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 pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  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 <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/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 <net/netns/generic.h>
  81 #include "bonding.h"
  82 #include "bond_3ad.h"
  83 #include "bond_alb.h"
  84
  85 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88 #define BOND_LINK_MON_INTERV    0
  89 #define BOND_LINK_ARP_INTERV    0
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int num_grat_arp = 1;
  93 static int num_unsol_na = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static struct bond_params bonding_defaults;
 109
 110 module_param(max_bonds, int, 0);
 111 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 112 module_param(num_grat_arp, int, 0644);
 113 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 114 module_param(num_unsol_na, int, 0644);
 115 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 116 module_param(miimon, int, 0);
 117 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 118 module_param(updelay, int, 0);
 119 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 120 module_param(downdelay, int, 0);
 121 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 122                             "in milliseconds");
 123 module_param(use_carrier, int, 0);
 124 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 125                               "0 for off, 1 for on (default)");
 126 module_param(mode, charp, 0);
 127 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 128                        "1 for active-backup, 2 for balance-xor, "
 129                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 130                        "6 for balance-alb");
 131 module_param(primary, charp, 0);
 132 MODULE_PARM_DESC(primary, "Primary network device to use");
 133 module_param(primary_reselect, charp, 0);
 134 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 135                                    "once it comes up; "
 136                                    "0 for always (default), "
 137                                    "1 for only if speed of primary is "
 138                                    "better, "
 139                                    "2 for only on active slave "
 140                                    "failure");
 141 module_param(lacp_rate, charp, 0);
 142 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 143                             "(slow/fast)");
 144 module_param(ad_select, charp, 0);
 145 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 146 module_param(xmit_hash_policy, charp, 0);
 147 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 148                                    ", 1 for layer 3+4");
 149 module_param(arp_interval, int, 0);
 150 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 151 module_param_array(arp_ip_target, charp, NULL, 0);
 152 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 153 module_param(arp_validate, charp, 0);
 154 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 155 module_param(fail_over_mac, charp, 0);
 156 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 157
 158 /*----------------------------- Global variables ----------------------------*/
 159
 160 static const char * const version =
 161         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 162
 163 int bond_net_id __read_mostly;
 164
 165 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 166 static int arp_ip_count;
 167 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 168 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 169 static int lacp_fast;
 170
 171
 172 const struct bond_parm_tbl bond_lacp_tbl[] = {
 173 {       "slow",         AD_LACP_SLOW},
 174 {       "fast",         AD_LACP_FAST},
 175 {       NULL,           -1},
 176 };
 177
 178 const struct bond_parm_tbl bond_mode_tbl[] = {
 179 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 180 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 181 {       "balance-xor",          BOND_MODE_XOR},
 182 {       "broadcast",            BOND_MODE_BROADCAST},
 183 {       "802.3ad",              BOND_MODE_8023AD},
 184 {       "balance-tlb",          BOND_MODE_TLB},
 185 {       "balance-alb",          BOND_MODE_ALB},
 186 {       NULL,                   -1},
 187 };
 188
 189 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 190 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 191 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 192 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 193 {       NULL,                   -1},
 194 };
 195
 196 const struct bond_parm_tbl arp_validate_tbl[] = {
 197 {       "none",                 BOND_ARP_VALIDATE_NONE},
 198 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 199 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 200 {       "all",                  BOND_ARP_VALIDATE_ALL},
 201 {       NULL,                   -1},
 202 };
 203
 204 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 205 {       "none",                 BOND_FOM_NONE},
 206 {       "active",               BOND_FOM_ACTIVE},
 207 {       "follow",               BOND_FOM_FOLLOW},
 208 {       NULL,                   -1},
 209 };
 210
 211 const struct bond_parm_tbl pri_reselect_tbl[] = {
 212 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 213 {       "better",               BOND_PRI_RESELECT_BETTER},
 214 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 215 {       NULL,                   -1},
 216 };
 217
 218 struct bond_parm_tbl ad_select_tbl[] = {
 219 {       "stable",       BOND_AD_STABLE},
 220 {       "bandwidth",    BOND_AD_BANDWIDTH},
 221 {       "count",        BOND_AD_COUNT},
 222 {       NULL,           -1},
 223 };
 224
 225 /*-------------------------- Forward declarations ---------------------------*/
 226
 227 static void bond_send_gratuitous_arp(struct bonding *bond);
 228 static int bond_init(struct net_device *bond_dev);
 229 static void bond_uninit(struct net_device *bond_dev);
 230
 231 /*---------------------------- General routines -----------------------------*/
 232
 233 static const char *bond_mode_name(int mode)
 234 {
 235         static const char *names[] = {
 236                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 237                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 238                 [BOND_MODE_XOR] = "load balancing (xor)",
 239                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 240                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 241                 [BOND_MODE_TLB] = "transmit load balancing",
 242                 [BOND_MODE_ALB] = "adaptive load balancing",
 243         };
 244
 245         if (mode < 0 || mode > BOND_MODE_ALB)
 246                 return "unknown";
 247
 248         return names[mode];
 249 }
 250
 251 /*---------------------------------- VLAN -----------------------------------*/
 252
 253 /**
 254  * bond_add_vlan - add a new vlan id on bond
 255  * @bond: bond that got the notification
 256  * @vlan_id: the vlan id to add
 257  *
 258  * Returns -ENOMEM if allocation failed.
 259  */
 260 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 261 {
 262         struct vlan_entry *vlan;
 263
 264         pr_debug("bond: %s, vlan id %d\n",
 265                  (bond ? bond->dev->name : "None"), vlan_id);
 266
 267         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 268         if (!vlan)
 269                 return -ENOMEM;
 270
 271         INIT_LIST_HEAD(&vlan->vlan_list);
 272         vlan->vlan_id = vlan_id;
 273
 274         write_lock_bh(&bond->lock);
 275
 276         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 277
 278         write_unlock_bh(&bond->lock);
 279
 280         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 281
 282         return 0;
 283 }
 284
 285 /**
 286  * bond_del_vlan - delete a vlan id from bond
 287  * @bond: bond that got the notification
 288  * @vlan_id: the vlan id to delete
 289  *
 290  * returns -ENODEV if @vlan_id was not found in @bond.
 291  */
 292 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 293 {
 294         struct vlan_entry *vlan;
 295         int res = -ENODEV;
 296
 297         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 298
 299         write_lock_bh(&bond->lock);
 300
 301         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 302                 if (vlan->vlan_id == vlan_id) {
 303                         list_del(&vlan->vlan_list);
 304
 305                         if (bond_is_lb(bond))
 306                                 bond_alb_clear_vlan(bond, vlan_id);
 307
 308                         pr_debug("removed VLAN ID %d from bond %s\n",
 309                                  vlan_id, bond->dev->name);
 310
 311                         kfree(vlan);
 312
 313                         if (list_empty(&bond->vlan_list) &&
 314                             (bond->slave_cnt == 0)) {
 315                                 /* Last VLAN removed and no slaves, so
 316                                  * restore block on adding VLANs. This will
 317                                  * be removed once new slaves that are not
 318                                  * VLAN challenged will be added.
 319                                  */
 320                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 321                         }
 322
 323                         res = 0;
 324                         goto out;
 325                 }
 326         }
 327
 328         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 329                  vlan_id, bond->dev->name);
 330
 331 out:
 332         write_unlock_bh(&bond->lock);
 333         return res;
 334 }
 335
 336 /**
 337  * bond_has_challenged_slaves
 338  * @bond: the bond we're working on
 339  *
 340  * Searches the slave list. Returns 1 if a vlan challenged slave
 341  * was found, 0 otherwise.
 342  *
 343  * Assumes bond->lock is held.
 344  */
 345 static int bond_has_challenged_slaves(struct bonding *bond)
 346 {
 347         struct slave *slave;
 348         int i;
 349
 350         bond_for_each_slave(bond, slave, i) {
 351                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 352                         pr_debug("found VLAN challenged slave - %s\n",
 353                                  slave->dev->name);
 354                         return 1;
 355                 }
 356         }
 357
 358         pr_debug("no VLAN challenged slaves found\n");
 359         return 0;
 360 }
 361
 362 /**
 363  * bond_next_vlan - safely skip to the next item in the vlans list.
 364  * @bond: the bond we're working on
 365  * @curr: item we're advancing from
 366  *
 367  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 368  * or @curr->next otherwise (even if it is @curr itself again).
 369  *
 370  * Caller must hold bond->lock
 371  */
 372 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 373 {
 374         struct vlan_entry *next, *last;
 375
 376         if (list_empty(&bond->vlan_list))
 377                 return NULL;
 378
 379         if (!curr) {
 380                 next = list_entry(bond->vlan_list.next,
 381                                   struct vlan_entry, vlan_list);
 382         } else {
 383                 last = list_entry(bond->vlan_list.prev,
 384                                   struct vlan_entry, vlan_list);
 385                 if (last == curr) {
 386                         next = list_entry(bond->vlan_list.next,
 387                                           struct vlan_entry, vlan_list);
 388                 } else {
 389                         next = list_entry(curr->vlan_list.next,
 390                                           struct vlan_entry, vlan_list);
 391                 }
 392         }
 393
 394         return next;
 395 }
 396
 397 /**
 398  * bond_dev_queue_xmit - Prepare skb for xmit.
 399  *
 400  * @bond: bond device that got this skb for tx.
 401  * @skb: hw accel VLAN tagged skb to transmit
 402  * @slave_dev: slave that is supposed to xmit this skbuff
 403  *
 404  * When the bond gets an skb to transmit that is
 405  * already hardware accelerated VLAN tagged, and it
 406  * needs to relay this skb to a slave that is not
 407  * hw accel capable, the skb needs to be "unaccelerated",
 408  * i.e. strip the hwaccel tag and re-insert it as part
 409  * of the payload.
 410  */
 411 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 412                         struct net_device *slave_dev)
 413 {
 414         unsigned short uninitialized_var(vlan_id);
 415
 416         if (!list_empty(&bond->vlan_list) &&
 417             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 418             vlan_get_tag(skb, &vlan_id) == 0) {
 419                 skb->dev = slave_dev;
 420                 skb = vlan_put_tag(skb, vlan_id);
 421                 if (!skb) {
 422                         /* vlan_put_tag() frees the skb in case of error,
 423                          * so return success here so the calling functions
 424                          * won't attempt to free is again.
 425                          */
 426                         return 0;
 427                 }
 428         } else {
 429                 skb->dev = slave_dev;
 430         }
 431
 432         skb->priority = 1;
 433         dev_queue_xmit(skb);
 434
 435         return 0;
 436 }
 437
 438 /*
 439  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 440  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 441  * lock because:
 442  * a. This operation is performed in IOCTL context,
 443  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 444  * c. Holding a lock with BH disabled while directly calling a base driver
 445  *    entry point is generally a BAD idea.
 446  *
 447  * The design of synchronization/protection for this operation in the 8021q
 448  * module is good for one or more VLAN devices over a single physical device
 449  * and cannot be extended for a teaming solution like bonding, so there is a
 450  * potential race condition here where a net device from the vlan group might
 451  * be referenced (either by a base driver or the 8021q code) while it is being
 452  * removed from the system. However, it turns out we're not making matters
 453  * worse, and if it works for regular VLAN usage it will work here too.
 454 */
 455
 456 /**
 457  * bond_vlan_rx_register - Propagates registration to slaves
 458  * @bond_dev: bonding net device that got called
 459  * @grp: vlan group being registered
 460  */
 461 static void bond_vlan_rx_register(struct net_device *bond_dev,
 462                                   struct vlan_group *grp)
 463 {
 464         struct bonding *bond = netdev_priv(bond_dev);
 465         struct slave *slave;
 466         int i;
 467
 468         bond->vlgrp = grp;
 469
 470         bond_for_each_slave(bond, slave, i) {
 471                 struct net_device *slave_dev = slave->dev;
 472                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 473
 474                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 475                     slave_ops->ndo_vlan_rx_register) {
 476                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 477                 }
 478         }
 479 }
 480
 481 /**
 482  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 483  * @bond_dev: bonding net device that got called
 484  * @vid: vlan id being added
 485  */
 486 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 487 {
 488         struct bonding *bond = netdev_priv(bond_dev);
 489         struct slave *slave;
 490         int i, res;
 491
 492         bond_for_each_slave(bond, slave, i) {
 493                 struct net_device *slave_dev = slave->dev;
 494                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 495
 496                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 497                     slave_ops->ndo_vlan_rx_add_vid) {
 498                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 499                 }
 500         }
 501
 502         res = bond_add_vlan(bond, vid);
 503         if (res) {
 504                 pr_err("%s: Error: Failed to add vlan id %d\n",
 505                        bond_dev->name, vid);
 506         }
 507 }
 508
 509 /**
 510  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 511  * @bond_dev: bonding net device that got called
 512  * @vid: vlan id being removed
 513  */
 514 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 515 {
 516         struct bonding *bond = netdev_priv(bond_dev);
 517         struct slave *slave;
 518         struct net_device *vlan_dev;
 519         int i, res;
 520
 521         bond_for_each_slave(bond, slave, i) {
 522                 struct net_device *slave_dev = slave->dev;
 523                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 524
 525                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 526                     slave_ops->ndo_vlan_rx_kill_vid) {
 527                         /* Save and then restore vlan_dev in the grp array,
 528                          * since the slave's driver might clear it.
 529                          */
 530                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 531                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 532                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 533                 }
 534         }
 535
 536         res = bond_del_vlan(bond, vid);
 537         if (res) {
 538                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 539                        bond_dev->name, vid);
 540         }
 541 }
 542
 543 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 544 {
 545         struct vlan_entry *vlan;
 546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 547
 548         write_lock_bh(&bond->lock);
 549
 550         if (list_empty(&bond->vlan_list))
 551                 goto out;
 552
 553         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 554             slave_ops->ndo_vlan_rx_register)
 555                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 556
 557         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 558             !(slave_ops->ndo_vlan_rx_add_vid))
 559                 goto out;
 560
 561         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 562                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 563
 564 out:
 565         write_unlock_bh(&bond->lock);
 566 }
 567
 568 static void bond_del_vlans_from_slave(struct bonding *bond,
 569                                       struct net_device *slave_dev)
 570 {
 571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 572         struct vlan_entry *vlan;
 573         struct net_device *vlan_dev;
 574
 575         write_lock_bh(&bond->lock);
 576
 577         if (list_empty(&bond->vlan_list))
 578                 goto out;
 579
 580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 581             !(slave_ops->ndo_vlan_rx_kill_vid))
 582                 goto unreg;
 583
 584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 585                 /* Save and then restore vlan_dev in the grp array,
 586                  * since the slave's driver might clear it.
 587                  */
 588                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 589                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 590                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 591         }
 592
 593 unreg:
 594         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 595             slave_ops->ndo_vlan_rx_register)
 596                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 597
 598 out:
 599         write_unlock_bh(&bond->lock);
 600 }
 601
 602 /*------------------------------- Link status -------------------------------*/
 603
 604 /*
 605  * Set the carrier state for the master according to the state of its
 606  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 607  * do special 802.3ad magic.
 608  *
 609  * Returns zero if carrier state does not change, nonzero if it does.
 610  */
 611 static int bond_set_carrier(struct bonding *bond)
 612 {
 613         struct slave *slave;
 614         int i;
 615
 616         if (bond->slave_cnt == 0)
 617                 goto down;
 618
 619         if (bond->params.mode == BOND_MODE_8023AD)
 620                 return bond_3ad_set_carrier(bond);
 621
 622         bond_for_each_slave(bond, slave, i) {
 623                 if (slave->link == BOND_LINK_UP) {
 624                         if (!netif_carrier_ok(bond->dev)) {
 625                                 netif_carrier_on(bond->dev);
 626                                 return 1;
 627                         }
 628                         return 0;
 629                 }
 630         }
 631
 632 down:
 633         if (netif_carrier_ok(bond->dev)) {
 634                 netif_carrier_off(bond->dev);
 635                 return 1;
 636         }
 637         return 0;
 638 }
 639
 640 /*
 641  * Get link speed and duplex from the slave's base driver
 642  * using ethtool. If for some reason the call fails or the
 643  * values are invalid, fake speed and duplex to 100/Full
 644  * and return error.
 645  */
 646 static int bond_update_speed_duplex(struct slave *slave)
 647 {
 648         struct net_device *slave_dev = slave->dev;
 649         struct ethtool_cmd etool;
 650         int res;
 651
 652         /* Fake speed and duplex */
 653         slave->speed = SPEED_100;
 654         slave->duplex = DUPLEX_FULL;
 655
 656         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 657                 return -1;
 658
 659         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 660         if (res < 0)
 661                 return -1;
 662
 663         switch (etool.speed) {
 664         case SPEED_10:
 665         case SPEED_100:
 666         case SPEED_1000:
 667         case SPEED_10000:
 668                 break;
 669         default:
 670                 return -1;
 671         }
 672
 673         switch (etool.duplex) {
 674         case DUPLEX_FULL:
 675         case DUPLEX_HALF:
 676                 break;
 677         default:
 678                 return -1;
 679         }
 680
 681         slave->speed = etool.speed;
 682         slave->duplex = etool.duplex;
 683
 684         return 0;
 685 }
 686
 687 /*
 688  * if <dev> supports MII link status reporting, check its link status.
 689  *
 690  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 691  * depending upon the setting of the use_carrier parameter.
 692  *
 693  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 694  * can't tell and just pretend it is), or 0, meaning that the link is
 695  * down.
 696  *
 697  * If reporting is non-zero, instead of faking link up, return -1 if
 698  * both ETHTOOL and MII ioctls fail (meaning the device does not
 699  * support them).  If use_carrier is set, return whatever it says.
 700  * It'd be nice if there was a good way to tell if a driver supports
 701  * netif_carrier, but there really isn't.
 702  */
 703 static int bond_check_dev_link(struct bonding *bond,
 704                                struct net_device *slave_dev, int reporting)
 705 {
 706         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 707         int (*ioctl)(struct net_device *, struct ifreq *, int);
 708         struct ifreq ifr;
 709         struct mii_ioctl_data *mii;
 710
 711         if (!reporting && !netif_running(slave_dev))
 712                 return 0;
 713
 714         if (bond->params.use_carrier)
 715                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 716
 717         /* Try to get link status using Ethtool first. */
 718         if (slave_dev->ethtool_ops) {
 719                 if (slave_dev->ethtool_ops->get_link) {
 720                         u32 link;
 721
 722                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 723
 724                         return link ? BMSR_LSTATUS : 0;
 725                 }
 726         }
 727
 728         /* Ethtool can't be used, fallback to MII ioctls. */
 729         ioctl = slave_ops->ndo_do_ioctl;
 730         if (ioctl) {
 731                 /* TODO: set pointer to correct ioctl on a per team member */
 732                 /*       bases to make this more efficient. that is, once  */
 733                 /*       we determine the correct ioctl, we will always    */
 734                 /*       call it and not the others for that team          */
 735                 /*       member.                                           */
 736
 737                 /*
 738                  * We cannot assume that SIOCGMIIPHY will also read a
 739                  * register; not all network drivers (e.g., e100)
 740                  * support that.
