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/netpoll.h>
  63 #include <linux/inetdevice.h>
  64 #include <linux/igmp.h>
  65 #include <linux/etherdevice.h>
  66 #include <linux/skbuff.h>
  67 #include <net/sock.h>
  68 #include <linux/rtnetlink.h>
  69 #include <linux/proc_fs.h>
  70 #include <linux/seq_file.h>
  71 #include <linux/smp.h>
  72 #include <linux/if_ether.h>
  73 #include <net/arp.h>
  74 #include <linux/mii.h>
  75 #include <linux/ethtool.h>
  76 #include <linux/if_vlan.h>
  77 #include <linux/if_bonding.h>
  78 #include <linux/jiffies.h>
  79 #include <linux/preempt.h>
  80 #include <net/route.h>
  81 #include <net/net_namespace.h>
  82 #include <net/netns/generic.h>
  83 #include "bonding.h"
  84 #include "bond_3ad.h"
  85 #include "bond_alb.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90 #define BOND_LINK_MON_INTERV    0
  91 #define BOND_LINK_ARP_INTERV    0
  92
  93 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  94 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  95 static int num_grat_arp = 1;
  96 static int num_unsol_na = 1;
  97 static int miimon       = BOND_LINK_MON_INTERV;
  98 static int updelay;
  99 static int downdelay;
 100 static int use_carrier  = 1;
 101 static char *mode;
 102 static char *primary;
 103 static char *primary_reselect;
 104 static char *lacp_rate;
 105 static char *ad_select;
 106 static char *xmit_hash_policy;
 107 static int arp_interval = BOND_LINK_ARP_INTERV;
 108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 109 static char *arp_validate;
 110 static char *fail_over_mac;
 111 static int all_slaves_active = 0;
 112 static struct bond_params bonding_defaults;
 113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 114
 115 module_param(max_bonds, int, 0);
 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 117 module_param(tx_queues, int, 0);
 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 119 module_param(num_grat_arp, int, 0644);
 120 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 121 module_param(num_unsol_na, int, 0644);
 122 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 123 module_param(miimon, int, 0);
 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 125 module_param(updelay, int, 0);
 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 127 module_param(downdelay, int, 0);
 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 129                             "in milliseconds");
 130 module_param(use_carrier, int, 0);
 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 132                               "0 for off, 1 for on (default)");
 133 module_param(mode, charp, 0);
 134 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 135                        "1 for active-backup, 2 for balance-xor, "
 136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 137                        "6 for balance-alb");
 138 module_param(primary, charp, 0);
 139 MODULE_PARM_DESC(primary, "Primary network device to use");
 140 module_param(primary_reselect, charp, 0);
 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 142                                    "once it comes up; "
 143                                    "0 for always (default), "
 144                                    "1 for only if speed of primary is "
 145                                    "better, "
 146                                    "2 for only on active slave "
 147                                    "failure");
 148 module_param(lacp_rate, charp, 0);
 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 150                             "(slow/fast)");
 151 module_param(ad_select, charp, 0);
 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 153 module_param(xmit_hash_policy, charp, 0);
 154 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 155                                    ", 1 for layer 3+4");
 156 module_param(arp_interval, int, 0);
 157 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 158 module_param_array(arp_ip_target, charp, NULL, 0);
 159 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 160 module_param(arp_validate, charp, 0);
 161 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 162 module_param(fail_over_mac, charp, 0);
 163 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 164 module_param(all_slaves_active, int, 0);
 165 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 166                                      "by setting active flag for all slaves.  "
 167                                      "0 for never (default), 1 for always.");
 168 module_param(resend_igmp, int, 0);
 169 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 170
 171 /*----------------------------- Global variables ----------------------------*/
 172
 173 #ifdef CONFIG_NET_POLL_CONTROLLER
 174 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 175 #endif
 176
 177 static const char * const version =
 178         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 179
 180 int bond_net_id __read_mostly;
 181
 182 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 183 static int arp_ip_count;
 184 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 185 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 186 static int lacp_fast;
 187
 188 const struct bond_parm_tbl bond_lacp_tbl[] = {
 189 {       "slow",         AD_LACP_SLOW},
 190 {       "fast",         AD_LACP_FAST},
 191 {       NULL,           -1},
 192 };
 193
 194 const struct bond_parm_tbl bond_mode_tbl[] = {
 195 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 196 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 197 {       "balance-xor",          BOND_MODE_XOR},
 198 {       "broadcast",            BOND_MODE_BROADCAST},
 199 {       "802.3ad",              BOND_MODE_8023AD},
 200 {       "balance-tlb",          BOND_MODE_TLB},
 201 {       "balance-alb",          BOND_MODE_ALB},
 202 {       NULL,                   -1},
 203 };
 204
 205 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 206 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 207 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 208 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl arp_validate_tbl[] = {
 213 {       "none",                 BOND_ARP_VALIDATE_NONE},
 214 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 215 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 216 {       "all",                  BOND_ARP_VALIDATE_ALL},
 217 {       NULL,                   -1},
 218 };
 219
 220 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 221 {       "none",                 BOND_FOM_NONE},
 222 {       "active",               BOND_FOM_ACTIVE},
 223 {       "follow",               BOND_FOM_FOLLOW},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl pri_reselect_tbl[] = {
 228 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 229 {       "better",               BOND_PRI_RESELECT_BETTER},
 230 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 231 {       NULL,                   -1},
 232 };
 233
 234 struct bond_parm_tbl ad_select_tbl[] = {
 235 {       "stable",       BOND_AD_STABLE},
 236 {       "bandwidth",    BOND_AD_BANDWIDTH},
 237 {       "count",        BOND_AD_COUNT},
 238 {       NULL,           -1},
 239 };
 240
 241 /*-------------------------- Forward declarations ---------------------------*/
 242
 243 static void bond_send_gratuitous_arp(struct bonding *bond);
 244 static int bond_init(struct net_device *bond_dev);
 245 static void bond_uninit(struct net_device *bond_dev);
 246
 247 /*---------------------------- General routines -----------------------------*/
 248
 249 static const char *bond_mode_name(int mode)
 250 {
 251         static const char *names[] = {
 252                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 253                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 254                 [BOND_MODE_XOR] = "load balancing (xor)",
 255                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 256                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 257                 [BOND_MODE_TLB] = "transmit load balancing",
 258                 [BOND_MODE_ALB] = "adaptive load balancing",
 259         };
 260
 261         if (mode < 0 || mode > BOND_MODE_ALB)
 262                 return "unknown";
 263
 264         return names[mode];
 265 }
 266
 267 /*---------------------------------- VLAN -----------------------------------*/
 268
 269 /**
 270  * bond_add_vlan - add a new vlan id on bond
 271  * @bond: bond that got the notification
 272  * @vlan_id: the vlan id to add
 273  *
 274  * Returns -ENOMEM if allocation failed.
 275  */
 276 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 277 {
 278         struct vlan_entry *vlan;
 279
 280         pr_debug("bond: %s, vlan id %d\n",
 281                  (bond ? bond->dev->name : "None"), vlan_id);
 282
 283         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 284         if (!vlan)
 285                 return -ENOMEM;
 286
 287         INIT_LIST_HEAD(&vlan->vlan_list);
 288         vlan->vlan_id = vlan_id;
 289
 290         write_lock_bh(&bond->lock);
 291
 292         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 293
 294         write_unlock_bh(&bond->lock);
 295
 296         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 297
 298         return 0;
 299 }
 300
 301 /**
 302  * bond_del_vlan - delete a vlan id from bond
 303  * @bond: bond that got the notification
 304  * @vlan_id: the vlan id to delete
 305  *
 306  * returns -ENODEV if @vlan_id was not found in @bond.
 307  */
 308 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 309 {
 310         struct vlan_entry *vlan;
 311         int res = -ENODEV;
 312
 313         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 314
 315         block_netpoll_tx();
 316         write_lock_bh(&bond->lock);
 317
 318         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 319                 if (vlan->vlan_id == vlan_id) {
 320                         list_del(&vlan->vlan_list);
 321
 322                         if (bond_is_lb(bond))
 323                                 bond_alb_clear_vlan(bond, vlan_id);
 324
 325                         pr_debug("removed VLAN ID %d from bond %s\n",
 326                                  vlan_id, bond->dev->name);
 327
 328                         kfree(vlan);
 329
 330                         if (list_empty(&bond->vlan_list) &&
 331                             (bond->slave_cnt == 0)) {
 332                                 /* Last VLAN removed and no slaves, so
 333                                  * restore block on adding VLANs. This will
 334                                  * be removed once new slaves that are not
 335                                  * VLAN challenged will be added.
 336                                  */
 337                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 338                         }
 339
 340                         res = 0;
 341                         goto out;
 342                 }
 343         }
 344
 345         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 346                  vlan_id, bond->dev->name);
 347
 348 out:
 349         write_unlock_bh(&bond->lock);
 350         unblock_netpoll_tx();
 351         return res;
 352 }
 353
 354 /**
 355  * bond_has_challenged_slaves
 356  * @bond: the bond we're working on
 357  *
 358  * Searches the slave list. Returns 1 if a vlan challenged slave
 359  * was found, 0 otherwise.
 360  *
 361  * Assumes bond->lock is held.
 362  */
 363 static int bond_has_challenged_slaves(struct bonding *bond)
 364 {
 365         struct slave *slave;
 366         int i;
 367
 368         bond_for_each_slave(bond, slave, i) {
 369                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 370                         pr_debug("found VLAN challenged slave - %s\n",
 371                                  slave->dev->name);
 372                         return 1;
 373                 }
 374         }
 375
 376         pr_debug("no VLAN challenged slaves found\n");
 377         return 0;
 378 }
 379
 380 /**
 381  * bond_next_vlan - safely skip to the next item in the vlans list.
 382  * @bond: the bond we're working on
 383  * @curr: item we're advancing from
 384  *
 385  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 386  * or @curr->next otherwise (even if it is @curr itself again).
 387  *
 388  * Caller must hold bond->lock
 389  */
 390 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 391 {
 392         struct vlan_entry *next, *last;
 393
 394         if (list_empty(&bond->vlan_list))
 395                 return NULL;
 396
 397         if (!curr) {
 398                 next = list_entry(bond->vlan_list.next,
 399                                   struct vlan_entry, vlan_list);
 400         } else {
 401                 last = list_entry(bond->vlan_list.prev,
 402                                   struct vlan_entry, vlan_list);
 403                 if (last == curr) {
 404                         next = list_entry(bond->vlan_list.next,
 405                                           struct vlan_entry, vlan_list);
 406                 } else {
 407                         next = list_entry(curr->vlan_list.next,
 408                                           struct vlan_entry, vlan_list);
 409                 }
 410         }
 411
 412         return next;
 413 }
 414
 415 /**
 416  * bond_dev_queue_xmit - Prepare skb for xmit.
 417  *
 418  * @bond: bond device that got this skb for tx.
 419  * @skb: hw accel VLAN tagged skb to transmit
 420  * @slave_dev: slave that is supposed to xmit this skbuff
 421  */
 422 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 423                         struct net_device *slave_dev)
 424 {
 425         skb->dev = slave_dev;
 426         skb->priority = 1;
 427 #ifdef CONFIG_NET_POLL_CONTROLLER
 428         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
 429                 struct netpoll *np = bond->dev->npinfo->netpoll;
 430                 slave_dev->npinfo = bond->dev->npinfo;
 431                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
 432                 netpoll_send_skb_on_dev(np, skb, slave_dev);
 433                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
 434         } else
 435 #endif
 436                 dev_queue_xmit(skb);
 437
 438         return 0;
 439 }
 440
 441 /*
 442  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 443  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 444  * lock because:
 445  * a. This operation is performed in IOCTL context,
 446  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 447  * c. Holding a lock with BH disabled while directly calling a base driver
 448  *    entry point is generally a BAD idea.
 449  *
 450  * The design of synchronization/protection for this operation in the 8021q
 451  * module is good for one or more VLAN devices over a single physical device
 452  * and cannot be extended for a teaming solution like bonding, so there is a
 453  * potential race condition here where a net device from the vlan group might
 454  * be referenced (either by a base driver or the 8021q code) while it is being
 455  * removed from the system. However, it turns out we're not making matters
 456  * worse, and if it works for regular VLAN usage it will work here too.
 457 */
 458
 459 /**
 460  * bond_vlan_rx_register - Propagates registration to slaves
 461  * @bond_dev: bonding net device that got called
 462  * @grp: vlan group being registered
 463  */
 464 static void bond_vlan_rx_register(struct net_device *bond_dev,
 465                                   struct vlan_group *grp)
 466 {
 467         struct bonding *bond = netdev_priv(bond_dev);
 468         struct slave *slave;
 469         int i;
 470
 471         write_lock_bh(&bond->lock);
 472         bond->vlgrp = grp;
 473         write_unlock_bh(&bond->lock);
 474
 475         bond_for_each_slave(bond, slave, i) {
 476                 struct net_device *slave_dev = slave->dev;
 477                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 478
 479                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 480                     slave_ops->ndo_vlan_rx_register) {
 481                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 482                 }
 483         }
 484 }
 485
 486 /**
 487  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 488  * @bond_dev: bonding net device that got called
 489  * @vid: vlan id being added
 490  */
 491 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 492 {
 493         struct bonding *bond = netdev_priv(bond_dev);
 494         struct slave *slave;
 495         int i, res;
 496
 497         bond_for_each_slave(bond, slave, i) {
 498                 struct net_device *slave_dev = slave->dev;
 499                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 500
 501                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 502                     slave_ops->ndo_vlan_rx_add_vid) {
 503                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 504                 }
 505         }
 506
 507         res = bond_add_vlan(bond, vid);
 508         if (res) {
 509                 pr_err("%s: Error: Failed to add vlan id %d\n",
 510                        bond_dev->name, vid);
 511         }
 512 }
 513
 514 /**
 515  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 516  * @bond_dev: bonding net device that got called
 517  * @vid: vlan id being removed
 518  */
 519 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 520 {
 521         struct bonding *bond = netdev_priv(bond_dev);
 522         struct slave *slave;
 523         struct net_device *vlan_dev;
 524         int i, res;
 525
 526         bond_for_each_slave(bond, slave, i) {
 527                 struct net_device *slave_dev = slave->dev;
 528                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 529
 530                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 531                     slave_ops->ndo_vlan_rx_kill_vid) {
 532                         /* Save and then restore vlan_dev in the grp array,
 533                          * since the slave's driver might clear it.
