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

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 //#define BONDING_DEBUG 1
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <asm/system.h>
  57 #include <asm/io.h>
  58 #include <asm/dma.h>
  59 #include <asm/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <net/route.h>
  79 #include <net/net_namespace.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int num_grat_arp = 1;
  92 static int miimon       = BOND_LINK_MON_INTERV;
  93 static int updelay      = 0;
  94 static int downdelay    = 0;
  95 static int use_carrier  = 1;
  96 static char *mode       = NULL;
  97 static char *primary    = NULL;
  98 static char *lacp_rate  = NULL;
  99 static char *xmit_hash_policy = NULL;
 100 static int arp_interval = BOND_LINK_ARP_INTERV;
 101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 102 static char *arp_validate = NULL;
 103 static char *fail_over_mac = NULL;
 104 struct bond_params bonding_defaults;
 105
 106 module_param(max_bonds, int, 0);
 107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 108 module_param(num_grat_arp, int, 0644);
 109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 110 module_param(miimon, int, 0);
 111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 112 module_param(updelay, int, 0);
 113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 114 module_param(downdelay, int, 0);
 115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 116                             "in milliseconds");
 117 module_param(use_carrier, int, 0);
 118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 119                               "0 for off, 1 for on (default)");
 120 module_param(mode, charp, 0);
 121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 122                        "1 for active-backup, 2 for balance-xor, "
 123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 124                        "6 for balance-alb");
 125 module_param(primary, charp, 0);
 126 MODULE_PARM_DESC(primary, "Primary network device to use");
 127 module_param(lacp_rate, charp, 0);
 128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 129                             "(slow/fast)");
 130 module_param(xmit_hash_policy, charp, 0);
 131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 132                                    ", 1 for layer 3+4");
 133 module_param(arp_interval, int, 0);
 134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 135 module_param_array(arp_ip_target, charp, NULL, 0);
 136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 137 module_param(arp_validate, charp, 0);
 138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 139 module_param(fail_over_mac, charp, 0);
 140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 141
 142 /*----------------------------- Global variables ----------------------------*/
 143
 144 static const char * const version =
 145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 146
 147 LIST_HEAD(bond_dev_list);
 148
 149 #ifdef CONFIG_PROC_FS
 150 static struct proc_dir_entry *bond_proc_dir = NULL;
 151 #endif
 152
 153 extern struct rw_semaphore bonding_rwsem;
 154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 155 static int arp_ip_count = 0;
 156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 158 static int lacp_fast    = 0;
 159
 160
 161 struct bond_parm_tbl bond_lacp_tbl[] = {
 162 {       "slow",         AD_LACP_SLOW},
 163 {       "fast",         AD_LACP_FAST},
 164 {       NULL,           -1},
 165 };
 166
 167 struct bond_parm_tbl bond_mode_tbl[] = {
 168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 170 {       "balance-xor",          BOND_MODE_XOR},
 171 {       "broadcast",            BOND_MODE_BROADCAST},
 172 {       "802.3ad",              BOND_MODE_8023AD},
 173 {       "balance-tlb",          BOND_MODE_TLB},
 174 {       "balance-alb",          BOND_MODE_ALB},
 175 {       NULL,                   -1},
 176 };
 177
 178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 182 {       NULL,                   -1},
 183 };
 184
 185 struct bond_parm_tbl arp_validate_tbl[] = {
 186 {       "none",                 BOND_ARP_VALIDATE_NONE},
 187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 189 {       "all",                  BOND_ARP_VALIDATE_ALL},
 190 {       NULL,                   -1},
 191 };
 192
 193 struct bond_parm_tbl fail_over_mac_tbl[] = {
 194 {       "none",                 BOND_FOM_NONE},
 195 {       "active",               BOND_FOM_ACTIVE},
 196 {       "follow",               BOND_FOM_FOLLOW},
 197 {       NULL,                   -1},
 198 };
 199
 200 /*-------------------------- Forward declarations ---------------------------*/
 201
 202 static void bond_send_gratuitous_arp(struct bonding *bond);
 203 static void bond_deinit(struct net_device *bond_dev);
 204
 205 /*---------------------------- General routines -----------------------------*/
 206
 207 static const char *bond_mode_name(int mode)
 208 {
 209         switch (mode) {
 210         case BOND_MODE_ROUNDROBIN :
 211                 return "load balancing (round-robin)";
 212         case BOND_MODE_ACTIVEBACKUP :
 213                 return "fault-tolerance (active-backup)";
 214         case BOND_MODE_XOR :
 215                 return "load balancing (xor)";
 216         case BOND_MODE_BROADCAST :
 217                 return "fault-tolerance (broadcast)";
 218         case BOND_MODE_8023AD:
 219                 return "IEEE 802.3ad Dynamic link aggregation";
 220         case BOND_MODE_TLB:
 221                 return "transmit load balancing";
 222         case BOND_MODE_ALB:
 223                 return "adaptive load balancing";
 224         default:
 225                 return "unknown";
 226         }
 227 }
 228
 229 /*---------------------------------- VLAN -----------------------------------*/
 230
 231 /**
 232  * bond_add_vlan - add a new vlan id on bond
 233  * @bond: bond that got the notification
 234  * @vlan_id: the vlan id to add
 235  *
 236  * Returns -ENOMEM if allocation failed.
 237  */
 238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 239 {
 240         struct vlan_entry *vlan;
 241
 242         dprintk("bond: %s, vlan id %d\n",
 243                 (bond ? bond->dev->name: "None"), vlan_id);
 244
 245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 246         if (!vlan) {
 247                 return -ENOMEM;
 248         }
 249
 250         INIT_LIST_HEAD(&vlan->vlan_list);
 251         vlan->vlan_id = vlan_id;
 252         vlan->vlan_ip = 0;
 253
 254         write_lock_bh(&bond->lock);
 255
 256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 257
 258         write_unlock_bh(&bond->lock);
 259
 260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 261
 262         return 0;
 263 }
 264
 265 /**
 266  * bond_del_vlan - delete a vlan id from bond
 267  * @bond: bond that got the notification
 268  * @vlan_id: the vlan id to delete
 269  *
 270  * returns -ENODEV if @vlan_id was not found in @bond.
 271  */
 272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 273 {
 274         struct vlan_entry *vlan;
 275         int res = -ENODEV;
 276
 277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 278
 279         write_lock_bh(&bond->lock);
 280
 281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 282                 if (vlan->vlan_id == vlan_id) {
 283                         list_del(&vlan->vlan_list);
 284
 285                         if ((bond->params.mode == BOND_MODE_TLB) ||
 286                             (bond->params.mode == BOND_MODE_ALB)) {
 287                                 bond_alb_clear_vlan(bond, vlan_id);
 288                         }
 289
 290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 291                                 bond->dev->name);
 292
 293                         kfree(vlan);
 294
 295                         if (list_empty(&bond->vlan_list) &&
 296                             (bond->slave_cnt == 0)) {
 297                                 /* Last VLAN removed and no slaves, so
 298                                  * restore block on adding VLANs. This will
 299                                  * be removed once new slaves that are not
 300                                  * VLAN challenged will be added.
 301                                  */
 302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 303                         }
 304
 305                         res = 0;
 306                         goto out;
 307                 }
 308         }
 309
 310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 311                 bond->dev->name);
 312
 313 out:
 314         write_unlock_bh(&bond->lock);
 315         return res;
 316 }
 317
 318 /**
 319  * bond_has_challenged_slaves
 320  * @bond: the bond we're working on
 321  *
 322  * Searches the slave list. Returns 1 if a vlan challenged slave
 323  * was found, 0 otherwise.
 324  *
 325  * Assumes bond->lock is held.
 326  */
 327 static int bond_has_challenged_slaves(struct bonding *bond)
 328 {
 329         struct slave *slave;
 330         int i;
 331
 332         bond_for_each_slave(bond, slave, i) {
 333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 334                         dprintk("found VLAN challenged slave - %s\n",
 335                                 slave->dev->name);
 336                         return 1;
 337                 }
 338         }
 339
 340         dprintk("no VLAN challenged slaves found\n");
 341         return 0;
 342 }
 343
 344 /**
 345  * bond_next_vlan - safely skip to the next item in the vlans list.
 346  * @bond: the bond we're working on
 347  * @curr: item we're advancing from
 348  *
 349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 350  * or @curr->next otherwise (even if it is @curr itself again).
 351  *
 352  * Caller must hold bond->lock
 353  */
 354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 355 {
 356         struct vlan_entry *next, *last;
 357
 358         if (list_empty(&bond->vlan_list)) {
 359                 return NULL;
 360         }
 361
 362         if (!curr) {
 363                 next = list_entry(bond->vlan_list.next,
 364                                   struct vlan_entry, vlan_list);
 365         } else {
 366                 last = list_entry(bond->vlan_list.prev,
 367                                   struct vlan_entry, vlan_list);
 368                 if (last == curr) {
 369                         next = list_entry(bond->vlan_list.next,
 370                                           struct vlan_entry, vlan_list);
 371                 } else {
 372                         next = list_entry(curr->vlan_list.next,
 373                                           struct vlan_entry, vlan_list);
 374                 }
 375         }
 376
 377         return next;
 378 }
 379
 380 /**
 381  * bond_dev_queue_xmit - Prepare skb for xmit.
 382  *
 383  * @bond: bond device that got this skb for tx.
 384  * @skb: hw accel VLAN tagged skb to transmit
 385  * @slave_dev: slave that is supposed to xmit this skbuff
 386  *
 387  * When the bond gets an skb to transmit that is
 388  * already hardware accelerated VLAN tagged, and it
 389  * needs to relay this skb to a slave that is not
 390  * hw accel capable, the skb needs to be "unaccelerated",
 391  * i.e. strip the hwaccel tag and re-insert it as part
 392  * of the payload.
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 395 {
 396         unsigned short uninitialized_var(vlan_id);
 397
 398         if (!list_empty(&bond->vlan_list) &&
 399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 400             vlan_get_tag(skb, &vlan_id) == 0) {
 401                 skb->dev = slave_dev;
 402                 skb = vlan_put_tag(skb, vlan_id);
 403                 if (!skb) {
 404                         /* vlan_put_tag() frees the skb in case of error,
 405                          * so return success here so the calling functions
 406                          * won't attempt to free is again.
 407                          */
 408                         return 0;
 409                 }
 410         } else {
 411                 skb->dev = slave_dev;
 412         }
 413
 414         skb->priority = 1;
 415         dev_queue_xmit(skb);
 416
 417         return 0;
 418 }
 419
 420 /*
 421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 423  * lock because:
 424  * a. This operation is performed in IOCTL context,
 425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 426  * c. Holding a lock with BH disabled while directly calling a base driver
 427  *    entry point is generally a BAD idea.
 428  *
 429  * The design of synchronization/protection for this operation in the 8021q
 430  * module is good for one or more VLAN devices over a single physical device
 431  * and cannot be extended for a teaming solution like bonding, so there is a
 432  * potential race condition here where a net device from the vlan group might
 433  * be referenced (either by a base driver or the 8021q code) while it is being
 434  * removed from the system. However, it turns out we're not making matters
 435  * worse, and if it works for regular VLAN usage it will work here too.
 436 */
 437
 438 /**
 439  * bond_vlan_rx_register - Propagates registration to slaves
 440  * @bond_dev: bonding net device that got called
 441  * @grp: vlan group being registered
 442  */
 443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 444 {
 445         struct bonding *bond = bond_dev->priv;
 446         struct slave *slave;
 447         int i;
 448
 449         bond->vlgrp = grp;
 450
 451         bond_for_each_slave(bond, slave, i) {
 452                 struct net_device *slave_dev = slave->dev;
 453
 454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 455                     slave_dev->vlan_rx_register) {
 456                         slave_dev->vlan_rx_register(slave_dev, grp);
 457                 }
 458         }
 459 }
 460
 461 /**
 462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being added
 465  */
 466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = bond_dev->priv;
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i) {
 473                 struct net_device *slave_dev = slave->dev;
 474
 475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 476                     slave_dev->vlan_rx_add_vid) {
 477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 478                 }
 479         }
 480
 481         res = bond_add_vlan(bond, vid);
 482         if (res) {
 483                 printk(KERN_ERR DRV_NAME
 484                        ": %s: Error: Failed to add vlan id %d\n",
 485                        bond_dev->name, vid);
 486         }
 487 }
 488
 489 /**
 490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 491  * @bond_dev: bonding net device that got called
 492  * @vid: vlan id being removed
 493  */
 494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 495 {
 496         struct bonding *bond = bond_dev->priv;
 497         struct slave *slave;
 498         struct net_device *vlan_dev;
 499         int i, res;
 500
 501         bond_for_each_slave(bond, slave, i) {
 502                 struct net_device *slave_dev = slave->dev;
 503
 504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 505                     slave_dev->vlan_rx_kill_vid) {
 506                         /* Save and then restore vlan_dev in the grp array,
 507                          * since the slave's driver might clear it.
 508                          */
 509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 512                 }
 513         }
 514
 515         res = bond_del_vlan(bond, vid);
 516         if (res) {
 517                 printk(KERN_ERR DRV_NAME
 518                        ": %s: Error: Failed to remove vlan id %d\n",
 519                        bond_dev->name, vid);
 520         }
 521 }
 522
 523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 524 {
 525         struct vlan_entry *vlan;
 526
 527         write_lock_bh(&bond->lock);
 528
 529         if (list_empty(&bond->vlan_list)) {
 530                 goto out;
 531         }
 532
 533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 534             slave_dev->vlan_rx_register) {
 535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 536         }
 537
 538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 539             !(slave_dev->vlan_rx_add_vid)) {
 540                 goto out;
 541         }
 542
 543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 545         }
 546
 547 out:
 548         write_unlock_bh(&bond->lock);
 549 }
 550
 551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 552 {
 553         struct vlan_entry *vlan;
 554         struct net_device *vlan_dev;
 555
 556         write_lock_bh(&bond->lock);
 557
 558         if (list_empty(&bond->vlan_list)) {
 559                 goto out;
 560         }
 561
 562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 563             !(slave_dev->vlan_rx_kill_vid)) {
 564                 goto unreg;
 565         }
 566
 567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 568                 /* Save and then restore vlan_dev in the grp array,
 569                  * since the slave's driver might clear it.
 570                  */
 571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 574         }
 575
 576 unreg:
 577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 578             slave_dev->vlan_rx_register) {
 579                 slave_dev->vlan_rx_register(slave_dev, NULL);
 580         }
 581
 582 out:
 583         write_unlock_bh(&bond->lock);
 584 }
 585
 586 /*------------------------------- Link status -------------------------------*/
 587
 588 /*
 589  * Set the carrier state for the master according to the state of its
 590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 591  * do special 802.3ad magic.
 592  *
 593  * Returns zero if carrier state does not change, nonzero if it does.
 594  */
 595 static int bond_set_carrier(struct bonding *bond)
 596 {
 597         struct slave *slave;
 598         int i;
 599
 600         if (bond->slave_cnt == 0)
 601                 goto down;
 602
 603         if (bond->params.mode == BOND_MODE_8023AD)
 604                 return bond_3ad_set_carrier(bond);
 605
 606         bond_for_each_slave(bond, slave, i) {
 607                 if (slave->link == BOND_LINK_UP) {
 608                         if (!netif_carrier_ok(bond->dev)) {
 609                                 netif_carrier_on(bond->dev);
 610                                 return 1;
 611                         }
 612                         return 0;
 613                 }
 614         }
 615
 616 down:
 617         if (netif_carrier_ok(bond->dev)) {
 618                 netif_carrier_off(bond->dev);
 619                 return 1;
 620         }
 621         return 0;
 622 }
 623
 624 /*
 625  * Get link speed and duplex from the slave's base driver
 626  * using ethtool. If for some reason the call fails or the
 627  * values are invalid, fake speed and duplex to 100/Full
 628  * and return error.
 629  */
 630 static int bond_update_speed_duplex(struct slave *slave)
 631 {
 632         struct net_device *slave_dev = slave->dev;
 633         struct ethtool_cmd etool;
 634         int res;
 635
 636         /* Fake speed and duplex */
 637         slave->speed = SPEED_100;
 638         slave->duplex = DUPLEX_FULL;
 639
 640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 641                 return -1;
 642
 643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 644         if (res < 0)
 645                 return -1;
 646
 647         switch (etool.speed) {
 648         case SPEED_10:
 649         case SPEED_100:
 650         case SPEED_1000:
 651         case SPEED_10000:
 652                 break;
 653         default:
 654                 return -1;
 655         }
 656
 657         switch (etool.duplex) {
 658         case DUPLEX_FULL:
 659         case DUPLEX_HALF:
 660                 break;
 661         default:
 662                 return -1;
 663         }
 664
 665         slave->speed = etool.speed;
 666         slave->duplex = etool.duplex;
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * if <dev> supports MII link status reporting, check its link status.
 673  *
 674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 675  * depening upon the setting of the use_carrier parameter.
 676  *
 677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 678  * can't tell and just pretend it is), or 0, meaning that the link is
 679  * down.
 680  *
 681  * If reporting is non-zero, instead of faking link up, return -1 if
 682  * both ETHTOOL and MII ioctls fail (meaning the device does not
 683  * support them).  If use_carrier is set, return whatever it says.
 684  * It'd be nice if there was a good way to tell if a driver supports
 685  * netif_carrier, but there really isn't.
 686  */
 687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 688 {
 689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 690         struct ifreq ifr;
 691         struct mii_ioctl_data *mii;
 692
 693         if (bond->params.use_carrier) {
 694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 695         }
 696
 697         ioctl = slave_dev->do_ioctl;
 698         if (ioctl) {
 699                 /* TODO: set pointer to correct ioctl on a per team member */
 700                 /*       bases to make this more efficient. that is, once  */
 701                 /*       we determine the correct ioctl, we will always    */
 702                 /*       call it and not the others for that team          */
 703                 /*       member.                                           */
 704
 705                 /*
 706                  * We cannot assume that SIOCGMIIPHY will also read a
 707                  * register; not all network drivers (e.g., e100)
 708                  * support that.
 709                  */
 710
 711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 713                 mii = if_mii(&ifr);
 714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 715                         mii->reg_num = MII_BMSR;
 716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 717                                 return (mii->val_out & BMSR_LSTATUS);
 718                         }
 719                 }
 720         }
 721
 722         /*
 723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 724          * attempt to get link status from it if the above MII ioctls fail.
 725          */
 726         if (slave_dev->ethtool_ops) {
 727                 if (slave_dev->ethtool_ops->get_link) {
 728                         u32 link;
 729
 730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 731
 732                         return link ? BMSR_LSTATUS : 0;
 733                 }
 734         }
 735
 736         /*
 737          * If reporting, report that either there's no dev->do_ioctl,
 738          * or both SIOCGMIIREG and get_link failed (meaning that we
 739          * cannot report link status).  If not reporting, pretend
 740          * we're ok.
