drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 //#define BONDING_DEBUG 1
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <asm/system.h>
  57 #include <asm/io.h>
  58 #include <asm/dma.h>
  59 #include <asm/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include "bonding.h"
  80 #include "bond_3ad.h"
  81 #include "bond_alb.h"
  82
  83 /*---------------------------- Module parameters ----------------------------*/
  84
  85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  86 #define BOND_LINK_MON_INTERV    0
  87 #define BOND_LINK_ARP_INTERV    0
  88
  89 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  90 static int miimon       = BOND_LINK_MON_INTERV;
  91 static int updelay      = 0;
  92 static int downdelay    = 0;
  93 static int use_carrier  = 1;
  94 static char *mode       = NULL;
  95 static char *primary    = NULL;
  96 static char *lacp_rate  = NULL;
  97 static char *xmit_hash_policy = NULL;
  98 static int arp_interval = BOND_LINK_ARP_INTERV;
  99 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 100 static char *arp_validate = NULL;
 101 static int fail_over_mac = 0;
 102 struct bond_params bonding_defaults;
 103
 104 module_param(max_bonds, int, 0);
 105 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 106 module_param(miimon, int, 0);
 107 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 108 module_param(updelay, int, 0);
 109 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 110 module_param(downdelay, int, 0);
 111 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 112                             "in milliseconds");
 113 module_param(use_carrier, int, 0);
 114 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 115                               "0 for off, 1 for on (default)");
 116 module_param(mode, charp, 0);
 117 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 118                        "1 for active-backup, 2 for balance-xor, "
 119                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 120                        "6 for balance-alb");
 121 module_param(primary, charp, 0);
 122 MODULE_PARM_DESC(primary, "Primary network device to use");
 123 module_param(lacp_rate, charp, 0);
 124 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 125                             "(slow/fast)");
 126 module_param(xmit_hash_policy, charp, 0);
 127 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 128                                    ", 1 for layer 3+4");
 129 module_param(arp_interval, int, 0);
 130 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 131 module_param_array(arp_ip_target, charp, NULL, 0);
 132 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 133 module_param(arp_validate, charp, 0);
 134 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 135 module_param(fail_over_mac, int, 0);
 136 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.");
 137
 138 /*----------------------------- Global variables ----------------------------*/
 139
 140 static const char * const version =
 141         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 142
 143 LIST_HEAD(bond_dev_list);
 144
 145 #ifdef CONFIG_PROC_FS
 146 static struct proc_dir_entry *bond_proc_dir = NULL;
 147 #endif
 148
 149 extern struct rw_semaphore bonding_rwsem;
 150 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 151 static int arp_ip_count = 0;
 152 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 153 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 154 static int lacp_fast    = 0;
 155
 156
 157 struct bond_parm_tbl bond_lacp_tbl[] = {
 158 {       "slow",         AD_LACP_SLOW},
 159 {       "fast",         AD_LACP_FAST},
 160 {       NULL,           -1},
 161 };
 162
 163 struct bond_parm_tbl bond_mode_tbl[] = {
 164 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 165 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 166 {       "balance-xor",          BOND_MODE_XOR},
 167 {       "broadcast",            BOND_MODE_BROADCAST},
 168 {       "802.3ad",              BOND_MODE_8023AD},
 169 {       "balance-tlb",          BOND_MODE_TLB},
 170 {       "balance-alb",          BOND_MODE_ALB},
 171 {       NULL,                   -1},
 172 };
 173
 174 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 175 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 176 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 177 {       NULL,                   -1},
 178 };
 179
 180 struct bond_parm_tbl arp_validate_tbl[] = {
 181 {       "none",                 BOND_ARP_VALIDATE_NONE},
 182 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 183 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 184 {       "all",                  BOND_ARP_VALIDATE_ALL},
 185 {       NULL,                   -1},
 186 };
 187
 188 /*-------------------------- Forward declarations ---------------------------*/
 189
 190 static void bond_send_gratuitous_arp(struct bonding *bond);
 191
 192 /*---------------------------- General routines -----------------------------*/
 193
 194 static const char *bond_mode_name(int mode)
 195 {
 196         switch (mode) {
 197         case BOND_MODE_ROUNDROBIN :
 198                 return "load balancing (round-robin)";
 199         case BOND_MODE_ACTIVEBACKUP :
 200                 return "fault-tolerance (active-backup)";
 201         case BOND_MODE_XOR :
 202                 return "load balancing (xor)";
 203         case BOND_MODE_BROADCAST :
 204                 return "fault-tolerance (broadcast)";
 205         case BOND_MODE_8023AD:
 206                 return "IEEE 802.3ad Dynamic link aggregation";
 207         case BOND_MODE_TLB:
 208                 return "transmit load balancing";
 209         case BOND_MODE_ALB:
 210                 return "adaptive load balancing";
 211         default:
 212                 return "unknown";
 213         }
 214 }
 215
 216 /*---------------------------------- VLAN -----------------------------------*/
 217
 218 /**
 219  * bond_add_vlan - add a new vlan id on bond
 220  * @bond: bond that got the notification
 221  * @vlan_id: the vlan id to add
 222  *
 223  * Returns -ENOMEM if allocation failed.
 224  */
 225 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 226 {
 227         struct vlan_entry *vlan;
 228
 229         dprintk("bond: %s, vlan id %d\n",
 230                 (bond ? bond->dev->name: "None"), vlan_id);
 231
 232         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 233         if (!vlan) {
 234                 return -ENOMEM;
 235         }
 236
 237         INIT_LIST_HEAD(&vlan->vlan_list);
 238         vlan->vlan_id = vlan_id;
 239         vlan->vlan_ip = 0;
 240
 241         write_lock_bh(&bond->lock);
 242
 243         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 244
 245         write_unlock_bh(&bond->lock);
 246
 247         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 248
 249         return 0;
 250 }
 251
 252 /**
 253  * bond_del_vlan - delete a vlan id from bond
 254  * @bond: bond that got the notification
 255  * @vlan_id: the vlan id to delete
 256  *
 257  * returns -ENODEV if @vlan_id was not found in @bond.
 258  */
 259 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 260 {
 261         struct vlan_entry *vlan, *next;
 262         int res = -ENODEV;
 263
 264         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 265
 266         write_lock_bh(&bond->lock);
 267
 268         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
 269                 if (vlan->vlan_id == vlan_id) {
 270                         list_del(&vlan->vlan_list);
 271
 272                         if ((bond->params.mode == BOND_MODE_TLB) ||
 273                             (bond->params.mode == BOND_MODE_ALB)) {
 274                                 bond_alb_clear_vlan(bond, vlan_id);
 275                         }
 276
 277                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 278                                 bond->dev->name);
 279
 280                         kfree(vlan);
 281
 282                         if (list_empty(&bond->vlan_list) &&
 283                             (bond->slave_cnt == 0)) {
 284                                 /* Last VLAN removed and no slaves, so
 285                                  * restore block on adding VLANs. This will
 286                                  * be removed once new slaves that are not
 287                                  * VLAN challenged will be added.
 288                                  */
 289                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 290                         }
 291
 292                         res = 0;
 293                         goto out;
 294                 }
 295         }
 296
 297         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 298                 bond->dev->name);
 299
 300 out:
 301         write_unlock_bh(&bond->lock);
 302         return res;
 303 }
 304
 305 /**
 306  * bond_has_challenged_slaves
 307  * @bond: the bond we're working on
 308  *
 309  * Searches the slave list. Returns 1 if a vlan challenged slave
 310  * was found, 0 otherwise.
 311  *
 312  * Assumes bond->lock is held.
 313  */
 314 static int bond_has_challenged_slaves(struct bonding *bond)
 315 {
 316         struct slave *slave;
 317         int i;
 318
 319         bond_for_each_slave(bond, slave, i) {
 320                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 321                         dprintk("found VLAN challenged slave - %s\n",
 322                                 slave->dev->name);
 323                         return 1;
 324                 }
 325         }
 326
 327         dprintk("no VLAN challenged slaves found\n");
 328         return 0;
 329 }
 330
 331 /**
 332  * bond_next_vlan - safely skip to the next item in the vlans list.
 333  * @bond: the bond we're working on
 334  * @curr: item we're advancing from
 335  *
 336  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 337  * or @curr->next otherwise (even if it is @curr itself again).
 338  *
 339  * Caller must hold bond->lock
 340  */
 341 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 342 {
 343         struct vlan_entry *next, *last;
 344
 345         if (list_empty(&bond->vlan_list)) {
 346                 return NULL;
 347         }
 348
 349         if (!curr) {
 350                 next = list_entry(bond->vlan_list.next,
 351                                   struct vlan_entry, vlan_list);
 352         } else {
 353                 last = list_entry(bond->vlan_list.prev,
 354                                   struct vlan_entry, vlan_list);
 355                 if (last == curr) {
 356                         next = list_entry(bond->vlan_list.next,
 357                                           struct vlan_entry, vlan_list);
 358                 } else {
 359                         next = list_entry(curr->vlan_list.next,
 360                                           struct vlan_entry, vlan_list);
 361                 }
 362         }
 363
 364         return next;
 365 }
 366
 367 /**
 368  * bond_dev_queue_xmit - Prepare skb for xmit.
 369  *
 370  * @bond: bond device that got this skb for tx.
 371  * @skb: hw accel VLAN tagged skb to transmit
 372  * @slave_dev: slave that is supposed to xmit this skbuff
 373  *
 374  * When the bond gets an skb to transmit that is
 375  * already hardware accelerated VLAN tagged, and it
 376  * needs to relay this skb to a slave that is not
 377  * hw accel capable, the skb needs to be "unaccelerated",
 378  * i.e. strip the hwaccel tag and re-insert it as part
 379  * of the payload.
 380  */
 381 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 382 {
 383         unsigned short vlan_id;
 384
 385         if (!list_empty(&bond->vlan_list) &&
 386             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 387             vlan_get_tag(skb, &vlan_id) == 0) {
 388                 skb->dev = slave_dev;
 389                 skb = vlan_put_tag(skb, vlan_id);
 390                 if (!skb) {
 391                         /* vlan_put_tag() frees the skb in case of error,
 392                          * so return success here so the calling functions
 393                          * won't attempt to free is again.
 394                          */
 395                         return 0;
 396                 }
 397         } else {
 398                 skb->dev = slave_dev;
 399         }
 400
 401         skb->priority = 1;
 402         dev_queue_xmit(skb);
 403
 404         return 0;
 405 }
 406
 407 /*
 408  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 409  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 410  * lock because:
 411  * a. This operation is performed in IOCTL context,
 412  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 413  * c. Holding a lock with BH disabled while directly calling a base driver
 414  *    entry point is generally a BAD idea.
 415  *
 416  * The design of synchronization/protection for this operation in the 8021q
 417  * module is good for one or more VLAN devices over a single physical device
 418  * and cannot be extended for a teaming solution like bonding, so there is a
 419  * potential race condition here where a net device from the vlan group might
 420  * be referenced (either by a base driver or the 8021q code) while it is being
 421  * removed from the system. However, it turns out we're not making matters
 422  * worse, and if it works for regular VLAN usage it will work here too.
 423 */
 424
 425 /**
 426  * bond_vlan_rx_register - Propagates registration to slaves
 427  * @bond_dev: bonding net device that got called
 428  * @grp: vlan group being registered
 429  */
 430 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 431 {
 432         struct bonding *bond = bond_dev->priv;
 433         struct slave *slave;
 434         int i;
 435
 436         bond->vlgrp = grp;
 437
 438         bond_for_each_slave(bond, slave, i) {
 439                 struct net_device *slave_dev = slave->dev;
 440
 441                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 442                     slave_dev->vlan_rx_register) {
 443                         slave_dev->vlan_rx_register(slave_dev, grp);
 444                 }
 445         }
 446 }
 447
 448 /**
 449  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 450  * @bond_dev: bonding net device that got called
 451  * @vid: vlan id being added
 452  */
 453 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 454 {
 455         struct bonding *bond = bond_dev->priv;
 456         struct slave *slave;
 457         int i, res;
 458
 459         bond_for_each_slave(bond, slave, i) {
 460                 struct net_device *slave_dev = slave->dev;
 461
 462                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 463                     slave_dev->vlan_rx_add_vid) {
 464                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 465                 }
 466         }
 467
 468         res = bond_add_vlan(bond, vid);
 469         if (res) {
 470                 printk(KERN_ERR DRV_NAME
 471                        ": %s: Error: Failed to add vlan id %d\n",
 472                        bond_dev->name, vid);
 473         }
 474 }
 475
 476 /**
 477  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 478  * @bond_dev: bonding net device that got called
 479  * @vid: vlan id being removed
 480  */
 481 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 482 {
 483         struct bonding *bond = bond_dev->priv;
 484         struct slave *slave;
 485         struct net_device *vlan_dev;
 486         int i, res;
 487
 488         bond_for_each_slave(bond, slave, i) {
 489                 struct net_device *slave_dev = slave->dev;
 490
 491                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 492                     slave_dev->vlan_rx_kill_vid) {
 493                         /* Save and then restore vlan_dev in the grp array,
 494                          * since the slave's driver might clear it.
 495                          */
 496                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 497                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 498                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 499                 }
 500         }
 501
 502         res = bond_del_vlan(bond, vid);
 503         if (res) {
 504                 printk(KERN_ERR DRV_NAME
 505                        ": %s: Error: Failed to remove vlan id %d\n",
 506                        bond_dev->name, vid);
 507         }
 508 }
 509
 510 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 511 {
 512         struct vlan_entry *vlan;
 513
 514         write_lock_bh(&bond->lock);
 515
 516         if (list_empty(&bond->vlan_list)) {
 517                 goto out;
 518         }
 519
 520         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 521             slave_dev->vlan_rx_register) {
 522                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 523         }
 524
 525         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 526             !(slave_dev->vlan_rx_add_vid)) {
 527                 goto out;
 528         }
 529
 530         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 531                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 532         }
 533
 534 out:
 535         write_unlock_bh(&bond->lock);
 536 }
 537
 538 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 539 {
 540         struct vlan_entry *vlan;
 541         struct net_device *vlan_dev;
 542
 543         write_lock_bh(&bond->lock);
 544
 545         if (list_empty(&bond->vlan_list)) {
 546                 goto out;
 547         }
 548
 549         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 550             !(slave_dev->vlan_rx_kill_vid)) {
 551                 goto unreg;
 552         }
 553
 554         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 555                 /* Save and then restore vlan_dev in the grp array,
 556                  * since the slave's driver might clear it.
 557                  */
 558                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 559                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 560                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 561         }
 562
 563 unreg:
 564         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 565             slave_dev->vlan_rx_register) {
 566                 slave_dev->vlan_rx_register(slave_dev, NULL);
 567         }
 568
 569 out:
 570         write_unlock_bh(&bond->lock);
 571 }
 572
 573 /*------------------------------- Link status -------------------------------*/
 574
 575 /*
 576  * Set the carrier state for the master according to the state of its
 577  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 578  * do special 802.3ad magic.
 579  *
 580  * Returns zero if carrier state does not change, nonzero if it does.
 581  */
 582 static int bond_set_carrier(struct bonding *bond)
 583 {
 584         struct slave *slave;
 585         int i;
 586
 587         if (bond->slave_cnt == 0)
 588                 goto down;
 589
 590         if (bond->params.mode == BOND_MODE_8023AD)
 591                 return bond_3ad_set_carrier(bond);
 592
 593         bond_for_each_slave(bond, slave, i) {
 594                 if (slave->link == BOND_LINK_UP) {
 595                         if (!netif_carrier_ok(bond->dev)) {
 596                                 netif_carrier_on(bond->dev);
 597                                 return 1;
 598                         }
 599                         return 0;
 600                 }
 601         }
 602
 603 down:
 604         if (netif_carrier_ok(bond->dev)) {
 605                 netif_carrier_off(bond->dev);
 606                 return 1;
 607         }
 608         return 0;
 609 }
 610
 611 /*
 612  * Get link speed and duplex from the slave's base driver
 613  * using ethtool. If for some reason the call fails or the
 614  * values are invalid, fake speed and duplex to 100/Full
 615  * and return error.
 616  */
 617 static int bond_update_speed_duplex(struct slave *slave)
 618 {
 619         struct net_device *slave_dev = slave->dev;
 620         struct ethtool_cmd etool;
 621         int res;
 622
 623         /* Fake speed and duplex */
 624         slave->speed = SPEED_100;
 625         slave->duplex = DUPLEX_FULL;
 626
 627         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 628                 return -1;
 629
 630         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 631         if (res < 0)
 632                 return -1;
 633
 634         switch (etool.speed) {
 635         case SPEED_10:
 636         case SPEED_100:
 637         case SPEED_1000:
 638         case SPEED_10000:
 639                 break;
 640         default:
 641                 return -1;
 642         }
 643
 644         switch (etool.duplex) {
 645         case DUPLEX_FULL:
 646         case DUPLEX_HALF:
 647                 break;
 648         default:
 649                 return -1;
 650         }
 651
 652         slave->speed = etool.speed;
 653         slave->duplex = etool.duplex;
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * if <dev> supports MII link status reporting, check its link status.
 660  *
 661  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 662  * depening upon the setting of the use_carrier parameter.
 663  *
 664  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 665  * can't tell and just pretend it is), or 0, meaning that the link is
 666  * down.
 667  *
 668  * If reporting is non-zero, instead of faking link up, return -1 if
 669  * both ETHTOOL and MII ioctls fail (meaning the device does not
 670  * support them).  If use_carrier is set, return whatever it says.
 671  * It'd be nice if there was a good way to tell if a driver supports
 672  * netif_carrier, but there really isn't.
 673  */
 674 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 675 {
 676         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 677         struct ifreq ifr;
 678         struct mii_ioctl_data *mii;
 679
 680         if (bond->params.use_carrier) {
 681                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 682         }
 683
 684         ioctl = slave_dev->do_ioctl;
 685         if (ioctl) {
 686                 /* TODO: set pointer to correct ioctl on a per team member */
 687                 /*       bases to make this more efficient. that is, once  */
 688                 /*       we determine the correct ioctl, we will always    */
 689                 /*       call it and not the others for that team          */
 690                 /*       member.                                           */
 691
 692                 /*
 693                  * We cannot assume that SIOCGMIIPHY will also read a
 694                  * register; not all network drivers (e.g., e100)
 695                  * support that.
 696                  */
 697
 698                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 699                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 700                 mii = if_mii(&ifr);
 701                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 702                         mii->reg_num = MII_BMSR;
 703                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 704                                 return (mii->val_out & BMSR_LSTATUS);
 705                         }
 706                 }
 707         }
 708
 709         /*
 710          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 711          * attempt to get link status from it if the above MII ioctls fail.
 712          */
 713         if (slave_dev->ethtool_ops) {
 714                 if (slave_dev->ethtool_ops->get_link) {
 715                         u32 link;
 716
 717                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 718
 719                         return link ? BMSR_LSTATUS : 0;
 720                 }
 721         }
 722
 723         /*
 724          * If reporting, report that either there's no dev->do_ioctl,
 725          * or both SIOCGMIIREG and get_link failed (meaning that we
 726          * cannot report link status).  If not reporting, pretend
 727          * we're ok.
