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

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 //#define BONDING_DEBUG 1
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <asm/system.h>
  57 #include <asm/io.h>
  58 #include <asm/dma.h>
  59 #include <asm/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <net/route.h>
  79 #include <net/net_namespace.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int num_grat_arp = 1;
  92 static int miimon       = BOND_LINK_MON_INTERV;
  93 static int updelay      = 0;
  94 static int downdelay    = 0;
  95 static int use_carrier  = 1;
  96 static char *mode       = NULL;
  97 static char *primary    = NULL;
  98 static char *lacp_rate  = NULL;
  99 static char *xmit_hash_policy = NULL;
 100 static int arp_interval = BOND_LINK_ARP_INTERV;
 101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
 102 static char *arp_validate = NULL;
 103 static char *fail_over_mac = NULL;
 104 struct bond_params bonding_defaults;
 105
 106 module_param(max_bonds, int, 0);
 107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 108 module_param(num_grat_arp, int, 0644);
 109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 110 module_param(miimon, int, 0);
 111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 112 module_param(updelay, int, 0);
 113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 114 module_param(downdelay, int, 0);
 115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 116                             "in milliseconds");
 117 module_param(use_carrier, int, 0);
 118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 119                               "0 for off, 1 for on (default)");
 120 module_param(mode, charp, 0);
 121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 122                        "1 for active-backup, 2 for balance-xor, "
 123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 124                        "6 for balance-alb");
 125 module_param(primary, charp, 0);
 126 MODULE_PARM_DESC(primary, "Primary network device to use");
 127 module_param(lacp_rate, charp, 0);
 128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 129                             "(slow/fast)");
 130 module_param(xmit_hash_policy, charp, 0);
 131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 132                                    ", 1 for layer 3+4");
 133 module_param(arp_interval, int, 0);
 134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 135 module_param_array(arp_ip_target, charp, NULL, 0);
 136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 137 module_param(arp_validate, charp, 0);
 138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 139 module_param(fail_over_mac, charp, 0);
 140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 141
 142 /*----------------------------- Global variables ----------------------------*/
 143
 144 static const char * const version =
 145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 146
 147 LIST_HEAD(bond_dev_list);
 148
 149 #ifdef CONFIG_PROC_FS
 150 static struct proc_dir_entry *bond_proc_dir = NULL;
 151 #endif
 152
 153 extern struct rw_semaphore bonding_rwsem;
 154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
 155 static int arp_ip_count = 0;
 156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
 158 static int lacp_fast    = 0;
 159
 160
 161 struct bond_parm_tbl bond_lacp_tbl[] = {
 162 {       "slow",         AD_LACP_SLOW},
 163 {       "fast",         AD_LACP_FAST},
 164 {       NULL,           -1},
 165 };
 166
 167 struct bond_parm_tbl bond_mode_tbl[] = {
 168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 170 {       "balance-xor",          BOND_MODE_XOR},
 171 {       "broadcast",            BOND_MODE_BROADCAST},
 172 {       "802.3ad",              BOND_MODE_8023AD},
 173 {       "balance-tlb",          BOND_MODE_TLB},
 174 {       "balance-alb",          BOND_MODE_ALB},
 175 {       NULL,                   -1},
 176 };
 177
 178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
 179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 182 {       NULL,                   -1},
 183 };
 184
 185 struct bond_parm_tbl arp_validate_tbl[] = {
 186 {       "none",                 BOND_ARP_VALIDATE_NONE},
 187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 189 {       "all",                  BOND_ARP_VALIDATE_ALL},
 190 {       NULL,                   -1},
 191 };
 192
 193 struct bond_parm_tbl fail_over_mac_tbl[] = {
 194 {       "none",                 BOND_FOM_NONE},
 195 {       "active",               BOND_FOM_ACTIVE},
 196 {       "follow",               BOND_FOM_FOLLOW},
 197 {       NULL,                   -1},
 198 };
 199
 200 /*-------------------------- Forward declarations ---------------------------*/
 201
 202 static void bond_send_gratuitous_arp(struct bonding *bond);
 203 static void bond_deinit(struct net_device *bond_dev);
 204
 205 /*---------------------------- General routines -----------------------------*/
 206
 207 static const char *bond_mode_name(int mode)
 208 {
 209         switch (mode) {
 210         case BOND_MODE_ROUNDROBIN :
 211                 return "load balancing (round-robin)";
 212         case BOND_MODE_ACTIVEBACKUP :
 213                 return "fault-tolerance (active-backup)";
 214         case BOND_MODE_XOR :
 215                 return "load balancing (xor)";
 216         case BOND_MODE_BROADCAST :
 217                 return "fault-tolerance (broadcast)";
 218         case BOND_MODE_8023AD:
 219                 return "IEEE 802.3ad Dynamic link aggregation";
 220         case BOND_MODE_TLB:
 221                 return "transmit load balancing";
 222         case BOND_MODE_ALB:
 223                 return "adaptive load balancing";
 224         default:
 225                 return "unknown";
 226         }
 227 }
 228
 229 /*---------------------------------- VLAN -----------------------------------*/
 230
 231 /**
 232  * bond_add_vlan - add a new vlan id on bond
 233  * @bond: bond that got the notification
 234  * @vlan_id: the vlan id to add
 235  *
 236  * Returns -ENOMEM if allocation failed.
 237  */
 238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 239 {
 240         struct vlan_entry *vlan;
 241
 242         dprintk("bond: %s, vlan id %d\n",
 243                 (bond ? bond->dev->name: "None"), vlan_id);
 244
 245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 246         if (!vlan) {
 247                 return -ENOMEM;
 248         }
 249
 250         INIT_LIST_HEAD(&vlan->vlan_list);
 251         vlan->vlan_id = vlan_id;
 252         vlan->vlan_ip = 0;
 253
 254         write_lock_bh(&bond->lock);
 255
 256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 257
 258         write_unlock_bh(&bond->lock);
 259
 260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 261
 262         return 0;
 263 }
 264
 265 /**
 266  * bond_del_vlan - delete a vlan id from bond
 267  * @bond: bond that got the notification
 268  * @vlan_id: the vlan id to delete
 269  *
 270  * returns -ENODEV if @vlan_id was not found in @bond.
 271  */
 272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 273 {
 274         struct vlan_entry *vlan;
 275         int res = -ENODEV;
 276
 277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 278
 279         write_lock_bh(&bond->lock);
 280
 281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 282                 if (vlan->vlan_id == vlan_id) {
 283                         list_del(&vlan->vlan_list);
 284
 285                         if ((bond->params.mode == BOND_MODE_TLB) ||
 286                             (bond->params.mode == BOND_MODE_ALB)) {
 287                                 bond_alb_clear_vlan(bond, vlan_id);
 288                         }
 289
 290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
 291                                 bond->dev->name);
 292
 293                         kfree(vlan);
 294
 295                         if (list_empty(&bond->vlan_list) &&
 296                             (bond->slave_cnt == 0)) {
 297                                 /* Last VLAN removed and no slaves, so
 298                                  * restore block on adding VLANs. This will
 299                                  * be removed once new slaves that are not
 300                                  * VLAN challenged will be added.
 301                                  */
 302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 303                         }
 304
 305                         res = 0;
 306                         goto out;
 307                 }
 308         }
 309
 310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 311                 bond->dev->name);
 312
 313 out:
 314         write_unlock_bh(&bond->lock);
 315         return res;
 316 }
 317
 318 /**
 319  * bond_has_challenged_slaves
 320  * @bond: the bond we're working on
 321  *
 322  * Searches the slave list. Returns 1 if a vlan challenged slave
 323  * was found, 0 otherwise.
 324  *
 325  * Assumes bond->lock is held.
 326  */
 327 static int bond_has_challenged_slaves(struct bonding *bond)
 328 {
 329         struct slave *slave;
 330         int i;
 331
 332         bond_for_each_slave(bond, slave, i) {
 333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 334                         dprintk("found VLAN challenged slave - %s\n",
 335                                 slave->dev->name);
 336                         return 1;
 337                 }
 338         }
 339
 340         dprintk("no VLAN challenged slaves found\n");
 341         return 0;
 342 }
 343
 344 /**
 345  * bond_next_vlan - safely skip to the next item in the vlans list.
 346  * @bond: the bond we're working on
 347  * @curr: item we're advancing from
 348  *
 349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 350  * or @curr->next otherwise (even if it is @curr itself again).
 351  *
 352  * Caller must hold bond->lock
 353  */
 354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 355 {
 356         struct vlan_entry *next, *last;
 357
 358         if (list_empty(&bond->vlan_list)) {
 359                 return NULL;
 360         }
 361
 362         if (!curr) {
 363                 next = list_entry(bond->vlan_list.next,
 364                                   struct vlan_entry, vlan_list);
 365         } else {
 366                 last = list_entry(bond->vlan_list.prev,
 367                                   struct vlan_entry, vlan_list);
 368                 if (last == curr) {
 369                         next = list_entry(bond->vlan_list.next,
 370                                           struct vlan_entry, vlan_list);
 371                 } else {
 372                         next = list_entry(curr->vlan_list.next,
 373                                           struct vlan_entry, vlan_list);
 374                 }
 375         }
 376
 377         return next;
 378 }
 379
 380 /**
 381  * bond_dev_queue_xmit - Prepare skb for xmit.
 382  *
 383  * @bond: bond device that got this skb for tx.
 384  * @skb: hw accel VLAN tagged skb to transmit
 385  * @slave_dev: slave that is supposed to xmit this skbuff
 386  *
 387  * When the bond gets an skb to transmit that is
 388  * already hardware accelerated VLAN tagged, and it
 389  * needs to relay this skb to a slave that is not
 390  * hw accel capable, the skb needs to be "unaccelerated",
 391  * i.e. strip the hwaccel tag and re-insert it as part
 392  * of the payload.
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
 395 {
 396         unsigned short uninitialized_var(vlan_id);
 397
 398         if (!list_empty(&bond->vlan_list) &&
 399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 400             vlan_get_tag(skb, &vlan_id) == 0) {
 401                 skb->dev = slave_dev;
 402                 skb = vlan_put_tag(skb, vlan_id);
 403                 if (!skb) {
 404                         /* vlan_put_tag() frees the skb in case of error,
 405                          * so return success here so the calling functions
 406                          * won't attempt to free is again.
 407                          */
 408                         return 0;
 409                 }
 410         } else {
 411                 skb->dev = slave_dev;
 412         }
 413
 414         skb->priority = 1;
 415         dev_queue_xmit(skb);
 416
 417         return 0;
 418 }
 419
 420 /*
 421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 423  * lock because:
 424  * a. This operation is performed in IOCTL context,
 425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 426  * c. Holding a lock with BH disabled while directly calling a base driver
 427  *    entry point is generally a BAD idea.
 428  *
 429  * The design of synchronization/protection for this operation in the 8021q
 430  * module is good for one or more VLAN devices over a single physical device
 431  * and cannot be extended for a teaming solution like bonding, so there is a
 432  * potential race condition here where a net device from the vlan group might
 433  * be referenced (either by a base driver or the 8021q code) while it is being
 434  * removed from the system. However, it turns out we're not making matters
 435  * worse, and if it works for regular VLAN usage it will work here too.
 436 */
 437
 438 /**
 439  * bond_vlan_rx_register - Propagates registration to slaves
 440  * @bond_dev: bonding net device that got called
 441  * @grp: vlan group being registered
 442  */
 443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
 444 {
 445         struct bonding *bond = bond_dev->priv;
 446         struct slave *slave;
 447         int i;
 448
 449         bond->vlgrp = grp;
 450
 451         bond_for_each_slave(bond, slave, i) {
 452                 struct net_device *slave_dev = slave->dev;
 453
 454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 455                     slave_dev->vlan_rx_register) {
 456                         slave_dev->vlan_rx_register(slave_dev, grp);
 457                 }
 458         }
 459 }
 460
 461 /**
 462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being added
 465  */
 466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = bond_dev->priv;
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i) {
 473                 struct net_device *slave_dev = slave->dev;
 474
 475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 476                     slave_dev->vlan_rx_add_vid) {
 477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
 478                 }
 479         }
 480
 481         res = bond_add_vlan(bond, vid);
 482         if (res) {
 483                 printk(KERN_ERR DRV_NAME
 484                        ": %s: Error: Failed to add vlan id %d\n",
 485                        bond_dev->name, vid);
 486         }
 487 }
 488
 489 /**
 490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 491  * @bond_dev: bonding net device that got called
 492  * @vid: vlan id being removed
 493  */
 494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 495 {
 496         struct bonding *bond = bond_dev->priv;
 497         struct slave *slave;
 498         struct net_device *vlan_dev;
 499         int i, res;
 500
 501         bond_for_each_slave(bond, slave, i) {
 502                 struct net_device *slave_dev = slave->dev;
 503
 504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 505                     slave_dev->vlan_rx_kill_vid) {
 506                         /* Save and then restore vlan_dev in the grp array,
 507                          * since the slave's driver might clear it.
 508                          */
 509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
 511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 512                 }
 513         }
 514
 515         res = bond_del_vlan(bond, vid);
 516         if (res) {
 517                 printk(KERN_ERR DRV_NAME
 518                        ": %s: Error: Failed to remove vlan id %d\n",
 519                        bond_dev->name, vid);
 520         }
 521 }
 522
 523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 524 {
 525         struct vlan_entry *vlan;
 526
 527         write_lock_bh(&bond->lock);
 528
 529         if (list_empty(&bond->vlan_list)) {
 530                 goto out;
 531         }
 532
 533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 534             slave_dev->vlan_rx_register) {
 535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
 536         }
 537
 538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 539             !(slave_dev->vlan_rx_add_vid)) {
 540                 goto out;
 541         }
 542
 543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 545         }
 546
 547 out:
 548         write_unlock_bh(&bond->lock);
 549 }
 550
 551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
 552 {
 553         struct vlan_entry *vlan;
 554         struct net_device *vlan_dev;
 555
 556         write_lock_bh(&bond->lock);
 557
 558         if (list_empty(&bond->vlan_list)) {
 559                 goto out;
 560         }
 561
 562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 563             !(slave_dev->vlan_rx_kill_vid)) {
 564                 goto unreg;
 565         }
 566
 567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 568                 /* Save and then restore vlan_dev in the grp array,
 569                  * since the slave's driver might clear it.
 570                  */
 571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 574         }
 575
 576 unreg:
 577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 578             slave_dev->vlan_rx_register) {
 579                 slave_dev->vlan_rx_register(slave_dev, NULL);
 580         }
 581
 582 out:
 583         write_unlock_bh(&bond->lock);
 584 }
 585
 586 /*------------------------------- Link status -------------------------------*/
 587
 588 /*
 589  * Set the carrier state for the master according to the state of its
 590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 591  * do special 802.3ad magic.
 592  *
 593  * Returns zero if carrier state does not change, nonzero if it does.
 594  */
 595 static int bond_set_carrier(struct bonding *bond)
 596 {
 597         struct slave *slave;
 598         int i;
 599
 600         if (bond->slave_cnt == 0)
 601                 goto down;
 602
 603         if (bond->params.mode == BOND_MODE_8023AD)
 604                 return bond_3ad_set_carrier(bond);
 605
 606         bond_for_each_slave(bond, slave, i) {
 607                 if (slave->link == BOND_LINK_UP) {
 608                         if (!netif_carrier_ok(bond->dev)) {
 609                                 netif_carrier_on(bond->dev);
 610                                 return 1;
 611                         }
 612                         return 0;
 613                 }
 614         }
 615
 616 down:
 617         if (netif_carrier_ok(bond->dev)) {
 618                 netif_carrier_off(bond->dev);
 619                 return 1;
 620         }
 621         return 0;
 622 }
 623
 624 /*
 625  * Get link speed and duplex from the slave's base driver
 626  * using ethtool. If for some reason the call fails or the
 627  * values are invalid, fake speed and duplex to 100/Full
 628  * and return error.
 629  */
 630 static int bond_update_speed_duplex(struct slave *slave)
 631 {
 632         struct net_device *slave_dev = slave->dev;
 633         struct ethtool_cmd etool;
 634         int res;
 635
 636         /* Fake speed and duplex */
 637         slave->speed = SPEED_100;
 638         slave->duplex = DUPLEX_FULL;
 639
 640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 641                 return -1;
 642
 643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 644         if (res < 0)
 645                 return -1;
 646
 647         switch (etool.speed) {
 648         case SPEED_10:
 649         case SPEED_100:
 650         case SPEED_1000:
 651         case SPEED_10000:
 652                 break;
 653         default:
 654                 return -1;
 655         }
 656
 657         switch (etool.duplex) {
 658         case DUPLEX_FULL:
 659         case DUPLEX_HALF:
 660                 break;
 661         default:
 662                 return -1;
 663         }
 664
 665         slave->speed = etool.speed;
 666         slave->duplex = etool.duplex;
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * if <dev> supports MII link status reporting, check its link status.
 673  *
 674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 675  * depening upon the setting of the use_carrier parameter.
 676  *
 677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 678  * can't tell and just pretend it is), or 0, meaning that the link is
 679  * down.
 680  *
 681  * If reporting is non-zero, instead of faking link up, return -1 if
 682  * both ETHTOOL and MII ioctls fail (meaning the device does not
 683  * support them).  If use_carrier is set, return whatever it says.
 684  * It'd be nice if there was a good way to tell if a driver supports
 685  * netif_carrier, but there really isn't.
 686  */
 687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
 688 {
 689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
 690         struct ifreq ifr;
 691         struct mii_ioctl_data *mii;
 692
 693         if (bond->params.use_carrier) {
 694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 695         }
 696
 697         ioctl = slave_dev->do_ioctl;
 698         if (ioctl) {
 699                 /* TODO: set pointer to correct ioctl on a per team member */
 700                 /*       bases to make this more efficient. that is, once  */
 701                 /*       we determine the correct ioctl, we will always    */
 702                 /*       call it and not the others for that team          */
 703                 /*       member.                                           */
 704
 705                 /*
 706                  * We cannot assume that SIOCGMIIPHY will also read a
 707                  * register; not all network drivers (e.g., e100)
 708                  * support that.
 709                  */
 710
 711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 713                 mii = if_mii(&ifr);
 714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 715                         mii->reg_num = MII_BMSR;
 716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
 717                                 return (mii->val_out & BMSR_LSTATUS);
 718                         }
 719                 }
 720         }
 721
 722         /*
 723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
 724          * attempt to get link status from it if the above MII ioctls fail.
 725          */
 726         if (slave_dev->ethtool_ops) {
 727                 if (slave_dev->ethtool_ops->get_link) {
 728                         u32 link;
 729
 730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 731
 732                         return link ? BMSR_LSTATUS : 0;
 733                 }
 734         }
 735
 736         /*
 737          * If reporting, report that either there's no dev->do_ioctl,
 738          * or both SIOCGMIIREG and get_link failed (meaning that we
 739          * cannot report link status).  If not reporting, pretend
 740          * we're ok.