 534                          */
 535                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 536                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 537                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 538                 }
 539         }
 540
 541         res = bond_del_vlan(bond, vid);
 542         if (res) {
 543                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 544                        bond_dev->name, vid);
 545         }
 546 }
 547
 548 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 549 {
 550         struct vlan_entry *vlan;
 551         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 552
 553         if (!bond->vlgrp)
 554                 return;
 555
 556         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 557             slave_ops->ndo_vlan_rx_register)
 558                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 559
 560         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 561             !(slave_ops->ndo_vlan_rx_add_vid))
 562                 return;
 563
 564         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 565                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 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         if (!bond->vlgrp)
 576                 return;
 577
 578         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 579             !(slave_ops->ndo_vlan_rx_kill_vid))
 580                 goto unreg;
 581
 582         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 583                 if (!vlan->vlan_id)
 584                         continue;
 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
 599 /*------------------------------- Link status -------------------------------*/
 600
 601 /*
 602  * Set the carrier state for the master according to the state of its
 603  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 604  * do special 802.3ad magic.
 605  *
 606  * Returns zero if carrier state does not change, nonzero if it does.
 607  */
 608 static int bond_set_carrier(struct bonding *bond)
 609 {
 610         struct slave *slave;
 611         int i;
 612
 613         if (bond->slave_cnt == 0)
 614                 goto down;
 615
 616         if (bond->params.mode == BOND_MODE_8023AD)
 617                 return bond_3ad_set_carrier(bond);
 618
 619         bond_for_each_slave(bond, slave, i) {
 620                 if (slave->link == BOND_LINK_UP) {
 621                         if (!netif_carrier_ok(bond->dev)) {
 622                                 netif_carrier_on(bond->dev);
 623                                 return 1;
 624                         }
 625                         return 0;
 626                 }
 627         }
 628
 629 down:
 630         if (netif_carrier_ok(bond->dev)) {
 631                 netif_carrier_off(bond->dev);
 632                 return 1;
 633         }
 634         return 0;
 635 }
 636
 637 /*
 638  * Get link speed and duplex from the slave's base driver
 639  * using ethtool. If for some reason the call fails or the
 640  * values are invalid, fake speed and duplex to 100/Full
 641  * and return error.
 642  */
 643 static int bond_update_speed_duplex(struct slave *slave)
 644 {
 645         struct net_device *slave_dev = slave->dev;
 646         struct ethtool_cmd etool;
 647         int res;
 648
 649         /* Fake speed and duplex */
 650         slave->speed = SPEED_100;
 651         slave->duplex = DUPLEX_FULL;
 652
 653         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 654                 return -1;
 655
 656         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 657         if (res < 0)
 658                 return -1;
 659
 660         switch (etool.speed) {
 661         case SPEED_10:
 662         case SPEED_100:
 663         case SPEED_1000:
 664         case SPEED_10000:
 665                 break;
 666         default:
 667                 return -1;
 668         }
 669
 670         switch (etool.duplex) {
 671         case DUPLEX_FULL:
 672         case DUPLEX_HALF:
 673                 break;
 674         default:
 675                 return -1;
 676         }
 677
 678         slave->speed = etool.speed;
 679         slave->duplex = etool.duplex;
 680
 681         return 0;
 682 }
 683
 684 /*
 685  * if <dev> supports MII link status reporting, check its link status.
 686  *
 687  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 688  * depending upon the setting of the use_carrier parameter.
 689  *
 690  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 691  * can't tell and just pretend it is), or 0, meaning that the link is
 692  * down.
 693  *
 694  * If reporting is non-zero, instead of faking link up, return -1 if
 695  * both ETHTOOL and MII ioctls fail (meaning the device does not
 696  * support them).  If use_carrier is set, return whatever it says.
 697  * It'd be nice if there was a good way to tell if a driver supports
 698  * netif_carrier, but there really isn't.
 699  */
 700 static int bond_check_dev_link(struct bonding *bond,
 701                                struct net_device *slave_dev, int reporting)
 702 {
 703         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 704         int (*ioctl)(struct net_device *, struct ifreq *, int);
 705         struct ifreq ifr;
 706         struct mii_ioctl_data *mii;
 707
 708         if (!reporting && !netif_running(slave_dev))
 709                 return 0;
 710
 711         if (bond->params.use_carrier)
 712                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 713
 714         /* Try to get link status using Ethtool first. */
 715         if (slave_dev->ethtool_ops) {
 716                 if (slave_dev->ethtool_ops->get_link) {
 717                         u32 link;
 718
 719                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 720
 721                         return link ? BMSR_LSTATUS : 0;
 722                 }
 723         }
 724
 725         /* Ethtool can't be used, fallback to MII ioctls. */
 726         ioctl = slave_ops->ndo_do_ioctl;
 727         if (ioctl) {
 728                 /* TODO: set pointer to correct ioctl on a per team member */
 729                 /*       bases to make this more efficient. that is, once  */
 730                 /*       we determine the correct ioctl, we will always    */
 731                 /*       call it and not the others for that team          */
 732                 /*       member.                                           */
 733
 734                 /*
 735                  * We cannot assume that SIOCGMIIPHY will also read a
 736                  * register; not all network drivers (e.g., e100)
 737                  * support that.
 738                  */
 739
 740                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 741                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 742                 mii = if_mii(&ifr);
 743                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 744                         mii->reg_num = MII_BMSR;
 745                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 746                                 return mii->val_out & BMSR_LSTATUS;
 747                 }
 748         }
 749
 750         /*
 751          * If reporting, report that either there's no dev->do_ioctl,
 752          * or both SIOCGMIIREG and get_link failed (meaning that we
 753          * cannot report link status).  If not reporting, pretend
 754          * we're ok.
 755          */
 756         return reporting ? -1 : BMSR_LSTATUS;
 757 }
 758
 759 /*----------------------------- Multicast list ------------------------------*/
 760
 761 /*
 762  * Push the promiscuity flag down to appropriate slaves
 763  */
 764 static int bond_set_promiscuity(struct bonding *bond, int inc)
 765 {
 766         int err = 0;
 767         if (USES_PRIMARY(bond->params.mode)) {
 768                 /* write lock already acquired */
 769                 if (bond->curr_active_slave) {
 770                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 771                                                   inc);
 772                 }
 773         } else {
 774                 struct slave *slave;
 775                 int i;
 776                 bond_for_each_slave(bond, slave, i) {
 777                         err = dev_set_promiscuity(slave->dev, inc);
 778                         if (err)
 779                                 return err;
 780                 }
 781         }
 782         return err;
 783 }
 784
 785 /*
 786  * Push the allmulti flag down to all slaves
 787  */
 788 static int bond_set_allmulti(struct bonding *bond, int inc)
 789 {
 790         int err = 0;
 791         if (USES_PRIMARY(bond->params.mode)) {
 792                 /* write lock already acquired */
 793                 if (bond->curr_active_slave) {
 794                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 795                                                inc);
 796                 }
 797         } else {
 798                 struct slave *slave;
 799                 int i;
 800                 bond_for_each_slave(bond, slave, i) {
 801                         err = dev_set_allmulti(slave->dev, inc);
 802                         if (err)
 803                                 return err;
 804                 }
 805         }
 806         return err;
 807 }
 808
 809 /*
 810  * Add a Multicast address to slaves
 811  * according to mode
 812  */
 813 static void bond_mc_add(struct bonding *bond, void *addr)
 814 {
 815         if (USES_PRIMARY(bond->params.mode)) {
 816                 /* write lock already acquired */
 817                 if (bond->curr_active_slave)
 818                         dev_mc_add(bond->curr_active_slave->dev, addr);
 819         } else {
 820                 struct slave *slave;
 821                 int i;
 822
 823                 bond_for_each_slave(bond, slave, i)
 824                         dev_mc_add(slave->dev, addr);
 825         }
 826 }
 827
 828 /*
 829  * Remove a multicast address from slave
 830  * according to mode
 831  */
 832 static void bond_mc_del(struct bonding *bond, void *addr)
 833 {
 834         if (USES_PRIMARY(bond->params.mode)) {
 835                 /* write lock already acquired */
 836                 if (bond->curr_active_slave)
 837                         dev_mc_del(bond->curr_active_slave->dev, addr);
 838         } else {
 839                 struct slave *slave;
 840                 int i;
 841                 bond_for_each_slave(bond, slave, i) {
 842                         dev_mc_del(slave->dev, addr);
 843                 }
 844         }
 845 }
 846
 847
 848 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 849 {
 850         struct in_device *in_dev;
 851
 852         rcu_read_lock();
 853         in_dev = __in_dev_get_rcu(dev);
 854         if (in_dev)
 855                 ip_mc_rejoin_groups(in_dev);
 856         rcu_read_unlock();
 857 }
 858
 859 /*
 860  * Retrieve the list of registered multicast addresses for the bonding
 861  * device and retransmit an IGMP JOIN request to the current active
 862  * slave.
 863  */
 864 static void bond_resend_igmp_join_requests(struct bonding *bond)
 865 {
 866         struct net_device *vlan_dev;
 867         struct vlan_entry *vlan;
 868
 869         read_lock(&bond->lock);
 870
 871         /* rejoin all groups on bond device */
 872         __bond_resend_igmp_join_requests(bond->dev);
 873
 874         /* rejoin all groups on vlan devices */
 875         if (bond->vlgrp) {
 876                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 877                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 878                                                          vlan->vlan_id);
 879                         if (vlan_dev)
 880                                 __bond_resend_igmp_join_requests(vlan_dev);
 881                 }
 882         }
 883
 884         if (--bond->igmp_retrans > 0)
 885                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 886
 887         read_unlock(&bond->lock);
 888 }
 889
 890 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 891 {
 892         struct bonding *bond = container_of(work, struct bonding,
 893                                                         mcast_work.work);
 894         bond_resend_igmp_join_requests(bond);
 895 }
 896
 897 /*
 898  * flush all members of flush->mc_list from device dev->mc_list
 899  */
 900 static void bond_mc_list_flush(struct net_device *bond_dev,
 901                                struct net_device *slave_dev)
 902 {
 903         struct bonding *bond = netdev_priv(bond_dev);
 904         struct netdev_hw_addr *ha;
 905
 906         netdev_for_each_mc_addr(ha, bond_dev)
 907                 dev_mc_del(slave_dev, ha->addr);
 908
 909         if (bond->params.mode == BOND_MODE_8023AD) {
 910                 /* del lacpdu mc addr from mc list */
 911                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 912
 913                 dev_mc_del(slave_dev, lacpdu_multicast);
 914         }
 915 }
 916
 917 /*--------------------------- Active slave change ---------------------------*/
 918
 919 /*
 920  * Update the mc list and multicast-related flags for the new and
 921  * old active slaves (if any) according to the multicast mode, and
 922  * promiscuous flags unconditionally.
 923  */
 924 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 925                          struct slave *old_active)
 926 {
 927         struct netdev_hw_addr *ha;
 928
 929         if (!USES_PRIMARY(bond->params.mode))
 930                 /* nothing to do -  mc list is already up-to-date on
 931                  * all slaves
 932                  */
 933                 return;
 934
 935         if (old_active) {
 936                 if (bond->dev->flags & IFF_PROMISC)
 937                         dev_set_promiscuity(old_active->dev, -1);
 938
 939                 if (bond->dev->flags & IFF_ALLMULTI)
 940                         dev_set_allmulti(old_active->dev, -1);
 941
 942                 netdev_for_each_mc_addr(ha, bond->dev)
 943                         dev_mc_del(old_active->dev, ha->addr);
 944         }
 945
 946         if (new_active) {
 947                 /* FIXME: Signal errors upstream. */
 948                 if (bond->dev->flags & IFF_PROMISC)
 949                         dev_set_promiscuity(new_active->dev, 1);
 950
 951                 if (bond->dev->flags & IFF_ALLMULTI)
 952                         dev_set_allmulti(new_active->dev, 1);
 953
 954                 netdev_for_each_mc_addr(ha, bond->dev)
 955                         dev_mc_add(new_active->dev, ha->addr);
 956         }
 957 }
 958
 959 /*
 960  * bond_do_fail_over_mac
 961  *
 962  * Perform special MAC address swapping for fail_over_mac settings
 963  *
 964  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 965  */
 966 static void bond_do_fail_over_mac(struct bonding *bond,
 967                                   struct slave *new_active,
 968                                   struct slave *old_active)
 969         __releases(&bond->curr_slave_lock)
 970         __releases(&bond->lock)
 971         __acquires(&bond->lock)
 972         __acquires(&bond->curr_slave_lock)
 973 {
 974         u8 tmp_mac[ETH_ALEN];
 975         struct sockaddr saddr;
 976         int rv;
 977
 978         switch (bond->params.fail_over_mac) {
 979         case BOND_FOM_ACTIVE:
 980                 if (new_active)
 981                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 982                                new_active->dev->addr_len);
 983                 break;
 984         case BOND_FOM_FOLLOW:
 985                 /*
 986                  * if new_active && old_active, swap them
 987                  * if just old_active, do nothing (going to no active slave)
 988                  * if just new_active, set new_active to bond's MAC
 989                  */
 990                 if (!new_active)
 991                         return;
 992
 993                 write_unlock_bh(&bond->curr_slave_lock);
 994                 read_unlock(&bond->lock);
 995
 996                 if (old_active) {
 997                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 998                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 999                                ETH_ALEN);
1000                         saddr.sa_family = new_active->dev->type;
1001                 } else {
1002                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1003                         saddr.sa_family = bond->dev->type;
1004                 }
1005
1006                 rv = dev_set_mac_address(new_active->dev, &saddr);
1007                 if (rv) {
1008                         pr_err("%s: Error %d setting MAC of slave %s\n",
1009                                bond->dev->name, -rv, new_active->dev->name);
1010                         goto out;
1011                 }
1012
1013                 if (!old_active)
1014                         goto out;
1015
1016                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1017                 saddr.sa_family = old_active->dev->type;
1018
1019                 rv = dev_set_mac_address(old_active->dev, &saddr);
1020                 if (rv)
1021                         pr_err("%s: Error %d setting MAC of slave %s\n",
1022                                bond->dev->name, -rv, new_active->dev->name);
1023 out:
1024                 read_lock(&bond->lock);
1025                 write_lock_bh(&bond->curr_slave_lock);
1026                 break;
1027         default:
1028                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1029                        bond->dev->name, bond->params.fail_over_mac);
1030                 break;
1031         }
1032
1033 }
1034
1035 static bool bond_should_change_active(struct bonding *bond)
1036 {
1037         struct slave *prim = bond->primary_slave;
1038         struct slave *curr = bond->curr_active_slave;
1039
1040         if (!prim || !curr || curr->link != BOND_LINK_UP)
1041                 return true;
1042         if (bond->force_primary) {
1043                 bond->force_primary = false;
1044                 return true;
1045         }
1046         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1047             (prim->speed < curr->speed ||
1048              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1049                 return false;
1050         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1051                 return false;
1052         return true;
1053 }
1054
1055 /**
1056  * find_best_interface - select the best available slave to be the active one
1057  * @bond: our bonding struct
1058  *
1059  * Warning: Caller must hold curr_slave_lock for writing.