 728          */
 729         return (reporting ? -1 : BMSR_LSTATUS);
 730 }
 731
 732 /*----------------------------- Multicast list ------------------------------*/
 733
 734 /*
 735  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 736  */
 737 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 738 {
 739         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 740                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 741 }
 742
 743 /*
 744  * returns dmi entry if found, NULL otherwise
 745  */
 746 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 747 {
 748         struct dev_mc_list *idmi;
 749
 750         for (idmi = mc_list; idmi; idmi = idmi->next) {
 751                 if (bond_is_dmi_same(dmi, idmi)) {
 752                         return idmi;
 753                 }
 754         }
 755
 756         return NULL;
 757 }
 758
 759 /*
 760  * Push the promiscuity flag down to appropriate slaves
 761  */
 762 static void bond_set_promiscuity(struct bonding *bond, int inc)
 763 {
 764         if (USES_PRIMARY(bond->params.mode)) {
 765                 /* write lock already acquired */
 766                 if (bond->curr_active_slave) {
 767                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
 768                 }
 769         } else {
 770                 struct slave *slave;
 771                 int i;
 772                 bond_for_each_slave(bond, slave, i) {
 773                         dev_set_promiscuity(slave->dev, inc);
 774                 }
 775         }
 776 }
 777
 778 /*
 779  * Push the allmulti flag down to all slaves
 780  */
 781 static void bond_set_allmulti(struct bonding *bond, int inc)
 782 {
 783         if (USES_PRIMARY(bond->params.mode)) {
 784                 /* write lock already acquired */
 785                 if (bond->curr_active_slave) {
 786                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
 787                 }
 788         } else {
 789                 struct slave *slave;
 790                 int i;
 791                 bond_for_each_slave(bond, slave, i) {
 792                         dev_set_allmulti(slave->dev, inc);
 793                 }
 794         }
 795 }
 796
 797 /*
 798  * Add a Multicast address to slaves
 799  * according to mode
 800  */
 801 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 802 {
 803         if (USES_PRIMARY(bond->params.mode)) {
 804                 /* write lock already acquired */
 805                 if (bond->curr_active_slave) {
 806                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 807                 }
 808         } else {
 809                 struct slave *slave;
 810                 int i;
 811                 bond_for_each_slave(bond, slave, i) {
 812                         dev_mc_add(slave->dev, addr, alen, 0);
 813                 }
 814         }
 815 }
 816
 817 /*
 818  * Remove a multicast address from slave
 819  * according to mode
 820  */
 821 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 822 {
 823         if (USES_PRIMARY(bond->params.mode)) {
 824                 /* write lock already acquired */
 825                 if (bond->curr_active_slave) {
 826                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 827                 }
 828         } else {
 829                 struct slave *slave;
 830                 int i;
 831                 bond_for_each_slave(bond, slave, i) {
 832                         dev_mc_delete(slave->dev, addr, alen, 0);
 833                 }
 834         }
 835 }
 836
 837
 838 /*
 839  * Retrieve the list of registered multicast addresses for the bonding
 840  * device and retransmit an IGMP JOIN request to the current active
 841  * slave.
 842  */
 843 static void bond_resend_igmp_join_requests(struct bonding *bond)
 844 {
 845         struct in_device *in_dev;
 846         struct ip_mc_list *im;
 847
 848         rcu_read_lock();
 849         in_dev = __in_dev_get_rcu(bond->dev);
 850         if (in_dev) {
 851                 for (im = in_dev->mc_list; im; im = im->next) {
 852                         ip_mc_rejoin_group(im);
 853                 }
 854         }
 855
 856         rcu_read_unlock();
 857 }
 858
 859 /*
 860  * Totally destroys the mc_list in bond
 861  */
 862 static void bond_mc_list_destroy(struct bonding *bond)
 863 {
 864         struct dev_mc_list *dmi;
 865
 866         dmi = bond->mc_list;
 867         while (dmi) {
 868                 bond->mc_list = dmi->next;
 869                 kfree(dmi);
 870                 dmi = bond->mc_list;
 871         }
 872         bond->mc_list = NULL;
 873 }
 874
 875 /*
 876  * Copy all the Multicast addresses from src to the bonding device dst
 877  */
 878 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 879                              gfp_t gfp_flag)
 880 {
 881         struct dev_mc_list *dmi, *new_dmi;
 882
 883         for (dmi = mc_list; dmi; dmi = dmi->next) {
 884                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 885
 886                 if (!new_dmi) {
 887                         /* FIXME: Potential memory leak !!! */
 888                         return -ENOMEM;
 889                 }
 890
 891                 new_dmi->next = bond->mc_list;
 892                 bond->mc_list = new_dmi;
 893                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 894                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 895                 new_dmi->dmi_users = dmi->dmi_users;
 896                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 897         }
 898
 899         return 0;
 900 }
 901
 902 /*
 903  * flush all members of flush->mc_list from device dev->mc_list
 904  */
 905 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 906 {
 907         struct bonding *bond = bond_dev->priv;
 908         struct dev_mc_list *dmi;
 909
 910         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 911                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 912         }
 913
 914         if (bond->params.mode == BOND_MODE_8023AD) {
 915                 /* del lacpdu mc addr from mc list */
 916                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 917
 918                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 919         }
 920 }
 921
 922 /*--------------------------- Active slave change ---------------------------*/
 923
 924 /*
 925  * Update the mc list and multicast-related flags for the new and
 926  * old active slaves (if any) according to the multicast mode, and
 927  * promiscuous flags unconditionally.
 928  */
 929 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 930 {
 931         struct dev_mc_list *dmi;
 932
 933         if (!USES_PRIMARY(bond->params.mode)) {
 934                 /* nothing to do -  mc list is already up-to-date on
 935                  * all slaves
 936                  */
 937                 return;
 938         }
 939
 940         if (old_active) {
 941                 if (bond->dev->flags & IFF_PROMISC) {
 942                         dev_set_promiscuity(old_active->dev, -1);
 943                 }
 944
 945                 if (bond->dev->flags & IFF_ALLMULTI) {
 946                         dev_set_allmulti(old_active->dev, -1);
 947                 }
 948
 949                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 950                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 951                 }
 952         }
 953
 954         if (new_active) {
 955                 if (bond->dev->flags & IFF_PROMISC) {
 956                         dev_set_promiscuity(new_active->dev, 1);
 957                 }
 958
 959                 if (bond->dev->flags & IFF_ALLMULTI) {
 960                         dev_set_allmulti(new_active->dev, 1);
 961                 }
 962
 963                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 964                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 965                 }
 966                 bond_resend_igmp_join_requests(bond);
 967         }
 968 }
 969
 970 /**
 971  * find_best_interface - select the best available slave to be the active one
 972  * @bond: our bonding struct
 973  *
 974  * Warning: Caller must hold curr_slave_lock for writing.
 975  */
 976 static struct slave *bond_find_best_slave(struct bonding *bond)
 977 {
 978         struct slave *new_active, *old_active;
 979         struct slave *bestslave = NULL;
 980         int mintime = bond->params.updelay;
 981         int i;
 982
 983         new_active = old_active = bond->curr_active_slave;
 984
 985         if (!new_active) { /* there were no active slaves left */
 986                 if (bond->slave_cnt > 0) {  /* found one slave */
 987                         new_active = bond->first_slave;
 988                 } else {
 989                         return NULL; /* still no slave, return NULL */
 990                 }
 991         }
 992
 993         /* first try the primary link; if arping, a link must tx/rx traffic
 994          * before it can be considered the curr_active_slave - also, we would skip
 995          * slaves between the curr_active_slave and primary_slave that may be up
 996          * and able to arp
 997          */
 998         if ((bond->primary_slave) &&
 999             (!bond->params.arp_interval) &&
1000             (IS_UP(bond->primary_slave->dev))) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (IS_UP(new_active->dev)) {
1009                         if (new_active->link == BOND_LINK_UP) {
1010                                 return new_active;
1011                         } else if (new_active->link == BOND_LINK_BACK) {
1012                                 /* link up, but waiting for stabilization */
1013                                 if (new_active->delay < mintime) {
1014                                         mintime = new_active->delay;
1015                                         bestslave = new_active;
1016                                 }
1017                         }
1018                 }
1019         }
1020
1021         return bestslave;
1022 }
1023
1024 /**
1025  * change_active_interface - change the active slave into the specified one
1026  * @bond: our bonding struct
1027  * @new: the new slave to make the active one
1028  *
1029  * Set the new slave to the bond's settings and unset them on the old
1030  * curr_active_slave.
1031  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1032  *
1033  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1034  * because it is apparently the best available slave we have, even though its
1035  * updelay hasn't timed out yet.
1036  *
1037  * Warning: Caller must hold curr_slave_lock for writing.
1038  */
1039 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1040 {
1041         struct slave *old_active = bond->curr_active_slave;
1042
1043         if (old_active == new_active) {
1044                 return;
1045         }
1046
1047         if (new_active) {
1048                 if (new_active->link == BOND_LINK_BACK) {
1049                         if (USES_PRIMARY(bond->params.mode)) {
1050                                 printk(KERN_INFO DRV_NAME
1051                                        ": %s: making interface %s the new "
1052                                        "active one %d ms earlier.\n",
1053                                        bond->dev->name, new_active->dev->name,
1054                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1055                         }
1056
1057                         new_active->delay = 0;
1058                         new_active->link = BOND_LINK_UP;
1059                         new_active->jiffies = jiffies;
1060
1061                         if (bond->params.mode == BOND_MODE_8023AD) {
1062                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1063                         }
1064
1065                         if ((bond->params.mode == BOND_MODE_TLB) ||
1066                             (bond->params.mode == BOND_MODE_ALB)) {
1067                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1068                         }
1069                 } else {
1070                         if (USES_PRIMARY(bond->params.mode)) {
1071                                 printk(KERN_INFO DRV_NAME
1072                                        ": %s: making interface %s the new "
1073                                        "active one.\n",
1074                                        bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode)) {
1080                 bond_mc_swap(bond, new_active, old_active);
1081         }
1082
1083         if ((bond->params.mode == BOND_MODE_TLB) ||
1084             (bond->params.mode == BOND_MODE_ALB)) {
1085                 bond_alb_handle_active_change(bond, new_active);
1086                 if (old_active)
1087                         bond_set_slave_inactive_flags(old_active);
1088                 if (new_active)
1089                         bond_set_slave_active_flags(new_active);
1090         } else {
1091                 bond->curr_active_slave = new_active;
1092         }
1093
1094         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1095                 if (old_active) {
1096                         bond_set_slave_inactive_flags(old_active);
1097                 }
1098
1099                 if (new_active) {
1100                         bond_set_slave_active_flags(new_active);
1101                 }
1102
1103                 /* when bonding does not set the slave MAC address, the bond MAC
1104                  * address is the one of the active slave.
1105                  */
1106                 if (new_active && bond->params.fail_over_mac)
1107                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1108                                 new_active->dev->addr_len);
1109                 if (bond->curr_active_slave &&
1110                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1111                                         &bond->curr_active_slave->dev->state)) {
1112                         dprintk("delaying gratuitous arp on %s\n",
1113                                 bond->curr_active_slave->dev->name);
1114                         bond->send_grat_arp = 1;
1115                 } else
1116                         bond_send_gratuitous_arp(bond);
1117         }
1118 }
1119
1120 /**
1121  * bond_select_active_slave - select a new active slave, if needed
1122  * @bond: our bonding struct
1123  *
1124  * This functions shoud be called when one of the following occurs:
1125  * - The old curr_active_slave has been released or lost its link.
1126  * - The primary_slave has got its link back.
1127  * - A slave has got its link back and there's no old curr_active_slave.
1128  *
1129  * Warning: Caller must hold curr_slave_lock for writing.
1130  */
1131 void bond_select_active_slave(struct bonding *bond)
1132 {
1133         struct slave *best_slave;
1134         int rv;
1135
1136         best_slave = bond_find_best_slave(bond);
1137         if (best_slave != bond->curr_active_slave) {
1138                 bond_change_active_slave(bond, best_slave);
1139                 rv = bond_set_carrier(bond);
1140                 if (!rv)
1141                         return;
1142
1143                 if (netif_carrier_ok(bond->dev)) {
1144                         printk(KERN_INFO DRV_NAME
1145                                ": %s: first active interface up!\n",
1146                                bond->dev->name);
1147                 } else {
1148                         printk(KERN_INFO DRV_NAME ": %s: "
1149                                "now running without any active interface !\n",
1150                                bond->dev->name);
1151                 }
1152         }
1153 }
1154
1155 /*--------------------------- slave list handling ---------------------------*/
1156
1157 /*
1158  * This function attaches the slave to the end of list.
1159  *
1160  * bond->lock held for writing by caller.
1161  */
1162 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1163 {
1164         if (bond->first_slave == NULL) { /* attaching the first slave */
1165                 new_slave->next = new_slave;
1166                 new_slave->prev = new_slave;
1167                 bond->first_slave = new_slave;
1168         } else {
1169                 new_slave->next = bond->first_slave;
1170                 new_slave->prev = bond->first_slave->prev;
1171                 new_slave->next->prev = new_slave;
1172                 new_slave->prev->next = new_slave;
1173         }
1174
1175         bond->slave_cnt++;
1176 }
1177
1178 /*
1179  * This function detaches the slave from the list.
1180  * WARNING: no check is made to verify if the slave effectively
1181  * belongs to <bond>.
1182  * Nothing is freed on return, structures are just unchained.
1183  * If any slave pointer in bond was pointing to <slave>,
1184  * it should be changed by the calling function.
1185  *
1186  * bond->lock held for writing by caller.
1187  */
1188 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1189 {
1190         if (slave->next) {
1191                 slave->next->prev = slave->prev;
1192         }
1193
1194         if (slave->prev) {
1195                 slave->prev->next = slave->next;
1196         }
1197
1198         if (bond->first_slave == slave) { /* slave is the first slave */
1199                 if (bond->slave_cnt > 1) { /* there are more slave */
1200                         bond->first_slave = slave->next;
1201                 } else {
1202                         bond->first_slave = NULL; /* slave was the last one */
1203                 }
1204         }
1205
1206         slave->next = NULL;
1207         slave->prev = NULL;
1208         bond->slave_cnt--;
1209 }
1210
1211 /*---------------------------------- IOCTL ----------------------------------*/
1212
1213 static int bond_sethwaddr(struct net_device *bond_dev,
1214                           struct net_device *slave_dev)
1215 {
1216         dprintk("bond_dev=%p\n", bond_dev);
1217         dprintk("slave_dev=%p\n", slave_dev);
1218         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1219         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1220         return 0;
1221 }
1222
1223 #define BOND_VLAN_FEATURES \
1224         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1225          NETIF_F_HW_VLAN_FILTER)
1226
1227 /*
1228  * Compute the common dev->feature set available to all slaves.  Some
1229  * feature bits are managed elsewhere, so preserve those feature bits
1230  * on the master device.
1231  */
1232 static int bond_compute_features(struct bonding *bond)
1233 {
1234         struct slave *slave;
1235         struct net_device *bond_dev = bond->dev;
1236         unsigned long features = bond_dev->features;
1237         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1238                                                 bond_dev->hard_header_len);
1239         int i;
1240
1241         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1242         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1243                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1244
1245         bond_for_each_slave(bond, slave, i) {
1246                 features = netdev_compute_features(features,
1247                                                    slave->dev->features);
1248                 if (slave->dev->hard_header_len > max_hard_header_len)
1249                         max_hard_header_len = slave->dev->hard_header_len;
1250         }
1251
1252         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1253         bond_dev->features = features;
1254         bond_dev->hard_header_len = max_hard_header_len;
1255
1256         return 0;
1257 }
1258
1259
1260 static void bond_setup_by_slave(struct net_device *bond_dev,
1261                                 struct net_device *slave_dev)
1262 {
1263         struct bonding *bond = bond_dev->priv;
1264
1265         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1266         bond_dev->header_ops            = slave_dev->header_ops;
1267
1268         bond_dev->type              = slave_dev->type;
1269         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1270         bond_dev->addr_len          = slave_dev->addr_len;
1271
1272         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1273                 slave_dev->addr_len);
1274         bond->setup_by_slave = 1;
1275 }
1276
1277 /* enslave device <slave> to bond device <master> */
1278 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1279 {
1280         struct bonding *bond = bond_dev->priv;
1281         struct slave *new_slave = NULL;
1282         struct dev_mc_list *dmi;
1283         struct sockaddr addr;
1284         int link_reporting;
1285         int old_features = bond_dev->features;
1286         int res = 0;
1287
1288         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1289                 slave_dev->do_ioctl == NULL) {
1290                 printk(KERN_WARNING DRV_NAME
1291                        ": %s: Warning: no link monitoring support for %s\n",
1292                        bond_dev->name, slave_dev->name);
1293         }
1294
1295         /* bond must be initialized by bond_open() before enslaving */
1296         if (!(bond_dev->flags & IFF_UP)) {
1297                 printk(KERN_WARNING DRV_NAME
1298                         " %s: master_dev is not up in bond_enslave\n",
1299                         bond_dev->name);
1300         }
1301
1302         /* already enslaved */
1303         if (slave_dev->flags & IFF_SLAVE) {
1304                 dprintk("Error, Device was already enslaved\n");
1305                 return -EBUSY;
1306         }
1307
1308         /* vlan challenged mutual exclusion */
1309         /* no need to lock since we're protected by rtnl_lock */
1310         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1311                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1312                 if (!list_empty(&bond->vlan_list)) {
1313                         printk(KERN_ERR DRV_NAME
1314                                ": %s: Error: cannot enslave VLAN "
1315                                "challenged slave %s on VLAN enabled "
1316                                "bond %s\n", bond_dev->name, slave_dev->name,
1317                                bond_dev->name);
1318                         return -EPERM;
1319                 } else {
1320                         printk(KERN_WARNING DRV_NAME
1321                                ": %s: Warning: enslaved VLAN challenged "
1322                                "slave %s. Adding VLANs will be blocked as "
1323                                "long as %s is part of bond %s\n",
1324                                bond_dev->name, slave_dev->name, slave_dev->name,
1325                                bond_dev->name);
1326                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1327                 }
1328         } else {
1329                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1330                 if (bond->slave_cnt == 0) {
1331                         /* First slave, and it is not VLAN challenged,
1332                          * so remove the block of adding VLANs over the bond.
1333                          */
1334                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1335                 }
1336         }
1337
1338         /*
1339          * Old ifenslave binaries are no longer supported.  These can
1340          * be identified with moderate accurary by the state of the slave:
1341          * the current ifenslave will set the interface down prior to
1342          * enslaving it; the old ifenslave will not.