 741          */
 742         return (reporting ? -1 : BMSR_LSTATUS);
 743 }
 744
 745 /*----------------------------- Multicast list ------------------------------*/
 746
 747 /*
 748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 749  */
 750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 751 {
 752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 754 }
 755
 756 /*
 757  * returns dmi entry if found, NULL otherwise
 758  */
 759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 760 {
 761         struct dev_mc_list *idmi;
 762
 763         for (idmi = mc_list; idmi; idmi = idmi->next) {
 764                 if (bond_is_dmi_same(dmi, idmi)) {
 765                         return idmi;
 766                 }
 767         }
 768
 769         return NULL;
 770 }
 771
 772 /*
 773  * Push the promiscuity flag down to appropriate slaves
 774  */
 775 static int bond_set_promiscuity(struct bonding *bond, int inc)
 776 {
 777         int err = 0;
 778         if (USES_PRIMARY(bond->params.mode)) {
 779                 /* write lock already acquired */
 780                 if (bond->curr_active_slave) {
 781                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 782                                                   inc);
 783                 }
 784         } else {
 785                 struct slave *slave;
 786                 int i;
 787                 bond_for_each_slave(bond, slave, i) {
 788                         err = dev_set_promiscuity(slave->dev, inc);
 789                         if (err)
 790                                 return err;
 791                 }
 792         }
 793         return err;
 794 }
 795
 796 /*
 797  * Push the allmulti flag down to all slaves
 798  */
 799 static int bond_set_allmulti(struct bonding *bond, int inc)
 800 {
 801         int err = 0;
 802         if (USES_PRIMARY(bond->params.mode)) {
 803                 /* write lock already acquired */
 804                 if (bond->curr_active_slave) {
 805                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 806                                                inc);
 807                 }
 808         } else {
 809                 struct slave *slave;
 810                 int i;
 811                 bond_for_each_slave(bond, slave, i) {
 812                         err = dev_set_allmulti(slave->dev, inc);
 813                         if (err)
 814                                 return err;
 815                 }
 816         }
 817         return err;
 818 }
 819
 820 /*
 821  * Add a Multicast address to slaves
 822  * according to mode
 823  */
 824 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 825 {
 826         if (USES_PRIMARY(bond->params.mode)) {
 827                 /* write lock already acquired */
 828                 if (bond->curr_active_slave) {
 829                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 830                 }
 831         } else {
 832                 struct slave *slave;
 833                 int i;
 834                 bond_for_each_slave(bond, slave, i) {
 835                         dev_mc_add(slave->dev, addr, alen, 0);
 836                 }
 837         }
 838 }
 839
 840 /*
 841  * Remove a multicast address from slave
 842  * according to mode
 843  */
 844 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 845 {
 846         if (USES_PRIMARY(bond->params.mode)) {
 847                 /* write lock already acquired */
 848                 if (bond->curr_active_slave) {
 849                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 850                 }
 851         } else {
 852                 struct slave *slave;
 853                 int i;
 854                 bond_for_each_slave(bond, slave, i) {
 855                         dev_mc_delete(slave->dev, addr, alen, 0);
 856                 }
 857         }
 858 }
 859
 860
 861 /*
 862  * Retrieve the list of registered multicast addresses for the bonding
 863  * device and retransmit an IGMP JOIN request to the current active
 864  * slave.
 865  */
 866 static void bond_resend_igmp_join_requests(struct bonding *bond)
 867 {
 868         struct in_device *in_dev;
 869         struct ip_mc_list *im;
 870
 871         rcu_read_lock();
 872         in_dev = __in_dev_get_rcu(bond->dev);
 873         if (in_dev) {
 874                 for (im = in_dev->mc_list; im; im = im->next) {
 875                         ip_mc_rejoin_group(im);
 876                 }
 877         }
 878
 879         rcu_read_unlock();
 880 }
 881
 882 /*
 883  * Totally destroys the mc_list in bond
 884  */
 885 static void bond_mc_list_destroy(struct bonding *bond)
 886 {
 887         struct dev_mc_list *dmi;
 888
 889         dmi = bond->mc_list;
 890         while (dmi) {
 891                 bond->mc_list = dmi->next;
 892                 kfree(dmi);
 893                 dmi = bond->mc_list;
 894         }
 895         bond->mc_list = NULL;
 896 }
 897
 898 /*
 899  * Copy all the Multicast addresses from src to the bonding device dst
 900  */
 901 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 902                              gfp_t gfp_flag)
 903 {
 904         struct dev_mc_list *dmi, *new_dmi;
 905
 906         for (dmi = mc_list; dmi; dmi = dmi->next) {
 907                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 908
 909                 if (!new_dmi) {
 910                         /* FIXME: Potential memory leak !!! */
 911                         return -ENOMEM;
 912                 }
 913
 914                 new_dmi->next = bond->mc_list;
 915                 bond->mc_list = new_dmi;
 916                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 917                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 918                 new_dmi->dmi_users = dmi->dmi_users;
 919                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 920         }
 921
 922         return 0;
 923 }
 924
 925 /*
 926  * flush all members of flush->mc_list from device dev->mc_list
 927  */
 928 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 929 {
 930         struct bonding *bond = bond_dev->priv;
 931         struct dev_mc_list *dmi;
 932
 933         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 934                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 935         }
 936
 937         if (bond->params.mode == BOND_MODE_8023AD) {
 938                 /* del lacpdu mc addr from mc list */
 939                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 940
 941                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 942         }
 943 }
 944
 945 /*--------------------------- Active slave change ---------------------------*/
 946
 947 /*
 948  * Update the mc list and multicast-related flags for the new and
 949  * old active slaves (if any) according to the multicast mode, and
 950  * promiscuous flags unconditionally.
 951  */
 952 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 953 {
 954         struct dev_mc_list *dmi;
 955
 956         if (!USES_PRIMARY(bond->params.mode)) {
 957                 /* nothing to do -  mc list is already up-to-date on
 958                  * all slaves
 959                  */
 960                 return;
 961         }
 962
 963         if (old_active) {
 964                 if (bond->dev->flags & IFF_PROMISC) {
 965                         dev_set_promiscuity(old_active->dev, -1);
 966                 }
 967
 968                 if (bond->dev->flags & IFF_ALLMULTI) {
 969                         dev_set_allmulti(old_active->dev, -1);
 970                 }
 971
 972                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 973                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 974                 }
 975         }
 976
 977         if (new_active) {
 978                 /* FIXME: Signal errors upstream. */
 979                 if (bond->dev->flags & IFF_PROMISC) {
 980                         dev_set_promiscuity(new_active->dev, 1);
 981                 }
 982
 983                 if (bond->dev->flags & IFF_ALLMULTI) {
 984                         dev_set_allmulti(new_active->dev, 1);
 985                 }
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 988                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 989                 }
 990                 bond_resend_igmp_join_requests(bond);
 991         }
 992 }
 993
 994 /*
 995  * bond_do_fail_over_mac
 996  *
 997  * Perform special MAC address swapping for fail_over_mac settings
 998  *
 999  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000  */
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002                                   struct slave *new_active,
1003                                   struct slave *old_active)
1004 {
1005         u8 tmp_mac[ETH_ALEN];
1006         struct sockaddr saddr;
1007         int rv;
1008
1009         switch (bond->params.fail_over_mac) {
1010         case BOND_FOM_ACTIVE:
1011                 if (new_active)
1012                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1013                                new_active->dev->addr_len);
1014                 break;
1015         case BOND_FOM_FOLLOW:
1016                 /*
1017                  * if new_active && old_active, swap them
1018                  * if just old_active, do nothing (going to no active slave)
1019                  * if just new_active, set new_active to bond's MAC
1020                  */
1021                 if (!new_active)
1022                         return;
1023
1024                 write_unlock_bh(&bond->curr_slave_lock);
1025                 read_unlock(&bond->lock);
1026
1027                 if (old_active) {
1028                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030                                ETH_ALEN);
1031                         saddr.sa_family = new_active->dev->type;
1032                 } else {
1033                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034                         saddr.sa_family = bond->dev->type;
1035                 }
1036
1037                 rv = dev_set_mac_address(new_active->dev, &saddr);
1038                 if (rv) {
1039                         printk(KERN_ERR DRV_NAME
1040                                ": %s: Error %d setting MAC of slave %s\n",
1041                                bond->dev->name, -rv, new_active->dev->name);
1042                         goto out;
1043                 }
1044
1045                 if (!old_active)
1046                         goto out;
1047
1048                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049                 saddr.sa_family = old_active->dev->type;
1050
1051                 rv = dev_set_mac_address(old_active->dev, &saddr);
1052                 if (rv)
1053                         printk(KERN_ERR DRV_NAME
1054                                ": %s: Error %d setting MAC of slave %s\n",
1055                                bond->dev->name, -rv, new_active->dev->name);
1056 out:
1057                 read_lock(&bond->lock);
1058                 write_lock_bh(&bond->curr_slave_lock);
1059                 break;
1060         default:
1061                 printk(KERN_ERR DRV_NAME
1062                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063                        bond->dev->name, bond->params.fail_over_mac);
1064                 break;
1065         }
1066
1067 }
1068
1069
1070 /**
1071  * find_best_interface - select the best available slave to be the active one
1072  * @bond: our bonding struct
1073  *
1074  * Warning: Caller must hold curr_slave_lock for writing.
1075  */
1076 static struct slave *bond_find_best_slave(struct bonding *bond)
1077 {
1078         struct slave *new_active, *old_active;
1079         struct slave *bestslave = NULL;
1080         int mintime = bond->params.updelay;
1081         int i;
1082
1083         new_active = old_active = bond->curr_active_slave;
1084
1085         if (!new_active) { /* there were no active slaves left */
1086                 if (bond->slave_cnt > 0) {  /* found one slave */
1087                         new_active = bond->first_slave;
1088                 } else {
1089                         return NULL; /* still no slave, return NULL */
1090                 }
1091         }
1092
1093         /* first try the primary link; if arping, a link must tx/rx traffic
1094          * before it can be considered the curr_active_slave - also, we would skip
1095          * slaves between the curr_active_slave and primary_slave that may be up
1096          * and able to arp
1097          */
1098         if ((bond->primary_slave) &&
1099             (!bond->params.arp_interval) &&
1100             (IS_UP(bond->primary_slave->dev))) {
1101                 new_active = bond->primary_slave;
1102         }
1103
1104         /* remember where to stop iterating over the slaves */
1105         old_active = new_active;
1106
1107         bond_for_each_slave_from(bond, new_active, i, old_active) {
1108                 if (IS_UP(new_active->dev)) {
1109                         if (new_active->link == BOND_LINK_UP) {
1110                                 return new_active;
1111                         } else if (new_active->link == BOND_LINK_BACK) {
1112                                 /* link up, but waiting for stabilization */
1113                                 if (new_active->delay < mintime) {
1114                                         mintime = new_active->delay;
1115                                         bestslave = new_active;
1116                                 }
1117                         }
1118                 }
1119         }
1120
1121         return bestslave;
1122 }
1123
1124 /**
1125  * change_active_interface - change the active slave into the specified one
1126  * @bond: our bonding struct
1127  * @new: the new slave to make the active one
1128  *
1129  * Set the new slave to the bond's settings and unset them on the old
1130  * curr_active_slave.
1131  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132  *
1133  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134  * because it is apparently the best available slave we have, even though its
1135  * updelay hasn't timed out yet.
1136  *
1137  * If new_active is not NULL, caller must hold bond->lock for read and
1138  * curr_slave_lock for write_bh.
1139  */
1140 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 {
1142         struct slave *old_active = bond->curr_active_slave;
1143
1144         if (old_active == new_active) {
1145                 return;
1146         }
1147
1148         if (new_active) {
1149                 new_active->jiffies = jiffies;
1150
1151                 if (new_active->link == BOND_LINK_BACK) {
1152                         if (USES_PRIMARY(bond->params.mode)) {
1153                                 printk(KERN_INFO DRV_NAME
1154                                        ": %s: making interface %s the new "
1155                                        "active one %d ms earlier.\n",
1156                                        bond->dev->name, new_active->dev->name,
1157                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158                         }
1159
1160                         new_active->delay = 0;
1161                         new_active->link = BOND_LINK_UP;
1162
1163                         if (bond->params.mode == BOND_MODE_8023AD) {
1164                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165                         }
1166
1167                         if ((bond->params.mode == BOND_MODE_TLB) ||
1168                             (bond->params.mode == BOND_MODE_ALB)) {
1169                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170                         }
1171                 } else {
1172                         if (USES_PRIMARY(bond->params.mode)) {
1173                                 printk(KERN_INFO DRV_NAME
1174                                        ": %s: making interface %s the new "
1175                                        "active one.\n",
1176                                        bond->dev->name, new_active->dev->name);
1177                         }
1178                 }
1179         }
1180
1181         if (USES_PRIMARY(bond->params.mode)) {
1182                 bond_mc_swap(bond, new_active, old_active);
1183         }
1184
1185         if ((bond->params.mode == BOND_MODE_TLB) ||
1186             (bond->params.mode == BOND_MODE_ALB)) {
1187                 bond_alb_handle_active_change(bond, new_active);
1188                 if (old_active)
1189                         bond_set_slave_inactive_flags(old_active);
1190                 if (new_active)
1191                         bond_set_slave_active_flags(new_active);
1192         } else {
1193                 bond->curr_active_slave = new_active;
1194         }
1195
1196         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197                 if (old_active) {
1198                         bond_set_slave_inactive_flags(old_active);
1199                 }
1200
1201                 if (new_active) {
1202                         bond_set_slave_active_flags(new_active);
1203
1204                         if (bond->params.fail_over_mac)
1205                                 bond_do_fail_over_mac(bond, new_active,
1206                                                       old_active);
1207
1208                         bond->send_grat_arp = bond->params.num_grat_arp;
1209                         bond_send_gratuitous_arp(bond);
1210
1211                         write_unlock_bh(&bond->curr_slave_lock);
1212                         read_unlock(&bond->lock);
1213
1214                         netdev_bonding_change(bond->dev);
1215
1216                         read_lock(&bond->lock);
1217                         write_lock_bh(&bond->curr_slave_lock);
1218                 }
1219         }
1220 }
1221
1222 /**
1223  * bond_select_active_slave - select a new active slave, if needed
1224  * @bond: our bonding struct
1225  *
1226  * This functions shoud be called when one of the following occurs:
1227  * - The old curr_active_slave has been released or lost its link.
1228  * - The primary_slave has got its link back.
1229  * - A slave has got its link back and there's no old curr_active_slave.
1230  *
1231  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1232  */
1233 void bond_select_active_slave(struct bonding *bond)
1234 {
1235         struct slave *best_slave;
1236         int rv;
1237
1238         best_slave = bond_find_best_slave(bond);
1239         if (best_slave != bond->curr_active_slave) {
1240                 bond_change_active_slave(bond, best_slave);
1241                 rv = bond_set_carrier(bond);
1242                 if (!rv)
1243                         return;
1244
1245                 if (netif_carrier_ok(bond->dev)) {
1246                         printk(KERN_INFO DRV_NAME
1247                                ": %s: first active interface up!\n",
1248                                bond->dev->name);
1249                 } else {
1250                         printk(KERN_INFO DRV_NAME ": %s: "
1251                                "now running without any active interface !\n",
1252                                bond->dev->name);
1253                 }
1254         }
1255 }
1256
1257 /*--------------------------- slave list handling ---------------------------*/
1258
1259 /*
1260  * This function attaches the slave to the end of list.
1261  *
1262  * bond->lock held for writing by caller.
1263  */
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265 {
1266         if (bond->first_slave == NULL) { /* attaching the first slave */
1267                 new_slave->next = new_slave;
1268                 new_slave->prev = new_slave;
1269                 bond->first_slave = new_slave;
1270         } else {
1271                 new_slave->next = bond->first_slave;
1272                 new_slave->prev = bond->first_slave->prev;
1273                 new_slave->next->prev = new_slave;
1274                 new_slave->prev->next = new_slave;
1275         }
1276
1277         bond->slave_cnt++;
1278 }
1279
1280 /*
1281  * This function detaches the slave from the list.
1282  * WARNING: no check is made to verify if the slave effectively
1283  * belongs to <bond>.
1284  * Nothing is freed on return, structures are just unchained.
1285  * If any slave pointer in bond was pointing to <slave>,
1286  * it should be changed by the calling function.
1287  *
1288  * bond->lock held for writing by caller.
1289  */
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291 {
1292         if (slave->next) {
1293                 slave->next->prev = slave->prev;
1294         }
1295
1296         if (slave->prev) {
1297                 slave->prev->next = slave->next;
1298         }
1299
1300         if (bond->first_slave == slave) { /* slave is the first slave */
1301                 if (bond->slave_cnt > 1) { /* there are more slave */
1302                         bond->first_slave = slave->next;
1303                 } else {
1304                         bond->first_slave = NULL; /* slave was the last one */
1305                 }
1306         }
1307
1308         slave->next = NULL;
1309         slave->prev = NULL;
1310         bond->slave_cnt--;
1311 }
1312
1313 /*---------------------------------- IOCTL ----------------------------------*/
1314
1315 static int bond_sethwaddr(struct net_device *bond_dev,
1316                           struct net_device *slave_dev)
1317 {
1318         dprintk("bond_dev=%p\n", bond_dev);
1319         dprintk("slave_dev=%p\n", slave_dev);
1320         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1321         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1322         return 0;
1323 }
1324
1325 #define BOND_VLAN_FEATURES \
1326         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327          NETIF_F_HW_VLAN_FILTER)
1328
1329 /*
1330  * Compute the common dev->feature set available to all slaves.  Some
1331  * feature bits are managed elsewhere, so preserve those feature bits
1332  * on the master device.
1333  */
1334 static int bond_compute_features(struct bonding *bond)
1335 {
1336         struct slave *slave;
1337         struct net_device *bond_dev = bond->dev;
1338         unsigned long features = bond_dev->features;
1339         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1340                                                 bond_dev->hard_header_len);
1341         int i;
1342
1343         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1344         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1345                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1346
1347         bond_for_each_slave(bond, slave, i) {
1348                 features = netdev_compute_features(features,
1349                                                    slave->dev->features);
1350                 if (slave->dev->hard_header_len > max_hard_header_len)
1351                         max_hard_header_len = slave->dev->hard_header_len;
1352         }
1353
1354         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1355         bond_dev->features = features;
1356         bond_dev->hard_header_len = max_hard_header_len;
1357
1358         return 0;
1359 }
1360
1361
1362 static void bond_setup_by_slave(struct net_device *bond_dev,
1363                                 struct net_device *slave_dev)
1364 {
1365         struct bonding *bond = bond_dev->priv;
1366
1367         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1368         bond_dev->header_ops            = slave_dev->header_ops;
1369
1370         bond_dev->type              = slave_dev->type;
1371         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1372         bond_dev->addr_len          = slave_dev->addr_len;
1373
1374         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1375                 slave_dev->addr_len);
1376         bond->setup_by_slave = 1;
1377 }
1378
1379 /* enslave device <slave> to bond device <master> */
1380 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1381 {
1382         struct bonding *bond = bond_dev->priv;
1383         struct slave *new_slave = NULL;
1384         struct dev_mc_list *dmi;
1385         struct sockaddr addr;
1386         int link_reporting;
1387         int old_features = bond_dev->features;
1388         int res = 0;
1389
1390         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1391                 slave_dev->do_ioctl == NULL) {
1392                 printk(KERN_WARNING DRV_NAME
1393                        ": %s: Warning: no link monitoring support for %s\n",
1394                        bond_dev->name, slave_dev->name);
1395         }
1396
1397         /* bond must be initialized by bond_open() before enslaving */
1398         if (!(bond_dev->flags & IFF_UP)) {
1399                 printk(KERN_WARNING DRV_NAME
1400                         " %s: master_dev is not up in bond_enslave\n",
1401                         bond_dev->name);
1402         }
1403
1404         /* already enslaved */
1405         if (slave_dev->flags & IFF_SLAVE) {
1406                 dprintk("Error, Device was already enslaved\n");
1407                 return -EBUSY;
1408         }
1409
1410         /* vlan challenged mutual exclusion */
1411         /* no need to lock since we're protected by rtnl_lock */
1412         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1413                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1414                 if (!list_empty(&bond->vlan_list)) {
1415                         printk(KERN_ERR DRV_NAME
1416                                ": %s: Error: cannot enslave VLAN "
1417                                "challenged slave %s on VLAN enabled "
1418                                "bond %s\n", bond_dev->name, slave_dev->name,
1419                                bond_dev->name);
1420                         return -EPERM;
1421                 } else {
1422                         printk(KERN_WARNING DRV_NAME
1423                                ": %s: Warning: enslaved VLAN challenged "
1424                                "slave %s. Adding VLANs will be blocked as "
1425                                "long as %s is part of bond %s\n",
1426                                bond_dev->name, slave_dev->name, slave_dev->name,
1427                                bond_dev->name);
1428                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1429                 }
1430         } else {
1431                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1432                 if (bond->slave_cnt == 0) {
1433                         /* First slave, and it is not VLAN challenged,
1434                          * so remove the block of adding VLANs over the bond.