 741          */
 742         return (reporting ? -1 : BMSR_LSTATUS);
 743 }
 744
 745 /*----------------------------- Multicast list ------------------------------*/
 746
 747 /*
 748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 749  */
 750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
 751 {
 752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 754 }
 755
 756 /*
 757  * returns dmi entry if found, NULL otherwise
 758  */
 759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
 760 {
 761         struct dev_mc_list *idmi;
 762
 763         for (idmi = mc_list; idmi; idmi = idmi->next) {
 764                 if (bond_is_dmi_same(dmi, idmi)) {
 765                         return idmi;
 766                 }
 767         }
 768
 769         return NULL;
 770 }
 771
 772 /*
 773  * Push the promiscuity flag down to appropriate slaves
 774  */
 775 static int bond_set_promiscuity(struct bonding *bond, int inc)
 776 {
 777         int err = 0;
 778         if (USES_PRIMARY(bond->params.mode)) {
 779                 /* write lock already acquired */
 780                 if (bond->curr_active_slave) {
 781                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 782                                                   inc);
 783                 }
 784         } else {
 785                 struct slave *slave;
 786                 int i;
 787                 bond_for_each_slave(bond, slave, i) {
 788                         err = dev_set_promiscuity(slave->dev, inc);
 789                         if (err)
 790                                 return err;
 791                 }
 792         }
 793         return err;
 794 }
 795
 796 /*
 797  * Push the allmulti flag down to all slaves
 798  */
 799 static int bond_set_allmulti(struct bonding *bond, int inc)
 800 {
 801         int err = 0;
 802         if (USES_PRIMARY(bond->params.mode)) {
 803                 /* write lock already acquired */
 804                 if (bond->curr_active_slave) {
 805                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 806                                                inc);
 807                 }
 808         } else {
 809                 struct slave *slave;
 810                 int i;
 811                 bond_for_each_slave(bond, slave, i) {
 812                         err = dev_set_allmulti(slave->dev, inc);
 813                         if (err)
 814                                 return err;
 815                 }
 816         }
 817         return err;
 818 }
 819
 820 /*
 821  * Add a Multicast address to slaves
 822  * according to mode
 823  */
 824 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 825 {
 826         if (USES_PRIMARY(bond->params.mode)) {
 827                 /* write lock already acquired */
 828                 if (bond->curr_active_slave) {
 829                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 830                 }
 831         } else {
 832                 struct slave *slave;
 833                 int i;
 834                 bond_for_each_slave(bond, slave, i) {
 835                         dev_mc_add(slave->dev, addr, alen, 0);
 836                 }
 837         }
 838 }
 839
 840 /*
 841  * Remove a multicast address from slave
 842  * according to mode
 843  */
 844 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 845 {
 846         if (USES_PRIMARY(bond->params.mode)) {
 847                 /* write lock already acquired */
 848                 if (bond->curr_active_slave) {
 849                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
 850                 }
 851         } else {
 852                 struct slave *slave;
 853                 int i;
 854                 bond_for_each_slave(bond, slave, i) {
 855                         dev_mc_delete(slave->dev, addr, alen, 0);
 856                 }
 857         }
 858 }
 859
 860
 861 /*
 862  * Retrieve the list of registered multicast addresses for the bonding
 863  * device and retransmit an IGMP JOIN request to the current active
 864  * slave.
 865  */
 866 static void bond_resend_igmp_join_requests(struct bonding *bond)
 867 {
 868         struct in_device *in_dev;
 869         struct ip_mc_list *im;
 870
 871         rcu_read_lock();
 872         in_dev = __in_dev_get_rcu(bond->dev);
 873         if (in_dev) {
 874                 for (im = in_dev->mc_list; im; im = im->next) {
 875                         ip_mc_rejoin_group(im);
 876                 }
 877         }
 878
 879         rcu_read_unlock();
 880 }
 881
 882 /*
 883  * Totally destroys the mc_list in bond
 884  */
 885 static void bond_mc_list_destroy(struct bonding *bond)
 886 {
 887         struct dev_mc_list *dmi;
 888
 889         dmi = bond->mc_list;
 890         while (dmi) {
 891                 bond->mc_list = dmi->next;
 892                 kfree(dmi);
 893                 dmi = bond->mc_list;
 894         }
 895         bond->mc_list = NULL;
 896 }
 897
 898 /*
 899  * Copy all the Multicast addresses from src to the bonding device dst
 900  */
 901 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 902                              gfp_t gfp_flag)
 903 {
 904         struct dev_mc_list *dmi, *new_dmi;
 905
 906         for (dmi = mc_list; dmi; dmi = dmi->next) {
 907                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 908
 909                 if (!new_dmi) {
 910                         /* FIXME: Potential memory leak !!! */
 911                         return -ENOMEM;
 912                 }
 913
 914                 new_dmi->next = bond->mc_list;
 915                 bond->mc_list = new_dmi;
 916                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 917                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 918                 new_dmi->dmi_users = dmi->dmi_users;
 919                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 920         }
 921
 922         return 0;
 923 }
 924
 925 /*
 926  * flush all members of flush->mc_list from device dev->mc_list
 927  */
 928 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
 929 {
 930         struct bonding *bond = bond_dev->priv;
 931         struct dev_mc_list *dmi;
 932
 933         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
 934                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 935         }
 936
 937         if (bond->params.mode == BOND_MODE_8023AD) {
 938                 /* del lacpdu mc addr from mc list */
 939                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 940
 941                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 942         }
 943 }
 944
 945 /*--------------------------- Active slave change ---------------------------*/
 946
 947 /*
 948  * Update the mc list and multicast-related flags for the new and
 949  * old active slaves (if any) according to the multicast mode, and
 950  * promiscuous flags unconditionally.
 951  */
 952 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
 953 {
 954         struct dev_mc_list *dmi;
 955
 956         if (!USES_PRIMARY(bond->params.mode)) {
 957                 /* nothing to do -  mc list is already up-to-date on
 958                  * all slaves
 959                  */
 960                 return;
 961         }
 962
 963         if (old_active) {
 964                 if (bond->dev->flags & IFF_PROMISC) {
 965                         dev_set_promiscuity(old_active->dev, -1);
 966                 }
 967
 968                 if (bond->dev->flags & IFF_ALLMULTI) {
 969                         dev_set_allmulti(old_active->dev, -1);
 970                 }
 971
 972                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 973                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 974                 }
 975         }
 976
 977         if (new_active) {
 978                 /* FIXME: Signal errors upstream. */
 979                 if (bond->dev->flags & IFF_PROMISC) {
 980                         dev_set_promiscuity(new_active->dev, 1);
 981                 }
 982
 983                 if (bond->dev->flags & IFF_ALLMULTI) {
 984                         dev_set_allmulti(new_active->dev, 1);
 985                 }
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
 988                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 989                 }
 990                 bond_resend_igmp_join_requests(bond);
 991         }
 992 }
 993
 994 /*
 995  * bond_do_fail_over_mac
 996  *
 997  * Perform special MAC address swapping for fail_over_mac settings
 998  *
 999  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000  */
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002                                   struct slave *new_active,
1003                                   struct slave *old_active)
1004 {
1005         u8 tmp_mac[ETH_ALEN];
1006         struct sockaddr saddr;
1007         int rv;
1008
1009         switch (bond->params.fail_over_mac) {
1010         case BOND_FOM_ACTIVE:
1011                 if (new_active)
1012                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1013                                new_active->dev->addr_len);
1014                 break;
1015         case BOND_FOM_FOLLOW:
1016                 /*
1017                  * if new_active && old_active, swap them
1018                  * if just old_active, do nothing (going to no active slave)
1019                  * if just new_active, set new_active to bond's MAC
1020                  */
1021                 if (!new_active)
1022                         return;
1023
1024                 write_unlock_bh(&bond->curr_slave_lock);
1025                 read_unlock(&bond->lock);
1026
1027                 if (old_active) {
1028                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030                                ETH_ALEN);
1031                         saddr.sa_family = new_active->dev->type;
1032                 } else {
1033                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034                         saddr.sa_family = bond->dev->type;
1035                 }
1036
1037                 rv = dev_set_mac_address(new_active->dev, &saddr);
1038                 if (rv) {
1039                         printk(KERN_ERR DRV_NAME
1040                                ": %s: Error %d setting MAC of slave %s\n",
1041                                bond->dev->name, -rv, new_active->dev->name);
1042                         goto out;
1043                 }
1044
1045                 if (!old_active)
1046                         goto out;
1047
1048                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049                 saddr.sa_family = old_active->dev->type;
1050
1051                 rv = dev_set_mac_address(old_active->dev, &saddr);
1052                 if (rv)
1053                         printk(KERN_ERR DRV_NAME
1054                                ": %s: Error %d setting MAC of slave %s\n",
1055                                bond->dev->name, -rv, new_active->dev->name);
1056 out:
1057                 read_lock(&bond->lock);
1058                 write_lock_bh(&bond->curr_slave_lock);
1059                 break;
1060         default:
1061                 printk(KERN_ERR DRV_NAME
1062                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063                        bond->dev->name, bond->params.fail_over_mac);
1064                 break;
1065         }
1066
1067 }
1068
1069
1070 /**
1071  * find_best_interface - select the best available slave to be the active one
1072  * @bond: our bonding struct
1073  *
1074  * Warning: Caller must hold curr_slave_lock for writing.
1075  */
1076 static struct slave *bond_find_best_slave(struct bonding *bond)
1077 {
1078         struct slave *new_active, *old_active;
1079         struct slave *bestslave = NULL;
1080         int mintime = bond->params.updelay;
1081         int i;
1082
1083         new_active = old_active = bond->curr_active_slave;
1084
1085         if (!new_active) { /* there were no active slaves left */
1086                 if (bond->slave_cnt > 0) {  /* found one slave */
1087                         new_active = bond->first_slave;
1088                 } else {
1089                         return NULL; /* still no slave, return NULL */
1090                 }
1091         }
1092
1093         /* first try the primary link; if arping, a link must tx/rx traffic
1094          * before it can be considered the curr_active_slave - also, we would skip
1095          * slaves between the curr_active_slave and primary_slave that may be up
1096          * and able to arp
1097          */
1098         if ((bond->primary_slave) &&
1099             (!bond->params.arp_interval) &&
1100             (IS_UP(bond->primary_slave->dev))) {
1101                 new_active = bond->primary_slave;
1102         }
1103
1104         /* remember where to stop iterating over the slaves */
1105         old_active = new_active;
1106
1107         bond_for_each_slave_from(bond, new_active, i, old_active) {
1108                 if (IS_UP(new_active->dev)) {
1109                         if (new_active->link == BOND_LINK_UP) {
1110                                 return new_active;
1111                         } else if (new_active->link == BOND_LINK_BACK) {
1112                                 /* link up, but waiting for stabilization */
1113                                 if (new_active->delay < mintime) {
1114                                         mintime = new_active->delay;
1115                                         bestslave = new_active;
1116                                 }
1117                         }
1118                 }
1119         }
1120
1121         return bestslave;
1122 }
1123
1124 /**
1125  * change_active_interface - change the active slave into the specified one
1126  * @bond: our bonding struct
1127  * @new: the new slave to make the active one
1128  *
1129  * Set the new slave to the bond's settings and unset them on the old
1130  * curr_active_slave.
1131  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132  *
1133  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134  * because it is apparently the best available slave we have, even though its
1135  * updelay hasn't timed out yet.
1136  *
1137  * If new_active is not NULL, caller must hold bond->lock for read and
1138  * curr_slave_lock for write_bh.
1139  */
1140 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 {
1142         struct slave *old_active = bond->curr_active_slave;
1143
1144         if (old_active == new_active) {
1145                 return;
1146         }
1147
1148         if (new_active) {
1149                 new_active->jiffies = jiffies;
1150
1151                 if (new_active->link == BOND_LINK_BACK) {
1152                         if (USES_PRIMARY(bond->params.mode)) {
1153                                 printk(KERN_INFO DRV_NAME
1154                                        ": %s: making interface %s the new "
1155                                        "active one %d ms earlier.\n",
1156                                        bond->dev->name, new_active->dev->name,
1157                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158                         }
1159
1160                         new_active->delay = 0;
1161                         new_active->link = BOND_LINK_UP;
1162
1163                         if (bond->params.mode == BOND_MODE_8023AD) {
1164                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165                         }
1166
1167                         if ((bond->params.mode == BOND_MODE_TLB) ||
1168                             (bond->params.mode == BOND_MODE_ALB)) {
1169                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170                         }
1171                 } else {
1172                         if (USES_PRIMARY(bond->params.mode)) {
1173                                 printk(KERN_INFO DRV_NAME
1174                                        ": %s: making interface %s the new "
1175                                        "active one.\n",
1176                                        bond->dev->name, new_active->dev->name);
1177                         }
1178                 }
1179         }
1180
1181         if (USES_PRIMARY(bond->params.mode)) {
1182                 bond_mc_swap(bond, new_active, old_active);
1183         }
1184
1185         if ((bond->params.mode == BOND_MODE_TLB) ||
1186             (bond->params.mode == BOND_MODE_ALB)) {
1187                 bond_alb_handle_active_change(bond, new_active);
1188                 if (old_active)
1189                         bond_set_slave_inactive_flags(old_active);
1190                 if (new_active)
1191                         bond_set_slave_active_flags(new_active);
1192         } else {
1193                 bond->curr_active_slave = new_active;
1194         }
1195
1196         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197                 if (old_active) {
1198                         bond_set_slave_inactive_flags(old_active);
1199                 }
1200
1201                 if (new_active) {
1202                         bond_set_slave_active_flags(new_active);
1203
1204                         if (bond->params.fail_over_mac)
1205                                 bond_do_fail_over_mac(bond, new_active,
1206                                                       old_active);
1207
1208                         bond->send_grat_arp = bond->params.num_grat_arp;
1209                         bond_send_gratuitous_arp(bond);
1210
1211                         write_unlock_bh(&bond->curr_slave_lock);
1212                         read_unlock(&bond->lock);
1213
1214                         netdev_bonding_change(bond->dev);
1215
1216                         read_lock(&bond->lock);
1217                         write_lock_bh(&bond->curr_slave_lock);
1218                 }
1219         }
1220 }
1221
1222 /**
1223  * bond_select_active_slave - select a new active slave, if needed
1224  * @bond: our bonding struct
1225  *
1226  * This functions shoud be called when one of the following occurs:
1227  * - The old curr_active_slave has been released or lost its link.
1228  * - The primary_slave has got its link back.
1229  * - A slave has got its link back and there's no old curr_active_slave.
1230  *
1231  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1232  */
1233 void bond_select_active_slave(struct bonding *bond)
1234 {
1235         struct slave *best_slave;
1236         int rv;
1237
1238         best_slave = bond_find_best_slave(bond);
1239         if (best_slave != bond->curr_active_slave) {
1240                 bond_change_active_slave(bond, best_slave);
1241                 rv = bond_set_carrier(bond);
1242                 if (!rv)
1243                         return;
1244
1245                 if (netif_carrier_ok(bond->dev)) {
1246                         printk(KERN_INFO DRV_NAME
1247                                ": %s: first active interface up!\n",
1248                                bond->dev->name);
1249                 } else {
1250                         printk(KERN_INFO DRV_NAME ": %s: "
1251                                "now running without any active interface !\n",
1252                                bond->dev->name);
1253                 }
1254         }
1255 }
1256
1257 /*--------------------------- slave list handling ---------------------------*/
1258
1259 /*
1260  * This function attaches the slave to the end of list.
1261  *
1262  * bond->lock held for writing by caller.
1263  */
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265 {
1266         if (bond->first_slave == NULL) { /* attaching the first slave */
1267                 new_slave->next = new_slave;
1268                 new_slave->prev = new_slave;
1269                 bond->first_slave = new_slave;
1270         } else {
1271                 new_slave->next = bond->first_slave;
1272                 new_slave->prev = bond->first_slave->prev;
1273                 new_slave->next->prev = new_slave;
1274                 new_slave->prev->next = new_slave;
1275         }
1276
1277         bond->slave_cnt++;
1278 }
1279
1280 /*
1281  * This function detaches the slave from the list.
1282  * WARNING: no check is made to verify if the slave effectively
1283  * belongs to <bond>.
1284  * Nothing is freed on return, structures are just unchained.
1285  * If any slave pointer in bond was pointing to <slave>,
1286  * it should be changed by the calling function.
1287  *
1288  * bond->lock held for writing by caller.
1289  */
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291 {
1292         if (slave->next) {
1293                 slave->next->prev = slave->prev;
1294         }
1295
1296         if (slave->prev) {
1297                 slave->prev->next = slave->next;
1298         }
1299
1300         if (bond->first_slave == slave) { /* slave is the first slave */
1301                 if (bond->slave_cnt > 1) { /* there are more slave */
1302                         bond->first_slave = slave->next;
1303                 } else {
1304                         bond->first_slave = NULL; /* slave was the last one */
1305                 }
1306         }
1307
1308         slave->next = NULL;
1309         slave->prev = NULL;
1310         bond->slave_cnt--;
1311 }
1312
1313 /*---------------------------------- IOCTL ----------------------------------*/
1314
1315 static int bond_sethwaddr(struct net_device *bond_dev,
1316                           struct net_device *slave_dev)
1317 {
1318         dprintk("bond_dev=%p\n", bond_dev);
1319         dprintk("slave_dev=%p\n", slave_dev);
1320         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1321         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1322         return 0;
1323 }
1324
1325 #define BOND_VLAN_FEATURES \
1326         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327          NETIF_F_HW_VLAN_FILTER)
1328
1329 /*
1330  * Compute the common dev->feature set available to all slaves.  Some
1331  * feature bits are managed elsewhere, so preserve those feature bits
1332  * on the master device.
1333  */
1334 static int bond_compute_features(struct bonding *bond)
1335 {
1336         struct slave *slave;
1337         struct net_device *bond_dev = bond->dev;
1338         unsigned long features = bond_dev->features;
1339         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1340                                                 bond_dev->hard_header_len);
1341         int i;
1342
1343         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1344         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1345
1346         if (!bond->first_slave)
1347                 goto done;
1348
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350
1351         bond_for_each_slave(bond, slave, i) {
1352                 features = netdev_increment_features(features,
1353                                                      slave->dev->features,
1354                                                      NETIF_F_ONE_FOR_ALL);
1355                 if (slave->dev->hard_header_len > max_hard_header_len)
1356                         max_hard_header_len = slave->dev->hard_header_len;
1357         }
1358
1359 done:
1360         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1361         bond_dev->features = netdev_fix_features(features, NULL);
1362         bond_dev->hard_header_len = max_hard_header_len;
1363
1364         return 0;
1365 }
1366
1367
1368 static void bond_setup_by_slave(struct net_device *bond_dev,
1369                                 struct net_device *slave_dev)
1370 {
1371         struct bonding *bond = bond_dev->priv;
1372
1373         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1374         bond_dev->header_ops            = slave_dev->header_ops;
1375
1376         bond_dev->type              = slave_dev->type;
1377         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1378         bond_dev->addr_len          = slave_dev->addr_len;
1379
1380         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1381                 slave_dev->addr_len);
1382         bond->setup_by_slave = 1;
1383 }
1384
1385 /* enslave device <slave> to bond device <master> */
1386 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1387 {
1388         struct bonding *bond = bond_dev->priv;
1389         struct slave *new_slave = NULL;
1390         struct dev_mc_list *dmi;
1391         struct sockaddr addr;
1392         int link_reporting;
1393         int old_features = bond_dev->features;
1394         int res = 0;
1395
1396         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397                 slave_dev->do_ioctl == NULL) {
1398                 printk(KERN_WARNING DRV_NAME
1399                        ": %s: Warning: no link monitoring support for %s\n",
1400                        bond_dev->name, slave_dev->name);
1401         }
1402
1403         /* bond must be initialized by bond_open() before enslaving */
1404         if (!(bond_dev->flags & IFF_UP)) {
1405                 printk(KERN_WARNING DRV_NAME
1406                         " %s: master_dev is not up in bond_enslave\n",
1407                         bond_dev->name);
1408         }
1409
1410         /* already enslaved */
1411         if (slave_dev->flags & IFF_SLAVE) {
1412                 dprintk("Error, Device was already enslaved\n");
1413                 return -EBUSY;
1414         }
1415
1416         /* vlan challenged mutual exclusion */
1417         /* no need to lock since we're protected by rtnl_lock */
1418         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420                 if (!list_empty(&bond->vlan_list)) {
1421                         printk(KERN_ERR DRV_NAME
1422                                ": %s: Error: cannot enslave VLAN "
1423                                "challenged slave %s on VLAN enabled "
1424                                "bond %s\n", bond_dev->name, slave_dev->name,
1425                                bond_dev->name);
1426                         return -EPERM;
1427                 } else {
1428                         printk(KERN_WARNING DRV_NAME
1429                                ": %s: Warning: enslaved VLAN challenged "
1430                                "slave %s. Adding VLANs will be blocked as "
1431                                "long as %s is part of bond %s\n",
1432                                bond_dev->name, slave_dev->name, slave_dev->name,
1433                                bond_dev->name);
1434                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1435                 }
1436         } else {
1437                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438                 if (bond->slave_cnt == 0) {
1439                         /* First slave, and it is not VLAN challenged,
1440                          * so remove the block of adding VLANs over the bond.