1343          */
1344         if ((slave_dev->flags & IFF_UP)) {
1345                 printk(KERN_ERR DRV_NAME ": %s is up. "
1346                        "This may be due to an out of date ifenslave.\n",
1347                        slave_dev->name);
1348                 res = -EPERM;
1349                 goto err_undo_flags;
1350         }
1351
1352         /* set bonding device ether type by slave - bonding netdevices are
1353          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1354          * there is a need to override some of the type dependent attribs/funcs.
1355          *
1356          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1357          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1358          */
1359         if (bond->slave_cnt == 0) {
1360                 if (slave_dev->type != ARPHRD_ETHER)
1361                         bond_setup_by_slave(bond_dev, slave_dev);
1362         } else if (bond_dev->type != slave_dev->type) {
1363                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1364                         "from other slaves (%d), can not enslave it.\n",
1365                         slave_dev->name,
1366                         slave_dev->type, bond_dev->type);
1367                         res = -EINVAL;
1368                         goto err_undo_flags;
1369         }
1370
1371         if (slave_dev->set_mac_address == NULL) {
1372                 if (bond->slave_cnt == 0) {
1373                         printk(KERN_WARNING DRV_NAME
1374                                ": %s: Warning: The first slave device "
1375                                "specified does not support setting the MAC "
1376                                "address. Enabling the fail_over_mac option.",
1377                                bond_dev->name);
1378                         bond->params.fail_over_mac = 1;
1379                 } else if (!bond->params.fail_over_mac) {
1380                         printk(KERN_ERR DRV_NAME
1381                                 ": %s: Error: The slave device specified "
1382                                 "does not support setting the MAC address, "
1383                                 "but fail_over_mac is not enabled.\n"
1384                                 , bond_dev->name);
1385                         res = -EOPNOTSUPP;
1386                         goto err_undo_flags;
1387                 }
1388         }
1389
1390         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1391         if (!new_slave) {
1392                 res = -ENOMEM;
1393                 goto err_undo_flags;
1394         }
1395
1396         /* save slave's original flags before calling
1397          * netdev_set_master and dev_open
1398          */
1399         new_slave->original_flags = slave_dev->flags;
1400
1401         /*
1402          * Save slave's original ("permanent") mac address for modes
1403          * that need it, and for restoring it upon release, and then
1404          * set it to the master's address
1405          */
1406         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1407
1408         if (!bond->params.fail_over_mac) {
1409                 /*
1410                  * Set slave to master's mac address.  The application already
1411                  * set the master's mac address to that of the first slave
1412                  */
1413                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1414                 addr.sa_family = slave_dev->type;
1415                 res = dev_set_mac_address(slave_dev, &addr);
1416                 if (res) {
1417                         dprintk("Error %d calling set_mac_address\n", res);
1418                         goto err_free;
1419                 }
1420         }
1421
1422         res = netdev_set_master(slave_dev, bond_dev);
1423         if (res) {
1424                 dprintk("Error %d calling netdev_set_master\n", res);
1425                 goto err_close;
1426         }
1427         /* open the slave since the application closed it */
1428         res = dev_open(slave_dev);
1429         if (res) {
1430                 dprintk("Openning slave %s failed\n", slave_dev->name);
1431                 goto err_restore_mac;
1432         }
1433
1434         new_slave->dev = slave_dev;
1435         slave_dev->priv_flags |= IFF_BONDING;
1436
1437         if ((bond->params.mode == BOND_MODE_TLB) ||
1438             (bond->params.mode == BOND_MODE_ALB)) {
1439                 /* bond_alb_init_slave() must be called before all other stages since
1440                  * it might fail and we do not want to have to undo everything
1441                  */
1442                 res = bond_alb_init_slave(bond, new_slave);
1443                 if (res) {
1444                         goto err_unset_master;
1445                 }
1446         }
1447
1448         /* If the mode USES_PRIMARY, then the new slave gets the
1449          * master's promisc (and mc) settings only if it becomes the
1450          * curr_active_slave, and that is taken care of later when calling
1451          * bond_change_active()
1452          */
1453         if (!USES_PRIMARY(bond->params.mode)) {
1454                 /* set promiscuity level to new slave */
1455                 if (bond_dev->flags & IFF_PROMISC) {
1456                         dev_set_promiscuity(slave_dev, 1);
1457                 }
1458
1459                 /* set allmulti level to new slave */
1460                 if (bond_dev->flags & IFF_ALLMULTI) {
1461                         dev_set_allmulti(slave_dev, 1);
1462                 }
1463
1464                 /* upload master's mc_list to new slave */
1465                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1466                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1467                 }
1468         }
1469
1470         if (bond->params.mode == BOND_MODE_8023AD) {
1471                 /* add lacpdu mc addr to mc list */
1472                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1473
1474                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1475         }
1476
1477         bond_add_vlans_on_slave(bond, slave_dev);
1478
1479         write_lock_bh(&bond->lock);
1480
1481         bond_attach_slave(bond, new_slave);
1482
1483         new_slave->delay = 0;
1484         new_slave->link_failure_count = 0;
1485
1486         bond_compute_features(bond);
1487
1488         new_slave->last_arp_rx = jiffies;
1489
1490         if (bond->params.miimon && !bond->params.use_carrier) {
1491                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1492
1493                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1494                         /*
1495                          * miimon is set but a bonded network driver
1496                          * does not support ETHTOOL/MII and
1497                          * arp_interval is not set.  Note: if
1498                          * use_carrier is enabled, we will never go
1499                          * here (because netif_carrier is always
1500                          * supported); thus, we don't need to change
1501                          * the messages for netif_carrier.
1502                          */
1503                         printk(KERN_WARNING DRV_NAME
1504                                ": %s: Warning: MII and ETHTOOL support not "
1505                                "available for interface %s, and "
1506                                "arp_interval/arp_ip_target module parameters "
1507                                "not specified, thus bonding will not detect "
1508                                "link failures! see bonding.txt for details.\n",
1509                                bond_dev->name, slave_dev->name);
1510                 } else if (link_reporting == -1) {
1511                         /* unable get link status using mii/ethtool */
1512                         printk(KERN_WARNING DRV_NAME
1513                                ": %s: Warning: can't get link status from "
1514                                "interface %s; the network driver associated "
1515                                "with this interface does not support MII or "
1516                                "ETHTOOL link status reporting, thus miimon "
1517                                "has no effect on this interface.\n",
1518                                bond_dev->name, slave_dev->name);
1519                 }
1520         }
1521
1522         /* check for initial state */
1523         if (!bond->params.miimon ||
1524             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1525                 if (bond->params.updelay) {
1526                         dprintk("Initial state of slave_dev is "
1527                                 "BOND_LINK_BACK\n");
1528                         new_slave->link  = BOND_LINK_BACK;
1529                         new_slave->delay = bond->params.updelay;
1530                 } else {
1531                         dprintk("Initial state of slave_dev is "
1532                                 "BOND_LINK_UP\n");
1533                         new_slave->link  = BOND_LINK_UP;
1534                 }
1535                 new_slave->jiffies = jiffies;
1536         } else {
1537                 dprintk("Initial state of slave_dev is "
1538                         "BOND_LINK_DOWN\n");
1539                 new_slave->link  = BOND_LINK_DOWN;
1540         }
1541
1542         if (bond_update_speed_duplex(new_slave) &&
1543             (new_slave->link != BOND_LINK_DOWN)) {
1544                 printk(KERN_WARNING DRV_NAME
1545                        ": %s: Warning: failed to get speed and duplex from %s, "
1546                        "assumed to be 100Mb/sec and Full.\n",
1547                        bond_dev->name, new_slave->dev->name);
1548
1549                 if (bond->params.mode == BOND_MODE_8023AD) {
1550                         printk(KERN_WARNING DRV_NAME
1551                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1552                                "support in base driver for proper aggregator "
1553                                "selection.\n", bond_dev->name);
1554                 }
1555         }
1556
1557         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1558                 /* if there is a primary slave, remember it */
1559                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1560                         bond->primary_slave = new_slave;
1561                 }
1562         }
1563
1564         switch (bond->params.mode) {
1565         case BOND_MODE_ACTIVEBACKUP:
1566                 bond_set_slave_inactive_flags(new_slave);
1567                 bond_select_active_slave(bond);
1568                 break;
1569         case BOND_MODE_8023AD:
1570                 /* in 802.3ad mode, the internal mechanism
1571                  * will activate the slaves in the selected
1572                  * aggregator
1573                  */
1574                 bond_set_slave_inactive_flags(new_slave);
1575                 /* if this is the first slave */
1576                 if (bond->slave_cnt == 1) {
1577                         SLAVE_AD_INFO(new_slave).id = 1;
1578                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1579                          * can be called only after the mac address of the bond is set
1580                          */
1581                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1582                                             bond->params.lacp_fast);
1583                 } else {
1584                         SLAVE_AD_INFO(new_slave).id =
1585                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1586                 }
1587
1588                 bond_3ad_bind_slave(new_slave);
1589                 break;
1590         case BOND_MODE_TLB:
1591         case BOND_MODE_ALB:
1592                 new_slave->state = BOND_STATE_ACTIVE;
1593                 if ((!bond->curr_active_slave) &&
1594                     (new_slave->link != BOND_LINK_DOWN)) {
1595                         /* first slave or no active slave yet, and this link
1596                          * is OK, so make this interface the active one
1597                          */
1598                         bond_change_active_slave(bond, new_slave);
1599                 } else {
1600                         bond_set_slave_inactive_flags(new_slave);
1601                 }
1602                 break;
1603         default:
1604                 dprintk("This slave is always active in trunk mode\n");
1605
1606                 /* always active in trunk mode */
1607                 new_slave->state = BOND_STATE_ACTIVE;
1608
1609                 /* In trunking mode there is little meaning to curr_active_slave
1610                  * anyway (it holds no special properties of the bond device),
1611                  * so we can change it without calling change_active_interface()
1612                  */
1613                 if (!bond->curr_active_slave) {
1614                         bond->curr_active_slave = new_slave;
1615                 }
1616                 break;
1617         } /* switch(bond_mode) */
1618
1619         bond_set_carrier(bond);
1620
1621         write_unlock_bh(&bond->lock);
1622
1623         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1624         if (res)
1625                 goto err_unset_master;
1626
1627         printk(KERN_INFO DRV_NAME
1628                ": %s: enslaving %s as a%s interface with a%s link.\n",
1629                bond_dev->name, slave_dev->name,
1630                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1631                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1632
1633         /* enslave is successful */
1634         return 0;
1635
1636 /* Undo stages on error */
1637 err_unset_master:
1638         netdev_set_master(slave_dev, NULL);
1639
1640 err_close:
1641         dev_close(slave_dev);
1642
1643 err_restore_mac:
1644         if (!bond->params.fail_over_mac) {
1645                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1646                 addr.sa_family = slave_dev->type;
1647                 dev_set_mac_address(slave_dev, &addr);
1648         }
1649
1650 err_free:
1651         kfree(new_slave);
1652
1653 err_undo_flags:
1654         bond_dev->features = old_features;
1655
1656         return res;
1657 }
1658
1659 /*
1660  * Try to release the slave device <slave> from the bond device <master>
1661  * It is legal to access curr_active_slave without a lock because all the function
1662  * is write-locked.
1663  *
1664  * The rules for slave state should be:
1665  *   for Active/Backup:
1666  *     Active stays on all backups go down
1667  *   for Bonded connections:
1668  *     The first up interface should be left on and all others downed.
1669  */
1670 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1671 {
1672         struct bonding *bond = bond_dev->priv;
1673         struct slave *slave, *oldcurrent;
1674         struct sockaddr addr;
1675         int mac_addr_differ;
1676         DECLARE_MAC_BUF(mac);
1677
1678         /* slave is not a slave or master is not master of this slave */
1679         if (!(slave_dev->flags & IFF_SLAVE) ||
1680             (slave_dev->master != bond_dev)) {
1681                 printk(KERN_ERR DRV_NAME
1682                        ": %s: Error: cannot release %s.\n",
1683                        bond_dev->name, slave_dev->name);
1684                 return -EINVAL;
1685         }
1686
1687         write_lock_bh(&bond->lock);
1688
1689         slave = bond_get_slave_by_dev(bond, slave_dev);
1690         if (!slave) {
1691                 /* not a slave of this bond */
1692                 printk(KERN_INFO DRV_NAME
1693                        ": %s: %s not enslaved\n",
1694                        bond_dev->name, slave_dev->name);
1695                 write_unlock_bh(&bond->lock);
1696                 return -EINVAL;
1697         }
1698
1699         mac_addr_differ = memcmp(bond_dev->dev_addr,
1700                                  slave->perm_hwaddr,
1701                                  ETH_ALEN);
1702         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1703                 printk(KERN_WARNING DRV_NAME
1704                        ": %s: Warning: the permanent HWaddr of %s - "
1705                        "%s - is still in use by %s. "
1706                        "Set the HWaddr of %s to a different address "
1707                        "to avoid conflicts.\n",
1708                        bond_dev->name,
1709                        slave_dev->name,
1710                        print_mac(mac, slave->perm_hwaddr),
1711                        bond_dev->name,
1712                        slave_dev->name);
1713         }
1714
1715         /* Inform AD package of unbinding of slave. */
1716         if (bond->params.mode == BOND_MODE_8023AD) {
1717                 /* must be called before the slave is
1718                  * detached from the list
1719                  */
1720                 bond_3ad_unbind_slave(slave);
1721         }
1722
1723         printk(KERN_INFO DRV_NAME
1724                ": %s: releasing %s interface %s\n",
1725                bond_dev->name,
1726                (slave->state == BOND_STATE_ACTIVE)
1727                ? "active" : "backup",
1728                slave_dev->name);
1729
1730         oldcurrent = bond->curr_active_slave;
1731
1732         bond->current_arp_slave = NULL;
1733
1734         /* release the slave from its bond */
1735         bond_detach_slave(bond, slave);
1736
1737         bond_compute_features(bond);
1738
1739         if (bond->primary_slave == slave) {
1740                 bond->primary_slave = NULL;
1741         }
1742
1743         if (oldcurrent == slave) {
1744                 bond_change_active_slave(bond, NULL);
1745         }
1746
1747         if ((bond->params.mode == BOND_MODE_TLB) ||
1748             (bond->params.mode == BOND_MODE_ALB)) {
1749                 /* Must be called only after the slave has been
1750                  * detached from the list and the curr_active_slave
1751                  * has been cleared (if our_slave == old_current),
1752                  * but before a new active slave is selected.
1753                  */
1754                 bond_alb_deinit_slave(bond, slave);
1755         }
1756
1757         if (oldcurrent == slave)
1758                 bond_select_active_slave(bond);
1759
1760         if (bond->slave_cnt == 0) {
1761                 bond_set_carrier(bond);
1762
1763                 /* if the last slave was removed, zero the mac address
1764                  * of the master so it will be set by the application
1765                  * to the mac address of the first slave
1766                  */
1767                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1768
1769                 if (list_empty(&bond->vlan_list)) {
1770                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1771                 } else {
1772                         printk(KERN_WARNING DRV_NAME
1773                                ": %s: Warning: clearing HW address of %s while it "
1774                                "still has VLANs.\n",
1775                                bond_dev->name, bond_dev->name);
1776                         printk(KERN_WARNING DRV_NAME
1777                                ": %s: When re-adding slaves, make sure the bond's "
1778                                "HW address matches its VLANs'.\n",
1779                                bond_dev->name);
1780                 }
1781         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1782                    !bond_has_challenged_slaves(bond)) {
1783                 printk(KERN_INFO DRV_NAME
1784                        ": %s: last VLAN challenged slave %s "
1785                        "left bond %s. VLAN blocking is removed\n",
1786                        bond_dev->name, slave_dev->name, bond_dev->name);
1787                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1788         }
1789
1790         write_unlock_bh(&bond->lock);
1791
1792         /* must do this from outside any spinlocks */
1793         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1794
1795         bond_del_vlans_from_slave(bond, slave_dev);
1796
1797         /* If the mode USES_PRIMARY, then we should only remove its
1798          * promisc and mc settings if it was the curr_active_slave, but that was
1799          * already taken care of above when we detached the slave
1800          */
1801         if (!USES_PRIMARY(bond->params.mode)) {
1802                 /* unset promiscuity level from slave */
1803                 if (bond_dev->flags & IFF_PROMISC) {
1804                         dev_set_promiscuity(slave_dev, -1);
1805                 }
1806
1807                 /* unset allmulti level from slave */
1808                 if (bond_dev->flags & IFF_ALLMULTI) {
1809                         dev_set_allmulti(slave_dev, -1);
1810                 }
1811
1812                 /* flush master's mc_list from slave */
1813                 bond_mc_list_flush(bond_dev, slave_dev);
1814         }
1815
1816         netdev_set_master(slave_dev, NULL);
1817
1818         /* close slave before restoring its mac address */
1819         dev_close(slave_dev);
1820
1821         if (!bond->params.fail_over_mac) {
1822                 /* restore original ("permanent") mac address */
1823                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1824                 addr.sa_family = slave_dev->type;
1825                 dev_set_mac_address(slave_dev, &addr);
1826         }
1827
1828         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1829                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1830                                    IFF_SLAVE_NEEDARP);
1831
1832         kfree(slave);
1833
1834         return 0;  /* deletion OK */
1835 }
1836
1837 /*
1838 * Destroy a bonding device.
1839 * Must be under rtnl_lock when this function is called.
1840 */
1841 void bond_destroy(struct bonding *bond)
1842 {
1843         bond_deinit(bond->dev);
1844         bond_destroy_sysfs_entry(bond);
1845         unregister_netdevice(bond->dev);
1846 }
1847
1848 /*
1849 * First release a slave and than destroy the bond if no more slaves iare left.
1850 * Must be under rtnl_lock when this function is called.
1851 */
1852 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1853 {
1854         struct bonding *bond = bond_dev->priv;
1855         int ret;
1856
1857         ret = bond_release(bond_dev, slave_dev);
1858         if ((ret == 0) && (bond->slave_cnt == 0)) {
1859                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1860                        bond_dev->name, bond_dev->name);
1861                 bond_destroy(bond);
1862         }
1863         return ret;
1864 }
1865
1866 /*
1867  * This function releases all slaves.
1868  */
1869 static int bond_release_all(struct net_device *bond_dev)
1870 {
1871         struct bonding *bond = bond_dev->priv;
1872         struct slave *slave;
1873         struct net_device *slave_dev;
1874         struct sockaddr addr;
1875
1876         write_lock_bh(&bond->lock);
1877
1878         netif_carrier_off(bond_dev);
1879
1880         if (bond->slave_cnt == 0) {
1881                 goto out;
1882         }
1883
1884         bond->current_arp_slave = NULL;
1885         bond->primary_slave = NULL;
1886         bond_change_active_slave(bond, NULL);
1887
1888         while ((slave = bond->first_slave) != NULL) {
1889                 /* Inform AD package of unbinding of slave
1890                  * before slave is detached from the list.
