accelerated capable. Special
4464          * care is taken in the various xmit functions
4465          * when there are slaves that are not hw accel
4466          * capable
4467          */
4468         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4469         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4470         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4471         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4472                                NETIF_F_HW_VLAN_RX |
4473                                NETIF_F_HW_VLAN_FILTER);
4474
4475 #ifdef CONFIG_PROC_FS
4476         bond_create_proc_entry(bond);
4477 #endif
4478         list_add_tail(&bond->bond_list, &bond_dev_list);
4479
4480         return 0;
4481 }
4482
4483 /* De-initialize device specific data.
4484  * Caller must hold rtnl_lock.
4485  */
4486 static void bond_deinit(struct net_device *bond_dev)
4487 {
4488         struct bonding *bond = bond_dev->priv;
4489
4490         list_del(&bond->bond_list);
4491
4492 #ifdef CONFIG_PROC_FS
4493         bond_remove_proc_entry(bond);
4494 #endif
4495 }
4496
4497 static void bond_work_cancel_all(struct bonding *bond)
4498 {
4499         write_lock_bh(&bond->lock);
4500         bond->kill_timers = 1;
4501         write_unlock_bh(&bond->lock);
4502
4503         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4504                 cancel_delayed_work(&bond->mii_work);
4505
4506         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4507                 cancel_delayed_work(&bond->arp_work);
4508
4509         if (bond->params.mode == BOND_MODE_ALB &&
4510             delayed_work_pending(&bond->alb_work))
4511                 cancel_delayed_work(&bond->alb_work);
4512
4513         if (bond->params.mode == BOND_MODE_8023AD &&
4514             delayed_work_pending(&bond->ad_work))
4515                 cancel_delayed_work(&bond->ad_work);
4516 }
4517
4518 /* Unregister and free all bond devices.
4519  * Caller must hold rtnl_lock.
4520  */
4521 static void bond_free_all(void)
4522 {
4523         struct bonding *bond, *nxt;
4524
4525         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4526                 struct net_device *bond_dev = bond->dev;
4527
4528                 bond_work_cancel_all(bond);
4529                 bond_mc_list_destroy(bond);
4530                 /* Release the bonded slaves */
4531                 bond_release_all(bond_dev);
4532                 bond_deinit(bond_dev);
4533                 unregister_netdevice(bond_dev);
4534         }
4535
4536 #ifdef CONFIG_PROC_FS
4537         bond_destroy_proc_dir();
4538 #endif
4539 }
4540
4541 /*------------------------- Module initialization ---------------------------*/
4542
4543 /*
4544  * Convert string input module parms.  Accept either the
4545  * number of the mode or its string name.  A bit complicated because
4546  * some mode names are substrings of other names, and calls from sysfs
4547  * may have whitespace in the name (trailing newlines, for example).
4548  */
4549 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4550 {
4551         int mode = -1, i, rv;
4552         char modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4553
4554         rv = sscanf(buf, "%d", &mode);
4555         if (!rv) {
4556                 rv = sscanf(buf, "%20s", modestr);
4557                 if (!rv)
4558                         return -1;
4559         }
4560
4561         for (i = 0; tbl[i].modename; i++) {
4562                 if (mode == tbl[i].mode)
4563                         return tbl[i].mode;
4564                 if (strcmp(modestr, tbl[i].modename) == 0)
4565                         return tbl[i].mode;
4566         }
4567
4568         return -1;
4569 }
4570
4571 static int bond_check_params(struct bond_params *params)
4572 {
4573         int arp_validate_value;
4574
4575         /*
4576          * Convert string parameters.