1435                          */
1436                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1437                 }
1438         }
1439
1440         /*
1441          * Old ifenslave binaries are no longer supported.  These can
1442          * be identified with moderate accurary by the state of the slave:
1443          * the current ifenslave will set the interface down prior to
1444          * enslaving it; the old ifenslave will not.
1445          */
1446         if ((slave_dev->flags & IFF_UP)) {
1447                 printk(KERN_ERR DRV_NAME ": %s is up. "
1448                        "This may be due to an out of date ifenslave.\n",
1449                        slave_dev->name);
1450                 res = -EPERM;
1451                 goto err_undo_flags;
1452         }
1453
1454         /* set bonding device ether type by slave - bonding netdevices are
1455          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1456          * there is a need to override some of the type dependent attribs/funcs.
1457          *
1458          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1459          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1460          */
1461         if (bond->slave_cnt == 0) {
1462                 if (slave_dev->type != ARPHRD_ETHER)
1463                         bond_setup_by_slave(bond_dev, slave_dev);
1464         } else if (bond_dev->type != slave_dev->type) {
1465                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1466                         "from other slaves (%d), can not enslave it.\n",
1467                         slave_dev->name,
1468                         slave_dev->type, bond_dev->type);
1469                         res = -EINVAL;
1470                         goto err_undo_flags;
1471         }
1472
1473         if (slave_dev->set_mac_address == NULL) {
1474                 if (bond->slave_cnt == 0) {
1475                         printk(KERN_WARNING DRV_NAME
1476                                ": %s: Warning: The first slave device "
1477                                "specified does not support setting the MAC "
1478                                "address. Setting fail_over_mac to active.",
1479                                bond_dev->name);
1480                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1481                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1482                         printk(KERN_ERR DRV_NAME
1483                                 ": %s: Error: The slave device specified "
1484                                 "does not support setting the MAC address, "
1485                                 "but fail_over_mac is not set to active.\n"
1486                                 , bond_dev->name);
1487                         res = -EOPNOTSUPP;
1488                         goto err_undo_flags;
1489                 }
1490         }
1491
1492         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1493         if (!new_slave) {
1494                 res = -ENOMEM;
1495                 goto err_undo_flags;
1496         }
1497
1498         /* save slave's original flags before calling
1499          * netdev_set_master and dev_open
1500          */
1501         new_slave->original_flags = slave_dev->flags;
1502
1503         /*
1504          * Save slave's original ("permanent") mac address for modes
1505          * that need it, and for restoring it upon release, and then
1506          * set it to the master's address
1507          */
1508         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1509
1510         if (!bond->params.fail_over_mac) {
1511                 /*
1512                  * Set slave to master's mac address.  The application already
1513                  * set the master's mac address to that of the first slave
1514                  */
1515                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1516                 addr.sa_family = slave_dev->type;
1517                 res = dev_set_mac_address(slave_dev, &addr);
1518                 if (res) {
1519                         dprintk("Error %d calling set_mac_address\n", res);
1520                         goto err_free;
1521                 }
1522         }
1523
1524         res = netdev_set_master(slave_dev, bond_dev);
1525         if (res) {
1526                 dprintk("Error %d calling netdev_set_master\n", res);
1527                 goto err_restore_mac;
1528         }
1529         /* open the slave since the application closed it */
1530         res = dev_open(slave_dev);
1531         if (res) {
1532                 dprintk("Openning slave %s failed\n", slave_dev->name);
1533                 goto err_unset_master;
1534         }
1535
1536         new_slave->dev = slave_dev;
1537         slave_dev->priv_flags |= IFF_BONDING;
1538
1539         if ((bond->params.mode == BOND_MODE_TLB) ||
1540             (bond->params.mode == BOND_MODE_ALB)) {
1541                 /* bond_alb_init_slave() must be called before all other stages since
1542                  * it might fail and we do not want to have to undo everything
1543                  */
1544                 res = bond_alb_init_slave(bond, new_slave);
1545                 if (res) {
1546                         goto err_close;
1547                 }
1548         }
1549
1550         /* If the mode USES_PRIMARY, then the new slave gets the
1551          * master's promisc (and mc) settings only if it becomes the
1552          * curr_active_slave, and that is taken care of later when calling
1553          * bond_change_active()
1554          */
1555         if (!USES_PRIMARY(bond->params.mode)) {
1556                 /* set promiscuity level to new slave */
1557                 if (bond_dev->flags & IFF_PROMISC) {
1558                         res = dev_set_promiscuity(slave_dev, 1);
1559                         if (res)
1560                                 goto err_close;
1561                 }
1562
1563                 /* set allmulti level to new slave */
1564                 if (bond_dev->flags & IFF_ALLMULTI) {
1565                         res = dev_set_allmulti(slave_dev, 1);
1566                         if (res)
1567                                 goto err_close;
1568                 }
1569
1570                 netif_addr_lock_bh(bond_dev);
1571                 /* upload master's mc_list to new slave */
1572                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1573                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1574                 }
1575                 netif_addr_unlock_bh(bond_dev);
1576         }
1577
1578         if (bond->params.mode == BOND_MODE_8023AD) {
1579                 /* add lacpdu mc addr to mc list */
1580                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1581
1582                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1583         }
1584
1585         bond_add_vlans_on_slave(bond, slave_dev);
1586
1587         write_lock_bh(&bond->lock);
1588
1589         bond_attach_slave(bond, new_slave);
1590
1591         new_slave->delay = 0;
1592         new_slave->link_failure_count = 0;
1593
1594         bond_compute_features(bond);
1595
1596         write_unlock_bh(&bond->lock);
1597
1598         read_lock(&bond->lock);
1599
1600         new_slave->last_arp_rx = jiffies;
1601
1602         if (bond->params.miimon && !bond->params.use_carrier) {
1603                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1604
1605                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1606                         /*
1607                          * miimon is set but a bonded network driver
1608                          * does not support ETHTOOL/MII and
1609                          * arp_interval is not set.  Note: if
1610                          * use_carrier is enabled, we will never go
1611                          * here (because netif_carrier is always
1612                          * supported); thus, we don't need to change
1613                          * the messages for netif_carrier.
1614                          */
1615                         printk(KERN_WARNING DRV_NAME
1616                                ": %s: Warning: MII and ETHTOOL support not "
1617                                "available for interface %s, and "
1618                                "arp_interval/arp_ip_target module parameters "
1619                                "not specified, thus bonding will not detect "
1620                                "link failures! see bonding.txt for details.\n",
1621                                bond_dev->name, slave_dev->name);
1622                 } else if (link_reporting == -1) {
1623                         /* unable get link status using mii/ethtool */
1624                         printk(KERN_WARNING DRV_NAME
1625                                ": %s: Warning: can't get link status from "
1626                                "interface %s; the network driver associated "
1627                                "with this interface does not support MII or "
1628                                "ETHTOOL link status reporting, thus miimon "
1629                                "has no effect on this interface.\n",
1630                                bond_dev->name, slave_dev->name);
1631                 }
1632         }
1633
1634         /* check for initial state */
1635         if (!bond->params.miimon ||
1636             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1637                 if (bond->params.updelay) {
1638                         dprintk("Initial state of slave_dev is "
1639                                 "BOND_LINK_BACK\n");
1640                         new_slave->link  = BOND_LINK_BACK;
1641                         new_slave->delay = bond->params.updelay;
1642                 } else {
1643                         dprintk("Initial state of slave_dev is "
1644                                 "BOND_LINK_UP\n");
1645                         new_slave->link  = BOND_LINK_UP;
1646                 }
1647                 new_slave->jiffies = jiffies;
1648         } else {
1649                 dprintk("Initial state of slave_dev is "
1650                         "BOND_LINK_DOWN\n");
1651                 new_slave->link  = BOND_LINK_DOWN;
1652         }
1653
1654         if (bond_update_speed_duplex(new_slave) &&
1655             (new_slave->link != BOND_LINK_DOWN)) {
1656                 printk(KERN_WARNING DRV_NAME
1657                        ": %s: Warning: failed to get speed and duplex from %s, "
1658                        "assumed to be 100Mb/sec and Full.\n",
1659                        bond_dev->name, new_slave->dev->name);
1660
1661                 if (bond->params.mode == BOND_MODE_8023AD) {
1662                         printk(KERN_WARNING DRV_NAME
1663                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1664                                "support in base driver for proper aggregator "
1665                                "selection.\n", bond_dev->name);
1666                 }
1667         }
1668
1669         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1670                 /* if there is a primary slave, remember it */
1671                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1672                         bond->primary_slave = new_slave;
1673                 }
1674         }
1675
1676         write_lock_bh(&bond->curr_slave_lock);
1677
1678         switch (bond->params.mode) {
1679         case BOND_MODE_ACTIVEBACKUP:
1680                 bond_set_slave_inactive_flags(new_slave);
1681                 bond_select_active_slave(bond);
1682                 break;
1683         case BOND_MODE_8023AD:
1684                 /* in 802.3ad mode, the internal mechanism
1685                  * will activate the slaves in the selected
1686                  * aggregator
1687                  */
1688                 bond_set_slave_inactive_flags(new_slave);
1689                 /* if this is the first slave */
1690                 if (bond->slave_cnt == 1) {
1691                         SLAVE_AD_INFO(new_slave).id = 1;
1692                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1693                          * can be called only after the mac address of the bond is set
1694                          */
1695                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1696                                             bond->params.lacp_fast);
1697                 } else {
1698                         SLAVE_AD_INFO(new_slave).id =
1699                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1700                 }
1701
1702                 bond_3ad_bind_slave(new_slave);
1703                 break;
1704         case BOND_MODE_TLB:
1705         case BOND_MODE_ALB:
1706                 new_slave->state = BOND_STATE_ACTIVE;
1707                 bond_set_slave_inactive_flags(new_slave);
1708                 break;
1709         default:
1710                 dprintk("This slave is always active in trunk mode\n");
1711
1712                 /* always active in trunk mode */
1713                 new_slave->state = BOND_STATE_ACTIVE;
1714
1715                 /* In trunking mode there is little meaning to curr_active_slave
1716                  * anyway (it holds no special properties of the bond device),
1717                  * so we can change it without calling change_active_interface()
1718                  */
1719                 if (!bond->curr_active_slave) {
1720                         bond->curr_active_slave = new_slave;
1721                 }
1722                 break;
1723         } /* switch(bond_mode) */
1724
1725         write_unlock_bh(&bond->curr_slave_lock);
1726
1727         bond_set_carrier(bond);
1728
1729         read_unlock(&bond->lock);
1730
1731         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1732         if (res)
1733                 goto err_close;
1734
1735         printk(KERN_INFO DRV_NAME
1736                ": %s: enslaving %s as a%s interface with a%s link.\n",
1737                bond_dev->name, slave_dev->name,
1738                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1739                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1740
1741         /* enslave is successful */
1742         return 0;
1743
1744 /* Undo stages on error */
1745 err_close:
1746         dev_close(slave_dev);
1747
1748 err_unset_master:
1749         netdev_set_master(slave_dev, NULL);
1750
1751 err_restore_mac:
1752         if (!bond->params.fail_over_mac) {
1753                 /* XXX TODO - fom follow mode needs to change master's
1754                  * MAC if this slave's MAC is in use by the bond, or at
1755                  * least print a warning.
1756                  */
1757                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1758                 addr.sa_family = slave_dev->type;
1759                 dev_set_mac_address(slave_dev, &addr);
1760         }
1761
1762 err_free:
1763         kfree(new_slave);
1764
1765 err_undo_flags:
1766         bond_dev->features = old_features;
1767
1768         return res;
1769 }
1770
1771 /*
1772  * Try to release the slave device <slave> from the bond device <master>
1773  * It is legal to access curr_active_slave without a lock because all the function
1774  * is write-locked.
1775  *
1776  * The rules for slave state should be:
1777  *   for Active/Backup:
1778  *     Active stays on all backups go down
1779  *   for Bonded connections:
1780  *     The first up interface should be left on and all others downed.
1781  */
1782 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1783 {
1784         struct bonding *bond = bond_dev->priv;
1785         struct slave *slave, *oldcurrent;
1786         struct sockaddr addr;
1787         int mac_addr_differ;
1788         DECLARE_MAC_BUF(mac);
1789
1790         /* slave is not a slave or master is not master of this slave */
1791         if (!(slave_dev->flags & IFF_SLAVE) ||
1792             (slave_dev->master != bond_dev)) {
1793                 printk(KERN_ERR DRV_NAME
1794                        ": %s: Error: cannot release %s.\n",
1795                        bond_dev->name, slave_dev->name);
1796                 return -EINVAL;
1797         }
1798
1799         write_lock_bh(&bond->lock);
1800
1801         slave = bond_get_slave_by_dev(bond, slave_dev);
1802         if (!slave) {
1803                 /* not a slave of this bond */
1804                 printk(KERN_INFO DRV_NAME
1805                        ": %s: %s not enslaved\n",
1806                        bond_dev->name, slave_dev->name);
1807                 write_unlock_bh(&bond->lock);
1808                 return -EINVAL;
1809         }
1810
1811         if (!bond->params.fail_over_mac) {
1812                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1813                                          ETH_ALEN);
1814                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1815                         printk(KERN_WARNING DRV_NAME
1816                                ": %s: Warning: the permanent HWaddr of %s - "
1817                                "%s - is still in use by %s. "
1818                                "Set the HWaddr of %s to a different address "
1819                                "to avoid conflicts.\n",
1820                                bond_dev->name, slave_dev->name,
1821                                print_mac(mac, slave->perm_hwaddr),
1822                                bond_dev->name, slave_dev->name);
1823         }
1824
1825         /* Inform AD package of unbinding of slave. */
1826         if (bond->params.mode == BOND_MODE_8023AD) {
1827                 /* must be called before the slave is
1828                  * detached from the list
1829                  */
1830                 bond_3ad_unbind_slave(slave);
1831         }
1832
1833         printk(KERN_INFO DRV_NAME
1834                ": %s: releasing %s interface %s\n",
1835                bond_dev->name,
1836                (slave->state == BOND_STATE_ACTIVE)
1837                ? "active" : "backup",
1838                slave_dev->name);
1839
1840         oldcurrent = bond->curr_active_slave;
1841
1842         bond->current_arp_slave = NULL;
1843
1844         /* release the slave from its bond */
1845         bond_detach_slave(bond, slave);
1846
1847         bond_compute_features(bond);
1848
1849         if (bond->primary_slave == slave) {
1850                 bond->primary_slave = NULL;
1851         }
1852
1853         if (oldcurrent == slave) {
1854                 bond_change_active_slave(bond, NULL);
1855         }
1856
1857         if ((bond->params.mode == BOND_MODE_TLB) ||
1858             (bond->params.mode == BOND_MODE_ALB)) {
1859                 /* Must be called only after the slave has been
1860                  * detached from the list and the curr_active_slave
1861                  * has been cleared (if our_slave == old_current),
1862                  * but before a new active slave is selected.
1863                  */
1864                 write_unlock_bh(&bond->lock);
1865                 bond_alb_deinit_slave(bond, slave);
1866                 write_lock_bh(&bond->lock);
1867         }
1868
1869         if (oldcurrent == slave) {
1870                 /*
1871                  * Note that we hold RTNL over this sequence, so there
1872                  * is no concern that another slave add/remove event
1873                  * will interfere.
1874                  */
1875                 write_unlock_bh(&bond->lock);
1876                 read_lock(&bond->lock);
1877                 write_lock_bh(&bond->curr_slave_lock);
1878
1879                 bond_select_active_slave(bond);
1880
1881                 write_unlock_bh(&bond->curr_slave_lock);
1882                 read_unlock(&bond->lock);
1883                 write_lock_bh(&bond->lock);
1884         }
1885
1886         if (bond->slave_cnt == 0) {
1887                 bond_set_carrier(bond);
1888
1889                 /* if the last slave was removed, zero the mac address
1890                  * of the master so it will be set by the application
1891                  * to the mac address of the first slave
1892                  */
1893                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1894
1895                 if (list_empty(&bond->vlan_list)) {
1896                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1897                 } else {
1898                         printk(KERN_WARNING DRV_NAME
1899                                ": %s: Warning: clearing HW address of %s while it "
1900                                "still has VLANs.\n",
1901                                bond_dev->name, bond_dev->name);
1902                         printk(KERN_WARNING DRV_NAME
1903                                ": %s: When re-adding slaves, make sure the bond's "
1904                                "HW address matches its VLANs'.\n",
1905                                bond_dev->name);
1906                 }
1907         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1908                    !bond_has_challenged_slaves(bond)) {
1909                 printk(KERN_INFO DRV_NAME
1910                        ": %s: last VLAN challenged slave %s "
1911                        "left bond %s. VLAN blocking is removed\n",
1912                        bond_dev->name, slave_dev->name, bond_dev->name);
1913                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1914         }
1915
1916         write_unlock_bh(&bond->lock);
1917
1918         /* must do this from outside any spinlocks */
1919         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1920
1921         bond_del_vlans_from_slave(bond, slave_dev);
1922
1923         /* If the mode USES_PRIMARY, then we should only remove its
1924          * promisc and mc settings if it was the curr_active_slave, but that was
1925          * already taken care of above when we detached the slave
1926          */
1927         if (!USES_PRIMARY(bond->params.mode)) {
1928                 /* unset promiscuity level from slave */
1929                 if (bond_dev->flags & IFF_PROMISC) {
1930                         dev_set_promiscuity(slave_dev, -1);
1931                 }
1932
1933                 /* unset allmulti level from slave */
1934                 if (bond_dev->flags & IFF_ALLMULTI) {
1935                         dev_set_allmulti(slave_dev, -1);
1936                 }
1937
1938                 /* flush master's mc_list from slave */
1939                 netif_addr_lock_bh(bond_dev);
1940                 bond_mc_list_flush(bond_dev, slave_dev);
1941                 netif_addr_unlock_bh(bond_dev);
1942         }
1943
1944         netdev_set_master(slave_dev, NULL);
1945
1946         /* close slave before restoring its mac address */
1947         dev_close(slave_dev);
1948
1949         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1950                 /* restore original ("permanent") mac address */
1951                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1952                 addr.sa_family = slave_dev->type;
1953                 dev_set_mac_address(slave_dev, &addr);
1954         }
1955
1956         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1957                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1958                                    IFF_SLAVE_NEEDARP);
1959
1960         kfree(slave);
1961
1962         return 0;  /* deletion OK */
1963 }
1964
1965 /*
1966 * Destroy a bonding device.
1967 * Must be under rtnl_lock when this function is called.
1968 */
1969 void bond_destroy(struct bonding *bond)
1970 {
1971         bond_deinit(bond->dev);
1972         bond_destroy_sysfs_entry(bond);
1973         unregister_netdevice(bond->dev);
1974 }
1975
1976 /*
1977 * First release a slave and than destroy the bond if no more slaves iare left.
1978 * Must be under rtnl_lock when this function is called.
1979 */
1980 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1981 {
1982         struct bonding *bond = bond_dev->priv;
1983         int ret;
1984
1985         ret = bond_release(bond_dev, slave_dev);
1986         if ((ret == 0) && (bond->slave_cnt == 0)) {
1987                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1988                        bond_dev->name, bond_dev->name);
1989                 bond_destroy(bond);
1990         }
1991         return ret;
1992 }
1993
1994 /*
1995  * This function releases all slaves.