1891                  */
1892                 if (bond->params.mode == BOND_MODE_8023AD) {
1893                         bond_3ad_unbind_slave(slave);
1894                 }
1895
1896                 slave_dev = slave->dev;
1897                 bond_detach_slave(bond, slave);
1898
1899                 if ((bond->params.mode == BOND_MODE_TLB) ||
1900                     (bond->params.mode == BOND_MODE_ALB)) {
1901                         /* must be called only after the slave
1902                          * has been detached from the list
1903                          */
1904                         bond_alb_deinit_slave(bond, slave);
1905                 }
1906
1907                 bond_compute_features(bond);
1908
1909                 /* now that the slave is detached, unlock and perform
1910                  * all the undo steps that should not be called from
1911                  * within a lock.
1912                  */
1913                 write_unlock_bh(&bond->lock);
1914
1915                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1916                 bond_del_vlans_from_slave(bond, slave_dev);
1917
1918                 /* If the mode USES_PRIMARY, then we should only remove its
1919                  * promisc and mc settings if it was the curr_active_slave, but that was
1920                  * already taken care of above when we detached the slave
1921                  */
1922                 if (!USES_PRIMARY(bond->params.mode)) {
1923                         /* unset promiscuity level from slave */
1924                         if (bond_dev->flags & IFF_PROMISC) {
1925                                 dev_set_promiscuity(slave_dev, -1);
1926                         }
1927
1928                         /* unset allmulti level from slave */
1929                         if (bond_dev->flags & IFF_ALLMULTI) {
1930                                 dev_set_allmulti(slave_dev, -1);
1931                         }
1932
1933                         /* flush master's mc_list from slave */
1934                         bond_mc_list_flush(bond_dev, slave_dev);
1935                 }
1936
1937                 netdev_set_master(slave_dev, NULL);
1938
1939                 /* close slave before restoring its mac address */
1940                 dev_close(slave_dev);
1941
1942                 if (!bond->params.fail_over_mac) {
1943                         /* restore original ("permanent") mac address*/
1944                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1945                         addr.sa_family = slave_dev->type;
1946                         dev_set_mac_address(slave_dev, &addr);
1947                 }
1948
1949                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1950                                            IFF_SLAVE_INACTIVE);
1951
1952                 kfree(slave);
1953
1954                 /* re-acquire the lock before getting the next slave */
1955                 write_lock_bh(&bond->lock);
1956         }
1957
1958         /* zero the mac address of the master so it will be
1959          * set by the application to the mac address of the
1960          * first slave
1961          */
1962         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1963
1964         if (list_empty(&bond->vlan_list)) {
1965                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1966         } else {
1967                 printk(KERN_WARNING DRV_NAME
1968                        ": %s: Warning: clearing HW address of %s while it "
1969                        "still has VLANs.\n",
1970                        bond_dev->name, bond_dev->name);
1971                 printk(KERN_WARNING DRV_NAME
1972                        ": %s: When re-adding slaves, make sure the bond's "
1973                        "HW address matches its VLANs'.\n",
1974                        bond_dev->name);
1975         }
1976
1977         printk(KERN_INFO DRV_NAME
1978                ": %s: released all slaves\n",
1979                bond_dev->name);
1980
1981 out:
1982         write_unlock_bh(&bond->lock);
1983
1984         return 0;
1985 }
1986
1987 /*
1988  * This function changes the active slave to slave <slave_dev>.
1989  * It returns -EINVAL in the following cases.
1990  *  - <slave_dev> is not found in the list.
1991  *  - There is not active slave now.
1992  *  - <slave_dev> is already active.
1993  *  - The link state of <slave_dev> is not BOND_LINK_UP.
1994  *  - <slave_dev> is not running.
1995  * In these cases, this fuction does nothing.
1996  * In the other cases, currnt_slave pointer is changed and 0 is returned.
1997  */
1998 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1999 {
2000         struct bonding *bond = bond_dev->priv;
2001         struct slave *old_active = NULL;
2002         struct slave *new_active = NULL;
2003         int res = 0;
2004
2005         if (!USES_PRIMARY(bond->params.mode)) {
2006                 return -EINVAL;
2007         }
2008
2009         /* Verify that master_dev is indeed the master of slave_dev */
2010         if (!(slave_dev->flags & IFF_SLAVE) ||
2011             (slave_dev->master != bond_dev)) {
2012                 return -EINVAL;
2013         }
2014
2015         write_lock_bh(&bond->lock);
2016
2017         old_active = bond->curr_active_slave;
2018         new_active = bond_get_slave_by_dev(bond, slave_dev);
2019
2020         /*
2021          * Changing to the current active: do nothing; return success.
2022          */
2023         if (new_active && (new_active == old_active)) {
2024                 write_unlock_bh(&bond->lock);
2025                 return 0;
2026         }
2027
2028         if ((new_active) &&
2029             (old_active) &&
2030             (new_active->link == BOND_LINK_UP) &&
2031             IS_UP(new_active->dev)) {
2032                 bond_change_active_slave(bond, new_active);
2033         } else {
2034                 res = -EINVAL;
2035         }
2036
2037         write_unlock_bh(&bond->lock);
2038
2039         return res;
2040 }
2041
2042 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2043 {
2044         struct bonding *bond = bond_dev->priv;
2045
2046         info->bond_mode = bond->params.mode;
2047         info->miimon = bond->params.miimon;
2048
2049         read_lock_bh(&bond->lock);
2050         info->num_slaves = bond->slave_cnt;
2051         read_unlock_bh(&bond->lock);
2052
2053         return 0;
2054 }
2055
2056 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2057 {
2058         struct bonding *bond = bond_dev->priv;
2059         struct slave *slave;
2060         int i, found = 0;
2061
2062         if (info->slave_id < 0) {
2063                 return -ENODEV;
2064         }
2065
2066         read_lock_bh(&bond->lock);
2067
2068         bond_for_each_slave(bond, slave, i) {
2069                 if (i == (int)info->slave_id) {
2070                         found = 1;
2071                         break;
2072                 }
2073         }
2074
2075         read_unlock_bh(&bond->lock);
2076
2077         if (found) {
2078                 strcpy(info->slave_name, slave->dev->name);
2079                 info->link = slave->link;
2080                 info->state = slave->state;
2081                 info->link_failure_count = slave->link_failure_count;
2082         } else {
2083                 return -ENODEV;
2084         }
2085
2086         return 0;
2087 }
2088
2089 /*-------------------------------- Monitoring -------------------------------*/
2090
2091 /* this function is called regularly to monitor each slave's link. */
2092 void bond_mii_monitor(struct work_struct *work)
2093 {
2094         struct bonding *bond = container_of(work, struct bonding,
2095                                             mii_work.work);
2096         struct slave *slave, *oldcurrent;
2097         int do_failover = 0;
2098         int delta_in_ticks;
2099         int i;
2100
2101         read_lock(&bond->lock);
2102
2103         delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2104
2105         if (bond->kill_timers) {
2106                 goto out;
2107         }
2108
2109         if (bond->slave_cnt == 0) {
2110                 goto re_arm;
2111         }
2112
2113         /* we will try to read the link status of each of our slaves, and
2114          * set their IFF_RUNNING flag appropriately. For each slave not
2115          * supporting MII status, we won't do anything so that a user-space
2116          * program could monitor the link itself if needed.
2117          */
2118
2119         if (bond->send_grat_arp) {
2120                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2121                                 &bond->curr_active_slave->dev->state))
2122                         dprintk("Needs to send gratuitous arp but not yet\n");
2123                 else {
2124                         dprintk("sending delayed gratuitous arp on on %s\n",
2125                                 bond->curr_active_slave->dev->name);
2126                         bond_send_gratuitous_arp(bond);
2127                         bond->send_grat_arp = 0;
2128                 }
2129         }
2130         read_lock(&bond->curr_slave_lock);
2131         oldcurrent = bond->curr_active_slave;
2132         read_unlock(&bond->curr_slave_lock);
2133
2134         bond_for_each_slave(bond, slave, i) {
2135                 struct net_device *slave_dev = slave->dev;
2136                 int link_state;
2137                 u16 old_speed = slave->speed;
2138                 u8 old_duplex = slave->duplex;
2139
2140                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2141
2142                 switch (slave->link) {
2143                 case BOND_LINK_UP:      /* the link was up */
2144                         if (link_state == BMSR_LSTATUS) {
2145                                 /* link stays up, nothing more to do */
2146                                 break;
2147                         } else { /* link going down */
2148                                 slave->link  = BOND_LINK_FAIL;
2149                                 slave->delay = bond->params.downdelay;
2150
2151                                 if (slave->link_failure_count < UINT_MAX) {
2152                                         slave->link_failure_count++;
2153                                 }
2154
2155                                 if (bond->params.downdelay) {
2156                                         printk(KERN_INFO DRV_NAME
2157                                                ": %s: link status down for %s "
2158                                                "interface %s, disabling it in "
2159                                                "%d ms.\n",
2160                                                bond->dev->name,
2161                                                IS_UP(slave_dev)
2162                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2163                                                   ? ((slave == oldcurrent)
2164                                                      ? "active " : "backup ")
2165                                                   : "")
2166                                                : "idle ",
2167                                                slave_dev->name,
2168                                                bond->params.downdelay * bond->params.miimon);
2169                                 }
2170                         }
2171                         /* no break ! fall through the BOND_LINK_FAIL test to
2172                            ensure proper action to be taken
2173                         */
2174                 case BOND_LINK_FAIL:    /* the link has just gone down */
2175                         if (link_state != BMSR_LSTATUS) {
2176                                 /* link stays down */
2177                                 if (slave->delay <= 0) {
2178                                         /* link down for too long time */
2179                                         slave->link = BOND_LINK_DOWN;
2180
2181                                         /* in active/backup mode, we must
2182                                          * completely disable this interface
2183                                          */
2184                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2185                                             (bond->params.mode == BOND_MODE_8023AD)) {
2186                                                 bond_set_slave_inactive_flags(slave);
2187                                         }
2188
2189                                         printk(KERN_INFO DRV_NAME
2190                                                ": %s: link status definitely "
2191                                                "down for interface %s, "
2192                                                "disabling it\n",
2193                                                bond->dev->name,
2194                                                slave_dev->name);
2195
2196                                         /* notify ad that the link status has changed */
2197                                         if (bond->params.mode == BOND_MODE_8023AD) {
2198                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2199                                         }
2200
2201                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2202                                             (bond->params.mode == BOND_MODE_ALB)) {
2203                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2204                                         }
2205
2206                                         if (slave == oldcurrent) {
2207                                                 do_failover = 1;
2208                                         }
2209                                 } else {
2210                                         slave->delay--;
2211                                 }
2212                         } else {
2213                                 /* link up again */
2214                                 slave->link  = BOND_LINK_UP;
2215                                 slave->jiffies = jiffies;
2216                                 printk(KERN_INFO DRV_NAME
2217                                        ": %s: link status up again after %d "
2218                                        "ms for interface %s.\n",
2219                                        bond->dev->name,
2220                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2221                                        slave_dev->name);
2222                         }
2223                         break;
2224                 case BOND_LINK_DOWN:    /* the link was down */
2225                         if (link_state != BMSR_LSTATUS) {
2226                                 /* the link stays down, nothing more to do */
2227                                 break;
2228                         } else {        /* link going up */
2229                                 slave->link  = BOND_LINK_BACK;
2230                                 slave->delay = bond->params.updelay;
2231
2232                                 if (bond->params.updelay) {
2233                                         /* if updelay == 0, no need to
2234                                            advertise about a 0 ms delay */
2235                                         printk(KERN_INFO DRV_NAME
2236                                                ": %s: link status up for "
2237                                                "interface %s, enabling it "
2238                                                "in %d ms.\n",
2239                                                bond->dev->name,
2240                                                slave_dev->name,
2241                                                bond->params.updelay * bond->params.miimon);
2242                                 }
2243                         }
2244                         /* no break ! fall through the BOND_LINK_BACK state in
2245                            case there's something to do.
2246                         */
2247                 case BOND_LINK_BACK:    /* the link has just come back */
2248                         if (link_state != BMSR_LSTATUS) {
2249                                 /* link down again */
2250                                 slave->link  = BOND_LINK_DOWN;
2251
2252                                 printk(KERN_INFO DRV_NAME
2253                                        ": %s: link status down again after %d "
2254                                        "ms for interface %s.\n",
2255                                        bond->dev->name,
2256                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2257                                        slave_dev->name);
2258                         } else {
2259                                 /* link stays up */
2260                                 if (slave->delay == 0) {
2261                                         /* now the link has been up for long time enough */
2262                                         slave->link = BOND_LINK_UP;
2263                                         slave->jiffies = jiffies;
2264
2265                                         if (bond->params.mode == BOND_MODE_8023AD) {
2266                                                 /* prevent it from being the active one */
2267                                                 slave->state = BOND_STATE_BACKUP;
2268                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2269                                                 /* make it immediately active */
2270                                                 slave->state = BOND_STATE_ACTIVE;
2271                                         } else if (slave != bond->primary_slave) {
2272                                                 /* prevent it from being the active one */
2273                                                 slave->state = BOND_STATE_BACKUP;
2274                                         }
2275
2276                                         printk(KERN_INFO DRV_NAME
2277                                                ": %s: link status definitely "
2278                                                "up for interface %s.\n",
2279                                                bond->dev->name,
2280                                                slave_dev->name);
2281
2282                                         /* notify ad that the link status has changed */
2283                                         if (bond->params.mode == BOND_MODE_8023AD) {
2284                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2285                                         }
2286
2287                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2288                                             (bond->params.mode == BOND_MODE_ALB)) {
2289                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2290                                         }
2291
2292                                         if ((!oldcurrent) ||
2293                                             (slave == bond->primary_slave)) {
2294                                                 do_failover = 1;
2295                                         }
2296                                 } else {
2297                                         slave->delay--;
2298                                 }
2299                         }
2300                         break;
2301                 default:
2302                         /* Should not happen */
2303                         printk(KERN_ERR DRV_NAME
2304                                ": %s: Error: %s Illegal value (link=%d)\n",
2305                                bond->dev->name,
2306                                slave->dev->name,
2307                                slave->link);
2308                         goto out;
2309                 } /* end of switch (slave->link) */
2310
2311                 bond_update_speed_duplex(slave);
2312
2313                 if (bond->params.mode == BOND_MODE_8023AD) {
2314                         if (old_speed != slave->speed) {
2315                                 bond_3ad_adapter_speed_changed(slave);
2316                         }
2317
2318                         if (old_duplex != slave->duplex) {
2319                                 bond_3ad_adapter_duplex_changed(slave);
2320                         }
2321                 }
2322
2323         } /* end of for */
2324
2325         if (do_failover) {
2326                 write_lock(&bond->curr_slave_lock);
2327
2328                 bond_select_active_slave(bond);
2329
2330                 write_unlock(&bond->curr_slave_lock);
2331         } else
2332                 bond_set_carrier(bond);
2333
2334 re_arm:
2335         if (bond->params.miimon)
2336                 queue_delayed_work(bond->wq, &bond->mii_work, delta_in_ticks);
2337 out:
2338         read_unlock(&bond->lock);
2339 }
2340
2341
2342 static __be32 bond_glean_dev_ip(struct net_device *dev)
2343 {
2344         struct in_device *idev;
2345         struct in_ifaddr *ifa;
2346         __be32 addr = 0;
2347
2348         if (!dev)
2349                 return 0;
2350
2351         rcu_read_lock();
2352         idev = __in_dev_get_rcu(dev);
2353         if (!idev)
2354                 goto out;
2355
2356         ifa = idev->ifa_list;
2357         if (!ifa)
2358                 goto out;
2359
2360         addr = ifa->ifa_local;
2361 out:
2362         rcu_read_unlock();
2363         return addr;
2364 }
2365
2366 static int bond_has_ip(struct bonding *bond)
2367 {
2368         struct vlan_entry *vlan, *vlan_next;
2369
2370         if (bond->master_ip)
2371                 return 1;
2372
2373         if (list_empty(&bond->vlan_list))
2374                 return 0;
2375
2376         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2377                                  vlan_list) {
2378                 if (vlan->vlan_ip)
2379                         return 1;
2380         }
2381
2382         return 0;
2383 }
2384
2385 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2386 {
2387         struct vlan_entry *vlan, *vlan_next;
2388
2389         if (ip == bond->master_ip)
2390                 return 1;
2391
2392         if (list_empty(&bond->vlan_list))
2393                 return 0;
2394
2395         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2396                                  vlan_list) {
2397                 if (ip == vlan->vlan_ip)
2398                         return 1;
2399         }
2400
2401         return 0;
2402 }
2403
2404 /*
2405  * We go to the (large) trouble of VLAN tagging ARP frames because
2406  * switches in VLAN mode (especially if ports are configured as
2407  * "native" to a VLAN) might not pass non-tagged frames.
2408  */
2409 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2410 {
2411         struct sk_buff *skb;
2412
2413         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2414                slave_dev->name, dest_ip, src_ip, vlan_id);
2415
2416         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2417                          NULL, slave_dev->dev_addr, NULL);
2418
2419         if (!skb) {
2420                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2421                 return;
2422         }
2423         if (vlan_id) {
2424                 skb = vlan_put_tag(skb, vlan_id);
2425                 if (!skb) {
2426                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2427                         return;
2428                 }
2429         }
2430         arp_xmit(skb);
2431 }
2432
2433
2434 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2435 {
2436         int i, vlan_id, rv;
2437         __be32 *targets = bond->params.arp_targets;
2438         struct vlan_entry *vlan, *vlan_next;
2439         struct net_device *vlan_dev;
2440         struct flowi fl;
2441         struct rtable *rt;
2442
2443         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2444                 if (!targets[i])
2445                         continue;
2446                 dprintk("basa: target %x\n", targets[i]);
2447                 if (list_empty(&bond->vlan_list)) {
2448                         dprintk("basa: empty vlan: arp_send\n");
2449                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2450                                       bond->master_ip, 0);
2451                         continue;
2452                 }
2453
2454                 /*
2455                  * If VLANs are configured, we do a route lookup to
2456                  * determine which VLAN interface would be used, so we
2457                  * can tag the ARP with the proper VLAN tag.
2458                  */
2459                 memset(&fl, 0, sizeof(fl));
2460                 fl.fl4_dst = targets[i];
2461                 fl.fl4_tos = RTO_ONLINK;
2462
2463                 rv = ip_route_output_key(&rt, &fl);
2464                 if (rv) {
2465                         if (net_ratelimit()) {
2466                                 printk(KERN_WARNING DRV_NAME
2467                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2468                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2469                         }
2470                         continue;
2471                 }
2472
2473                 /*
2474                  * This target is not on a VLAN
2475                  */
2476                 if (rt->u.dst.dev == bond->dev) {
2477                         ip_rt_put(rt);
2478                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2479                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2480                                       bond->master_ip, 0);
2481                         continue;
2482                 }
2483
2484                 vlan_id = 0;
2485                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2486                                          vlan_list) {
2487                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2488                         if (vlan_dev == rt->u.dst.dev) {
2489                                 vlan_id = vlan->vlan_id;
2490                                 dprintk("basa: vlan match on %s %d\n",
2491                                        vlan_dev->name, vlan_id);
2492                                 break;
2493                         }
2494                 }
2495
2496                 if (vlan_id) {
2497                         ip_rt_put(rt);
2498                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2499                                       vlan->vlan_ip, vlan_id);
2500                         continue;
2501                 }
2502
2503                 if (net_ratelimit()) {
2504                         printk(KERN_WARNING DRV_NAME
2505                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2506                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2507                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2508                 }
2509                 ip_rt_put(rt);
2510         }
2511 }
2512
2513 /*
2514  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2515  * for each VLAN above us.