1060  */
1061 static struct slave *bond_find_best_slave(struct bonding *bond)
1062 {
1063         struct slave *new_active, *old_active;
1064         struct slave *bestslave = NULL;
1065         int mintime = bond->params.updelay;
1066         int i;
1067
1068         new_active = bond->curr_active_slave;
1069
1070         if (!new_active) { /* there were no active slaves left */
1071                 if (bond->slave_cnt > 0)   /* found one slave */
1072                         new_active = bond->first_slave;
1073                 else
1074                         return NULL; /* still no slave, return NULL */
1075         }
1076
1077         if ((bond->primary_slave) &&
1078             bond->primary_slave->link == BOND_LINK_UP &&
1079             bond_should_change_active(bond)) {
1080                 new_active = bond->primary_slave;
1081         }
1082
1083         /* remember where to stop iterating over the slaves */
1084         old_active = new_active;
1085
1086         bond_for_each_slave_from(bond, new_active, i, old_active) {
1087                 if (new_active->link == BOND_LINK_UP) {
1088                         return new_active;
1089                 } else if (new_active->link == BOND_LINK_BACK &&
1090                            IS_UP(new_active->dev)) {
1091                         /* link up, but waiting for stabilization */
1092                         if (new_active->delay < mintime) {
1093                                 mintime = new_active->delay;
1094                                 bestslave = new_active;
1095                         }
1096                 }
1097         }
1098
1099         return bestslave;
1100 }
1101
1102 /**
1103  * change_active_interface - change the active slave into the specified one
1104  * @bond: our bonding struct
1105  * @new: the new slave to make the active one
1106  *
1107  * Set the new slave to the bond's settings and unset them on the old
1108  * curr_active_slave.
1109  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1110  *
1111  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1112  * because it is apparently the best available slave we have, even though its
1113  * updelay hasn't timed out yet.
1114  *
1115  * If new_active is not NULL, caller must hold bond->lock for read and
1116  * curr_slave_lock for write_bh.
1117  */
1118 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1119 {
1120         struct slave *old_active = bond->curr_active_slave;
1121
1122         if (old_active == new_active)
1123                 return;
1124
1125         if (new_active) {
1126                 new_active->jiffies = jiffies;
1127
1128                 if (new_active->link == BOND_LINK_BACK) {
1129                         if (USES_PRIMARY(bond->params.mode)) {
1130                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1131                                         bond->dev->name, new_active->dev->name,
1132                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1133                         }
1134
1135                         new_active->delay = 0;
1136                         new_active->link = BOND_LINK_UP;
1137
1138                         if (bond->params.mode == BOND_MODE_8023AD)
1139                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1140
1141                         if (bond_is_lb(bond))
1142                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1143                 } else {
1144                         if (USES_PRIMARY(bond->params.mode)) {
1145                                 pr_info("%s: making interface %s the new active one.\n",
1146                                         bond->dev->name, new_active->dev->name);
1147                         }
1148                 }
1149         }
1150
1151         if (USES_PRIMARY(bond->params.mode))
1152                 bond_mc_swap(bond, new_active, old_active);
1153
1154         if (bond_is_lb(bond)) {
1155                 bond_alb_handle_active_change(bond, new_active);
1156                 if (old_active)
1157                         bond_set_slave_inactive_flags(old_active);
1158                 if (new_active)
1159                         bond_set_slave_active_flags(new_active);
1160         } else {
1161                 bond->curr_active_slave = new_active;
1162         }
1163
1164         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1165                 if (old_active)
1166                         bond_set_slave_inactive_flags(old_active);
1167
1168                 if (new_active) {
1169                         bond_set_slave_active_flags(new_active);
1170
1171                         if (bond->params.fail_over_mac)
1172                                 bond_do_fail_over_mac(bond, new_active,
1173                                                       old_active);
1174
1175                         if (netif_running(bond->dev)) {
1176                                 bond->send_grat_arp = bond->params.num_grat_arp;
1177                                 bond_send_gratuitous_arp(bond);
1178
1179                                 bond->send_unsol_na = bond->params.num_unsol_na;
1180                                 bond_send_unsolicited_na(bond);
1181                         }
1182
1183                         write_unlock_bh(&bond->curr_slave_lock);
1184                         read_unlock(&bond->lock);
1185
1186                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1187
1188                         read_lock(&bond->lock);
1189                         write_lock_bh(&bond->curr_slave_lock);
1190                 }
1191         }
1192
1193         /* resend IGMP joins since active slave has changed or
1194          * all were sent on curr_active_slave */
1195         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1196              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1197             netif_running(bond->dev)) {
1198                 bond->igmp_retrans = bond->params.resend_igmp;
1199                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1200         }
1201 }
1202
1203 /**
1204  * bond_select_active_slave - select a new active slave, if needed
1205  * @bond: our bonding struct
1206  *
1207  * This functions should be called when one of the following occurs:
1208  * - The old curr_active_slave has been released or lost its link.
1209  * - The primary_slave has got its link back.
1210  * - A slave has got its link back and there's no old curr_active_slave.
1211  *
1212  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1213  */
1214 void bond_select_active_slave(struct bonding *bond)
1215 {
1216         struct slave *best_slave;
1217         int rv;
1218
1219         best_slave = bond_find_best_slave(bond);
1220         if (best_slave != bond->curr_active_slave) {
1221                 bond_change_active_slave(bond, best_slave);
1222                 rv = bond_set_carrier(bond);
1223                 if (!rv)
1224                         return;
1225
1226                 if (netif_carrier_ok(bond->dev)) {
1227                         pr_info("%s: first active interface up!\n",
1228                                 bond->dev->name);
1229                 } else {
1230                         pr_info("%s: now running without any active interface !\n",
1231                                 bond->dev->name);
1232                 }
1233         }
1234 }
1235
1236 /*--------------------------- slave list handling ---------------------------*/
1237
1238 /*
1239  * This function attaches the slave to the end of list.
1240  *
1241  * bond->lock held for writing by caller.
1242  */
1243 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1244 {
1245         if (bond->first_slave == NULL) { /* attaching the first slave */
1246                 new_slave->next = new_slave;
1247                 new_slave->prev = new_slave;
1248                 bond->first_slave = new_slave;
1249         } else {
1250                 new_slave->next = bond->first_slave;
1251                 new_slave->prev = bond->first_slave->prev;
1252                 new_slave->next->prev = new_slave;
1253                 new_slave->prev->next = new_slave;
1254         }
1255
1256         bond->slave_cnt++;
1257 }
1258
1259 /*
1260  * This function detaches the slave from the list.
1261  * WARNING: no check is made to verify if the slave effectively
1262  * belongs to <bond>.
1263  * Nothing is freed on return, structures are just unchained.
1264  * If any slave pointer in bond was pointing to <slave>,
1265  * it should be changed by the calling function.
1266  *
1267  * bond->lock held for writing by caller.
1268  */
1269 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1270 {
1271         if (slave->next)
1272                 slave->next->prev = slave->prev;
1273
1274         if (slave->prev)
1275                 slave->prev->next = slave->next;
1276
1277         if (bond->first_slave == slave) { /* slave is the first slave */
1278                 if (bond->slave_cnt > 1) { /* there are more slave */
1279                         bond->first_slave = slave->next;
1280                 } else {
1281                         bond->first_slave = NULL; /* slave was the last one */
1282                 }
1283         }
1284
1285         slave->next = NULL;
1286         slave->prev = NULL;
1287         bond->slave_cnt--;
1288 }
1289
1290 #ifdef CONFIG_NET_POLL_CONTROLLER
1291 /*
1292  * You must hold read lock on bond->lock before calling this.
1293  */
1294 static bool slaves_support_netpoll(struct net_device *bond_dev)
1295 {
1296         struct bonding *bond = netdev_priv(bond_dev);
1297         struct slave *slave;
1298         int i = 0;
1299         bool ret = true;
1300
1301         bond_for_each_slave(bond, slave, i) {
1302                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1303                     !slave->dev->netdev_ops->ndo_poll_controller)
1304                         ret = false;
1305         }
1306         return i != 0 && ret;
1307 }
1308
1309 static void bond_poll_controller(struct net_device *bond_dev)
1310 {
1311         struct bonding *bond = netdev_priv(bond_dev);
1312         struct slave *slave;
1313         int i;
1314
1315         bond_for_each_slave(bond, slave, i) {
1316                 if (slave->dev && IS_UP(slave->dev))
1317                         netpoll_poll_dev(slave->dev);
1318         }
1319 }
1320
1321 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1322 {
1323         struct bonding *bond = netdev_priv(bond_dev);
1324         struct slave *slave;
1325         const struct net_device_ops *ops;
1326         int i;
1327
1328         read_lock(&bond->lock);
1329         bond_dev->npinfo = NULL;
1330         bond_for_each_slave(bond, slave, i) {
1331                 if (slave->dev) {
1332                         ops = slave->dev->netdev_ops;
1333                         if (ops->ndo_netpoll_cleanup)
1334                                 ops->ndo_netpoll_cleanup(slave->dev);
1335                         else
1336                                 slave->dev->npinfo = NULL;
1337                 }
1338         }
1339         read_unlock(&bond->lock);
1340 }
1341
1342 #else
1343
1344 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1345 {
1346 }
1347
1348 #endif
1349
1350 /*---------------------------------- IOCTL ----------------------------------*/
1351
1352 static int bond_sethwaddr(struct net_device *bond_dev,
1353                           struct net_device *slave_dev)
1354 {
1355         pr_debug("bond_dev=%p\n", bond_dev);
1356         pr_debug("slave_dev=%p\n", slave_dev);
1357         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1358         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1359         return 0;
1360 }
1361
1362 #define BOND_VLAN_FEATURES \
1363         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1364          NETIF_F_HW_VLAN_FILTER)
1365
1366 /*
1367  * Compute the common dev->feature set available to all slaves.  Some
1368  * feature bits are managed elsewhere, so preserve those feature bits
1369  * on the master device.
1370  */
1371 static int bond_compute_features(struct bonding *bond)
1372 {
1373         struct slave *slave;
1374         struct net_device *bond_dev = bond->dev;
1375         unsigned long features = bond_dev->features;
1376         unsigned long vlan_features = 0;
1377         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1378                                                 bond_dev->hard_header_len);
1379         int i;
1380
1381         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1382         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1383
1384         if (!bond->first_slave)
1385                 goto done;
1386
1387         features &= ~NETIF_F_ONE_FOR_ALL;
1388
1389         vlan_features = bond->first_slave->dev->vlan_features;
1390         bond_for_each_slave(bond, slave, i) {
1391                 features = netdev_increment_features(features,
1392                                                      slave->dev->features,
1393                                                      NETIF_F_ONE_FOR_ALL);
1394                 vlan_features = netdev_increment_features(vlan_features,
1395                                                         slave->dev->vlan_features,
1396                                                         NETIF_F_ONE_FOR_ALL);
1397                 if (slave->dev->hard_header_len > max_hard_header_len)
1398                         max_hard_header_len = slave->dev->hard_header_len;
1399         }
1400
1401 done:
1402         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1403         bond_dev->features = netdev_fix_features(features, NULL);
1404         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1405         bond_dev->hard_header_len = max_hard_header_len;
1406
1407         return 0;
1408 }
1409
1410 static void bond_setup_by_slave(struct net_device *bond_dev,
1411                                 struct net_device *slave_dev)
1412 {
1413         struct bonding *bond = netdev_priv(bond_dev);
1414
1415         bond_dev->header_ops        = slave_dev->header_ops;
1416
1417         bond_dev->type              = slave_dev->type;
1418         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1419         bond_dev->addr_len          = slave_dev->addr_len;
1420
1421         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422                 slave_dev->addr_len);
1423         bond->setup_by_slave = 1;
1424 }
1425
1426 /* enslave device <slave> to bond device <master> */
1427 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1428 {
1429         struct bonding *bond = netdev_priv(bond_dev);
1430         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1431         struct slave *new_slave = NULL;
1432         struct netdev_hw_addr *ha;
1433         struct sockaddr addr;
1434         int link_reporting;
1435         int old_features = bond_dev->features;
1436         int res = 0;
1437
1438         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1439                 slave_ops->ndo_do_ioctl == NULL) {
1440                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1441                            bond_dev->name, slave_dev->name);
1442         }
1443
1444         /* bond must be initialized by bond_open() before enslaving */
1445         if (!(bond_dev->flags & IFF_UP)) {
1446                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1447                            bond_dev->name);
1448         }
1449
1450         /* already enslaved */
1451         if (slave_dev->flags & IFF_SLAVE) {
1452                 pr_debug("Error, Device was already enslaved\n");
1453                 return -EBUSY;
1454         }
1455
1456         /* vlan challenged mutual exclusion */
1457         /* no need to lock since we're protected by rtnl_lock */
1458         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1459                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1460                 if (bond->vlgrp) {
1461                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1462                                bond_dev->name, slave_dev->name, bond_dev->name);
1463                         return -EPERM;
1464                 } else {
1465                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1466                                    bond_dev->name, slave_dev->name,
1467                                    slave_dev->name, bond_dev->name);
1468                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1469                 }
1470         } else {
1471                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1472                 if (bond->slave_cnt == 0) {
1473                         /* First slave, and it is not VLAN challenged,
1474                          * so remove the block of adding VLANs over the bond.