4577          */
4578         if (mode) {
4579                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4580                 if (bond_mode == -1) {
4581                         printk(KERN_ERR DRV_NAME
4582                                ": Error: Invalid bonding mode \"%s\"\n",
4583                                mode == NULL ? "NULL" : mode);
4584                         return -EINVAL;
4585                 }
4586         }
4587
4588         if (xmit_hash_policy) {
4589                 if ((bond_mode != BOND_MODE_XOR) &&
4590                     (bond_mode != BOND_MODE_8023AD)) {
4591                         printk(KERN_INFO DRV_NAME
4592                                ": xor_mode param is irrelevant in mode %s\n",
4593                                bond_mode_name(bond_mode));
4594                 } else {
4595                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4596                                                         xmit_hashtype_tbl);
4597                         if (xmit_hashtype == -1) {
4598                                 printk(KERN_ERR DRV_NAME
4599                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4600                                 xmit_hash_policy == NULL ? "NULL" :
4601                                        xmit_hash_policy);
4602                                 return -EINVAL;
4603                         }
4604                 }
4605         }
4606
4607         if (lacp_rate) {
4608                 if (bond_mode != BOND_MODE_8023AD) {
4609                         printk(KERN_INFO DRV_NAME
4610                                ": lacp_rate param is irrelevant in mode %s\n",
4611                                bond_mode_name(bond_mode));
4612                 } else {
4613                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4614                         if (lacp_fast == -1) {
4615                                 printk(KERN_ERR DRV_NAME
4616                                        ": Error: Invalid lacp rate \"%s\"\n",
4617                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4618                                 return -EINVAL;
4619                         }
4620                 }
4621         }
4622
4623         if (max_bonds < 1 || max_bonds > INT_MAX) {
4624                 printk(KERN_WARNING DRV_NAME
4625                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4626                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4627                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4628                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4629         }
4630
4631         if (miimon < 0) {
4632                 printk(KERN_WARNING DRV_NAME
4633                        ": Warning: miimon module parameter (%d), "
4634                        "not in range 0-%d, so it was reset to %d\n",
4635                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4636                 miimon = BOND_LINK_MON_INTERV;
4637         }
4638
4639         if (updelay < 0) {
4640                 printk(KERN_WARNING DRV_NAME
4641                        ": Warning: updelay module parameter (%d), "
4642                        "not in range 0-%d, so it was reset to 0\n",
4643                        updelay, INT_MAX);
4644                 updelay = 0;
4645         }
4646
4647         if (downdelay < 0) {
4648                 printk(KERN_WARNING DRV_NAME
4649                        ": Warning: downdelay module parameter (%d), "
4650                        "not in range 0-%d, so it was reset to 0\n",
4651                        downdelay, INT_MAX);
4652                 downdelay = 0;
4653         }
4654
4655         if ((use_carrier != 0) && (use_carrier != 1)) {
4656                 printk(KERN_WARNING DRV_NAME
4657                        ": Warning: use_carrier module parameter (%d), "
4658                        "not of valid value (0/1), so it was set to 1\n",
4659                        use_carrier);
4660                 use_carrier = 1;
4661         }
4662
4663         /* reset values for 802.3ad */
4664         if (bond_mode == BOND_MODE_8023AD) {
4665                 if (!miimon) {
4666                         printk(KERN_WARNING DRV_NAME
4667                                ": Warning: miimon must be specified, "
4668                                "otherwise bonding will not detect link "
4669                                "failure, speed and duplex which are "
4670                                "essential for 802.3ad operation\n");
4671                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4672                         miimon = 100;
4673                 }
4674         }
4675
4676         /* reset values for TLB/ALB */
4677         if ((bond_mode == BOND_MODE_TLB) ||
4678             (bond_mode == BOND_MODE_ALB)) {
4679                 if (!miimon) {
4680                         printk(KERN_WARNING DRV_NAME
4681                                ": Warning: miimon must be specified, "
4682                                "otherwise bonding will not detect link "
4683                                "failure and link speed which are essential "
4684                                "for TLB/ALB load balancing\n");
4685                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4686                         miimon = 100;
4687                 }
4688         }
4689
4690         if (bond_mode == BOND_MODE_ALB) {
4691                 printk(KERN_NOTICE DRV_NAME
4692                        ": In ALB mode you might experience client "
4693                        "disconnections upon reconnection of a link if the "
4694                        "bonding module updelay parameter (%d msec) is "
4695                        "incompatible with the forwarding delay time of the "
4696                        "switch\n",
4697                        updelay);
4698         }
4699
4700         if (!miimon) {
4701                 if (updelay || downdelay) {
4702                         /* just warn the user the up/down delay will have
4703                          * no effect since miimon is zero...