2516  */
2517 static void bond_send_gratuitous_arp(struct bonding *bond)
2518 {
2519         struct slave *slave = bond->curr_active_slave;
2520         struct vlan_entry *vlan;
2521         struct net_device *vlan_dev;
2522
2523         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2524                                 slave ? slave->dev->name : "NULL");
2525         if (!slave)
2526                 return;
2527
2528         if (bond->master_ip) {
2529                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2530                                 bond->master_ip, 0);
2531         }
2532
2533         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2534                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2535                 if (vlan->vlan_ip) {
2536                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2537                                       vlan->vlan_ip, vlan->vlan_id);
2538                 }
2539         }
2540 }
2541
2542 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2543 {
2544         int i;
2545         __be32 *targets = bond->params.arp_targets;
2546
2547         targets = bond->params.arp_targets;
2548         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2549                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2550                         "%u.%u.%u.%u bhti(tip) %d\n",
2551                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2552                        bond_has_this_ip(bond, tip));
2553                 if (sip == targets[i]) {
2554                         if (bond_has_this_ip(bond, tip))
2555                                 slave->last_arp_rx = jiffies;
2556                         return;
2557                 }
2558         }
2559 }
2560
2561 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2562 {
2563         struct arphdr *arp;
2564         struct slave *slave;
2565         struct bonding *bond;
2566         unsigned char *arp_ptr;
2567         __be32 sip, tip;
2568
2569         if (dev->nd_net != &init_net)
2570                 goto out;
2571
2572         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2573                 goto out;
2574
2575         bond = dev->priv;
2576         read_lock(&bond->lock);
2577
2578         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2579                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2580                 orig_dev ? orig_dev->name : "NULL");
2581
2582         slave = bond_get_slave_by_dev(bond, orig_dev);
2583         if (!slave || !slave_do_arp_validate(bond, slave))
2584                 goto out_unlock;
2585
2586         /* ARP header, plus 2 device addresses, plus 2 IP addresses.  */
2587         if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2588                                  (2 * dev->addr_len) +
2589                                  (2 * sizeof(u32)))))
2590                 goto out_unlock;
2591
2592         arp = arp_hdr(skb);
2593         if (arp->ar_hln != dev->addr_len ||
2594             skb->pkt_type == PACKET_OTHERHOST ||
2595             skb->pkt_type == PACKET_LOOPBACK ||
2596             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2597             arp->ar_pro != htons(ETH_P_IP) ||
2598             arp->ar_pln != 4)
2599                 goto out_unlock;
2600
2601         arp_ptr = (unsigned char *)(arp + 1);
2602         arp_ptr += dev->addr_len;
2603         memcpy(&sip, arp_ptr, 4);
2604         arp_ptr += 4 + dev->addr_len;
2605         memcpy(&tip, arp_ptr, 4);
2606
2607         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2608                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2609                 slave->state, bond->params.arp_validate,
2610                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2611
2612         /*
2613          * Backup slaves won't see the ARP reply, but do come through
2614          * here for each ARP probe (so we swap the sip/tip to validate
2615          * the probe).  In a "redundant switch, common router" type of
2616          * configuration, the ARP probe will (hopefully) travel from
2617          * the active, through one switch, the router, then the other
2618          * switch before reaching the backup.
2619          */
2620         if (slave->state == BOND_STATE_ACTIVE)
2621                 bond_validate_arp(bond, slave, sip, tip);
2622         else
2623                 bond_validate_arp(bond, slave, tip, sip);
2624
2625 out_unlock:
2626         read_unlock(&bond->lock);
2627 out:
2628         dev_kfree_skb(skb);
2629         return NET_RX_SUCCESS;
2630 }
2631
2632 /*
2633  * this function is called regularly to monitor each slave's link
2634  * ensuring that traffic is being sent and received when arp monitoring
2635  * is used in load-balancing mode. if the adapter has been dormant, then an
2636  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2637  * arp monitoring in active backup mode.
2638  */
2639 void bond_loadbalance_arp_mon(struct work_struct *work)
2640 {
2641         struct bonding *bond = container_of(work, struct bonding,
2642                                             arp_work.work);
2643         struct slave *slave, *oldcurrent;
2644         int do_failover = 0;
2645         int delta_in_ticks;
2646         int i;
2647
2648         read_lock(&bond->lock);
2649
2650         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2651
2652         if (bond->kill_timers) {
2653                 goto out;
2654         }
2655
2656         if (bond->slave_cnt == 0) {
2657                 goto re_arm;
2658         }
2659
2660         read_lock(&bond->curr_slave_lock);
2661         oldcurrent = bond->curr_active_slave;
2662         read_unlock(&bond->curr_slave_lock);
2663
2664         /* see if any of the previous devices are up now (i.e. they have
2665          * xmt and rcv traffic). the curr_active_slave does not come into
2666          * the picture unless it is null. also, slave->jiffies is not needed
2667          * here because we send an arp on each slave and give a slave as
2668          * long as it needs to get the tx/rx within the delta.
2669          * TODO: what about up/down delay in arp mode? it wasn't here before
2670          *       so it can wait
2671          */
2672         bond_for_each_slave(bond, slave, i) {
2673                 if (slave->link != BOND_LINK_UP) {
2674                         if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2675                             ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2676
2677                                 slave->link  = BOND_LINK_UP;
2678                                 slave->state = BOND_STATE_ACTIVE;
2679
2680                                 /* primary_slave has no meaning in round-robin
2681                                  * mode. the window of a slave being up and
2682                                  * curr_active_slave being null after enslaving
2683                                  * is closed.
2684                                  */
2685                                 if (!oldcurrent) {
2686                                         printk(KERN_INFO DRV_NAME
2687                                                ": %s: link status definitely "
2688                                                "up for interface %s, ",
2689                                                bond->dev->name,
2690                                                slave->dev->name);
2691                                         do_failover = 1;
2692                                 } else {
2693                                         printk(KERN_INFO DRV_NAME
2694                                                ": %s: interface %s is now up\n",
2695                                                bond->dev->name,
2696                                                slave->dev->name);
2697                                 }
2698                         }
2699                 } else {
2700                         /* slave->link == BOND_LINK_UP */
2701
2702                         /* not all switches will respond to an arp request
2703                          * when the source ip is 0, so don't take the link down
2704                          * if we don't know our ip yet
2705                          */
2706                         if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2707                             (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2708                              bond_has_ip(bond))) {
2709
2710                                 slave->link  = BOND_LINK_DOWN;
2711                                 slave->state = BOND_STATE_BACKUP;
2712
2713                                 if (slave->link_failure_count < UINT_MAX) {
2714                                         slave->link_failure_count++;
2715                                 }
2716
2717                                 printk(KERN_INFO DRV_NAME
2718                                        ": %s: interface %s is now down.\n",
2719                                        bond->dev->name,
2720                                        slave->dev->name);
2721
2722                                 if (slave == oldcurrent) {
2723                                         do_failover = 1;
2724                                 }
2725                         }
2726                 }
2727
2728                 /* note: if switch is in round-robin mode, all links
2729                  * must tx arp to ensure all links rx an arp - otherwise
2730                  * links may oscillate or not come up at all; if switch is
2731                  * in something like xor mode, there is nothing we can
2732                  * do - all replies will be rx'ed on same link causing slaves
2733                  * to be unstable during low/no traffic periods
2734                  */
2735                 if (IS_UP(slave->dev)) {
2736                         bond_arp_send_all(bond, slave);
2737                 }
2738         }
2739
2740         if (do_failover) {
2741                 write_lock(&bond->curr_slave_lock);
2742
2743                 bond_select_active_slave(bond);
2744
2745                 write_unlock(&bond->curr_slave_lock);
2746         }
2747
2748 re_arm:
2749         if (bond->params.arp_interval)
2750                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2751 out:
2752         read_unlock(&bond->lock);
2753 }
2754
2755 /*
2756  * When using arp monitoring in active-backup mode, this function is
2757  * called to determine if any backup slaves have went down or a new
2758  * current slave needs to be found.
2759  * The backup slaves never generate traffic, they are considered up by merely
2760  * receiving traffic. If the current slave goes down, each backup slave will
2761  * be given the opportunity to tx/rx an arp before being taken down - this
2762  * prevents all slaves from being taken down due to the current slave not
2763  * sending any traffic for the backups to receive. The arps are not necessarily
2764  * necessary, any tx and rx traffic will keep the current slave up. While any
2765  * rx traffic will keep the backup slaves up, the current slave is responsible
2766  * for generating traffic to keep them up regardless of any other traffic they
2767  * may have received.
2768  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2769  */
2770 void bond_activebackup_arp_mon(struct work_struct *work)
2771 {
2772         struct bonding *bond = container_of(work, struct bonding,
2773                                             arp_work.work);
2774         struct slave *slave;
2775         int delta_in_ticks;
2776         int i;
2777
2778         read_lock(&bond->lock);
2779
2780         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2781
2782         if (bond->kill_timers) {
2783                 goto out;
2784         }
2785
2786         if (bond->slave_cnt == 0) {
2787                 goto re_arm;
2788         }
2789
2790         /* determine if any slave has come up or any backup slave has
2791          * gone down
2792          * TODO: what about up/down delay in arp mode? it wasn't here before
2793          *       so it can wait
2794          */
2795         bond_for_each_slave(bond, slave, i) {
2796                 if (slave->link != BOND_LINK_UP) {
2797                         if ((jiffies - slave_last_rx(bond, slave)) <=
2798                              delta_in_ticks) {
2799
2800                                 slave->link = BOND_LINK_UP;
2801
2802                                 write_lock(&bond->curr_slave_lock);
2803
2804                                 if ((!bond->curr_active_slave) &&
2805                                     ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2806                                         bond_change_active_slave(bond, slave);
2807                                         bond->current_arp_slave = NULL;
2808                                 } else if (bond->curr_active_slave != slave) {
2809                                         /* this slave has just come up but we
2810                                          * already have a current slave; this
2811                                          * can also happen if bond_enslave adds
2812                                          * a new slave that is up while we are
2813                                          * searching for a new slave
2814                                          */
2815                                         bond_set_slave_inactive_flags(slave);
2816                                         bond->current_arp_slave = NULL;
2817                                 }
2818
2819                                 bond_set_carrier(bond);
2820
2821                                 if (slave == bond->curr_active_slave) {
2822                                         printk(KERN_INFO DRV_NAME
2823                                                ": %s: %s is up and now the "
2824                                                "active interface\n",
2825                                                bond->dev->name,
2826                                                slave->dev->name);
2827                                         netif_carrier_on(bond->dev);
2828                                 } else {
2829                                         printk(KERN_INFO DRV_NAME
2830                                                ": %s: backup interface %s is "
2831                                                "now up\n",
2832                                                bond->dev->name,
2833                                                slave->dev->name);
2834                                 }
2835
2836                                 write_unlock(&bond->curr_slave_lock);
2837                         }
2838                 } else {
2839                         read_lock(&bond->curr_slave_lock);
2840
2841                         if ((slave != bond->curr_active_slave) &&
2842                             (!bond->current_arp_slave) &&
2843                             (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2844                              bond_has_ip(bond))) {
2845                                 /* a backup slave has gone down; three times
2846                                  * the delta allows the current slave to be
2847                                  * taken out before the backup slave.
2848                                  * note: a non-null current_arp_slave indicates
2849                                  * the curr_active_slave went down and we are
2850                                  * searching for a new one; under this
2851                                  * condition we only take the curr_active_slave
2852                                  * down - this gives each slave a chance to
2853                                  * tx/rx traffic before being taken out
2854                                  */
2855
2856                                 read_unlock(&bond->curr_slave_lock);
2857
2858                                 slave->link  = BOND_LINK_DOWN;
2859
2860                                 if (slave->link_failure_count < UINT_MAX) {
2861                                         slave->link_failure_count++;
2862                                 }
2863
2864                                 bond_set_slave_inactive_flags(slave);
2865
2866                                 printk(KERN_INFO DRV_NAME
2867                                        ": %s: backup interface %s is now down\n",
2868                                        bond->dev->name,
2869                                        slave->dev->name);
2870                         } else {
2871                                 read_unlock(&bond->curr_slave_lock);
2872                         }
2873                 }
2874         }
2875
2876         read_lock(&bond->curr_slave_lock);
2877         slave = bond->curr_active_slave;
2878         read_unlock(&bond->curr_slave_lock);
2879
2880         if (slave) {
2881                 /* if we have sent traffic in the past 2*arp_intervals but
2882                  * haven't xmit and rx traffic in that time interval, select
2883                  * a different slave. slave->jiffies is only updated when
2884                  * a slave first becomes the curr_active_slave - not necessarily
2885                  * after every arp; this ensures the slave has a full 2*delta
2886                  * before being taken out. if a primary is being used, check
2887                  * if it is up and needs to take over as the curr_active_slave
2888                  */
2889                 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2890             (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2891              bond_has_ip(bond))) &&
2892                     ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2893
2894                         slave->link  = BOND_LINK_DOWN;
2895
2896                         if (slave->link_failure_count < UINT_MAX) {
2897                                 slave->link_failure_count++;
2898                         }
2899
2900                         printk(KERN_INFO DRV_NAME
2901                                ": %s: link status down for active interface "
2902                                "%s, disabling it\n",
2903                                bond->dev->name,
2904                                slave->dev->name);
2905
2906                         write_lock(&bond->curr_slave_lock);
2907
2908                         bond_select_active_slave(bond);
2909                         slave = bond->curr_active_slave;
2910
2911                         write_unlock(&bond->curr_slave_lock);
2912
2913                         bond->current_arp_slave = slave;
2914
2915                         if (slave) {
2916                                 slave->jiffies = jiffies;
2917                         }
2918                 } else if ((bond->primary_slave) &&
2919                            (bond->primary_slave != slave) &&
2920                            (bond->primary_slave->link == BOND_LINK_UP)) {
2921                         /* at this point, slave is the curr_active_slave */
2922                         printk(KERN_INFO DRV_NAME
2923                                ": %s: changing from interface %s to primary "
2924                                "interface %s\n",
2925                                bond->dev->name,
2926                                slave->dev->name,
2927                                bond->primary_slave->dev->name);
2928
2929                         /* primary is up so switch to it */
2930                         write_lock(&bond->curr_slave_lock);
2931                         bond_change_active_slave(bond, bond->primary_slave);
2932                         write_unlock(&bond->curr_slave_lock);
2933
2934                         slave = bond->primary_slave;
2935                         slave->jiffies = jiffies;
2936                 } else {
2937                         bond->current_arp_slave = NULL;
2938                 }
2939
2940                 /* the current slave must tx an arp to ensure backup slaves
2941                  * rx traffic
2942                  */
2943                 if (slave && bond_has_ip(bond)) {
2944                         bond_arp_send_all(bond, slave);
2945                 }
2946         }
2947
2948         /* if we don't have a curr_active_slave, search for the next available
2949          * backup slave from the current_arp_slave and make it the candidate
2950          * for becoming the curr_active_slave
2951          */
2952         if (!slave) {
2953                 if (!bond->current_arp_slave) {
2954                         bond->current_arp_slave = bond->first_slave;
2955                 }
2956
2957                 if (bond->current_arp_slave) {
2958                         bond_set_slave_inactive_flags(bond->current_arp_slave);
2959
2960                         /* search for next candidate */
2961                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2962                                 if (IS_UP(slave->dev)) {
2963                                         slave->link = BOND_LINK_BACK;
2964                                         bond_set_slave_active_flags(slave);
2965                                         bond_arp_send_all(bond, slave);
2966                                         slave->jiffies = jiffies;
2967                                         bond->current_arp_slave = slave;
2968                                         break;
2969                                 }
2970
2971                                 /* if the link state is up at this point, we
2972                                  * mark it down - this can happen if we have
2973                                  * simultaneous link failures and
2974                                  * reselect_active_interface doesn't make this
2975                                  * one the current slave so it is still marked
2976                                  * up when it is actually down
2977                                  */
2978                                 if (slave->link == BOND_LINK_UP) {
2979                                         slave->link  = BOND_LINK_DOWN;
2980                                         if (slave->link_failure_count < UINT_MAX) {
2981                                                 slave->link_failure_count++;
2982                                         }
2983
2984                                         bond_set_slave_inactive_flags(slave);
2985
2986                                         printk(KERN_INFO DRV_NAME
2987                                                ": %s: backup interface %s is "
2988                                                "now down.\n",
2989                                                bond->dev->name,
2990                                                slave->dev->name);
2991                                 }
2992                         }
2993                 }
2994         }
2995
2996 re_arm:
2997         if (bond->params.arp_interval) {
2998                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2999         }
3000 out:
3001         read_unlock(&bond->lock);
3002 }
3003
3004 /*------------------------------ proc/seq_file-------------------------------*/
3005
3006 #ifdef CONFIG_PROC_FS
3007
3008 #define SEQ_START_TOKEN ((void *)1)
3009
3010 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3011 {
3012         struct bonding *bond = seq->private;
3013         loff_t off = 0;
3014         struct slave *slave;
3015         int i;
3016
3017         /* make sure the bond won't be taken away */
3018         read_lock(&dev_base_lock);
3019         read_lock_bh(&bond->lock);
3020
3021         if (*pos == 0) {
3022                 return SEQ_START_TOKEN;
3023         }
3024
3025         bond_for_each_slave(bond, slave, i) {
3026                 if (++off == *pos) {
3027                         return slave;
3028                 }
3029         }
3030
3031         return NULL;
3032 }
3033
3034 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3035 {
3036         struct bonding *bond = seq->private;
3037         struct slave *slave = v;
3038
3039         ++*pos;
3040         if (v == SEQ_START_TOKEN) {
3041                 return bond->first_slave;
3042         }
3043
3044         slave = slave->next;
3045
3046         return (slave == bond->first_slave) ? NULL : slave;
3047 }
3048
3049 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3050 {
3051         struct bonding *bond = seq->private;
3052
3053         read_unlock_bh(&bond->lock);
3054         read_unlock(&dev_base_lock);
3055 }
3056
3057 static void bond_info_show_master(struct seq_file *seq)
3058 {
3059         struct bonding *bond = seq->private;
3060         struct slave *curr;
3061         int i;
3062         u32 target;
3063
3064         read_lock(&bond->curr_slave_lock);
3065         curr = bond->curr_active_slave;
3066         read_unlock(&bond->curr_slave_lock);
3067
3068         seq_printf(seq, "Bonding Mode: %s",
3069                    bond_mode_name(bond->params.mode));
3070
3071         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3072             bond->params.fail_over_mac)
3073                 seq_printf(seq, " (fail_over_mac)");
3074
3075         seq_printf(seq, "\n");
3076
3077         if (bond->params.mode == BOND_MODE_XOR ||
3078                 bond->params.mode == BOND_MODE_8023AD) {
3079                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3080                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3081                         bond->params.xmit_policy);
3082         }
3083
3084         if (USES_PRIMARY(bond->params.mode)) {
3085                 seq_printf(seq, "Primary Slave: %s\n",
3086                            (bond->primary_slave) ?