1996  */
1997 static int bond_release_all(struct net_device *bond_dev)
1998 {
1999         struct bonding *bond = bond_dev->priv;
2000         struct slave *slave;
2001         struct net_device *slave_dev;
2002         struct sockaddr addr;
2003
2004         write_lock_bh(&bond->lock);
2005
2006         netif_carrier_off(bond_dev);
2007
2008         if (bond->slave_cnt == 0) {
2009                 goto out;
2010         }
2011
2012         bond->current_arp_slave = NULL;
2013         bond->primary_slave = NULL;
2014         bond_change_active_slave(bond, NULL);
2015
2016         while ((slave = bond->first_slave) != NULL) {
2017                 /* Inform AD package of unbinding of slave
2018                  * before slave is detached from the list.
2019                  */
2020                 if (bond->params.mode == BOND_MODE_8023AD) {
2021                         bond_3ad_unbind_slave(slave);
2022                 }
2023
2024                 slave_dev = slave->dev;
2025                 bond_detach_slave(bond, slave);
2026
2027                 /* now that the slave is detached, unlock and perform
2028                  * all the undo steps that should not be called from
2029                  * within a lock.
2030                  */
2031                 write_unlock_bh(&bond->lock);
2032
2033                 if ((bond->params.mode == BOND_MODE_TLB) ||
2034                     (bond->params.mode == BOND_MODE_ALB)) {
2035                         /* must be called only after the slave
2036                          * has been detached from the list
2037                          */
2038                         bond_alb_deinit_slave(bond, slave);
2039                 }
2040
2041                 bond_compute_features(bond);
2042
2043                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2044                 bond_del_vlans_from_slave(bond, slave_dev);
2045
2046                 /* If the mode USES_PRIMARY, then we should only remove its
2047                  * promisc and mc settings if it was the curr_active_slave, but that was
2048                  * already taken care of above when we detached the slave
2049                  */
2050                 if (!USES_PRIMARY(bond->params.mode)) {
2051                         /* unset promiscuity level from slave */
2052                         if (bond_dev->flags & IFF_PROMISC) {
2053                                 dev_set_promiscuity(slave_dev, -1);
2054                         }
2055
2056                         /* unset allmulti level from slave */
2057                         if (bond_dev->flags & IFF_ALLMULTI) {
2058                                 dev_set_allmulti(slave_dev, -1);
2059                         }
2060
2061                         /* flush master's mc_list from slave */
2062                         netif_addr_lock_bh(bond_dev);
2063                         bond_mc_list_flush(bond_dev, slave_dev);
2064                         netif_addr_unlock_bh(bond_dev);
2065                 }
2066
2067                 netdev_set_master(slave_dev, NULL);
2068
2069                 /* close slave before restoring its mac address */
2070                 dev_close(slave_dev);
2071
2072                 if (!bond->params.fail_over_mac) {
2073                         /* restore original ("permanent") mac address*/
2074                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2075                         addr.sa_family = slave_dev->type;
2076                         dev_set_mac_address(slave_dev, &addr);
2077                 }
2078
2079                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2080                                            IFF_SLAVE_INACTIVE);
2081
2082                 kfree(slave);
2083
2084                 /* re-acquire the lock before getting the next slave */
2085                 write_lock_bh(&bond->lock);
2086         }
2087
2088         /* zero the mac address of the master so it will be
2089          * set by the application to the mac address of the
2090          * first slave
2091          */
2092         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2093
2094         if (list_empty(&bond->vlan_list)) {
2095                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2096         } else {
2097                 printk(KERN_WARNING DRV_NAME
2098                        ": %s: Warning: clearing HW address of %s while it "
2099                        "still has VLANs.\n",
2100                        bond_dev->name, bond_dev->name);
2101                 printk(KERN_WARNING DRV_NAME
2102                        ": %s: When re-adding slaves, make sure the bond's "
2103                        "HW address matches its VLANs'.\n",
2104                        bond_dev->name);
2105         }
2106
2107         printk(KERN_INFO DRV_NAME
2108                ": %s: released all slaves\n",
2109                bond_dev->name);
2110
2111 out:
2112         write_unlock_bh(&bond->lock);
2113
2114         return 0;
2115 }
2116
2117 /*
2118  * This function changes the active slave to slave <slave_dev>.
2119  * It returns -EINVAL in the following cases.
2120  *  - <slave_dev> is not found in the list.
2121  *  - There is not active slave now.
2122  *  - <slave_dev> is already active.
2123  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2124  *  - <slave_dev> is not running.
2125  * In these cases, this fuction does nothing.
2126  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2127  */
2128 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2129 {
2130         struct bonding *bond = bond_dev->priv;
2131         struct slave *old_active = NULL;
2132         struct slave *new_active = NULL;
2133         int res = 0;
2134
2135         if (!USES_PRIMARY(bond->params.mode)) {
2136                 return -EINVAL;
2137         }
2138
2139         /* Verify that master_dev is indeed the master of slave_dev */
2140         if (!(slave_dev->flags & IFF_SLAVE) ||
2141             (slave_dev->master != bond_dev)) {
2142                 return -EINVAL;
2143         }
2144
2145         read_lock(&bond->lock);
2146
2147         read_lock(&bond->curr_slave_lock);
2148         old_active = bond->curr_active_slave;
2149         read_unlock(&bond->curr_slave_lock);
2150
2151         new_active = bond_get_slave_by_dev(bond, slave_dev);
2152
2153         /*
2154          * Changing to the current active: do nothing; return success.
2155          */
2156         if (new_active && (new_active == old_active)) {
2157                 read_unlock(&bond->lock);
2158                 return 0;
2159         }
2160
2161         if ((new_active) &&
2162             (old_active) &&
2163             (new_active->link == BOND_LINK_UP) &&
2164             IS_UP(new_active->dev)) {
2165                 write_lock_bh(&bond->curr_slave_lock);
2166                 bond_change_active_slave(bond, new_active);
2167                 write_unlock_bh(&bond->curr_slave_lock);
2168         } else {
2169                 res = -EINVAL;
2170         }
2171
2172         read_unlock(&bond->lock);
2173
2174         return res;
2175 }
2176
2177 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2178 {
2179         struct bonding *bond = bond_dev->priv;
2180
2181         info->bond_mode = bond->params.mode;
2182         info->miimon = bond->params.miimon;
2183
2184         read_lock(&bond->lock);
2185         info->num_slaves = bond->slave_cnt;
2186         read_unlock(&bond->lock);
2187
2188         return 0;
2189 }
2190
2191 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2192 {
2193         struct bonding *bond = bond_dev->priv;
2194         struct slave *slave;
2195         int i, found = 0;
2196
2197         if (info->slave_id < 0) {
2198                 return -ENODEV;
2199         }
2200
2201         read_lock(&bond->lock);
2202
2203         bond_for_each_slave(bond, slave, i) {
2204                 if (i == (int)info->slave_id) {
2205                         found = 1;
2206                         break;
2207                 }
2208         }
2209
2210         read_unlock(&bond->lock);
2211
2212         if (found) {
2213                 strcpy(info->slave_name, slave->dev->name);
2214                 info->link = slave->link;
2215                 info->state = slave->state;
2216                 info->link_failure_count = slave->link_failure_count;
2217         } else {
2218                 return -ENODEV;
2219         }
2220
2221         return 0;
2222 }
2223
2224 /*-------------------------------- Monitoring -------------------------------*/
2225
2226
2227 static int bond_miimon_inspect(struct bonding *bond)
2228 {
2229         struct slave *slave;
2230         int i, link_state, commit = 0;
2231
2232         bond_for_each_slave(bond, slave, i) {
2233                 slave->new_link = BOND_LINK_NOCHANGE;
2234
2235                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2236
2237                 switch (slave->link) {
2238                 case BOND_LINK_UP:
2239                         if (link_state)
2240                                 continue;
2241
2242                         slave->link = BOND_LINK_FAIL;
2243                         slave->delay = bond->params.downdelay;
2244                         if (slave->delay) {
2245                                 printk(KERN_INFO DRV_NAME
2246                                        ": %s: link status down for %s"
2247                                        "interface %s, disabling it in %d ms.\n",
2248                                        bond->dev->name,
2249                                        (bond->params.mode ==
2250                                         BOND_MODE_ACTIVEBACKUP) ?
2251                                        ((slave->state == BOND_STATE_ACTIVE) ?
2252                                         "active " : "backup ") : "",
2253                                        slave->dev->name,
2254                                        bond->params.downdelay * bond->params.miimon);
2255                         }
2256                         /*FALLTHRU*/
2257                 case BOND_LINK_FAIL:
2258                         if (link_state) {
2259                                 /*
2260                                  * recovered before downdelay expired
2261                                  */
2262                                 slave->link = BOND_LINK_UP;
2263                                 slave->jiffies = jiffies;
2264                                 printk(KERN_INFO DRV_NAME
2265                                        ": %s: link status up again after %d "
2266                                        "ms for interface %s.\n",
2267                                        bond->dev->name,
2268                                        (bond->params.downdelay - slave->delay) *
2269                                        bond->params.miimon,
2270                                        slave->dev->name);
2271                                 continue;
2272                         }
2273
2274                         if (slave->delay <= 0) {
2275                                 slave->new_link = BOND_LINK_DOWN;
2276                                 commit++;
2277                                 continue;
2278                         }
2279
2280                         slave->delay--;
2281                         break;
2282
2283                 case BOND_LINK_DOWN:
2284                         if (!link_state)
2285                                 continue;
2286
2287                         slave->link = BOND_LINK_BACK;
2288                         slave->delay = bond->params.updelay;
2289
2290                         if (slave->delay) {
2291                                 printk(KERN_INFO DRV_NAME
2292                                        ": %s: link status up for "
2293                                        "interface %s, enabling it in %d ms.\n",
2294                                        bond->dev->name, slave->dev->name,
2295                                        bond->params.updelay *
2296                                        bond->params.miimon);
2297                         }
2298                         /*FALLTHRU*/
2299                 case BOND_LINK_BACK:
2300                         if (!link_state) {
2301                                 slave->link = BOND_LINK_DOWN;
2302                                 printk(KERN_INFO DRV_NAME
2303                                        ": %s: link status down again after %d "
2304                                        "ms for interface %s.\n",
2305                                        bond->dev->name,
2306                                        (bond->params.updelay - slave->delay) *
2307                                        bond->params.miimon,
2308                                        slave->dev->name);
2309
2310                                 continue;
2311                         }
2312
2313                         if (slave->delay <= 0) {
2314                                 slave->new_link = BOND_LINK_UP;
2315                                 commit++;
2316                                 continue;
2317                         }
2318
2319                         slave->delay--;
2320                         break;
2321                 }
2322         }
2323
2324         return commit;
2325 }
2326
2327 static void bond_miimon_commit(struct bonding *bond)
2328 {
2329         struct slave *slave;
2330         int i;
2331
2332         bond_for_each_slave(bond, slave, i) {
2333                 switch (slave->new_link) {
2334                 case BOND_LINK_NOCHANGE:
2335                         continue;
2336
2337                 case BOND_LINK_UP:
2338                         slave->link = BOND_LINK_UP;
2339                         slave->jiffies = jiffies;
2340
2341                         if (bond->params.mode == BOND_MODE_8023AD) {
2342                                 /* prevent it from being the active one */
2343                                 slave->state = BOND_STATE_BACKUP;
2344                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2345                                 /* make it immediately active */
2346                                 slave->state = BOND_STATE_ACTIVE;
2347                         } else if (slave != bond->primary_slave) {
2348                                 /* prevent it from being the active one */
2349                                 slave->state = BOND_STATE_BACKUP;
2350                         }
2351
2352                         printk(KERN_INFO DRV_NAME
2353                                ": %s: link status definitely "
2354                                "up for interface %s.\n",
2355                                bond->dev->name, slave->dev->name);
2356
2357                         /* notify ad that the link status has changed */
2358                         if (bond->params.mode == BOND_MODE_8023AD)
2359                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2360
2361                         if ((bond->params.mode == BOND_MODE_TLB) ||
2362                             (bond->params.mode == BOND_MODE_ALB))
2363                                 bond_alb_handle_link_change(bond, slave,
2364                                                             BOND_LINK_UP);
2365
2366                         if (!bond->curr_active_slave ||
2367                             (slave == bond->primary_slave))
2368                                 goto do_failover;
2369
2370                         continue;
2371
2372                 case BOND_LINK_DOWN:
2373                         slave->link = BOND_LINK_DOWN;
2374
2375                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2376                             bond->params.mode == BOND_MODE_8023AD)
2377                                 bond_set_slave_inactive_flags(slave);
2378
2379                         printk(KERN_INFO DRV_NAME
2380                                ": %s: link status definitely down for "
2381                                "interface %s, disabling it\n",
2382                                bond->dev->name, slave->dev->name);
2383
2384                         if (bond->params.mode == BOND_MODE_8023AD)
2385                                 bond_3ad_handle_link_change(slave,
2386                                                             BOND_LINK_DOWN);
2387
2388                         if (bond->params.mode == BOND_MODE_TLB ||
2389                             bond->params.mode == BOND_MODE_ALB)
2390                                 bond_alb_handle_link_change(bond, slave,
2391                                                             BOND_LINK_DOWN);
2392
2393                         if (slave == bond->curr_active_slave)
2394                                 goto do_failover;
2395
2396                         continue;
2397
2398                 default:
2399                         printk(KERN_ERR DRV_NAME
2400                                ": %s: invalid new link %d on slave %s\n",
2401                                bond->dev->name, slave->new_link,
2402                                slave->dev->name);
2403                         slave->new_link = BOND_LINK_NOCHANGE;
2404
2405                         continue;
2406                 }
2407
2408 do_failover:
2409                 ASSERT_RTNL();
2410                 write_lock_bh(&bond->curr_slave_lock);
2411                 bond_select_active_slave(bond);
2412                 write_unlock_bh(&bond->curr_slave_lock);
2413         }
2414
2415         bond_set_carrier(bond);
2416 }
2417
2418 /*
2419  * bond_mii_monitor
2420  *
2421  * Really a wrapper that splits the mii monitor into two phases: an
2422  * inspection, then (if inspection indicates something needs to be done)
2423  * an acquisition of appropriate locks followed by a commit phase to
2424  * implement whatever link state changes are indicated.
2425  */
2426 void bond_mii_monitor(struct work_struct *work)
2427 {
2428         struct bonding *bond = container_of(work, struct bonding,
2429                                             mii_work.work);
2430
2431         read_lock(&bond->lock);
2432         if (bond->kill_timers)
2433                 goto out;
2434
2435         if (bond->slave_cnt == 0)
2436                 goto re_arm;
2437
2438         if (bond->send_grat_arp) {
2439                 read_lock(&bond->curr_slave_lock);
2440                 bond_send_gratuitous_arp(bond);
2441                 read_unlock(&bond->curr_slave_lock);
2442         }
2443
2444         if (bond_miimon_inspect(bond)) {
2445                 read_unlock(&bond->lock);
2446                 rtnl_lock();
2447                 read_lock(&bond->lock);
2448
2449                 bond_miimon_commit(bond);
2450
2451                 read_unlock(&bond->lock);
2452                 rtnl_unlock();  /* might sleep, hold no other locks */
2453                 read_lock(&bond->lock);
2454         }
2455
2456 re_arm:
2457         if (bond->params.miimon)
2458                 queue_delayed_work(bond->wq, &bond->mii_work,
2459                                    msecs_to_jiffies(bond->params.miimon));
2460 out:
2461         read_unlock(&bond->lock);
2462 }
2463
2464 static __be32 bond_glean_dev_ip(struct net_device *dev)
2465 {
2466         struct in_device *idev;
2467         struct in_ifaddr *ifa;
2468         __be32 addr = 0;
2469
2470         if (!dev)
2471                 return 0;
2472
2473         rcu_read_lock();
2474         idev = __in_dev_get_rcu(dev);
2475         if (!idev)
2476                 goto out;
2477
2478         ifa = idev->ifa_list;
2479         if (!ifa)
2480                 goto out;
2481
2482         addr = ifa->ifa_local;
2483 out:
2484         rcu_read_unlock();
2485         return addr;
2486 }
2487
2488 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2489 {
2490         struct vlan_entry *vlan;
2491
2492         if (ip == bond->master_ip)
2493                 return 1;
2494
2495         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2496                 if (ip == vlan->vlan_ip)
2497                         return 1;
2498         }
2499
2500         return 0;
2501 }
2502
2503 /*
2504  * We go to the (large) trouble of VLAN tagging ARP frames because
2505  * switches in VLAN mode (especially if ports are configured as
2506  * "native" to a VLAN) might not pass non-tagged frames.
2507  */
2508 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2509 {
2510         struct sk_buff *skb;
2511
2512         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2513                slave_dev->name, dest_ip, src_ip, vlan_id);
2514
2515         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2516                          NULL, slave_dev->dev_addr, NULL);
2517
2518         if (!skb) {
2519                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2520                 return;
2521         }
2522         if (vlan_id) {
2523                 skb = vlan_put_tag(skb, vlan_id);
2524                 if (!skb) {
2525                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2526                         return;
2527                 }
2528         }
2529         arp_xmit(skb);
2530 }
2531
2532
2533 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2534 {
2535         int i, vlan_id, rv;
2536         __be32 *targets = bond->params.arp_targets;
2537         struct vlan_entry *vlan;
2538         struct net_device *vlan_dev;
2539         struct flowi fl;
2540         struct rtable *rt;
2541
2542         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2543                 if (!targets[i])
2544                         continue;
2545                 dprintk("basa: target %x\n", targets[i]);
2546                 if (list_empty(&bond->vlan_list)) {
2547                         dprintk("basa: empty vlan: arp_send\n");
2548                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2549                                       bond->master_ip, 0);
2550                         continue;
2551                 }
2552
2553                 /*
2554                  * If VLANs are configured, we do a route lookup to
2555                  * determine which VLAN interface would be used, so we
2556                  * can tag the ARP with the proper VLAN tag.
2557                  */
2558                 memset(&fl, 0, sizeof(fl));
2559                 fl.fl4_dst = targets[i];
2560                 fl.fl4_tos = RTO_ONLINK;
2561
2562                 rv = ip_route_output_key(&init_net, &rt, &fl);
2563                 if (rv) {
2564                         if (net_ratelimit()) {
2565                                 printk(KERN_WARNING DRV_NAME
2566                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2567                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2568                         }
2569                         continue;
2570                 }
2571
2572                 /*
2573                  * This target is not on a VLAN
2574                  */
2575                 if (rt->u.dst.dev == bond->dev) {
2576                         ip_rt_put(rt);
2577                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2578                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2579                                       bond->master_ip, 0);
2580                         continue;
2581                 }
2582
2583                 vlan_id = 0;
2584                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2585                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2586                         if (vlan_dev == rt->u.dst.dev) {
2587                                 vlan_id = vlan->vlan_id;
2588                                 dprintk("basa: vlan match on %s %d\n",
2589                                        vlan_dev->name, vlan_id);
2590                                 break;
2591                         }
2592                 }
2593
2594                 if (vlan_id) {
2595                         ip_rt_put(rt);
2596                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2597                                       vlan->vlan_ip, vlan_id);
2598                         continue;
2599                 }
2600
2601                 if (net_ratelimit()) {
2602                         printk(KERN_WARNING DRV_NAME
2603                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2604                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2605                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2606                 }
2607                 ip_rt_put(rt);
2608         }
2609 }
2610
2611 /*
2612  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2613  * for each VLAN above us.