 741                  */
 742
 743                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 744                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 745                 mii = if_mii(&ifr);
 746                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 747                         mii->reg_num = MII_BMSR;
 748                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 749                                 return mii->val_out & BMSR_LSTATUS;
 750                 }
 751         }
 752
 753         /*
 754          * If reporting, report that either there's no dev->do_ioctl,
 755          * or both SIOCGMIIREG and get_link failed (meaning that we
 756          * cannot report link status).  If not reporting, pretend
 757          * we're ok.
 758          */
 759         return reporting ? -1 : BMSR_LSTATUS;
 760 }
 761
 762 /*----------------------------- Multicast list ------------------------------*/
 763
 764 /*
 765  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 766  */
 767 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
 768                                    const struct dev_mc_list *dmi2)
 769 {
 770         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 771                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 772 }
 773
 774 /*
 775  * returns dmi entry if found, NULL otherwise
 776  */
 777 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
 778                                                  struct dev_mc_list *mc_list)
 779 {
 780         struct dev_mc_list *idmi;
 781
 782         for (idmi = mc_list; idmi; idmi = idmi->next) {
 783                 if (bond_is_dmi_same(dmi, idmi))
 784                         return idmi;
 785         }
 786
 787         return NULL;
 788 }
 789
 790 /*
 791  * Push the promiscuity flag down to appropriate slaves
 792  */
 793 static int bond_set_promiscuity(struct bonding *bond, int inc)
 794 {
 795         int err = 0;
 796         if (USES_PRIMARY(bond->params.mode)) {
 797                 /* write lock already acquired */
 798                 if (bond->curr_active_slave) {
 799                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 800                                                   inc);
 801                 }
 802         } else {
 803                 struct slave *slave;
 804                 int i;
 805                 bond_for_each_slave(bond, slave, i) {
 806                         err = dev_set_promiscuity(slave->dev, inc);
 807                         if (err)
 808                                 return err;
 809                 }
 810         }
 811         return err;
 812 }
 813
 814 /*
 815  * Push the allmulti flag down to all slaves
 816  */
 817 static int bond_set_allmulti(struct bonding *bond, int inc)
 818 {
 819         int err = 0;
 820         if (USES_PRIMARY(bond->params.mode)) {
 821                 /* write lock already acquired */
 822                 if (bond->curr_active_slave) {
 823                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 824                                                inc);
 825                 }
 826         } else {
 827                 struct slave *slave;
 828                 int i;
 829                 bond_for_each_slave(bond, slave, i) {
 830                         err = dev_set_allmulti(slave->dev, inc);
 831                         if (err)
 832                                 return err;
 833                 }
 834         }
 835         return err;
 836 }
 837
 838 /*
 839  * Add a Multicast address to slaves
 840  * according to mode
 841  */
 842 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 843 {
 844         if (USES_PRIMARY(bond->params.mode)) {
 845                 /* write lock already acquired */
 846                 if (bond->curr_active_slave)
 847                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 848         } else {
 849                 struct slave *slave;
 850                 int i;
 851
 852                 bond_for_each_slave(bond, slave, i)
 853                         dev_mc_add(slave->dev, addr, alen, 0);
 854         }
 855 }
 856
 857 /*
 858  * Remove a multicast address from slave
 859  * according to mode
 860  */
 861 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 862 {
 863         if (USES_PRIMARY(bond->params.mode)) {
 864                 /* write lock already acquired */
 865                 if (bond->curr_active_slave)
 866                         dev_mc_delete(bond->curr_active_slave->dev, addr,
 867                                       alen, 0);
 868         } else {
 869                 struct slave *slave;
 870                 int i;
 871                 bond_for_each_slave(bond, slave, i) {
 872                         dev_mc_delete(slave->dev, addr, alen, 0);
 873                 }
 874         }
 875 }
 876
 877
 878 /*
 879  * Retrieve the list of registered multicast addresses for the bonding
 880  * device and retransmit an IGMP JOIN request to the current active
 881  * slave.
 882  */
 883 static void bond_resend_igmp_join_requests(struct bonding *bond)
 884 {
 885         struct in_device *in_dev;
 886         struct ip_mc_list *im;
 887
 888         rcu_read_lock();
 889         in_dev = __in_dev_get_rcu(bond->dev);
 890         if (in_dev) {
 891                 for (im = in_dev->mc_list; im; im = im->next)
 892                         ip_mc_rejoin_group(im);
 893         }
 894
 895         rcu_read_unlock();
 896 }
 897
 898 /*
 899  * Totally destroys the mc_list in bond
 900  */
 901 static void bond_mc_list_destroy(struct bonding *bond)
 902 {
 903         struct dev_mc_list *dmi;
 904
 905         dmi = bond->mc_list;
 906         while (dmi) {
 907                 bond->mc_list = dmi->next;
 908                 kfree(dmi);
 909                 dmi = bond->mc_list;
 910         }
 911
 912         bond->mc_list = NULL;
 913 }
 914
 915 /*
 916  * Copy all the Multicast addresses from src to the bonding device dst
 917  */
 918 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 919                              gfp_t gfp_flag)
 920 {
 921         struct dev_mc_list *dmi, *new_dmi;
 922
 923         for (dmi = mc_list; dmi; dmi = dmi->next) {
 924                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 925
 926                 if (!new_dmi) {
 927                         /* FIXME: Potential memory leak !!! */
 928                         return -ENOMEM;
 929                 }
 930
 931                 new_dmi->next = bond->mc_list;
 932                 bond->mc_list = new_dmi;
 933                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 934                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 935                 new_dmi->dmi_users = dmi->dmi_users;
 936                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 937         }
 938
 939         return 0;
 940 }
 941
 942 /*
 943  * flush all members of flush->mc_list from device dev->mc_list
 944  */
 945 static void bond_mc_list_flush(struct net_device *bond_dev,
 946                                struct net_device *slave_dev)
 947 {
 948         struct bonding *bond = netdev_priv(bond_dev);
 949         struct dev_mc_list *dmi;
 950
 951         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
 952                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 953
 954         if (bond->params.mode == BOND_MODE_8023AD) {
 955                 /* del lacpdu mc addr from mc list */
 956                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 957
 958                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 959         }
 960 }
 961
 962 /*--------------------------- Active slave change ---------------------------*/
 963
 964 /*
 965  * Update the mc list and multicast-related flags for the new and
 966  * old active slaves (if any) according to the multicast mode, and
 967  * promiscuous flags unconditionally.
 968  */
 969 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 970                          struct slave *old_active)
 971 {
 972         struct dev_mc_list *dmi;
 973
 974         if (!USES_PRIMARY(bond->params.mode))
 975                 /* nothing to do -  mc list is already up-to-date on
 976                  * all slaves
 977                  */
 978                 return;
 979
 980         if (old_active) {
 981                 if (bond->dev->flags & IFF_PROMISC)
 982                         dev_set_promiscuity(old_active->dev, -1);
 983
 984                 if (bond->dev->flags & IFF_ALLMULTI)
 985                         dev_set_allmulti(old_active->dev, -1);
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 988                         dev_mc_delete(old_active->dev, dmi->dmi_addr,
 989                                       dmi->dmi_addrlen, 0);
 990         }
 991
 992         if (new_active) {
 993                 /* FIXME: Signal errors upstream. */
 994                 if (bond->dev->flags & IFF_PROMISC)
 995                         dev_set_promiscuity(new_active->dev, 1);
 996
 997                 if (bond->dev->flags & IFF_ALLMULTI)
 998                         dev_set_allmulti(new_active->dev, 1);
 999
1000                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
1001                         dev_mc_add(new_active->dev, dmi->dmi_addr,
1002                                    dmi->dmi_addrlen, 0);
1003                 bond_resend_igmp_join_requests(bond);
1004         }
1005 }
1006
1007 /*
1008  * bond_do_fail_over_mac
1009  *
1010  * Perform special MAC address swapping for fail_over_mac settings
1011  *
1012  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1013  */
1014 static void bond_do_fail_over_mac(struct bonding *bond,
1015                                   struct slave *new_active,
1016                                   struct slave *old_active)
1017         __releases(&bond->curr_slave_lock)
1018         __releases(&bond->lock)
1019         __acquires(&bond->lock)
1020         __acquires(&bond->curr_slave_lock)
1021 {
1022         u8 tmp_mac[ETH_ALEN];
1023         struct sockaddr saddr;
1024         int rv;
1025
1026         switch (bond->params.fail_over_mac) {
1027         case BOND_FOM_ACTIVE:
1028                 if (new_active)
1029                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1030                                new_active->dev->addr_len);
1031                 break;
1032         case BOND_FOM_FOLLOW:
1033                 /*
1034                  * if new_active && old_active, swap them
1035                  * if just old_active, do nothing (going to no active slave)
1036                  * if just new_active, set new_active to bond's MAC
1037                  */
1038                 if (!new_active)
1039                         return;
1040
1041                 write_unlock_bh(&bond->curr_slave_lock);
1042                 read_unlock(&bond->lock);
1043
1044                 if (old_active) {
1045                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1046                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1047                                ETH_ALEN);
1048                         saddr.sa_family = new_active->dev->type;
1049                 } else {
1050                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1051                         saddr.sa_family = bond->dev->type;
1052                 }
1053
1054                 rv = dev_set_mac_address(new_active->dev, &saddr);
1055                 if (rv) {
1056                         pr_err("%s: Error %d setting MAC of slave %s\n",
1057                                bond->dev->name, -rv, new_active->dev->name);
1058                         goto out;
1059                 }
1060
1061                 if (!old_active)
1062                         goto out;
1063
1064                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1065                 saddr.sa_family = old_active->dev->type;
1066
1067                 rv = dev_set_mac_address(old_active->dev, &saddr);
1068                 if (rv)
1069                         pr_err("%s: Error %d setting MAC of slave %s\n",
1070                                bond->dev->name, -rv, new_active->dev->name);
1071 out:
1072                 read_lock(&bond->lock);
1073                 write_lock_bh(&bond->curr_slave_lock);
1074                 break;
1075         default:
1076                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1077                        bond->dev->name, bond->params.fail_over_mac);
1078                 break;
1079         }
1080
1081 }
1082
1083 static bool bond_should_change_active(struct bonding *bond)
1084 {
1085         struct slave *prim = bond->primary_slave;
1086         struct slave *curr = bond->curr_active_slave;
1087
1088         if (!prim || !curr || curr->link != BOND_LINK_UP)
1089                 return true;
1090         if (bond->force_primary) {
1091                 bond->force_primary = false;
1092                 return true;
1093         }
1094         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1095             (prim->speed < curr->speed ||
1096              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1097                 return false;
1098         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1099                 return false;
1100         return true;
1101 }
1102
1103 /**
1104  * find_best_interface - select the best available slave to be the active one
1105  * @bond: our bonding struct
1106  *
1107  * Warning: Caller must hold curr_slave_lock for writing.
1108  */
1109 static struct slave *bond_find_best_slave(struct bonding *bond)
1110 {
1111         struct slave *new_active, *old_active;
1112         struct slave *bestslave = NULL;
1113         int mintime = bond->params.updelay;
1114         int i;
1115
1116         new_active = bond->curr_active_slave;
1117
1118         if (!new_active) { /* there were no active slaves left */
1119                 if (bond->slave_cnt > 0)   /* found one slave */
1120                         new_active = bond->first_slave;
1121                 else
1122                         return NULL; /* still no slave, return NULL */
1123         }
1124
1125         if ((bond->primary_slave) &&
1126             bond->primary_slave->link == BOND_LINK_UP &&
1127             bond_should_change_active(bond)) {
1128                 new_active = bond->primary_slave;
1129         }
1130
1131         /* remember where to stop iterating over the slaves */
1132         old_active = new_active;
1133
1134         bond_for_each_slave_from(bond, new_active, i, old_active) {
1135                 if (new_active->link == BOND_LINK_UP) {
1136                         return new_active;
1137                 } else if (new_active->link == BOND_LINK_BACK &&
1138                            IS_UP(new_active->dev)) {
1139                         /* link up, but waiting for stabilization */
1140                         if (new_active->delay < mintime) {
1141                                 mintime = new_active->delay;
1142                                 bestslave = new_active;
1143                         }
1144                 }
1145         }
1146
1147         return bestslave;
1148 }
1149
1150 /**
1151  * change_active_interface - change the active slave into the specified one
1152  * @bond: our bonding struct
1153  * @new: the new slave to make the active one
1154  *
1155  * Set the new slave to the bond's settings and unset them on the old
1156  * curr_active_slave.
1157  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1158  *
1159  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1160  * because it is apparently the best available slave we have, even though its
1161  * updelay hasn't timed out yet.
1162  *
1163  * If new_active is not NULL, caller must hold bond->lock for read and
1164  * curr_slave_lock for write_bh.
1165  */
1166 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1167 {
1168         struct slave *old_active = bond->curr_active_slave;
1169
1170         if (old_active == new_active)
1171                 return;
1172
1173         if (new_active) {
1174                 new_active->jiffies = jiffies;
1175
1176                 if (new_active->link == BOND_LINK_BACK) {
1177                         if (USES_PRIMARY(bond->params.mode)) {
1178                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1179                                         bond->dev->name, new_active->dev->name,
1180                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1181                         }
1182
1183                         new_active->delay = 0;
1184                         new_active->link = BOND_LINK_UP;
1185
1186                         if (bond->params.mode == BOND_MODE_8023AD)
1187                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1188
1189                         if (bond_is_lb(bond))
1190                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1191                 } else {
1192                         if (USES_PRIMARY(bond->params.mode)) {
1193                                 pr_info("%s: making interface %s the new active one.\n",
1194                                         bond->dev->name, new_active->dev->name);
1195                         }
1196                 }
1197         }
1198
1199         if (USES_PRIMARY(bond->params.mode))
1200                 bond_mc_swap(bond, new_active, old_active);
1201
1202         if (bond_is_lb(bond)) {
1203                 bond_alb_handle_active_change(bond, new_active);
1204                 if (old_active)
1205                         bond_set_slave_inactive_flags(old_active);
1206                 if (new_active)
1207                         bond_set_slave_active_flags(new_active);
1208         } else {
1209                 bond->curr_active_slave = new_active;
1210         }
1211
1212         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1213                 if (old_active)
1214                         bond_set_slave_inactive_flags(old_active);
1215
1216                 if (new_active) {
1217                         bond_set_slave_active_flags(new_active);
1218
1219                         if (bond->params.fail_over_mac)
1220                                 bond_do_fail_over_mac(bond, new_active,
1221                                                       old_active);
1222
1223                         bond->send_grat_arp = bond->params.num_grat_arp;
1224                         bond_send_gratuitous_arp(bond);
1225
1226                         bond->send_unsol_na = bond->params.num_unsol_na;
1227                         bond_send_unsolicited_na(bond);
1228
1229                         write_unlock_bh(&bond->curr_slave_lock);
1230                         read_unlock(&bond->lock);
1231
1232                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1233
1234                         read_lock(&bond->lock);
1235                         write_lock_bh(&bond->curr_slave_lock);
1236                 }
1237         }
1238 }
1239
1240 /**
1241  * bond_select_active_slave - select a new active slave, if needed
1242  * @bond: our bonding struct
1243  *
1244  * This functions should be called when one of the following occurs:
1245  * - The old curr_active_slave has been released or lost its link.
1246  * - The primary_slave has got its link back.
1247  * - A slave has got its link back and there's no old curr_active_slave.
1248  *
1249  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1250  */
1251 void bond_select_active_slave(struct bonding *bond)
1252 {
1253         struct slave *best_slave;
1254         int rv;
1255
1256         best_slave = bond_find_best_slave(bond);
1257         if (best_slave != bond->curr_active_slave) {
1258                 bond_change_active_slave(bond, best_slave);
1259                 rv = bond_set_carrier(bond);
1260                 if (!rv)
1261                         return;
1262
1263                 if (netif_carrier_ok(bond->dev)) {
1264                         pr_info("%s: first active interface up!\n",
1265                                 bond->dev->name);
1266                 } else {
1267                         pr_info("%s: now running without any active interface !\n",
1268                                 bond->dev->name);
1269                 }
1270         }
1271 }
1272
1273 /*--------------------------- slave list handling ---------------------------*/
1274
1275 /*
1276  * This function attaches the slave to the end of list.
1277  *
1278  * bond->lock held for writing by caller.
1279  */
1280 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1281 {
1282         if (bond->first_slave == NULL) { /* attaching the first slave */
1283                 new_slave->next = new_slave;
1284                 new_slave->prev = new_slave;
1285                 bond->first_slave = new_slave;
1286         } else {
1287                 new_slave->next = bond->first_slave;
1288                 new_slave->prev = bond->first_slave->prev;
1289                 new_slave->next->prev = new_slave;
1290                 new_slave->prev->next = new_slave;
1291         }
1292
1293         bond->slave_cnt++;
1294 }
1295
1296 /*
1297  * This function detaches the slave from the list.
1298  * WARNING: no check is made to verify if the slave effectively
1299  * belongs to <bond>.
1300  * Nothing is freed on return, structures are just unchained.
1301  * If any slave pointer in bond was pointing to <slave>,
1302  * it should be changed by the calling function.
1303  *
1304  * bond->lock held for writing by caller.
1305  */
1306 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1307 {
1308         if (slave->next)
1309                 slave->next->prev = slave->prev;
1310
1311         if (slave->prev)
1312                 slave->prev->next = slave->next;
1313
1314         if (bond->first_slave == slave) { /* slave is the first slave */
1315                 if (bond->slave_cnt > 1) { /* there are more slave */
1316                         bond->first_slave = slave->next;
1317                 } else {
1318                         bond->first_slave = NULL; /* slave was the last one */
1319                 }
1320         }
1321
1322         slave->next = NULL;
1323         slave->prev = NULL;
1324         bond->slave_cnt--;
1325 }
1326
1327 /*---------------------------------- IOCTL ----------------------------------*/
1328
1329 static int bond_sethwaddr(struct net_device *bond_dev,
1330                           struct net_device *slave_dev)
1331 {
1332         pr_debug("bond_dev=%p\n", bond_dev);
1333         pr_debug("slave_dev=%p\n", slave_dev);
1334         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1335         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1336         return 0;
1337 }
1338
1339 #define BOND_VLAN_FEATURES \
1340         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1341          NETIF_F_HW_VLAN_FILTER)
1342
1343 /*
1344  * Compute the common dev->feature set available to all slaves.  Some
1345  * feature bits are managed elsewhere, so preserve those feature bits
1346  * on the master device.