3087                            bond->primary_slave->dev->name : "None");
3088
3089                 seq_printf(seq, "Currently Active Slave: %s\n",
3090                            (curr) ? curr->dev->name : "None");
3091         }
3092
3093         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3094                    "up" : "down");
3095         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3096         seq_printf(seq, "Up Delay (ms): %d\n",
3097                    bond->params.updelay * bond->params.miimon);
3098         seq_printf(seq, "Down Delay (ms): %d\n",
3099                    bond->params.downdelay * bond->params.miimon);
3100
3101
3102         /* ARP information */
3103         if(bond->params.arp_interval > 0) {
3104                 int printed=0;
3105                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3106                                 bond->params.arp_interval);
3107
3108                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3109
3110                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3111                         if (!bond->params.arp_targets[i])
3112                                 continue;
3113                         if (printed)
3114                                 seq_printf(seq, ",");
3115                         target = ntohl(bond->params.arp_targets[i]);
3116                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3117                         printed = 1;
3118                 }
3119                 seq_printf(seq, "\n");
3120         }
3121
3122         if (bond->params.mode == BOND_MODE_8023AD) {
3123                 struct ad_info ad_info;
3124                 DECLARE_MAC_BUF(mac);
3125
3126                 seq_puts(seq, "\n802.3ad info\n");
3127                 seq_printf(seq, "LACP rate: %s\n",
3128                            (bond->params.lacp_fast) ? "fast" : "slow");
3129
3130                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3131                         seq_printf(seq, "bond %s has no active aggregator\n",
3132                                    bond->dev->name);
3133                 } else {
3134                         seq_printf(seq, "Active Aggregator Info:\n");
3135
3136                         seq_printf(seq, "\tAggregator ID: %d\n",
3137                                    ad_info.aggregator_id);
3138                         seq_printf(seq, "\tNumber of ports: %d\n",
3139                                    ad_info.ports);
3140                         seq_printf(seq, "\tActor Key: %d\n",
3141                                    ad_info.actor_key);
3142                         seq_printf(seq, "\tPartner Key: %d\n",
3143                                    ad_info.partner_key);
3144                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3145                                    print_mac(mac, ad_info.partner_system));
3146                 }
3147         }
3148 }
3149
3150 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3151 {
3152         struct bonding *bond = seq->private;
3153         DECLARE_MAC_BUF(mac);
3154
3155         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3156         seq_printf(seq, "MII Status: %s\n",
3157                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3158         seq_printf(seq, "Link Failure Count: %u\n",
3159                    slave->link_failure_count);
3160
3161         seq_printf(seq,
3162                    "Permanent HW addr: %s\n",
3163                    print_mac(mac, slave->perm_hwaddr));
3164
3165         if (bond->params.mode == BOND_MODE_8023AD) {
3166                 const struct aggregator *agg
3167                         = SLAVE_AD_INFO(slave).port.aggregator;
3168
3169                 if (agg) {
3170                         seq_printf(seq, "Aggregator ID: %d\n",
3171                                    agg->aggregator_identifier);
3172                 } else {
3173                         seq_puts(seq, "Aggregator ID: N/A\n");
3174                 }
3175         }
3176 }
3177
3178 static int bond_info_seq_show(struct seq_file *seq, void *v)
3179 {
3180         if (v == SEQ_START_TOKEN) {
3181                 seq_printf(seq, "%s\n", version);
3182                 bond_info_show_master(seq);
3183         } else {
3184                 bond_info_show_slave(seq, v);
3185         }
3186
3187         return 0;
3188 }
3189
3190 static struct seq_operations bond_info_seq_ops = {
3191         .start = bond_info_seq_start,
3192         .next  = bond_info_seq_next,
3193         .stop  = bond_info_seq_stop,
3194         .show  = bond_info_seq_show,
3195 };
3196
3197 static int bond_info_open(struct inode *inode, struct file *file)
3198 {
3199         struct seq_file *seq;
3200         struct proc_dir_entry *proc;
3201         int res;
3202
3203         res = seq_open(file, &bond_info_seq_ops);
3204         if (!res) {
3205                 /* recover the pointer buried in proc_dir_entry data */
3206                 seq = file->private_data;
3207                 proc = PDE(inode);
3208                 seq->private = proc->data;
3209         }
3210
3211         return res;
3212 }
3213
3214 static const struct file_operations bond_info_fops = {
3215         .owner   = THIS_MODULE,
3216         .open    = bond_info_open,
3217         .read    = seq_read,
3218         .llseek  = seq_lseek,
3219         .release = seq_release,
3220 };
3221
3222 static int bond_create_proc_entry(struct bonding *bond)
3223 {
3224         struct net_device *bond_dev = bond->dev;
3225
3226         if (bond_proc_dir) {
3227                 bond->proc_entry = create_proc_entry(bond_dev->name,
3228                                                      S_IRUGO,
3229                                                      bond_proc_dir);
3230                 if (bond->proc_entry == NULL) {
3231                         printk(KERN_WARNING DRV_NAME
3232                                ": Warning: Cannot create /proc/net/%s/%s\n",
3233                                DRV_NAME, bond_dev->name);
3234                 } else {
3235                         bond->proc_entry->data = bond;
3236                         bond->proc_entry->proc_fops = &bond_info_fops;
3237                         bond->proc_entry->owner = THIS_MODULE;
3238                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3239                 }
3240         }
3241
3242         return 0;
3243 }
3244
3245 static void bond_remove_proc_entry(struct bonding *bond)
3246 {
3247         if (bond_proc_dir && bond->proc_entry) {
3248                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3249                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3250                 bond->proc_entry = NULL;
3251         }
3252 }
3253
3254 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3255  * Caller must hold rtnl_lock.
3256  */
3257 static void bond_create_proc_dir(void)
3258 {
3259         int len = strlen(DRV_NAME);
3260
3261         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3262              bond_proc_dir = bond_proc_dir->next) {
3263                 if ((bond_proc_dir->namelen == len) &&
3264                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3265                         break;
3266                 }
3267         }
3268
3269         if (!bond_proc_dir) {
3270                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3271                 if (bond_proc_dir) {
3272                         bond_proc_dir->owner = THIS_MODULE;
3273                 } else {
3274                         printk(KERN_WARNING DRV_NAME
3275                                 ": Warning: cannot create /proc/net/%s\n",
3276                                 DRV_NAME);
3277                 }
3278         }
3279 }
3280
3281 /* Destroy the bonding directory under /proc/net, if empty.
3282  * Caller must hold rtnl_lock.
3283  */
3284 static void bond_destroy_proc_dir(void)
3285 {
3286         struct proc_dir_entry *de;
3287
3288         if (!bond_proc_dir) {
3289                 return;
3290         }
3291
3292         /* verify that the /proc dir is empty */
3293         for (de = bond_proc_dir->subdir; de; de = de->next) {
3294                 /* ignore . and .. */
3295                 if (*(de->name) != '.') {
3296                         break;
3297                 }
3298         }
3299
3300         if (de) {
3301                 if (bond_proc_dir->owner == THIS_MODULE) {
3302                         bond_proc_dir->owner = NULL;
3303                 }
3304         } else {
3305                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3306                 bond_proc_dir = NULL;
3307         }
3308 }
3309 #endif /* CONFIG_PROC_FS */
3310
3311 /*-------------------------- netdev event handling --------------------------*/
3312
3313 /*
3314  * Change device name
3315  */
3316 static int bond_event_changename(struct bonding *bond)
3317 {
3318 #ifdef CONFIG_PROC_FS
3319         bond_remove_proc_entry(bond);
3320         bond_create_proc_entry(bond);
3321 #endif
3322         down_write(&(bonding_rwsem));
3323         bond_destroy_sysfs_entry(bond);
3324         bond_create_sysfs_entry(bond);
3325         up_write(&(bonding_rwsem));
3326         return NOTIFY_DONE;
3327 }
3328
3329 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3330 {
3331         struct bonding *event_bond = bond_dev->priv;
3332
3333         switch (event) {
3334         case NETDEV_CHANGENAME:
3335                 return bond_event_changename(event_bond);
3336         case NETDEV_UNREGISTER:
3337                 /*
3338                  * TODO: remove a bond from the list?
3339                  */
3340                 break;
3341         default:
3342                 break;
3343         }
3344
3345         return NOTIFY_DONE;
3346 }
3347
3348 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3349 {
3350         struct net_device *bond_dev = slave_dev->master;
3351         struct bonding *bond = bond_dev->priv;
3352
3353         switch (event) {
3354         case NETDEV_UNREGISTER:
3355                 if (bond_dev) {
3356                         if (bond->setup_by_slave)
3357                                 bond_release_and_destroy(bond_dev, slave_dev);
3358                         else
3359                                 bond_release(bond_dev, slave_dev);
3360                 }
3361                 break;
3362         case NETDEV_CHANGE:
3363                 /*
3364                  * TODO: is this what we get if somebody
3365                  * sets up a hierarchical bond, then rmmod's
3366                  * one of the slave bonding devices?
3367                  */
3368                 break;
3369         case NETDEV_DOWN:
3370                 /*
3371                  * ... Or is it this?
3372                  */
3373                 break;
3374         case NETDEV_CHANGEMTU:
3375                 /*
3376                  * TODO: Should slaves be allowed to
3377                  * independently alter their MTU?  For
3378                  * an active-backup bond, slaves need
3379                  * not be the same type of device, so
3380                  * MTUs may vary.  For other modes,
3381                  * slaves arguably should have the
3382                  * same MTUs. To do this, we'd need to
3383                  * take over the slave's change_mtu
3384                  * function for the duration of their
3385                  * servitude.
3386                  */
3387                 break;
3388         case NETDEV_CHANGENAME:
3389                 /*
3390                  * TODO: handle changing the primary's name
3391                  */
3392                 break;
3393         case NETDEV_FEAT_CHANGE:
3394                 bond_compute_features(bond);
3395                 break;
3396         default:
3397                 break;
3398         }
3399
3400         return NOTIFY_DONE;
3401 }
3402
3403 /*
3404  * bond_netdev_event: handle netdev notifier chain events.
3405  *
3406  * This function receives events for the netdev chain.  The caller (an
3407  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3408  * locks for us to safely manipulate the slave devices (RTNL lock,
3409  * dev_probe_lock).
3410  */
3411 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3412 {
3413         struct net_device *event_dev = (struct net_device *)ptr;
3414
3415         if (event_dev->nd_net != &init_net)
3416                 return NOTIFY_DONE;
3417
3418         dprintk("event_dev: %s, event: %lx\n",
3419                 (event_dev ? event_dev->name : "None"),
3420                 event);
3421
3422         if (!(event_dev->priv_flags & IFF_BONDING))
3423                 return NOTIFY_DONE;
3424
3425         if (event_dev->flags & IFF_MASTER) {
3426                 dprintk("IFF_MASTER\n");
3427                 return bond_master_netdev_event(event, event_dev);
3428         }
3429
3430         if (event_dev->flags & IFF_SLAVE) {
3431                 dprintk("IFF_SLAVE\n");
3432                 return bond_slave_netdev_event(event, event_dev);
3433         }
3434
3435         return NOTIFY_DONE;
3436 }
3437
3438 /*
3439  * bond_inetaddr_event: handle inetaddr notifier chain events.
3440  *
3441  * We keep track of device IPs primarily to use as source addresses in
3442  * ARP monitor probes (rather than spewing out broadcasts all the time).
3443  *
3444  * We track one IP for the main device (if it has one), plus one per VLAN.
3445  */
3446 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3447 {
3448         struct in_ifaddr *ifa = ptr;
3449         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3450         struct bonding *bond, *bond_next;
3451         struct vlan_entry *vlan, *vlan_next;
3452
3453         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3454                 if (bond->dev == event_dev) {
3455                         switch (event) {
3456                         case NETDEV_UP:
3457                                 bond->master_ip = ifa->ifa_local;
3458                                 return NOTIFY_OK;
3459                         case NETDEV_DOWN:
3460                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3461                                 return NOTIFY_OK;
3462                         default:
3463                                 return NOTIFY_DONE;
3464                         }
3465                 }
3466
3467                 if (list_empty(&bond->vlan_list))
3468                         continue;
3469
3470                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3471                                          vlan_list) {
3472                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3473                         if (vlan_dev == event_dev) {
3474                                 switch (event) {
3475                                 case NETDEV_UP:
3476                                         vlan->vlan_ip = ifa->ifa_local;
3477                                         return NOTIFY_OK;
3478                                 case NETDEV_DOWN:
3479                                         vlan->vlan_ip =
3480                                                 bond_glean_dev_ip(vlan_dev);
3481                                         return NOTIFY_OK;
3482                                 default:
3483                                         return NOTIFY_DONE;
3484                                 }
3485                         }
3486                 }
3487         }
3488         return NOTIFY_DONE;
3489 }
3490
3491 static struct notifier_block bond_netdev_notifier = {
3492         .notifier_call = bond_netdev_event,
3493 };
3494
3495 static struct notifier_block bond_inetaddr_notifier = {
3496         .notifier_call = bond_inetaddr_event,
3497 };
3498
3499 /*-------------------------- Packet type handling ---------------------------*/
3500
3501 /* register to receive lacpdus on a bond */
3502 static void bond_register_lacpdu(struct bonding *bond)
3503 {
3504         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3505
3506         /* initialize packet type */
3507         pk_type->type = PKT_TYPE_LACPDU;
3508         pk_type->dev = bond->dev;
3509         pk_type->func = bond_3ad_lacpdu_recv;
3510
3511         dev_add_pack(pk_type);
3512 }
3513
3514 /* unregister to receive lacpdus on a bond */
3515 static void bond_unregister_lacpdu(struct bonding *bond)
3516 {
3517         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3518 }
3519
3520 void bond_register_arp(struct bonding *bond)
3521 {
3522         struct packet_type *pt = &bond->arp_mon_pt;
3523
3524         if (pt->type)
3525                 return;
3526
3527         pt->type = htons(ETH_P_ARP);
3528         pt->dev = bond->dev;
3529         pt->func = bond_arp_rcv;
3530         dev_add_pack(pt);
3531 }
3532
3533 void bond_unregister_arp(struct bonding *bond)
3534 {
3535         struct packet_type *pt = &bond->arp_mon_pt;
3536
3537         dev_remove_pack(pt);
3538         pt->type = 0;
3539 }
3540
3541 /*---------------------------- Hashing Policies -----------------------------*/
3542
3543 /*
3544  * Hash for the output device based upon layer 3 and layer 4 data. If
3545  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3546  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3547  */
3548 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3549                                     struct net_device *bond_dev, int count)
3550 {
3551         struct ethhdr *data = (struct ethhdr *)skb->data;
3552         struct iphdr *iph = ip_hdr(skb);
3553         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3554         int layer4_xor = 0;
3555
3556         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3557                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3558                     (iph->protocol == IPPROTO_TCP ||
3559                      iph->protocol == IPPROTO_UDP)) {
3560                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3561                 }
3562                 return (layer4_xor ^
3563                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3564
3565         }
3566
3567         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3568 }
3569
3570 /*
3571  * Hash for the output device based upon layer 2 data
3572  */
3573 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3574                                    struct net_device *bond_dev, int count)
3575 {
3576         struct ethhdr *data = (struct ethhdr *)skb->data;
3577
3578         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3579 }
3580
3581 /*-------------------------- Device entry points ----------------------------*/
3582
3583 static int bond_open(struct net_device *bond_dev)
3584 {
3585         struct bonding *bond = bond_dev->priv;
3586
3587         bond->kill_timers = 0;
3588
3589         if ((bond->params.mode == BOND_MODE_TLB) ||
3590             (bond->params.mode == BOND_MODE_ALB)) {
3591                 /* bond_alb_initialize must be called before the timer
3592                  * is started.
3593                  */
3594                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3595                         /* something went wrong - fail the open operation */
3596                         return -1;
3597                 }
3598
3599                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3600                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3601         }
3602
3603         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3604                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3605                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3606         }
3607
3608         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3609                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3610                         INIT_DELAYED_WORK(&bond->arp_work,
3611                                           bond_activebackup_arp_mon);
3612                 else
3613                         INIT_DELAYED_WORK(&bond->arp_work,
3614                                           bond_loadbalance_arp_mon);
3615
3616                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3617                 if (bond->params.arp_validate)
3618                         bond_register_arp(bond);
3619         }
3620
3621         if (bond->params.mode == BOND_MODE_8023AD) {
3622                 INIT_DELAYED_WORK(&bond->ad_work, bond_alb_monitor);
3623                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3624                 /* register to receive LACPDUs */
3625                 bond_register_lacpdu(bond);
3626         }
3627
3628         return 0;
3629 }
3630
3631 static int bond_close(struct net_device *bond_dev)
3632 {
3633         struct bonding *bond = bond_dev->priv;
3634
3635         if (bond->params.mode == BOND_MODE_8023AD) {
3636                 /* Unregister the receive of LACPDUs */
3637                 bond_unregister_lacpdu(bond);
3638         }
3639
3640         if (bond->params.arp_validate)
3641                 bond_unregister_arp(bond);
3642
3643         write_lock_bh(&bond->lock);
3644
3645
3646         /* signal timers not to re-arm */
3647         bond->kill_timers = 1;
3648
3649         write_unlock_bh(&bond->lock);
3650
3651         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3652                 cancel_delayed_work(&bond->mii_work);
3653         }
3654
3655         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3656                 cancel_delayed_work(&bond->arp_work);
3657         }
3658
3659         switch (bond->params.mode) {
3660         case BOND_MODE_8023AD:
3661                 cancel_delayed_work(&bond->ad_work);
3662                 break;
3663         case BOND_MODE_TLB:
3664         case BOND_MODE_ALB:
3665                 cancel_delayed_work(&bond->alb_work);
3666                 break;
3667         default:
3668                 break;
3669         }
3670
3671
3672         if ((bond->params.mode == BOND_MODE_TLB) ||
3673             (bond->params.mode == BOND_MODE_ALB)) {
3674                 /* Must be called only after all
3675                  * slaves have been released
3676                  */
3677                 bond_alb_deinitialize(bond);
3678         }
3679
3680         return 0;
3681 }
3682
3683 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3684 {
3685         struct bonding *bond = bond_dev->priv;
3686         struct net_device_stats *stats = &(bond->stats), *sstats;
3687         struct slave *slave;
3688         int i;
3689
3690         memset(stats, 0, sizeof(struct net_device_stats));
3691
3692         read_lock_bh(&bond->lock);
3693
3694         bond_for_each_slave(bond, slave, i) {
3695                 sstats = slave->dev->get_stats(slave->dev);
3696                 stats->rx_packets += sstats->rx_packets;
3697                 stats->rx_bytes += sstats->rx_bytes;
3698                 stats->rx_errors += sstats->rx_errors;
3699                 stats->rx_dropped += sstats->rx_dropped;
3700
3701                 stats->tx_packets += sstats->tx_packets;
3702                 stats->tx_bytes += sstats->tx_bytes;
3703                 stats->tx_errors += sstats->tx_errors;
3704                 stats->tx_dropped += sstats->tx_dropped;
3705
3706                 stats->multicast += sstats->multicast;
3707                 stats->collisions += sstats->collisions;
3708
3709                 stats->rx_length_errors += sstats->rx_length_errors;
3710                 stats->rx_over_errors += sstats->rx_over_errors;
3711                 stats->rx_crc_errors += sstats->rx_crc_errors;
3712                 stats->rx_frame_errors += sstats->rx_frame_errors;
3713                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3714                 stats->rx_missed_errors += sstats->rx_missed_errors;
3715
3716                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3717                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3718                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3719                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3720                 stats->tx_window_errors += sstats->tx_window_errors;
3721         }
3722
3723         read_unlock_bh(&bond->lock);
3724
3725         return stats;
3726 }
3727
3728 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3729 {
3730         struct net_device *slave_dev = NULL;
3731         struct ifbond k_binfo;
3732         struct ifbond __user *u_binfo = NULL;
3733         struct ifslave k_sinfo;
3734         struct ifslave __user *u_sinfo = NULL;
3735         struct mii_ioctl_data *mii = NULL;
3736         int res = 0;
3737
3738         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3739                 bond_dev->name, cmd);
3740
3741         switch (cmd) {
3742         case SIOCGMIIPHY:
3743                 mii = if_mii(ifr);
3744                 if (!mii) {
3745                         return -EINVAL;
3746                 }
3747                 mii->phy_id = 0;
3748                 /* Fall Through */
3749         case SIOCGMIIREG:
3750                 /*
3751                  * We do this again just in case we were called by SIOCGMIIREG
3752                  * instead of SIOCGMIIPHY.