2614  *
2615  * Caller must hold curr_slave_lock for read or better
2616  */
2617 static void bond_send_gratuitous_arp(struct bonding *bond)
2618 {
2619         struct slave *slave = bond->curr_active_slave;
2620         struct vlan_entry *vlan;
2621         struct net_device *vlan_dev;
2622
2623         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2624                                 slave ? slave->dev->name : "NULL");
2625
2626         if (!slave || !bond->send_grat_arp ||
2627             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2628                 return;
2629
2630         bond->send_grat_arp--;
2631
2632         if (bond->master_ip) {
2633                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2634                                 bond->master_ip, 0);
2635         }
2636
2637         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2638                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2639                 if (vlan->vlan_ip) {
2640                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2641                                       vlan->vlan_ip, vlan->vlan_id);
2642                 }
2643         }
2644 }
2645
2646 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2647 {
2648         int i;
2649         __be32 *targets = bond->params.arp_targets;
2650
2651         targets = bond->params.arp_targets;
2652         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2653                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2654                         "%u.%u.%u.%u bhti(tip) %d\n",
2655                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2656                        bond_has_this_ip(bond, tip));
2657                 if (sip == targets[i]) {
2658                         if (bond_has_this_ip(bond, tip))
2659                                 slave->last_arp_rx = jiffies;
2660                         return;
2661                 }
2662         }
2663 }
2664
2665 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2666 {
2667         struct arphdr *arp;
2668         struct slave *slave;
2669         struct bonding *bond;
2670         unsigned char *arp_ptr;
2671         __be32 sip, tip;
2672
2673         if (dev_net(dev) != &init_net)
2674                 goto out;
2675
2676         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2677                 goto out;
2678
2679         bond = dev->priv;
2680         read_lock(&bond->lock);
2681
2682         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2683                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2684                 orig_dev ? orig_dev->name : "NULL");
2685
2686         slave = bond_get_slave_by_dev(bond, orig_dev);
2687         if (!slave || !slave_do_arp_validate(bond, slave))
2688                 goto out_unlock;
2689
2690         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2691                 goto out_unlock;
2692
2693         arp = arp_hdr(skb);
2694         if (arp->ar_hln != dev->addr_len ||
2695             skb->pkt_type == PACKET_OTHERHOST ||
2696             skb->pkt_type == PACKET_LOOPBACK ||
2697             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2698             arp->ar_pro != htons(ETH_P_IP) ||
2699             arp->ar_pln != 4)
2700                 goto out_unlock;
2701
2702         arp_ptr = (unsigned char *)(arp + 1);
2703         arp_ptr += dev->addr_len;
2704         memcpy(&sip, arp_ptr, 4);
2705         arp_ptr += 4 + dev->addr_len;
2706         memcpy(&tip, arp_ptr, 4);
2707
2708         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2709                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2710                 slave->state, bond->params.arp_validate,
2711                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2712
2713         /*
2714          * Backup slaves won't see the ARP reply, but do come through
2715          * here for each ARP probe (so we swap the sip/tip to validate
2716          * the probe).  In a "redundant switch, common router" type of
2717          * configuration, the ARP probe will (hopefully) travel from
2718          * the active, through one switch, the router, then the other
2719          * switch before reaching the backup.
2720          */
2721         if (slave->state == BOND_STATE_ACTIVE)
2722                 bond_validate_arp(bond, slave, sip, tip);
2723         else
2724                 bond_validate_arp(bond, slave, tip, sip);
2725
2726 out_unlock:
2727         read_unlock(&bond->lock);
2728 out:
2729         dev_kfree_skb(skb);
2730         return NET_RX_SUCCESS;
2731 }
2732
2733 /*
2734  * this function is called regularly to monitor each slave's link
2735  * ensuring that traffic is being sent and received when arp monitoring
2736  * is used in load-balancing mode. if the adapter has been dormant, then an
2737  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2738  * arp monitoring in active backup mode.
2739  */
2740 void bond_loadbalance_arp_mon(struct work_struct *work)
2741 {
2742         struct bonding *bond = container_of(work, struct bonding,
2743                                             arp_work.work);
2744         struct slave *slave, *oldcurrent;
2745         int do_failover = 0;
2746         int delta_in_ticks;
2747         int i;
2748
2749         read_lock(&bond->lock);
2750
2751         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2752
2753         if (bond->kill_timers) {
2754                 goto out;
2755         }
2756
2757         if (bond->slave_cnt == 0) {
2758                 goto re_arm;
2759         }
2760
2761         read_lock(&bond->curr_slave_lock);
2762         oldcurrent = bond->curr_active_slave;
2763         read_unlock(&bond->curr_slave_lock);
2764
2765         /* see if any of the previous devices are up now (i.e. they have
2766          * xmt and rcv traffic). the curr_active_slave does not come into
2767          * the picture unless it is null. also, slave->jiffies is not needed
2768          * here because we send an arp on each slave and give a slave as
2769          * long as it needs to get the tx/rx within the delta.
2770          * TODO: what about up/down delay in arp mode? it wasn't here before
2771          *       so it can wait
2772          */
2773         bond_for_each_slave(bond, slave, i) {
2774                 if (slave->link != BOND_LINK_UP) {
2775                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2776                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2777
2778                                 slave->link  = BOND_LINK_UP;
2779                                 slave->state = BOND_STATE_ACTIVE;
2780
2781                                 /* primary_slave has no meaning in round-robin
2782                                  * mode. the window of a slave being up and
2783                                  * curr_active_slave being null after enslaving
2784                                  * is closed.
2785                                  */
2786                                 if (!oldcurrent) {
2787                                         printk(KERN_INFO DRV_NAME
2788                                                ": %s: link status definitely "
2789                                                "up for interface %s, ",
2790                                                bond->dev->name,
2791                                                slave->dev->name);
2792                                         do_failover = 1;
2793                                 } else {
2794                                         printk(KERN_INFO DRV_NAME
2795                                                ": %s: interface %s is now up\n",
2796                                                bond->dev->name,
2797                                                slave->dev->name);
2798                                 }
2799                         }
2800                 } else {
2801                         /* slave->link == BOND_LINK_UP */
2802
2803                         /* not all switches will respond to an arp request
2804                          * when the source ip is 0, so don't take the link down
2805                          * if we don't know our ip yet
2806                          */
2807                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2808                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2809
2810                                 slave->link  = BOND_LINK_DOWN;
2811                                 slave->state = BOND_STATE_BACKUP;
2812
2813                                 if (slave->link_failure_count < UINT_MAX) {
2814                                         slave->link_failure_count++;
2815                                 }
2816
2817                                 printk(KERN_INFO DRV_NAME
2818                                        ": %s: interface %s is now down.\n",
2819                                        bond->dev->name,
2820                                        slave->dev->name);
2821
2822                                 if (slave == oldcurrent) {
2823                                         do_failover = 1;
2824                                 }
2825                         }
2826                 }
2827
2828                 /* note: if switch is in round-robin mode, all links
2829                  * must tx arp to ensure all links rx an arp - otherwise
2830                  * links may oscillate or not come up at all; if switch is
2831                  * in something like xor mode, there is nothing we can
2832                  * do - all replies will be rx'ed on same link causing slaves
2833                  * to be unstable during low/no traffic periods
2834                  */
2835                 if (IS_UP(slave->dev)) {
2836                         bond_arp_send_all(bond, slave);
2837                 }
2838         }
2839
2840         if (do_failover) {
2841                 write_lock_bh(&bond->curr_slave_lock);
2842
2843                 bond_select_active_slave(bond);
2844
2845                 write_unlock_bh(&bond->curr_slave_lock);
2846         }
2847
2848 re_arm:
2849         if (bond->params.arp_interval)
2850                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2851 out:
2852         read_unlock(&bond->lock);
2853 }
2854
2855 /*
2856  * Called to inspect slaves for active-backup mode ARP monitor link state
2857  * changes.  Sets new_link in slaves to specify what action should take
2858  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2859  * to link states must be committed.
2860  *
2861  * Called with bond->lock held for read.
2862  */
2863 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2864 {
2865         struct slave *slave;
2866         int i, commit = 0;
2867
2868         bond_for_each_slave(bond, slave, i) {
2869                 slave->new_link = BOND_LINK_NOCHANGE;
2870
2871                 if (slave->link != BOND_LINK_UP) {
2872                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2873                                            delta_in_ticks)) {
2874                                 slave->new_link = BOND_LINK_UP;
2875                                 commit++;
2876                         }
2877
2878                         continue;
2879                 }
2880
2881                 /*
2882                  * Give slaves 2*delta after being enslaved or made
2883                  * active.  This avoids bouncing, as the last receive
2884                  * times need a full ARP monitor cycle to be updated.
2885                  */
2886                 if (!time_after_eq(jiffies, slave->jiffies +
2887                                    2 * delta_in_ticks))
2888                         continue;
2889
2890                 /*
2891                  * Backup slave is down if:
2892                  * - No current_arp_slave AND
2893                  * - more than 3*delta since last receive AND
2894                  * - the bond has an IP address
2895                  *
2896                  * Note: a non-null current_arp_slave indicates
2897                  * the curr_active_slave went down and we are
2898                  * searching for a new one; under this condition
2899                  * we only take the curr_active_slave down - this
2900                  * gives each slave a chance to tx/rx traffic
2901                  * before being taken out
2902                  */
2903                 if (slave->state == BOND_STATE_BACKUP &&
2904                     !bond->current_arp_slave &&
2905                     time_after(jiffies, slave_last_rx(bond, slave) +
2906                                3 * delta_in_ticks)) {
2907                         slave->new_link = BOND_LINK_DOWN;
2908                         commit++;
2909                 }
2910
2911                 /*
2912                  * Active slave is down if:
2913                  * - more than 2*delta since transmitting OR
2914                  * - (more than 2*delta since receive AND
2915                  *    the bond has an IP address)
2916                  */
2917                 if ((slave->state == BOND_STATE_ACTIVE) &&
2918                     (time_after_eq(jiffies, slave->dev->trans_start +
2919                                     2 * delta_in_ticks) ||
2920                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2921                                      + 2 * delta_in_ticks)))) {
2922                         slave->new_link = BOND_LINK_DOWN;
2923                         commit++;
2924                 }
2925         }
2926
2927         read_lock(&bond->curr_slave_lock);
2928
2929         /*
2930          * Trigger a commit if the primary option setting has changed.
2931          */
2932         if (bond->primary_slave &&
2933             (bond->primary_slave != bond->curr_active_slave) &&
2934             (bond->primary_slave->link == BOND_LINK_UP))
2935                 commit++;
2936
2937         read_unlock(&bond->curr_slave_lock);
2938
2939         return commit;
2940 }
2941
2942 /*
2943  * Called to commit link state changes noted by inspection step of
2944  * active-backup mode ARP monitor.
2945  *
2946  * Called with RTNL and bond->lock for read.
2947  */
2948 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2949 {
2950         struct slave *slave;
2951         int i;
2952
2953         bond_for_each_slave(bond, slave, i) {
2954                 switch (slave->new_link) {
2955                 case BOND_LINK_NOCHANGE:
2956                         continue;
2957
2958                 case BOND_LINK_UP:
2959                         write_lock_bh(&bond->curr_slave_lock);
2960
2961                         if (!bond->curr_active_slave &&
2962                             time_before_eq(jiffies, slave->dev->trans_start +
2963                                            delta_in_ticks)) {
2964                                 slave->link = BOND_LINK_UP;
2965                                 bond_change_active_slave(bond, slave);
2966                                 bond->current_arp_slave = NULL;
2967
2968                                 printk(KERN_INFO DRV_NAME
2969                                        ": %s: %s is up and now the "
2970                                        "active interface\n",
2971                                        bond->dev->name, slave->dev->name);
2972
2973                         } else if (bond->curr_active_slave != slave) {
2974                                 /* this slave has just come up but we
2975                                  * already have a current slave; this can
2976                                  * also happen if bond_enslave adds a new
2977                                  * slave that is up while we are searching
2978                                  * for a new slave
2979                                  */
2980                                 slave->link = BOND_LINK_UP;
2981                                 bond_set_slave_inactive_flags(slave);
2982                                 bond->current_arp_slave = NULL;
2983
2984                                 printk(KERN_INFO DRV_NAME
2985                                        ": %s: backup interface %s is now up\n",
2986                                        bond->dev->name, slave->dev->name);
2987                         }
2988
2989                         write_unlock_bh(&bond->curr_slave_lock);
2990
2991                         break;
2992
2993                 case BOND_LINK_DOWN:
2994                         if (slave->link_failure_count < UINT_MAX)
2995                                 slave->link_failure_count++;
2996
2997                         slave->link = BOND_LINK_DOWN;
2998
2999                         if (slave == bond->curr_active_slave) {
3000                                 printk(KERN_INFO DRV_NAME
3001                                        ": %s: link status down for active "
3002                                        "interface %s, disabling it\n",
3003                                        bond->dev->name, slave->dev->name);
3004
3005                                 bond_set_slave_inactive_flags(slave);
3006
3007                                 write_lock_bh(&bond->curr_slave_lock);
3008
3009                                 bond_select_active_slave(bond);
3010                                 if (bond->curr_active_slave)
3011                                         bond->curr_active_slave->jiffies =
3012                                                 jiffies;
3013
3014                                 write_unlock_bh(&bond->curr_slave_lock);
3015
3016                                 bond->current_arp_slave = NULL;
3017
3018                         } else if (slave->state == BOND_STATE_BACKUP) {
3019                                 printk(KERN_INFO DRV_NAME
3020                                        ": %s: backup interface %s is now down\n",
3021                                        bond->dev->name, slave->dev->name);
3022
3023                                 bond_set_slave_inactive_flags(slave);
3024                         }
3025                         break;
3026
3027                 default:
3028                         printk(KERN_ERR DRV_NAME
3029                                ": %s: impossible: new_link %d on slave %s\n",
3030                                bond->dev->name, slave->new_link,
3031                                slave->dev->name);
3032                 }
3033         }
3034
3035         /*
3036          * No race with changes to primary via sysfs, as we hold rtnl.
3037          */
3038         if (bond->primary_slave &&
3039             (bond->primary_slave != bond->curr_active_slave) &&
3040             (bond->primary_slave->link == BOND_LINK_UP)) {
3041                 write_lock_bh(&bond->curr_slave_lock);
3042                 bond_change_active_slave(bond, bond->primary_slave);
3043                 write_unlock_bh(&bond->curr_slave_lock);
3044         }
3045
3046         bond_set_carrier(bond);
3047 }
3048
3049 /*
3050  * Send ARP probes for active-backup mode ARP monitor.
3051  *
3052  * Called with bond->lock held for read.
3053  */
3054 static void bond_ab_arp_probe(struct bonding *bond)
3055 {
3056         struct slave *slave;
3057         int i;
3058
3059         read_lock(&bond->curr_slave_lock);
3060
3061         if (bond->current_arp_slave && bond->curr_active_slave)
3062                 printk("PROBE: c_arp %s && cas %s BAD\n",
3063                        bond->current_arp_slave->dev->name,
3064                        bond->curr_active_slave->dev->name);
3065
3066         if (bond->curr_active_slave) {
3067                 bond_arp_send_all(bond, bond->curr_active_slave);
3068                 read_unlock(&bond->curr_slave_lock);
3069                 return;
3070         }
3071
3072         read_unlock(&bond->curr_slave_lock);
3073
3074         /* if we don't have a curr_active_slave, search for the next available
3075          * backup slave from the current_arp_slave and make it the candidate
3076          * for becoming the curr_active_slave
3077          */
3078
3079         if (!bond->current_arp_slave) {
3080                 bond->current_arp_slave = bond->first_slave;
3081                 if (!bond->current_arp_slave)
3082                         return;
3083         }
3084
3085         bond_set_slave_inactive_flags(bond->current_arp_slave);
3086
3087         /* search for next candidate */
3088         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3089                 if (IS_UP(slave->dev)) {
3090                         slave->link = BOND_LINK_BACK;
3091                         bond_set_slave_active_flags(slave);
3092                         bond_arp_send_all(bond, slave);
3093                         slave->jiffies = jiffies;
3094                         bond->current_arp_slave = slave;
3095                         break;
3096                 }
3097
3098                 /* if the link state is up at this point, we
3099                  * mark it down - this can happen if we have
3100                  * simultaneous link failures and
3101                  * reselect_active_interface doesn't make this
3102                  * one the current slave so it is still marked
3103                  * up when it is actually down
3104                  */
3105                 if (slave->link == BOND_LINK_UP) {
3106                         slave->link = BOND_LINK_DOWN;
3107                         if (slave->link_failure_count < UINT_MAX)
3108                                 slave->link_failure_count++;
3109
3110                         bond_set_slave_inactive_flags(slave);
3111
3112                         printk(KERN_INFO DRV_NAME
3113                                ": %s: backup interface %s is now down.\n",
3114                                bond->dev->name, slave->dev->name);
3115                 }
3116         }
3117 }
3118
3119 void bond_activebackup_arp_mon(struct work_struct *work)
3120 {
3121         struct bonding *bond = container_of(work, struct bonding,
3122                                             arp_work.work);
3123         int delta_in_ticks;
3124
3125         read_lock(&bond->lock);
3126
3127         if (bond->kill_timers)
3128                 goto out;
3129
3130         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3131
3132         if (bond->slave_cnt == 0)
3133                 goto re_arm;
3134
3135         if (bond->send_grat_arp) {
3136                 read_lock(&bond->curr_slave_lock);
3137                 bond_send_gratuitous_arp(bond);
3138                 read_unlock(&bond->curr_slave_lock);
3139         }
3140
3141         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3142                 read_unlock(&bond->lock);
3143                 rtnl_lock();
3144                 read_lock(&bond->lock);
3145
3146                 bond_ab_arp_commit(bond, delta_in_ticks);
3147
3148                 read_unlock(&bond->lock);
3149                 rtnl_unlock();
3150                 read_lock(&bond->lock);
3151         }
3152
3153         bond_ab_arp_probe(bond);
3154
3155 re_arm:
3156         if (bond->params.arp_interval) {
3157                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3158         }
3159 out:
3160         read_unlock(&bond->lock);
3161 }
3162
3163 /*------------------------------ proc/seq_file-------------------------------*/
3164
3165 #ifdef CONFIG_PROC_FS
3166
3167 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3168 {
3169         struct bonding *bond = seq->private;
3170         loff_t off = 0;
3171         struct slave *slave;
3172         int i;
3173
3174         /* make sure the bond won't be taken away */
3175         read_lock(&dev_base_lock);
3176         read_lock(&bond->lock);
3177
3178         if (*pos == 0) {
3179                 return SEQ_START_TOKEN;
3180         }
3181
3182         bond_for_each_slave(bond, slave, i) {
3183                 if (++off == *pos) {
3184                         return slave;
3185                 }
3186         }
3187
3188         return NULL;
3189 }
3190
3191 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3192 {
3193         struct bonding *bond = seq->private;
3194         struct slave *slave = v;
3195
3196         ++*pos;
3197         if (v == SEQ_START_TOKEN) {
3198                 return bond->first_slave;
3199         }
3200
3201         slave = slave->next;
3202
3203         return (slave == bond->first_slave) ? NULL : slave;
3204 }
3205
3206 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3207 {
3208         struct bonding *bond = seq->private;
3209
3210         read_unlock(&bond->lock);
3211         read_unlock(&dev_base_lock);
3212 }
3213
3214 static void bond_info_show_master(struct seq_file *seq)
3215 {
3216         struct bonding *bond = seq->private;
3217         struct slave *curr;
3218         int i;
3219         u32 target;
3220
3221         read_lock(&bond->curr_slave_lock);
3222         curr = bond->curr_active_slave;
3223         read_unlock(&bond->curr_slave_lock);
3224
3225         seq_printf(seq, "Bonding Mode: %s",
3226                    bond_mode_name(bond->params.mode));
3227
3228         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3229             bond->params.fail_over_mac)
3230                 seq_printf(seq, " (fail_over_mac %s)",
3231                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3232
3233         seq_printf(seq, "\n");
3234
3235         if (bond->params.mode == BOND_MODE_XOR ||
3236                 bond->params.mode == BOND_MODE_8023AD) {
3237                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3238                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3239                         bond->params.xmit_policy);
3240         }
3241
3242         if (USES_PRIMARY(bond->params.mode)) {
3243                 seq_printf(seq, "Primary Slave: %s\n",
3244                            (bond->primary_slave) ?