1347  */
1348 static int bond_compute_features(struct bonding *bond)
1349 {
1350         struct slave *slave;
1351         struct net_device *bond_dev = bond->dev;
1352         unsigned long features = bond_dev->features;
1353         unsigned long vlan_features = 0;
1354         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1355                                                 bond_dev->hard_header_len);
1356         int i;
1357
1358         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1359         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1360
1361         if (!bond->first_slave)
1362                 goto done;
1363
1364         features &= ~NETIF_F_ONE_FOR_ALL;
1365
1366         vlan_features = bond->first_slave->dev->vlan_features;
1367         bond_for_each_slave(bond, slave, i) {
1368                 features = netdev_increment_features(features,
1369                                                      slave->dev->features,
1370                                                      NETIF_F_ONE_FOR_ALL);
1371                 vlan_features = netdev_increment_features(vlan_features,
1372                                                         slave->dev->vlan_features,
1373                                                         NETIF_F_ONE_FOR_ALL);
1374                 if (slave->dev->hard_header_len > max_hard_header_len)
1375                         max_hard_header_len = slave->dev->hard_header_len;
1376         }
1377
1378 done:
1379         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1380         bond_dev->features = netdev_fix_features(features, NULL);
1381         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1382         bond_dev->hard_header_len = max_hard_header_len;
1383
1384         return 0;
1385 }
1386
1387 static void bond_setup_by_slave(struct net_device *bond_dev,
1388                                 struct net_device *slave_dev)
1389 {
1390         struct bonding *bond = netdev_priv(bond_dev);
1391
1392         bond_dev->header_ops        = slave_dev->header_ops;
1393
1394         bond_dev->type              = slave_dev->type;
1395         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1396         bond_dev->addr_len          = slave_dev->addr_len;
1397
1398         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1399                 slave_dev->addr_len);
1400         bond->setup_by_slave = 1;
1401 }
1402
1403 /* enslave device <slave> to bond device <master> */
1404 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1405 {
1406         struct bonding *bond = netdev_priv(bond_dev);
1407         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1408         struct slave *new_slave = NULL;
1409         struct dev_mc_list *dmi;
1410         struct sockaddr addr;
1411         int link_reporting;
1412         int old_features = bond_dev->features;
1413         int res = 0;
1414
1415         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1416                 slave_ops->ndo_do_ioctl == NULL) {
1417                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1418                            bond_dev->name, slave_dev->name);
1419         }
1420
1421         /* bond must be initialized by bond_open() before enslaving */
1422         if (!(bond_dev->flags & IFF_UP)) {
1423                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1424                            bond_dev->name);
1425         }
1426
1427         /* already enslaved */
1428         if (slave_dev->flags & IFF_SLAVE) {
1429                 pr_debug("Error, Device was already enslaved\n");
1430                 return -EBUSY;
1431         }
1432
1433         /* vlan challenged mutual exclusion */
1434         /* no need to lock since we're protected by rtnl_lock */
1435         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1436                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1437                 if (!list_empty(&bond->vlan_list)) {
1438                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1439                                bond_dev->name, slave_dev->name, bond_dev->name);
1440                         return -EPERM;
1441                 } else {
1442                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1443                                    bond_dev->name, slave_dev->name,
1444                                    slave_dev->name, bond_dev->name);
1445                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1446                 }
1447         } else {
1448                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1449                 if (bond->slave_cnt == 0) {
1450                         /* First slave, and it is not VLAN challenged,
1451                          * so remove the block of adding VLANs over the bond.
1452                          */
1453                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1454                 }
1455         }
1456
1457         /*
1458          * Old ifenslave binaries are no longer supported.  These can
1459          * be identified with moderate accuracy by the state of the slave:
1460          * the current ifenslave will set the interface down prior to
1461          * enslaving it; the old ifenslave will not.
1462          */
1463         if ((slave_dev->flags & IFF_UP)) {
1464                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1465                        slave_dev->name);
1466                 res = -EPERM;
1467                 goto err_undo_flags;
1468         }
1469
1470         /* set bonding device ether type by slave - bonding netdevices are
1471          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1472          * there is a need to override some of the type dependent attribs/funcs.
1473          *
1474          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1475          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1476          */
1477         if (bond->slave_cnt == 0) {
1478                 if (bond_dev->type != slave_dev->type) {
1479                         pr_debug("%s: change device type from %d to %d\n",
1480                                  bond_dev->name,
1481                                  bond_dev->type, slave_dev->type);
1482
1483                         res = netdev_bonding_change(bond_dev,
1484                                                     NETDEV_PRE_TYPE_CHANGE);
1485                         res = notifier_to_errno(res);
1486                         if (res) {
1487                                 pr_err("%s: refused to change device type\n",
1488                                        bond_dev->name);
1489                                 res = -EBUSY;
1490                                 goto err_undo_flags;
1491                         }
1492
1493                         /* Flush unicast and multicast addresses */
1494                         dev_uc_flush(bond_dev);
1495                         dev_addr_discard(bond_dev);
1496
1497                         if (slave_dev->type != ARPHRD_ETHER)
1498                                 bond_setup_by_slave(bond_dev, slave_dev);
1499                         else
1500                                 ether_setup(bond_dev);
1501
1502                         netdev_bonding_change(bond_dev,
1503                                               NETDEV_POST_TYPE_CHANGE);
1504                 }
1505         } else if (bond_dev->type != slave_dev->type) {
1506                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1507                        slave_dev->name,
1508                        slave_dev->type, bond_dev->type);
1509                 res = -EINVAL;
1510                 goto err_undo_flags;
1511         }
1512
1513         if (slave_ops->ndo_set_mac_address == NULL) {
1514                 if (bond->slave_cnt == 0) {
1515                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1516                                    bond_dev->name);
1517                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1518                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1519                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1520                                bond_dev->name);
1521                         res = -EOPNOTSUPP;
1522                         goto err_undo_flags;
1523                 }
1524         }
1525
1526         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1527         if (!new_slave) {
1528                 res = -ENOMEM;
1529                 goto err_undo_flags;
1530         }
1531
1532         /* save slave's original flags before calling
1533          * netdev_set_master and dev_open
1534          */
1535         new_slave->original_flags = slave_dev->flags;
1536
1537         /*
1538          * Save slave's original ("permanent") mac address for modes
1539          * that need it, and for restoring it upon release, and then
1540          * set it to the master's address
1541          */
1542         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1543
1544         if (!bond->params.fail_over_mac) {
1545                 /*
1546                  * Set slave to master's mac address.  The application already
1547                  * set the master's mac address to that of the first slave
1548                  */
1549                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1550                 addr.sa_family = slave_dev->type;
1551                 res = dev_set_mac_address(slave_dev, &addr);
1552                 if (res) {
1553                         pr_debug("Error %d calling set_mac_address\n", res);
1554                         goto err_free;
1555                 }
1556         }
1557
1558         res = netdev_set_master(slave_dev, bond_dev);
1559         if (res) {
1560                 pr_debug("Error %d calling netdev_set_master\n", res);
1561                 goto err_restore_mac;
1562         }
1563         /* open the slave since the application closed it */
1564         res = dev_open(slave_dev);
1565         if (res) {
1566                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1567                 goto err_unset_master;
1568         }
1569
1570         new_slave->dev = slave_dev;
1571         slave_dev->priv_flags |= IFF_BONDING;
1572
1573         if (bond_is_lb(bond)) {
1574                 /* bond_alb_init_slave() must be called before all other stages since
1575                  * it might fail and we do not want to have to undo everything
1576                  */
1577                 res = bond_alb_init_slave(bond, new_slave);
1578                 if (res)
1579                         goto err_close;
1580         }
1581
1582         /* If the mode USES_PRIMARY, then the new slave gets the
1583          * master's promisc (and mc) settings only if it becomes the
1584          * curr_active_slave, and that is taken care of later when calling
1585          * bond_change_active()
1586          */
1587         if (!USES_PRIMARY(bond->params.mode)) {
1588                 /* set promiscuity level to new slave */
1589                 if (bond_dev->flags & IFF_PROMISC) {
1590                         res = dev_set_promiscuity(slave_dev, 1);
1591                         if (res)
1592                                 goto err_close;
1593                 }
1594
1595                 /* set allmulti level to new slave */
1596                 if (bond_dev->flags & IFF_ALLMULTI) {
1597                         res = dev_set_allmulti(slave_dev, 1);
1598                         if (res)
1599                                 goto err_close;
1600                 }
1601
1602                 netif_addr_lock_bh(bond_dev);
1603                 /* upload master's mc_list to new slave */
1604                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1605                         dev_mc_add(slave_dev, dmi->dmi_addr,
1606                                    dmi->dmi_addrlen, 0);
1607                 netif_addr_unlock_bh(bond_dev);
1608         }
1609
1610         if (bond->params.mode == BOND_MODE_8023AD) {
1611                 /* add lacpdu mc addr to mc list */
1612                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1613
1614                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1615         }
1616
1617         bond_add_vlans_on_slave(bond, slave_dev);
1618
1619         write_lock_bh(&bond->lock);
1620
1621         bond_attach_slave(bond, new_slave);
1622
1623         new_slave->delay = 0;
1624         new_slave->link_failure_count = 0;
1625
1626         bond_compute_features(bond);
1627
1628         write_unlock_bh(&bond->lock);
1629
1630         read_lock(&bond->lock);
1631
1632         new_slave->last_arp_rx = jiffies;
1633
1634         if (bond->params.miimon && !bond->params.use_carrier) {
1635                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1636
1637                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1638                         /*
1639                          * miimon is set but a bonded network driver
1640                          * does not support ETHTOOL/MII and
1641                          * arp_interval is not set.  Note: if
1642                          * use_carrier is enabled, we will never go
1643                          * here (because netif_carrier is always
1644                          * supported); thus, we don't need to change
1645                          * the messages for netif_carrier.
1646                          */
1647                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1648                                bond_dev->name, slave_dev->name);
1649                 } else if (link_reporting == -1) {
1650                         /* unable get link status using mii/ethtool */
1651                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1652                                    bond_dev->name, slave_dev->name);
1653                 }
1654         }
1655
1656         /* check for initial state */
1657         if (!bond->params.miimon ||
1658             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1659                 if (bond->params.updelay) {
1660                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1661                         new_slave->link  = BOND_LINK_BACK;
1662                         new_slave->delay = bond->params.updelay;
1663                 } else {
1664                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1665                         new_slave->link  = BOND_LINK_UP;
1666                 }
1667                 new_slave->jiffies = jiffies;
1668         } else {
1669                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1670                 new_slave->link  = BOND_LINK_DOWN;
1671         }
1672
1673         if (bond_update_speed_duplex(new_slave) &&
1674             (new_slave->link != BOND_LINK_DOWN)) {
1675                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1676                            bond_dev->name, new_slave->dev->name);
1677
1678                 if (bond->params.mode == BOND_MODE_8023AD) {
1679                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1680                                    bond_dev->name);
1681                 }
1682         }
1683
1684         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1685                 /* if there is a primary slave, remember it */
1686                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1687                         bond->primary_slave = new_slave;
1688                         bond->force_primary = true;
1689                 }
1690         }
1691
1692         write_lock_bh(&bond->curr_slave_lock);
1693
1694         switch (bond->params.mode) {
1695         case BOND_MODE_ACTIVEBACKUP:
1696                 bond_set_slave_inactive_flags(new_slave);
1697                 bond_select_active_slave(bond);
1698                 break;
1699         case BOND_MODE_8023AD:
1700                 /* in 802.3ad mode, the internal mechanism
1701                  * will activate the slaves in the selected
1702                  * aggregator
1703                  */
1704                 bond_set_slave_inactive_flags(new_slave);
1705                 /* if this is the first slave */
1706                 if (bond->slave_cnt == 1) {
1707                         SLAVE_AD_INFO(new_slave).id = 1;
1708                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1709                          * can be called only after the mac address of the bond is set
1710                          */
1711                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1712                                             bond->params.lacp_fast);
1713                 } else {
1714                         SLAVE_AD_INFO(new_slave).id =
1715                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1716                 }
1717
1718                 bond_3ad_bind_slave(new_slave);
1719                 break;
1720         case BOND_MODE_TLB:
1721         case BOND_MODE_ALB:
1722                 new_slave->state = BOND_STATE_ACTIVE;
1723                 bond_set_slave_inactive_flags(new_slave);
1724                 bond_select_active_slave(bond);
1725                 break;
1726         default:
1727                 pr_debug("This slave is always active in trunk mode\n");
1728
1729                 /* always active in trunk mode */
1730                 new_slave->state = BOND_STATE_ACTIVE;
1731
1732                 /* In trunking mode there is little meaning to curr_active_slave
1733                  * anyway (it holds no special properties of the bond device),
1734                  * so we can change it without calling change_active_interface()
1735                  */
1736                 if (!bond->curr_active_slave)
1737                         bond->curr_active_slave = new_slave;
1738
1739                 break;
1740         } /* switch(bond_mode) */
1741
1742         write_unlock_bh(&bond->curr_slave_lock);
1743
1744         bond_set_carrier(bond);
1745
1746         read_unlock(&bond->lock);
1747
1748         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1749         if (res)
1750                 goto err_close;
1751
1752         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1753                 bond_dev->name, slave_dev->name,
1754                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1755                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1756
1757         /* enslave is successful */
1758         return 0;
1759
1760 /* Undo stages on error */
1761 err_close:
1762         dev_close(slave_dev);
1763
1764 err_unset_master:
1765         netdev_set_master(slave_dev, NULL);
1766
1767 err_restore_mac:
1768         if (!bond->params.fail_over_mac) {
1769                 /* XXX TODO - fom follow mode needs to change master's
1770                  * MAC if this slave's MAC is in use by the bond, or at
1771                  * least print a warning.
1772                  */
1773                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1774                 addr.sa_family = slave_dev->type;
1775                 dev_set_mac_address(slave_dev, &addr);
1776         }
1777
1778 err_free:
1779         kfree(new_slave);
1780
1781 err_undo_flags:
1782         bond_dev->features = old_features;
1783
1784         return res;
1785 }
1786
1787 /*
1788  * Try to release the slave device <slave> from the bond device <master>
1789  * It is legal to access curr_active_slave without a lock because all the function
1790  * is write-locked.
1791  *
1792  * The rules for slave state should be:
1793  *   for Active/Backup:
1794  *     Active stays on all backups go down
1795  *   for Bonded connections:
1796  *     The first up interface should be left on and all others downed.
1797  */
1798 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1799 {
1800         struct bonding *bond = netdev_priv(bond_dev);
1801         struct slave *slave, *oldcurrent;
1802         struct sockaddr addr;
1803
1804         /* slave is not a slave or master is not master of this slave */
1805         if (!(slave_dev->flags & IFF_SLAVE) ||
1806             (slave_dev->master != bond_dev)) {
1807                 pr_err("%s: Error: cannot release %s.\n",
1808                        bond_dev->name, slave_dev->name);
1809                 return -EINVAL;
1810         }
1811
1812         write_lock_bh(&bond->lock);
1813
1814         slave = bond_get_slave_by_dev(bond, slave_dev);
1815         if (!slave) {
1816                 /* not a slave of this bond */
1817                 pr_info("%s: %s not enslaved\n",
1818                         bond_dev->name, slave_dev->name);
1819                 write_unlock_bh(&bond->lock);
1820                 return -EINVAL;
1821         }
1822
1823         if (!bond->params.fail_over_mac) {
1824                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1825                     bond->slave_cnt > 1)
1826                         pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1827                                    bond_dev->name, slave_dev->name,
1828                                    slave->perm_hwaddr,
1829                                    bond_dev->name, slave_dev->name);
1830         }
1831
1832         /* Inform AD package of unbinding of slave. */
1833         if (bond->params.mode == BOND_MODE_8023AD) {
1834                 /* must be called before the slave is
1835                  * detached from the list
1836                  */
1837                 bond_3ad_unbind_slave(slave);
1838         }
1839
1840         pr_info("%s: releasing %s interface %s\n",
1841                 bond_dev->name,
1842                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1843                 slave_dev->name);
1844
1845         oldcurrent = bond->curr_active_slave;
1846
1847         bond->current_arp_slave = NULL;
1848
1849         /* release the slave from its bond */
1850         bond_detach_slave(bond, slave);
1851
1852         bond_compute_features(bond);
1853
1854         if (bond->primary_slave == slave)
1855                 bond->primary_slave = NULL;
1856
1857         if (oldcurrent == slave)
1858                 bond_change_active_slave(bond, NULL);
1859
1860         if (bond_is_lb(bond)) {
1861                 /* Must be called only after the slave has been
1862                  * detached from the list and the curr_active_slave
1863                  * has been cleared (if our_slave == old_current),
1864                  * but before a new active slave is selected.
1865                  */
1866                 write_unlock_bh(&bond->lock);
1867                 bond_alb_deinit_slave(bond, slave);
1868                 write_lock_bh(&bond->lock);
1869         }
1870
1871         if (oldcurrent == slave) {
1872                 /*
1873                  * Note that we hold RTNL over this sequence, so there
1874                  * is no concern that another slave add/remove event
1875                  * will interfere.
1876                  */
1877                 write_unlock_bh(&bond->lock);
1878                 read_lock(&bond->lock);
1879                 write_lock_bh(&bond->curr_slave_lock);
1880
1881                 bond_select_active_slave(bond);
1882
1883                 write_unlock_bh(&bond->curr_slave_lock);
1884                 read_unlock(&bond->lock);
1885                 write_lock_bh(&bond->lock);
1886         }
1887
1888         if (bond->slave_cnt == 0) {
1889                 bond_set_carrier(bond);
1890
1891                 /* if the last slave was removed, zero the mac address
1892                  * of the master so it will be set by the application
1893                  * to the mac address of the first slave
1894                  */
1895                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1896
1897                 if (list_empty(&bond->vlan_list)) {
1898                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1899                 } else {
1900                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1901                                    bond_dev->name, bond_dev->name);
1902                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1903                                    bond_dev->name);
1904                 }
1905         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1906                    !bond_has_challenged_slaves(bond)) {
1907                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1908                         bond_dev->name, slave_dev->name, bond_dev->name);
1909                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1910         }
1911
1912         write_unlock_bh(&bond->lock);
1913
1914         /* must do this from outside any spinlocks */
1915         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1916
1917         bond_del_vlans_from_slave(bond, slave_dev);
1918
1919         /* If the mode USES_PRIMARY, then we should only remove its
1920          * promisc and mc settings if it was the curr_active_slave, but that was
1921          * already taken care of above when we detached the slave
1922          */
1923         if (!USES_PRIMARY(bond->params.mode)) {
1924                 /* unset promiscuity level from slave */
1925                 if (bond_dev->flags & IFF_PROMISC)
1926                         dev_set_promiscuity(slave_dev, -1);
1927
1928                 /* unset allmulti level from slave */
1929                 if (bond_dev->flags & IFF_ALLMULTI)
1930                         dev_set_allmulti(slave_dev, -1);
1931
1932                 /* flush master's mc_list from slave */
1933                 netif_addr_lock_bh(bond_dev);
1934                 bond_mc_list_flush(bond_dev, slave_dev);
1935                 netif_addr_unlock_bh(bond_dev);
1936         }
1937
1938         netdev_set_master(slave_dev, NULL);
1939
1940         /* close slave before restoring its mac address */
1941         dev_close(slave_dev);
1942
1943         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1944                 /* restore original ("permanent") mac address */
1945                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1946                 addr.sa_family = slave_dev->type;
1947                 dev_set_mac_address(slave_dev, &addr);
1948         }
1949
1950         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1951                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1952                                    IFF_SLAVE_NEEDARP);
1953
1954         kfree(slave);
1955
1956         return 0;  /* deletion OK */
1957 }
1958
1959 /*
1960 * First release a slave and than destroy the bond if no more slaves are left.