1441                          */
1442                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1443                 }
1444         }
1445
1446         /*
1447          * Old ifenslave binaries are no longer supported.  These can
1448          * be identified with moderate accurary by the state of the slave:
1449          * the current ifenslave will set the interface down prior to
1450          * enslaving it; the old ifenslave will not.
1451          */
1452         if ((slave_dev->flags & IFF_UP)) {
1453                 printk(KERN_ERR DRV_NAME ": %s is up. "
1454                        "This may be due to an out of date ifenslave.\n",
1455                        slave_dev->name);
1456                 res = -EPERM;
1457                 goto err_undo_flags;
1458         }
1459
1460         /* set bonding device ether type by slave - bonding netdevices are
1461          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462          * there is a need to override some of the type dependent attribs/funcs.
1463          *
1464          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1466          */
1467         if (bond->slave_cnt == 0) {
1468                 if (slave_dev->type != ARPHRD_ETHER)
1469                         bond_setup_by_slave(bond_dev, slave_dev);
1470         } else if (bond_dev->type != slave_dev->type) {
1471                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1472                         "from other slaves (%d), can not enslave it.\n",
1473                         slave_dev->name,
1474                         slave_dev->type, bond_dev->type);
1475                         res = -EINVAL;
1476                         goto err_undo_flags;
1477         }
1478
1479         if (slave_dev->set_mac_address == NULL) {
1480                 if (bond->slave_cnt == 0) {
1481                         printk(KERN_WARNING DRV_NAME
1482                                ": %s: Warning: The first slave device "
1483                                "specified does not support setting the MAC "
1484                                "address. Setting fail_over_mac to active.",
1485                                bond_dev->name);
1486                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1488                         printk(KERN_ERR DRV_NAME
1489                                 ": %s: Error: The slave device specified "
1490                                 "does not support setting the MAC address, "
1491                                 "but fail_over_mac is not set to active.\n"
1492                                 , bond_dev->name);
1493                         res = -EOPNOTSUPP;
1494                         goto err_undo_flags;
1495                 }
1496         }
1497
1498         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1499         if (!new_slave) {
1500                 res = -ENOMEM;
1501                 goto err_undo_flags;
1502         }
1503
1504         /* save slave's original flags before calling
1505          * netdev_set_master and dev_open
1506          */
1507         new_slave->original_flags = slave_dev->flags;
1508
1509         /*
1510          * Save slave's original ("permanent") mac address for modes
1511          * that need it, and for restoring it upon release, and then
1512          * set it to the master's address
1513          */
1514         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1515
1516         if (!bond->params.fail_over_mac) {
1517                 /*
1518                  * Set slave to master's mac address.  The application already
1519                  * set the master's mac address to that of the first slave
1520                  */
1521                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1522                 addr.sa_family = slave_dev->type;
1523                 res = dev_set_mac_address(slave_dev, &addr);
1524                 if (res) {
1525                         dprintk("Error %d calling set_mac_address\n", res);
1526                         goto err_free;
1527                 }
1528         }
1529
1530         res = netdev_set_master(slave_dev, bond_dev);
1531         if (res) {
1532                 dprintk("Error %d calling netdev_set_master\n", res);
1533                 goto err_restore_mac;
1534         }
1535         /* open the slave since the application closed it */
1536         res = dev_open(slave_dev);
1537         if (res) {
1538                 dprintk("Openning slave %s failed\n", slave_dev->name);
1539                 goto err_unset_master;
1540         }
1541
1542         new_slave->dev = slave_dev;
1543         slave_dev->priv_flags |= IFF_BONDING;
1544
1545         if ((bond->params.mode == BOND_MODE_TLB) ||
1546             (bond->params.mode == BOND_MODE_ALB)) {
1547                 /* bond_alb_init_slave() must be called before all other stages since
1548                  * it might fail and we do not want to have to undo everything
1549                  */
1550                 res = bond_alb_init_slave(bond, new_slave);
1551                 if (res) {
1552                         goto err_close;
1553                 }
1554         }
1555
1556         /* If the mode USES_PRIMARY, then the new slave gets the
1557          * master's promisc (and mc) settings only if it becomes the
1558          * curr_active_slave, and that is taken care of later when calling
1559          * bond_change_active()
1560          */
1561         if (!USES_PRIMARY(bond->params.mode)) {
1562                 /* set promiscuity level to new slave */
1563                 if (bond_dev->flags & IFF_PROMISC) {
1564                         res = dev_set_promiscuity(slave_dev, 1);
1565                         if (res)
1566                                 goto err_close;
1567                 }
1568
1569                 /* set allmulti level to new slave */
1570                 if (bond_dev->flags & IFF_ALLMULTI) {
1571                         res = dev_set_allmulti(slave_dev, 1);
1572                         if (res)
1573                                 goto err_close;
1574                 }
1575
1576                 netif_addr_lock_bh(bond_dev);
1577                 /* upload master's mc_list to new slave */
1578                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1579                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1580                 }
1581                 netif_addr_unlock_bh(bond_dev);
1582         }
1583
1584         if (bond->params.mode == BOND_MODE_8023AD) {
1585                 /* add lacpdu mc addr to mc list */
1586                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1587
1588                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1589         }
1590
1591         bond_add_vlans_on_slave(bond, slave_dev);
1592
1593         write_lock_bh(&bond->lock);
1594
1595         bond_attach_slave(bond, new_slave);
1596
1597         new_slave->delay = 0;
1598         new_slave->link_failure_count = 0;
1599
1600         bond_compute_features(bond);
1601
1602         write_unlock_bh(&bond->lock);
1603
1604         read_lock(&bond->lock);
1605
1606         new_slave->last_arp_rx = jiffies;
1607
1608         if (bond->params.miimon && !bond->params.use_carrier) {
1609                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1610
1611                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1612                         /*
1613                          * miimon is set but a bonded network driver
1614                          * does not support ETHTOOL/MII and
1615                          * arp_interval is not set.  Note: if
1616                          * use_carrier is enabled, we will never go
1617                          * here (because netif_carrier is always
1618                          * supported); thus, we don't need to change
1619                          * the messages for netif_carrier.
1620                          */
1621                         printk(KERN_WARNING DRV_NAME
1622                                ": %s: Warning: MII and ETHTOOL support not "
1623                                "available for interface %s, and "
1624                                "arp_interval/arp_ip_target module parameters "
1625                                "not specified, thus bonding will not detect "
1626                                "link failures! see bonding.txt for details.\n",
1627                                bond_dev->name, slave_dev->name);
1628                 } else if (link_reporting == -1) {
1629                         /* unable get link status using mii/ethtool */
1630                         printk(KERN_WARNING DRV_NAME
1631                                ": %s: Warning: can't get link status from "
1632                                "interface %s; the network driver associated "
1633                                "with this interface does not support MII or "
1634                                "ETHTOOL link status reporting, thus miimon "
1635                                "has no effect on this interface.\n",
1636                                bond_dev->name, slave_dev->name);
1637                 }
1638         }
1639
1640         /* check for initial state */
1641         if (!bond->params.miimon ||
1642             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1643                 if (bond->params.updelay) {
1644                         dprintk("Initial state of slave_dev is "
1645                                 "BOND_LINK_BACK\n");
1646                         new_slave->link  = BOND_LINK_BACK;
1647                         new_slave->delay = bond->params.updelay;
1648                 } else {
1649                         dprintk("Initial state of slave_dev is "
1650                                 "BOND_LINK_UP\n");
1651                         new_slave->link  = BOND_LINK_UP;
1652                 }
1653                 new_slave->jiffies = jiffies;
1654         } else {
1655                 dprintk("Initial state of slave_dev is "
1656                         "BOND_LINK_DOWN\n");
1657                 new_slave->link  = BOND_LINK_DOWN;
1658         }
1659
1660         if (bond_update_speed_duplex(new_slave) &&
1661             (new_slave->link != BOND_LINK_DOWN)) {
1662                 printk(KERN_WARNING DRV_NAME
1663                        ": %s: Warning: failed to get speed and duplex from %s, "
1664                        "assumed to be 100Mb/sec and Full.\n",
1665                        bond_dev->name, new_slave->dev->name);
1666
1667                 if (bond->params.mode == BOND_MODE_8023AD) {
1668                         printk(KERN_WARNING DRV_NAME
1669                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1670                                "support in base driver for proper aggregator "
1671                                "selection.\n", bond_dev->name);
1672                 }
1673         }
1674
1675         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1676                 /* if there is a primary slave, remember it */
1677                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1678                         bond->primary_slave = new_slave;
1679                 }
1680         }
1681
1682         write_lock_bh(&bond->curr_slave_lock);
1683
1684         switch (bond->params.mode) {
1685         case BOND_MODE_ACTIVEBACKUP:
1686                 bond_set_slave_inactive_flags(new_slave);
1687                 bond_select_active_slave(bond);
1688                 break;
1689         case BOND_MODE_8023AD:
1690                 /* in 802.3ad mode, the internal mechanism
1691                  * will activate the slaves in the selected
1692                  * aggregator
1693                  */
1694                 bond_set_slave_inactive_flags(new_slave);
1695                 /* if this is the first slave */
1696                 if (bond->slave_cnt == 1) {
1697                         SLAVE_AD_INFO(new_slave).id = 1;
1698                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1699                          * can be called only after the mac address of the bond is set
1700                          */
1701                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1702                                             bond->params.lacp_fast);
1703                 } else {
1704                         SLAVE_AD_INFO(new_slave).id =
1705                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1706                 }
1707
1708                 bond_3ad_bind_slave(new_slave);
1709                 break;
1710         case BOND_MODE_TLB:
1711         case BOND_MODE_ALB:
1712                 new_slave->state = BOND_STATE_ACTIVE;
1713                 bond_set_slave_inactive_flags(new_slave);
1714                 break;
1715         default:
1716                 dprintk("This slave is always active in trunk mode\n");
1717
1718                 /* always active in trunk mode */
1719                 new_slave->state = BOND_STATE_ACTIVE;
1720
1721                 /* In trunking mode there is little meaning to curr_active_slave
1722                  * anyway (it holds no special properties of the bond device),
1723                  * so we can change it without calling change_active_interface()
1724                  */
1725                 if (!bond->curr_active_slave) {
1726                         bond->curr_active_slave = new_slave;
1727                 }
1728                 break;
1729         } /* switch(bond_mode) */
1730
1731         write_unlock_bh(&bond->curr_slave_lock);
1732
1733         bond_set_carrier(bond);
1734
1735         read_unlock(&bond->lock);
1736
1737         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738         if (res)
1739                 goto err_close;
1740
1741         printk(KERN_INFO DRV_NAME
1742                ": %s: enslaving %s as a%s interface with a%s link.\n",
1743                bond_dev->name, slave_dev->name,
1744                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1745                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1746
1747         /* enslave is successful */
1748         return 0;
1749
1750 /* Undo stages on error */
1751 err_close:
1752         dev_close(slave_dev);
1753
1754 err_unset_master:
1755         netdev_set_master(slave_dev, NULL);
1756
1757 err_restore_mac:
1758         if (!bond->params.fail_over_mac) {
1759                 /* XXX TODO - fom follow mode needs to change master's
1760                  * MAC if this slave's MAC is in use by the bond, or at
1761                  * least print a warning.
1762                  */
1763                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1764                 addr.sa_family = slave_dev->type;
1765                 dev_set_mac_address(slave_dev, &addr);
1766         }
1767
1768 err_free:
1769         kfree(new_slave);
1770
1771 err_undo_flags:
1772         bond_dev->features = old_features;
1773
1774         return res;
1775 }
1776
1777 /*
1778  * Try to release the slave device <slave> from the bond device <master>
1779  * It is legal to access curr_active_slave without a lock because all the function
1780  * is write-locked.
1781  *
1782  * The rules for slave state should be:
1783  *   for Active/Backup:
1784  *     Active stays on all backups go down
1785  *   for Bonded connections:
1786  *     The first up interface should be left on and all others downed.
1787  */
1788 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1789 {
1790         struct bonding *bond = bond_dev->priv;
1791         struct slave *slave, *oldcurrent;
1792         struct sockaddr addr;
1793         int mac_addr_differ;
1794         DECLARE_MAC_BUF(mac);
1795
1796         /* slave is not a slave or master is not master of this slave */
1797         if (!(slave_dev->flags & IFF_SLAVE) ||
1798             (slave_dev->master != bond_dev)) {
1799                 printk(KERN_ERR DRV_NAME
1800                        ": %s: Error: cannot release %s.\n",
1801                        bond_dev->name, slave_dev->name);
1802                 return -EINVAL;
1803         }
1804
1805         write_lock_bh(&bond->lock);
1806
1807         slave = bond_get_slave_by_dev(bond, slave_dev);
1808         if (!slave) {
1809                 /* not a slave of this bond */
1810                 printk(KERN_INFO DRV_NAME
1811                        ": %s: %s not enslaved\n",
1812                        bond_dev->name, slave_dev->name);
1813                 write_unlock_bh(&bond->lock);
1814                 return -EINVAL;
1815         }
1816
1817         if (!bond->params.fail_over_mac) {
1818                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1819                                          ETH_ALEN);
1820                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1821                         printk(KERN_WARNING DRV_NAME
1822                                ": %s: Warning: the permanent HWaddr of %s - "
1823                                "%s - is still in use by %s. "
1824                                "Set the HWaddr of %s to a different address "
1825                                "to avoid conflicts.\n",
1826                                bond_dev->name, slave_dev->name,
1827                                print_mac(mac, slave->perm_hwaddr),
1828                                bond_dev->name, slave_dev->name);
1829         }
1830
1831         /* Inform AD package of unbinding of slave. */
1832         if (bond->params.mode == BOND_MODE_8023AD) {
1833                 /* must be called before the slave is
1834                  * detached from the list
1835                  */
1836                 bond_3ad_unbind_slave(slave);
1837         }
1838
1839         printk(KERN_INFO DRV_NAME
1840                ": %s: releasing %s interface %s\n",
1841                bond_dev->name,
1842                (slave->state == BOND_STATE_ACTIVE)
1843                ? "active" : "backup",
1844                slave_dev->name);
1845
1846         oldcurrent = bond->curr_active_slave;
1847
1848         bond->current_arp_slave = NULL;
1849
1850         /* release the slave from its bond */
1851         bond_detach_slave(bond, slave);
1852
1853         bond_compute_features(bond);
1854
1855         if (bond->primary_slave == slave) {
1856                 bond->primary_slave = NULL;
1857         }
1858
1859         if (oldcurrent == slave) {
1860                 bond_change_active_slave(bond, NULL);
1861         }
1862
1863         if ((bond->params.mode == BOND_MODE_TLB) ||
1864             (bond->params.mode == BOND_MODE_ALB)) {
1865                 /* Must be called only after the slave has been
1866                  * detached from the list and the curr_active_slave
1867                  * has been cleared (if our_slave == old_current),
1868                  * but before a new active slave is selected.
1869                  */
1870                 write_unlock_bh(&bond->lock);
1871                 bond_alb_deinit_slave(bond, slave);
1872                 write_lock_bh(&bond->lock);
1873         }
1874
1875         if (oldcurrent == slave) {
1876                 /*
1877                  * Note that we hold RTNL over this sequence, so there
1878                  * is no concern that another slave add/remove event
1879                  * will interfere.
1880                  */
1881                 write_unlock_bh(&bond->lock);
1882                 read_lock(&bond->lock);
1883                 write_lock_bh(&bond->curr_slave_lock);
1884
1885                 bond_select_active_slave(bond);
1886
1887                 write_unlock_bh(&bond->curr_slave_lock);
1888                 read_unlock(&bond->lock);
1889                 write_lock_bh(&bond->lock);
1890         }
1891
1892         if (bond->slave_cnt == 0) {
1893                 bond_set_carrier(bond);
1894
1895                 /* if the last slave was removed, zero the mac address
1896                  * of the master so it will be set by the application
1897                  * to the mac address of the first slave
1898                  */
1899                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901                 if (list_empty(&bond->vlan_list)) {
1902                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903                 } else {
1904                         printk(KERN_WARNING DRV_NAME
1905                                ": %s: Warning: clearing HW address of %s while it "
1906                                "still has VLANs.\n",
1907                                bond_dev->name, bond_dev->name);
1908                         printk(KERN_WARNING DRV_NAME
1909                                ": %s: When re-adding slaves, make sure the bond's "
1910                                "HW address matches its VLANs'.\n",
1911                                bond_dev->name);
1912                 }
1913         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914                    !bond_has_challenged_slaves(bond)) {
1915                 printk(KERN_INFO DRV_NAME
1916                        ": %s: last VLAN challenged slave %s "
1917                        "left bond %s. VLAN blocking is removed\n",
1918                        bond_dev->name, slave_dev->name, bond_dev->name);
1919                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920         }
1921
1922         write_unlock_bh(&bond->lock);
1923
1924         /* must do this from outside any spinlocks */
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927         bond_del_vlans_from_slave(bond, slave_dev);
1928
1929         /* If the mode USES_PRIMARY, then we should only remove its
1930          * promisc and mc settings if it was the curr_active_slave, but that was
1931          * already taken care of above when we detached the slave
1932          */
1933         if (!USES_PRIMARY(bond->params.mode)) {
1934                 /* unset promiscuity level from slave */
1935                 if (bond_dev->flags & IFF_PROMISC) {
1936                         dev_set_promiscuity(slave_dev, -1);
1937                 }
1938
1939                 /* unset allmulti level from slave */
1940                 if (bond_dev->flags & IFF_ALLMULTI) {
1941                         dev_set_allmulti(slave_dev, -1);
1942                 }
1943
1944                 /* flush master's mc_list from slave */
1945                 netif_addr_lock_bh(bond_dev);
1946                 bond_mc_list_flush(bond_dev, slave_dev);
1947                 netif_addr_unlock_bh(bond_dev);
1948         }
1949
1950         netdev_set_master(slave_dev, NULL);
1951
1952         /* close slave before restoring its mac address */
1953         dev_close(slave_dev);
1954
1955         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1956                 /* restore original ("permanent") mac address */
1957                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958                 addr.sa_family = slave_dev->type;
1959                 dev_set_mac_address(slave_dev, &addr);
1960         }
1961
1962         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1963                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1964                                    IFF_SLAVE_NEEDARP);
1965
1966         kfree(slave);
1967
1968         return 0;  /* deletion OK */
1969 }
1970
1971 /*
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1974 */
1975 void bond_destroy(struct bonding *bond)
1976 {
1977         bond_deinit(bond->dev);
1978         bond_destroy_sysfs_entry(bond);
1979         unregister_netdevice(bond->dev);
1980 }
1981
1982 /*
1983 * First release a slave and than destroy the bond if no more slaves iare left.
1984 * Must be under rtnl_lock when this function is called.
1985 */
1986 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1987 {
1988         struct bonding *bond = bond_dev->priv;
1989         int ret;
1990
1991         ret = bond_release(bond_dev, slave_dev);
1992         if ((ret == 0) && (bond->slave_cnt == 0)) {
1993                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1994                        bond_dev->name, bond_dev->name);
1995                 bond_destroy(bond);
1996         }
1997         return ret;
1998 }
1999
2000 /*
2001  * This function releases all slaves.