3245                            bond->primary_slave->dev->name : "None");
3246
3247                 seq_printf(seq, "Currently Active Slave: %s\n",
3248                            (curr) ? curr->dev->name : "None");
3249         }
3250
3251         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3252                    "up" : "down");
3253         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3254         seq_printf(seq, "Up Delay (ms): %d\n",
3255                    bond->params.updelay * bond->params.miimon);
3256         seq_printf(seq, "Down Delay (ms): %d\n",
3257                    bond->params.downdelay * bond->params.miimon);
3258
3259
3260         /* ARP information */
3261         if(bond->params.arp_interval > 0) {
3262                 int printed=0;
3263                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3264                                 bond->params.arp_interval);
3265
3266                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3267
3268                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3269                         if (!bond->params.arp_targets[i])
3270                                 continue;
3271                         if (printed)
3272                                 seq_printf(seq, ",");
3273                         target = ntohl(bond->params.arp_targets[i]);
3274                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3275                         printed = 1;
3276                 }
3277                 seq_printf(seq, "\n");
3278         }
3279
3280         if (bond->params.mode == BOND_MODE_8023AD) {
3281                 struct ad_info ad_info;
3282                 DECLARE_MAC_BUF(mac);
3283
3284                 seq_puts(seq, "\n802.3ad info\n");
3285                 seq_printf(seq, "LACP rate: %s\n",
3286                            (bond->params.lacp_fast) ? "fast" : "slow");
3287
3288                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3289                         seq_printf(seq, "bond %s has no active aggregator\n",
3290                                    bond->dev->name);
3291                 } else {
3292                         seq_printf(seq, "Active Aggregator Info:\n");
3293
3294                         seq_printf(seq, "\tAggregator ID: %d\n",
3295                                    ad_info.aggregator_id);
3296                         seq_printf(seq, "\tNumber of ports: %d\n",
3297                                    ad_info.ports);
3298                         seq_printf(seq, "\tActor Key: %d\n",
3299                                    ad_info.actor_key);
3300                         seq_printf(seq, "\tPartner Key: %d\n",
3301                                    ad_info.partner_key);
3302                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3303                                    print_mac(mac, ad_info.partner_system));
3304                 }
3305         }
3306 }
3307
3308 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3309 {
3310         struct bonding *bond = seq->private;
3311         DECLARE_MAC_BUF(mac);
3312
3313         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3314         seq_printf(seq, "MII Status: %s\n",
3315                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3316         seq_printf(seq, "Link Failure Count: %u\n",
3317                    slave->link_failure_count);
3318
3319         seq_printf(seq,
3320                    "Permanent HW addr: %s\n",
3321                    print_mac(mac, slave->perm_hwaddr));
3322
3323         if (bond->params.mode == BOND_MODE_8023AD) {
3324                 const struct aggregator *agg
3325                         = SLAVE_AD_INFO(slave).port.aggregator;
3326
3327                 if (agg) {
3328                         seq_printf(seq, "Aggregator ID: %d\n",
3329                                    agg->aggregator_identifier);
3330                 } else {
3331                         seq_puts(seq, "Aggregator ID: N/A\n");
3332                 }
3333         }
3334 }
3335
3336 static int bond_info_seq_show(struct seq_file *seq, void *v)
3337 {
3338         if (v == SEQ_START_TOKEN) {
3339                 seq_printf(seq, "%s\n", version);
3340                 bond_info_show_master(seq);
3341         } else {
3342                 bond_info_show_slave(seq, v);
3343         }
3344
3345         return 0;
3346 }
3347
3348 static struct seq_operations bond_info_seq_ops = {
3349         .start = bond_info_seq_start,
3350         .next  = bond_info_seq_next,
3351         .stop  = bond_info_seq_stop,
3352         .show  = bond_info_seq_show,
3353 };
3354
3355 static int bond_info_open(struct inode *inode, struct file *file)
3356 {
3357         struct seq_file *seq;
3358         struct proc_dir_entry *proc;
3359         int res;
3360
3361         res = seq_open(file, &bond_info_seq_ops);
3362         if (!res) {
3363                 /* recover the pointer buried in proc_dir_entry data */
3364                 seq = file->private_data;
3365                 proc = PDE(inode);
3366                 seq->private = proc->data;
3367         }
3368
3369         return res;
3370 }
3371
3372 static const struct file_operations bond_info_fops = {
3373         .owner   = THIS_MODULE,
3374         .open    = bond_info_open,
3375         .read    = seq_read,
3376         .llseek  = seq_lseek,
3377         .release = seq_release,
3378 };
3379
3380 static int bond_create_proc_entry(struct bonding *bond)
3381 {
3382         struct net_device *bond_dev = bond->dev;
3383
3384         if (bond_proc_dir) {
3385                 bond->proc_entry = proc_create_data(bond_dev->name,
3386                                                     S_IRUGO, bond_proc_dir,
3387                                                     &bond_info_fops, bond);
3388                 if (bond->proc_entry == NULL) {
3389                         printk(KERN_WARNING DRV_NAME
3390                                ": Warning: Cannot create /proc/net/%s/%s\n",
3391                                DRV_NAME, bond_dev->name);
3392                 } else {
3393                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3394                 }
3395         }
3396
3397         return 0;
3398 }
3399
3400 static void bond_remove_proc_entry(struct bonding *bond)
3401 {
3402         if (bond_proc_dir && bond->proc_entry) {
3403                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3404                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3405                 bond->proc_entry = NULL;
3406         }
3407 }
3408
3409 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3410  * Caller must hold rtnl_lock.
3411  */
3412 static void bond_create_proc_dir(void)
3413 {
3414         int len = strlen(DRV_NAME);
3415
3416         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3417              bond_proc_dir = bond_proc_dir->next) {
3418                 if ((bond_proc_dir->namelen == len) &&
3419                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3420                         break;
3421                 }
3422         }
3423
3424         if (!bond_proc_dir) {
3425                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3426                 if (bond_proc_dir) {
3427                         bond_proc_dir->owner = THIS_MODULE;
3428                 } else {
3429                         printk(KERN_WARNING DRV_NAME
3430                                 ": Warning: cannot create /proc/net/%s\n",
3431                                 DRV_NAME);
3432                 }
3433         }
3434 }
3435
3436 /* Destroy the bonding directory under /proc/net, if empty.
3437  * Caller must hold rtnl_lock.
3438  */
3439 static void bond_destroy_proc_dir(void)
3440 {
3441         struct proc_dir_entry *de;
3442
3443         if (!bond_proc_dir) {
3444                 return;
3445         }
3446
3447         /* verify that the /proc dir is empty */
3448         for (de = bond_proc_dir->subdir; de; de = de->next) {
3449                 /* ignore . and .. */
3450                 if (*(de->name) != '.') {
3451                         break;
3452                 }
3453         }
3454
3455         if (de) {
3456                 if (bond_proc_dir->owner == THIS_MODULE) {
3457                         bond_proc_dir->owner = NULL;
3458                 }
3459         } else {
3460                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3461                 bond_proc_dir = NULL;
3462         }
3463 }
3464 #endif /* CONFIG_PROC_FS */
3465
3466 /*-------------------------- netdev event handling --------------------------*/
3467
3468 /*
3469  * Change device name
3470  */
3471 static int bond_event_changename(struct bonding *bond)
3472 {
3473 #ifdef CONFIG_PROC_FS
3474         bond_remove_proc_entry(bond);
3475         bond_create_proc_entry(bond);
3476 #endif
3477         down_write(&(bonding_rwsem));
3478         bond_destroy_sysfs_entry(bond);
3479         bond_create_sysfs_entry(bond);
3480         up_write(&(bonding_rwsem));
3481         return NOTIFY_DONE;
3482 }
3483
3484 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3485 {
3486         struct bonding *event_bond = bond_dev->priv;
3487
3488         switch (event) {
3489         case NETDEV_CHANGENAME:
3490                 return bond_event_changename(event_bond);
3491         case NETDEV_UNREGISTER:
3492                 bond_release_all(event_bond->dev);
3493                 break;
3494         default:
3495                 break;
3496         }
3497
3498         return NOTIFY_DONE;
3499 }
3500
3501 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3502 {
3503         struct net_device *bond_dev = slave_dev->master;
3504         struct bonding *bond = bond_dev->priv;
3505
3506         switch (event) {
3507         case NETDEV_UNREGISTER:
3508                 if (bond_dev) {
3509                         if (bond->setup_by_slave)
3510                                 bond_release_and_destroy(bond_dev, slave_dev);
3511                         else
3512                                 bond_release(bond_dev, slave_dev);
3513                 }
3514                 break;
3515         case NETDEV_CHANGE:
3516                 /*
3517                  * TODO: is this what we get if somebody
3518                  * sets up a hierarchical bond, then rmmod's
3519                  * one of the slave bonding devices?
3520                  */
3521                 break;
3522         case NETDEV_DOWN:
3523                 /*
3524                  * ... Or is it this?
3525                  */
3526                 break;
3527         case NETDEV_CHANGEMTU:
3528                 /*
3529                  * TODO: Should slaves be allowed to
3530                  * independently alter their MTU?  For
3531                  * an active-backup bond, slaves need
3532                  * not be the same type of device, so
3533                  * MTUs may vary.  For other modes,
3534                  * slaves arguably should have the
3535                  * same MTUs. To do this, we'd need to
3536                  * take over the slave's change_mtu
3537                  * function for the duration of their
3538                  * servitude.
3539                  */
3540                 break;
3541         case NETDEV_CHANGENAME:
3542                 /*
3543                  * TODO: handle changing the primary's name
3544                  */
3545                 break;
3546         case NETDEV_FEAT_CHANGE:
3547                 bond_compute_features(bond);
3548                 break;
3549         default:
3550                 break;
3551         }
3552
3553         return NOTIFY_DONE;
3554 }
3555
3556 /*
3557  * bond_netdev_event: handle netdev notifier chain events.
3558  *
3559  * This function receives events for the netdev chain.  The caller (an
3560  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3561  * locks for us to safely manipulate the slave devices (RTNL lock,
3562  * dev_probe_lock).
3563  */
3564 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3565 {
3566         struct net_device *event_dev = (struct net_device *)ptr;
3567
3568         if (dev_net(event_dev) != &init_net)
3569                 return NOTIFY_DONE;
3570
3571         dprintk("event_dev: %s, event: %lx\n",
3572                 (event_dev ? event_dev->name : "None"),
3573                 event);
3574
3575         if (!(event_dev->priv_flags & IFF_BONDING))
3576                 return NOTIFY_DONE;
3577
3578         if (event_dev->flags & IFF_MASTER) {
3579                 dprintk("IFF_MASTER\n");
3580                 return bond_master_netdev_event(event, event_dev);
3581         }
3582
3583         if (event_dev->flags & IFF_SLAVE) {
3584                 dprintk("IFF_SLAVE\n");
3585                 return bond_slave_netdev_event(event, event_dev);
3586         }
3587
3588         return NOTIFY_DONE;
3589 }
3590
3591 /*
3592  * bond_inetaddr_event: handle inetaddr notifier chain events.
3593  *
3594  * We keep track of device IPs primarily to use as source addresses in
3595  * ARP monitor probes (rather than spewing out broadcasts all the time).
3596  *
3597  * We track one IP for the main device (if it has one), plus one per VLAN.
3598  */
3599 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3600 {
3601         struct in_ifaddr *ifa = ptr;
3602         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3603         struct bonding *bond;
3604         struct vlan_entry *vlan;
3605
3606         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3607                 return NOTIFY_DONE;
3608
3609         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3610                 if (bond->dev == event_dev) {
3611                         switch (event) {
3612                         case NETDEV_UP:
3613                                 bond->master_ip = ifa->ifa_local;
3614                                 return NOTIFY_OK;
3615                         case NETDEV_DOWN:
3616                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3617                                 return NOTIFY_OK;
3618                         default:
3619                                 return NOTIFY_DONE;
3620                         }
3621                 }
3622
3623                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3624                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3625                         if (vlan_dev == event_dev) {
3626                                 switch (event) {
3627                                 case NETDEV_UP:
3628                                         vlan->vlan_ip = ifa->ifa_local;
3629                                         return NOTIFY_OK;
3630                                 case NETDEV_DOWN:
3631                                         vlan->vlan_ip =
3632                                                 bond_glean_dev_ip(vlan_dev);
3633                                         return NOTIFY_OK;
3634                                 default:
3635                                         return NOTIFY_DONE;
3636                                 }
3637                         }
3638                 }
3639         }
3640         return NOTIFY_DONE;
3641 }
3642
3643 static struct notifier_block bond_netdev_notifier = {
3644         .notifier_call = bond_netdev_event,
3645 };
3646
3647 static struct notifier_block bond_inetaddr_notifier = {
3648         .notifier_call = bond_inetaddr_event,
3649 };
3650
3651 /*-------------------------- Packet type handling ---------------------------*/
3652
3653 /* register to receive lacpdus on a bond */
3654 static void bond_register_lacpdu(struct bonding *bond)
3655 {
3656         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3657
3658         /* initialize packet type */
3659         pk_type->type = PKT_TYPE_LACPDU;
3660         pk_type->dev = bond->dev;
3661         pk_type->func = bond_3ad_lacpdu_recv;
3662
3663         dev_add_pack(pk_type);
3664 }
3665
3666 /* unregister to receive lacpdus on a bond */
3667 static void bond_unregister_lacpdu(struct bonding *bond)
3668 {
3669         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3670 }
3671
3672 void bond_register_arp(struct bonding *bond)
3673 {
3674         struct packet_type *pt = &bond->arp_mon_pt;
3675
3676         if (pt->type)
3677                 return;
3678
3679         pt->type = htons(ETH_P_ARP);
3680         pt->dev = bond->dev;
3681         pt->func = bond_arp_rcv;
3682         dev_add_pack(pt);
3683 }
3684
3685 void bond_unregister_arp(struct bonding *bond)
3686 {
3687         struct packet_type *pt = &bond->arp_mon_pt;
3688
3689         dev_remove_pack(pt);
3690         pt->type = 0;
3691 }
3692
3693 /*---------------------------- Hashing Policies -----------------------------*/
3694
3695 /*
3696  * Hash for the output device based upon layer 2 and layer 3 data. If
3697  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3698  */
3699 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3700                                      struct net_device *bond_dev, int count)
3701 {
3702         struct ethhdr *data = (struct ethhdr *)skb->data;
3703         struct iphdr *iph = ip_hdr(skb);
3704
3705         if (skb->protocol == htons(ETH_P_IP)) {
3706                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3707                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3708         }
3709
3710         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3711 }
3712
3713 /*
3714  * Hash for the output device based upon layer 3 and layer 4 data. If
3715  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3716  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3717  */
3718 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3719                                     struct net_device *bond_dev, int count)
3720 {
3721         struct ethhdr *data = (struct ethhdr *)skb->data;
3722         struct iphdr *iph = ip_hdr(skb);
3723         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3724         int layer4_xor = 0;
3725
3726         if (skb->protocol == htons(ETH_P_IP)) {
3727                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3728                     (iph->protocol == IPPROTO_TCP ||
3729                      iph->protocol == IPPROTO_UDP)) {
3730                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3731                 }
3732                 return (layer4_xor ^
3733                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3734
3735         }
3736
3737         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3738 }
3739
3740 /*
3741  * Hash for the output device based upon layer 2 data
3742  */
3743 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3744                                    struct net_device *bond_dev, int count)
3745 {
3746         struct ethhdr *data = (struct ethhdr *)skb->data;
3747
3748         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3749 }
3750
3751 /*-------------------------- Device entry points ----------------------------*/
3752
3753 static int bond_open(struct net_device *bond_dev)
3754 {
3755         struct bonding *bond = bond_dev->priv;
3756
3757         bond->kill_timers = 0;
3758
3759         if ((bond->params.mode == BOND_MODE_TLB) ||
3760             (bond->params.mode == BOND_MODE_ALB)) {
3761                 /* bond_alb_initialize must be called before the timer
3762                  * is started.
3763                  */
3764                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3765                         /* something went wrong - fail the open operation */
3766                         return -1;
3767                 }
3768
3769                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3770                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3771         }
3772
3773         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3774                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3775                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3776         }
3777
3778         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3779                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3780                         INIT_DELAYED_WORK(&bond->arp_work,
3781                                           bond_activebackup_arp_mon);
3782                 else
3783                         INIT_DELAYED_WORK(&bond->arp_work,
3784                                           bond_loadbalance_arp_mon);
3785
3786                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3787                 if (bond->params.arp_validate)
3788                         bond_register_arp(bond);
3789         }
3790
3791         if (bond->params.mode == BOND_MODE_8023AD) {
3792                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3793                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3794                 /* register to receive LACPDUs */
3795                 bond_register_lacpdu(bond);
3796         }
3797
3798         return 0;
3799 }
3800
3801 static int bond_close(struct net_device *bond_dev)
3802 {
3803         struct bonding *bond = bond_dev->priv;
3804
3805         if (bond->params.mode == BOND_MODE_8023AD) {
3806                 /* Unregister the receive of LACPDUs */
3807                 bond_unregister_lacpdu(bond);
3808         }
3809
3810         if (bond->params.arp_validate)
3811                 bond_unregister_arp(bond);
3812
3813         write_lock_bh(&bond->lock);
3814
3815         bond->send_grat_arp = 0;
3816
3817         /* signal timers not to re-arm */
3818         bond->kill_timers = 1;
3819
3820         write_unlock_bh(&bond->lock);
3821
3822         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3823                 cancel_delayed_work(&bond->mii_work);
3824         }
3825
3826         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3827                 cancel_delayed_work(&bond->arp_work);
3828         }
3829
3830         switch (bond->params.mode) {
3831         case BOND_MODE_8023AD:
3832                 cancel_delayed_work(&bond->ad_work);
3833                 break;
3834         case BOND_MODE_TLB:
3835         case BOND_MODE_ALB:
3836                 cancel_delayed_work(&bond->alb_work);
3837                 break;
3838         default:
3839                 break;
3840         }
3841
3842
3843         if ((bond->params.mode == BOND_MODE_TLB) ||
3844             (bond->params.mode == BOND_MODE_ALB)) {
3845                 /* Must be called only after all
3846                  * slaves have been released
3847                  */
3848                 bond_alb_deinitialize(bond);
3849         }
3850
3851         return 0;
3852 }
3853
3854 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3855 {
3856         struct bonding *bond = bond_dev->priv;
3857         struct net_device_stats *stats = &(bond->stats), *sstats;
3858         struct net_device_stats local_stats;
3859         struct slave *slave;
3860         int i;
3861
3862         memset(&local_stats, 0, sizeof(struct net_device_stats));
3863
3864         read_lock_bh(&bond->lock);
3865
3866         bond_for_each_slave(bond, slave, i) {
3867                 sstats = slave->dev->get_stats(slave->dev);
3868                 local_stats.rx_packets += sstats->rx_packets;
3869                 local_stats.rx_bytes += sstats->rx_bytes;
3870                 local_stats.rx_errors += sstats->rx_errors;
3871                 local_stats.rx_dropped += sstats->rx_dropped;
3872
3873                 local_stats.tx_packets += sstats->tx_packets;
3874                 local_stats.tx_bytes += sstats->tx_bytes;
3875                 local_stats.tx_errors += sstats->tx_errors;
3876                 local_stats.tx_dropped += sstats->tx_dropped;
3877
3878                 local_stats.multicast += sstats->multicast;
3879                 local_stats.collisions += sstats->collisions;
3880
3881                 local_stats.rx_length_errors += sstats->rx_length_errors;
3882                 local_stats.rx_over_errors += sstats->rx_over_errors;
3883                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3884                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3885                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3886                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3887
3888                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3889                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3890                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3891                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3892                 local_stats.tx_window_errors += sstats->tx_window_errors;
3893         }
3894
3895         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3896
3897         read_unlock_bh(&bond->lock);
3898
3899         return stats;
3900 }
3901
3902 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3903 {
3904         struct net_device *slave_dev = NULL;
3905         struct ifbond k_binfo;
3906         struct ifbond __user *u_binfo = NULL;
3907         struct ifslave k_sinfo;
3908         struct ifslave __user *u_sinfo = NULL;
3909         struct mii_ioctl_data *mii = NULL;
3910         int res = 0;
3911
3912         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3913                 bond_dev->name, cmd);
3914
3915         switch (cmd) {
3916         case SIOCGMIIPHY:
3917                 mii = if_mii(ifr);
3918                 if (!mii) {
3919                         return -EINVAL;
3920                 }
3921                 mii->phy_id = 0;
3922                 /* Fall Through */
3923         case SIOCGMIIREG:
3924                 /*
3925                  * We do this again just in case we were called by SIOCGMIIREG
3926                  * instead of SIOCGMIIPHY.