3753                  */
3754                 mii = if_mii(ifr);
3755                 if (!mii) {
3756                         return -EINVAL;
3757                 }
3758
3759                 if (mii->reg_num == 1) {
3760                         struct bonding *bond = bond_dev->priv;
3761                         mii->val_out = 0;
3762                         read_lock_bh(&bond->lock);
3763                         read_lock(&bond->curr_slave_lock);
3764                         if (netif_carrier_ok(bond->dev)) {
3765                                 mii->val_out = BMSR_LSTATUS;
3766                         }
3767                         read_unlock(&bond->curr_slave_lock);
3768                         read_unlock_bh(&bond->lock);
3769                 }
3770
3771                 return 0;
3772         case BOND_INFO_QUERY_OLD:
3773         case SIOCBONDINFOQUERY:
3774                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3775
3776                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3777                         return -EFAULT;
3778                 }
3779
3780                 res = bond_info_query(bond_dev, &k_binfo);
3781                 if (res == 0) {
3782                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3783                                 return -EFAULT;
3784                         }
3785                 }
3786
3787                 return res;
3788         case BOND_SLAVE_INFO_QUERY_OLD:
3789         case SIOCBONDSLAVEINFOQUERY:
3790                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3791
3792                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3793                         return -EFAULT;
3794                 }
3795
3796                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3797                 if (res == 0) {
3798                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3799                                 return -EFAULT;
3800                         }
3801                 }
3802
3803                 return res;
3804         default:
3805                 /* Go on */
3806                 break;
3807         }
3808
3809         if (!capable(CAP_NET_ADMIN)) {
3810                 return -EPERM;
3811         }
3812
3813         down_write(&(bonding_rwsem));
3814         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3815
3816         dprintk("slave_dev=%p: \n", slave_dev);
3817
3818         if (!slave_dev) {
3819                 res = -ENODEV;
3820         } else {
3821                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3822                 switch (cmd) {
3823                 case BOND_ENSLAVE_OLD:
3824                 case SIOCBONDENSLAVE:
3825                         res = bond_enslave(bond_dev, slave_dev);
3826                         break;
3827                 case BOND_RELEASE_OLD:
3828                 case SIOCBONDRELEASE:
3829                         res = bond_release(bond_dev, slave_dev);
3830                         break;
3831                 case BOND_SETHWADDR_OLD:
3832                 case SIOCBONDSETHWADDR:
3833                         res = bond_sethwaddr(bond_dev, slave_dev);
3834                         break;
3835                 case BOND_CHANGE_ACTIVE_OLD:
3836                 case SIOCBONDCHANGEACTIVE:
3837                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3838                         break;
3839                 default:
3840                         res = -EOPNOTSUPP;
3841                 }
3842
3843                 dev_put(slave_dev);
3844         }
3845
3846         up_write(&(bonding_rwsem));
3847         return res;
3848 }
3849
3850 static void bond_set_multicast_list(struct net_device *bond_dev)
3851 {
3852         struct bonding *bond = bond_dev->priv;
3853         struct dev_mc_list *dmi;
3854
3855         write_lock_bh(&bond->lock);
3856
3857         /*
3858          * Do promisc before checking multicast_mode
3859          */
3860         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3861                 bond_set_promiscuity(bond, 1);
3862         }
3863
3864         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3865                 bond_set_promiscuity(bond, -1);
3866         }
3867
3868         /* set allmulti flag to slaves */
3869         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3870                 bond_set_allmulti(bond, 1);
3871         }
3872
3873         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3874                 bond_set_allmulti(bond, -1);
3875         }
3876
3877         bond->flags = bond_dev->flags;
3878
3879         /* looking for addresses to add to slaves' mc list */
3880         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3881                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3882                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3883                 }
3884         }
3885
3886         /* looking for addresses to delete from slaves' list */
3887         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3888                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3889                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3890                 }
3891         }
3892
3893         /* save master's multicast list */
3894         bond_mc_list_destroy(bond);
3895         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3896
3897         write_unlock_bh(&bond->lock);
3898 }
3899
3900 /*
3901  * Change the MTU of all of a master's slaves to match the master
3902  */
3903 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3904 {
3905         struct bonding *bond = bond_dev->priv;
3906         struct slave *slave, *stop_at;
3907         int res = 0;
3908         int i;
3909
3910         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3911                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3912
3913         /* Can't hold bond->lock with bh disabled here since
3914          * some base drivers panic. On the other hand we can't
3915          * hold bond->lock without bh disabled because we'll
3916          * deadlock. The only solution is to rely on the fact
3917          * that we're under rtnl_lock here, and the slaves
3918          * list won't change. This doesn't solve the problem
3919          * of setting the slave's MTU while it is
3920          * transmitting, but the assumption is that the base
3921          * driver can handle that.
3922          *
3923          * TODO: figure out a way to safely iterate the slaves
3924          * list, but without holding a lock around the actual
3925          * call to the base driver.
3926          */
3927
3928         bond_for_each_slave(bond, slave, i) {
3929                 dprintk("s %p s->p %p c_m %p\n", slave,
3930                         slave->prev, slave->dev->change_mtu);
3931
3932                 res = dev_set_mtu(slave->dev, new_mtu);
3933
3934                 if (res) {
3935                         /* If we failed to set the slave's mtu to the new value
3936                          * we must abort the operation even in ACTIVE_BACKUP
3937                          * mode, because if we allow the backup slaves to have
3938                          * different mtu values than the active slave we'll
3939                          * need to change their mtu when doing a failover. That
3940                          * means changing their mtu from timer context, which
3941                          * is probably not a good idea.
3942                          */
3943                         dprintk("err %d %s\n", res, slave->dev->name);
3944                         goto unwind;
3945                 }
3946         }
3947
3948         bond_dev->mtu = new_mtu;
3949
3950         return 0;
3951
3952 unwind:
3953         /* unwind from head to the slave that failed */
3954         stop_at = slave;
3955         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3956                 int tmp_res;
3957
3958                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3959                 if (tmp_res) {
3960                         dprintk("unwind err %d dev %s\n", tmp_res,
3961                                 slave->dev->name);
3962                 }
3963         }
3964
3965         return res;
3966 }
3967
3968 /*
3969  * Change HW address
3970  *
3971  * Note that many devices must be down to change the HW address, and
3972  * downing the master releases all slaves.  We can make bonds full of
3973  * bonding devices to test this, however.
3974  */
3975 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3976 {
3977         struct bonding *bond = bond_dev->priv;
3978         struct sockaddr *sa = addr, tmp_sa;
3979         struct slave *slave, *stop_at;
3980         int res = 0;
3981         int i;
3982
3983         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3984
3985         /*
3986          * If fail_over_mac is enabled, do nothing and return success.
3987          * Returning an error causes ifenslave to fail.
3988          */
3989         if (bond->params.fail_over_mac)
3990                 return 0;
3991
3992         if (!is_valid_ether_addr(sa->sa_data)) {
3993                 return -EADDRNOTAVAIL;
3994         }
3995
3996         /* Can't hold bond->lock with bh disabled here since
3997          * some base drivers panic. On the other hand we can't
3998          * hold bond->lock without bh disabled because we'll
3999          * deadlock. The only solution is to rely on the fact
4000          * that we're under rtnl_lock here, and the slaves
4001          * list won't change. This doesn't solve the problem
4002          * of setting the slave's hw address while it is
4003          * transmitting, but the assumption is that the base
4004          * driver can handle that.
4005          *
4006          * TODO: figure out a way to safely iterate the slaves
4007          * list, but without holding a lock around the actual
4008          * call to the base driver.
4009          */
4010
4011         bond_for_each_slave(bond, slave, i) {
4012                 dprintk("slave %p %s\n", slave, slave->dev->name);
4013
4014                 if (slave->dev->set_mac_address == NULL) {
4015                         res = -EOPNOTSUPP;
4016                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4017                         goto unwind;
4018                 }
4019
4020                 res = dev_set_mac_address(slave->dev, addr);
4021                 if (res) {
4022                         /* TODO: consider downing the slave
4023                          * and retry ?
4024                          * User should expect communications
4025                          * breakage anyway until ARP finish
4026                          * updating, so...
4027                          */
4028                         dprintk("err %d %s\n", res, slave->dev->name);
4029                         goto unwind;
4030                 }
4031         }
4032
4033         /* success */
4034         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4035         return 0;
4036
4037 unwind:
4038         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4039         tmp_sa.sa_family = bond_dev->type;
4040
4041         /* unwind from head to the slave that failed */
4042         stop_at = slave;
4043         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4044                 int tmp_res;
4045
4046                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4047                 if (tmp_res) {
4048                         dprintk("unwind err %d dev %s\n", tmp_res,
4049                                 slave->dev->name);
4050                 }
4051         }
4052
4053         return res;
4054 }
4055
4056 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4057 {
4058         struct bonding *bond = bond_dev->priv;
4059         struct slave *slave, *start_at;
4060         int i;
4061         int res = 1;
4062
4063         read_lock(&bond->lock);
4064
4065         if (!BOND_IS_OK(bond)) {
4066                 goto out;
4067         }
4068
4069         read_lock(&bond->curr_slave_lock);
4070         slave = start_at = bond->curr_active_slave;
4071         read_unlock(&bond->curr_slave_lock);
4072
4073         if (!slave) {
4074                 goto out;
4075         }
4076
4077         bond_for_each_slave_from(bond, slave, i, start_at) {
4078                 if (IS_UP(slave->dev) &&
4079                     (slave->link == BOND_LINK_UP) &&
4080                     (slave->state == BOND_STATE_ACTIVE)) {
4081                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4082
4083                         write_lock(&bond->curr_slave_lock);
4084                         bond->curr_active_slave = slave->next;
4085                         write_unlock(&bond->curr_slave_lock);
4086
4087                         break;
4088                 }
4089         }
4090
4091
4092 out:
4093         if (res) {
4094                 /* no suitable interface, frame not sent */
4095                 dev_kfree_skb(skb);
4096         }
4097         read_unlock(&bond->lock);
4098         return 0;
4099 }
4100
4101
4102 /*
4103  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4104  * the bond has a usable interface.
4105  */
4106 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4107 {
4108         struct bonding *bond = bond_dev->priv;
4109         int res = 1;
4110
4111         read_lock(&bond->lock);
4112         read_lock(&bond->curr_slave_lock);
4113
4114         if (!BOND_IS_OK(bond)) {
4115                 goto out;
4116         }
4117
4118         if (!bond->curr_active_slave)
4119                 goto out;
4120
4121         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4122
4123 out:
4124         if (res) {
4125                 /* no suitable interface, frame not sent */
4126                 dev_kfree_skb(skb);
4127         }
4128         read_unlock(&bond->curr_slave_lock);
4129         read_unlock(&bond->lock);
4130         return 0;
4131 }
4132
4133 /*
4134  * In bond_xmit_xor() , we determine the output device by using a pre-
4135  * determined xmit_hash_policy(), If the selected device is not enabled,
4136  * find the next active slave.
4137  */
4138 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4139 {
4140         struct bonding *bond = bond_dev->priv;
4141         struct slave *slave, *start_at;
4142         int slave_no;
4143         int i;
4144         int res = 1;
4145
4146         read_lock(&bond->lock);
4147
4148         if (!BOND_IS_OK(bond)) {
4149                 goto out;
4150         }
4151
4152         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4153
4154         bond_for_each_slave(bond, slave, i) {
4155                 slave_no--;
4156                 if (slave_no < 0) {
4157                         break;
4158                 }
4159         }
4160
4161         start_at = slave;
4162
4163         bond_for_each_slave_from(bond, slave, i, start_at) {
4164                 if (IS_UP(slave->dev) &&
4165                     (slave->link == BOND_LINK_UP) &&
4166                     (slave->state == BOND_STATE_ACTIVE)) {
4167                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4168                         break;
4169                 }
4170         }
4171
4172 out:
4173         if (res) {
4174                 /* no suitable interface, frame not sent */
4175                 dev_kfree_skb(skb);
4176         }
4177         read_unlock(&bond->lock);
4178         return 0;
4179 }
4180
4181 /*
4182  * in broadcast mode, we send everything to all usable interfaces.
4183  */
4184 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4185 {
4186         struct bonding *bond = bond_dev->priv;
4187         struct slave *slave, *start_at;
4188         struct net_device *tx_dev = NULL;
4189         int i;
4190         int res = 1;
4191
4192         read_lock(&bond->lock);
4193
4194         if (!BOND_IS_OK(bond)) {
4195                 goto out;
4196         }
4197
4198         read_lock(&bond->curr_slave_lock);
4199         start_at = bond->curr_active_slave;
4200         read_unlock(&bond->curr_slave_lock);
4201
4202         if (!start_at) {
4203                 goto out;
4204         }
4205
4206         bond_for_each_slave_from(bond, slave, i, start_at) {
4207                 if (IS_UP(slave->dev) &&
4208                     (slave->link == BOND_LINK_UP) &&
4209                     (slave->state == BOND_STATE_ACTIVE)) {
4210                         if (tx_dev) {
4211                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4212                                 if (!skb2) {
4213                                         printk(KERN_ERR DRV_NAME
4214                                                ": %s: Error: bond_xmit_broadcast(): "
4215                                                "skb_clone() failed\n",
4216                                                bond_dev->name);
4217                                         continue;
4218                                 }
4219
4220                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4221                                 if (res) {
4222                                         dev_kfree_skb(skb2);
4223                                         continue;
4224                                 }
4225                         }
4226                         tx_dev = slave->dev;
4227                 }
4228         }
4229
4230         if (tx_dev) {
4231                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4232         }
4233
4234 out:
4235         if (res) {
4236                 /* no suitable interface, frame not sent */
4237                 dev_kfree_skb(skb);
4238         }
4239         /* frame sent to all suitable interfaces */
4240         read_unlock(&bond->lock);
4241         return 0;
4242 }
4243
4244 /*------------------------- Device initialization ---------------------------*/
4245
4246 /*
4247  * set bond mode specific net device operations
4248  */
4249 void bond_set_mode_ops(struct bonding *bond, int mode)
4250 {
4251         struct net_device *bond_dev = bond->dev;
4252
4253         switch (mode) {
4254         case BOND_MODE_ROUNDROBIN:
4255                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4256                 break;
4257         case BOND_MODE_ACTIVEBACKUP:
4258                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4259                 break;
4260         case BOND_MODE_XOR:
4261                 bond_dev->hard_start_xmit = bond_xmit_xor;
4262                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4263                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4264                 else
4265                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4266                 break;
4267         case BOND_MODE_BROADCAST:
4268                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4269                 break;
4270         case BOND_MODE_8023AD:
4271                 bond_set_master_3ad_flags(bond);
4272                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4273                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4274                         bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4275                 else
4276                         bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4277                 break;
4278         case BOND_MODE_ALB:
4279                 bond_set_master_alb_flags(bond);
4280                 /* FALLTHRU */
4281         case BOND_MODE_TLB:
4282                 bond_dev->hard_start_xmit = bond_alb_xmit;
4283                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4284                 break;
4285         default:
4286                 /* Should never happen, mode already checked */
4287                 printk(KERN_ERR DRV_NAME
4288                        ": %s: Error: Unknown bonding mode %d\n",
4289                        bond_dev->name,
4290                        mode);
4291                 break;
4292         }
4293 }
4294
4295 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4296                                     struct ethtool_drvinfo *drvinfo)
4297 {
4298         strncpy(drvinfo->driver, DRV_NAME, 32);
4299         strncpy(drvinfo->version, DRV_VERSION, 32);
4300         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4301 }
4302
4303 static const struct ethtool_ops bond_ethtool_ops = {
4304         .get_drvinfo            = bond_ethtool_get_drvinfo,
4305 };
4306
4307 /*
4308  * Does not allocate but creates a /proc entry.
4309  * Allowed to fail.