2002  */
2003 static int bond_release_all(struct net_device *bond_dev)
2004 {
2005         struct bonding *bond = bond_dev->priv;
2006         struct slave *slave;
2007         struct net_device *slave_dev;
2008         struct sockaddr addr;
2009
2010         write_lock_bh(&bond->lock);
2011
2012         netif_carrier_off(bond_dev);
2013
2014         if (bond->slave_cnt == 0) {
2015                 goto out;
2016         }
2017
2018         bond->current_arp_slave = NULL;
2019         bond->primary_slave = NULL;
2020         bond_change_active_slave(bond, NULL);
2021
2022         while ((slave = bond->first_slave) != NULL) {
2023                 /* Inform AD package of unbinding of slave
2024                  * before slave is detached from the list.
2025                  */
2026                 if (bond->params.mode == BOND_MODE_8023AD) {
2027                         bond_3ad_unbind_slave(slave);
2028                 }
2029
2030                 slave_dev = slave->dev;
2031                 bond_detach_slave(bond, slave);
2032
2033                 /* now that the slave is detached, unlock and perform
2034                  * all the undo steps that should not be called from
2035                  * within a lock.
2036                  */
2037                 write_unlock_bh(&bond->lock);
2038
2039                 if ((bond->params.mode == BOND_MODE_TLB) ||
2040                     (bond->params.mode == BOND_MODE_ALB)) {
2041                         /* must be called only after the slave
2042                          * has been detached from the list
2043                          */
2044                         bond_alb_deinit_slave(bond, slave);
2045                 }
2046
2047                 bond_compute_features(bond);
2048
2049                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2050                 bond_del_vlans_from_slave(bond, slave_dev);
2051
2052                 /* If the mode USES_PRIMARY, then we should only remove its
2053                  * promisc and mc settings if it was the curr_active_slave, but that was
2054                  * already taken care of above when we detached the slave
2055                  */
2056                 if (!USES_PRIMARY(bond->params.mode)) {
2057                         /* unset promiscuity level from slave */
2058                         if (bond_dev->flags & IFF_PROMISC) {
2059                                 dev_set_promiscuity(slave_dev, -1);
2060                         }
2061
2062                         /* unset allmulti level from slave */
2063                         if (bond_dev->flags & IFF_ALLMULTI) {
2064                                 dev_set_allmulti(slave_dev, -1);
2065                         }
2066
2067                         /* flush master's mc_list from slave */
2068                         netif_addr_lock_bh(bond_dev);
2069                         bond_mc_list_flush(bond_dev, slave_dev);
2070                         netif_addr_unlock_bh(bond_dev);
2071                 }
2072
2073                 netdev_set_master(slave_dev, NULL);
2074
2075                 /* close slave before restoring its mac address */
2076                 dev_close(slave_dev);
2077
2078                 if (!bond->params.fail_over_mac) {
2079                         /* restore original ("permanent") mac address*/
2080                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2081                         addr.sa_family = slave_dev->type;
2082                         dev_set_mac_address(slave_dev, &addr);
2083                 }
2084
2085                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2086                                            IFF_SLAVE_INACTIVE);
2087
2088                 kfree(slave);
2089
2090                 /* re-acquire the lock before getting the next slave */
2091                 write_lock_bh(&bond->lock);
2092         }
2093
2094         /* zero the mac address of the master so it will be
2095          * set by the application to the mac address of the
2096          * first slave
2097          */
2098         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2099
2100         if (list_empty(&bond->vlan_list)) {
2101                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2102         } else {
2103                 printk(KERN_WARNING DRV_NAME
2104                        ": %s: Warning: clearing HW address of %s while it "
2105                        "still has VLANs.\n",
2106                        bond_dev->name, bond_dev->name);
2107                 printk(KERN_WARNING DRV_NAME
2108                        ": %s: When re-adding slaves, make sure the bond's "
2109                        "HW address matches its VLANs'.\n",
2110                        bond_dev->name);
2111         }
2112
2113         printk(KERN_INFO DRV_NAME
2114                ": %s: released all slaves\n",
2115                bond_dev->name);
2116
2117 out:
2118         write_unlock_bh(&bond->lock);
2119
2120         return 0;
2121 }
2122
2123 /*
2124  * This function changes the active slave to slave <slave_dev>.
2125  * It returns -EINVAL in the following cases.
2126  *  - <slave_dev> is not found in the list.
2127  *  - There is not active slave now.
2128  *  - <slave_dev> is already active.
2129  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2130  *  - <slave_dev> is not running.
2131  * In these cases, this fuction does nothing.
2132  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2133  */
2134 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2135 {
2136         struct bonding *bond = bond_dev->priv;
2137         struct slave *old_active = NULL;
2138         struct slave *new_active = NULL;
2139         int res = 0;
2140
2141         if (!USES_PRIMARY(bond->params.mode)) {
2142                 return -EINVAL;
2143         }
2144
2145         /* Verify that master_dev is indeed the master of slave_dev */
2146         if (!(slave_dev->flags & IFF_SLAVE) ||
2147             (slave_dev->master != bond_dev)) {
2148                 return -EINVAL;
2149         }
2150
2151         read_lock(&bond->lock);
2152
2153         read_lock(&bond->curr_slave_lock);
2154         old_active = bond->curr_active_slave;
2155         read_unlock(&bond->curr_slave_lock);
2156
2157         new_active = bond_get_slave_by_dev(bond, slave_dev);
2158
2159         /*
2160          * Changing to the current active: do nothing; return success.
2161          */
2162         if (new_active && (new_active == old_active)) {
2163                 read_unlock(&bond->lock);
2164                 return 0;
2165         }
2166
2167         if ((new_active) &&
2168             (old_active) &&
2169             (new_active->link == BOND_LINK_UP) &&
2170             IS_UP(new_active->dev)) {
2171                 write_lock_bh(&bond->curr_slave_lock);
2172                 bond_change_active_slave(bond, new_active);
2173                 write_unlock_bh(&bond->curr_slave_lock);
2174         } else {
2175                 res = -EINVAL;
2176         }
2177
2178         read_unlock(&bond->lock);
2179
2180         return res;
2181 }
2182
2183 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2184 {
2185         struct bonding *bond = bond_dev->priv;
2186
2187         info->bond_mode = bond->params.mode;
2188         info->miimon = bond->params.miimon;
2189
2190         read_lock(&bond->lock);
2191         info->num_slaves = bond->slave_cnt;
2192         read_unlock(&bond->lock);
2193
2194         return 0;
2195 }
2196
2197 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2198 {
2199         struct bonding *bond = bond_dev->priv;
2200         struct slave *slave;
2201         int i, found = 0;
2202
2203         if (info->slave_id < 0) {
2204                 return -ENODEV;
2205         }
2206
2207         read_lock(&bond->lock);
2208
2209         bond_for_each_slave(bond, slave, i) {
2210                 if (i == (int)info->slave_id) {
2211                         found = 1;
2212                         break;
2213                 }
2214         }
2215
2216         read_unlock(&bond->lock);
2217
2218         if (found) {
2219                 strcpy(info->slave_name, slave->dev->name);
2220                 info->link = slave->link;
2221                 info->state = slave->state;
2222                 info->link_failure_count = slave->link_failure_count;
2223         } else {
2224                 return -ENODEV;
2225         }
2226
2227         return 0;
2228 }
2229
2230 /*-------------------------------- Monitoring -------------------------------*/
2231
2232
2233 static int bond_miimon_inspect(struct bonding *bond)
2234 {
2235         struct slave *slave;
2236         int i, link_state, commit = 0;
2237
2238         bond_for_each_slave(bond, slave, i) {
2239                 slave->new_link = BOND_LINK_NOCHANGE;
2240
2241                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2242
2243                 switch (slave->link) {
2244                 case BOND_LINK_UP:
2245                         if (link_state)
2246                                 continue;
2247
2248                         slave->link = BOND_LINK_FAIL;
2249                         slave->delay = bond->params.downdelay;
2250                         if (slave->delay) {
2251                                 printk(KERN_INFO DRV_NAME
2252                                        ": %s: link status down for %s"
2253                                        "interface %s, disabling it in %d ms.\n",
2254                                        bond->dev->name,
2255                                        (bond->params.mode ==
2256                                         BOND_MODE_ACTIVEBACKUP) ?
2257                                        ((slave->state == BOND_STATE_ACTIVE) ?
2258                                         "active " : "backup ") : "",
2259                                        slave->dev->name,
2260                                        bond->params.downdelay * bond->params.miimon);
2261                         }
2262                         /*FALLTHRU*/
2263                 case BOND_LINK_FAIL:
2264                         if (link_state) {
2265                                 /*
2266                                  * recovered before downdelay expired
2267                                  */
2268                                 slave->link = BOND_LINK_UP;
2269                                 slave->jiffies = jiffies;
2270                                 printk(KERN_INFO DRV_NAME
2271                                        ": %s: link status up again after %d "
2272                                        "ms for interface %s.\n",
2273                                        bond->dev->name,
2274                                        (bond->params.downdelay - slave->delay) *
2275                                        bond->params.miimon,
2276                                        slave->dev->name);
2277                                 continue;
2278                         }
2279
2280                         if (slave->delay <= 0) {
2281                                 slave->new_link = BOND_LINK_DOWN;
2282                                 commit++;
2283                                 continue;
2284                         }
2285
2286                         slave->delay--;
2287                         break;
2288
2289                 case BOND_LINK_DOWN:
2290                         if (!link_state)
2291                                 continue;
2292
2293                         slave->link = BOND_LINK_BACK;
2294                         slave->delay = bond->params.updelay;
2295
2296                         if (slave->delay) {
2297                                 printk(KERN_INFO DRV_NAME
2298                                        ": %s: link status up for "
2299                                        "interface %s, enabling it in %d ms.\n",
2300                                        bond->dev->name, slave->dev->name,
2301                                        bond->params.updelay *
2302                                        bond->params.miimon);
2303                         }
2304                         /*FALLTHRU*/
2305                 case BOND_LINK_BACK:
2306                         if (!link_state) {
2307                                 slave->link = BOND_LINK_DOWN;
2308                                 printk(KERN_INFO DRV_NAME
2309                                        ": %s: link status down again after %d "
2310                                        "ms for interface %s.\n",
2311                                        bond->dev->name,
2312                                        (bond->params.updelay - slave->delay) *
2313                                        bond->params.miimon,
2314                                        slave->dev->name);
2315
2316                                 continue;
2317                         }
2318
2319                         if (slave->delay <= 0) {
2320                                 slave->new_link = BOND_LINK_UP;
2321                                 commit++;
2322                                 continue;
2323                         }
2324
2325                         slave->delay--;
2326                         break;
2327                 }
2328         }
2329
2330         return commit;
2331 }
2332
2333 static void bond_miimon_commit(struct bonding *bond)
2334 {
2335         struct slave *slave;
2336         int i;
2337
2338         bond_for_each_slave(bond, slave, i) {
2339                 switch (slave->new_link) {
2340                 case BOND_LINK_NOCHANGE:
2341                         continue;
2342
2343                 case BOND_LINK_UP:
2344                         slave->link = BOND_LINK_UP;
2345                         slave->jiffies = jiffies;
2346
2347                         if (bond->params.mode == BOND_MODE_8023AD) {
2348                                 /* prevent it from being the active one */
2349                                 slave->state = BOND_STATE_BACKUP;
2350                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2351                                 /* make it immediately active */
2352                                 slave->state = BOND_STATE_ACTIVE;
2353                         } else if (slave != bond->primary_slave) {
2354                                 /* prevent it from being the active one */
2355                                 slave->state = BOND_STATE_BACKUP;
2356                         }
2357
2358                         printk(KERN_INFO DRV_NAME
2359                                ": %s: link status definitely "
2360                                "up for interface %s.\n",
2361                                bond->dev->name, slave->dev->name);
2362
2363                         /* notify ad that the link status has changed */
2364                         if (bond->params.mode == BOND_MODE_8023AD)
2365                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2366
2367                         if ((bond->params.mode == BOND_MODE_TLB) ||
2368                             (bond->params.mode == BOND_MODE_ALB))
2369                                 bond_alb_handle_link_change(bond, slave,
2370                                                             BOND_LINK_UP);
2371
2372                         if (!bond->curr_active_slave ||
2373                             (slave == bond->primary_slave))
2374                                 goto do_failover;
2375
2376                         continue;
2377
2378                 case BOND_LINK_DOWN:
2379                         slave->link = BOND_LINK_DOWN;
2380
2381                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2382                             bond->params.mode == BOND_MODE_8023AD)
2383                                 bond_set_slave_inactive_flags(slave);
2384
2385                         printk(KERN_INFO DRV_NAME
2386                                ": %s: link status definitely down for "
2387                                "interface %s, disabling it\n",
2388                                bond->dev->name, slave->dev->name);
2389
2390                         if (bond->params.mode == BOND_MODE_8023AD)
2391                                 bond_3ad_handle_link_change(slave,
2392                                                             BOND_LINK_DOWN);
2393
2394                         if (bond->params.mode == BOND_MODE_TLB ||
2395                             bond->params.mode == BOND_MODE_ALB)
2396                                 bond_alb_handle_link_change(bond, slave,
2397                                                             BOND_LINK_DOWN);
2398
2399                         if (slave == bond->curr_active_slave)
2400                                 goto do_failover;
2401
2402                         continue;
2403
2404                 default:
2405                         printk(KERN_ERR DRV_NAME
2406                                ": %s: invalid new link %d on slave %s\n",
2407                                bond->dev->name, slave->new_link,
2408                                slave->dev->name);
2409                         slave->new_link = BOND_LINK_NOCHANGE;
2410
2411                         continue;
2412                 }
2413
2414 do_failover:
2415                 ASSERT_RTNL();
2416                 write_lock_bh(&bond->curr_slave_lock);
2417                 bond_select_active_slave(bond);
2418                 write_unlock_bh(&bond->curr_slave_lock);
2419         }
2420
2421         bond_set_carrier(bond);
2422 }
2423
2424 /*
2425  * bond_mii_monitor
2426  *
2427  * Really a wrapper that splits the mii monitor into two phases: an
2428  * inspection, then (if inspection indicates something needs to be done)
2429  * an acquisition of appropriate locks followed by a commit phase to
2430  * implement whatever link state changes are indicated.
2431  */
2432 void bond_mii_monitor(struct work_struct *work)
2433 {
2434         struct bonding *bond = container_of(work, struct bonding,
2435                                             mii_work.work);
2436
2437         read_lock(&bond->lock);
2438         if (bond->kill_timers)
2439                 goto out;
2440
2441         if (bond->slave_cnt == 0)
2442                 goto re_arm;
2443
2444         if (bond->send_grat_arp) {
2445                 read_lock(&bond->curr_slave_lock);
2446                 bond_send_gratuitous_arp(bond);
2447                 read_unlock(&bond->curr_slave_lock);
2448         }
2449
2450         if (bond_miimon_inspect(bond)) {
2451                 read_unlock(&bond->lock);
2452                 rtnl_lock();
2453                 read_lock(&bond->lock);
2454
2455                 bond_miimon_commit(bond);
2456
2457                 read_unlock(&bond->lock);
2458                 rtnl_unlock();  /* might sleep, hold no other locks */
2459                 read_lock(&bond->lock);
2460         }
2461
2462 re_arm:
2463         if (bond->params.miimon)
2464                 queue_delayed_work(bond->wq, &bond->mii_work,
2465                                    msecs_to_jiffies(bond->params.miimon));
2466 out:
2467         read_unlock(&bond->lock);
2468 }
2469
2470 static __be32 bond_glean_dev_ip(struct net_device *dev)
2471 {
2472         struct in_device *idev;
2473         struct in_ifaddr *ifa;
2474         __be32 addr = 0;
2475
2476         if (!dev)
2477                 return 0;
2478
2479         rcu_read_lock();
2480         idev = __in_dev_get_rcu(dev);
2481         if (!idev)
2482                 goto out;
2483
2484         ifa = idev->ifa_list;
2485         if (!ifa)
2486                 goto out;
2487
2488         addr = ifa->ifa_local;
2489 out:
2490         rcu_read_unlock();
2491         return addr;
2492 }
2493
2494 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2495 {
2496         struct vlan_entry *vlan;
2497
2498         if (ip == bond->master_ip)
2499                 return 1;
2500
2501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2502                 if (ip == vlan->vlan_ip)
2503                         return 1;
2504         }
2505
2506         return 0;
2507 }
2508
2509 /*
2510  * We go to the (large) trouble of VLAN tagging ARP frames because
2511  * switches in VLAN mode (especially if ports are configured as
2512  * "native" to a VLAN) might not pass non-tagged frames.
2513  */
2514 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2515 {
2516         struct sk_buff *skb;
2517
2518         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2519                slave_dev->name, dest_ip, src_ip, vlan_id);
2520
2521         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2522                          NULL, slave_dev->dev_addr, NULL);
2523
2524         if (!skb) {
2525                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2526                 return;
2527         }
2528         if (vlan_id) {
2529                 skb = vlan_put_tag(skb, vlan_id);
2530                 if (!skb) {
2531                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2532                         return;
2533                 }
2534         }
2535         arp_xmit(skb);
2536 }
2537
2538
2539 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2540 {
2541         int i, vlan_id, rv;
2542         __be32 *targets = bond->params.arp_targets;
2543         struct vlan_entry *vlan;
2544         struct net_device *vlan_dev;
2545         struct flowi fl;
2546         struct rtable *rt;
2547
2548         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2549                 if (!targets[i])
2550                         continue;
2551                 dprintk("basa: target %x\n", targets[i]);
2552                 if (list_empty(&bond->vlan_list)) {
2553                         dprintk("basa: empty vlan: arp_send\n");
2554                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2555                                       bond->master_ip, 0);
2556                         continue;
2557                 }
2558
2559                 /*
2560                  * If VLANs are configured, we do a route lookup to
2561                  * determine which VLAN interface would be used, so we
2562                  * can tag the ARP with the proper VLAN tag.
2563                  */
2564                 memset(&fl, 0, sizeof(fl));
2565                 fl.fl4_dst = targets[i];
2566                 fl.fl4_tos = RTO_ONLINK;
2567
2568                 rv = ip_route_output_key(&init_net, &rt, &fl);
2569                 if (rv) {
2570                         if (net_ratelimit()) {
2571                                 printk(KERN_WARNING DRV_NAME
2572                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2573                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2574                         }
2575                         continue;
2576                 }
2577
2578                 /*
2579                  * This target is not on a VLAN
2580                  */
2581                 if (rt->u.dst.dev == bond->dev) {
2582                         ip_rt_put(rt);
2583                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2584                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2585                                       bond->master_ip, 0);
2586                         continue;
2587                 }
2588
2589                 vlan_id = 0;
2590                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2591                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2592                         if (vlan_dev == rt->u.dst.dev) {
2593                                 vlan_id = vlan->vlan_id;
2594                                 dprintk("basa: vlan match on %s %d\n",
2595                                        vlan_dev->name, vlan_id);
2596                                 break;
2597                         }
2598                 }
2599
2600                 if (vlan_id) {
2601                         ip_rt_put(rt);
2602                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2603                                       vlan->vlan_ip, vlan_id);
2604                         continue;
2605                 }
2606
2607                 if (net_ratelimit()) {
2608                         printk(KERN_WARNING DRV_NAME
2609                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2610                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2611                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2612                 }
2613                 ip_rt_put(rt);
2614         }
2615 }
2616
2617 /*
2618  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2619  * for each VLAN above us.