4704                          */
4705                         printk(KERN_WARNING DRV_NAME
4706                                ": Warning: miimon module parameter not set "
4707                                "and updelay (%d) or downdelay (%d) module "
4708                                "parameter is set; updelay and downdelay have "
4709                                "no effect unless miimon is set\n",
4710                                updelay, downdelay);
4711                 }
4712         } else {
4713                 /* don't allow arp monitoring */
4714                 if (arp_interval) {
4715                         printk(KERN_WARNING DRV_NAME
4716                                ": Warning: miimon (%d) and arp_interval (%d) "
4717                                "can't be used simultaneously, disabling ARP "
4718                                "monitoring\n",
4719                                miimon, arp_interval);
4720                         arp_interval = 0;
4721                 }
4722
4723                 if ((updelay % miimon) != 0) {
4724                         printk(KERN_WARNING DRV_NAME
4725                                ": Warning: updelay (%d) is not a multiple "
4726                                "of miimon (%d), updelay rounded to %d ms\n",
4727                                updelay, miimon, (updelay / miimon) * miimon);
4728                 }
4729
4730                 updelay /= miimon;
4731
4732                 if ((downdelay % miimon) != 0) {
4733                         printk(KERN_WARNING DRV_NAME
4734                                ": Warning: downdelay (%d) is not a multiple "
4735                                "of miimon (%d), downdelay rounded to %d ms\n",
4736                                downdelay, miimon,
4737                                (downdelay / miimon) * miimon);
4738                 }
4739
4740                 downdelay /= miimon;
4741         }
4742
4743         if (arp_interval < 0) {
4744                 printk(KERN_WARNING DRV_NAME
4745                        ": Warning: arp_interval module parameter (%d) "
4746                        ", not in range 0-%d, so it was reset to %d\n",
4747                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4748                 arp_interval = BOND_LINK_ARP_INTERV;
4749         }
4750
4751         for (arp_ip_count = 0;
4752              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4753              arp_ip_count++) {
4754                 /* not complete check, but should be good enough to
4755                    catch mistakes */
4756                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4757                         printk(KERN_WARNING DRV_NAME
4758                                ": Warning: bad arp_ip_target module parameter "
4759                                "(%s), ARP monitoring will not be performed\n",
4760                                arp_ip_target[arp_ip_count]);
4761                         arp_interval = 0;
4762                 } else {
4763                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4764                         arp_target[arp_ip_count] = ip;
4765                 }
4766         }
4767
4768         if (arp_interval && !arp_ip_count) {
4769                 /* don't allow arping if no arp_ip_target given... */
4770                 printk(KERN_WARNING DRV_NAME
4771                        ": Warning: arp_interval module parameter (%d) "
4772                        "specified without providing an arp_ip_target "
4773                        "parameter, arp_interval was reset to 0\n",
4774                        arp_interval);
4775                 arp_interval = 0;
4776         }
4777
4778         if (arp_validate) {
4779                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4780                         printk(KERN_ERR DRV_NAME
4781                ": arp_validate only supported in active-backup mode\n");
4782                         return -EINVAL;
4783                 }
4784                 if (!arp_interval) {
4785                         printk(KERN_ERR DRV_NAME
4786                                ": arp_validate requires arp_interval\n");
4787                         return -EINVAL;
4788                 }
4789
4790                 arp_validate_value = bond_parse_parm(arp_validate,
4791                                                      arp_validate_tbl);
4792                 if (arp_validate_value == -1) {
4793                         printk(KERN_ERR DRV_NAME
4794                                ": Error: invalid arp_validate \"%s\"\n",
4795                                arp_validate == NULL ? "NULL" : arp_validate);
4796                         return -EINVAL;
4797                 }
4798         } else
4799                 arp_validate_value = 0;
4800
4801         if (miimon) {
4802                 printk(KERN_INFO DRV_NAME
4803                        ": MII link monitoring set to %d ms\n",
4804                        miimon);
4805         } else if (arp_interval) {
4806                 int i;
4807
4808                 printk(KERN_INFO DRV_NAME
4809                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4810                        arp_interval,
4811                        arp_validate_tbl[arp_validate_value].modename,
4812                        arp_ip_count);
4813
4814                 for (i = 0; i < arp_ip_count; i++)
4815                         printk (" %s", arp_ip_target[i]);
4816
4817                 printk("\n");
4818
4819         } else {
4820                 /* miimon and arp_interval not set, we need one so things
4821                  * work as expected, see bonding.txt for details
4822                  */
4823                 printk(KERN_WARNING DRV_NAME
4824                        ": Warning: either miimon or arp_interval and "
4825                        "arp_ip_target module parameters must be specified, "
4826                        "otherwise bonding will not detect link failures! see "
4827                        "bonding.txt for details.\n");
4828         }
4829
4830         if (primary && !USES_PRIMARY(bond_mode)) {
4831                 /* currently, using a primary only makes sense
4832                  * in active backup, TLB or ALB modes
4833                  */
4834                 printk(KERN_WARNING DRV_NAME
4835                        ": Warning: %s primary device specified but has no "
4836                        "effect in %s mode\n",
4837                        primary, bond_mode_name(bond_mode));
4838                 primary = NULL;
4839         }
4840
4841         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4842                 printk(KERN_WARNING DRV_NAME
4843                        ": Warning: fail_over_mac only affects "
4844                        "active-backup mode.\n");
4845
4846         /* fill params struct with the proper values */
4847         params->mode = bond_mode;
4848         params->xmit_policy = xmit_hashtype;
4849         params->miimon = miimon;
4850         params->arp_interval = arp_interval;
4851         params->arp_validate = arp_validate_value;
4852         params->updelay = updelay;
4853         params->downdelay = downdelay;
4854         params->use_carrier = use_carrier;
4855         params->lacp_fast = lacp_fast;
4856         params->primary[0] = 0;
4857         params->fail_over_mac = fail_over_mac;
4858
4859         if (primary) {
4860                 strncpy(params->primary, primary, IFNAMSIZ);
4861                 params->primary[IFNAMSIZ - 1] = 0;
4862         }
4863
4864         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4865
4866         return 0;
4867 }
4868
4869 static struct lock_class_key bonding_netdev_xmit_lock_key;
4870
4871 /* Create a new bond based on the specified name and bonding parameters.