1475                          */
1476                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1477                 }
1478         }
1479
1480         /*
1481          * Old ifenslave binaries are no longer supported.  These can
1482          * be identified with moderate accuracy by the state of the slave:
1483          * the current ifenslave will set the interface down prior to
1484          * enslaving it; the old ifenslave will not.
1485          */
1486         if ((slave_dev->flags & IFF_UP)) {
1487                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1488                        slave_dev->name);
1489                 res = -EPERM;
1490                 goto err_undo_flags;
1491         }
1492
1493         /* set bonding device ether type by slave - bonding netdevices are
1494          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1495          * there is a need to override some of the type dependent attribs/funcs.
1496          *
1497          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1498          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1499          */
1500         if (bond->slave_cnt == 0) {
1501                 if (bond_dev->type != slave_dev->type) {
1502                         pr_debug("%s: change device type from %d to %d\n",
1503                                  bond_dev->name,
1504                                  bond_dev->type, slave_dev->type);
1505
1506                         res = netdev_bonding_change(bond_dev,
1507                                                     NETDEV_PRE_TYPE_CHANGE);
1508                         res = notifier_to_errno(res);
1509                         if (res) {
1510                                 pr_err("%s: refused to change device type\n",
1511                                        bond_dev->name);
1512                                 res = -EBUSY;
1513                                 goto err_undo_flags;
1514                         }
1515
1516                         /* Flush unicast and multicast addresses */
1517                         dev_uc_flush(bond_dev);
1518                         dev_mc_flush(bond_dev);
1519
1520                         if (slave_dev->type != ARPHRD_ETHER)
1521                                 bond_setup_by_slave(bond_dev, slave_dev);
1522                         else
1523                                 ether_setup(bond_dev);
1524
1525                         netdev_bonding_change(bond_dev,
1526                                               NETDEV_POST_TYPE_CHANGE);
1527                 }
1528         } else if (bond_dev->type != slave_dev->type) {
1529                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1530                        slave_dev->name,
1531                        slave_dev->type, bond_dev->type);
1532                 res = -EINVAL;
1533                 goto err_undo_flags;
1534         }
1535
1536         if (slave_ops->ndo_set_mac_address == NULL) {
1537                 if (bond->slave_cnt == 0) {
1538                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1539                                    bond_dev->name);
1540                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1541                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1542                         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",
1543                                bond_dev->name);
1544                         res = -EOPNOTSUPP;
1545                         goto err_undo_flags;
1546                 }
1547         }
1548
1549         /* If this is the first slave, then we need to set the master's hardware
1550          * address to be the same as the slave's. */
1551         if (is_zero_ether_addr(bond->dev->dev_addr))
1552                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1553                        slave_dev->addr_len);
1554
1555
1556         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1557         if (!new_slave) {
1558                 res = -ENOMEM;
1559                 goto err_undo_flags;
1560         }
1561
1562         /*
1563          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1564          * is set via sysfs or module option if desired.
1565          */
1566         new_slave->queue_id = 0;
1567
1568         /* Save slave's original mtu and then set it to match the bond */
1569         new_slave->original_mtu = slave_dev->mtu;
1570         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1571         if (res) {
1572                 pr_debug("Error %d calling dev_set_mtu\n", res);
1573                 goto err_free;
1574         }
1575
1576         /*
1577          * Save slave's original ("permanent") mac address for modes
1578          * that need it, and for restoring it upon release, and then
1579          * set it to the master's address
1580          */
1581         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1582
1583         if (!bond->params.fail_over_mac) {
1584                 /*
1585                  * Set slave to master's mac address.  The application already
1586                  * set the master's mac address to that of the first slave
1587                  */
1588                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1589                 addr.sa_family = slave_dev->type;
1590                 res = dev_set_mac_address(slave_dev, &addr);
1591                 if (res) {
1592                         pr_debug("Error %d calling set_mac_address\n", res);
1593                         goto err_restore_mtu;
1594                 }
1595         }
1596
1597         res = netdev_set_master(slave_dev, bond_dev);
1598         if (res) {
1599                 pr_debug("Error %d calling netdev_set_master\n", res);
1600                 goto err_restore_mac;
1601         }
1602         /* open the slave since the application closed it */
1603         res = dev_open(slave_dev);
1604         if (res) {
1605                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1606                 goto err_unset_master;
1607         }
1608
1609         new_slave->dev = slave_dev;
1610         slave_dev->priv_flags |= IFF_BONDING;
1611
1612         if (bond_is_lb(bond)) {
1613                 /* bond_alb_init_slave() must be called before all other stages since
1614                  * it might fail and we do not want to have to undo everything
1615                  */
1616                 res = bond_alb_init_slave(bond, new_slave);
1617                 if (res)
1618                         goto err_close;
1619         }
1620
1621         /* If the mode USES_PRIMARY, then the new slave gets the
1622          * master's promisc (and mc) settings only if it becomes the
1623          * curr_active_slave, and that is taken care of later when calling
1624          * bond_change_active()
1625          */
1626         if (!USES_PRIMARY(bond->params.mode)) {
1627                 /* set promiscuity level to new slave */
1628                 if (bond_dev->flags & IFF_PROMISC) {
1629                         res = dev_set_promiscuity(slave_dev, 1);
1630                         if (res)
1631                                 goto err_close;
1632                 }
1633
1634                 /* set allmulti level to new slave */
1635                 if (bond_dev->flags & IFF_ALLMULTI) {
1636                         res = dev_set_allmulti(slave_dev, 1);
1637                         if (res)
1638                                 goto err_close;
1639                 }
1640
1641                 netif_addr_lock_bh(bond_dev);
1642                 /* upload master's mc_list to new slave */
1643                 netdev_for_each_mc_addr(ha, bond_dev)
1644                         dev_mc_add(slave_dev, ha->addr);
1645                 netif_addr_unlock_bh(bond_dev);
1646         }
1647
1648         if (bond->params.mode == BOND_MODE_8023AD) {
1649                 /* add lacpdu mc addr to mc list */
1650                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1651
1652                 dev_mc_add(slave_dev, lacpdu_multicast);
1653         }
1654
1655         bond_add_vlans_on_slave(bond, slave_dev);
1656
1657         write_lock_bh(&bond->lock);
1658
1659         bond_attach_slave(bond, new_slave);
1660
1661         new_slave->delay = 0;
1662         new_slave->link_failure_count = 0;
1663
1664         bond_compute_features(bond);
1665
1666         write_unlock_bh(&bond->lock);
1667
1668         read_lock(&bond->lock);
1669
1670         new_slave->last_arp_rx = jiffies;
1671
1672         if (bond->params.miimon && !bond->params.use_carrier) {
1673                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1674
1675                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1676                         /*
1677                          * miimon is set but a bonded network driver
1678                          * does not support ETHTOOL/MII and
1679                          * arp_interval is not set.  Note: if
1680                          * use_carrier is enabled, we will never go
1681                          * here (because netif_carrier is always
1682                          * supported); thus, we don't need to change
1683                          * the messages for netif_carrier.
1684                          */
1685                         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",
1686                                bond_dev->name, slave_dev->name);
1687                 } else if (link_reporting == -1) {
1688                         /* unable get link status using mii/ethtool */
1689                         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",
1690                                    bond_dev->name, slave_dev->name);
1691                 }
1692         }
1693
1694         /* check for initial state */
1695         if (!bond->params.miimon ||
1696             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1697                 if (bond->params.updelay) {
1698                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1699                         new_slave->link  = BOND_LINK_BACK;
1700                         new_slave->delay = bond->params.updelay;
1701                 } else {
1702                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1703                         new_slave->link  = BOND_LINK_UP;
1704                 }
1705                 new_slave->jiffies = jiffies;
1706         } else {
1707                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1708                 new_slave->link  = BOND_LINK_DOWN;
1709         }
1710
1711         if (bond_update_speed_duplex(new_slave) &&
1712             (new_slave->link != BOND_LINK_DOWN)) {
1713                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1714                            bond_dev->name, new_slave->dev->name);
1715
1716                 if (bond->params.mode == BOND_MODE_8023AD) {
1717                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1718                                    bond_dev->name);
1719                 }
1720         }
1721
1722         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1723                 /* if there is a primary slave, remember it */
1724                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1725                         bond->primary_slave = new_slave;
1726                         bond->force_primary = true;
1727                 }
1728         }
1729
1730         write_lock_bh(&bond->curr_slave_lock);
1731
1732         switch (bond->params.mode) {
1733         case BOND_MODE_ACTIVEBACKUP:
1734                 bond_set_slave_inactive_flags(new_slave);
1735                 bond_select_active_slave(bond);
1736                 break;
1737         case BOND_MODE_8023AD:
1738                 /* in 802.3ad mode, the internal mechanism
1739                  * will activate the slaves in the selected
1740                  * aggregator
1741                  */
1742                 bond_set_slave_inactive_flags(new_slave);
1743                 /* if this is the first slave */
1744                 if (bond->slave_cnt == 1) {
1745                         SLAVE_AD_INFO(new_slave).id = 1;
1746                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1747                          * can be called only after the mac address of the bond is set
1748                          */
1749                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1750                                             bond->params.lacp_fast);
1751                 } else {
1752                         SLAVE_AD_INFO(new_slave).id =
1753                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1754                 }
1755
1756                 bond_3ad_bind_slave(new_slave);
1757                 break;
1758         case BOND_MODE_TLB:
1759         case BOND_MODE_ALB:
1760                 new_slave->state = BOND_STATE_ACTIVE;
1761                 bond_set_slave_inactive_flags(new_slave);
1762                 bond_select_active_slave(bond);
1763                 break;
1764         default:
1765                 pr_debug("This slave is always active in trunk mode\n");
1766
1767                 /* always active in trunk mode */
1768                 new_slave->state = BOND_STATE_ACTIVE;
1769
1770                 /* In trunking mode there is little meaning to curr_active_slave
1771                  * anyway (it holds no special properties of the bond device),
1772                  * so we can change it without calling change_active_interface()
1773                  */
1774                 if (!bond->curr_active_slave)
1775                         bond->curr_active_slave = new_slave;
1776
1777                 break;
1778         } /* switch(bond_mode) */
1779
1780         write_unlock_bh(&bond->curr_slave_lock);
1781
1782         bond_set_carrier(bond);
1783
1784 #ifdef CONFIG_NET_POLL_CONTROLLER
1785         if (slaves_support_netpoll(bond_dev)) {
1786                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1787                 if (bond_dev->npinfo)
1788                         slave_dev->npinfo = bond_dev->npinfo;
1789         } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1790                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1791                 pr_info("New slave device %s does not support netpoll\n",
1792                         slave_dev->name);
1793                 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1794         }
1795 #endif
1796         read_unlock(&bond->lock);
1797
1798         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1799         if (res)
1800                 goto err_close;
1801
1802         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1803                 bond_dev->name, slave_dev->name,
1804                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1805                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1806
1807         /* enslave is successful */
1808         return 0;
1809
1810 /* Undo stages on error */
1811 err_close:
1812         dev_close(slave_dev);
1813
1814 err_unset_master:
1815         netdev_set_master(slave_dev, NULL);
1816
1817 err_restore_mac:
1818         if (!bond->params.fail_over_mac) {
1819                 /* XXX TODO - fom follow mode needs to change master's
1820                  * MAC if this slave's MAC is in use by the bond, or at
1821                  * least print a warning.
1822                  */
1823                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1824                 addr.sa_family = slave_dev->type;
1825                 dev_set_mac_address(slave_dev, &addr);
1826         }
1827
1828 err_restore_mtu:
1829         dev_set_mtu(slave_dev, new_slave->original_mtu);
1830
1831 err_free:
1832         kfree(new_slave);
1833
1834 err_undo_flags:
1835         bond_dev->features = old_features;
1836
1837         return res;
1838 }
1839
1840 /*
1841  * Try to release the slave device <slave> from the bond device <master>
1842  * It is legal to access curr_active_slave without a lock because all the function
1843  * is write-locked.
1844  *
1845  * The rules for slave state should be:
1846  *   for Active/Backup:
1847  *     Active stays on all backups go down
1848  *   for Bonded connections:
1849  *     The first up interface should be left on and all others downed.
1850  */
1851 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1852 {
1853         struct bonding *bond = netdev_priv(bond_dev);
1854         struct slave *slave, *oldcurrent;
1855         struct sockaddr addr;
1856
1857         /* slave is not a slave or master is not master of this slave */
1858         if (!(slave_dev->flags & IFF_SLAVE) ||
1859             (slave_dev->master != bond_dev)) {
1860                 pr_err("%s: Error: cannot release %s.\n",
1861                        bond_dev->name, slave_dev->name);
1862                 return -EINVAL;
1863         }
1864
1865         block_netpoll_tx();
1866         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1867         write_lock_bh(&bond->lock);
1868
1869         slave = bond_get_slave_by_dev(bond, slave_dev);
1870         if (!slave) {
1871                 /* not a slave of this bond */
1872                 pr_info("%s: %s not enslaved\n",
1873                         bond_dev->name, slave_dev->name);
1874                 write_unlock_bh(&bond->lock);
1875                 unblock_netpoll_tx();
1876                 return -EINVAL;
1877         }
1878
1879         if (!bond->params.fail_over_mac) {
1880                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1881                     bond->slave_cnt > 1)
1882                         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",
1883                                    bond_dev->name, slave_dev->name,
1884                                    slave->perm_hwaddr,
1885                                    bond_dev->name, slave_dev->name);
1886         }
1887
1888         /* Inform AD package of unbinding of slave. */
1889         if (bond->params.mode == BOND_MODE_8023AD) {
1890                 /* must be called before the slave is
1891                  * detached from the list
1892                  */
1893                 bond_3ad_unbind_slave(slave);
1894         }
1895
1896         pr_info("%s: releasing %s interface %s\n",
1897                 bond_dev->name,
1898                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1899                 slave_dev->name);
1900
1901         oldcurrent = bond->curr_active_slave;
1902
1903         bond->current_arp_slave = NULL;
1904
1905         /* release the slave from its bond */
1906         bond_detach_slave(bond, slave);
1907
1908         bond_compute_features(bond);
1909
1910         if (bond->primary_slave == slave)
1911                 bond->primary_slave = NULL;
1912
1913         if (oldcurrent == slave)
1914                 bond_change_active_slave(bond, NULL);
1915
1916         if (bond_is_lb(bond)) {
1917                 /* Must be called only after the slave has been
1918                  * detached from the list and the curr_active_slave
1919                  * has been cleared (if our_slave == old_current),
1920                  * but before a new active slave is selected.