2620  *
2621  * Caller must hold curr_slave_lock for read or better
2622  */
2623 static void bond_send_gratuitous_arp(struct bonding *bond)
2624 {
2625         struct slave *slave = bond->curr_active_slave;
2626         struct vlan_entry *vlan;
2627         struct net_device *vlan_dev;
2628
2629         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2630                                 slave ? slave->dev->name : "NULL");
2631
2632         if (!slave || !bond->send_grat_arp ||
2633             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2634                 return;
2635
2636         bond->send_grat_arp--;
2637
2638         if (bond->master_ip) {
2639                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2640                                 bond->master_ip, 0);
2641         }
2642
2643         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2644                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2645                 if (vlan->vlan_ip) {
2646                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2647                                       vlan->vlan_ip, vlan->vlan_id);
2648                 }
2649         }
2650 }
2651
2652 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2653 {
2654         int i;
2655         __be32 *targets = bond->params.arp_targets;
2656
2657         targets = bond->params.arp_targets;
2658         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2659                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2660                         "%u.%u.%u.%u bhti(tip) %d\n",
2661                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2662                        bond_has_this_ip(bond, tip));
2663                 if (sip == targets[i]) {
2664                         if (bond_has_this_ip(bond, tip))
2665                                 slave->last_arp_rx = jiffies;
2666                         return;
2667                 }
2668         }
2669 }
2670
2671 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2672 {
2673         struct arphdr *arp;
2674         struct slave *slave;
2675         struct bonding *bond;
2676         unsigned char *arp_ptr;
2677         __be32 sip, tip;
2678
2679         if (dev_net(dev) != &init_net)
2680                 goto out;
2681
2682         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2683                 goto out;
2684
2685         bond = dev->priv;
2686         read_lock(&bond->lock);
2687
2688         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2689                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2690                 orig_dev ? orig_dev->name : "NULL");
2691
2692         slave = bond_get_slave_by_dev(bond, orig_dev);
2693         if (!slave || !slave_do_arp_validate(bond, slave))
2694                 goto out_unlock;
2695
2696         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2697                 goto out_unlock;
2698
2699         arp = arp_hdr(skb);
2700         if (arp->ar_hln != dev->addr_len ||
2701             skb->pkt_type == PACKET_OTHERHOST ||
2702             skb->pkt_type == PACKET_LOOPBACK ||
2703             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2704             arp->ar_pro != htons(ETH_P_IP) ||
2705             arp->ar_pln != 4)
2706                 goto out_unlock;
2707
2708         arp_ptr = (unsigned char *)(arp + 1);
2709         arp_ptr += dev->addr_len;
2710         memcpy(&sip, arp_ptr, 4);
2711         arp_ptr += 4 + dev->addr_len;
2712         memcpy(&tip, arp_ptr, 4);
2713
2714         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2715                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2716                 slave->state, bond->params.arp_validate,
2717                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2718
2719         /*
2720          * Backup slaves won't see the ARP reply, but do come through
2721          * here for each ARP probe (so we swap the sip/tip to validate
2722          * the probe).  In a "redundant switch, common router" type of
2723          * configuration, the ARP probe will (hopefully) travel from
2724          * the active, through one switch, the router, then the other
2725          * switch before reaching the backup.
2726          */
2727         if (slave->state == BOND_STATE_ACTIVE)
2728                 bond_validate_arp(bond, slave, sip, tip);
2729         else
2730                 bond_validate_arp(bond, slave, tip, sip);
2731
2732 out_unlock:
2733         read_unlock(&bond->lock);
2734 out:
2735         dev_kfree_skb(skb);
2736         return NET_RX_SUCCESS;
2737 }
2738
2739 /*
2740  * this function is called regularly to monitor each slave's link
2741  * ensuring that traffic is being sent and received when arp monitoring
2742  * is used in load-balancing mode. if the adapter has been dormant, then an
2743  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2744  * arp monitoring in active backup mode.
2745  */
2746 void bond_loadbalance_arp_mon(struct work_struct *work)
2747 {
2748         struct bonding *bond = container_of(work, struct bonding,
2749                                             arp_work.work);
2750         struct slave *slave, *oldcurrent;
2751         int do_failover = 0;
2752         int delta_in_ticks;
2753         int i;
2754
2755         read_lock(&bond->lock);
2756
2757         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2758
2759         if (bond->kill_timers) {
2760                 goto out;
2761         }
2762
2763         if (bond->slave_cnt == 0) {
2764                 goto re_arm;
2765         }
2766
2767         read_lock(&bond->curr_slave_lock);
2768         oldcurrent = bond->curr_active_slave;
2769         read_unlock(&bond->curr_slave_lock);
2770
2771         /* see if any of the previous devices are up now (i.e. they have
2772          * xmt and rcv traffic). the curr_active_slave does not come into
2773          * the picture unless it is null. also, slave->jiffies is not needed
2774          * here because we send an arp on each slave and give a slave as
2775          * long as it needs to get the tx/rx within the delta.
2776          * TODO: what about up/down delay in arp mode? it wasn't here before
2777          *       so it can wait
2778          */
2779         bond_for_each_slave(bond, slave, i) {
2780                 if (slave->link != BOND_LINK_UP) {
2781                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2782                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2783
2784                                 slave->link  = BOND_LINK_UP;
2785                                 slave->state = BOND_STATE_ACTIVE;
2786
2787                                 /* primary_slave has no meaning in round-robin
2788                                  * mode. the window of a slave being up and
2789                                  * curr_active_slave being null after enslaving
2790                                  * is closed.
2791                                  */
2792                                 if (!oldcurrent) {
2793                                         printk(KERN_INFO DRV_NAME
2794                                                ": %s: link status definitely "
2795                                                "up for interface %s, ",
2796                                                bond->dev->name,
2797                                                slave->dev->name);
2798                                         do_failover = 1;
2799                                 } else {
2800                                         printk(KERN_INFO DRV_NAME
2801                                                ": %s: interface %s is now up\n",
2802                                                bond->dev->name,
2803                                                slave->dev->name);
2804                                 }
2805                         }
2806                 } else {
2807                         /* slave->link == BOND_LINK_UP */
2808
2809                         /* not all switches will respond to an arp request
2810                          * when the source ip is 0, so don't take the link down
2811                          * if we don't know our ip yet
2812                          */
2813                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2814                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2815
2816                                 slave->link  = BOND_LINK_DOWN;
2817                                 slave->state = BOND_STATE_BACKUP;
2818
2819                                 if (slave->link_failure_count < UINT_MAX) {
2820                                         slave->link_failure_count++;
2821                                 }
2822
2823                                 printk(KERN_INFO DRV_NAME
2824                                        ": %s: interface %s is now down.\n",
2825                                        bond->dev->name,
2826                                        slave->dev->name);
2827
2828                                 if (slave == oldcurrent) {
2829                                         do_failover = 1;
2830                                 }
2831                         }
2832                 }
2833
2834                 /* note: if switch is in round-robin mode, all links
2835                  * must tx arp to ensure all links rx an arp - otherwise
2836                  * links may oscillate or not come up at all; if switch is
2837                  * in something like xor mode, there is nothing we can
2838                  * do - all replies will be rx'ed on same link causing slaves
2839                  * to be unstable during low/no traffic periods
2840                  */
2841                 if (IS_UP(slave->dev)) {
2842                         bond_arp_send_all(bond, slave);
2843                 }
2844         }
2845
2846         if (do_failover) {
2847                 write_lock_bh(&bond->curr_slave_lock);
2848
2849                 bond_select_active_slave(bond);
2850
2851                 write_unlock_bh(&bond->curr_slave_lock);
2852         }
2853
2854 re_arm:
2855         if (bond->params.arp_interval)
2856                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2857 out:
2858         read_unlock(&bond->lock);
2859 }
2860
2861 /*
2862  * Called to inspect slaves for active-backup mode ARP monitor link state
2863  * changes.  Sets new_link in slaves to specify what action should take
2864  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2865  * to link states must be committed.
2866  *
2867  * Called with bond->lock held for read.
2868  */
2869 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2870 {
2871         struct slave *slave;
2872         int i, commit = 0;
2873
2874         bond_for_each_slave(bond, slave, i) {
2875                 slave->new_link = BOND_LINK_NOCHANGE;
2876
2877                 if (slave->link != BOND_LINK_UP) {
2878                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2879                                            delta_in_ticks)) {
2880                                 slave->new_link = BOND_LINK_UP;
2881                                 commit++;
2882                         }
2883
2884                         continue;
2885                 }
2886
2887                 /*
2888                  * Give slaves 2*delta after being enslaved or made
2889                  * active.  This avoids bouncing, as the last receive
2890                  * times need a full ARP monitor cycle to be updated.
2891                  */
2892                 if (!time_after_eq(jiffies, slave->jiffies +
2893                                    2 * delta_in_ticks))
2894                         continue;
2895
2896                 /*
2897                  * Backup slave is down if:
2898                  * - No current_arp_slave AND
2899                  * - more than 3*delta since last receive AND
2900                  * - the bond has an IP address
2901                  *
2902                  * Note: a non-null current_arp_slave indicates
2903                  * the curr_active_slave went down and we are
2904                  * searching for a new one; under this condition
2905                  * we only take the curr_active_slave down - this
2906                  * gives each slave a chance to tx/rx traffic
2907                  * before being taken out
2908                  */
2909                 if (slave->state == BOND_STATE_BACKUP &&
2910                     !bond->current_arp_slave &&
2911                     time_after(jiffies, slave_last_rx(bond, slave) +
2912                                3 * delta_in_ticks)) {
2913                         slave->new_link = BOND_LINK_DOWN;
2914                         commit++;
2915                 }
2916
2917                 /*
2918                  * Active slave is down if:
2919                  * - more than 2*delta since transmitting OR
2920                  * - (more than 2*delta since receive AND
2921                  *    the bond has an IP address)
2922                  */
2923                 if ((slave->state == BOND_STATE_ACTIVE) &&
2924                     (time_after_eq(jiffies, slave->dev->trans_start +
2925                                     2 * delta_in_ticks) ||
2926                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2927                                      + 2 * delta_in_ticks)))) {
2928                         slave->new_link = BOND_LINK_DOWN;
2929                         commit++;
2930                 }
2931         }
2932
2933         read_lock(&bond->curr_slave_lock);
2934
2935         /*
2936          * Trigger a commit if the primary option setting has changed.
2937          */
2938         if (bond->primary_slave &&
2939             (bond->primary_slave != bond->curr_active_slave) &&
2940             (bond->primary_slave->link == BOND_LINK_UP))
2941                 commit++;
2942
2943         read_unlock(&bond->curr_slave_lock);
2944
2945         return commit;
2946 }
2947
2948 /*
2949  * Called to commit link state changes noted by inspection step of
2950  * active-backup mode ARP monitor.
2951  *
2952  * Called with RTNL and bond->lock for read.
2953  */
2954 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2955 {
2956         struct slave *slave;
2957         int i;
2958
2959         bond_for_each_slave(bond, slave, i) {
2960                 switch (slave->new_link) {
2961                 case BOND_LINK_NOCHANGE:
2962                         continue;
2963
2964                 case BOND_LINK_UP:
2965                         write_lock_bh(&bond->curr_slave_lock);
2966
2967                         if (!bond->curr_active_slave &&
2968                             time_before_eq(jiffies, slave->dev->trans_start +
2969                                            delta_in_ticks)) {
2970                                 slave->link = BOND_LINK_UP;
2971                                 bond_change_active_slave(bond, slave);
2972                                 bond->current_arp_slave = NULL;
2973
2974                                 printk(KERN_INFO DRV_NAME
2975                                        ": %s: %s is up and now the "
2976                                        "active interface\n",
2977                                        bond->dev->name, slave->dev->name);
2978
2979                         } else if (bond->curr_active_slave != slave) {
2980                                 /* this slave has just come up but we
2981                                  * already have a current slave; this can
2982                                  * also happen if bond_enslave adds a new
2983                                  * slave that is up while we are searching
2984                                  * for a new slave
2985                                  */
2986                                 slave->link = BOND_LINK_UP;
2987                                 bond_set_slave_inactive_flags(slave);
2988                                 bond->current_arp_slave = NULL;
2989
2990                                 printk(KERN_INFO DRV_NAME
2991                                        ": %s: backup interface %s is now up\n",
2992                                        bond->dev->name, slave->dev->name);
2993                         }
2994
2995                         write_unlock_bh(&bond->curr_slave_lock);
2996
2997                         break;
2998
2999                 case BOND_LINK_DOWN:
3000                         if (slave->link_failure_count < UINT_MAX)
3001                                 slave->link_failure_count++;
3002
3003                         slave->link = BOND_LINK_DOWN;
3004
3005                         if (slave == bond->curr_active_slave) {
3006                                 printk(KERN_INFO DRV_NAME
3007                                        ": %s: link status down for active "
3008                                        "interface %s, disabling it\n",
3009                                        bond->dev->name, slave->dev->name);
3010
3011                                 bond_set_slave_inactive_flags(slave);
3012
3013                                 write_lock_bh(&bond->curr_slave_lock);
3014
3015                                 bond_select_active_slave(bond);
3016                                 if (bond->curr_active_slave)
3017                                         bond->curr_active_slave->jiffies =
3018                                                 jiffies;
3019
3020                                 write_unlock_bh(&bond->curr_slave_lock);
3021
3022                                 bond->current_arp_slave = NULL;
3023
3024                         } else if (slave->state == BOND_STATE_BACKUP) {
3025                                 printk(KERN_INFO DRV_NAME
3026                                        ": %s: backup interface %s is now down\n",
3027                                        bond->dev->name, slave->dev->name);
3028
3029                                 bond_set_slave_inactive_flags(slave);
3030                         }
3031                         break;
3032
3033                 default:
3034                         printk(KERN_ERR DRV_NAME
3035                                ": %s: impossible: new_link %d on slave %s\n",
3036                                bond->dev->name, slave->new_link,
3037                                slave->dev->name);
3038                 }
3039         }
3040
3041         /*
3042          * No race with changes to primary via sysfs, as we hold rtnl.
3043          */
3044         if (bond->primary_slave &&
3045             (bond->primary_slave != bond->curr_active_slave) &&
3046             (bond->primary_slave->link == BOND_LINK_UP)) {
3047                 write_lock_bh(&bond->curr_slave_lock);
3048                 bond_change_active_slave(bond, bond->primary_slave);
3049                 write_unlock_bh(&bond->curr_slave_lock);
3050         }
3051
3052         bond_set_carrier(bond);
3053 }
3054
3055 /*
3056  * Send ARP probes for active-backup mode ARP monitor.
3057  *
3058  * Called with bond->lock held for read.
3059  */
3060 static void bond_ab_arp_probe(struct bonding *bond)
3061 {
3062         struct slave *slave;
3063         int i;
3064
3065         read_lock(&bond->curr_slave_lock);
3066
3067         if (bond->current_arp_slave && bond->curr_active_slave)
3068                 printk("PROBE: c_arp %s && cas %s BAD\n",
3069                        bond->current_arp_slave->dev->name,
3070                        bond->curr_active_slave->dev->name);
3071
3072         if (bond->curr_active_slave) {
3073                 bond_arp_send_all(bond, bond->curr_active_slave);
3074                 read_unlock(&bond->curr_slave_lock);
3075                 return;
3076         }
3077
3078         read_unlock(&bond->curr_slave_lock);
3079
3080         /* if we don't have a curr_active_slave, search for the next available
3081          * backup slave from the current_arp_slave and make it the candidate
3082          * for becoming the curr_active_slave
3083          */
3084
3085         if (!bond->current_arp_slave) {
3086                 bond->current_arp_slave = bond->first_slave;
3087                 if (!bond->current_arp_slave)
3088                         return;
3089         }
3090
3091         bond_set_slave_inactive_flags(bond->current_arp_slave);
3092
3093         /* search for next candidate */
3094         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3095                 if (IS_UP(slave->dev)) {
3096                         slave->link = BOND_LINK_BACK;
3097                         bond_set_slave_active_flags(slave);
3098                         bond_arp_send_all(bond, slave);
3099                         slave->jiffies = jiffies;
3100                         bond->current_arp_slave = slave;
3101                         break;
3102                 }
3103
3104                 /* if the link state is up at this point, we
3105                  * mark it down - this can happen if we have
3106                  * simultaneous link failures and
3107                  * reselect_active_interface doesn't make this
3108                  * one the current slave so it is still marked
3109                  * up when it is actually down
3110                  */
3111                 if (slave->link == BOND_LINK_UP) {
3112                         slave->link = BOND_LINK_DOWN;
3113                         if (slave->link_failure_count < UINT_MAX)
3114                                 slave->link_failure_count++;
3115
3116                         bond_set_slave_inactive_flags(slave);
3117
3118                         printk(KERN_INFO DRV_NAME
3119                                ": %s: backup interface %s is now down.\n",
3120                                bond->dev->name, slave->dev->name);
3121                 }
3122         }
3123 }
3124
3125 void bond_activebackup_arp_mon(struct work_struct *work)
3126 {
3127         struct bonding *bond = container_of(work, struct bonding,
3128                                             arp_work.work);
3129         int delta_in_ticks;
3130
3131         read_lock(&bond->lock);
3132
3133         if (bond->kill_timers)
3134                 goto out;
3135
3136         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3137
3138         if (bond->slave_cnt == 0)
3139                 goto re_arm;
3140
3141         if (bond->send_grat_arp) {
3142                 read_lock(&bond->curr_slave_lock);
3143                 bond_send_gratuitous_arp(bond);
3144                 read_unlock(&bond->curr_slave_lock);
3145         }
3146
3147         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3148                 read_unlock(&bond->lock);
3149                 rtnl_lock();
3150                 read_lock(&bond->lock);
3151
3152                 bond_ab_arp_commit(bond, delta_in_ticks);
3153
3154                 read_unlock(&bond->lock);
3155                 rtnl_unlock();
3156                 read_lock(&bond->lock);
3157         }
3158
3159         bond_ab_arp_probe(bond);
3160
3161 re_arm:
3162         if (bond->params.arp_interval) {
3163                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3164         }
3165 out:
3166         read_unlock(&bond->lock);
3167 }
3168
3169 /*------------------------------ proc/seq_file-------------------------------*/
3170
3171 #ifdef CONFIG_PROC_FS
3172
3173 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3174 {
3175         struct bonding *bond = seq->private;
3176         loff_t off = 0;
3177         struct slave *slave;
3178         int i;
3179
3180         /* make sure the bond won't be taken away */
3181         read_lock(&dev_base_lock);
3182         read_lock(&bond->lock);
3183
3184         if (*pos == 0) {
3185                 return SEQ_START_TOKEN;
3186         }
3187
3188         bond_for_each_slave(bond, slave, i) {
3189                 if (++off == *pos) {
3190                         return slave;
3191                 }
3192         }
3193
3194         return NULL;
3195 }
3196
3197 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3198 {
3199         struct bonding *bond = seq->private;
3200         struct slave *slave = v;
3201
3202         ++*pos;
3203         if (v == SEQ_START_TOKEN) {
3204                 return bond->first_slave;
3205         }
3206
3207         slave = slave->next;
3208
3209         return (slave == bond->first_slave) ? NULL : slave;
3210 }
3211
3212 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3213 {
3214         struct bonding *bond = seq->private;
3215
3216         read_unlock(&bond->lock);
3217         read_unlock(&dev_base_lock);
3218 }
3219
3220 static void bond_info_show_master(struct seq_file *seq)
3221 {
3222         struct bonding *bond = seq->private;
3223         struct slave *curr;
3224         int i;
3225         u32 target;
3226
3227         read_lock(&bond->curr_slave_lock);
3228         curr = bond->curr_active_slave;
3229         read_unlock(&bond->curr_slave_lock);
3230
3231         seq_printf(seq, "Bonding Mode: %s",
3232                    bond_mode_name(bond->params.mode));
3233
3234         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3235             bond->params.fail_over_mac)
3236                 seq_printf(seq, " (fail_over_mac %s)",
3237                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3238
3239         seq_printf(seq, "\n");
3240
3241         if (bond->params.mode == BOND_MODE_XOR ||
3242                 bond->params.mode == BOND_MODE_8023AD) {
3243                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3244                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3245                         bond->params.xmit_policy);
3246         }
3247
3248         if (USES_PRIMARY(bond->params.mode)) {
3249                 seq_printf(seq, "Primary Slave: %s\n",
3250                            (bond->primary_slave) ?