1961 * Must be under rtnl_lock when this function is called.
1962 */
1963 int  bond_release_and_destroy(struct net_device *bond_dev,
1964                               struct net_device *slave_dev)
1965 {
1966         struct bonding *bond = netdev_priv(bond_dev);
1967         int ret;
1968
1969         ret = bond_release(bond_dev, slave_dev);
1970         if ((ret == 0) && (bond->slave_cnt == 0)) {
1971                 pr_info("%s: destroying bond %s.\n",
1972                         bond_dev->name, bond_dev->name);
1973                 unregister_netdevice(bond_dev);
1974         }
1975         return ret;
1976 }
1977
1978 /*
1979  * This function releases all slaves.
1980  */
1981 static int bond_release_all(struct net_device *bond_dev)
1982 {
1983         struct bonding *bond = netdev_priv(bond_dev);
1984         struct slave *slave;
1985         struct net_device *slave_dev;
1986         struct sockaddr addr;
1987
1988         write_lock_bh(&bond->lock);
1989
1990         netif_carrier_off(bond_dev);
1991
1992         if (bond->slave_cnt == 0)
1993                 goto out;
1994
1995         bond->current_arp_slave = NULL;
1996         bond->primary_slave = NULL;
1997         bond_change_active_slave(bond, NULL);
1998
1999         while ((slave = bond->first_slave) != NULL) {
2000                 /* Inform AD package of unbinding of slave
2001                  * before slave is detached from the list.
2002                  */
2003                 if (bond->params.mode == BOND_MODE_8023AD)
2004                         bond_3ad_unbind_slave(slave);
2005
2006                 slave_dev = slave->dev;
2007                 bond_detach_slave(bond, slave);
2008
2009                 /* now that the slave is detached, unlock and perform
2010                  * all the undo steps that should not be called from
2011                  * within a lock.
2012                  */
2013                 write_unlock_bh(&bond->lock);
2014
2015                 if (bond_is_lb(bond)) {
2016                         /* must be called only after the slave
2017                          * has been detached from the list
2018                          */
2019                         bond_alb_deinit_slave(bond, slave);
2020                 }
2021
2022                 bond_compute_features(bond);
2023
2024                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2025                 bond_del_vlans_from_slave(bond, slave_dev);
2026
2027                 /* If the mode USES_PRIMARY, then we should only remove its
2028                  * promisc and mc settings if it was the curr_active_slave, but that was
2029                  * already taken care of above when we detached the slave
2030                  */
2031                 if (!USES_PRIMARY(bond->params.mode)) {
2032                         /* unset promiscuity level from slave */
2033                         if (bond_dev->flags & IFF_PROMISC)
2034                                 dev_set_promiscuity(slave_dev, -1);
2035
2036                         /* unset allmulti level from slave */
2037                         if (bond_dev->flags & IFF_ALLMULTI)
2038                                 dev_set_allmulti(slave_dev, -1);
2039
2040                         /* flush master's mc_list from slave */
2041                         netif_addr_lock_bh(bond_dev);
2042                         bond_mc_list_flush(bond_dev, slave_dev);
2043                         netif_addr_unlock_bh(bond_dev);
2044                 }
2045
2046                 netdev_set_master(slave_dev, NULL);
2047
2048                 /* close slave before restoring its mac address */
2049                 dev_close(slave_dev);
2050
2051                 if (!bond->params.fail_over_mac) {
2052                         /* restore original ("permanent") mac address*/
2053                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2054                         addr.sa_family = slave_dev->type;
2055                         dev_set_mac_address(slave_dev, &addr);
2056                 }
2057
2058                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2059                                            IFF_SLAVE_INACTIVE);
2060
2061                 kfree(slave);
2062
2063                 /* re-acquire the lock before getting the next slave */
2064                 write_lock_bh(&bond->lock);
2065         }
2066
2067         /* zero the mac address of the master so it will be
2068          * set by the application to the mac address of the
2069          * first slave
2070          */
2071         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2072
2073         if (list_empty(&bond->vlan_list))
2074                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2075         else {
2076                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2077                            bond_dev->name, bond_dev->name);
2078                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2079                            bond_dev->name);
2080         }
2081
2082         pr_info("%s: released all slaves\n", bond_dev->name);
2083
2084 out:
2085         write_unlock_bh(&bond->lock);
2086
2087         return 0;
2088 }
2089
2090 /*
2091  * This function changes the active slave to slave <slave_dev>.
2092  * It returns -EINVAL in the following cases.
2093  *  - <slave_dev> is not found in the list.
2094  *  - There is not active slave now.
2095  *  - <slave_dev> is already active.
2096  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2097  *  - <slave_dev> is not running.
2098  * In these cases, this function does nothing.
2099  * In the other cases, current_slave pointer is changed and 0 is returned.
2100  */
2101 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2102 {
2103         struct bonding *bond = netdev_priv(bond_dev);
2104         struct slave *old_active = NULL;
2105         struct slave *new_active = NULL;
2106         int res = 0;
2107
2108         if (!USES_PRIMARY(bond->params.mode))
2109                 return -EINVAL;
2110
2111         /* Verify that master_dev is indeed the master of slave_dev */
2112         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2113                 return -EINVAL;
2114
2115         read_lock(&bond->lock);
2116
2117         read_lock(&bond->curr_slave_lock);
2118         old_active = bond->curr_active_slave;
2119         read_unlock(&bond->curr_slave_lock);
2120
2121         new_active = bond_get_slave_by_dev(bond, slave_dev);
2122
2123         /*
2124          * Changing to the current active: do nothing; return success.
2125          */
2126         if (new_active && (new_active == old_active)) {
2127                 read_unlock(&bond->lock);
2128                 return 0;
2129         }
2130
2131         if ((new_active) &&
2132             (old_active) &&
2133             (new_active->link == BOND_LINK_UP) &&
2134             IS_UP(new_active->dev)) {
2135                 write_lock_bh(&bond->curr_slave_lock);
2136                 bond_change_active_slave(bond, new_active);
2137                 write_unlock_bh(&bond->curr_slave_lock);
2138         } else
2139                 res = -EINVAL;
2140
2141         read_unlock(&bond->lock);
2142
2143         return res;
2144 }
2145
2146 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2147 {
2148         struct bonding *bond = netdev_priv(bond_dev);
2149
2150         info->bond_mode = bond->params.mode;
2151         info->miimon = bond->params.miimon;
2152
2153         read_lock(&bond->lock);
2154         info->num_slaves = bond->slave_cnt;
2155         read_unlock(&bond->lock);
2156
2157         return 0;
2158 }
2159
2160 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2161 {
2162         struct bonding *bond = netdev_priv(bond_dev);
2163         struct slave *slave;
2164         int i, res = -ENODEV;
2165
2166         read_lock(&bond->lock);
2167
2168         bond_for_each_slave(bond, slave, i) {
2169                 if (i == (int)info->slave_id) {
2170                         res = 0;
2171                         strcpy(info->slave_name, slave->dev->name);
2172                         info->link = slave->link;
2173                         info->state = slave->state;
2174                         info->link_failure_count = slave->link_failure_count;
2175                         break;
2176                 }
2177         }
2178
2179         read_unlock(&bond->lock);
2180
2181         return res;
2182 }
2183
2184 /*-------------------------------- Monitoring -------------------------------*/
2185
2186
2187 static int bond_miimon_inspect(struct bonding *bond)
2188 {
2189         struct slave *slave;
2190         int i, link_state, commit = 0;
2191         bool ignore_updelay;
2192
2193         ignore_updelay = !bond->curr_active_slave ? true : false;
2194
2195         bond_for_each_slave(bond, slave, i) {
2196                 slave->new_link = BOND_LINK_NOCHANGE;
2197
2198                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2199
2200                 switch (slave->link) {
2201                 case BOND_LINK_UP:
2202                         if (link_state)
2203                                 continue;
2204
2205                         slave->link = BOND_LINK_FAIL;
2206                         slave->delay = bond->params.downdelay;
2207                         if (slave->delay) {
2208                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2209                                         bond->dev->name,
2210                                         (bond->params.mode ==
2211                                          BOND_MODE_ACTIVEBACKUP) ?
2212                                         ((slave->state == BOND_STATE_ACTIVE) ?
2213                                          "active " : "backup ") : "",
2214                                         slave->dev->name,
2215                                         bond->params.downdelay * bond->params.miimon);
2216                         }
2217                         /*FALLTHRU*/
2218                 case BOND_LINK_FAIL:
2219                         if (link_state) {
2220                                 /*
2221                                  * recovered before downdelay expired
2222                                  */
2223                                 slave->link = BOND_LINK_UP;
2224                                 slave->jiffies = jiffies;
2225                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2226                                         bond->dev->name,
2227                                         (bond->params.downdelay - slave->delay) *
2228                                         bond->params.miimon,
2229                                         slave->dev->name);
2230                                 continue;
2231                         }
2232
2233                         if (slave->delay <= 0) {
2234                                 slave->new_link = BOND_LINK_DOWN;
2235                                 commit++;
2236                                 continue;
2237                         }
2238
2239                         slave->delay--;
2240                         break;
2241
2242                 case BOND_LINK_DOWN:
2243                         if (!link_state)
2244                                 continue;
2245
2246                         slave->link = BOND_LINK_BACK;
2247                         slave->delay = bond->params.updelay;
2248
2249                         if (slave->delay) {
2250                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2251                                         bond->dev->name, slave->dev->name,
2252                                         ignore_updelay ? 0 :
2253                                         bond->params.updelay *
2254                                         bond->params.miimon);
2255                         }
2256                         /*FALLTHRU*/
2257                 case BOND_LINK_BACK:
2258                         if (!link_state) {
2259                                 slave->link = BOND_LINK_DOWN;
2260                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2261                                         bond->dev->name,
2262                                         (bond->params.updelay - slave->delay) *
2263                                         bond->params.miimon,
2264                                         slave->dev->name);
2265
2266                                 continue;
2267                         }
2268
2269                         if (ignore_updelay)
2270                                 slave->delay = 0;
2271
2272                         if (slave->delay <= 0) {
2273                                 slave->new_link = BOND_LINK_UP;
2274                                 commit++;
2275                                 ignore_updelay = false;
2276                                 continue;
2277                         }
2278
2279                         slave->delay--;
2280                         break;
2281                 }
2282         }
2283
2284         return commit;
2285 }
2286
2287 static void bond_miimon_commit(struct bonding *bond)
2288 {
2289         struct slave *slave;
2290         int i;
2291
2292         bond_for_each_slave(bond, slave, i) {
2293                 switch (slave->new_link) {
2294                 case BOND_LINK_NOCHANGE:
2295                         continue;
2296
2297                 case BOND_LINK_UP:
2298                         slave->link = BOND_LINK_UP;
2299                         slave->jiffies = jiffies;
2300
2301                         if (bond->params.mode == BOND_MODE_8023AD) {
2302                                 /* prevent it from being the active one */
2303                                 slave->state = BOND_STATE_BACKUP;
2304                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2305                                 /* make it immediately active */
2306                                 slave->state = BOND_STATE_ACTIVE;
2307                         } else if (slave != bond->primary_slave) {
2308                                 /* prevent it from being the active one */
2309                                 slave->state = BOND_STATE_BACKUP;
2310                         }
2311
2312                         pr_info("%s: link status definitely up for interface %s.\n",
2313                                 bond->dev->name, slave->dev->name);
2314
2315                         /* notify ad that the link status has changed */
2316                         if (bond->params.mode == BOND_MODE_8023AD)
2317                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2318
2319                         if (bond_is_lb(bond))
2320                                 bond_alb_handle_link_change(bond, slave,
2321                                                             BOND_LINK_UP);
2322
2323                         if (!bond->curr_active_slave ||
2324                             (slave == bond->primary_slave))
2325                                 goto do_failover;
2326
2327                         continue;
2328
2329                 case BOND_LINK_DOWN:
2330                         if (slave->link_failure_count < UINT_MAX)
2331                                 slave->link_failure_count++;
2332
2333                         slave->link = BOND_LINK_DOWN;
2334
2335                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2336                             bond->params.mode == BOND_MODE_8023AD)
2337                                 bond_set_slave_inactive_flags(slave);
2338
2339                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2340                                 bond->dev->name, slave->dev->name);
2341
2342                         if (bond->params.mode == BOND_MODE_8023AD)
2343                                 bond_3ad_handle_link_change(slave,
2344                                                             BOND_LINK_DOWN);
2345
2346                         if (bond_is_lb(bond))
2347                                 bond_alb_handle_link_change(bond, slave,
2348                                                             BOND_LINK_DOWN);
2349
2350                         if (slave == bond->curr_active_slave)
2351                                 goto do_failover;
2352
2353                         continue;
2354
2355                 default:
2356                         pr_err("%s: invalid new link %d on slave %s\n",
2357                                bond->dev->name, slave->new_link,
2358                                slave->dev->name);
2359                         slave->new_link = BOND_LINK_NOCHANGE;
2360
2361                         continue;
2362                 }
2363
2364 do_failover:
2365                 ASSERT_RTNL();
2366                 write_lock_bh(&bond->curr_slave_lock);
2367                 bond_select_active_slave(bond);
2368                 write_unlock_bh(&bond->curr_slave_lock);
2369         }
2370
2371         bond_set_carrier(bond);
2372 }
2373
2374 /*
2375  * bond_mii_monitor
2376  *
2377  * Really a wrapper that splits the mii monitor into two phases: an
2378  * inspection, then (if inspection indicates something needs to be done)
2379  * an acquisition of appropriate locks followed by a commit phase to
2380  * implement whatever link state changes are indicated.
2381  */
2382 void bond_mii_monitor(struct work_struct *work)
2383 {
2384         struct bonding *bond = container_of(work, struct bonding,
2385                                             mii_work.work);
2386
2387         read_lock(&bond->lock);
2388         if (bond->kill_timers)
2389                 goto out;
2390
2391         if (bond->slave_cnt == 0)
2392                 goto re_arm;
2393
2394         if (bond->send_grat_arp) {
2395                 read_lock(&bond->curr_slave_lock);
2396                 bond_send_gratuitous_arp(bond);
2397                 read_unlock(&bond->curr_slave_lock);
2398         }
2399
2400         if (bond->send_unsol_na) {
2401                 read_lock(&bond->curr_slave_lock);
2402                 bond_send_unsolicited_na(bond);
2403                 read_unlock(&bond->curr_slave_lock);
2404         }
2405
2406         if (bond_miimon_inspect(bond)) {
2407                 read_unlock(&bond->lock);
2408                 rtnl_lock();
2409                 read_lock(&bond->lock);
2410
2411                 bond_miimon_commit(bond);
2412
2413                 read_unlock(&bond->lock);
2414                 rtnl_unlock();  /* might sleep, hold no other locks */
2415                 read_lock(&bond->lock);
2416         }
2417
2418 re_arm:
2419         if (bond->params.miimon)
2420                 queue_delayed_work(bond->wq, &bond->mii_work,
2421                                    msecs_to_jiffies(bond->params.miimon));
2422 out:
2423         read_unlock(&bond->lock);
2424 }
2425
2426 static __be32 bond_glean_dev_ip(struct net_device *dev)
2427 {
2428         struct in_device *idev;
2429         struct in_ifaddr *ifa;
2430         __be32 addr = 0;
2431
2432         if (!dev)
2433                 return 0;
2434
2435         rcu_read_lock();
2436         idev = __in_dev_get_rcu(dev);
2437         if (!idev)
2438                 goto out;
2439
2440         ifa = idev->ifa_list;
2441         if (!ifa)
2442                 goto out;
2443
2444         addr = ifa->ifa_local;
2445 out:
2446         rcu_read_unlock();
2447         return addr;
2448 }
2449
2450 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2451 {
2452         struct vlan_entry *vlan;
2453
2454         if (ip == bond->master_ip)
2455                 return 1;
2456
2457         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2458                 if (ip == vlan->vlan_ip)
2459                         return 1;
2460         }
2461
2462         return 0;
2463 }
2464
2465 /*
2466  * We go to the (large) trouble of VLAN tagging ARP frames because
2467  * switches in VLAN mode (especially if ports are configured as
2468  * "native" to a VLAN) might not pass non-tagged frames.
2469  */
2470 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2471 {
2472         struct sk_buff *skb;
2473
2474         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2475                  slave_dev->name, dest_ip, src_ip, vlan_id);
2476
2477         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2478                          NULL, slave_dev->dev_addr, NULL);
2479
2480         if (!skb) {
2481                 pr_err("ARP packet allocation failed\n");
2482                 return;
2483         }
2484         if (vlan_id) {
2485                 skb = vlan_put_tag(skb, vlan_id);
2486                 if (!skb) {
2487                         pr_err("failed to insert VLAN tag\n");
2488                         return;
2489                 }
2490         }
2491         arp_xmit(skb);
2492 }
2493
2494
2495 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2496 {
2497         int i, vlan_id, rv;
2498         __be32 *targets = bond->params.arp_targets;
2499         struct vlan_entry *vlan;
2500         struct net_device *vlan_dev;
2501         struct flowi fl;
2502         struct rtable *rt;
2503
2504         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2505                 if (!targets[i])
2506                         break;
2507                 pr_debug("basa: target %x\n", targets[i]);
2508                 if (list_empty(&bond->vlan_list)) {
2509                         pr_debug("basa: empty vlan: arp_send\n");
2510                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2511                                       bond->master_ip, 0);
2512                         continue;
2513                 }
2514
2515                 /*
2516                  * If VLANs are configured, we do a route lookup to
2517                  * determine which VLAN interface would be used, so we
2518                  * can tag the ARP with the proper VLAN tag.
2519                  */
2520                 memset(&fl, 0, sizeof(fl));
2521                 fl.fl4_dst = targets[i];
2522                 fl.fl4_tos = RTO_ONLINK;
2523
2524                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2525                 if (rv) {
2526                         if (net_ratelimit()) {
2527                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2528                                            bond->dev->name, &fl.fl4_dst);
2529                         }
2530                         continue;
2531                 }
2532
2533                 /*
2534                  * This target is not on a VLAN
2535                  */
2536                 if (rt->u.dst.dev == bond->dev) {
2537                         ip_rt_put(rt);
2538                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2539                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2540                                       bond->master_ip, 0);
2541                         continue;
2542                 }
2543
2544                 vlan_id = 0;
2545                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2546                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2547                         if (vlan_dev == rt->u.dst.dev) {
2548                                 vlan_id = vlan->vlan_id;
2549                                 pr_debug("basa: vlan match on %s %d\n",
2550                                        vlan_dev->name, vlan_id);
2551                                 break;
2552                         }
2553                 }
2554
2555                 if (vlan_id) {
2556                         ip_rt_put(rt);
2557                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2558                                       vlan->vlan_ip, vlan_id);
2559                         continue;
2560                 }
2561
2562                 if (net_ratelimit()) {
2563                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2564                                    bond->dev->name, &fl.fl4_dst,
2565                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2566                 }
2567                 ip_rt_put(rt);
2568         }
2569 }
2570
2571 /*
2572  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2573  * for each VLAN above us.