1921                  */
1922                 write_unlock_bh(&bond->lock);
1923                 bond_alb_deinit_slave(bond, slave);
1924                 write_lock_bh(&bond->lock);
1925         }
1926
1927         if (oldcurrent == slave) {
1928                 /*
1929                  * Note that we hold RTNL over this sequence, so there
1930                  * is no concern that another slave add/remove event
1931                  * will interfere.
1932                  */
1933                 write_unlock_bh(&bond->lock);
1934                 read_lock(&bond->lock);
1935                 write_lock_bh(&bond->curr_slave_lock);
1936
1937                 bond_select_active_slave(bond);
1938
1939                 write_unlock_bh(&bond->curr_slave_lock);
1940                 read_unlock(&bond->lock);
1941                 write_lock_bh(&bond->lock);
1942         }
1943
1944         if (bond->slave_cnt == 0) {
1945                 bond_set_carrier(bond);
1946
1947                 /* if the last slave was removed, zero the mac address
1948                  * of the master so it will be set by the application
1949                  * to the mac address of the first slave
1950                  */
1951                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1952
1953                 if (!bond->vlgrp) {
1954                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1955                 } else {
1956                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1957                                    bond_dev->name, bond_dev->name);
1958                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1959                                    bond_dev->name);
1960                 }
1961         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1962                    !bond_has_challenged_slaves(bond)) {
1963                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1964                         bond_dev->name, slave_dev->name, bond_dev->name);
1965                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1966         }
1967
1968         write_unlock_bh(&bond->lock);
1969         unblock_netpoll_tx();
1970
1971         /* must do this from outside any spinlocks */
1972         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1973
1974         bond_del_vlans_from_slave(bond, slave_dev);
1975
1976         /* If the mode USES_PRIMARY, then we should only remove its
1977          * promisc and mc settings if it was the curr_active_slave, but that was
1978          * already taken care of above when we detached the slave
1979          */
1980         if (!USES_PRIMARY(bond->params.mode)) {
1981                 /* unset promiscuity level from slave */
1982                 if (bond_dev->flags & IFF_PROMISC)
1983                         dev_set_promiscuity(slave_dev, -1);
1984
1985                 /* unset allmulti level from slave */
1986                 if (bond_dev->flags & IFF_ALLMULTI)
1987                         dev_set_allmulti(slave_dev, -1);
1988
1989                 /* flush master's mc_list from slave */
1990                 netif_addr_lock_bh(bond_dev);
1991                 bond_mc_list_flush(bond_dev, slave_dev);
1992                 netif_addr_unlock_bh(bond_dev);
1993         }
1994
1995         netdev_set_master(slave_dev, NULL);
1996
1997 #ifdef CONFIG_NET_POLL_CONTROLLER
1998         read_lock_bh(&bond->lock);
1999
2000         if (slaves_support_netpoll(bond_dev))
2001                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2002         read_unlock_bh(&bond->lock);
2003         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2004                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2005         else
2006                 slave_dev->npinfo = NULL;
2007 #endif
2008
2009         /* close slave before restoring its mac address */
2010         dev_close(slave_dev);
2011
2012         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2013                 /* restore original ("permanent") mac address */
2014                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2015                 addr.sa_family = slave_dev->type;
2016                 dev_set_mac_address(slave_dev, &addr);
2017         }
2018
2019         dev_set_mtu(slave_dev, slave->original_mtu);
2020
2021         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2022                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2023                                    IFF_SLAVE_NEEDARP);
2024
2025         kfree(slave);
2026
2027         return 0;  /* deletion OK */
2028 }
2029
2030 /*
2031 * First release a slave and than destroy the bond if no more slaves are left.
2032 * Must be under rtnl_lock when this function is called.
2033 */
2034 static int  bond_release_and_destroy(struct net_device *bond_dev,
2035                                      struct net_device *slave_dev)
2036 {
2037         struct bonding *bond = netdev_priv(bond_dev);
2038         int ret;
2039
2040         ret = bond_release(bond_dev, slave_dev);
2041         if ((ret == 0) && (bond->slave_cnt == 0)) {
2042                 pr_info("%s: destroying bond %s.\n",
2043                         bond_dev->name, bond_dev->name);
2044                 unregister_netdevice(bond_dev);
2045         }
2046         return ret;
2047 }
2048
2049 /*
2050  * This function releases all slaves.
2051  */
2052 static int bond_release_all(struct net_device *bond_dev)
2053 {
2054         struct bonding *bond = netdev_priv(bond_dev);
2055         struct slave *slave;
2056         struct net_device *slave_dev;
2057         struct sockaddr addr;
2058
2059         write_lock_bh(&bond->lock);
2060
2061         netif_carrier_off(bond_dev);
2062
2063         if (bond->slave_cnt == 0)
2064                 goto out;
2065
2066         bond->current_arp_slave = NULL;
2067         bond->primary_slave = NULL;
2068         bond_change_active_slave(bond, NULL);
2069
2070         while ((slave = bond->first_slave) != NULL) {
2071                 /* Inform AD package of unbinding of slave
2072                  * before slave is detached from the list.
2073                  */
2074                 if (bond->params.mode == BOND_MODE_8023AD)
2075                         bond_3ad_unbind_slave(slave);
2076
2077                 slave_dev = slave->dev;
2078                 bond_detach_slave(bond, slave);
2079
2080                 /* now that the slave is detached, unlock and perform
2081                  * all the undo steps that should not be called from
2082                  * within a lock.
2083                  */
2084                 write_unlock_bh(&bond->lock);
2085
2086                 if (bond_is_lb(bond)) {
2087                         /* must be called only after the slave
2088                          * has been detached from the list
2089                          */
2090                         bond_alb_deinit_slave(bond, slave);
2091                 }
2092
2093                 bond_compute_features(bond);
2094
2095                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2096                 bond_del_vlans_from_slave(bond, slave_dev);
2097
2098                 /* If the mode USES_PRIMARY, then we should only remove its
2099                  * promisc and mc settings if it was the curr_active_slave, but that was
2100                  * already taken care of above when we detached the slave
2101                  */
2102                 if (!USES_PRIMARY(bond->params.mode)) {
2103                         /* unset promiscuity level from slave */
2104                         if (bond_dev->flags & IFF_PROMISC)
2105                                 dev_set_promiscuity(slave_dev, -1);
2106
2107                         /* unset allmulti level from slave */
2108                         if (bond_dev->flags & IFF_ALLMULTI)
2109                                 dev_set_allmulti(slave_dev, -1);
2110
2111                         /* flush master's mc_list from slave */
2112                         netif_addr_lock_bh(bond_dev);
2113                         bond_mc_list_flush(bond_dev, slave_dev);
2114                         netif_addr_unlock_bh(bond_dev);
2115                 }
2116
2117                 netdev_set_master(slave_dev, NULL);
2118
2119                 /* close slave before restoring its mac address */
2120                 dev_close(slave_dev);
2121
2122                 if (!bond->params.fail_over_mac) {
2123                         /* restore original ("permanent") mac address*/
2124                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2125                         addr.sa_family = slave_dev->type;
2126                         dev_set_mac_address(slave_dev, &addr);
2127                 }
2128
2129                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2130                                            IFF_SLAVE_INACTIVE);
2131
2132                 kfree(slave);
2133
2134                 /* re-acquire the lock before getting the next slave */
2135                 write_lock_bh(&bond->lock);
2136         }
2137
2138         /* zero the mac address of the master so it will be
2139          * set by the application to the mac address of the
2140          * first slave
2141          */
2142         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2143
2144         if (!bond->vlgrp) {
2145                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2146         } else {
2147                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2148                            bond_dev->name, bond_dev->name);
2149                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2150                            bond_dev->name);
2151         }
2152
2153         pr_info("%s: released all slaves\n", bond_dev->name);
2154
2155 out:
2156         write_unlock_bh(&bond->lock);
2157         return 0;
2158 }
2159
2160 /*
2161  * This function changes the active slave to slave <slave_dev>.
2162  * It returns -EINVAL in the following cases.
2163  *  - <slave_dev> is not found in the list.
2164  *  - There is not active slave now.
2165  *  - <slave_dev> is already active.
2166  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2167  *  - <slave_dev> is not running.
2168  * In these cases, this function does nothing.
2169  * In the other cases, current_slave pointer is changed and 0 is returned.
2170  */
2171 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2172 {
2173         struct bonding *bond = netdev_priv(bond_dev);
2174         struct slave *old_active = NULL;
2175         struct slave *new_active = NULL;
2176         int res = 0;
2177
2178         if (!USES_PRIMARY(bond->params.mode))
2179                 return -EINVAL;
2180
2181         /* Verify that master_dev is indeed the master of slave_dev */
2182         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2183                 return -EINVAL;
2184
2185         read_lock(&bond->lock);
2186
2187         read_lock(&bond->curr_slave_lock);
2188         old_active = bond->curr_active_slave;
2189         read_unlock(&bond->curr_slave_lock);
2190
2191         new_active = bond_get_slave_by_dev(bond, slave_dev);
2192
2193         /*
2194          * Changing to the current active: do nothing; return success.
2195          */
2196         if (new_active && (new_active == old_active)) {
2197                 read_unlock(&bond->lock);
2198                 return 0;
2199         }
2200
2201         if ((new_active) &&
2202             (old_active) &&
2203             (new_active->link == BOND_LINK_UP) &&
2204             IS_UP(new_active->dev)) {
2205                 block_netpoll_tx();
2206                 write_lock_bh(&bond->curr_slave_lock);
2207                 bond_change_active_slave(bond, new_active);
2208                 write_unlock_bh(&bond->curr_slave_lock);
2209                 unblock_netpoll_tx();
2210         } else
2211                 res = -EINVAL;
2212
2213         read_unlock(&bond->lock);
2214
2215         return res;
2216 }
2217
2218 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2219 {
2220         struct bonding *bond = netdev_priv(bond_dev);
2221
2222         info->bond_mode = bond->params.mode;
2223         info->miimon = bond->params.miimon;
2224
2225         read_lock(&bond->lock);
2226         info->num_slaves = bond->slave_cnt;
2227         read_unlock(&bond->lock);
2228
2229         return 0;
2230 }
2231
2232 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2233 {
2234         struct bonding *bond = netdev_priv(bond_dev);
2235         struct slave *slave;
2236         int i, res = -ENODEV;
2237
2238         read_lock(&bond->lock);
2239
2240         bond_for_each_slave(bond, slave, i) {
2241                 if (i == (int)info->slave_id) {
2242                         res = 0;
2243                         strcpy(info->slave_name, slave->dev->name);
2244                         info->link = slave->link;
2245                         info->state = slave->state;
2246                         info->link_failure_count = slave->link_failure_count;
2247                         break;
2248                 }
2249         }
2250
2251         read_unlock(&bond->lock);
2252
2253         return res;
2254 }
2255
2256 /*-------------------------------- Monitoring -------------------------------*/
2257
2258
2259 static int bond_miimon_inspect(struct bonding *bond)
2260 {
2261         struct slave *slave;
2262         int i, link_state, commit = 0;
2263         bool ignore_updelay;
2264
2265         ignore_updelay = !bond->curr_active_slave ? true : false;
2266
2267         bond_for_each_slave(bond, slave, i) {
2268                 slave->new_link = BOND_LINK_NOCHANGE;
2269
2270                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2271
2272                 switch (slave->link) {
2273                 case BOND_LINK_UP:
2274                         if (link_state)
2275                                 continue;
2276
2277                         slave->link = BOND_LINK_FAIL;
2278                         slave->delay = bond->params.downdelay;
2279                         if (slave->delay) {
2280                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2281                                         bond->dev->name,
2282                                         (bond->params.mode ==
2283                                          BOND_MODE_ACTIVEBACKUP) ?
2284                                         ((slave->state == BOND_STATE_ACTIVE) ?