3251                            bond->primary_slave->dev->name : "None");
3252
3253                 seq_printf(seq, "Currently Active Slave: %s\n",
3254                            (curr) ? curr->dev->name : "None");
3255         }
3256
3257         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3258                    "up" : "down");
3259         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3260         seq_printf(seq, "Up Delay (ms): %d\n",
3261                    bond->params.updelay * bond->params.miimon);
3262         seq_printf(seq, "Down Delay (ms): %d\n",
3263                    bond->params.downdelay * bond->params.miimon);
3264
3265
3266         /* ARP information */
3267         if(bond->params.arp_interval > 0) {
3268                 int printed=0;
3269                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3270                                 bond->params.arp_interval);
3271
3272                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3273
3274                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3275                         if (!bond->params.arp_targets[i])
3276                                 continue;
3277                         if (printed)
3278                                 seq_printf(seq, ",");
3279                         target = ntohl(bond->params.arp_targets[i]);
3280                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3281                         printed = 1;
3282                 }
3283                 seq_printf(seq, "\n");
3284         }
3285
3286         if (bond->params.mode == BOND_MODE_8023AD) {
3287                 struct ad_info ad_info;
3288                 DECLARE_MAC_BUF(mac);
3289
3290                 seq_puts(seq, "\n802.3ad info\n");
3291                 seq_printf(seq, "LACP rate: %s\n",
3292                            (bond->params.lacp_fast) ? "fast" : "slow");
3293
3294                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3295                         seq_printf(seq, "bond %s has no active aggregator\n",
3296                                    bond->dev->name);
3297                 } else {
3298                         seq_printf(seq, "Active Aggregator Info:\n");
3299
3300                         seq_printf(seq, "\tAggregator ID: %d\n",
3301                                    ad_info.aggregator_id);
3302                         seq_printf(seq, "\tNumber of ports: %d\n",
3303                                    ad_info.ports);
3304                         seq_printf(seq, "\tActor Key: %d\n",
3305                                    ad_info.actor_key);
3306                         seq_printf(seq, "\tPartner Key: %d\n",
3307                                    ad_info.partner_key);
3308                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3309                                    print_mac(mac, ad_info.partner_system));
3310                 }
3311         }
3312 }
3313
3314 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3315 {
3316         struct bonding *bond = seq->private;
3317         DECLARE_MAC_BUF(mac);
3318
3319         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3320         seq_printf(seq, "MII Status: %s\n",
3321                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3322         seq_printf(seq, "Link Failure Count: %u\n",
3323                    slave->link_failure_count);
3324
3325         seq_printf(seq,
3326                    "Permanent HW addr: %s\n",
3327                    print_mac(mac, slave->perm_hwaddr));
3328
3329         if (bond->params.mode == BOND_MODE_8023AD) {
3330                 const struct aggregator *agg
3331                         = SLAVE_AD_INFO(slave).port.aggregator;
3332
3333                 if (agg) {
3334                         seq_printf(seq, "Aggregator ID: %d\n",
3335                                    agg->aggregator_identifier);
3336                 } else {
3337                         seq_puts(seq, "Aggregator ID: N/A\n");
3338                 }
3339         }
3340 }
3341
3342 static int bond_info_seq_show(struct seq_file *seq, void *v)
3343 {
3344         if (v == SEQ_START_TOKEN) {
3345                 seq_printf(seq, "%s\n", version);
3346                 bond_info_show_master(seq);
3347         } else {
3348                 bond_info_show_slave(seq, v);
3349         }
3350
3351         return 0;
3352 }
3353
3354 static struct seq_operations bond_info_seq_ops = {
3355         .start = bond_info_seq_start,
3356         .next  = bond_info_seq_next,
3357         .stop  = bond_info_seq_stop,
3358         .show  = bond_info_seq_show,
3359 };
3360
3361 static int bond_info_open(struct inode *inode, struct file *file)
3362 {
3363         struct seq_file *seq;
3364         struct proc_dir_entry *proc;
3365         int res;
3366
3367         res = seq_open(file, &bond_info_seq_ops);
3368         if (!res) {
3369                 /* recover the pointer buried in proc_dir_entry data */
3370                 seq = file->private_data;
3371                 proc = PDE(inode);
3372                 seq->private = proc->data;
3373         }
3374
3375         return res;
3376 }
3377
3378 static const struct file_operations bond_info_fops = {
3379         .owner   = THIS_MODULE,
3380         .open    = bond_info_open,
3381         .read    = seq_read,
3382         .llseek  = seq_lseek,
3383         .release = seq_release,
3384 };
3385
3386 static int bond_create_proc_entry(struct bonding *bond)
3387 {
3388         struct net_device *bond_dev = bond->dev;
3389
3390         if (bond_proc_dir) {
3391                 bond->proc_entry = proc_create_data(bond_dev->name,
3392                                                     S_IRUGO, bond_proc_dir,
3393                                                     &bond_info_fops, bond);
3394                 if (bond->proc_entry == NULL) {
3395                         printk(KERN_WARNING DRV_NAME
3396                                ": Warning: Cannot create /proc/net/%s/%s\n",
3397                                DRV_NAME, bond_dev->name);
3398                 } else {
3399                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3400                 }
3401         }
3402
3403         return 0;
3404 }
3405
3406 static void bond_remove_proc_entry(struct bonding *bond)
3407 {
3408         if (bond_proc_dir && bond->proc_entry) {
3409                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3410                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3411                 bond->proc_entry = NULL;
3412         }
3413 }
3414
3415 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3416  * Caller must hold rtnl_lock.
3417  */
3418 static void bond_create_proc_dir(void)
3419 {
3420         int len = strlen(DRV_NAME);
3421
3422         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3423              bond_proc_dir = bond_proc_dir->next) {
3424                 if ((bond_proc_dir->namelen == len) &&
3425                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3426                         break;
3427                 }
3428         }
3429
3430         if (!bond_proc_dir) {
3431                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3432                 if (bond_proc_dir) {
3433                         bond_proc_dir->owner = THIS_MODULE;
3434                 } else {
3435                         printk(KERN_WARNING DRV_NAME
3436                                 ": Warning: cannot create /proc/net/%s\n",
3437                                 DRV_NAME);
3438                 }
3439         }
3440 }
3441
3442 /* Destroy the bonding directory under /proc/net, if empty.
3443  * Caller must hold rtnl_lock.
3444  */
3445 static void bond_destroy_proc_dir(void)
3446 {
3447         struct proc_dir_entry *de;
3448
3449         if (!bond_proc_dir) {
3450                 return;
3451         }
3452
3453         /* verify that the /proc dir is empty */
3454         for (de = bond_proc_dir->subdir; de; de = de->next) {
3455                 /* ignore . and .. */
3456                 if (*(de->name) != '.') {
3457                         break;
3458                 }
3459         }
3460
3461         if (de) {
3462                 if (bond_proc_dir->owner == THIS_MODULE) {
3463                         bond_proc_dir->owner = NULL;
3464                 }
3465         } else {
3466                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3467                 bond_proc_dir = NULL;
3468         }
3469 }
3470 #endif /* CONFIG_PROC_FS */
3471
3472 /*-------------------------- netdev event handling --------------------------*/
3473
3474 /*
3475  * Change device name
3476  */
3477 static int bond_event_changename(struct bonding *bond)
3478 {
3479 #ifdef CONFIG_PROC_FS
3480         bond_remove_proc_entry(bond);
3481         bond_create_proc_entry(bond);
3482 #endif
3483         down_write(&(bonding_rwsem));
3484         bond_destroy_sysfs_entry(bond);
3485         bond_create_sysfs_entry(bond);
3486         up_write(&(bonding_rwsem));
3487         return NOTIFY_DONE;
3488 }
3489
3490 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3491 {
3492         struct bonding *event_bond = bond_dev->priv;
3493
3494         switch (event) {
3495         case NETDEV_CHANGENAME:
3496                 return bond_event_changename(event_bond);
3497         case NETDEV_UNREGISTER:
3498                 bond_release_all(event_bond->dev);
3499                 break;
3500         default:
3501                 break;
3502         }
3503
3504         return NOTIFY_DONE;
3505 }
3506
3507 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3508 {
3509         struct net_device *bond_dev = slave_dev->master;
3510         struct bonding *bond = bond_dev->priv;
3511
3512         switch (event) {
3513         case NETDEV_UNREGISTER:
3514                 if (bond_dev) {
3515                         if (bond->setup_by_slave)
3516                                 bond_release_and_destroy(bond_dev, slave_dev);
3517                         else
3518                                 bond_release(bond_dev, slave_dev);
3519                 }
3520                 break;
3521         case NETDEV_CHANGE:
3522                 /*
3523                  * TODO: is this what we get if somebody
3524                  * sets up a hierarchical bond, then rmmod's
3525                  * one of the slave bonding devices?
3526                  */
3527                 break;
3528         case NETDEV_DOWN:
3529                 /*
3530                  * ... Or is it this?
3531                  */
3532                 break;
3533         case NETDEV_CHANGEMTU:
3534                 /*
3535                  * TODO: Should slaves be allowed to
3536                  * independently alter their MTU?  For
3537                  * an active-backup bond, slaves need
3538                  * not be the same type of device, so
3539                  * MTUs may vary.  For other modes,
3540                  * slaves arguably should have the
3541                  * same MTUs. To do this, we'd need to
3542                  * take over the slave's change_mtu
3543                  * function for the duration of their
3544                  * servitude.
3545                  */
3546                 break;
3547         case NETDEV_CHANGENAME:
3548                 /*
3549                  * TODO: handle changing the primary's name
3550                  */
3551                 break;
3552         case NETDEV_FEAT_CHANGE:
3553                 bond_compute_features(bond);
3554                 break;
3555         default:
3556                 break;
3557         }
3558
3559         return NOTIFY_DONE;
3560 }
3561
3562 /*
3563  * bond_netdev_event: handle netdev notifier chain events.
3564  *
3565  * This function receives events for the netdev chain.  The caller (an
3566  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3567  * locks for us to safely manipulate the slave devices (RTNL lock,
3568  * dev_probe_lock).
3569  */
3570 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3571 {
3572         struct net_device *event_dev = (struct net_device *)ptr;
3573
3574         if (dev_net(event_dev) != &init_net)
3575                 return NOTIFY_DONE;
3576
3577         dprintk("event_dev: %s, event: %lx\n",
3578                 (event_dev ? event_dev->name : "None"),
3579                 event);
3580
3581         if (!(event_dev->priv_flags & IFF_BONDING))
3582                 return NOTIFY_DONE;
3583
3584         if (event_dev->flags & IFF_MASTER) {
3585                 dprintk("IFF_MASTER\n");
3586                 return bond_master_netdev_event(event, event_dev);
3587         }
3588
3589         if (event_dev->flags & IFF_SLAVE) {
3590                 dprintk("IFF_SLAVE\n");
3591                 return bond_slave_netdev_event(event, event_dev);
3592         }
3593
3594         return NOTIFY_DONE;
3595 }
3596
3597 /*
3598  * bond_inetaddr_event: handle inetaddr notifier chain events.
3599  *
3600  * We keep track of device IPs primarily to use as source addresses in
3601  * ARP monitor probes (rather than spewing out broadcasts all the time).
3602  *
3603  * We track one IP for the main device (if it has one), plus one per VLAN.
3604  */
3605 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3606 {
3607         struct in_ifaddr *ifa = ptr;
3608         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3609         struct bonding *bond;
3610         struct vlan_entry *vlan;
3611
3612         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3613                 return NOTIFY_DONE;
3614
3615         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3616                 if (bond->dev == event_dev) {
3617                         switch (event) {
3618                         case NETDEV_UP:
3619                                 bond->master_ip = ifa->ifa_local;
3620                                 return NOTIFY_OK;
3621                         case NETDEV_DOWN:
3622                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3623                                 return NOTIFY_OK;
3624                         default:
3625                                 return NOTIFY_DONE;
3626                         }
3627                 }
3628
3629                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3630                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3631                         if (vlan_dev == event_dev) {
3632                                 switch (event) {
3633                                 case NETDEV_UP:
3634                                         vlan->vlan_ip = ifa->ifa_local;
3635                                         return NOTIFY_OK;
3636                                 case NETDEV_DOWN:
3637                                         vlan->vlan_ip =
3638                                                 bond_glean_dev_ip(vlan_dev);
3639                                         return NOTIFY_OK;
3640                                 default:
3641                                         return NOTIFY_DONE;
3642                                 }
3643                         }
3644                 }
3645         }
3646         return NOTIFY_DONE;
3647 }
3648
3649 static struct notifier_block bond_netdev_notifier = {
3650         .notifier_call = bond_netdev_event,
3651 };
3652
3653 static struct notifier_block bond_inetaddr_notifier = {
3654         .notifier_call = bond_inetaddr_event,
3655 };
3656
3657 /*-------------------------- Packet type handling ---------------------------*/
3658
3659 /* register to receive lacpdus on a bond */
3660 static void bond_register_lacpdu(struct bonding *bond)
3661 {
3662         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3663
3664         /* initialize packet type */
3665         pk_type->type = PKT_TYPE_LACPDU;
3666         pk_type->dev = bond->dev;
3667         pk_type->func = bond_3ad_lacpdu_recv;
3668
3669         dev_add_pack(pk_type);
3670 }
3671
3672 /* unregister to receive lacpdus on a bond */
3673 static void bond_unregister_lacpdu(struct bonding *bond)
3674 {
3675         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3676 }
3677
3678 void bond_register_arp(struct bonding *bond)
3679 {
3680         struct packet_type *pt = &bond->arp_mon_pt;
3681
3682         if (pt->type)
3683                 return;
3684
3685         pt->type = htons(ETH_P_ARP);
3686         pt->dev = bond->dev;
3687         pt->func = bond_arp_rcv;
3688         dev_add_pack(pt);
3689 }
3690
3691 void bond_unregister_arp(struct bonding *bond)
3692 {
3693         struct packet_type *pt = &bond->arp_mon_pt;
3694
3695         dev_remove_pack(pt);
3696         pt->type = 0;
3697 }
3698
3699 /*---------------------------- Hashing Policies -----------------------------*/
3700
3701 /*
3702  * Hash for the output device based upon layer 2 and layer 3 data. If
3703  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3704  */
3705 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3706                                      struct net_device *bond_dev, int count)
3707 {
3708         struct ethhdr *data = (struct ethhdr *)skb->data;
3709         struct iphdr *iph = ip_hdr(skb);
3710
3711         if (skb->protocol == htons(ETH_P_IP)) {
3712                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3713                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3714         }
3715
3716         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3717 }
3718
3719 /*
3720  * Hash for the output device based upon layer 3 and layer 4 data. If
3721  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3722  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3723  */
3724 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3725                                     struct net_device *bond_dev, int count)
3726 {
3727         struct ethhdr *data = (struct ethhdr *)skb->data;
3728         struct iphdr *iph = ip_hdr(skb);
3729         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3730         int layer4_xor = 0;
3731
3732         if (skb->protocol == htons(ETH_P_IP)) {
3733                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3734                     (iph->protocol == IPPROTO_TCP ||
3735                      iph->protocol == IPPROTO_UDP)) {
3736                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3737                 }
3738                 return (layer4_xor ^
3739                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3740
3741         }
3742
3743         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3744 }
3745
3746 /*
3747  * Hash for the output device based upon layer 2 data
3748  */
3749 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3750                                    struct net_device *bond_dev, int count)
3751 {
3752         struct ethhdr *data = (struct ethhdr *)skb->data;
3753
3754         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3755 }
3756
3757 /*-------------------------- Device entry points ----------------------------*/
3758
3759 static int bond_open(struct net_device *bond_dev)
3760 {
3761         struct bonding *bond = bond_dev->priv;
3762
3763         bond->kill_timers = 0;
3764
3765         if ((bond->params.mode == BOND_MODE_TLB) ||
3766             (bond->params.mode == BOND_MODE_ALB)) {
3767                 /* bond_alb_initialize must be called before the timer
3768                  * is started.
3769                  */
3770                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3771                         /* something went wrong - fail the open operation */
3772                         return -1;
3773                 }
3774
3775                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3776                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3777         }
3778
3779         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3780                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3781                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3782         }
3783
3784         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3785                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3786                         INIT_DELAYED_WORK(&bond->arp_work,
3787                                           bond_activebackup_arp_mon);
3788                 else
3789                         INIT_DELAYED_WORK(&bond->arp_work,
3790                                           bond_loadbalance_arp_mon);
3791
3792                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3793                 if (bond->params.arp_validate)
3794                         bond_register_arp(bond);
3795         }
3796
3797         if (bond->params.mode == BOND_MODE_8023AD) {
3798                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3799                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3800                 /* register to receive LACPDUs */
3801                 bond_register_lacpdu(bond);
3802         }
3803
3804         return 0;
3805 }
3806
3807 static int bond_close(struct net_device *bond_dev)
3808 {
3809         struct bonding *bond = bond_dev->priv;
3810
3811         if (bond->params.mode == BOND_MODE_8023AD) {
3812                 /* Unregister the receive of LACPDUs */
3813                 bond_unregister_lacpdu(bond);
3814         }
3815
3816         if (bond->params.arp_validate)
3817                 bond_unregister_arp(bond);
3818
3819         write_lock_bh(&bond->lock);
3820
3821         bond->send_grat_arp = 0;
3822
3823         /* signal timers not to re-arm */
3824         bond->kill_timers = 1;
3825
3826         write_unlock_bh(&bond->lock);
3827
3828         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3829                 cancel_delayed_work(&bond->mii_work);
3830         }
3831
3832         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3833                 cancel_delayed_work(&bond->arp_work);
3834         }
3835
3836         switch (bond->params.mode) {
3837         case BOND_MODE_8023AD:
3838                 cancel_delayed_work(&bond->ad_work);
3839                 break;
3840         case BOND_MODE_TLB:
3841         case BOND_MODE_ALB:
3842                 cancel_delayed_work(&bond->alb_work);
3843                 break;
3844         default:
3845                 break;
3846         }
3847
3848
3849         if ((bond->params.mode == BOND_MODE_TLB) ||
3850             (bond->params.mode == BOND_MODE_ALB)) {
3851                 /* Must be called only after all
3852                  * slaves have been released
3853                  */
3854                 bond_alb_deinitialize(bond);
3855         }
3856
3857         return 0;
3858 }
3859
3860 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3861 {
3862         struct bonding *bond = bond_dev->priv;
3863         struct net_device_stats *stats = &(bond->stats), *sstats;
3864         struct net_device_stats local_stats;
3865         struct slave *slave;
3866         int i;
3867
3868         memset(&local_stats, 0, sizeof(struct net_device_stats));
3869
3870         read_lock_bh(&bond->lock);
3871
3872         bond_for_each_slave(bond, slave, i) {
3873                 sstats = slave->dev->get_stats(slave->dev);
3874                 local_stats.rx_packets += sstats->rx_packets;
3875                 local_stats.rx_bytes += sstats->rx_bytes;
3876                 local_stats.rx_errors += sstats->rx_errors;
3877                 local_stats.rx_dropped += sstats->rx_dropped;
3878
3879                 local_stats.tx_packets += sstats->tx_packets;
3880                 local_stats.tx_bytes += sstats->tx_bytes;
3881                 local_stats.tx_errors += sstats->tx_errors;
3882                 local_stats.tx_dropped += sstats->tx_dropped;
3883
3884                 local_stats.multicast += sstats->multicast;
3885                 local_stats.collisions += sstats->collisions;
3886
3887                 local_stats.rx_length_errors += sstats->rx_length_errors;
3888                 local_stats.rx_over_errors += sstats->rx_over_errors;
3889                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3890                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3891                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3892                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3893
3894                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3895                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3896                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3897                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3898                 local_stats.tx_window_errors += sstats->tx_window_errors;
3899         }
3900
3901         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3902
3903         read_unlock_bh(&bond->lock);
3904
3905         return stats;
3906 }
3907
3908 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3909 {
3910         struct net_device *slave_dev = NULL;
3911         struct ifbond k_binfo;
3912         struct ifbond __user *u_binfo = NULL;
3913         struct ifslave k_sinfo;
3914         struct ifslave __user *u_sinfo = NULL;
3915         struct mii_ioctl_data *mii = NULL;
3916         int res = 0;
3917
3918         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3919                 bond_dev->name, cmd);
3920
3921         switch (cmd) {
3922         case SIOCGMIIPHY:
3923                 mii = if_mii(ifr);
3924                 if (!mii) {
3925                         return -EINVAL;
3926                 }
3927                 mii->phy_id = 0;
3928                 /* Fall Through */
3929         case SIOCGMIIREG:
3930                 /*
3931                  * We do this again just in case we were called by SIOCGMIIREG
3932                  * instead of SIOCGMIIPHY.