2574  *
2575  * Caller must hold curr_slave_lock for read or better
2576  */
2577 static void bond_send_gratuitous_arp(struct bonding *bond)
2578 {
2579         struct slave *slave = bond->curr_active_slave;
2580         struct vlan_entry *vlan;
2581         struct net_device *vlan_dev;
2582
2583         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2584                  bond->dev->name, slave ? slave->dev->name : "NULL");
2585
2586         if (!slave || !bond->send_grat_arp ||
2587             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2588                 return;
2589
2590         bond->send_grat_arp--;
2591
2592         if (bond->master_ip) {
2593                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2594                                 bond->master_ip, 0);
2595         }
2596
2597         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2598                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2599                 if (vlan->vlan_ip) {
2600                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2601                                       vlan->vlan_ip, vlan->vlan_id);
2602                 }
2603         }
2604 }
2605
2606 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2607 {
2608         int i;
2609         __be32 *targets = bond->params.arp_targets;
2610
2611         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2612                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2613                          &sip, &tip, i, &targets[i],
2614                          bond_has_this_ip(bond, tip));
2615                 if (sip == targets[i]) {
2616                         if (bond_has_this_ip(bond, tip))
2617                                 slave->last_arp_rx = jiffies;
2618                         return;
2619                 }
2620         }
2621 }
2622
2623 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2624 {
2625         struct arphdr *arp;
2626         struct slave *slave;
2627         struct bonding *bond;
2628         unsigned char *arp_ptr;
2629         __be32 sip, tip;
2630
2631         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2632                 /*
2633                  * When using VLANS and bonding, dev and oriv_dev may be
2634                  * incorrect if the physical interface supports VLAN
2635                  * acceleration.  With this change ARP validation now
2636                  * works for hosts only reachable on the VLAN interface.
2637                  */
2638                 dev = vlan_dev_real_dev(dev);
2639                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2640         }
2641
2642         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2643                 goto out;
2644
2645         bond = netdev_priv(dev);
2646         read_lock(&bond->lock);
2647
2648         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2649                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2650                  orig_dev ? orig_dev->name : "NULL");
2651
2652         slave = bond_get_slave_by_dev(bond, orig_dev);
2653         if (!slave || !slave_do_arp_validate(bond, slave))
2654                 goto out_unlock;
2655
2656         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2657                 goto out_unlock;
2658
2659         arp = arp_hdr(skb);
2660         if (arp->ar_hln != dev->addr_len ||
2661             skb->pkt_type == PACKET_OTHERHOST ||
2662             skb->pkt_type == PACKET_LOOPBACK ||
2663             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2664             arp->ar_pro != htons(ETH_P_IP) ||
2665             arp->ar_pln != 4)
2666                 goto out_unlock;
2667
2668         arp_ptr = (unsigned char *)(arp + 1);
2669         arp_ptr += dev->addr_len;
2670         memcpy(&sip, arp_ptr, 4);
2671         arp_ptr += 4 + dev->addr_len;
2672         memcpy(&tip, arp_ptr, 4);
2673
2674         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2675                  bond->dev->name, slave->dev->name, slave->state,
2676                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2677                  &sip, &tip);
2678
2679         /*
2680          * Backup slaves won't see the ARP reply, but do come through
2681          * here for each ARP probe (so we swap the sip/tip to validate
2682          * the probe).  In a "redundant switch, common router" type of
2683          * configuration, the ARP probe will (hopefully) travel from
2684          * the active, through one switch, the router, then the other
2685          * switch before reaching the backup.
2686          */
2687         if (slave->state == BOND_STATE_ACTIVE)
2688                 bond_validate_arp(bond, slave, sip, tip);
2689         else
2690                 bond_validate_arp(bond, slave, tip, sip);
2691
2692 out_unlock:
2693         read_unlock(&bond->lock);
2694 out:
2695         dev_kfree_skb(skb);
2696         return NET_RX_SUCCESS;
2697 }
2698
2699 /*
2700  * this function is called regularly to monitor each slave's link
2701  * ensuring that traffic is being sent and received when arp monitoring
2702  * is used in load-balancing mode. if the adapter has been dormant, then an
2703  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2704  * arp monitoring in active backup mode.
2705  */
2706 void bond_loadbalance_arp_mon(struct work_struct *work)
2707 {
2708         struct bonding *bond = container_of(work, struct bonding,
2709                                             arp_work.work);
2710         struct slave *slave, *oldcurrent;
2711         int do_failover = 0;
2712         int delta_in_ticks;
2713         int i;
2714
2715         read_lock(&bond->lock);
2716
2717         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2718
2719         if (bond->kill_timers)
2720                 goto out;
2721
2722         if (bond->slave_cnt == 0)
2723                 goto re_arm;
2724
2725         read_lock(&bond->curr_slave_lock);
2726         oldcurrent = bond->curr_active_slave;
2727         read_unlock(&bond->curr_slave_lock);
2728
2729         /* see if any of the previous devices are up now (i.e. they have
2730          * xmt and rcv traffic). the curr_active_slave does not come into
2731          * the picture unless it is null. also, slave->jiffies is not needed
2732          * here because we send an arp on each slave and give a slave as
2733          * long as it needs to get the tx/rx within the delta.
2734          * TODO: what about up/down delay in arp mode? it wasn't here before
2735          *       so it can wait
2736          */
2737         bond_for_each_slave(bond, slave, i) {
2738                 if (slave->link != BOND_LINK_UP) {
2739                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2740                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2741
2742                                 slave->link  = BOND_LINK_UP;
2743                                 slave->state = BOND_STATE_ACTIVE;
2744
2745                                 /* primary_slave has no meaning in round-robin
2746                                  * mode. the window of a slave being up and
2747                                  * curr_active_slave being null after enslaving
2748                                  * is closed.
2749                                  */
2750                                 if (!oldcurrent) {
2751                                         pr_info("%s: link status definitely up for interface %s, ",
2752                                                 bond->dev->name,
2753                                                 slave->dev->name);
2754                                         do_failover = 1;
2755                                 } else {
2756                                         pr_info("%s: interface %s is now up\n",
2757                                                 bond->dev->name,
2758                                                 slave->dev->name);
2759                                 }
2760                         }
2761                 } else {
2762                         /* slave->link == BOND_LINK_UP */
2763
2764                         /* not all switches will respond to an arp request
2765                          * when the source ip is 0, so don't take the link down
2766                          * if we don't know our ip yet
2767                          */
2768                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2769                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2770
2771                                 slave->link  = BOND_LINK_DOWN;
2772                                 slave->state = BOND_STATE_BACKUP;
2773
2774                                 if (slave->link_failure_count < UINT_MAX)
2775                                         slave->link_failure_count++;
2776
2777                                 pr_info("%s: interface %s is now down.\n",
2778                                         bond->dev->name,
2779                                         slave->dev->name);
2780
2781                                 if (slave == oldcurrent)
2782                                         do_failover = 1;
2783                         }
2784                 }
2785
2786                 /* note: if switch is in round-robin mode, all links
2787                  * must tx arp to ensure all links rx an arp - otherwise
2788                  * links may oscillate or not come up at all; if switch is
2789                  * in something like xor mode, there is nothing we can
2790                  * do - all replies will be rx'ed on same link causing slaves
2791                  * to be unstable during low/no traffic periods
2792                  */
2793                 if (IS_UP(slave->dev))
2794                         bond_arp_send_all(bond, slave);
2795         }
2796
2797         if (do_failover) {
2798                 write_lock_bh(&bond->curr_slave_lock);
2799
2800                 bond_select_active_slave(bond);
2801
2802                 write_unlock_bh(&bond->curr_slave_lock);
2803         }
2804
2805 re_arm:
2806         if (bond->params.arp_interval)
2807                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2808 out:
2809         read_unlock(&bond->lock);
2810 }
2811
2812 /*
2813  * Called to inspect slaves for active-backup mode ARP monitor link state
2814  * changes.  Sets new_link in slaves to specify what action should take
2815  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2816  * to link states must be committed.
2817  *
2818  * Called with bond->lock held for read.
2819  */
2820 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2821 {
2822         struct slave *slave;
2823         int i, commit = 0;
2824
2825         bond_for_each_slave(bond, slave, i) {
2826                 slave->new_link = BOND_LINK_NOCHANGE;
2827
2828                 if (slave->link != BOND_LINK_UP) {
2829                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2830                                            delta_in_ticks)) {
2831                                 slave->new_link = BOND_LINK_UP;
2832                                 commit++;
2833                         }
2834
2835                         continue;
2836                 }
2837
2838                 /*
2839                  * Give slaves 2*delta after being enslaved or made
2840                  * active.  This avoids bouncing, as the last receive
2841                  * times need a full ARP monitor cycle to be updated.
2842                  */
2843                 if (!time_after_eq(jiffies, slave->jiffies +
2844                                    2 * delta_in_ticks))
2845                         continue;
2846
2847                 /*
2848                  * Backup slave is down if:
2849                  * - No current_arp_slave AND
2850                  * - more than 3*delta since last receive AND
2851                  * - the bond has an IP address
2852                  *
2853                  * Note: a non-null current_arp_slave indicates
2854                  * the curr_active_slave went down and we are
2855                  * searching for a new one; under this condition
2856                  * we only take the curr_active_slave down - this
2857                  * gives each slave a chance to tx/rx traffic
2858                  * before being taken out
2859                  */
2860                 if (slave->state == BOND_STATE_BACKUP &&
2861                     !bond->current_arp_slave &&
2862                     time_after(jiffies, slave_last_rx(bond, slave) +
2863                                3 * delta_in_ticks)) {
2864                         slave->new_link = BOND_LINK_DOWN;
2865                         commit++;
2866                 }
2867
2868                 /*
2869                  * Active slave is down if:
2870                  * - more than 2*delta since transmitting OR
2871                  * - (more than 2*delta since receive AND
2872                  *    the bond has an IP address)
2873                  */
2874                 if ((slave->state == BOND_STATE_ACTIVE) &&
2875                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2876                                     2 * delta_in_ticks) ||
2877                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2878                                      + 2 * delta_in_ticks)))) {
2879                         slave->new_link = BOND_LINK_DOWN;
2880                         commit++;
2881                 }
2882         }
2883
2884         return commit;
2885 }
2886
2887 /*
2888  * Called to commit link state changes noted by inspection step of
2889  * active-backup mode ARP monitor.
2890  *
2891  * Called with RTNL and bond->lock for read.
2892  */
2893 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2894 {
2895         struct slave *slave;
2896         int i;
2897
2898         bond_for_each_slave(bond, slave, i) {
2899                 switch (slave->new_link) {
2900                 case BOND_LINK_NOCHANGE:
2901                         continue;
2902
2903                 case BOND_LINK_UP:
2904                         if ((!bond->curr_active_slave &&
2905                              time_before_eq(jiffies,
2906                                             dev_trans_start(slave->dev) +
2907                                             delta_in_ticks)) ||
2908                             bond->curr_active_slave != slave) {
2909                                 slave->link = BOND_LINK_UP;
2910                                 bond->current_arp_slave = NULL;
2911
2912                                 pr_info("%s: link status definitely up for interface %s.\n",
2913                                         bond->dev->name, slave->dev->name);
2914
2915                                 if (!bond->curr_active_slave ||
2916                                     (slave == bond->primary_slave))
2917                                         goto do_failover;
2918
2919                         }
2920
2921                         continue;
2922
2923                 case BOND_LINK_DOWN:
2924                         if (slave->link_failure_count < UINT_MAX)
2925                                 slave->link_failure_count++;
2926
2927                         slave->link = BOND_LINK_DOWN;
2928                         bond_set_slave_inactive_flags(slave);
2929
2930                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2931                                 bond->dev->name, slave->dev->name);
2932
2933                         if (slave == bond->curr_active_slave) {
2934                                 bond->current_arp_slave = NULL;
2935                                 goto do_failover;
2936                         }
2937
2938                         continue;
2939
2940                 default:
2941                         pr_err("%s: impossible: new_link %d on slave %s\n",
2942                                bond->dev->name, slave->new_link,
2943                                slave->dev->name);
2944                         continue;
2945                 }
2946
2947 do_failover:
2948                 ASSERT_RTNL();
2949                 write_lock_bh(&bond->curr_slave_lock);
2950                 bond_select_active_slave(bond);
2951                 write_unlock_bh(&bond->curr_slave_lock);
2952         }
2953
2954         bond_set_carrier(bond);
2955 }
2956
2957 /*
2958  * Send ARP probes for active-backup mode ARP monitor.
2959  *
2960  * Called with bond->lock held for read.
2961  */
2962 static void bond_ab_arp_probe(struct bonding *bond)
2963 {
2964         struct slave *slave;
2965         int i;
2966
2967         read_lock(&bond->curr_slave_lock);
2968
2969         if (bond->current_arp_slave && bond->curr_active_slave)
2970                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2971                         bond->current_arp_slave->dev->name,
2972                         bond->curr_active_slave->dev->name);
2973
2974         if (bond->curr_active_slave) {
2975                 bond_arp_send_all(bond, bond->curr_active_slave);
2976                 read_unlock(&bond->curr_slave_lock);
2977                 return;
2978         }
2979
2980         read_unlock(&bond->curr_slave_lock);
2981
2982         /* if we don't have a curr_active_slave, search for the next available
2983          * backup slave from the current_arp_slave and make it the candidate
2984          * for becoming the curr_active_slave
2985          */
2986
2987         if (!bond->current_arp_slave) {
2988                 bond->current_arp_slave = bond->first_slave;
2989                 if (!bond->current_arp_slave)
2990                         return;
2991         }
2992
2993         bond_set_slave_inactive_flags(bond->current_arp_slave);
2994
2995         /* search for next candidate */
2996         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2997                 if (IS_UP(slave->dev)) {
2998                         slave->link = BOND_LINK_BACK;
2999                         bond_set_slave_active_flags(slave);
3000                         bond_arp_send_all(bond, slave);
3001                         slave->jiffies = jiffies;
3002                         bond->current_arp_slave = slave;
3003                         break;
3004                 }
3005
3006                 /* if the link state is up at this point, we
3007                  * mark it down - this can happen if we have
3008                  * simultaneous link failures and
3009                  * reselect_active_interface doesn't make this
3010                  * one the current slave so it is still marked
3011                  * up when it is actually down
3012                  */
3013                 if (slave->link == BOND_LINK_UP) {
3014                         slave->link = BOND_LINK_DOWN;
3015                         if (slave->link_failure_count < UINT_MAX)
3016                                 slave->link_failure_count++;
3017
3018                         bond_set_slave_inactive_flags(slave);
3019
3020                         pr_info("%s: backup interface %s is now down.\n",
3021                                 bond->dev->name, slave->dev->name);
3022                 }
3023         }
3024 }
3025
3026 void bond_activebackup_arp_mon(struct work_struct *work)
3027 {
3028         struct bonding *bond = container_of(work, struct bonding,
3029                                             arp_work.work);
3030         int delta_in_ticks;
3031
3032         read_lock(&bond->lock);
3033
3034         if (bond->kill_timers)
3035                 goto out;
3036
3037         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3038
3039         if (bond->slave_cnt == 0)
3040                 goto re_arm;
3041
3042         if (bond->send_grat_arp) {
3043                 read_lock(&bond->curr_slave_lock);
3044                 bond_send_gratuitous_arp(bond);
3045                 read_unlock(&bond->curr_slave_lock);
3046         }
3047
3048         if (bond->send_unsol_na) {
3049                 read_lock(&bond->curr_slave_lock);
3050                 bond_send_unsolicited_na(bond);
3051                 read_unlock(&bond->curr_slave_lock);
3052         }
3053
3054         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3055                 read_unlock(&bond->lock);
3056                 rtnl_lock();
3057                 read_lock(&bond->lock);
3058
3059                 bond_ab_arp_commit(bond, delta_in_ticks);
3060
3061                 read_unlock(&bond->lock);
3062                 rtnl_unlock();
3063                 read_lock(&bond->lock);
3064         }
3065
3066         bond_ab_arp_probe(bond);
3067
3068 re_arm:
3069         if (bond->params.arp_interval)
3070                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3071 out:
3072         read_unlock(&bond->lock);
3073 }
3074
3075 /*------------------------------ proc/seq_file-------------------------------*/
3076
3077 #ifdef CONFIG_PROC_FS
3078
3079 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3080         __acquires(&dev_base_lock)
3081         __acquires(&bond->lock)
3082 {
3083         struct bonding *bond = seq->private;
3084         loff_t off = 0;
3085         struct slave *slave;
3086         int i;
3087
3088         /* make sure the bond won't be taken away */
3089         read_lock(&dev_base_lock);
3090         read_lock(&bond->lock);
3091
3092         if (*pos == 0)
3093                 return SEQ_START_TOKEN;
3094
3095         bond_for_each_slave(bond, slave, i) {
3096                 if (++off == *pos)
3097                         return slave;
3098         }
3099
3100         return NULL;
3101 }
3102
3103 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3104 {
3105         struct bonding *bond = seq->private;
3106         struct slave *slave = v;
3107
3108         ++*pos;
3109         if (v == SEQ_START_TOKEN)
3110                 return bond->first_slave;
3111
3112         slave = slave->next;
3113
3114         return (slave == bond->first_slave) ? NULL : slave;
3115 }
3116
3117 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3118         __releases(&bond->lock)
3119         __releases(&dev_base_lock)
3120 {
3121         struct bonding *bond = seq->private;
3122
3123         read_unlock(&bond->lock);
3124         read_unlock(&dev_base_lock);
3125 }
3126
3127 static void bond_info_show_master(struct seq_file *seq)
3128 {
3129         struct bonding *bond = seq->private;
3130         struct slave *curr;
3131         int i;
3132
3133         read_lock(&bond->curr_slave_lock);
3134         curr = bond->curr_active_slave;
3135         read_unlock(&bond->curr_slave_lock);
3136
3137         seq_printf(seq, "Bonding Mode: %s",
3138                    bond_mode_name(bond->params.mode));
3139
3140         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3141             bond->params.fail_over_mac)
3142                 seq_printf(seq, " (fail_over_mac %s)",
3143                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3144
3145         seq_printf(seq, "\n");
3146
3147         if (bond->params.mode == BOND_MODE_XOR ||
3148                 bond->params.mode == BOND_MODE_8023AD) {
3149                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3150                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3151                         bond->params.xmit_policy);
3152         }
3153
3154         if (USES_PRIMARY(bond->params.mode)) {
3155                 seq_printf(seq, "Primary Slave: %s",
3156                            (bond->primary_slave) ?