4310  */
4311 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4312 {
4313         struct bonding *bond = bond_dev->priv;
4314
4315         dprintk("Begin bond_init for %s\n", bond_dev->name);
4316
4317         /* initialize rwlocks */
4318         rwlock_init(&bond->lock);
4319         rwlock_init(&bond->curr_slave_lock);
4320
4321         bond->params = *params; /* copy params struct */
4322
4323         bond->wq = create_singlethread_workqueue(bond_dev->name);
4324         if (!bond->wq)
4325                 return -ENOMEM;
4326
4327         /* Initialize pointers */
4328         bond->first_slave = NULL;
4329         bond->curr_active_slave = NULL;
4330         bond->current_arp_slave = NULL;
4331         bond->primary_slave = NULL;
4332         bond->dev = bond_dev;
4333         bond->send_grat_arp = 0;
4334         bond->setup_by_slave = 0;
4335         INIT_LIST_HEAD(&bond->vlan_list);
4336
4337         /* Initialize the device entry points */
4338         bond_dev->open = bond_open;
4339         bond_dev->stop = bond_close;
4340         bond_dev->get_stats = bond_get_stats;
4341         bond_dev->do_ioctl = bond_do_ioctl;
4342         bond_dev->ethtool_ops = &bond_ethtool_ops;
4343         bond_dev->set_multicast_list = bond_set_multicast_list;
4344         bond_dev->change_mtu = bond_change_mtu;
4345         bond_dev->set_mac_address = bond_set_mac_address;
4346
4347         bond_set_mode_ops(bond, bond->params.mode);
4348
4349         bond_dev->destructor = free_netdev;
4350
4351         /* Initialize the device options */
4352         bond_dev->tx_queue_len = 0;
4353         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4354         bond_dev->priv_flags |= IFF_BONDING;
4355
4356         /* At first, we block adding VLANs. That's the only way to
4357          * prevent problems that occur when adding VLANs over an
4358          * empty bond. The block will be removed once non-challenged
4359          * slaves are enslaved.
4360          */
4361         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4362
4363         /* don't acquire bond device's netif_tx_lock when
4364          * transmitting */
4365         bond_dev->features |= NETIF_F_LLTX;
4366
4367         /* By default, we declare the bond to be fully
4368          * VLAN hardware accelerated capable. Special
4369          * care is taken in the various xmit functions
4370          * when there are slaves that are not hw accel
4371          * capable
4372          */
4373         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4374         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4375         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4376         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4377                                NETIF_F_HW_VLAN_RX |
4378                                NETIF_F_HW_VLAN_FILTER);
4379
4380 #ifdef CONFIG_PROC_FS
4381         bond_create_proc_entry(bond);
4382 #endif
4383         list_add_tail(&bond->bond_list, &bond_dev_list);
4384
4385         return 0;
4386 }
4387
4388 /* De-initialize device specific data.
4389  * Caller must hold rtnl_lock.
4390  */
4391 void bond_deinit(struct net_device *bond_dev)
4392 {
4393         struct bonding *bond = bond_dev->priv;
4394
4395         list_del(&bond->bond_list);
4396
4397 #ifdef CONFIG_PROC_FS
4398         bond_remove_proc_entry(bond);
4399 #endif
4400 }
4401
4402 /* Unregister and free all bond devices.
4403  * Caller must hold rtnl_lock.
4404  */
4405 static void bond_free_all(void)
4406 {
4407         struct bonding *bond, *nxt;
4408
4409         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4410                 struct net_device *bond_dev = bond->dev;
4411
4412                 bond_mc_list_destroy(bond);
4413                 /* Release the bonded slaves */
4414                 bond_release_all(bond_dev);
4415                 bond_deinit(bond_dev);
4416                 unregister_netdevice(bond_dev);
4417         }
4418
4419 #ifdef CONFIG_PROC_FS
4420         bond_destroy_proc_dir();
4421 #endif
4422 }
4423
4424 /*------------------------- Module initialization ---------------------------*/
4425
4426 /*
4427  * Convert string input module parms.  Accept either the
4428  * number of the mode or its string name.
4429  */
4430 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4431 {
4432         int i;
4433
4434         for (i = 0; tbl[i].modename; i++) {
4435                 if ((isdigit(*mode_arg) &&
4436                      tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4437                     (strncmp(mode_arg, tbl[i].modename,
4438                              strlen(tbl[i].modename)) == 0)) {
4439                         return tbl[i].mode;
4440                 }
4441         }
4442
4443         return -1;
4444 }
4445
4446 static int bond_check_params(struct bond_params *params)
4447 {
4448         int arp_validate_value;
4449
4450         /*
4451          * Convert string parameters.
4452          */
4453         if (mode) {
4454                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4455                 if (bond_mode == -1) {
4456                         printk(KERN_ERR DRV_NAME
4457                                ": Error: Invalid bonding mode \"%s\"\n",
4458                                mode == NULL ? "NULL" : mode);
4459                         return -EINVAL;
4460                 }
4461         }
4462
4463         if (xmit_hash_policy) {
4464                 if ((bond_mode != BOND_MODE_XOR) &&
4465                     (bond_mode != BOND_MODE_8023AD)) {
4466                         printk(KERN_INFO DRV_NAME
4467                                ": xor_mode param is irrelevant in mode %s\n",
4468                                bond_mode_name(bond_mode));
4469                 } else {
4470                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4471                                                         xmit_hashtype_tbl);
4472                         if (xmit_hashtype == -1) {
4473                                 printk(KERN_ERR DRV_NAME
4474                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4475                                 xmit_hash_policy == NULL ? "NULL" :
4476                                        xmit_hash_policy);
4477                                 return -EINVAL;
4478                         }
4479                 }
4480         }
4481
4482         if (lacp_rate) {
4483                 if (bond_mode != BOND_MODE_8023AD) {
4484                         printk(KERN_INFO DRV_NAME
4485                                ": lacp_rate param is irrelevant in mode %s\n",
4486                                bond_mode_name(bond_mode));
4487                 } else {
4488                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4489                         if (lacp_fast == -1) {
4490                                 printk(KERN_ERR DRV_NAME
4491                                        ": Error: Invalid lacp rate \"%s\"\n",
4492                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4493                                 return -EINVAL;
4494                         }
4495                 }
4496         }
4497
4498         if (max_bonds < 1 || max_bonds > INT_MAX) {
4499                 printk(KERN_WARNING DRV_NAME
4500                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4501                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4502                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4503                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4504         }
4505
4506         if (miimon < 0) {
4507                 printk(KERN_WARNING DRV_NAME
4508                        ": Warning: miimon module parameter (%d), "
4509                        "not in range 0-%d, so it was reset to %d\n",
4510                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4511                 miimon = BOND_LINK_MON_INTERV;
4512         }
4513
4514         if (updelay < 0) {
4515                 printk(KERN_WARNING DRV_NAME
4516                        ": Warning: updelay module parameter (%d), "
4517                        "not in range 0-%d, so it was reset to 0\n",
4518                        updelay, INT_MAX);
4519                 updelay = 0;
4520         }
4521
4522         if (downdelay < 0) {
4523                 printk(KERN_WARNING DRV_NAME
4524                        ": Warning: downdelay module parameter (%d), "
4525                        "not in range 0-%d, so it was reset to 0\n",
4526                        downdelay, INT_MAX);
4527                 downdelay = 0;
4528         }
4529
4530         if ((use_carrier != 0) && (use_carrier != 1)) {
4531                 printk(KERN_WARNING DRV_NAME
4532                        ": Warning: use_carrier module parameter (%d), "
4533                        "not of valid value (0/1), so it was set to 1\n",
4534                        use_carrier);
4535                 use_carrier = 1;
4536         }
4537
4538         /* reset values for 802.3ad */
4539         if (bond_mode == BOND_MODE_8023AD) {
4540                 if (!miimon) {
4541                         printk(KERN_WARNING DRV_NAME
4542                                ": Warning: miimon must be specified, "
4543                                "otherwise bonding will not detect link "
4544                                "failure, speed and duplex which are "
4545                                "essential for 802.3ad operation\n");
4546                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4547                         miimon = 100;
4548                 }
4549         }
4550
4551         /* reset values for TLB/ALB */
4552         if ((bond_mode == BOND_MODE_TLB) ||
4553             (bond_mode == BOND_MODE_ALB)) {
4554                 if (!miimon) {
4555                         printk(KERN_WARNING DRV_NAME
4556                                ": Warning: miimon must be specified, "
4557                                "otherwise bonding will not detect link "
4558                                "failure and link speed which are essential "
4559                                "for TLB/ALB load balancing\n");
4560                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4561                         miimon = 100;
4562                 }
4563         }
4564
4565         if (bond_mode == BOND_MODE_ALB) {
4566                 printk(KERN_NOTICE DRV_NAME
4567                        ": In ALB mode you might experience client "
4568                        "disconnections upon reconnection of a link if the "
4569                        "bonding module updelay parameter (%d msec) is "
4570                        "incompatible with the forwarding delay time of the "
4571                        "switch\n",
4572                        updelay);
4573         }
4574
4575         if (!miimon) {
4576                 if (updelay || downdelay) {
4577                         /* just warn the user the up/down delay will have
4578                          * no effect since miimon is zero...
4579                          */
4580                         printk(KERN_WARNING DRV_NAME
4581                                ": Warning: miimon module parameter not set "
4582                                "and updelay (%d) or downdelay (%d) module "
4583                                "parameter is set; updelay and downdelay have "
4584                                "no effect unless miimon is set\n",
4585                                updelay, downdelay);
4586                 }
4587         } else {
4588                 /* don't allow arp monitoring */
4589                 if (arp_interval) {
4590                         printk(KERN_WARNING DRV_NAME
4591                                ": Warning: miimon (%d) and arp_interval (%d) "
4592                                "can't be used simultaneously, disabling ARP "
4593                                "monitoring\n",
4594                                miimon, arp_interval);
4595                         arp_interval = 0;
4596                 }
4597
4598                 if ((updelay % miimon) != 0) {
4599                         printk(KERN_WARNING DRV_NAME
4600                                ": Warning: updelay (%d) is not a multiple "
4601                                "of miimon (%d), updelay rounded to %d ms\n",
4602                                updelay, miimon, (updelay / miimon) * miimon);
4603                 }
4604
4605                 updelay /= miimon;
4606
4607                 if ((downdelay % miimon) != 0) {
4608                         printk(KERN_WARNING DRV_NAME
4609                                ": Warning: downdelay (%d) is not a multiple "
4610                                "of miimon (%d), downdelay rounded to %d ms\n",
4611                                downdelay, miimon,
4612                                (downdelay / miimon) * miimon);
4613                 }
4614
4615                 downdelay /= miimon;
4616         }
4617
4618         if (arp_interval < 0) {
4619                 printk(KERN_WARNING DRV_NAME
4620                        ": Warning: arp_interval module parameter (%d) "
4621                        ", not in range 0-%d, so it was reset to %d\n",
4622                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4623                 arp_interval = BOND_LINK_ARP_INTERV;
4624         }
4625
4626         for (arp_ip_count = 0;
4627              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4628              arp_ip_count++) {
4629                 /* not complete check, but should be good enough to
4630                    catch mistakes */
4631                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4632                         printk(KERN_WARNING DRV_NAME
4633                                ": Warning: bad arp_ip_target module parameter "
4634                                "(%s), ARP monitoring will not be performed\n",
4635                                arp_ip_target[arp_ip_count]);
4636                         arp_interval = 0;
4637                 } else {
4638                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4639                         arp_target[arp_ip_count] = ip;
4640                 }
4641         }
4642
4643         if (arp_interval && !arp_ip_count) {
4644                 /* don't allow arping if no arp_ip_target given... */
4645                 printk(KERN_WARNING DRV_NAME
4646                        ": Warning: arp_interval module parameter (%d) "
4647                        "specified without providing an arp_ip_target "
4648                        "parameter, arp_interval was reset to 0\n",
4649                        arp_interval);
4650                 arp_interval = 0;
4651         }
4652
4653         if (arp_validate) {
4654                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4655                         printk(KERN_ERR DRV_NAME
4656                ": arp_validate only supported in active-backup mode\n");
4657                         return -EINVAL;
4658                 }
4659                 if (!arp_interval) {
4660                         printk(KERN_ERR DRV_NAME
4661                                ": arp_validate requires arp_interval\n");
4662                         return -EINVAL;
4663                 }
4664
4665                 arp_validate_value = bond_parse_parm(arp_validate,
4666                                                      arp_validate_tbl);
4667                 if (arp_validate_value == -1) {
4668                         printk(KERN_ERR DRV_NAME
4669                                ": Error: invalid arp_validate \"%s\"\n",
4670                                arp_validate == NULL ? "NULL" : arp_validate);
4671                         return -EINVAL;
4672                 }
4673         } else
4674                 arp_validate_value = 0;
4675
4676         if (miimon) {
4677                 printk(KERN_INFO DRV_NAME
4678                        ": MII link monitoring set to %d ms\n",
4679                        miimon);
4680         } else if (arp_interval) {
4681                 int i;
4682
4683                 printk(KERN_INFO DRV_NAME
4684                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4685                        arp_interval,
4686                        arp_validate_tbl[arp_validate_value].modename,
4687                        arp_ip_count);
4688
4689                 for (i = 0; i < arp_ip_count; i++)
4690                         printk (" %s", arp_ip_target[i]);
4691
4692                 printk("\n");
4693
4694         } else {
4695                 /* miimon and arp_interval not set, we need one so things
4696                  * work as expected, see bonding.txt for details
4697                  */
4698                 printk(KERN_WARNING DRV_NAME
4699                        ": Warning: either miimon or arp_interval and "
4700                        "arp_ip_target module parameters must be specified, "
4701                        "otherwise bonding will not detect link failures! see "
4702                        "bonding.txt for details.\n");
4703         }
4704
4705         if (primary && !USES_PRIMARY(bond_mode)) {
4706                 /* currently, using a primary only makes sense
4707                  * in active backup, TLB or ALB modes
4708                  */
4709                 printk(KERN_WARNING DRV_NAME
4710                        ": Warning: %s primary device specified but has no "
4711                        "effect in %s mode\n",
4712                        primary, bond_mode_name(bond_mode));
4713                 primary = NULL;
4714         }
4715
4716         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4717                 printk(KERN_WARNING DRV_NAME
4718                        ": Warning: fail_over_mac only affects "
4719                        "active-backup mode.\n");
4720
4721         /* fill params struct with the proper values */
4722         params->mode = bond_mode;
4723         params->xmit_policy = xmit_hashtype;
4724         params->miimon = miimon;
4725         params->arp_interval = arp_interval;
4726         params->arp_validate = arp_validate_value;
4727         params->updelay = updelay;
4728         params->downdelay = downdelay;
4729         params->use_carrier = use_carrier;
4730         params->lacp_fast = lacp_fast;
4731         params->primary[0] = 0;
4732         params->fail_over_mac = fail_over_mac;
4733
4734         if (primary) {
4735                 strncpy(params->primary, primary, IFNAMSIZ);
4736                 params->primary[IFNAMSIZ - 1] = 0;
4737         }
4738
4739         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4740
4741         return 0;
4742 }
4743
4744 static struct lock_class_key bonding_netdev_xmit_lock_key;
4745
4746 /* Create a new bond based on the specified name and bonding parameters.
4747  * If name is NULL, obtain a suitable "bond%d" name for us.
4748  * Caller must NOT hold rtnl_lock; we need to release it here before we
4749  * set up our sysfs entries.
4750  */
4751 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4752 {
4753         struct net_device *bond_dev;
4754         int res;
4755
4756         rtnl_lock();
4757         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4758                                 ether_setup);
4759         if (!bond_dev) {
4760                 printk(KERN_ERR DRV_NAME
4761                        ": %s: eek! can't alloc netdev!\n",
4762                        name);
4763                 res = -ENOMEM;
4764                 goto out_rtnl;
4765         }
4766
4767         if (!name) {
4768                 res = dev_alloc_name(bond_dev, "bond%d");
4769                 if (res < 0)
4770                         goto out_netdev;
4771         }
4772
4773         /* bond_init() must be called after dev_alloc_name() (for the
4774          * /proc files), but before register_netdevice(), because we
4775          * need to set function pointers.
4776          */
4777
4778         res = bond_init(bond_dev, params);
4779         if (res < 0) {
4780                 goto out_netdev;
4781         }
4782
4783         res = register_netdevice(bond_dev);
4784         if (res < 0) {
4785                 goto out_bond;
4786         }
4787
4788         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4789
4790         if (newbond)
4791                 *newbond = bond_dev->priv;
4792
4793         netif_carrier_off(bond_dev);
4794
4795         rtnl_unlock(); /* allows sysfs registration of net device */
4796         res = bond_create_sysfs_entry(bond_dev->priv);
4797         if (res < 0) {
4798                 rtnl_lock();
4799                 goto out_bond;
4800         }
4801
4802         return 0;
4803
4804 out_bond:
4805         bond_deinit(bond_dev);
4806 out_netdev:
4807         free_netdev(bond_dev);
4808 out_rtnl:
4809         rtnl_unlock();
4810         return res;
4811 }
4812
4813 static void bond_work_cancel_all(struct bonding *bond)
4814 {
4815         write_lock_bh(&bond->lock);
4816         bond->kill_timers = 1;
4817         write_unlock_bh(&bond->lock);
4818
4819         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4820                 cancel_delayed_work(&bond->mii_work);
4821
4822         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4823                 cancel_delayed_work(&bond->arp_work);
4824
4825         if (bond->params.mode == BOND_MODE_ALB &&
4826             delayed_work_pending(&bond->alb_work))
4827                 cancel_delayed_work(&bond->alb_work);
4828
4829         if (bond->params.mode == BOND_MODE_8023AD &&
4830             delayed_work_pending(&bond->ad_work))
4831                 cancel_delayed_work(&bond->ad_work);
4832 }
4833
4834 static int __init bonding_init(void)
4835 {
4836         int i;
4837         int res;
4838         struct bonding *bond, *nxt;
4839
4840         printk(KERN_INFO "%s", version);
4841
4842         res = bond_check_params(&bonding_defaults);
4843         if (res) {
4844                 goto out;
4845         }
4846
4847 #ifdef CONFIG_PROC_FS
4848         bond_create_proc_dir();
4849 #endif
4850         for (i = 0; i < max_bonds; i++) {
4851                 res = bond_create(NULL, &bonding_defaults, NULL);
4852                 if (res)
4853                         goto err;
4854         }
4855
4856         res = bond_create_sysfs();
4857         if (res)
4858                 goto err;
4859
4860         register_netdevice_notifier(&bond_netdev_notifier);
4861         register_inetaddr_notifier(&bond_inetaddr_notifier);
4862
4863         goto out;
4864 err:
4865         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4866                 bond_work_cancel_all(bond);
4867                 destroy_workqueue(bond->wq);
4868         }
4869
4870         rtnl_lock();
4871         bond_free_all();
4872         bond_destroy_sysfs();
4873         rtnl_unlock();
4874 out:
4875         return res;
4876
4877 }
4878
4879 static void __exit bonding_exit(void)
4880 {
4881         unregister_netdevice_notifier(&bond_netdev_notifier);
4882         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4883
4884         rtnl_lock();
4885         bond_free_all();
4886         bond_destroy_sysfs();
4887         rtnl_unlock();
4888 }
4889
4890 module_init(bonding_init);
4891 module_exit(bonding_exit);
4892 MODULE_LICENSE("GPL");
4893 MODULE_VERSION(DRV_VERSION);
4894 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4895 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4896 MODULE_SUPPORTED_DEVICE("most ethernet devices");
4897
4898 /*
4899  * Local variables:
4900  *  c-indent-level: 8
4901  *  c-basic-offset: 8
4902  *  tab-width: 8
4903  * End:
4904  */
4905