3933                  */
3934                 mii = if_mii(ifr);
3935                 if (!mii) {
3936                         return -EINVAL;
3937                 }
3938
3939                 if (mii->reg_num == 1) {
3940                         struct bonding *bond = bond_dev->priv;
3941                         mii->val_out = 0;
3942                         read_lock(&bond->lock);
3943                         read_lock(&bond->curr_slave_lock);
3944                         if (netif_carrier_ok(bond->dev)) {
3945                                 mii->val_out = BMSR_LSTATUS;
3946                         }
3947                         read_unlock(&bond->curr_slave_lock);
3948                         read_unlock(&bond->lock);
3949                 }
3950
3951                 return 0;
3952         case BOND_INFO_QUERY_OLD:
3953         case SIOCBONDINFOQUERY:
3954                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3955
3956                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3957                         return -EFAULT;
3958                 }
3959
3960                 res = bond_info_query(bond_dev, &k_binfo);
3961                 if (res == 0) {
3962                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3963                                 return -EFAULT;
3964                         }
3965                 }
3966
3967                 return res;
3968         case BOND_SLAVE_INFO_QUERY_OLD:
3969         case SIOCBONDSLAVEINFOQUERY:
3970                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3971
3972                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3973                         return -EFAULT;
3974                 }
3975
3976                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3977                 if (res == 0) {
3978                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3979                                 return -EFAULT;
3980                         }
3981                 }
3982
3983                 return res;
3984         default:
3985                 /* Go on */
3986                 break;
3987         }
3988
3989         if (!capable(CAP_NET_ADMIN)) {
3990                 return -EPERM;
3991         }
3992
3993         down_write(&(bonding_rwsem));
3994         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3995
3996         dprintk("slave_dev=%p: \n", slave_dev);
3997
3998         if (!slave_dev) {
3999                 res = -ENODEV;
4000         } else {
4001                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4002                 switch (cmd) {
4003                 case BOND_ENSLAVE_OLD:
4004                 case SIOCBONDENSLAVE:
4005                         res = bond_enslave(bond_dev, slave_dev);
4006                         break;
4007                 case BOND_RELEASE_OLD:
4008                 case SIOCBONDRELEASE:
4009                         res = bond_release(bond_dev, slave_dev);
4010                         break;
4011                 case BOND_SETHWADDR_OLD:
4012                 case SIOCBONDSETHWADDR:
4013                         res = bond_sethwaddr(bond_dev, slave_dev);
4014                         break;
4015                 case BOND_CHANGE_ACTIVE_OLD:
4016                 case SIOCBONDCHANGEACTIVE:
4017                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4018                         break;
4019                 default:
4020                         res = -EOPNOTSUPP;
4021                 }
4022
4023                 dev_put(slave_dev);
4024         }
4025
4026         up_write(&(bonding_rwsem));
4027         return res;
4028 }
4029
4030 static void bond_set_multicast_list(struct net_device *bond_dev)
4031 {
4032         struct bonding *bond = bond_dev->priv;
4033         struct dev_mc_list *dmi;
4034
4035         /*
4036          * Do promisc before checking multicast_mode
4037          */
4038         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4039                 /*
4040                  * FIXME: Need to handle the error when one of the multi-slaves
4041                  * encounters error.
4042                  */
4043                 bond_set_promiscuity(bond, 1);
4044         }
4045
4046         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4047                 bond_set_promiscuity(bond, -1);
4048         }
4049
4050         /* set allmulti flag to slaves */
4051         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4052                 /*
4053                  * FIXME: Need to handle the error when one of the multi-slaves
4054                  * encounters error.
4055                  */
4056                 bond_set_allmulti(bond, 1);
4057         }
4058
4059         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4060                 bond_set_allmulti(bond, -1);
4061         }
4062
4063         read_lock(&bond->lock);
4064
4065         bond->flags = bond_dev->flags;
4066
4067         /* looking for addresses to add to slaves' mc list */
4068         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4069                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4070                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4071                 }
4072         }
4073
4074         /* looking for addresses to delete from slaves' list */
4075         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4076                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4077                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4078                 }
4079         }
4080
4081         /* save master's multicast list */
4082         bond_mc_list_destroy(bond);
4083         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4084
4085         read_unlock(&bond->lock);
4086 }
4087
4088 /*
4089  * Change the MTU of all of a master's slaves to match the master
4090  */
4091 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4092 {
4093         struct bonding *bond = bond_dev->priv;
4094         struct slave *slave, *stop_at;
4095         int res = 0;
4096         int i;
4097
4098         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4099                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4100
4101         /* Can't hold bond->lock with bh disabled here since
4102          * some base drivers panic. On the other hand we can't
4103          * hold bond->lock without bh disabled because we'll
4104          * deadlock. The only solution is to rely on the fact
4105          * that we're under rtnl_lock here, and the slaves
4106          * list won't change. This doesn't solve the problem
4107          * of setting the slave's MTU while it is
4108          * transmitting, but the assumption is that the base
4109          * driver can handle that.
4110          *
4111          * TODO: figure out a way to safely iterate the slaves
4112          * list, but without holding a lock around the actual
4113          * call to the base driver.
4114          */
4115
4116         bond_for_each_slave(bond, slave, i) {
4117                 dprintk("s %p s->p %p c_m %p\n", slave,
4118                         slave->prev, slave->dev->change_mtu);
4119
4120                 res = dev_set_mtu(slave->dev, new_mtu);
4121
4122                 if (res) {
4123                         /* If we failed to set the slave's mtu to the new value
4124                          * we must abort the operation even in ACTIVE_BACKUP
4125                          * mode, because if we allow the backup slaves to have
4126                          * different mtu values than the active slave we'll
4127                          * need to change their mtu when doing a failover. That
4128                          * means changing their mtu from timer context, which
4129                          * is probably not a good idea.
4130                          */
4131                         dprintk("err %d %s\n", res, slave->dev->name);
4132                         goto unwind;
4133                 }
4134         }
4135
4136         bond_dev->mtu = new_mtu;
4137
4138         return 0;
4139
4140 unwind:
4141         /* unwind from head to the slave that failed */
4142         stop_at = slave;
4143         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4144                 int tmp_res;
4145
4146                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4147                 if (tmp_res) {
4148                         dprintk("unwind err %d dev %s\n", tmp_res,
4149                                 slave->dev->name);
4150                 }
4151         }
4152
4153         return res;
4154 }
4155
4156 /*
4157  * Change HW address
4158  *
4159  * Note that many devices must be down to change the HW address, and
4160  * downing the master releases all slaves.  We can make bonds full of
4161  * bonding devices to test this, however.
4162  */
4163 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4164 {
4165         struct bonding *bond = bond_dev->priv;
4166         struct sockaddr *sa = addr, tmp_sa;
4167         struct slave *slave, *stop_at;
4168         int res = 0;
4169         int i;
4170
4171         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4172
4173         /*
4174          * If fail_over_mac is set to active, do nothing and return
4175          * success.  Returning an error causes ifenslave to fail.
4176          */
4177         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4178                 return 0;
4179
4180         if (!is_valid_ether_addr(sa->sa_data)) {
4181                 return -EADDRNOTAVAIL;
4182         }
4183
4184         /* Can't hold bond->lock with bh disabled here since
4185          * some base drivers panic. On the other hand we can't
4186          * hold bond->lock without bh disabled because we'll
4187          * deadlock. The only solution is to rely on the fact
4188          * that we're under rtnl_lock here, and the slaves
4189          * list won't change. This doesn't solve the problem
4190          * of setting the slave's hw address while it is
4191          * transmitting, but the assumption is that the base
4192          * driver can handle that.
4193          *
4194          * TODO: figure out a way to safely iterate the slaves
4195          * list, but without holding a lock around the actual
4196          * call to the base driver.
4197          */
4198
4199         bond_for_each_slave(bond, slave, i) {
4200                 dprintk("slave %p %s\n", slave, slave->dev->name);
4201
4202                 if (slave->dev->set_mac_address == NULL) {
4203                         res = -EOPNOTSUPP;
4204                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4205                         goto unwind;
4206                 }
4207
4208                 res = dev_set_mac_address(slave->dev, addr);
4209                 if (res) {
4210                         /* TODO: consider downing the slave
4211                          * and retry ?
4212                          * User should expect communications
4213                          * breakage anyway until ARP finish
4214                          * updating, so...
4215                          */
4216                         dprintk("err %d %s\n", res, slave->dev->name);
4217                         goto unwind;
4218                 }
4219         }
4220
4221         /* success */
4222         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4223         return 0;
4224
4225 unwind:
4226         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4227         tmp_sa.sa_family = bond_dev->type;
4228
4229         /* unwind from head to the slave that failed */
4230         stop_at = slave;
4231         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4232                 int tmp_res;
4233
4234                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4235                 if (tmp_res) {
4236                         dprintk("unwind err %d dev %s\n", tmp_res,
4237                                 slave->dev->name);
4238                 }
4239         }
4240
4241         return res;
4242 }
4243
4244 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4245 {
4246         struct bonding *bond = bond_dev->priv;
4247         struct slave *slave, *start_at;
4248         int i, slave_no, res = 1;
4249
4250         read_lock(&bond->lock);
4251
4252         if (!BOND_IS_OK(bond)) {
4253                 goto out;
4254         }
4255
4256         /*
4257          * Concurrent TX may collide on rr_tx_counter; we accept that
4258          * as being rare enough not to justify using an atomic op here
4259          */
4260         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4261
4262         bond_for_each_slave(bond, slave, i) {
4263                 slave_no--;
4264                 if (slave_no < 0) {
4265                         break;
4266                 }
4267         }
4268
4269         start_at = slave;
4270         bond_for_each_slave_from(bond, slave, i, start_at) {
4271                 if (IS_UP(slave->dev) &&
4272                     (slave->link == BOND_LINK_UP) &&
4273                     (slave->state == BOND_STATE_ACTIVE)) {
4274                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4275                         break;
4276                 }
4277         }
4278
4279 out:
4280         if (res) {
4281                 /* no suitable interface, frame not sent */
4282                 dev_kfree_skb(skb);
4283         }
4284         read_unlock(&bond->lock);
4285         return 0;
4286 }
4287
4288
4289 /*
4290  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4291  * the bond has a usable interface.
4292  */
4293 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4294 {
4295         struct bonding *bond = bond_dev->priv;
4296         int res = 1;
4297
4298         read_lock(&bond->lock);
4299         read_lock(&bond->curr_slave_lock);
4300
4301         if (!BOND_IS_OK(bond)) {
4302                 goto out;
4303         }
4304
4305         if (!bond->curr_active_slave)
4306                 goto out;
4307
4308         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4309
4310 out:
4311         if (res) {
4312                 /* no suitable interface, frame not sent */
4313                 dev_kfree_skb(skb);
4314         }
4315         read_unlock(&bond->curr_slave_lock);
4316         read_unlock(&bond->lock);
4317         return 0;
4318 }
4319
4320 /*
4321  * In bond_xmit_xor() , we determine the output device by using a pre-
4322  * determined xmit_hash_policy(), If the selected device is not enabled,
4323  * find the next active slave.
4324  */
4325 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4326 {
4327         struct bonding *bond = bond_dev->priv;
4328         struct slave *slave, *start_at;
4329         int slave_no;
4330         int i;
4331         int res = 1;
4332
4333         read_lock(&bond->lock);
4334
4335         if (!BOND_IS_OK(bond)) {
4336                 goto out;
4337         }
4338
4339         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4340
4341         bond_for_each_slave(bond, slave, i) {
4342                 slave_no--;
4343                 if (slave_no < 0) {
4344                         break;
4345                 }
4346         }
4347
4348         start_at = slave;
4349
4350         bond_for_each_slave_from(bond, slave, i, start_at) {
4351                 if (IS_UP(slave->dev) &&
4352                     (slave->link == BOND_LINK_UP) &&
4353                     (slave->state == BOND_STATE_ACTIVE)) {
4354                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4355                         break;
4356                 }
4357         }
4358
4359 out:
4360         if (res) {
4361                 /* no suitable interface, frame not sent */
4362                 dev_kfree_skb(skb);
4363         }
4364         read_unlock(&bond->lock);
4365         return 0;
4366 }
4367
4368 /*
4369  * in broadcast mode, we send everything to all usable interfaces.
4370  */
4371 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4372 {
4373         struct bonding *bond = bond_dev->priv;
4374         struct slave *slave, *start_at;
4375         struct net_device *tx_dev = NULL;
4376         int i;
4377         int res = 1;
4378
4379         read_lock(&bond->lock);
4380
4381         if (!BOND_IS_OK(bond)) {
4382                 goto out;
4383         }
4384
4385         read_lock(&bond->curr_slave_lock);
4386         start_at = bond->curr_active_slave;
4387         read_unlock(&bond->curr_slave_lock);
4388
4389         if (!start_at) {
4390                 goto out;
4391         }
4392
4393         bond_for_each_slave_from(bond, slave, i, start_at) {
4394                 if (IS_UP(slave->dev) &&
4395                     (slave->link == BOND_LINK_UP) &&
4396                     (slave->state == BOND_STATE_ACTIVE)) {
4397                         if (tx_dev) {
4398                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4399                                 if (!skb2) {
4400                                         printk(KERN_ERR DRV_NAME
4401                                                ": %s: Error: bond_xmit_broadcast(): "
4402                                                "skb_clone() failed\n",
4403                                                bond_dev->name);
4404                                         continue;
4405                                 }
4406
4407                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4408                                 if (res) {
4409                                         dev_kfree_skb(skb2);
4410                                         continue;
4411                                 }
4412                         }
4413                         tx_dev = slave->dev;
4414                 }
4415         }
4416
4417         if (tx_dev) {
4418                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4419         }
4420
4421 out:
4422         if (res) {
4423                 /* no suitable interface, frame not sent */
4424                 dev_kfree_skb(skb);
4425         }
4426         /* frame sent to all suitable interfaces */
4427         read_unlock(&bond->lock);
4428         return 0;
4429 }
4430
4431 /*------------------------- Device initialization ---------------------------*/
4432
4433 static void bond_set_xmit_hash_policy(struct bonding *bond)
4434 {
4435         switch (bond->params.xmit_policy) {
4436         case BOND_XMIT_POLICY_LAYER23:
4437                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4438                 break;
4439         case BOND_XMIT_POLICY_LAYER34:
4440                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4441                 break;
4442         case BOND_XMIT_POLICY_LAYER2:
4443         default:
4444                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4445                 break;
4446         }
4447 }
4448
4449 /*
4450  * set bond mode specific net device operations
4451  */
4452 void bond_set_mode_ops(struct bonding *bond, int mode)
4453 {
4454         struct net_device *bond_dev = bond->dev;
4455
4456         switch (mode) {
4457         case BOND_MODE_ROUNDROBIN:
4458                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4459                 break;
4460         case BOND_MODE_ACTIVEBACKUP:
4461                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4462                 break;
4463         case BOND_MODE_XOR:
4464                 bond_dev->hard_start_xmit = bond_xmit_xor;
4465                 bond_set_xmit_hash_policy(bond);
4466                 break;
4467         case BOND_MODE_BROADCAST:
4468                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4469                 break;
4470         case BOND_MODE_8023AD:
4471                 bond_set_master_3ad_flags(bond);
4472                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4473                 bond_set_xmit_hash_policy(bond);
4474                 break;
4475         case BOND_MODE_ALB:
4476                 bond_set_master_alb_flags(bond);
4477                 /* FALLTHRU */
4478         case BOND_MODE_TLB:
4479                 bond_dev->hard_start_xmit = bond_alb_xmit;
4480                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4481                 break;
4482         default:
4483                 /* Should never happen, mode already checked */
4484                 printk(KERN_ERR DRV_NAME
4485                        ": %s: Error: Unknown bonding mode %d\n",
4486                        bond_dev->name,
4487                        mode);
4488                 break;
4489         }
4490 }
4491
4492 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4493                                     struct ethtool_drvinfo *drvinfo)
4494 {
4495         strncpy(drvinfo->driver, DRV_NAME, 32);
4496         strncpy(drvinfo->version, DRV_VERSION, 32);
4497         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4498 }
4499
4500 static const struct ethtool_ops bond_ethtool_ops = {
4501         .get_drvinfo            = bond_ethtool_get_drvinfo,
4502         .get_link               = ethtool_op_get_link,
4503         .get_tx_csum            = ethtool_op_get_tx_csum,
4504         .get_sg                 = ethtool_op_get_sg,
4505         .get_tso                = ethtool_op_get_tso,
4506         .get_ufo                = ethtool_op_get_ufo,
4507         .get_flags              = ethtool_op_get_flags,
4508 };
4509
4510 /*
4511  * Does not allocate but creates a /proc entry.
4512  * Allowed to fail.
4513  */
4514 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4515 {
4516         struct bonding *bond = bond_dev->priv;
4517
4518         dprintk("Begin bond_init for %s\n", bond_dev->name);
4519
4520         /* initialize rwlocks */
4521         rwlock_init(&bond->lock);
4522         rwlock_init(&bond->curr_slave_lock);
4523
4524         bond->params = *params; /* copy params struct */
4525
4526         bond->wq = create_singlethread_workqueue(bond_dev->name);
4527         if (!bond->wq)
4528                 return -ENOMEM;
4529
4530         /* Initialize pointers */
4531         bond->first_slave = NULL;
4532         bond->curr_active_slave = NULL;
4533         bond->current_arp_slave = NULL;
4534         bond->primary_slave = NULL;
4535         bond->dev = bond_dev;
4536         bond->send_grat_arp = 0;
4537         bond->setup_by_slave = 0;
4538         INIT_LIST_HEAD(&bond->vlan_list);
4539
4540         /* Initialize the device entry points */
4541         bond_dev->open = bond_open;
4542         bond_dev->stop = bond_close;
4543         bond_dev->get_stats = bond_get_stats;
4544         bond_dev->do_ioctl = bond_do_ioctl;
4545         bond_dev->ethtool_ops = &bond_ethtool_ops;
4546         bond_dev->set_multicast_list = bond_set_multicast_list;
4547         bond_dev->change_mtu = bond_change_mtu;
4548         bond_dev->set_mac_address = bond_set_mac_address;
4549         bond_dev->validate_addr = NULL;
4550
4551         bond_set_mode_ops(bond, bond->params.mode);
4552
4553         bond_dev->destructor = free_netdev;
4554
4555         /* Initialize the device options */
4556         bond_dev->tx_queue_len = 0;
4557         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4558         bond_dev->priv_flags |= IFF_BONDING;
4559
4560         /* At first, we block adding VLANs. That's the only way to
4561          * prevent problems that occur when adding VLANs over an
4562          * empty bond. The block will be removed once non-challenged
4563          * slaves are enslaved.
4564          */
4565         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4566
4567         /* don't acquire bond device's netif_tx_lock when
4568          * transmitting */
4569         bond_dev->features |= NETIF_F_LLTX;
4570
4571         /* By default, we declare the bond to be fully
4572          * VLAN hardware accelerated capable. Special
4573          * care is taken in the various xmit functions
4574          * when there are slaves that are not hw accel
4575          * capable
4576          */
4577         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4578         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4579         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4580         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4581                                NETIF_F_HW_VLAN_RX |
4582                                NETIF_F_HW_VLAN_FILTER);
4583
4584 #ifdef CONFIG_PROC_FS
4585         bond_create_proc_entry(bond);
4586 #endif
4587         list_add_tail(&bond->bond_list, &bond_dev_list);
4588
4589         return 0;
4590 }
4591
4592 /* De-initialize device specific data.