2285                                          "active " : "backup ") : "",
2286                                         slave->dev->name,
2287                                         bond->params.downdelay * bond->params.miimon);
2288                         }
2289                         /*FALLTHRU*/
2290                 case BOND_LINK_FAIL:
2291                         if (link_state) {
2292                                 /*
2293                                  * recovered before downdelay expired
2294                                  */
2295                                 slave->link = BOND_LINK_UP;
2296                                 slave->jiffies = jiffies;
2297                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2298                                         bond->dev->name,
2299                                         (bond->params.downdelay - slave->delay) *
2300                                         bond->params.miimon,
2301                                         slave->dev->name);
2302                                 continue;
2303                         }
2304
2305                         if (slave->delay <= 0) {
2306                                 slave->new_link = BOND_LINK_DOWN;
2307                                 commit++;
2308                                 continue;
2309                         }
2310
2311                         slave->delay--;
2312                         break;
2313
2314                 case BOND_LINK_DOWN:
2315                         if (!link_state)
2316                                 continue;
2317
2318                         slave->link = BOND_LINK_BACK;
2319                         slave->delay = bond->params.updelay;
2320
2321                         if (slave->delay) {
2322                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2323                                         bond->dev->name, slave->dev->name,
2324                                         ignore_updelay ? 0 :
2325                                         bond->params.updelay *
2326                                         bond->params.miimon);
2327                         }
2328                         /*FALLTHRU*/
2329                 case BOND_LINK_BACK:
2330                         if (!link_state) {
2331                                 slave->link = BOND_LINK_DOWN;
2332                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2333                                         bond->dev->name,
2334                                         (bond->params.updelay - slave->delay) *
2335                                         bond->params.miimon,
2336                                         slave->dev->name);
2337
2338                                 continue;
2339                         }
2340
2341                         if (ignore_updelay)
2342                                 slave->delay = 0;
2343
2344                         if (slave->delay <= 0) {
2345                                 slave->new_link = BOND_LINK_UP;
2346                                 commit++;
2347                                 ignore_updelay = false;
2348                                 continue;
2349                         }
2350
2351                         slave->delay--;
2352                         break;
2353                 }
2354         }
2355
2356         return commit;
2357 }
2358
2359 static void bond_miimon_commit(struct bonding *bond)
2360 {
2361         struct slave *slave;
2362         int i;
2363
2364         bond_for_each_slave(bond, slave, i) {
2365                 switch (slave->new_link) {
2366                 case BOND_LINK_NOCHANGE:
2367                         continue;
2368
2369                 case BOND_LINK_UP:
2370                         slave->link = BOND_LINK_UP;
2371                         slave->jiffies = jiffies;
2372
2373                         if (bond->params.mode == BOND_MODE_8023AD) {
2374                                 /* prevent it from being the active one */
2375                                 slave->state = BOND_STATE_BACKUP;
2376                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2377                                 /* make it immediately active */
2378                                 slave->state = BOND_STATE_ACTIVE;
2379                         } else if (slave != bond->primary_slave) {
2380                                 /* prevent it from being the active one */
2381                                 slave->state = BOND_STATE_BACKUP;
2382                         }
2383
2384                         bond_update_speed_duplex(slave);
2385
2386                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2387                                 bond->dev->name, slave->dev->name,
2388                                 slave->speed, slave->duplex ? "full" : "half");
2389
2390                         /* notify ad that the link status has changed */
2391                         if (bond->params.mode == BOND_MODE_8023AD)
2392                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2393
2394                         if (bond_is_lb(bond))
2395                                 bond_alb_handle_link_change(bond, slave,
2396                                                             BOND_LINK_UP);
2397
2398                         if (!bond->curr_active_slave ||
2399                             (slave == bond->primary_slave))
2400                                 goto do_failover;
2401
2402                         continue;
2403
2404                 case BOND_LINK_DOWN:
2405                         if (slave->link_failure_count < UINT_MAX)
2406                                 slave->link_failure_count++;
2407
2408                         slave->link = BOND_LINK_DOWN;
2409
2410                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2411                             bond->params.mode == BOND_MODE_8023AD)
2412                                 bond_set_slave_inactive_flags(slave);
2413
2414                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2415                                 bond->dev->name, slave->dev->name);
2416
2417                         if (bond->params.mode == BOND_MODE_8023AD)
2418                                 bond_3ad_handle_link_change(slave,
2419                                                             BOND_LINK_DOWN);
2420
2421                         if (bond_is_lb(bond))
2422                                 bond_alb_handle_link_change(bond, slave,
2423                                                             BOND_LINK_DOWN);
2424
2425                         if (slave == bond->curr_active_slave)
2426                                 goto do_failover;
2427
2428                         continue;
2429
2430                 default:
2431                         pr_err("%s: invalid new link %d on slave %s\n",
2432                                bond->dev->name, slave->new_link,
2433                                slave->dev->name);
2434                         slave->new_link = BOND_LINK_NOCHANGE;
2435
2436                         continue;
2437                 }
2438
2439 do_failover:
2440                 ASSERT_RTNL();
2441                 block_netpoll_tx();
2442                 write_lock_bh(&bond->curr_slave_lock);
2443                 bond_select_active_slave(bond);
2444                 write_unlock_bh(&bond->curr_slave_lock);
2445                 unblock_netpoll_tx();
2446         }
2447
2448         bond_set_carrier(bond);
2449 }
2450
2451 /*
2452  * bond_mii_monitor
2453  *
2454  * Really a wrapper that splits the mii monitor into two phases: an
2455  * inspection, then (if inspection indicates something needs to be done)
2456  * an acquisition of appropriate locks followed by a commit phase to
2457  * implement whatever link state changes are indicated.
2458  */
2459 void bond_mii_monitor(struct work_struct *work)
2460 {
2461         struct bonding *bond = container_of(work, struct bonding,
2462                                             mii_work.work);
2463
2464         read_lock(&bond->lock);
2465         if (bond->kill_timers)
2466                 goto out;
2467
2468         if (bond->slave_cnt == 0)
2469                 goto re_arm;
2470
2471         if (bond->send_grat_arp) {
2472                 read_lock(&bond->curr_slave_lock);
2473                 bond_send_gratuitous_arp(bond);
2474                 read_unlock(&bond->curr_slave_lock);
2475         }
2476
2477         if (bond->send_unsol_na) {
2478                 read_lock(&bond->curr_slave_lock);
2479                 bond_send_unsolicited_na(bond);
2480                 read_unlock(&bond->curr_slave_lock);
2481         }
2482
2483         if (bond_miimon_inspect(bond)) {
2484                 read_unlock(&bond->lock);
2485                 rtnl_lock();
2486                 read_lock(&bond->lock);
2487
2488                 bond_miimon_commit(bond);
2489
2490                 read_unlock(&bond->lock);
2491                 rtnl_unlock();  /* might sleep, hold no other locks */
2492                 read_lock(&bond->lock);
2493         }
2494
2495 re_arm:
2496         if (bond->params.miimon)
2497                 queue_delayed_work(bond->wq, &bond->mii_work,
2498                                    msecs_to_jiffies(bond->params.miimon));
2499 out:
2500         read_unlock(&bond->lock);
2501 }
2502
2503 static __be32 bond_glean_dev_ip(struct net_device *dev)
2504 {
2505         struct in_device *idev;
2506         struct in_ifaddr *ifa;
2507         __be32 addr = 0;
2508
2509         if (!dev)
2510                 return 0;
2511
2512         rcu_read_lock();
2513         idev = __in_dev_get_rcu(dev);
2514         if (!idev)
2515                 goto out;
2516
2517         ifa = idev->ifa_list;
2518         if (!ifa)
2519                 goto out;
2520
2521         addr = ifa->ifa_local;
2522 out:
2523         rcu_read_unlock();
2524         return addr;
2525 }
2526
2527 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2528 {
2529         struct vlan_entry *vlan;
2530
2531         if (ip == bond->master_ip)
2532                 return 1;
2533
2534         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2535                 if (ip == vlan->vlan_ip)
2536                         return 1;
2537         }
2538
2539         return 0;
2540 }
2541
2542 /*
2543  * We go to the (large) trouble of VLAN tagging ARP frames because
2544  * switches in VLAN mode (especially if ports are configured as
2545  * "native" to a VLAN) might not pass non-tagged frames.
2546  */
2547 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2548 {
2549         struct sk_buff *skb;
2550
2551         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2552                  slave_dev->name, dest_ip, src_ip, vlan_id);
2553
2554         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2555                          NULL, slave_dev->dev_addr, NULL);
2556
2557         if (!skb) {
2558                 pr_err("ARP packet allocation failed\n");
2559                 return;
2560         }
2561         if (vlan_id) {
2562                 skb = vlan_put_tag(skb, vlan_id);
2563                 if (!skb) {
2564                         pr_err("failed to insert VLAN tag\n");
2565                         return;
2566                 }
2567         }
2568         arp_xmit(skb);
2569 }
2570
2571
2572 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2573 {
2574         int i, vlan_id, rv;
2575         __be32 *targets = bond->params.arp_targets;
2576         struct vlan_entry *vlan;
2577         struct net_device *vlan_dev;
2578         struct flowi fl;
2579         struct rtable *rt;
2580
2581         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2582                 if (!targets[i])
2583                         break;
2584                 pr_debug("basa: target %x\n", targets[i]);
2585                 if (!bond->vlgrp) {
2586                         pr_debug("basa: empty vlan: arp_send\n");
2587                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2588                                       bond->master_ip, 0);
2589                         continue;
2590                 }
2591
2592                 /*
2593                  * If VLANs are configured, we do a route lookup to
2594                  * determine which VLAN interface would be used, so we
2595                  * can tag the ARP with the proper VLAN tag.
2596                  */
2597                 memset(&fl, 0, sizeof(fl));
2598                 fl.fl4_dst = targets[i];
2599                 fl.fl4_tos = RTO_ONLINK;
2600
2601                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2602                 if (rv) {
2603                         if (net_ratelimit()) {
2604                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2605                                            bond->dev->name, &fl.fl4_dst);
2606                         }
2607                         continue;
2608                 }
2609
2610                 /*
2611                  * This target is not on a VLAN
2612                  */
2613                 if (rt->dst.dev == bond->dev) {
2614                         ip_rt_put(rt);
2615                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2616                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617                                       bond->master_ip, 0);
2618                         continue;
2619                 }
2620
2621                 vlan_id = 0;
2622                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2623                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2624                         if (vlan_dev == rt->dst.dev) {
2625                                 vlan_id = vlan->vlan_id;
2626                                 pr_debug("basa: vlan match on %s %d\n",
2627                                        vlan_dev->name, vlan_id);
2628                                 break;
2629                         }
2630                 }
2631
2632                 if (vlan_id) {
2633                         ip_rt_put(rt);
2634                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2635                                       vlan->vlan_ip, vlan_id);
2636                         continue;
2637                 }
2638
2639                 if (net_ratelimit()) {
2640                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2641                                    bond->dev->name, &fl.fl4_dst,
2642                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2643                 }
2644                 ip_rt_put(rt);
2645         }
2646 }
2647
2648 /*
2649  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2650  * for each VLAN above us.
2651  *
2652  * Caller must hold curr_slave_lock for read or better
2653  */
2654 static void bond_send_gratuitous_arp(struct bonding *bond)
2655 {
2656         struct slave *slave = bond->curr_active_slave;
2657         struct vlan_entry *vlan;
2658         struct net_device *vlan_dev;
2659
2660         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2661                  bond->dev->name, slave ? slave->dev->name : "NULL");
2662
2663         if (!slave || !bond->send_grat_arp ||
2664             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2665                 return;
2666
2667         bond->send_grat_arp--;
2668
2669         if (bond->master_ip) {
2670                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2671                                 bond->master_ip, 0);
2672         }
2673
2674         if (!bond->vlgrp)
2675                 return;
2676
2677         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2678                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2679                 if (vlan->vlan_ip) {
2680                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2681                                       vlan->vlan_ip, vlan->vlan_id);
2682                 }
2683         }
2684 }
2685
2686 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2687 {
2688         int i;
2689         __be32 *targets = bond->params.arp_targets;
2690
2691         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2692                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2693                          &sip, &tip, i, &targets[i],
2694                          bond_has_this_ip(bond, tip));
2695                 if (sip == targets[i]) {
2696                         if (bond_has_this_ip(bond, tip))
2697                                 slave->last_arp_rx = jiffies;
2698                         return;
2699                 }
2700         }
2701 }
2702
2703 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2704 {
2705         struct arphdr *arp;
2706         struct slave *slave;
2707         struct bonding *bond;
2708         unsigned char *arp_ptr;
2709         __be32 sip, tip;
2710
2711         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2712                 /*
2713                  * When using VLANS and bonding, dev and oriv_dev may be
2714                  * incorrect if the physical interface supports VLAN
2715                  * acceleration.  With this change ARP validation now
2716                  * works for hosts only reachable on the VLAN interface.
2717                  */
2718                 dev = vlan_dev_real_dev(dev);
2719                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2720         }
2721
2722         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2723                 goto out;
2724
2725         bond = netdev_priv(dev);
2726         read_lock(&bond->lock);
2727
2728         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2729                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2730                  orig_dev ? orig_dev->name : "NULL");
2731
2732         slave = bond_get_slave_by_dev(bond, orig_dev);
2733         if (!slave || !slave_do_arp_validate(bond, slave))
2734                 goto out_unlock;
2735
2736         skb = skb_share_check(skb, GFP_ATOMIC);
2737         if (!skb)
2738                 goto out_unlock;
2739
2740         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2741                 goto out_unlock;
2742
2743         arp = arp_hdr(skb);
2744         if (arp->ar_hln != dev->addr_len ||
2745             skb->pkt_type == PACKET_OTHERHOST ||
2746             skb->pkt_type == PACKET_LOOPBACK ||
2747             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2748             arp->ar_pro != htons(ETH_P_IP) ||
2749             arp->ar_pln != 4)
2750                 goto out_unlock;
2751
2752         arp_ptr = (unsigned char *)(arp + 1);
2753         arp_ptr += dev->addr_len;
2754         memcpy(&sip, arp_ptr, 4);
2755         arp_ptr += 4 + dev->addr_len;
2756         memcpy(&tip, arp_ptr, 4);
2757
2758         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2759                  bond->dev->name, slave->dev->name, slave->state,
2760                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2761                  &sip, &tip);
2762
2763         /*
2764          * Backup slaves won't see the ARP reply, but do come through
2765          * here for each ARP probe (so we swap the sip/tip to validate
2766          * the probe).  In a "redundant switch, common router" type of
2767          * configuration, the ARP probe will (hopefully) travel from
2768          * the active, through one switch, the router, then the other
2769          * switch before reaching the backup.
2770          */
2771         if (slave->state == BOND_STATE_ACTIVE)
2772                 bond_validate_arp(bond, slave, sip, tip);
2773         else
2774                 bond_validate_arp(bond, slave, tip, sip);
2775
2776 out_unlock:
2777         read_unlock(&bond->lock);
2778 out:
2779         dev_kfree_skb(skb);
2780         return NET_RX_SUCCESS;
2781 }
2782
2783 /*
2784  * this function is called regularly to monitor each slave's link
2785  * ensuring that traffic is being sent and received when arp monitoring
2786  * is used in load-balancing mode. if the adapter has been dormant, then an
2787  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2788  * arp monitoring in active backup mode.
2789  */
2790 void bond_loadbalance_arp_mon(struct work_struct *work)
2791 {
2792         struct bonding *bond = container_of(work, struct bonding,
2793                                             arp_work.work);
2794         struct slave *slave, *oldcurrent;
2795         int do_failover = 0;
2796         int delta_in_ticks;
2797         int i;
2798
2799         read_lock(&bond->lock);
2800
2801         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2802
2803         if (bond->kill_timers)
2804                 goto out;
2805
2806         if (bond->slave_cnt == 0)
2807                 goto re_arm;
2808
2809         read_lock(&bond->curr_slave_lock);
2810         oldcurrent = bond->curr_active_slave;
2811         read_unlock(&bond->curr_slave_lock);
2812
2813         /* see if any of the previous devices are up now (i.e. they have
2814          * xmt and rcv traffic). the curr_active_slave does not come into
2815          * the picture unless it is null. also, slave->jiffies is not needed
2816          * here because we send an arp on each slave and give a slave as
2817          * long as it needs to get the tx/rx within the delta.