3927                  */
3928                 mii = if_mii(ifr);
3929                 if (!mii) {
3930                         return -EINVAL;
3931                 }
3932
3933                 if (mii->reg_num == 1) {
3934                         struct bonding *bond = bond_dev->priv;
3935                         mii->val_out = 0;
3936                         read_lock(&bond->lock);
3937                         read_lock(&bond->curr_slave_lock);
3938                         if (netif_carrier_ok(bond->dev)) {
3939                                 mii->val_out = BMSR_LSTATUS;
3940                         }
3941                         read_unlock(&bond->curr_slave_lock);
3942                         read_unlock(&bond->lock);
3943                 }
3944
3945                 return 0;
3946         case BOND_INFO_QUERY_OLD:
3947         case SIOCBONDINFOQUERY:
3948                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3949
3950                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3951                         return -EFAULT;
3952                 }
3953
3954                 res = bond_info_query(bond_dev, &k_binfo);
3955                 if (res == 0) {
3956                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3957                                 return -EFAULT;
3958                         }
3959                 }
3960
3961                 return res;
3962         case BOND_SLAVE_INFO_QUERY_OLD:
3963         case SIOCBONDSLAVEINFOQUERY:
3964                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3965
3966                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3967                         return -EFAULT;
3968                 }
3969
3970                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3971                 if (res == 0) {
3972                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3973                                 return -EFAULT;
3974                         }
3975                 }
3976
3977                 return res;
3978         default:
3979                 /* Go on */
3980                 break;
3981         }
3982
3983         if (!capable(CAP_NET_ADMIN)) {
3984                 return -EPERM;
3985         }
3986
3987         down_write(&(bonding_rwsem));
3988         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3989
3990         dprintk("slave_dev=%p: \n", slave_dev);
3991
3992         if (!slave_dev) {
3993                 res = -ENODEV;
3994         } else {
3995                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3996                 switch (cmd) {
3997                 case BOND_ENSLAVE_OLD:
3998                 case SIOCBONDENSLAVE:
3999                         res = bond_enslave(bond_dev, slave_dev);
4000                         break;
4001                 case BOND_RELEASE_OLD:
4002                 case SIOCBONDRELEASE:
4003                         res = bond_release(bond_dev, slave_dev);
4004                         break;
4005                 case BOND_SETHWADDR_OLD:
4006                 case SIOCBONDSETHWADDR:
4007                         res = bond_sethwaddr(bond_dev, slave_dev);
4008                         break;
4009                 case BOND_CHANGE_ACTIVE_OLD:
4010                 case SIOCBONDCHANGEACTIVE:
4011                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4012                         break;
4013                 default:
4014                         res = -EOPNOTSUPP;
4015                 }
4016
4017                 dev_put(slave_dev);
4018         }
4019
4020         up_write(&(bonding_rwsem));
4021         return res;
4022 }
4023
4024 static void bond_set_multicast_list(struct net_device *bond_dev)
4025 {
4026         struct bonding *bond = bond_dev->priv;
4027         struct dev_mc_list *dmi;
4028
4029         /*
4030          * Do promisc before checking multicast_mode
4031          */
4032         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4033                 /*
4034                  * FIXME: Need to handle the error when one of the multi-slaves
4035                  * encounters error.
4036                  */
4037                 bond_set_promiscuity(bond, 1);
4038         }
4039
4040         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4041                 bond_set_promiscuity(bond, -1);
4042         }
4043
4044         /* set allmulti flag to slaves */
4045         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4046                 /*
4047                  * FIXME: Need to handle the error when one of the multi-slaves
4048                  * encounters error.
4049                  */
4050                 bond_set_allmulti(bond, 1);
4051         }
4052
4053         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4054                 bond_set_allmulti(bond, -1);
4055         }
4056
4057         read_lock(&bond->lock);
4058
4059         bond->flags = bond_dev->flags;
4060
4061         /* looking for addresses to add to slaves' mc list */
4062         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4063                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4064                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4065                 }
4066         }
4067
4068         /* looking for addresses to delete from slaves' list */
4069         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4070                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4071                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4072                 }
4073         }
4074
4075         /* save master's multicast list */
4076         bond_mc_list_destroy(bond);
4077         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4078
4079         read_unlock(&bond->lock);
4080 }
4081
4082 /*
4083  * Change the MTU of all of a master's slaves to match the master
4084  */
4085 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4086 {
4087         struct bonding *bond = bond_dev->priv;
4088         struct slave *slave, *stop_at;
4089         int res = 0;
4090         int i;
4091
4092         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4093                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4094
4095         /* Can't hold bond->lock with bh disabled here since
4096          * some base drivers panic. On the other hand we can't
4097          * hold bond->lock without bh disabled because we'll
4098          * deadlock. The only solution is to rely on the fact
4099          * that we're under rtnl_lock here, and the slaves
4100          * list won't change. This doesn't solve the problem
4101          * of setting the slave's MTU while it is
4102          * transmitting, but the assumption is that the base
4103          * driver can handle that.
4104          *
4105          * TODO: figure out a way to safely iterate the slaves
4106          * list, but without holding a lock around the actual
4107          * call to the base driver.
4108          */
4109
4110         bond_for_each_slave(bond, slave, i) {
4111                 dprintk("s %p s->p %p c_m %p\n", slave,
4112                         slave->prev, slave->dev->change_mtu);
4113
4114                 res = dev_set_mtu(slave->dev, new_mtu);
4115
4116                 if (res) {
4117                         /* If we failed to set the slave's mtu to the new value
4118                          * we must abort the operation even in ACTIVE_BACKUP
4119                          * mode, because if we allow the backup slaves to have
4120                          * different mtu values than the active slave we'll
4121                          * need to change their mtu when doing a failover. That
4122                          * means changing their mtu from timer context, which
4123                          * is probably not a good idea.
4124                          */
4125                         dprintk("err %d %s\n", res, slave->dev->name);
4126                         goto unwind;
4127                 }
4128         }
4129
4130         bond_dev->mtu = new_mtu;
4131
4132         return 0;
4133
4134 unwind:
4135         /* unwind from head to the slave that failed */
4136         stop_at = slave;
4137         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4138                 int tmp_res;
4139
4140                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4141                 if (tmp_res) {
4142                         dprintk("unwind err %d dev %s\n", tmp_res,
4143                                 slave->dev->name);
4144                 }
4145         }
4146
4147         return res;
4148 }
4149
4150 /*
4151  * Change HW address
4152  *
4153  * Note that many devices must be down to change the HW address, and
4154  * downing the master releases all slaves.  We can make bonds full of
4155  * bonding devices to test this, however.
4156  */
4157 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4158 {
4159         struct bonding *bond = bond_dev->priv;
4160         struct sockaddr *sa = addr, tmp_sa;
4161         struct slave *slave, *stop_at;
4162         int res = 0;
4163         int i;
4164
4165         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4166
4167         /*
4168          * If fail_over_mac is set to active, do nothing and return
4169          * success.  Returning an error causes ifenslave to fail.
4170          */
4171         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4172                 return 0;
4173
4174         if (!is_valid_ether_addr(sa->sa_data)) {
4175                 return -EADDRNOTAVAIL;
4176         }
4177
4178         /* Can't hold bond->lock with bh disabled here since
4179          * some base drivers panic. On the other hand we can't
4180          * hold bond->lock without bh disabled because we'll
4181          * deadlock. The only solution is to rely on the fact
4182          * that we're under rtnl_lock here, and the slaves
4183          * list won't change. This doesn't solve the problem
4184          * of setting the slave's hw address while it is
4185          * transmitting, but the assumption is that the base
4186          * driver can handle that.
4187          *
4188          * TODO: figure out a way to safely iterate the slaves
4189          * list, but without holding a lock around the actual
4190          * call to the base driver.
4191          */
4192
4193         bond_for_each_slave(bond, slave, i) {
4194                 dprintk("slave %p %s\n", slave, slave->dev->name);
4195
4196                 if (slave->dev->set_mac_address == NULL) {
4197                         res = -EOPNOTSUPP;
4198                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4199                         goto unwind;
4200                 }
4201
4202                 res = dev_set_mac_address(slave->dev, addr);
4203                 if (res) {
4204                         /* TODO: consider downing the slave
4205                          * and retry ?
4206                          * User should expect communications
4207                          * breakage anyway until ARP finish
4208                          * updating, so...
4209                          */
4210                         dprintk("err %d %s\n", res, slave->dev->name);
4211                         goto unwind;
4212                 }
4213         }
4214
4215         /* success */
4216         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4217         return 0;
4218
4219 unwind:
4220         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4221         tmp_sa.sa_family = bond_dev->type;
4222
4223         /* unwind from head to the slave that failed */
4224         stop_at = slave;
4225         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4226                 int tmp_res;
4227
4228                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4229                 if (tmp_res) {
4230                         dprintk("unwind err %d dev %s\n", tmp_res,
4231                                 slave->dev->name);
4232                 }
4233         }
4234
4235         return res;
4236 }
4237
4238 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4239 {
4240         struct bonding *bond = bond_dev->priv;
4241         struct slave *slave, *start_at;
4242         int i, slave_no, res = 1;
4243
4244         read_lock(&bond->lock);
4245
4246         if (!BOND_IS_OK(bond)) {
4247                 goto out;
4248         }
4249
4250         /*
4251          * Concurrent TX may collide on rr_tx_counter; we accept that
4252          * as being rare enough not to justify using an atomic op here
4253          */
4254         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4255
4256         bond_for_each_slave(bond, slave, i) {
4257                 slave_no--;
4258                 if (slave_no < 0) {
4259                         break;
4260                 }
4261         }
4262
4263         start_at = slave;
4264         bond_for_each_slave_from(bond, slave, i, start_at) {
4265                 if (IS_UP(slave->dev) &&
4266                     (slave->link == BOND_LINK_UP) &&
4267                     (slave->state == BOND_STATE_ACTIVE)) {
4268                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4269                         break;
4270                 }
4271         }
4272
4273 out:
4274         if (res) {
4275                 /* no suitable interface, frame not sent */
4276                 dev_kfree_skb(skb);
4277         }
4278         read_unlock(&bond->lock);
4279         return 0;
4280 }
4281
4282
4283 /*
4284  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4285  * the bond has a usable interface.
4286  */
4287 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4288 {
4289         struct bonding *bond = bond_dev->priv;
4290         int res = 1;
4291
4292         read_lock(&bond->lock);
4293         read_lock(&bond->curr_slave_lock);
4294
4295         if (!BOND_IS_OK(bond)) {
4296                 goto out;
4297         }
4298
4299         if (!bond->curr_active_slave)
4300                 goto out;
4301
4302         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4303
4304 out:
4305         if (res) {
4306                 /* no suitable interface, frame not sent */
4307                 dev_kfree_skb(skb);
4308         }
4309         read_unlock(&bond->curr_slave_lock);
4310         read_unlock(&bond->lock);
4311         return 0;
4312 }
4313
4314 /*
4315  * In bond_xmit_xor() , we determine the output device by using a pre-
4316  * determined xmit_hash_policy(), If the selected device is not enabled,
4317  * find the next active slave.
4318  */
4319 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4320 {
4321         struct bonding *bond = bond_dev->priv;
4322         struct slave *slave, *start_at;
4323         int slave_no;
4324         int i;
4325         int res = 1;
4326
4327         read_lock(&bond->lock);
4328
4329         if (!BOND_IS_OK(bond)) {
4330                 goto out;
4331         }
4332
4333         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4334
4335         bond_for_each_slave(bond, slave, i) {
4336                 slave_no--;
4337                 if (slave_no < 0) {
4338                         break;
4339                 }
4340         }
4341
4342         start_at = slave;
4343
4344         bond_for_each_slave_from(bond, slave, i, start_at) {
4345                 if (IS_UP(slave->dev) &&
4346                     (slave->link == BOND_LINK_UP) &&
4347                     (slave->state == BOND_STATE_ACTIVE)) {
4348                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4349                         break;
4350                 }
4351         }
4352
4353 out:
4354         if (res) {
4355                 /* no suitable interface, frame not sent */
4356                 dev_kfree_skb(skb);
4357         }
4358         read_unlock(&bond->lock);
4359         return 0;
4360 }
4361
4362 /*
4363  * in broadcast mode, we send everything to all usable interfaces.
4364  */
4365 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4366 {
4367         struct bonding *bond = bond_dev->priv;
4368         struct slave *slave, *start_at;
4369         struct net_device *tx_dev = NULL;
4370         int i;
4371         int res = 1;
4372
4373         read_lock(&bond->lock);
4374
4375         if (!BOND_IS_OK(bond)) {
4376                 goto out;
4377         }
4378
4379         read_lock(&bond->curr_slave_lock);
4380         start_at = bond->curr_active_slave;
4381         read_unlock(&bond->curr_slave_lock);
4382
4383         if (!start_at) {
4384                 goto out;
4385         }
4386
4387         bond_for_each_slave_from(bond, slave, i, start_at) {
4388                 if (IS_UP(slave->dev) &&
4389                     (slave->link == BOND_LINK_UP) &&
4390                     (slave->state == BOND_STATE_ACTIVE)) {
4391                         if (tx_dev) {
4392                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4393                                 if (!skb2) {
4394                                         printk(KERN_ERR DRV_NAME
4395                                                ": %s: Error: bond_xmit_broadcast(): "
4396                                                "skb_clone() failed\n",
4397                                                bond_dev->name);
4398                                         continue;
4399                                 }
4400
4401                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4402                                 if (res) {
4403                                         dev_kfree_skb(skb2);
4404                                         continue;
4405                                 }
4406                         }
4407                         tx_dev = slave->dev;
4408                 }
4409         }
4410
4411         if (tx_dev) {
4412                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4413         }
4414
4415 out:
4416         if (res) {
4417                 /* no suitable interface, frame not sent */
4418                 dev_kfree_skb(skb);
4419         }
4420         /* frame sent to all suitable interfaces */
4421         read_unlock(&bond->lock);
4422         return 0;
4423 }
4424
4425 /*------------------------- Device initialization ---------------------------*/
4426
4427 static void bond_set_xmit_hash_policy(struct bonding *bond)
4428 {
4429         switch (bond->params.xmit_policy) {
4430         case BOND_XMIT_POLICY_LAYER23:
4431                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4432                 break;
4433         case BOND_XMIT_POLICY_LAYER34:
4434                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4435                 break;
4436         case BOND_XMIT_POLICY_LAYER2:
4437         default:
4438                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4439                 break;
4440         }
4441 }
4442
4443 /*
4444  * set bond mode specific net device operations
4445  */
4446 void bond_set_mode_ops(struct bonding *bond, int mode)
4447 {
4448         struct net_device *bond_dev = bond->dev;
4449
4450         switch (mode) {
4451         case BOND_MODE_ROUNDROBIN:
4452                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4453                 break;
4454         case BOND_MODE_ACTIVEBACKUP:
4455                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4456                 break;
4457         case BOND_MODE_XOR:
4458                 bond_dev->hard_start_xmit = bond_xmit_xor;
4459                 bond_set_xmit_hash_policy(bond);
4460                 break;
4461         case BOND_MODE_BROADCAST:
4462                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4463                 break;
4464         case BOND_MODE_8023AD:
4465                 bond_set_master_3ad_flags(bond);
4466                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4467                 bond_set_xmit_hash_policy(bond);
4468                 break;
4469         case BOND_MODE_ALB:
4470                 bond_set_master_alb_flags(bond);
4471                 /* FALLTHRU */
4472         case BOND_MODE_TLB:
4473                 bond_dev->hard_start_xmit = bond_alb_xmit;
4474                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4475                 break;
4476         default:
4477                 /* Should never happen, mode already checked */
4478                 printk(KERN_ERR DRV_NAME
4479                        ": %s: Error: Unknown bonding mode %d\n",
4480                        bond_dev->name,
4481                        mode);
4482                 break;
4483         }
4484 }
4485
4486 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4487                                     struct ethtool_drvinfo *drvinfo)
4488 {
4489         strncpy(drvinfo->driver, DRV_NAME, 32);
4490         strncpy(drvinfo->version, DRV_VERSION, 32);
4491         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4492 }
4493
4494 static const struct ethtool_ops bond_ethtool_ops = {
4495         .get_drvinfo            = bond_ethtool_get_drvinfo,
4496         .get_link               = ethtool_op_get_link,
4497         .get_tx_csum            = ethtool_op_get_tx_csum,
4498         .get_sg                 = ethtool_op_get_sg,
4499         .get_tso                = ethtool_op_get_tso,
4500         .get_ufo                = ethtool_op_get_ufo,
4501         .get_flags              = ethtool_op_get_flags,
4502 };
4503
4504 /*
4505  * Does not allocate but creates a /proc entry.
4506  * Allowed to fail.
4507  */
4508 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4509 {
4510         struct bonding *bond = bond_dev->priv;
4511
4512         dprintk("Begin bond_init for %s\n", bond_dev->name);
4513
4514         /* initialize rwlocks */
4515         rwlock_init(&bond->lock);
4516         rwlock_init(&bond->curr_slave_lock);
4517
4518         bond->params = *params; /* copy params struct */
4519
4520         bond->wq = create_singlethread_workqueue(bond_dev->name);
4521         if (!bond->wq)
4522                 return -ENOMEM;
4523
4524         /* Initialize pointers */
4525         bond->first_slave = NULL;
4526         bond->curr_active_slave = NULL;
4527         bond->current_arp_slave = NULL;
4528         bond->primary_slave = NULL;
4529         bond->dev = bond_dev;
4530         bond->send_grat_arp = 0;
4531         bond->setup_by_slave = 0;
4532         INIT_LIST_HEAD(&bond->vlan_list);
4533
4534         /* Initialize the device entry points */
4535         bond_dev->open = bond_open;
4536         bond_dev->stop = bond_close;
4537         bond_dev->get_stats = bond_get_stats;
4538         bond_dev->do_ioctl = bond_do_ioctl;
4539         bond_dev->ethtool_ops = &bond_ethtool_ops;
4540         bond_dev->set_multicast_list = bond_set_multicast_list;
4541         bond_dev->change_mtu = bond_change_mtu;
4542         bond_dev->set_mac_address = bond_set_mac_address;
4543         bond_dev->validate_addr = NULL;
4544
4545         bond_set_mode_ops(bond, bond->params.mode);
4546
4547         bond_dev->destructor = free_netdev;
4548
4549         /* Initialize the device options */
4550         bond_dev->tx_queue_len = 0;
4551         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4552         bond_dev->priv_flags |= IFF_BONDING;
4553
4554         /* At first, we block adding VLANs. That's the only way to
4555          * prevent problems that occur when adding VLANs over an
4556          * empty bond. The block will be removed once non-challenged
4557          * slaves are enslaved.
4558          */
4559         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4560
4561         /* don't acquire bond device's netif_tx_lock when
4562          * transmitting */
4563         bond_dev->features |= NETIF_F_LLTX;
4564
4565         /* By default, we declare the bond to be fully
4566          * VLAN hardware accelerated capable. Special
4567          * care is taken in the various xmit functions
4568          * when there are slaves that are not hw accel
4569          * capable
4570          */
4571         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4572         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4573         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4574         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4575                                NETIF_F_HW_VLAN_RX |
4576                                NETIF_F_HW_VLAN_FILTER);
4577
4578 #ifdef CONFIG_PROC_FS
4579         bond_create_proc_entry(bond);
4580 #endif
4581         list_add_tail(&bond->bond_list, &bond_dev_list);
4582
4583         return 0;
4584 }
4585
4586 /* De-initialize device specific data.