3157                            bond->primary_slave->dev->name : "None");
3158                 if (bond->primary_slave)
3159                         seq_printf(seq, " (primary_reselect %s)",
3160                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3161
3162                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3163                            (curr) ? curr->dev->name : "None");
3164         }
3165
3166         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3167                    "up" : "down");
3168         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3169         seq_printf(seq, "Up Delay (ms): %d\n",
3170                    bond->params.updelay * bond->params.miimon);
3171         seq_printf(seq, "Down Delay (ms): %d\n",
3172                    bond->params.downdelay * bond->params.miimon);
3173
3174
3175         /* ARP information */
3176         if (bond->params.arp_interval > 0) {
3177                 int printed = 0;
3178                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3179                                 bond->params.arp_interval);
3180
3181                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3182
3183                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3184                         if (!bond->params.arp_targets[i])
3185                                 break;
3186                         if (printed)
3187                                 seq_printf(seq, ",");
3188                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3189                         printed = 1;
3190                 }
3191                 seq_printf(seq, "\n");
3192         }
3193
3194         if (bond->params.mode == BOND_MODE_8023AD) {
3195                 struct ad_info ad_info;
3196
3197                 seq_puts(seq, "\n802.3ad info\n");
3198                 seq_printf(seq, "LACP rate: %s\n",
3199                            (bond->params.lacp_fast) ? "fast" : "slow");
3200                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3201                            ad_select_tbl[bond->params.ad_select].modename);
3202
3203                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3204                         seq_printf(seq, "bond %s has no active aggregator\n",
3205                                    bond->dev->name);
3206                 } else {
3207                         seq_printf(seq, "Active Aggregator Info:\n");
3208
3209                         seq_printf(seq, "\tAggregator ID: %d\n",
3210                                    ad_info.aggregator_id);
3211                         seq_printf(seq, "\tNumber of ports: %d\n",
3212                                    ad_info.ports);
3213                         seq_printf(seq, "\tActor Key: %d\n",
3214                                    ad_info.actor_key);
3215                         seq_printf(seq, "\tPartner Key: %d\n",
3216                                    ad_info.partner_key);
3217                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3218                                    ad_info.partner_system);
3219                 }
3220         }
3221 }
3222
3223 static void bond_info_show_slave(struct seq_file *seq,
3224                                  const struct slave *slave)
3225 {
3226         struct bonding *bond = seq->private;
3227
3228         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3229         seq_printf(seq, "MII Status: %s\n",
3230                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3231         seq_printf(seq, "Link Failure Count: %u\n",
3232                    slave->link_failure_count);
3233
3234         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3235
3236         if (bond->params.mode == BOND_MODE_8023AD) {
3237                 const struct aggregator *agg
3238                         = SLAVE_AD_INFO(slave).port.aggregator;
3239
3240                 if (agg)
3241                         seq_printf(seq, "Aggregator ID: %d\n",
3242                                    agg->aggregator_identifier);
3243                 else
3244                         seq_puts(seq, "Aggregator ID: N/A\n");
3245         }
3246 }
3247
3248 static int bond_info_seq_show(struct seq_file *seq, void *v)
3249 {
3250         if (v == SEQ_START_TOKEN) {
3251                 seq_printf(seq, "%s\n", version);
3252                 bond_info_show_master(seq);
3253         } else
3254                 bond_info_show_slave(seq, v);
3255
3256         return 0;
3257 }
3258
3259 static const struct seq_operations bond_info_seq_ops = {
3260         .start = bond_info_seq_start,
3261         .next  = bond_info_seq_next,
3262         .stop  = bond_info_seq_stop,
3263         .show  = bond_info_seq_show,
3264 };
3265
3266 static int bond_info_open(struct inode *inode, struct file *file)
3267 {
3268         struct seq_file *seq;
3269         struct proc_dir_entry *proc;
3270         int res;
3271
3272         res = seq_open(file, &bond_info_seq_ops);
3273         if (!res) {
3274                 /* recover the pointer buried in proc_dir_entry data */
3275                 seq = file->private_data;
3276                 proc = PDE(inode);
3277                 seq->private = proc->data;
3278         }
3279
3280         return res;
3281 }
3282
3283 static const struct file_operations bond_info_fops = {
3284         .owner   = THIS_MODULE,
3285         .open    = bond_info_open,
3286         .read    = seq_read,
3287         .llseek  = seq_lseek,
3288         .release = seq_release,
3289 };
3290
3291 static void bond_create_proc_entry(struct bonding *bond)
3292 {
3293         struct net_device *bond_dev = bond->dev;
3294         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3295
3296         if (bn->proc_dir) {
3297                 bond->proc_entry = proc_create_data(bond_dev->name,
3298                                                     S_IRUGO, bn->proc_dir,
3299                                                     &bond_info_fops, bond);
3300                 if (bond->proc_entry == NULL)
3301                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3302                                    DRV_NAME, bond_dev->name);
3303                 else
3304                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3305         }
3306 }
3307
3308 static void bond_remove_proc_entry(struct bonding *bond)
3309 {
3310         struct net_device *bond_dev = bond->dev;
3311         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3312
3313         if (bn->proc_dir && bond->proc_entry) {
3314                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3315                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3316                 bond->proc_entry = NULL;
3317         }
3318 }
3319
3320 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3321  * Caller must hold rtnl_lock.
3322  */
3323 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3324 {
3325         if (!bn->proc_dir) {
3326                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3327                 if (!bn->proc_dir)
3328                         pr_warning("Warning: cannot create /proc/net/%s\n",
3329                                    DRV_NAME);
3330         }
3331 }
3332
3333 /* Destroy the bonding directory under /proc/net, if empty.
3334  * Caller must hold rtnl_lock.
3335  */
3336 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3337 {
3338         if (bn->proc_dir) {
3339                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3340                 bn->proc_dir = NULL;
3341         }
3342 }
3343
3344 #else /* !CONFIG_PROC_FS */
3345
3346 static void bond_create_proc_entry(struct bonding *bond)
3347 {
3348 }
3349
3350 static void bond_remove_proc_entry(struct bonding *bond)
3351 {
3352 }
3353
3354 static inline void bond_create_proc_dir(struct bond_net *bn)
3355 {
3356 }
3357
3358 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3359 {
3360 }
3361
3362 #endif /* CONFIG_PROC_FS */
3363
3364
3365 /*-------------------------- netdev event handling --------------------------*/
3366
3367 /*
3368  * Change device name
3369  */
3370 static int bond_event_changename(struct bonding *bond)
3371 {
3372         bond_remove_proc_entry(bond);
3373         bond_create_proc_entry(bond);
3374
3375         return NOTIFY_DONE;
3376 }
3377
3378 static int bond_master_netdev_event(unsigned long event,
3379                                     struct net_device *bond_dev)
3380 {
3381         struct bonding *event_bond = netdev_priv(bond_dev);
3382
3383         switch (event) {
3384         case NETDEV_CHANGENAME:
3385                 return bond_event_changename(event_bond);
3386         default:
3387                 break;
3388         }
3389
3390         return NOTIFY_DONE;
3391 }
3392
3393 static int bond_slave_netdev_event(unsigned long event,
3394                                    struct net_device *slave_dev)
3395 {
3396         struct net_device *bond_dev = slave_dev->master;
3397         struct bonding *bond = netdev_priv(bond_dev);
3398
3399         switch (event) {
3400         case NETDEV_UNREGISTER:
3401                 if (bond_dev) {
3402                         if (bond->setup_by_slave)
3403                                 bond_release_and_destroy(bond_dev, slave_dev);
3404                         else
3405                                 bond_release(bond_dev, slave_dev);
3406                 }
3407                 break;
3408         case NETDEV_CHANGE:
3409                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3410                         struct slave *slave;
3411
3412                         slave = bond_get_slave_by_dev(bond, slave_dev);
3413                         if (slave) {
3414                                 u16 old_speed = slave->speed;
3415                                 u16 old_duplex = slave->duplex;
3416
3417                                 bond_update_speed_duplex(slave);
3418
3419                                 if (bond_is_lb(bond))
3420                                         break;
3421
3422                                 if (old_speed != slave->speed)
3423                                         bond_3ad_adapter_speed_changed(slave);
3424                                 if (old_duplex != slave->duplex)
3425                                         bond_3ad_adapter_duplex_changed(slave);
3426                         }
3427                 }
3428
3429                 break;
3430         case NETDEV_DOWN:
3431                 /*
3432                  * ... Or is it this?
3433                  */
3434                 break;
3435         case NETDEV_CHANGEMTU:
3436                 /*
3437                  * TODO: Should slaves be allowed to
3438                  * independently alter their MTU?  For
3439                  * an active-backup bond, slaves need
3440                  * not be the same type of device, so
3441                  * MTUs may vary.  For other modes,
3442                  * slaves arguably should have the
3443                  * same MTUs. To do this, we'd need to
3444                  * take over the slave's change_mtu
3445                  * function for the duration of their
3446                  * servitude.
3447                  */
3448                 break;
3449         case NETDEV_CHANGENAME:
3450                 /*
3451                  * TODO: handle changing the primary's name
3452                  */
3453                 break;
3454         case NETDEV_FEAT_CHANGE:
3455                 bond_compute_features(bond);
3456                 break;
3457         default:
3458                 break;
3459         }
3460
3461         return NOTIFY_DONE;
3462 }
3463
3464 /*
3465  * bond_netdev_event: handle netdev notifier chain events.
3466  *
3467  * This function receives events for the netdev chain.  The caller (an
3468  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3469  * locks for us to safely manipulate the slave devices (RTNL lock,
3470  * dev_probe_lock).
3471  */
3472 static int bond_netdev_event(struct notifier_block *this,
3473                              unsigned long event, void *ptr)
3474 {
3475         struct net_device *event_dev = (struct net_device *)ptr;
3476
3477         pr_debug("event_dev: %s, event: %lx\n",
3478                  event_dev ? event_dev->name : "None",
3479                  event);
3480
3481         if (!(event_dev->priv_flags & IFF_BONDING))
3482                 return NOTIFY_DONE;
3483
3484         if (event_dev->flags & IFF_MASTER) {
3485                 pr_debug("IFF_MASTER\n");
3486                 return bond_master_netdev_event(event, event_dev);
3487         }
3488
3489         if (event_dev->flags & IFF_SLAVE) {
3490                 pr_debug("IFF_SLAVE\n");
3491                 return bond_slave_netdev_event(event, event_dev);
3492         }
3493
3494         return NOTIFY_DONE;
3495 }
3496
3497 /*
3498  * bond_inetaddr_event: handle inetaddr notifier chain events.
3499  *
3500  * We keep track of device IPs primarily to use as source addresses in
3501  * ARP monitor probes (rather than spewing out broadcasts all the time).
3502  *
3503  * We track one IP for the main device (if it has one), plus one per VLAN.
3504  */
3505 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3506 {
3507         struct in_ifaddr *ifa = ptr;
3508         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3509         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3510         struct bonding *bond;
3511         struct vlan_entry *vlan;
3512
3513         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3514                 if (bond->dev == event_dev) {
3515                         switch (event) {
3516                         case NETDEV_UP:
3517                                 bond->master_ip = ifa->ifa_local;
3518                                 return NOTIFY_OK;
3519                         case NETDEV_DOWN:
3520                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3521                                 return NOTIFY_OK;
3522                         default:
3523                                 return NOTIFY_DONE;
3524                         }
3525                 }
3526
3527                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3528                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3529                         if (vlan_dev == event_dev) {
3530                                 switch (event) {
3531                                 case NETDEV_UP:
3532                                         vlan->vlan_ip = ifa->ifa_local;
3533                                         return NOTIFY_OK;
3534                                 case NETDEV_DOWN:
3535                                         vlan->vlan_ip =
3536                                                 bond_glean_dev_ip(vlan_dev);
3537                                         return NOTIFY_OK;
3538                                 default:
3539                                         return NOTIFY_DONE;
3540                                 }
3541                         }
3542                 }
3543         }
3544         return NOTIFY_DONE;
3545 }
3546
3547 static struct notifier_block bond_netdev_notifier = {
3548         .notifier_call = bond_netdev_event,
3549 };
3550
3551 static struct notifier_block bond_inetaddr_notifier = {
3552         .notifier_call = bond_inetaddr_event,
3553 };
3554
3555 /*-------------------------- Packet type handling ---------------------------*/
3556
3557 /* register to receive lacpdus on a bond */
3558 static void bond_register_lacpdu(struct bonding *bond)
3559 {
3560         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3561
3562         /* initialize packet type */
3563         pk_type->type = PKT_TYPE_LACPDU;
3564         pk_type->dev = bond->dev;
3565         pk_type->func = bond_3ad_lacpdu_recv;
3566
3567         dev_add_pack(pk_type);
3568 }
3569
3570 /* unregister to receive lacpdus on a bond */
3571 static void bond_unregister_lacpdu(struct bonding *bond)
3572 {
3573         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3574 }
3575
3576 void bond_register_arp(struct bonding *bond)
3577 {
3578         struct packet_type *pt = &bond->arp_mon_pt;
3579
3580         if (pt->type)
3581                 return;
3582
3583         pt->type = htons(ETH_P_ARP);
3584         pt->dev = bond->dev;
3585         pt->func = bond_arp_rcv;
3586         dev_add_pack(pt);
3587 }
3588
3589 void bond_unregister_arp(struct bonding *bond)
3590 {
3591         struct packet_type *pt = &bond->arp_mon_pt;
3592
3593         dev_remove_pack(pt);
3594         pt->type = 0;
3595 }
3596
3597 /*---------------------------- Hashing Policies -----------------------------*/
3598
3599 /*
3600  * Hash for the output device based upon layer 2 and layer 3 data. If
3601  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3602  */
3603 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3604 {
3605         struct ethhdr *data = (struct ethhdr *)skb->data;
3606         struct iphdr *iph = ip_hdr(skb);
3607
3608         if (skb->protocol == htons(ETH_P_IP)) {
3609                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3610                         (data->h_dest[5] ^ data->h_source[5])) % count;
3611         }
3612
3613         return (data->h_dest[5] ^ data->h_source[5]) % count;
3614 }
3615
3616 /*
3617  * Hash for the output device based upon layer 3 and layer 4 data. If
3618  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3619  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3620  */
3621 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3622 {
3623         struct ethhdr *data = (struct ethhdr *)skb->data;
3624         struct iphdr *iph = ip_hdr(skb);
3625         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3626         int layer4_xor = 0;
3627
3628         if (skb->protocol == htons(ETH_P_IP)) {
3629                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3630                     (iph->protocol == IPPROTO_TCP ||
3631                      iph->protocol == IPPROTO_UDP)) {
3632                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3633                 }
3634                 return (layer4_xor ^
3635                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3636
3637         }
3638
3639         return (data->h_dest[5] ^ data->h_source[5]) % count;
3640 }
3641
3642 /*
3643  * Hash for the output device based upon layer 2 data
3644  */
3645 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3646 {
3647         struct ethhdr *data = (struct ethhdr *)skb->data;
3648
3649         return (data->h_dest[5] ^ data->h_source[5]) % count;
3650 }
3651
3652 /*-------------------------- Device entry points ----------------------------*/
3653
3654 static int bond_open(struct net_device *bond_dev)
3655 {
3656         struct bonding *bond = netdev_priv(bond_dev);
3657
3658         bond->kill_timers = 0;
3659
3660         if (bond_is_lb(bond)) {
3661                 /* bond_alb_initialize must be called before the timer
3662                  * is started.
3663                  */
3664                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3665                         /* something went wrong - fail the open operation */
3666                         return -ENOMEM;
3667                 }
3668
3669                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3670                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3671         }
3672
3673         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3674                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3675                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3676         }
3677
3678         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3679                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3680                         INIT_DELAYED_WORK(&bond->arp_work,
3681                                           bond_activebackup_arp_mon);
3682                 else
3683                         INIT_DELAYED_WORK(&bond->arp_work,
3684                                           bond_loadbalance_arp_mon);
3685
3686                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3687                 if (bond->params.arp_validate)
3688                         bond_register_arp(bond);
3689         }
3690
3691         if (bond->params.mode == BOND_MODE_8023AD) {
3692                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3693                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3694                 /* register to receive LACPDUs */
3695                 bond_register_lacpdu(bond);
3696                 bond_3ad_initiate_agg_selection(bond, 1);
3697         }
3698
3699         return 0;
3700 }
3701
3702 static int bond_close(struct net_device *bond_dev)
3703 {
3704         struct bonding *bond = netdev_priv(bond_dev);
3705
3706         if (bond->params.mode == BOND_MODE_8023AD) {
3707                 /* Unregister the receive of LACPDUs */
3708                 bond_unregister_lacpdu(bond);
3709         }
3710
3711         if (bond->params.arp_validate)
3712                 bond_unregister_arp(bond);
3713
3714         write_lock_bh(&bond->lock);
3715
3716         bond->send_grat_arp = 0;
3717         bond->send_unsol_na = 0;
3718
3719         /* signal timers not to re-arm */
3720         bond->kill_timers = 1;
3721
3722         write_unlock_bh(&bond->lock);
3723
3724         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3725                 cancel_delayed_work(&bond->mii_work);
3726         }
3727
3728         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3729                 cancel_delayed_work(&bond->arp_work);
3730         }
3731
3732         switch (bond->params.mode) {
3733         case BOND_MODE_8023AD:
3734                 cancel_delayed_work(&bond->ad_work);
3735                 break;
3736         case BOND_MODE_TLB:
3737         case BOND_MODE_ALB:
3738                 cancel_delayed_work(&bond->alb_work);
3739                 break;
3740         default:
3741                 break;
3742         }
3743
3744
3745         if (bond_is_lb(bond)) {
3746                 /* Must be called only after all
3747                  * slaves have been released
3748                  */
3749                 bond_alb_deinitialize(bond);
3750         }
3751
3752         return 0;
3753 }
3754
3755 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3756 {
3757         struct bonding *bond = netdev_priv(bond_dev);
3758         struct net_device_stats *stats = &bond_dev->stats;
3759         struct net_device_stats local_stats;
3760         struct slave *slave;
3761         int i;
3762
3763         memset(&local_stats, 0, sizeof(struct net_device_stats));
3764
3765         read_lock_bh(&bond->lock);
3766
3767         bond_for_each_slave(bond, slave, i) {
3768                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3769
3770                 local_stats.rx_packets += sstats->rx_packets;
3771                 local_stats.rx_bytes += sstats->rx_bytes;
3772                 local_stats.rx_errors += sstats->rx_errors;
3773                 local_stats.rx_dropped += sstats->rx_dropped;
3774
3775                 local_stats.tx_packets += sstats->tx_packets;
3776                 local_stats.tx_bytes += sstats->tx_bytes;
3777                 local_stats.tx_errors += sstats->tx_errors;
3778                 local_stats.tx_dropped += sstats->tx_dropped;
3779
3780                 local_stats.multicast += sstats->multicast;
3781                 local_stats.collisions += sstats->collisions;
3782
3783                 local_stats.rx_length_errors += sstats->rx_length_errors;
3784                 local_stats.rx_over_errors += sstats->rx_over_errors;
3785                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3786                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3787                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3788                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3789
3790                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3791                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3792                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3793                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3794                 local_stats.tx_window_errors += sstats->tx_window_errors;
3795         }
3796
3797         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3798
3799         read_unlock_bh(&bond->lock);
3800
3801         return stats;
3802 }
3803
3804 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3805 {
3806         struct net_device *slave_dev = NULL;
3807         struct ifbond k_binfo;
3808         struct ifbond __user *u_binfo = NULL;
3809         struct ifslave k_sinfo;
3810         struct ifslave __user *u_sinfo = NULL;
3811         struct mii_ioctl_data *mii = NULL;
3812         int res = 0;
3813
3814         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3815
3816         switch (cmd) {
3817         case SIOCGMIIPHY:
3818                 mii = if_mii(ifr);
3819                 if (!mii)
3820                         return -EINVAL;
3821
3822                 mii->phy_id = 0;
3823                 /* Fall Through */
3824         case SIOCGMIIREG:
3825                 /*
3826                  * We do this again just in case we were called by SIOCGMIIREG
3827                  * instead of SIOCGMIIPHY.