2818          * TODO: what about up/down delay in arp mode? it wasn't here before
2819          *       so it can wait
2820          */
2821         bond_for_each_slave(bond, slave, i) {
2822                 unsigned long trans_start = dev_trans_start(slave->dev);
2823
2824                 if (slave->link != BOND_LINK_UP) {
2825                         if (time_in_range(jiffies,
2826                                 trans_start - delta_in_ticks,
2827                                 trans_start + delta_in_ticks) &&
2828                             time_in_range(jiffies,
2829                                 slave->dev->last_rx - delta_in_ticks,
2830                                 slave->dev->last_rx + delta_in_ticks)) {
2831
2832                                 slave->link  = BOND_LINK_UP;
2833                                 slave->state = BOND_STATE_ACTIVE;
2834
2835                                 /* primary_slave has no meaning in round-robin
2836                                  * mode. the window of a slave being up and
2837                                  * curr_active_slave being null after enslaving
2838                                  * is closed.
2839                                  */
2840                                 if (!oldcurrent) {
2841                                         pr_info("%s: link status definitely up for interface %s, ",
2842                                                 bond->dev->name,
2843                                                 slave->dev->name);
2844                                         do_failover = 1;
2845                                 } else {
2846                                         pr_info("%s: interface %s is now up\n",
2847                                                 bond->dev->name,
2848                                                 slave->dev->name);
2849                                 }
2850                         }
2851                 } else {
2852                         /* slave->link == BOND_LINK_UP */
2853
2854                         /* not all switches will respond to an arp request
2855                          * when the source ip is 0, so don't take the link down
2856                          * if we don't know our ip yet
2857                          */
2858                         if (!time_in_range(jiffies,
2859                                 trans_start - delta_in_ticks,
2860                                 trans_start + 2 * delta_in_ticks) ||
2861                             !time_in_range(jiffies,
2862                                 slave->dev->last_rx - delta_in_ticks,
2863                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2864
2865                                 slave->link  = BOND_LINK_DOWN;
2866                                 slave->state = BOND_STATE_BACKUP;
2867
2868                                 if (slave->link_failure_count < UINT_MAX)
2869                                         slave->link_failure_count++;
2870
2871                                 pr_info("%s: interface %s is now down.\n",
2872                                         bond->dev->name,
2873                                         slave->dev->name);
2874
2875                                 if (slave == oldcurrent)
2876                                         do_failover = 1;
2877                         }
2878                 }
2879
2880                 /* note: if switch is in round-robin mode, all links
2881                  * must tx arp to ensure all links rx an arp - otherwise
2882                  * links may oscillate or not come up at all; if switch is
2883                  * in something like xor mode, there is nothing we can
2884                  * do - all replies will be rx'ed on same link causing slaves
2885                  * to be unstable during low/no traffic periods
2886                  */
2887                 if (IS_UP(slave->dev))
2888                         bond_arp_send_all(bond, slave);
2889         }
2890
2891         if (do_failover) {
2892                 block_netpoll_tx();
2893                 write_lock_bh(&bond->curr_slave_lock);
2894
2895                 bond_select_active_slave(bond);
2896
2897                 write_unlock_bh(&bond->curr_slave_lock);
2898                 unblock_netpoll_tx();
2899         }
2900
2901 re_arm:
2902         if (bond->params.arp_interval)
2903                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2904 out:
2905         read_unlock(&bond->lock);
2906 }
2907
2908 /*
2909  * Called to inspect slaves for active-backup mode ARP monitor link state
2910  * changes.  Sets new_link in slaves to specify what action should take
2911  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2912  * to link states must be committed.
2913  *
2914  * Called with bond->lock held for read.
2915  */
2916 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2917 {
2918         struct slave *slave;
2919         int i, commit = 0;
2920         unsigned long trans_start;
2921
2922         bond_for_each_slave(bond, slave, i) {
2923                 slave->new_link = BOND_LINK_NOCHANGE;
2924
2925                 if (slave->link != BOND_LINK_UP) {
2926                         if (time_in_range(jiffies,
2927                                 slave_last_rx(bond, slave) - delta_in_ticks,
2928                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2929
2930                                 slave->new_link = BOND_LINK_UP;
2931                                 commit++;
2932                         }
2933
2934                         continue;
2935                 }
2936
2937                 /*
2938                  * Give slaves 2*delta after being enslaved or made
2939                  * active.  This avoids bouncing, as the last receive
2940                  * times need a full ARP monitor cycle to be updated.
2941                  */
2942                 if (time_in_range(jiffies,
2943                                   slave->jiffies - delta_in_ticks,
2944                                   slave->jiffies + 2 * delta_in_ticks))
2945                         continue;
2946
2947                 /*
2948                  * Backup slave is down if:
2949                  * - No current_arp_slave AND
2950                  * - more than 3*delta since last receive AND
2951                  * - the bond has an IP address
2952                  *
2953                  * Note: a non-null current_arp_slave indicates
2954                  * the curr_active_slave went down and we are
2955                  * searching for a new one; under this condition
2956                  * we only take the curr_active_slave down - this
2957                  * gives each slave a chance to tx/rx traffic
2958                  * before being taken out
2959                  */
2960                 if (slave->state == BOND_STATE_BACKUP &&
2961                     !bond->current_arp_slave &&
2962                     !time_in_range(jiffies,
2963                         slave_last_rx(bond, slave) - delta_in_ticks,
2964                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2965
2966                         slave->new_link = BOND_LINK_DOWN;
2967                         commit++;
2968                 }
2969
2970                 /*
2971                  * Active slave is down if:
2972                  * - more than 2*delta since transmitting OR
2973                  * - (more than 2*delta since receive AND
2974                  *    the bond has an IP address)
2975                  */
2976                 trans_start = dev_trans_start(slave->dev);
2977                 if ((slave->state == BOND_STATE_ACTIVE) &&
2978                     (!time_in_range(jiffies,
2979                         trans_start - delta_in_ticks,
2980                         trans_start + 2 * delta_in_ticks) ||
2981                      !time_in_range(jiffies,
2982                         slave_last_rx(bond, slave) - delta_in_ticks,
2983                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2984
2985                         slave->new_link = BOND_LINK_DOWN;
2986                         commit++;
2987                 }
2988         }
2989
2990         return commit;
2991 }
2992
2993 /*
2994  * Called to commit link state changes noted by inspection step of
2995  * active-backup mode ARP monitor.
2996  *
2997  * Called with RTNL and bond->lock for read.
2998  */
2999 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3000 {
3001         struct slave *slave;
3002         int i;
3003         unsigned long trans_start;
3004
3005         bond_for_each_slave(bond, slave, i) {
3006                 switch (slave->new_link) {
3007                 case BOND_LINK_NOCHANGE:
3008                         continue;
3009
3010                 case BOND_LINK_UP:
3011                         trans_start = dev_trans_start(slave->dev);
3012                         if ((!bond->curr_active_slave &&
3013                              time_in_range(jiffies,
3014                                            trans_start - delta_in_ticks,
3015                                            trans_start + delta_in_ticks)) ||
3016                             bond->curr_active_slave != slave) {
3017                                 slave->link = BOND_LINK_UP;
3018                                 bond->current_arp_slave = NULL;
3019
3020                                 pr_info("%s: link status definitely up for interface %s.\n",
3021                                         bond->dev->name, slave->dev->name);
3022
3023                                 if (!bond->curr_active_slave ||
3024                                     (slave == bond->primary_slave))
3025                                         goto do_failover;
3026
3027                         }
3028
3029                         continue;
3030
3031                 case BOND_LINK_DOWN:
3032                         if (slave->link_failure_count < UINT_MAX)
3033                                 slave->link_failure_count++;
3034
3035                         slave->link = BOND_LINK_DOWN;
3036                         bond_set_slave_inactive_flags(slave);
3037
3038                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3039                                 bond->dev->name, slave->dev->name);
3040
3041                         if (slave == bond->curr_active_slave) {
3042                                 bond->current_arp_slave = NULL;
3043                                 goto do_failover;
3044                         }
3045
3046                         continue;
3047
3048                 default:
3049                         pr_err("%s: impossible: new_link %d on slave %s\n",
3050                                bond->dev->name, slave->new_link,
3051                                slave->dev->name);
3052                         continue;
3053                 }
3054
3055 do_failover:
3056                 ASSERT_RTNL();
3057                 block_netpoll_tx();
3058                 write_lock_bh(&bond->curr_slave_lock);
3059                 bond_select_active_slave(bond);
3060                 write_unlock_bh(&bond->curr_slave_lock);
3061                 unblock_netpoll_tx();
3062         }
3063
3064         bond_set_carrier(bond);
3065 }
3066
3067 /*
3068  * Send ARP probes for active-backup mode ARP monitor.
3069  *
3070  * Called with bond->lock held for read.
3071  */
3072 static void bond_ab_arp_probe(struct bonding *bond)
3073 {
3074         struct slave *slave;
3075         int i;
3076
3077         read_lock(&bond->curr_slave_lock);
3078
3079         if (bond->current_arp_slave && bond->curr_active_slave)
3080                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3081                         bond->current_arp_slave->dev->name,
3082                         bond->curr_active_slave->dev->name);
3083
3084         if (bond->curr_active_slave) {
3085                 bond_arp_send_all(bond, bond->curr_active_slave);
3086                 read_unlock(&bond->curr_slave_lock);
3087                 return;
3088         }
3089
3090         read_unlock(&bond->curr_slave_lock);
3091
3092         /* if we don't have a curr_active_slave, search for the next available
3093          * backup slave from the current_arp_slave and make it the candidate
3094          * for becoming the curr_active_slave
3095          */
3096
3097         if (!bond->current_arp_slave) {
3098                 bond->current_arp_slave = bond->first_slave;
3099                 if (!bond->current_arp_slave)
3100                         return;
3101         }
3102
3103         bond_set_slave_inactive_flags(bond->current_arp_slave);
3104
3105         /* search for next candidate */
3106         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3107                 if (IS_UP(slave->dev)) {
3108                         slave->link = BOND_LINK_BACK;
3109                         bond_set_slave_active_flags(slave);
3110                         bond_arp_send_all(bond, slave);
3111                         slave->jiffies = jiffies;
3112                         bond->current_arp_slave = slave;
3113                         break;
3114                 }
3115
3116                 /* if the link state is up at this point, we
3117                  * mark it down - this can happen if we have
3118                  * simultaneous link failures and
3119                  * reselect_active_interface doesn't make this
3120                  * one the current slave so it is still marked
3121                  * up when it is actually down
3122                  */
3123                 if (slave->link == BOND_LINK_UP) {
3124                         slave->link = BOND_LINK_DOWN;
3125                         if (slave->link_failure_count < UINT_MAX)
3126                                 slave->link_failure_count++;
3127
3128                         bond_set_slave_inactive_flags(slave);
3129
3130                         pr_info("%s: backup interface %s is now down.\n",
3131                                 bond->dev->name, slave->dev->name);
3132                 }
3133         }
3134 }
3135
3136 void bond_activebackup_arp_mon(struct work_struct *work)
3137 {
3138         struct bonding *bond = container_of(work, struct bonding,
3139                                             arp_work.work);
3140         int delta_in_ticks;
3141
3142         read_lock(&bond->lock);
3143
3144         if (bond->kill_timers)
3145                 goto out;
3146
3147         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3148
3149         if (bond->slave_cnt == 0)
3150                 goto re_arm;
3151
3152         if (bond->send_grat_arp) {
3153                 read_lock(&bond->curr_slave_lock);
3154                 bond_send_gratuitous_arp(bond);
3155                 read_unlock(&bond->curr_slave_lock);
3156         }
3157
3158         if (bond->send_unsol_na) {
3159                 read_lock(&bond->curr_slave_lock);
3160                 bond_send_unsolicited_na(bond);
3161                 read_unlock(&bond->curr_slave_lock);
3162         }
3163
3164         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3165                 read_unlock(&bond->lock);
3166                 rtnl_lock();
3167                 read_lock(&bond->lock);
3168
3169                 bond_ab_arp_commit(bond, delta_in_ticks);
3170
3171                 read_unlock(&bond->lock);
3172                 rtnl_unlock();
3173                 read_lock(&bond->lock);
3174         }
3175
3176         bond_ab_arp_probe(bond);
3177
3178 re_arm:
3179         if (bond->params.arp_interval)
3180                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3181 out:
3182         read_unlock(&bond->lock);
3183 }
3184
3185 /*------------------------------ proc/seq_file-------------------------------*/
3186
3187 #ifdef CONFIG_PROC_FS
3188
3189 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3190         __acquires(RCU)
3191         __acquires(&bond->lock)
3192 {
3193         struct bonding *bond = seq->private;
3194         loff_t off = 0;
3195         struct slave *slave;
3196         int i;
3197
3198         /* make sure the bond won't be taken away */
3199         rcu_read_lock();
3200         read_lock(&bond->lock);
3201
3202         if (*pos == 0)
3203                 return SEQ_START_TOKEN;
3204
3205         bond_for_each_slave(bond, slave, i) {
3206                 if (++off == *pos)
3207                         return slave;
3208         }
3209
3210         return NULL;
3211 }
3212
3213 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3214 {
3215         struct bonding *bond = seq->private;
3216         struct slave *slave = v;
3217
3218         ++*pos;
3219         if (v == SEQ_START_TOKEN)
3220                 return bond->first_slave;
3221
3222         slave = slave->next;
3223
3224         return (slave == bond->first_slave) ? NULL : slave;
3225 }
3226
3227 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3228         __releases(&bond->lock)
3229         __releases(RCU)
3230 {
3231         struct bonding *bond = seq->private;
3232
3233         read_unlock(&bond->lock);
3234         rcu_read_unlock();
3235 }
3236
3237 static void bond_info_show_master(struct seq_file *seq)
3238 {
3239         struct bonding *bond = seq->private;
3240         struct slave *curr;
3241         int i;
3242
3243         read_lock(&bond->curr_slave_lock);
3244         curr = bond->curr_active_slave;
3245         read_unlock(&bond->curr_slave_lock);
3246
3247         seq_printf(seq, "Bonding Mode: %s",
3248                    bond_mode_name(bond->params.mode));
3249
3250         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3251             bond->params.fail_over_mac)
3252                 seq_printf(seq, " (fail_over_mac %s)",
3253                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3254
3255         seq_printf(seq, "\n");
3256
3257         if (bond->params.mode == BOND_MODE_XOR ||
3258                 bond->params.mode == BOND_MODE_8023AD) {
3259                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3260                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3261                         bond->params.xmit_policy);
3262         }
3263
3264         if (USES_PRIMARY(bond->params.mode)) {
3265                 seq_printf(seq, "Primary Slave: %s",
3266                            (bond->primary_slave) ?