4872  * If name is NULL, obtain a suitable "bond%d" name for us.
4873  * Caller must NOT hold rtnl_lock; we need to release it here before we
4874  * set up our sysfs entries.
4875  */
4876 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4877 {
4878         struct net_device *bond_dev;
4879         struct bonding *bond, *nxt;
4880         int res;
4881
4882         rtnl_lock();
4883         down_write(&bonding_rwsem);
4884
4885         /* Check to see if the bond already exists. */
4886         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list)
4887                 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
4888                         printk(KERN_ERR DRV_NAME
4889                                ": cannot add bond %s; it already exists\n",
4890                                name);
4891                         res = -EPERM;
4892                         goto out_rtnl;
4893                 }
4894
4895         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4896                                 ether_setup);
4897         if (!bond_dev) {
4898                 printk(KERN_ERR DRV_NAME
4899                        ": %s: eek! can't alloc netdev!\n",
4900                        name);
4901                 res = -ENOMEM;
4902                 goto out_rtnl;
4903         }
4904
4905         if (!name) {
4906                 res = dev_alloc_name(bond_dev, "bond%d");
4907                 if (res < 0)
4908                         goto out_netdev;
4909         }
4910
4911         /* bond_init() must be called after dev_alloc_name() (for the
4912          * /proc files), but before register_netdevice(), because we
4913          * need to set function pointers.
4914          */
4915
4916         res = bond_init(bond_dev, params);
4917         if (res < 0) {
4918                 goto out_netdev;
4919         }
4920
4921         res = register_netdevice(bond_dev);
4922         if (res < 0) {
4923                 goto out_bond;
4924         }
4925
4926         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4927
4928         if (newbond)
4929                 *newbond = bond_dev->priv;
4930
4931         netif_carrier_off(bond_dev);
4932
4933         up_write(&bonding_rwsem);
4934         rtnl_unlock(); /* allows sysfs registration of net device */
4935         res = bond_create_sysfs_entry(bond_dev->priv);
4936         if (res < 0) {
4937                 rtnl_lock();
4938                 down_write(&bonding_rwsem);
4939                 goto out_bond;
4940         }
4941
4942         return 0;
4943
4944 out_bond:
4945         bond_deinit(bond_dev);
4946 out_netdev:
4947         free_netdev(bond_dev);
4948 out_rtnl:
4949         up_write(&bonding_rwsem);
4950         rtnl_unlock();
4951         return res;
4952 }
4953
4954 static int __init bonding_init(void)
4955 {
4956         int i;
4957         int res;
4958         struct bonding *bond, *nxt;
4959
4960         printk(KERN_INFO "%s", version);
4961
4962         res = bond_check_params(&bonding_defaults);
4963         if (res) {
4964                 goto out;
4965         }
4966
4967 #ifdef CONFIG_PROC_FS
4968         bond_create_proc_dir();
4969 #endif
4970
4971         init_rwsem(&bonding_rwsem);
4972
4973         for (i = 0; i < max_bonds; i++) {
4974                 res = bond_create(NULL, &bonding_defaults, NULL);
4975                 if (res)
4976                         goto err;
4977         }
4978
4979         res = bond_create_sysfs();
4980         if (res)
4981                 goto err;
4982
4983         register_netdevice_notifier(&bond_netdev_notifier);
4984         register_inetaddr_notifier(&bond_inetaddr_notifier);
4985
4986         goto out;
4987 err:
4988         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4989                 bond_work_cancel_all(bond);
4990                 destroy_workqueue(bond->wq);
4991         }
4992
4993         rtnl_lock();
4994         bond_free_all();
4995         bond_destroy_sysfs();
4996         rtnl_unlock();
4997 out:
4998         return res;
4999
5000 }
5001
5002 static void __exit bonding_exit(void)
5003 {
5004         unregister_netdevice_notifier(&bond_netdev_notifier);
5005         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5006
5007         rtnl_lock();
5008         bond_free_all();
5009         bond_destroy_sysfs();
5010         rtnl_unlock();
5011 }
5012
5013 module_init(bonding_init);
5014 module_exit(bonding_exit);
5015 MODULE_LICENSE("GPL");
5016 MODULE_VERSION(DRV_VERSION);
5017 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5018 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5019 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5020
5021 /*
5022  * Local variables:
5023  *  c-indent-level: 8
5024  *  c-basic-offset: 8
5025  *  tab-width: 8
5026  * End:
5027  */
5028