3828                  */
3829                 mii = if_mii(ifr);
3830                 if (!mii)
3831                         return -EINVAL;
3832
3833
3834                 if (mii->reg_num == 1) {
3835                         struct bonding *bond = netdev_priv(bond_dev);
3836                         mii->val_out = 0;
3837                         read_lock(&bond->lock);
3838                         read_lock(&bond->curr_slave_lock);
3839                         if (netif_carrier_ok(bond->dev))
3840                                 mii->val_out = BMSR_LSTATUS;
3841
3842                         read_unlock(&bond->curr_slave_lock);
3843                         read_unlock(&bond->lock);
3844                 }
3845
3846                 return 0;
3847         case BOND_INFO_QUERY_OLD:
3848         case SIOCBONDINFOQUERY:
3849                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3850
3851                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3852                         return -EFAULT;
3853
3854                 res = bond_info_query(bond_dev, &k_binfo);
3855                 if (res == 0 &&
3856                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3857                         return -EFAULT;
3858
3859                 return res;
3860         case BOND_SLAVE_INFO_QUERY_OLD:
3861         case SIOCBONDSLAVEINFOQUERY:
3862                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3863
3864                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3865                         return -EFAULT;
3866
3867                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3868                 if (res == 0 &&
3869                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3870                         return -EFAULT;
3871
3872                 return res;
3873         default:
3874                 /* Go on */
3875                 break;
3876         }
3877
3878         if (!capable(CAP_NET_ADMIN))
3879                 return -EPERM;
3880
3881         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3882
3883         pr_debug("slave_dev=%p:\n", slave_dev);
3884
3885         if (!slave_dev)
3886                 res = -ENODEV;
3887         else {
3888                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3889                 switch (cmd) {
3890                 case BOND_ENSLAVE_OLD:
3891                 case SIOCBONDENSLAVE:
3892                         res = bond_enslave(bond_dev, slave_dev);
3893                         break;
3894                 case BOND_RELEASE_OLD:
3895                 case SIOCBONDRELEASE:
3896                         res = bond_release(bond_dev, slave_dev);
3897                         break;
3898                 case BOND_SETHWADDR_OLD:
3899                 case SIOCBONDSETHWADDR:
3900                         res = bond_sethwaddr(bond_dev, slave_dev);
3901                         break;
3902                 case BOND_CHANGE_ACTIVE_OLD:
3903                 case SIOCBONDCHANGEACTIVE:
3904                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3905                         break;
3906                 default:
3907                         res = -EOPNOTSUPP;
3908                 }
3909
3910                 dev_put(slave_dev);
3911         }
3912
3913         return res;
3914 }
3915
3916 static void bond_set_multicast_list(struct net_device *bond_dev)
3917 {
3918         struct bonding *bond = netdev_priv(bond_dev);
3919         struct dev_mc_list *dmi;
3920
3921         /*
3922          * Do promisc before checking multicast_mode
3923          */
3924         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3925                 /*
3926                  * FIXME: Need to handle the error when one of the multi-slaves
3927                  * encounters error.
3928                  */
3929                 bond_set_promiscuity(bond, 1);
3930
3931
3932         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3933                 bond_set_promiscuity(bond, -1);
3934
3935
3936         /* set allmulti flag to slaves */
3937         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3938                 /*
3939                  * FIXME: Need to handle the error when one of the multi-slaves
3940                  * encounters error.
3941                  */
3942                 bond_set_allmulti(bond, 1);
3943
3944
3945         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3946                 bond_set_allmulti(bond, -1);
3947
3948
3949         read_lock(&bond->lock);
3950
3951         bond->flags = bond_dev->flags;
3952
3953         /* looking for addresses to add to slaves' mc list */
3954         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3955                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
3956                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3957         }
3958
3959         /* looking for addresses to delete from slaves' list */
3960         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3961                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
3962                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3963         }
3964
3965         /* save master's multicast list */
3966         bond_mc_list_destroy(bond);
3967         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3968
3969         read_unlock(&bond->lock);
3970 }
3971
3972 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3973 {
3974         struct bonding *bond = netdev_priv(dev);
3975         struct slave *slave = bond->first_slave;
3976
3977         if (slave) {
3978                 const struct net_device_ops *slave_ops
3979                         = slave->dev->netdev_ops;
3980                 if (slave_ops->ndo_neigh_setup)
3981                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3982         }
3983         return 0;
3984 }
3985
3986 /*
3987  * Change the MTU of all of a master's slaves to match the master
3988  */
3989 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3990 {
3991         struct bonding *bond = netdev_priv(bond_dev);
3992         struct slave *slave, *stop_at;
3993         int res = 0;
3994         int i;
3995
3996         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3997                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3998
3999         /* Can't hold bond->lock with bh disabled here since
4000          * some base drivers panic. On the other hand we can't
4001          * hold bond->lock without bh disabled because we'll
4002          * deadlock. The only solution is to rely on the fact
4003          * that we're under rtnl_lock here, and the slaves
4004          * list won't change. This doesn't solve the problem
4005          * of setting the slave's MTU while it is
4006          * transmitting, but the assumption is that the base
4007          * driver can handle that.
4008          *
4009          * TODO: figure out a way to safely iterate the slaves
4010          * list, but without holding a lock around the actual
4011          * call to the base driver.
4012          */
4013
4014         bond_for_each_slave(bond, slave, i) {
4015                 pr_debug("s %p s->p %p c_m %p\n",
4016                          slave,
4017                          slave->prev,
4018                          slave->dev->netdev_ops->ndo_change_mtu);
4019
4020                 res = dev_set_mtu(slave->dev, new_mtu);
4021
4022                 if (res) {
4023                         /* If we failed to set the slave's mtu to the new value
4024                          * we must abort the operation even in ACTIVE_BACKUP
4025                          * mode, because if we allow the backup slaves to have
4026                          * different mtu values than the active slave we'll
4027                          * need to change their mtu when doing a failover. That
4028                          * means changing their mtu from timer context, which
4029                          * is probably not a good idea.
4030                          */
4031                         pr_debug("err %d %s\n", res, slave->dev->name);
4032                         goto unwind;
4033                 }
4034         }
4035
4036         bond_dev->mtu = new_mtu;
4037
4038         return 0;
4039
4040 unwind:
4041         /* unwind from head to the slave that failed */
4042         stop_at = slave;
4043         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4044                 int tmp_res;
4045
4046                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4047                 if (tmp_res) {
4048                         pr_debug("unwind err %d dev %s\n",
4049                                  tmp_res, slave->dev->name);
4050                 }
4051         }
4052
4053         return res;
4054 }
4055
4056 /*
4057  * Change HW address
4058  *
4059  * Note that many devices must be down to change the HW address, and
4060  * downing the master releases all slaves.  We can make bonds full of
4061  * bonding devices to test this, however.
4062  */
4063 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4064 {
4065         struct bonding *bond = netdev_priv(bond_dev);
4066         struct sockaddr *sa = addr, tmp_sa;
4067         struct slave *slave, *stop_at;
4068         int res = 0;
4069         int i;
4070
4071         if (bond->params.mode == BOND_MODE_ALB)
4072                 return bond_alb_set_mac_address(bond_dev, addr);
4073
4074
4075         pr_debug("bond=%p, name=%s\n",
4076                  bond, bond_dev ? bond_dev->name : "None");
4077
4078         /*
4079          * If fail_over_mac is set to active, do nothing and return
4080          * success.  Returning an error causes ifenslave to fail.
4081          */
4082         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4083                 return 0;
4084
4085         if (!is_valid_ether_addr(sa->sa_data))
4086                 return -EADDRNOTAVAIL;
4087
4088         /* Can't hold bond->lock with bh disabled here since
4089          * some base drivers panic. On the other hand we can't
4090          * hold bond->lock without bh disabled because we'll
4091          * deadlock. The only solution is to rely on the fact
4092          * that we're under rtnl_lock here, and the slaves
4093          * list won't change. This doesn't solve the problem
4094          * of setting the slave's hw address while it is
4095          * transmitting, but the assumption is that the base
4096          * driver can handle that.
4097          *
4098          * TODO: figure out a way to safely iterate the slaves
4099          * list, but without holding a lock around the actual
4100          * call to the base driver.
4101          */
4102
4103         bond_for_each_slave(bond, slave, i) {
4104                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4105                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4106
4107                 if (slave_ops->ndo_set_mac_address == NULL) {
4108                         res = -EOPNOTSUPP;
4109                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4110                         goto unwind;
4111                 }
4112
4113                 res = dev_set_mac_address(slave->dev, addr);
4114                 if (res) {
4115                         /* TODO: consider downing the slave
4116                          * and retry ?
4117                          * User should expect communications
4118                          * breakage anyway until ARP finish
4119                          * updating, so...
4120                          */
4121                         pr_debug("err %d %s\n", res, slave->dev->name);
4122                         goto unwind;
4123                 }
4124         }
4125
4126         /* success */
4127         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4128         return 0;
4129
4130 unwind:
4131         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4132         tmp_sa.sa_family = bond_dev->type;
4133
4134         /* unwind from head to the slave that failed */
4135         stop_at = slave;
4136         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4137                 int tmp_res;
4138
4139                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4140                 if (tmp_res) {
4141                         pr_debug("unwind err %d dev %s\n",
4142                                  tmp_res, slave->dev->name);
4143                 }
4144         }
4145
4146         return res;
4147 }
4148
4149 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4150 {
4151         struct bonding *bond = netdev_priv(bond_dev);
4152         struct slave *slave, *start_at;
4153         int i, slave_no, res = 1;
4154
4155         read_lock(&bond->lock);
4156
4157         if (!BOND_IS_OK(bond))
4158                 goto out;
4159
4160         /*
4161          * Concurrent TX may collide on rr_tx_counter; we accept that
4162          * as being rare enough not to justify using an atomic op here
4163          */
4164         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4165
4166         bond_for_each_slave(bond, slave, i) {
4167                 slave_no--;
4168                 if (slave_no < 0)
4169                         break;
4170         }
4171
4172         start_at = slave;
4173         bond_for_each_slave_from(bond, slave, i, start_at) {
4174                 if (IS_UP(slave->dev) &&
4175                     (slave->link == BOND_LINK_UP) &&
4176                     (slave->state == BOND_STATE_ACTIVE)) {
4177                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4178                         break;
4179                 }
4180         }
4181
4182 out:
4183         if (res) {
4184                 /* no suitable interface, frame not sent */
4185                 dev_kfree_skb(skb);
4186         }
4187         read_unlock(&bond->lock);
4188         return NETDEV_TX_OK;
4189 }
4190
4191
4192 /*
4193  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4194  * the bond has a usable interface.
4195  */
4196 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4197 {
4198         struct bonding *bond = netdev_priv(bond_dev);
4199         int res = 1;
4200
4201         read_lock(&bond->lock);
4202         read_lock(&bond->curr_slave_lock);
4203
4204         if (!BOND_IS_OK(bond))
4205                 goto out;
4206
4207         if (!bond->curr_active_slave)
4208                 goto out;
4209
4210         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4211
4212 out:
4213         if (res)
4214                 /* no suitable interface, frame not sent */
4215                 dev_kfree_skb(skb);
4216
4217         read_unlock(&bond->curr_slave_lock);
4218         read_unlock(&bond->lock);
4219         return NETDEV_TX_OK;
4220 }
4221
4222 /*
4223  * In bond_xmit_xor() , we determine the output device by using a pre-
4224  * determined xmit_hash_policy(), If the selected device is not enabled,
4225  * find the next active slave.
4226  */
4227 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4228 {
4229         struct bonding *bond = netdev_priv(bond_dev);
4230         struct slave *slave, *start_at;
4231         int slave_no;
4232         int i;
4233         int res = 1;
4234
4235         read_lock(&bond->lock);
4236
4237         if (!BOND_IS_OK(bond))
4238                 goto out;
4239
4240         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4241
4242         bond_for_each_slave(bond, slave, i) {
4243                 slave_no--;
4244                 if (slave_no < 0)
4245                         break;
4246         }
4247
4248         start_at = slave;
4249
4250         bond_for_each_slave_from(bond, slave, i, start_at) {
4251                 if (IS_UP(slave->dev) &&
4252                     (slave->link == BOND_LINK_UP) &&
4253                     (slave->state == BOND_STATE_ACTIVE)) {
4254                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4255                         break;
4256                 }
4257         }
4258
4259 out:
4260         if (res) {
4261                 /* no suitable interface, frame not sent */
4262                 dev_kfree_skb(skb);
4263         }
4264         read_unlock(&bond->lock);
4265         return NETDEV_TX_OK;
4266 }
4267
4268 /*
4269  * in broadcast mode, we send everything to all usable interfaces.
4270  */
4271 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4272 {
4273         struct bonding *bond = netdev_priv(bond_dev);
4274         struct slave *slave, *start_at;
4275         struct net_device *tx_dev = NULL;
4276         int i;
4277         int res = 1;
4278
4279         read_lock(&bond->lock);
4280
4281         if (!BOND_IS_OK(bond))
4282                 goto out;
4283
4284         read_lock(&bond->curr_slave_lock);
4285         start_at = bond->curr_active_slave;
4286         read_unlock(&bond->curr_slave_lock);
4287
4288         if (!start_at)
4289                 goto out;
4290
4291         bond_for_each_slave_from(bond, slave, i, start_at) {
4292                 if (IS_UP(slave->dev) &&
4293                     (slave->link == BOND_LINK_UP) &&
4294                     (slave->state == BOND_STATE_ACTIVE)) {
4295                         if (tx_dev) {
4296                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4297                                 if (!skb2) {
4298                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4299                                                bond_dev->name);
4300                                         continue;
4301                                 }
4302
4303                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4304                                 if (res) {
4305                                         dev_kfree_skb(skb2);
4306                                         continue;
4307                                 }
4308                         }
4309                         tx_dev = slave->dev;
4310                 }
4311         }
4312
4313         if (tx_dev)
4314                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4315
4316 out:
4317         if (res)
4318                 /* no suitable interface, frame not sent */
4319                 dev_kfree_skb(skb);
4320
4321         /* frame sent to all suitable interfaces */
4322         read_unlock(&bond->lock);
4323         return NETDEV_TX_OK;
4324 }
4325
4326 /*------------------------- Device initialization ---------------------------*/
4327
4328 static void bond_set_xmit_hash_policy(struct bonding *bond)
4329 {
4330         switch (bond->params.xmit_policy) {
4331         case BOND_XMIT_POLICY_LAYER23:
4332                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4333                 break;
4334         case BOND_XMIT_POLICY_LAYER34:
4335                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4336                 break;
4337         case BOND_XMIT_POLICY_LAYER2:
4338         default:
4339                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4340                 break;
4341         }
4342 }
4343
4344 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4345 {
4346         const struct bonding *bond = netdev_priv(dev);
4347
4348         switch (bond->params.mode) {
4349         case BOND_MODE_ROUNDROBIN:
4350                 return bond_xmit_roundrobin(skb, dev);
4351         case BOND_MODE_ACTIVEBACKUP:
4352                 return bond_xmit_activebackup(skb, dev);
4353         case BOND_MODE_XOR:
4354                 return bond_xmit_xor(skb, dev);
4355         case BOND_MODE_BROADCAST:
4356                 return bond_xmit_broadcast(skb, dev);
4357         case BOND_MODE_8023AD:
4358                 return bond_3ad_xmit_xor(skb, dev);
4359         case BOND_MODE_ALB:
4360         case BOND_MODE_TLB:
4361                 return bond_alb_xmit(skb, dev);
4362         default:
4363                 /* Should never happen, mode already checked */
4364                 pr_err("%s: Error: Unknown bonding mode %d\n",
4365                        dev->name, bond->params.mode);
4366                 WARN_ON_ONCE(1);
4367                 dev_kfree_skb(skb);
4368                 return NETDEV_TX_OK;
4369         }
4370 }
4371
4372
4373 /*
4374  * set bond mode specific net device operations
4375  */
4376 void bond_set_mode_ops(struct bonding *bond, int mode)
4377 {
4378         struct net_device *bond_dev = bond->dev;
4379
4380         switch (mode) {
4381         case BOND_MODE_ROUNDROBIN:
4382                 break;
4383         case BOND_MODE_ACTIVEBACKUP:
4384                 break;
4385         case BOND_MODE_XOR:
4386                 bond_set_xmit_hash_policy(bond);
4387                 break;
4388         case BOND_MODE_BROADCAST:
4389                 break;
4390         case BOND_MODE_8023AD:
4391                 bond_set_master_3ad_flags(bond);
4392                 bond_set_xmit_hash_policy(bond);
4393                 break;
4394         case BOND_MODE_ALB:
4395                 bond_set_master_alb_flags(bond);
4396                 /* FALLTHRU */
4397         case BOND_MODE_TLB:
4398                 break;
4399         default:
4400                 /* Should never happen, mode already checked */
4401                 pr_err("%s: Error: Unknown bonding mode %d\n",
4402                        bond_dev->name, mode);
4403                 break;
4404         }
4405 }
4406
4407 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4408                                     struct ethtool_drvinfo *drvinfo)
4409 {
4410         strncpy(drvinfo->driver, DRV_NAME, 32);
4411         strncpy(drvinfo->version, DRV_VERSION, 32);
4412         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4413 }
4414
4415 static const struct ethtool_ops bond_ethtool_ops = {
4416         .get_drvinfo            = bond_ethtool_get_drvinfo,
4417         .get_link               = ethtool_op_get_link,
4418         .get_tx_csum            = ethtool_op_get_tx_csum,
4419         .get_sg                 = ethtool_op_get_sg,
4420         .get_tso                = ethtool_op_get_tso,
4421         .get_ufo                = ethtool_op_get_ufo,
4422         .get_flags              = ethtool_op_get_flags,
4423 };
4424
4425 static const struct net_device_ops bond_netdev_ops = {
4426         .ndo_init               = bond_init,
4427         .ndo_uninit             = bond_uninit,
4428         .ndo_open               = bond_open,
4429         .ndo_stop               = bond_close,
4430         .ndo_start_xmit         = bond_start_xmit,
4431         .ndo_get_stats          = bond_get_stats,
4432         .ndo_do_ioctl           = bond_do_ioctl,
4433         .ndo_set_multicast_list = bond_set_multicast_list,
4434         .ndo_change_mtu         = bond_change_mtu,
4435         .ndo_set_mac_address    = bond_set_mac_address,
4436         .ndo_neigh_setup        = bond_neigh_setup,
4437         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4438         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4439         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4440 };
4441
4442 static void bond_setup(struct net_device *bond_dev)
4443 {
4444         struct bonding *bond = netdev_priv(bond_dev);
4445
4446         /* initialize rwlocks */
4447         rwlock_init(&bond->lock);
4448         rwlock_init(&bond->curr_slave_lock);
4449
4450         bond->params = bonding_defaults;
4451
4452         /* Initialize pointers */
4453         bond->dev = bond_dev;
4454         INIT_LIST_HEAD(&bond->vlan_list);
4455
4456         /* Initialize the device entry points */
4457         ether_setup(bond_dev);
4458         bond_dev->netdev_ops = &bond_netdev_ops;
4459         bond_dev->ethtool_ops = &bond_ethtool_ops;
4460         bond_set_mode_ops(bond, bond->params.mode);
4461
4462         bond_dev->destructor = free_netdev;
4463
4464         /* Initialize the device options */
4465         bond_dev->tx_queue_len = 0;
4466         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4467         bond_dev->priv_flags |= IFF_BONDING;
4468         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4469
4470         if (bond->params.arp_interval)
4471                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4472
4473         /* At first, we block adding VLANs. That's the only way to
4474          * prevent problems that occur when adding VLANs over an
4475          * empty bond. The block will be removed once non-challenged
4476          * slaves are enslaved.