4587  * Caller must hold rtnl_lock.
4588  */
4589 static void bond_deinit(struct net_device *bond_dev)
4590 {
4591         struct bonding *bond = bond_dev->priv;
4592
4593         list_del(&bond->bond_list);
4594
4595 #ifdef CONFIG_PROC_FS
4596         bond_remove_proc_entry(bond);
4597 #endif
4598 }
4599
4600 static void bond_work_cancel_all(struct bonding *bond)
4601 {
4602         write_lock_bh(&bond->lock);
4603         bond->kill_timers = 1;
4604         write_unlock_bh(&bond->lock);
4605
4606         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4607                 cancel_delayed_work(&bond->mii_work);
4608
4609         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4610                 cancel_delayed_work(&bond->arp_work);
4611
4612         if (bond->params.mode == BOND_MODE_ALB &&
4613             delayed_work_pending(&bond->alb_work))
4614                 cancel_delayed_work(&bond->alb_work);
4615
4616         if (bond->params.mode == BOND_MODE_8023AD &&
4617             delayed_work_pending(&bond->ad_work))
4618                 cancel_delayed_work(&bond->ad_work);
4619 }
4620
4621 /* Unregister and free all bond devices.
4622  * Caller must hold rtnl_lock.
4623  */
4624 static void bond_free_all(void)
4625 {
4626         struct bonding *bond, *nxt;
4627
4628         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4629                 struct net_device *bond_dev = bond->dev;
4630
4631                 bond_work_cancel_all(bond);
4632                 netif_addr_lock_bh(bond_dev);
4633                 bond_mc_list_destroy(bond);
4634                 netif_addr_unlock_bh(bond_dev);
4635                 /* Release the bonded slaves */
4636                 bond_release_all(bond_dev);
4637                 bond_destroy(bond);
4638         }
4639
4640 #ifdef CONFIG_PROC_FS
4641         bond_destroy_proc_dir();
4642 #endif
4643 }
4644
4645 /*------------------------- Module initialization ---------------------------*/
4646
4647 /*
4648  * Convert string input module parms.  Accept either the
4649  * number of the mode or its string name.  A bit complicated because
4650  * some mode names are substrings of other names, and calls from sysfs
4651  * may have whitespace in the name (trailing newlines, for example).
4652  */
4653 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4654 {
4655         int mode = -1, i, rv;
4656         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4657
4658         for (p = (char *)buf; *p; p++)
4659                 if (!(isdigit(*p) || isspace(*p)))
4660                         break;
4661
4662         if (*p)
4663                 rv = sscanf(buf, "%20s", modestr);
4664         else
4665                 rv = sscanf(buf, "%d", &mode);
4666
4667         if (!rv)
4668                 return -1;
4669
4670         for (i = 0; tbl[i].modename; i++) {
4671                 if (mode == tbl[i].mode)
4672                         return tbl[i].mode;
4673                 if (strcmp(modestr, tbl[i].modename) == 0)
4674                         return tbl[i].mode;
4675         }
4676
4677         return -1;
4678 }
4679
4680 static int bond_check_params(struct bond_params *params)
4681 {
4682         int arp_validate_value, fail_over_mac_value;
4683
4684         /*
4685          * Convert string parameters.
4686          */
4687         if (mode) {
4688                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4689                 if (bond_mode == -1) {
4690                         printk(KERN_ERR DRV_NAME
4691                                ": Error: Invalid bonding mode \"%s\"\n",
4692                                mode == NULL ? "NULL" : mode);
4693                         return -EINVAL;
4694                 }
4695         }
4696
4697         if (xmit_hash_policy) {
4698                 if ((bond_mode != BOND_MODE_XOR) &&
4699                     (bond_mode != BOND_MODE_8023AD)) {
4700                         printk(KERN_INFO DRV_NAME
4701                                ": xor_mode param is irrelevant in mode %s\n",
4702                                bond_mode_name(bond_mode));
4703                 } else {
4704                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4705                                                         xmit_hashtype_tbl);
4706                         if (xmit_hashtype == -1) {
4707                                 printk(KERN_ERR DRV_NAME
4708                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4709                                 xmit_hash_policy == NULL ? "NULL" :
4710                                        xmit_hash_policy);
4711                                 return -EINVAL;
4712                         }
4713                 }
4714         }
4715
4716         if (lacp_rate) {
4717                 if (bond_mode != BOND_MODE_8023AD) {
4718                         printk(KERN_INFO DRV_NAME
4719                                ": lacp_rate param is irrelevant in mode %s\n",
4720                                bond_mode_name(bond_mode));
4721                 } else {
4722                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4723                         if (lacp_fast == -1) {
4724                                 printk(KERN_ERR DRV_NAME
4725                                        ": Error: Invalid lacp rate \"%s\"\n",
4726                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4727                                 return -EINVAL;
4728                         }
4729                 }
4730         }
4731
4732         if (max_bonds < 0 || max_bonds > INT_MAX) {
4733                 printk(KERN_WARNING DRV_NAME
4734                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4735                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4736                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4737                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4738         }
4739
4740         if (miimon < 0) {
4741                 printk(KERN_WARNING DRV_NAME
4742                        ": Warning: miimon module parameter (%d), "
4743                        "not in range 0-%d, so it was reset to %d\n",
4744                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4745                 miimon = BOND_LINK_MON_INTERV;
4746         }
4747
4748         if (updelay < 0) {
4749                 printk(KERN_WARNING DRV_NAME
4750                        ": Warning: updelay module parameter (%d), "
4751                        "not in range 0-%d, so it was reset to 0\n",
4752                        updelay, INT_MAX);
4753                 updelay = 0;
4754         }
4755
4756         if (downdelay < 0) {
4757                 printk(KERN_WARNING DRV_NAME
4758                        ": Warning: downdelay module parameter (%d), "
4759                        "not in range 0-%d, so it was reset to 0\n",
4760                        downdelay, INT_MAX);
4761                 downdelay = 0;
4762         }
4763
4764         if ((use_carrier != 0) && (use_carrier != 1)) {
4765                 printk(KERN_WARNING DRV_NAME
4766                        ": Warning: use_carrier module parameter (%d), "
4767                        "not of valid value (0/1), so it was set to 1\n",
4768                        use_carrier);
4769                 use_carrier = 1;
4770         }
4771
4772         if (num_grat_arp < 0 || num_grat_arp > 255) {
4773                 printk(KERN_WARNING DRV_NAME
4774                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4775                        "was reset to 1 \n", num_grat_arp);
4776                 num_grat_arp = 1;
4777         }
4778
4779         /* reset values for 802.3ad */
4780         if (bond_mode == BOND_MODE_8023AD) {
4781                 if (!miimon) {
4782                         printk(KERN_WARNING DRV_NAME
4783                                ": Warning: miimon must be specified, "
4784                                "otherwise bonding will not detect link "
4785                                "failure, speed and duplex which are "
4786                                "essential for 802.3ad operation\n");
4787                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4788                         miimon = 100;
4789                 }
4790         }
4791
4792         /* reset values for TLB/ALB */
4793         if ((bond_mode == BOND_MODE_TLB) ||
4794             (bond_mode == BOND_MODE_ALB)) {
4795                 if (!miimon) {
4796                         printk(KERN_WARNING DRV_NAME
4797                                ": Warning: miimon must be specified, "
4798                                "otherwise bonding will not detect link "
4799                                "failure and link speed which are essential "
4800                                "for TLB/ALB load balancing\n");
4801                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4802                         miimon = 100;
4803                 }
4804         }
4805
4806         if (bond_mode == BOND_MODE_ALB) {
4807                 printk(KERN_NOTICE DRV_NAME
4808                        ": In ALB mode you might experience client "
4809                        "disconnections upon reconnection of a link if the "
4810                        "bonding module updelay parameter (%d msec) is "
4811                        "incompatible with the forwarding delay time of the "
4812                        "switch\n",
4813                        updelay);
4814         }
4815
4816         if (!miimon) {
4817                 if (updelay || downdelay) {
4818                         /* just warn the user the up/down delay will have
4819                          * no effect since miimon is zero...
4820                          */
4821                         printk(KERN_WARNING DRV_NAME
4822                                ": Warning: miimon module parameter not set "
4823                                "and updelay (%d) or downdelay (%d) module "
4824                                "parameter is set; updelay and downdelay have "
4825                                "no effect unless miimon is set\n",
4826                                updelay, downdelay);
4827                 }
4828         } else {
4829                 /* don't allow arp monitoring */
4830                 if (arp_interval) {
4831                         printk(KERN_WARNING DRV_NAME
4832                                ": Warning: miimon (%d) and arp_interval (%d) "
4833                                "can't be used simultaneously, disabling ARP "
4834                                "monitoring\n",
4835                                miimon, arp_interval);
4836                         arp_interval = 0;
4837                 }
4838
4839                 if ((updelay % miimon) != 0) {
4840                         printk(KERN_WARNING DRV_NAME
4841                                ": Warning: updelay (%d) is not a multiple "
4842                                "of miimon (%d), updelay rounded to %d ms\n",
4843                                updelay, miimon, (updelay / miimon) * miimon);
4844                 }
4845
4846                 updelay /= miimon;
4847
4848                 if ((downdelay % miimon) != 0) {
4849                         printk(KERN_WARNING DRV_NAME
4850                                ": Warning: downdelay (%d) is not a multiple "
4851                                "of miimon (%d), downdelay rounded to %d ms\n",
4852                                downdelay, miimon,
4853                                (downdelay / miimon) * miimon);
4854                 }
4855
4856                 downdelay /= miimon;
4857         }
4858
4859         if (arp_interval < 0) {
4860                 printk(KERN_WARNING DRV_NAME
4861                        ": Warning: arp_interval module parameter (%d) "
4862                        ", not in range 0-%d, so it was reset to %d\n",
4863                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4864                 arp_interval = BOND_LINK_ARP_INTERV;
4865         }
4866
4867         for (arp_ip_count = 0;
4868              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4869              arp_ip_count++) {
4870                 /* not complete check, but should be good enough to
4871                    catch mistakes */
4872                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4873                         printk(KERN_WARNING DRV_NAME
4874                                ": Warning: bad arp_ip_target module parameter "
4875                                "(%s), ARP monitoring will not be performed\n",
4876                                arp_ip_target[arp_ip_count]);
4877                         arp_interval = 0;
4878                 } else {
4879                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4880                         arp_target[arp_ip_count] = ip;
4881                 }
4882         }
4883
4884         if (arp_interval && !arp_ip_count) {
4885                 /* don't allow arping if no arp_ip_target given... */
4886                 printk(KERN_WARNING DRV_NAME
4887                        ": Warning: arp_interval module parameter (%d) "
4888                        "specified without providing an arp_ip_target "
4889                        "parameter, arp_interval was reset to 0\n",
4890                        arp_interval);
4891                 arp_interval = 0;
4892         }
4893
4894         if (arp_validate) {
4895                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4896                         printk(KERN_ERR DRV_NAME
4897                ": arp_validate only supported in active-backup mode\n");
4898                         return -EINVAL;
4899                 }
4900                 if (!arp_interval) {
4901                         printk(KERN_ERR DRV_NAME
4902                                ": arp_validate requires arp_interval\n");
4903                         return -EINVAL;
4904                 }
4905
4906                 arp_validate_value = bond_parse_parm(arp_validate,
4907                                                      arp_validate_tbl);
4908                 if (arp_validate_value == -1) {
4909                         printk(KERN_ERR DRV_NAME
4910                                ": Error: invalid arp_validate \"%s\"\n",
4911                                arp_validate == NULL ? "NULL" : arp_validate);
4912                         return -EINVAL;
4913                 }
4914         } else
4915                 arp_validate_value = 0;
4916
4917         if (miimon) {
4918                 printk(KERN_INFO DRV_NAME
4919                        ": MII link monitoring set to %d ms\n",
4920                        miimon);
4921         } else if (arp_interval) {
4922                 int i;
4923
4924                 printk(KERN_INFO DRV_NAME
4925                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4926                        arp_interval,
4927                        arp_validate_tbl[arp_validate_value].modename,
4928                        arp_ip_count);
4929
4930                 for (i = 0; i < arp_ip_count; i++)
4931                         printk (" %s", arp_ip_target[i]);
4932
4933                 printk("\n");
4934
4935         } else if (max_bonds) {
4936                 /* miimon and arp_interval not set, we need one so things
4937                  * work as expected, see bonding.txt for details
4938                  */
4939                 printk(KERN_WARNING DRV_NAME
4940                        ": Warning: either miimon or arp_interval and "
4941                        "arp_ip_target module parameters must be specified, "
4942                        "otherwise bonding will not detect link failures! see "
4943                        "bonding.txt for details.\n");
4944         }
4945
4946         if (primary && !USES_PRIMARY(bond_mode)) {
4947                 /* currently, using a primary only makes sense
4948                  * in active backup, TLB or ALB modes
4949                  */
4950                 printk(KERN_WARNING DRV_NAME
4951                        ": Warning: %s primary device specified but has no "
4952                        "effect in %s mode\n",
4953                        primary, bond_mode_name(bond_mode));
4954                 primary = NULL;
4955         }
4956
4957         if (fail_over_mac) {
4958                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4959                                                       fail_over_mac_tbl);
4960                 if (fail_over_mac_value == -1) {
4961                         printk(KERN_ERR DRV_NAME
4962                                ": Error: invalid fail_over_mac \"%s\"\n",
4963                                arp_validate == NULL ? "NULL" : arp_validate);
4964                         return -EINVAL;
4965                 }
4966
4967                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4968                         printk(KERN_WARNING DRV_NAME
4969                                ": Warning: fail_over_mac only affects "
4970                                "active-backup mode.\n");
4971         } else {
4972                 fail_over_mac_value = BOND_FOM_NONE;
4973         }
4974
4975         /* fill params struct with the proper values */
4976         params->mode = bond_mode;
4977         params->xmit_policy = xmit_hashtype;
4978         params->miimon = miimon;
4979         params->num_grat_arp = num_grat_arp;
4980         params->arp_interval = arp_interval;
4981         params->arp_validate = arp_validate_value;
4982         params->updelay = updelay;
4983         params->downdelay = downdelay;
4984         params->use_carrier = use_carrier;
4985         params->lacp_fast = lacp_fast;
4986         params->primary[0] = 0;
4987         params->fail_over_mac = fail_over_mac_value;
4988
4989         if (primary) {
4990                 strncpy(params->primary, primary, IFNAMSIZ);
4991                 params->primary[IFNAMSIZ - 1] = 0;
4992         }
4993
4994         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4995
4996         return 0;
4997 }
4998
4999 static struct lock_class_key bonding_netdev_xmit_lock_key;
5000 static struct lock_class_key bonding_netdev_addr_lock_key;
5001
5002 static void bond_set_lockdep_class_one(struct net_device *dev,
5003                                        struct netdev_queue *txq,
5004                                        void *_unused)
5005 {
5006         lockdep_set_class(&txq->_xmit_lock,
5007                           &bonding_netdev_xmit_lock_key);
5008 }
5009
5010 static void bond_set_lockdep_class(struct net_device *dev)
5011 {
5012         lockdep_set_class(&dev->addr_list_lock,
5013                           &bonding_netdev_addr_lock_key);
5014         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5015 }
5016
5017 /* Create a new bond based on the specified name and bonding parameters.
5018  * If name is NULL, obtain a suitable "bond%d" name for us.
5019  * Caller must NOT hold rtnl_lock; we need to release it here before we
5020  * set up our sysfs entries.
5021  */
5022 int bond_create(char *name, struct bond_params *params)
5023 {
5024         struct net_device *bond_dev;
5025         struct bonding *bond;
5026         int res;
5027
5028         rtnl_lock();
5029         down_write(&bonding_rwsem);
5030
5031         /* Check to see if the bond already exists. */
5032         if (name) {
5033                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5034                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5035                                 printk(KERN_ERR DRV_NAME
5036                                ": cannot add bond %s; it already exists\n",
5037                                        name);
5038                                 res = -EPERM;
5039                                 goto out_rtnl;
5040                         }
5041         }
5042
5043         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5044                                 ether_setup);
5045         if (!bond_dev) {
5046                 printk(KERN_ERR DRV_NAME
5047                        ": %s: eek! can't alloc netdev!\n",
5048                        name);
5049                 res = -ENOMEM;
5050                 goto out_rtnl;
5051         }
5052
5053         if (!name) {
5054                 res = dev_alloc_name(bond_dev, "bond%d");
5055                 if (res < 0)
5056                         goto out_netdev;
5057         }
5058
5059         /* bond_init() must be called after dev_alloc_name() (for the
5060          * /proc files), but before register_netdevice(), because we
5061          * need to set function pointers.
5062          */
5063
5064         res = bond_init(bond_dev, params);
5065         if (res < 0) {
5066                 goto out_netdev;
5067         }
5068
5069         res = register_netdevice(bond_dev);
5070         if (res < 0) {
5071                 goto out_bond;
5072         }
5073
5074         bond_set_lockdep_class(bond_dev);
5075
5076         netif_carrier_off(bond_dev);
5077
5078         up_write(&bonding_rwsem);
5079         rtnl_unlock(); /* allows sysfs registration of net device */
5080         res = bond_create_sysfs_entry(bond_dev->priv);
5081         if (res < 0) {
5082                 rtnl_lock();
5083                 down_write(&bonding_rwsem);
5084                 bond_deinit(bond_dev);
5085                 unregister_netdevice(bond_dev);
5086                 goto out_rtnl;
5087         }
5088
5089         return 0;
5090
5091 out_bond:
5092         bond_deinit(bond_dev);
5093 out_netdev:
5094         free_netdev(bond_dev);
5095 out_rtnl:
5096         up_write(&bonding_rwsem);
5097         rtnl_unlock();
5098         return res;
5099 }
5100
5101 static int __init bonding_init(void)
5102 {
5103         int i;
5104         int res;
5105         struct bonding *bond;
5106
5107         printk(KERN_INFO "%s", version);
5108
5109         res = bond_check_params(&bonding_defaults);
5110         if (res) {
5111                 goto out;
5112         }
5113
5114 #ifdef CONFIG_PROC_FS
5115         bond_create_proc_dir();
5116 #endif
5117
5118         init_rwsem(&bonding_rwsem);
5119
5120         for (i = 0; i < max_bonds; i++) {
5121                 res = bond_create(NULL, &bonding_defaults);
5122                 if (res)
5123                         goto err;
5124         }
5125
5126         res = bond_create_sysfs();
5127         if (res)
5128                 goto err;
5129
5130         register_netdevice_notifier(&bond_netdev_notifier);
5131         register_inetaddr_notifier(&bond_inetaddr_notifier);
5132
5133         goto out;
5134 err:
5135         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5136                 bond_work_cancel_all(bond);
5137                 destroy_workqueue(bond->wq);
5138         }
5139
5140         bond_destroy_sysfs();
5141
5142         rtnl_lock();
5143         bond_free_all();
5144         rtnl_unlock();
5145 out:
5146         return res;
5147
5148 }
5149
5150 static void __exit bonding_exit(void)
5151 {
5152         unregister_netdevice_notifier(&bond_netdev_notifier);
5153         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5154
5155         bond_destroy_sysfs();
5156
5157         rtnl_lock();
5158         bond_free_all();
5159         rtnl_unlock();
5160 }
5161
5162 module_init(bonding_init);
5163 module_exit(bonding_exit);
5164 MODULE_LICENSE("GPL");
5165 MODULE_VERSION(DRV_VERSION);
5166 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5167 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5168 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5169
5170 /*
5171  * Local variables:
5172  *  c-indent-level: 8
5173  *  c-basic-offset: 8
5174  *  tab-width: 8
5175  * End:
5176  */
5177