4593  * Caller must hold rtnl_lock.
4594  */
4595 static void bond_deinit(struct net_device *bond_dev)
4596 {
4597         struct bonding *bond = bond_dev->priv;
4598
4599         list_del(&bond->bond_list);
4600
4601 #ifdef CONFIG_PROC_FS
4602         bond_remove_proc_entry(bond);
4603 #endif
4604 }
4605
4606 static void bond_work_cancel_all(struct bonding *bond)
4607 {
4608         write_lock_bh(&bond->lock);
4609         bond->kill_timers = 1;
4610         write_unlock_bh(&bond->lock);
4611
4612         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4613                 cancel_delayed_work(&bond->mii_work);
4614
4615         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4616                 cancel_delayed_work(&bond->arp_work);
4617
4618         if (bond->params.mode == BOND_MODE_ALB &&
4619             delayed_work_pending(&bond->alb_work))
4620                 cancel_delayed_work(&bond->alb_work);
4621
4622         if (bond->params.mode == BOND_MODE_8023AD &&
4623             delayed_work_pending(&bond->ad_work))
4624                 cancel_delayed_work(&bond->ad_work);
4625 }
4626
4627 /* Unregister and free all bond devices.
4628  * Caller must hold rtnl_lock.
4629  */
4630 static void bond_free_all(void)
4631 {
4632         struct bonding *bond, *nxt;
4633
4634         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4635                 struct net_device *bond_dev = bond->dev;
4636
4637                 bond_work_cancel_all(bond);
4638                 netif_addr_lock_bh(bond_dev);
4639                 bond_mc_list_destroy(bond);
4640                 netif_addr_unlock_bh(bond_dev);
4641                 /* Release the bonded slaves */
4642                 bond_release_all(bond_dev);
4643                 bond_destroy(bond);
4644         }
4645
4646 #ifdef CONFIG_PROC_FS
4647         bond_destroy_proc_dir();
4648 #endif
4649 }
4650
4651 /*------------------------- Module initialization ---------------------------*/
4652
4653 /*
4654  * Convert string input module parms.  Accept either the
4655  * number of the mode or its string name.  A bit complicated because
4656  * some mode names are substrings of other names, and calls from sysfs
4657  * may have whitespace in the name (trailing newlines, for example).
4658  */
4659 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4660 {
4661         int mode = -1, i, rv;
4662         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4663
4664         for (p = (char *)buf; *p; p++)
4665                 if (!(isdigit(*p) || isspace(*p)))
4666                         break;
4667
4668         if (*p)
4669                 rv = sscanf(buf, "%20s", modestr);
4670         else
4671                 rv = sscanf(buf, "%d", &mode);
4672
4673         if (!rv)
4674                 return -1;
4675
4676         for (i = 0; tbl[i].modename; i++) {
4677                 if (mode == tbl[i].mode)
4678                         return tbl[i].mode;
4679                 if (strcmp(modestr, tbl[i].modename) == 0)
4680                         return tbl[i].mode;
4681         }
4682
4683         return -1;
4684 }
4685
4686 static int bond_check_params(struct bond_params *params)
4687 {
4688         int arp_validate_value, fail_over_mac_value;
4689
4690         /*
4691          * Convert string parameters.
4692          */
4693         if (mode) {
4694                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4695                 if (bond_mode == -1) {
4696                         printk(KERN_ERR DRV_NAME
4697                                ": Error: Invalid bonding mode \"%s\"\n",
4698                                mode == NULL ? "NULL" : mode);
4699                         return -EINVAL;
4700                 }
4701         }
4702
4703         if (xmit_hash_policy) {
4704                 if ((bond_mode != BOND_MODE_XOR) &&
4705                     (bond_mode != BOND_MODE_8023AD)) {
4706                         printk(KERN_INFO DRV_NAME
4707                                ": xor_mode param is irrelevant in mode %s\n",
4708                                bond_mode_name(bond_mode));
4709                 } else {
4710                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4711                                                         xmit_hashtype_tbl);
4712                         if (xmit_hashtype == -1) {
4713                                 printk(KERN_ERR DRV_NAME
4714                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4715                                 xmit_hash_policy == NULL ? "NULL" :
4716                                        xmit_hash_policy);
4717                                 return -EINVAL;
4718                         }
4719                 }
4720         }
4721
4722         if (lacp_rate) {
4723                 if (bond_mode != BOND_MODE_8023AD) {
4724                         printk(KERN_INFO DRV_NAME
4725                                ": lacp_rate param is irrelevant in mode %s\n",
4726                                bond_mode_name(bond_mode));
4727                 } else {
4728                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4729                         if (lacp_fast == -1) {
4730                                 printk(KERN_ERR DRV_NAME
4731                                        ": Error: Invalid lacp rate \"%s\"\n",
4732                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4733                                 return -EINVAL;
4734                         }
4735                 }
4736         }
4737
4738         if (max_bonds < 0 || max_bonds > INT_MAX) {
4739                 printk(KERN_WARNING DRV_NAME
4740                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4741                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4742                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4743                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4744         }
4745
4746         if (miimon < 0) {
4747                 printk(KERN_WARNING DRV_NAME
4748                        ": Warning: miimon module parameter (%d), "
4749                        "not in range 0-%d, so it was reset to %d\n",
4750                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4751                 miimon = BOND_LINK_MON_INTERV;
4752         }
4753
4754         if (updelay < 0) {
4755                 printk(KERN_WARNING DRV_NAME
4756                        ": Warning: updelay module parameter (%d), "
4757                        "not in range 0-%d, so it was reset to 0\n",
4758                        updelay, INT_MAX);
4759                 updelay = 0;
4760         }
4761
4762         if (downdelay < 0) {
4763                 printk(KERN_WARNING DRV_NAME
4764                        ": Warning: downdelay module parameter (%d), "
4765                        "not in range 0-%d, so it was reset to 0\n",
4766                        downdelay, INT_MAX);
4767                 downdelay = 0;
4768         }
4769
4770         if ((use_carrier != 0) && (use_carrier != 1)) {
4771                 printk(KERN_WARNING DRV_NAME
4772                        ": Warning: use_carrier module parameter (%d), "
4773                        "not of valid value (0/1), so it was set to 1\n",
4774                        use_carrier);
4775                 use_carrier = 1;
4776         }
4777
4778         if (num_grat_arp < 0 || num_grat_arp > 255) {
4779                 printk(KERN_WARNING DRV_NAME
4780                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4781                        "was reset to 1 \n", num_grat_arp);
4782                 num_grat_arp = 1;
4783         }
4784
4785         /* reset values for 802.3ad */
4786         if (bond_mode == BOND_MODE_8023AD) {
4787                 if (!miimon) {
4788                         printk(KERN_WARNING DRV_NAME
4789                                ": Warning: miimon must be specified, "
4790                                "otherwise bonding will not detect link "
4791                                "failure, speed and duplex which are "
4792                                "essential for 802.3ad operation\n");
4793                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4794                         miimon = 100;
4795                 }
4796         }
4797
4798         /* reset values for TLB/ALB */
4799         if ((bond_mode == BOND_MODE_TLB) ||
4800             (bond_mode == BOND_MODE_ALB)) {
4801                 if (!miimon) {
4802                         printk(KERN_WARNING DRV_NAME
4803                                ": Warning: miimon must be specified, "
4804                                "otherwise bonding will not detect link "
4805                                "failure and link speed which are essential "
4806                                "for TLB/ALB load balancing\n");
4807                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4808                         miimon = 100;
4809                 }
4810         }
4811
4812         if (bond_mode == BOND_MODE_ALB) {
4813                 printk(KERN_NOTICE DRV_NAME
4814                        ": In ALB mode you might experience client "
4815                        "disconnections upon reconnection of a link if the "
4816                        "bonding module updelay parameter (%d msec) is "
4817                        "incompatible with the forwarding delay time of the "
4818                        "switch\n",
4819                        updelay);
4820         }
4821
4822         if (!miimon) {
4823                 if (updelay || downdelay) {
4824                         /* just warn the user the up/down delay will have
4825                          * no effect since miimon is zero...
4826                          */
4827                         printk(KERN_WARNING DRV_NAME
4828                                ": Warning: miimon module parameter not set "
4829                                "and updelay (%d) or downdelay (%d) module "
4830                                "parameter is set; updelay and downdelay have "
4831                                "no effect unless miimon is set\n",
4832                                updelay, downdelay);
4833                 }
4834         } else {
4835                 /* don't allow arp monitoring */
4836                 if (arp_interval) {
4837                         printk(KERN_WARNING DRV_NAME
4838                                ": Warning: miimon (%d) and arp_interval (%d) "
4839                                "can't be used simultaneously, disabling ARP "
4840                                "monitoring\n",
4841                                miimon, arp_interval);
4842                         arp_interval = 0;
4843                 }
4844
4845                 if ((updelay % miimon) != 0) {
4846                         printk(KERN_WARNING DRV_NAME
4847                                ": Warning: updelay (%d) is not a multiple "
4848                                "of miimon (%d), updelay rounded to %d ms\n",
4849                                updelay, miimon, (updelay / miimon) * miimon);
4850                 }
4851
4852                 updelay /= miimon;
4853
4854                 if ((downdelay % miimon) != 0) {
4855                         printk(KERN_WARNING DRV_NAME
4856                                ": Warning: downdelay (%d) is not a multiple "
4857                                "of miimon (%d), downdelay rounded to %d ms\n",
4858                                downdelay, miimon,
4859                                (downdelay / miimon) * miimon);
4860                 }
4861
4862                 downdelay /= miimon;
4863         }
4864
4865         if (arp_interval < 0) {
4866                 printk(KERN_WARNING DRV_NAME
4867                        ": Warning: arp_interval module parameter (%d) "
4868                        ", not in range 0-%d, so it was reset to %d\n",
4869                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4870                 arp_interval = BOND_LINK_ARP_INTERV;
4871         }
4872
4873         for (arp_ip_count = 0;
4874              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4875              arp_ip_count++) {
4876                 /* not complete check, but should be good enough to
4877                    catch mistakes */
4878                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4879                         printk(KERN_WARNING DRV_NAME
4880                                ": Warning: bad arp_ip_target module parameter "
4881                                "(%s), ARP monitoring will not be performed\n",
4882                                arp_ip_target[arp_ip_count]);
4883                         arp_interval = 0;
4884                 } else {
4885                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4886                         arp_target[arp_ip_count] = ip;
4887                 }
4888         }
4889
4890         if (arp_interval && !arp_ip_count) {
4891                 /* don't allow arping if no arp_ip_target given... */
4892                 printk(KERN_WARNING DRV_NAME
4893                        ": Warning: arp_interval module parameter (%d) "
4894                        "specified without providing an arp_ip_target "
4895                        "parameter, arp_interval was reset to 0\n",
4896                        arp_interval);
4897                 arp_interval = 0;
4898         }
4899
4900         if (arp_validate) {
4901                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4902                         printk(KERN_ERR DRV_NAME
4903                ": arp_validate only supported in active-backup mode\n");
4904                         return -EINVAL;
4905                 }
4906                 if (!arp_interval) {
4907                         printk(KERN_ERR DRV_NAME
4908                                ": arp_validate requires arp_interval\n");
4909                         return -EINVAL;
4910                 }
4911
4912                 arp_validate_value = bond_parse_parm(arp_validate,
4913                                                      arp_validate_tbl);
4914                 if (arp_validate_value == -1) {
4915                         printk(KERN_ERR DRV_NAME
4916                                ": Error: invalid arp_validate \"%s\"\n",
4917                                arp_validate == NULL ? "NULL" : arp_validate);
4918                         return -EINVAL;
4919                 }
4920         } else
4921                 arp_validate_value = 0;
4922
4923         if (miimon) {
4924                 printk(KERN_INFO DRV_NAME
4925                        ": MII link monitoring set to %d ms\n",
4926                        miimon);
4927         } else if (arp_interval) {
4928                 int i;
4929
4930                 printk(KERN_INFO DRV_NAME
4931                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4932                        arp_interval,
4933                        arp_validate_tbl[arp_validate_value].modename,
4934                        arp_ip_count);
4935
4936                 for (i = 0; i < arp_ip_count; i++)
4937                         printk (" %s", arp_ip_target[i]);
4938
4939                 printk("\n");
4940
4941         } else if (max_bonds) {
4942                 /* miimon and arp_interval not set, we need one so things
4943                  * work as expected, see bonding.txt for details
4944                  */
4945                 printk(KERN_WARNING DRV_NAME
4946                        ": Warning: either miimon or arp_interval and "
4947                        "arp_ip_target module parameters must be specified, "
4948                        "otherwise bonding will not detect link failures! see "
4949                        "bonding.txt for details.\n");
4950         }
4951
4952         if (primary && !USES_PRIMARY(bond_mode)) {
4953                 /* currently, using a primary only makes sense
4954                  * in active backup, TLB or ALB modes
4955                  */
4956                 printk(KERN_WARNING DRV_NAME
4957                        ": Warning: %s primary device specified but has no "
4958                        "effect in %s mode\n",
4959                        primary, bond_mode_name(bond_mode));
4960                 primary = NULL;
4961         }
4962
4963         if (fail_over_mac) {
4964                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4965                                                       fail_over_mac_tbl);
4966                 if (fail_over_mac_value == -1) {
4967                         printk(KERN_ERR DRV_NAME
4968                                ": Error: invalid fail_over_mac \"%s\"\n",
4969                                arp_validate == NULL ? "NULL" : arp_validate);
4970                         return -EINVAL;
4971                 }
4972
4973                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4974                         printk(KERN_WARNING DRV_NAME
4975                                ": Warning: fail_over_mac only affects "
4976                                "active-backup mode.\n");
4977         } else {
4978                 fail_over_mac_value = BOND_FOM_NONE;
4979         }
4980
4981         /* fill params struct with the proper values */
4982         params->mode = bond_mode;
4983         params->xmit_policy = xmit_hashtype;
4984         params->miimon = miimon;
4985         params->num_grat_arp = num_grat_arp;
4986         params->arp_interval = arp_interval;
4987         params->arp_validate = arp_validate_value;
4988         params->updelay = updelay;
4989         params->downdelay = downdelay;
4990         params->use_carrier = use_carrier;
4991         params->lacp_fast = lacp_fast;
4992         params->primary[0] = 0;
4993         params->fail_over_mac = fail_over_mac_value;
4994
4995         if (primary) {
4996                 strncpy(params->primary, primary, IFNAMSIZ);
4997                 params->primary[IFNAMSIZ - 1] = 0;
4998         }
4999
5000         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5001
5002         return 0;
5003 }
5004
5005 static struct lock_class_key bonding_netdev_xmit_lock_key;
5006 static struct lock_class_key bonding_netdev_addr_lock_key;
5007
5008 static void bond_set_lockdep_class_one(struct net_device *dev,
5009                                        struct netdev_queue *txq,
5010                                        void *_unused)
5011 {
5012         lockdep_set_class(&txq->_xmit_lock,
5013                           &bonding_netdev_xmit_lock_key);
5014 }
5015
5016 static void bond_set_lockdep_class(struct net_device *dev)
5017 {
5018         lockdep_set_class(&dev->addr_list_lock,
5019                           &bonding_netdev_addr_lock_key);
5020         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5021 }
5022
5023 /* Create a new bond based on the specified name and bonding parameters.
5024  * If name is NULL, obtain a suitable "bond%d" name for us.
5025  * Caller must NOT hold rtnl_lock; we need to release it here before we
5026  * set up our sysfs entries.
5027  */
5028 int bond_create(char *name, struct bond_params *params)
5029 {
5030         struct net_device *bond_dev;
5031         struct bonding *bond;
5032         int res;
5033
5034         rtnl_lock();
5035         down_write(&bonding_rwsem);
5036
5037         /* Check to see if the bond already exists. */
5038         if (name) {
5039                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5040                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5041                                 printk(KERN_ERR DRV_NAME
5042                                ": cannot add bond %s; it already exists\n",
5043                                        name);
5044                                 res = -EPERM;
5045                                 goto out_rtnl;
5046                         }
5047         }
5048
5049         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5050                                 ether_setup);
5051         if (!bond_dev) {
5052                 printk(KERN_ERR DRV_NAME
5053                        ": %s: eek! can't alloc netdev!\n",
5054                        name);
5055                 res = -ENOMEM;
5056                 goto out_rtnl;
5057         }
5058
5059         if (!name) {
5060                 res = dev_alloc_name(bond_dev, "bond%d");
5061                 if (res < 0)
5062                         goto out_netdev;
5063         }
5064
5065         /* bond_init() must be called after dev_alloc_name() (for the
5066          * /proc files), but before register_netdevice(), because we
5067          * need to set function pointers.
5068          */
5069
5070         res = bond_init(bond_dev, params);
5071         if (res < 0) {
5072                 goto out_netdev;
5073         }
5074
5075         res = register_netdevice(bond_dev);
5076         if (res < 0) {
5077                 goto out_bond;
5078         }
5079
5080         bond_set_lockdep_class(bond_dev);
5081
5082         netif_carrier_off(bond_dev);
5083
5084         up_write(&bonding_rwsem);
5085         rtnl_unlock(); /* allows sysfs registration of net device */
5086         res = bond_create_sysfs_entry(bond_dev->priv);
5087         if (res < 0) {
5088                 rtnl_lock();
5089                 down_write(&bonding_rwsem);
5090                 bond_deinit(bond_dev);
5091                 unregister_netdevice(bond_dev);
5092                 goto out_rtnl;
5093         }
5094
5095         return 0;
5096
5097 out_bond:
5098         bond_deinit(bond_dev);
5099 out_netdev:
5100         free_netdev(bond_dev);
5101 out_rtnl:
5102         up_write(&bonding_rwsem);
5103         rtnl_unlock();
5104         return res;
5105 }
5106
5107 static int __init bonding_init(void)
5108 {
5109         int i;
5110         int res;
5111         struct bonding *bond;
5112
5113         printk(KERN_INFO "%s", version);
5114
5115         res = bond_check_params(&bonding_defaults);
5116         if (res) {
5117                 goto out;
5118         }
5119
5120 #ifdef CONFIG_PROC_FS
5121         bond_create_proc_dir();
5122 #endif
5123
5124         init_rwsem(&bonding_rwsem);
5125
5126         for (i = 0; i < max_bonds; i++) {
5127                 res = bond_create(NULL, &bonding_defaults);
5128                 if (res)
5129                         goto err;
5130         }
5131
5132         res = bond_create_sysfs();
5133         if (res)
5134                 goto err;
5135
5136         register_netdevice_notifier(&bond_netdev_notifier);
5137         register_inetaddr_notifier(&bond_inetaddr_notifier);
5138
5139         goto out;
5140 err:
5141         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5142                 bond_work_cancel_all(bond);
5143                 destroy_workqueue(bond->wq);
5144         }
5145
5146         bond_destroy_sysfs();
5147
5148         rtnl_lock();
5149         bond_free_all();
5150         rtnl_unlock();
5151 out:
5152         return res;
5153
5154 }
5155
5156 static void __exit bonding_exit(void)
5157 {
5158         unregister_netdevice_notifier(&bond_netdev_notifier);
5159         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5160
5161         bond_destroy_sysfs();
5162
5163         rtnl_lock();
5164         bond_free_all();
5165         rtnl_unlock();
5166 }
5167
5168 module_init(bonding_init);
5169 module_exit(bonding_exit);
5170 MODULE_LICENSE("GPL");
5171 MODULE_VERSION(DRV_VERSION);
5172 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5173 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5174 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5175
5176 /*
5177  * Local variables:
5178  *  c-indent-level: 8
5179  *  c-basic-offset: 8
5180  *  tab-width: 8
5181  * End:
5182  */
5183