4477          */
4478         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4479
4480         /* don't acquire bond device's netif_tx_lock when
4481          * transmitting */
4482         bond_dev->features |= NETIF_F_LLTX;
4483
4484         /* By default, we declare the bond to be fully
4485          * VLAN hardware accelerated capable. Special
4486          * care is taken in the various xmit functions
4487          * when there are slaves that are not hw accel
4488          * capable
4489          */
4490         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4491                                NETIF_F_HW_VLAN_RX |
4492                                NETIF_F_HW_VLAN_FILTER);
4493
4494 }
4495
4496 static void bond_work_cancel_all(struct bonding *bond)
4497 {
4498         write_lock_bh(&bond->lock);
4499         bond->kill_timers = 1;
4500         write_unlock_bh(&bond->lock);
4501
4502         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4503                 cancel_delayed_work(&bond->mii_work);
4504
4505         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4506                 cancel_delayed_work(&bond->arp_work);
4507
4508         if (bond->params.mode == BOND_MODE_ALB &&
4509             delayed_work_pending(&bond->alb_work))
4510                 cancel_delayed_work(&bond->alb_work);
4511
4512         if (bond->params.mode == BOND_MODE_8023AD &&
4513             delayed_work_pending(&bond->ad_work))
4514                 cancel_delayed_work(&bond->ad_work);
4515 }
4516
4517 /*
4518 * Destroy a bonding device.
4519 * Must be under rtnl_lock when this function is called.
4520 */
4521 static void bond_uninit(struct net_device *bond_dev)
4522 {
4523         struct bonding *bond = netdev_priv(bond_dev);
4524
4525         /* Release the bonded slaves */
4526         bond_release_all(bond_dev);
4527
4528         list_del(&bond->bond_list);
4529
4530         bond_work_cancel_all(bond);
4531
4532         bond_remove_proc_entry(bond);
4533
4534         if (bond->wq)
4535                 destroy_workqueue(bond->wq);
4536
4537         netif_addr_lock_bh(bond_dev);
4538         bond_mc_list_destroy(bond);
4539         netif_addr_unlock_bh(bond_dev);
4540 }
4541
4542 /*------------------------- Module initialization ---------------------------*/
4543
4544 /*
4545  * Convert string input module parms.  Accept either the
4546  * number of the mode or its string name.  A bit complicated because
4547  * some mode names are substrings of other names, and calls from sysfs
4548  * may have whitespace in the name (trailing newlines, for example).
4549  */
4550 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4551 {
4552         int modeint = -1, i, rv;
4553         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4554
4555         for (p = (char *)buf; *p; p++)
4556                 if (!(isdigit(*p) || isspace(*p)))
4557                         break;
4558
4559         if (*p)
4560                 rv = sscanf(buf, "%20s", modestr);
4561         else
4562                 rv = sscanf(buf, "%d", &modeint);
4563
4564         if (!rv)
4565                 return -1;
4566
4567         for (i = 0; tbl[i].modename; i++) {
4568                 if (modeint == tbl[i].mode)
4569                         return tbl[i].mode;
4570                 if (strcmp(modestr, tbl[i].modename) == 0)
4571                         return tbl[i].mode;
4572         }
4573
4574         return -1;
4575 }
4576
4577 static int bond_check_params(struct bond_params *params)
4578 {
4579         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4580
4581         /*
4582          * Convert string parameters.
4583          */
4584         if (mode) {
4585                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4586                 if (bond_mode == -1) {
4587                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4588                                mode == NULL ? "NULL" : mode);
4589                         return -EINVAL;
4590                 }
4591         }
4592
4593         if (xmit_hash_policy) {
4594                 if ((bond_mode != BOND_MODE_XOR) &&
4595                     (bond_mode != BOND_MODE_8023AD)) {
4596                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4597                                bond_mode_name(bond_mode));
4598                 } else {
4599                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4600                                                         xmit_hashtype_tbl);
4601                         if (xmit_hashtype == -1) {
4602                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4603                                        xmit_hash_policy == NULL ? "NULL" :
4604                                        xmit_hash_policy);
4605                                 return -EINVAL;
4606                         }
4607                 }
4608         }
4609
4610         if (lacp_rate) {
4611                 if (bond_mode != BOND_MODE_8023AD) {
4612                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4613                                 bond_mode_name(bond_mode));
4614                 } else {
4615                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4616                         if (lacp_fast == -1) {
4617                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4618                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4619                                 return -EINVAL;
4620                         }
4621                 }
4622         }
4623
4624         if (ad_select) {
4625                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4626                 if (params->ad_select == -1) {
4627                         pr_err("Error: Invalid ad_select \"%s\"\n",
4628                                ad_select == NULL ? "NULL" : ad_select);
4629                         return -EINVAL;
4630                 }
4631
4632                 if (bond_mode != BOND_MODE_8023AD) {
4633                         pr_warning("ad_select param only affects 802.3ad mode\n");
4634                 }
4635         } else {
4636                 params->ad_select = BOND_AD_STABLE;
4637         }
4638
4639         if (max_bonds < 0) {
4640                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4641                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4642                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4643         }
4644
4645         if (miimon < 0) {
4646                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4647                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4648                 miimon = BOND_LINK_MON_INTERV;
4649         }
4650
4651         if (updelay < 0) {
4652                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4653                            updelay, INT_MAX);
4654                 updelay = 0;
4655         }
4656
4657         if (downdelay < 0) {
4658                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4659                            downdelay, INT_MAX);
4660                 downdelay = 0;
4661         }
4662
4663         if ((use_carrier != 0) && (use_carrier != 1)) {
4664                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4665                            use_carrier);
4666                 use_carrier = 1;
4667         }
4668
4669         if (num_grat_arp < 0 || num_grat_arp > 255) {
4670                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4671                            num_grat_arp);
4672                 num_grat_arp = 1;
4673         }
4674
4675         if (num_unsol_na < 0 || num_unsol_na > 255) {
4676                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4677                            num_unsol_na);
4678                 num_unsol_na = 1;
4679         }
4680
4681         /* reset values for 802.3ad */
4682         if (bond_mode == BOND_MODE_8023AD) {
4683                 if (!miimon) {
4684                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4685                         pr_warning("Forcing miimon to 100msec\n");
4686                         miimon = 100;
4687                 }
4688         }
4689
4690         /* reset values for TLB/ALB */
4691         if ((bond_mode == BOND_MODE_TLB) ||
4692             (bond_mode == BOND_MODE_ALB)) {
4693                 if (!miimon) {
4694                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4695                         pr_warning("Forcing miimon to 100msec\n");
4696                         miimon = 100;
4697                 }
4698         }
4699
4700         if (bond_mode == BOND_MODE_ALB) {
4701                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4702                           updelay);
4703         }
4704
4705         if (!miimon) {
4706                 if (updelay || downdelay) {
4707                         /* just warn the user the up/down delay will have
4708                          * no effect since miimon is zero...
4709                          */
4710                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4711                                    updelay, downdelay);
4712                 }
4713         } else {
4714                 /* don't allow arp monitoring */
4715                 if (arp_interval) {
4716                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4717                                    miimon, arp_interval);
4718                         arp_interval = 0;
4719                 }
4720
4721                 if ((updelay % miimon) != 0) {
4722                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4723                                    updelay, miimon,
4724                                    (updelay / miimon) * miimon);
4725                 }
4726
4727                 updelay /= miimon;
4728
4729                 if ((downdelay % miimon) != 0) {
4730                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4731                                    downdelay, miimon,
4732                                    (downdelay / miimon) * miimon);
4733                 }
4734
4735                 downdelay /= miimon;
4736         }
4737
4738         if (arp_interval < 0) {
4739                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4740                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4741                 arp_interval = BOND_LINK_ARP_INTERV;
4742         }
4743
4744         for (arp_ip_count = 0;
4745              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4746              arp_ip_count++) {
4747                 /* not complete check, but should be good enough to
4748                    catch mistakes */
4749                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4750                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4751                                    arp_ip_target[arp_ip_count]);
4752                         arp_interval = 0;
4753                 } else {
4754                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4755                         arp_target[arp_ip_count] = ip;
4756                 }
4757         }
4758
4759         if (arp_interval && !arp_ip_count) {
4760                 /* don't allow arping if no arp_ip_target given... */
4761                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4762                            arp_interval);
4763                 arp_interval = 0;
4764         }
4765
4766         if (arp_validate) {
4767                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4768                         pr_err("arp_validate only supported in active-backup mode\n");
4769                         return -EINVAL;
4770                 }
4771                 if (!arp_interval) {
4772                         pr_err("arp_validate requires arp_interval\n");
4773                         return -EINVAL;
4774                 }
4775
4776                 arp_validate_value = bond_parse_parm(arp_validate,
4777                                                      arp_validate_tbl);
4778                 if (arp_validate_value == -1) {
4779                         pr_err("Error: invalid arp_validate \"%s\"\n",
4780                                arp_validate == NULL ? "NULL" : arp_validate);
4781                         return -EINVAL;
4782                 }
4783         } else
4784                 arp_validate_value = 0;
4785
4786         if (miimon) {
4787                 pr_info("MII link monitoring set to %d ms\n", miimon);
4788         } else if (arp_interval) {
4789                 int i;
4790
4791                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4792                         arp_interval,
4793                         arp_validate_tbl[arp_validate_value].modename,
4794                         arp_ip_count);
4795
4796                 for (i = 0; i < arp_ip_count; i++)
4797                         pr_info(" %s", arp_ip_target[i]);
4798
4799                 pr_info("\n");
4800
4801         } else if (max_bonds) {
4802                 /* miimon and arp_interval not set, we need one so things
4803                  * work as expected, see bonding.txt for details
4804                  */
4805                 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4806         }
4807
4808         if (primary && !USES_PRIMARY(bond_mode)) {
4809                 /* currently, using a primary only makes sense
4810                  * in active backup, TLB or ALB modes
4811                  */
4812                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4813                            primary, bond_mode_name(bond_mode));
4814                 primary = NULL;
4815         }
4816
4817         if (primary && primary_reselect) {
4818                 primary_reselect_value = bond_parse_parm(primary_reselect,
4819                                                          pri_reselect_tbl);
4820                 if (primary_reselect_value == -1) {
4821                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4822                                primary_reselect ==
4823                                         NULL ? "NULL" : primary_reselect);
4824                         return -EINVAL;
4825                 }
4826         } else {
4827                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4828         }
4829
4830         if (fail_over_mac) {
4831                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4832                                                       fail_over_mac_tbl);
4833                 if (fail_over_mac_value == -1) {
4834                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4835                                arp_validate == NULL ? "NULL" : arp_validate);
4836                         return -EINVAL;
4837                 }
4838
4839                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4840                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4841         } else {
4842                 fail_over_mac_value = BOND_FOM_NONE;
4843         }
4844
4845         /* fill params struct with the proper values */
4846         params->mode = bond_mode;
4847         params->xmit_policy = xmit_hashtype;
4848         params->miimon = miimon;
4849         params->num_grat_arp = num_grat_arp;
4850         params->num_unsol_na = num_unsol_na;
4851         params->arp_interval = arp_interval;
4852         params->arp_validate = arp_validate_value;
4853         params->updelay = updelay;
4854         params->downdelay = downdelay;
4855         params->use_carrier = use_carrier;
4856         params->lacp_fast = lacp_fast;
4857         params->primary[0] = 0;
4858         params->primary_reselect = primary_reselect_value;
4859         params->fail_over_mac = fail_over_mac_value;
4860
4861         if (primary) {
4862                 strncpy(params->primary, primary, IFNAMSIZ);
4863                 params->primary[IFNAMSIZ - 1] = 0;
4864         }
4865
4866         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4867
4868         return 0;
4869 }
4870
4871 static struct lock_class_key bonding_netdev_xmit_lock_key;
4872 static struct lock_class_key bonding_netdev_addr_lock_key;
4873
4874 static void bond_set_lockdep_class_one(struct net_device *dev,
4875                                        struct netdev_queue *txq,
4876                                        void *_unused)
4877 {
4878         lockdep_set_class(&txq->_xmit_lock,
4879                           &bonding_netdev_xmit_lock_key);
4880 }
4881
4882 static void bond_set_lockdep_class(struct net_device *dev)
4883 {
4884         lockdep_set_class(&dev->addr_list_lock,
4885                           &bonding_netdev_addr_lock_key);
4886         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4887 }
4888
4889 /*
4890  * Called from registration process
4891  */
4892 static int bond_init(struct net_device *bond_dev)
4893 {
4894         struct bonding *bond = netdev_priv(bond_dev);
4895         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4896
4897         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4898
4899         bond->wq = create_singlethread_workqueue(bond_dev->name);
4900         if (!bond->wq)
4901                 return -ENOMEM;
4902
4903         bond_set_lockdep_class(bond_dev);
4904
4905         netif_carrier_off(bond_dev);
4906
4907         bond_create_proc_entry(bond);
4908         list_add_tail(&bond->bond_list, &bn->dev_list);
4909
4910         bond_prepare_sysfs_group(bond);
4911         return 0;
4912 }
4913
4914 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4915 {
4916         if (tb[IFLA_ADDRESS]) {
4917                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4918                         return -EINVAL;
4919                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4920                         return -EADDRNOTAVAIL;
4921         }
4922         return 0;
4923 }
4924
4925 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4926         .kind           = "bond",
4927         .priv_size      = sizeof(struct bonding),
4928         .setup          = bond_setup,
4929         .validate       = bond_validate,
4930 };
4931
4932 /* Create a new bond based on the specified name and bonding parameters.
4933  * If name is NULL, obtain a suitable "bond%d" name for us.
4934  * Caller must NOT hold rtnl_lock; we need to release it here before we
4935  * set up our sysfs entries.
4936  */
4937 int bond_create(struct net *net, const char *name)
4938 {
4939         struct net_device *bond_dev;
4940         int res;
4941
4942         rtnl_lock();
4943
4944         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4945                                 bond_setup);
4946         if (!bond_dev) {
4947                 pr_err("%s: eek! can't alloc netdev!\n", name);
4948                 res = -ENOMEM;
4949                 goto out;
4950         }
4951
4952         dev_net_set(bond_dev, net);
4953         bond_dev->rtnl_link_ops = &bond_link_ops;
4954
4955         if (!name) {
4956                 res = dev_alloc_name(bond_dev, "bond%d");
4957                 if (res < 0)
4958                         goto out_netdev;
4959         }
4960
4961         res = register_netdevice(bond_dev);
4962         if (res < 0)
4963                 goto out_netdev;
4964
4965 out:
4966         rtnl_unlock();
4967         return res;
4968 out_netdev:
4969         free_netdev(bond_dev);
4970         goto out;
4971 }
4972
4973 static int __net_init bond_net_init(struct net *net)
4974 {
4975         struct bond_net *bn = net_generic(net, bond_net_id);
4976
4977         bn->net = net;
4978         INIT_LIST_HEAD(&bn->dev_list);
4979
4980         bond_create_proc_dir(bn);
4981
4982         return 0;
4983 }
4984
4985 static void __net_exit bond_net_exit(struct net *net)
4986 {
4987         struct bond_net *bn = net_generic(net, bond_net_id);
4988
4989         bond_destroy_proc_dir(bn);
4990 }
4991
4992 static struct pernet_operations bond_net_ops = {
4993         .init = bond_net_init,
4994         .exit = bond_net_exit,
4995         .id   = &bond_net_id,
4996         .size = sizeof(struct bond_net),
4997 };
4998
4999 static int __init bonding_init(void)
5000 {
5001         int i;
5002         int res;
5003
5004         pr_info("%s", version);
5005
5006         res = bond_check_params(&bonding_defaults);
5007         if (res)
5008                 goto out;
5009
5010         res = register_pernet_subsys(&bond_net_ops);
5011         if (res)
5012                 goto out;
5013
5014         res = rtnl_link_register(&bond_link_ops);
5015         if (res)
5016                 goto err_link;
5017
5018         for (i = 0; i < max_bonds; i++) {
5019                 res = bond_create(&init_net, NULL);
5020                 if (res)
5021                         goto err;
5022         }
5023
5024         res = bond_create_sysfs();
5025         if (res)
5026                 goto err;
5027
5028         register_netdevice_notifier(&bond_netdev_notifier);
5029         register_inetaddr_notifier(&bond_inetaddr_notifier);
5030         bond_register_ipv6_notifier();
5031 out:
5032         return res;
5033 err:
5034         rtnl_link_unregister(&bond_link_ops);
5035 err_link:
5036         unregister_pernet_subsys(&bond_net_ops);
5037         goto out;
5038
5039 }
5040
5041 static void __exit bonding_exit(void)
5042 {
5043         unregister_netdevice_notifier(&bond_netdev_notifier);
5044         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5045         bond_unregister_ipv6_notifier();
5046
5047         bond_destroy_sysfs();
5048
5049         rtnl_link_unregister(&bond_link_ops);
5050         unregister_pernet_subsys(&bond_net_ops);
5051 }
5052
5053 module_init(bonding_init);
5054 module_exit(bonding_exit);
5055 MODULE_LICENSE("GPL");
5056 MODULE_VERSION(DRV_VERSION);
5057 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5058 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5059 MODULE_ALIAS_RTNL_LINK("bond");