3267                            bond->primary_slave->dev->name : "None");
3268                 if (bond->primary_slave)
3269                         seq_printf(seq, " (primary_reselect %s)",
3270                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3271
3272                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3273                            (curr) ? curr->dev->name : "None");
3274         }
3275
3276         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3277                    "up" : "down");
3278         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3279         seq_printf(seq, "Up Delay (ms): %d\n",
3280                    bond->params.updelay * bond->params.miimon);
3281         seq_printf(seq, "Down Delay (ms): %d\n",
3282                    bond->params.downdelay * bond->params.miimon);
3283
3284
3285         /* ARP information */
3286         if (bond->params.arp_interval > 0) {
3287                 int printed = 0;
3288                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3289                                 bond->params.arp_interval);
3290
3291                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3292
3293                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3294                         if (!bond->params.arp_targets[i])
3295                                 break;
3296                         if (printed)
3297                                 seq_printf(seq, ",");
3298                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3299                         printed = 1;
3300                 }
3301                 seq_printf(seq, "\n");
3302         }
3303
3304         if (bond->params.mode == BOND_MODE_8023AD) {
3305                 struct ad_info ad_info;
3306
3307                 seq_puts(seq, "\n802.3ad info\n");
3308                 seq_printf(seq, "LACP rate: %s\n",
3309                            (bond->params.lacp_fast) ? "fast" : "slow");
3310                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3311                            ad_select_tbl[bond->params.ad_select].modename);
3312
3313                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3314                         seq_printf(seq, "bond %s has no active aggregator\n",
3315                                    bond->dev->name);
3316                 } else {
3317                         seq_printf(seq, "Active Aggregator Info:\n");
3318
3319                         seq_printf(seq, "\tAggregator ID: %d\n",
3320                                    ad_info.aggregator_id);
3321                         seq_printf(seq, "\tNumber of ports: %d\n",
3322                                    ad_info.ports);
3323                         seq_printf(seq, "\tActor Key: %d\n",
3324                                    ad_info.actor_key);
3325                         seq_printf(seq, "\tPartner Key: %d\n",
3326                                    ad_info.partner_key);
3327                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3328                                    ad_info.partner_system);
3329                 }
3330         }
3331 }
3332
3333 static void bond_info_show_slave(struct seq_file *seq,
3334                                  const struct slave *slave)
3335 {
3336         struct bonding *bond = seq->private;
3337
3338         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3339         seq_printf(seq, "MII Status: %s\n",
3340                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3341         seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3342         seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3343         seq_printf(seq, "Link Failure Count: %u\n",
3344                    slave->link_failure_count);
3345
3346         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3347
3348         if (bond->params.mode == BOND_MODE_8023AD) {
3349                 const struct aggregator *agg
3350                         = SLAVE_AD_INFO(slave).port.aggregator;
3351
3352                 if (agg)
3353                         seq_printf(seq, "Aggregator ID: %d\n",
3354                                    agg->aggregator_identifier);
3355                 else
3356                         seq_puts(seq, "Aggregator ID: N/A\n");
3357         }
3358         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3359 }
3360
3361 static int bond_info_seq_show(struct seq_file *seq, void *v)
3362 {
3363         if (v == SEQ_START_TOKEN) {
3364                 seq_printf(seq, "%s\n", version);
3365                 bond_info_show_master(seq);
3366         } else
3367                 bond_info_show_slave(seq, v);
3368
3369         return 0;
3370 }
3371
3372 static const struct seq_operations bond_info_seq_ops = {
3373         .start = bond_info_seq_start,
3374         .next  = bond_info_seq_next,
3375         .stop  = bond_info_seq_stop,
3376         .show  = bond_info_seq_show,
3377 };
3378
3379 static int bond_info_open(struct inode *inode, struct file *file)
3380 {
3381         struct seq_file *seq;
3382         struct proc_dir_entry *proc;
3383         int res;
3384
3385         res = seq_open(file, &bond_info_seq_ops);
3386         if (!res) {
3387                 /* recover the pointer buried in proc_dir_entry data */
3388                 seq = file->private_data;
3389                 proc = PDE(inode);
3390                 seq->private = proc->data;
3391         }
3392
3393         return res;
3394 }
3395
3396 static const struct file_operations bond_info_fops = {
3397         .owner   = THIS_MODULE,
3398         .open    = bond_info_open,
3399         .read    = seq_read,
3400         .llseek  = seq_lseek,
3401         .release = seq_release,
3402 };
3403
3404 static void bond_create_proc_entry(struct bonding *bond)
3405 {
3406         struct net_device *bond_dev = bond->dev;
3407         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3408
3409         if (bn->proc_dir) {
3410                 bond->proc_entry = proc_create_data(bond_dev->name,
3411                                                     S_IRUGO, bn->proc_dir,
3412                                                     &bond_info_fops, bond);
3413                 if (bond->proc_entry == NULL)
3414                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3415                                    DRV_NAME, bond_dev->name);
3416                 else
3417                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3418         }
3419 }
3420
3421 static void bond_remove_proc_entry(struct bonding *bond)
3422 {
3423         struct net_device *bond_dev = bond->dev;
3424         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3425
3426         if (bn->proc_dir && bond->proc_entry) {
3427                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3428                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3429                 bond->proc_entry = NULL;
3430         }
3431 }
3432
3433 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3434  * Caller must hold rtnl_lock.
3435  */
3436 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3437 {
3438         if (!bn->proc_dir) {
3439                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3440                 if (!bn->proc_dir)
3441                         pr_warning("Warning: cannot create /proc/net/%s\n",
3442                                    DRV_NAME);
3443         }
3444 }
3445
3446 /* Destroy the bonding directory under /proc/net, if empty.
3447  * Caller must hold rtnl_lock.
3448  */
3449 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3450 {
3451         if (bn->proc_dir) {
3452                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3453                 bn->proc_dir = NULL;
3454         }
3455 }
3456
3457 #else /* !CONFIG_PROC_FS */
3458
3459 static void bond_create_proc_entry(struct bonding *bond)
3460 {
3461 }
3462
3463 static void bond_remove_proc_entry(struct bonding *bond)
3464 {
3465 }
3466
3467 static inline void bond_create_proc_dir(struct bond_net *bn)
3468 {
3469 }
3470
3471 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3472 {
3473 }
3474
3475 #endif /* CONFIG_PROC_FS */
3476
3477
3478 /*-------------------------- netdev event handling --------------------------*/
3479
3480 /*
3481  * Change device name
3482  */
3483 static int bond_event_changename(struct bonding *bond)
3484 {
3485         bond_remove_proc_entry(bond);
3486         bond_create_proc_entry(bond);
3487
3488         bond_debug_reregister(bond);
3489
3490         return NOTIFY_DONE;
3491 }
3492
3493 static int bond_master_netdev_event(unsigned long event,
3494                                     struct net_device *bond_dev)
3495 {
3496         struct bonding *event_bond = netdev_priv(bond_dev);
3497
3498         switch (event) {
3499         case NETDEV_CHANGENAME:
3500                 return bond_event_changename(event_bond);
3501         default:
3502                 break;
3503         }
3504
3505         return NOTIFY_DONE;
3506 }
3507
3508 static int bond_slave_netdev_event(unsigned long event,
3509                                    struct net_device *slave_dev)
3510 {
3511         struct net_device *bond_dev = slave_dev->master;
3512         struct bonding *bond = netdev_priv(bond_dev);
3513
3514         switch (event) {
3515         case NETDEV_UNREGISTER:
3516                 if (bond_dev) {
3517                         if (bond->setup_by_slave)
3518                                 bond_release_and_destroy(bond_dev, slave_dev);
3519                         else
3520                                 bond_release(bond_dev, slave_dev);
3521                 }
3522                 break;
3523         case NETDEV_CHANGE:
3524                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3525                         struct slave *slave;
3526
3527                         slave = bond_get_slave_by_dev(bond, slave_dev);
3528                         if (slave) {
3529                                 u16 old_speed = slave->speed;
3530                                 u16 old_duplex = slave->duplex;
3531
3532                                 bond_update_speed_duplex(slave);
3533
3534                                 if (bond_is_lb(bond))
3535                                         break;
3536
3537                                 if (old_speed != slave->speed)
3538                                         bond_3ad_adapter_speed_changed(slave);
3539                                 if (old_duplex != slave->duplex)
3540                                         bond_3ad_adapter_duplex_changed(slave);
3541                         }
3542                 }
3543
3544                 break;
3545         case NETDEV_DOWN:
3546                 /*
3547                  * ... Or is it this?
3548                  */
3549                 break;
3550         case NETDEV_CHANGEMTU:
3551                 /*
3552                  * TODO: Should slaves be allowed to
3553                  * independently alter their MTU?  For
3554                  * an active-backup bond, slaves need
3555                  * not be the same type of device, so
3556                  * MTUs may vary.  For other modes,
3557                  * slaves arguably should have the
3558                  * same MTUs. To do this, we'd need to
3559                  * take over the slave's change_mtu
3560                  * function for the duration of their
3561                  * servitude.
3562                  */
3563                 break;
3564         case NETDEV_CHANGENAME:
3565                 /*
3566                  * TODO: handle changing the primary's name
3567                  */
3568                 break;
3569         case NETDEV_FEAT_CHANGE:
3570                 bond_compute_features(bond);
3571                 break;
3572         default:
3573                 break;
3574         }
3575
3576         return NOTIFY_DONE;
3577 }
3578
3579 /*
3580  * bond_netdev_event: handle netdev notifier chain events.
3581  *
3582  * This function receives events for the netdev chain.  The caller (an
3583  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3584  * locks for us to safely manipulate the slave devices (RTNL lock,
3585  * dev_probe_lock).
3586  */
3587 static int bond_netdev_event(struct notifier_block *this,
3588                              unsigned long event, void *ptr)
3589 {
3590         struct net_device *event_dev = (struct net_device *)ptr;
3591
3592         pr_debug("event_dev: %s, event: %lx\n",
3593                  event_dev ? event_dev->name : "None",
3594                  event);
3595
3596         if (!(event_dev->priv_flags & IFF_BONDING))
3597                 return NOTIFY_DONE;
3598
3599         if (event_dev->flags & IFF_MASTER) {
3600                 pr_debug("IFF_MASTER\n");
3601                 return bond_master_netdev_event(event, event_dev);
3602         }
3603
3604         if (event_dev->flags & IFF_SLAVE) {
3605                 pr_debug("IFF_SLAVE\n");
3606                 return bond_slave_netdev_event(event, event_dev);
3607         }
3608
3609         return NOTIFY_DONE;
3610 }
3611
3612 /*
3613  * bond_inetaddr_event: handle inetaddr notifier chain events.
3614  *
3615  * We keep track of device IPs primarily to use as source addresses in
3616  * ARP monitor probes (rather than spewing out broadcasts all the time).
3617  *
3618  * We track one IP for the main device (if it has one), plus one per VLAN.
3619  */
3620 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3621 {
3622         struct in_ifaddr *ifa = ptr;
3623         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3624         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3625         struct bonding *bond;
3626         struct vlan_entry *vlan;
3627
3628         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3629                 if (bond->dev == event_dev) {
3630                         switch (event) {
3631                         case NETDEV_UP:
3632                                 bond->master_ip = ifa->ifa_local;
3633                                 return NOTIFY_OK;
3634                         case NETDEV_DOWN:
3635                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3636                                 return NOTIFY_OK;
3637                         default:
3638                                 return NOTIFY_DONE;
3639                         }
3640                 }
3641
3642                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3643                         if (!bond->vlgrp)
3644                                 continue;
3645                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3646                         if (vlan_dev == event_dev) {
3647                                 switch (event) {
3648                                 case NETDEV_UP:
3649                                         vlan->vlan_ip = ifa->ifa_local;
3650                                         return NOTIFY_OK;
3651                                 case NETDEV_DOWN:
3652                                         vlan->vlan_ip =
3653                                                 bond_glean_dev_ip(vlan_dev);
3654                                         return NOTIFY_OK;
3655                                 default:
3656                                         return NOTIFY_DONE;
3657                                 }
3658                         }
3659                 }
3660         }
3661         return NOTIFY_DONE;
3662 }
3663
3664 static struct notifier_block bond_netdev_notifier = {
3665         .notifier_call = bond_netdev_event,
3666 };
3667
3668 static struct notifier_block bond_inetaddr_notifier = {
3669         .notifier_call = bond_inetaddr_event,
3670 };
3671
3672 /*-------------------------- Packet type handling ---------------------------*/
3673
3674 /* register to receive lacpdus on a bond */
3675 static void bond_register_lacpdu(struct bonding *bond)
3676 {
3677         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3678
3679         /* initialize packet type */
3680         pk_type->type = PKT_TYPE_LACPDU;
3681         pk_type->dev = bond->dev;
3682         pk_type->func = bond_3ad_lacpdu_recv;
3683
3684         dev_add_pack(pk_type);
3685 }
3686
3687 /* unregister to receive lacpdus on a bond */
3688 static void bond_unregister_lacpdu(struct bonding *bond)
3689 {
3690         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3691 }
3692
3693 void bond_register_arp(struct bonding *bond)
3694 {
3695         struct packet_type *pt = &bond->arp_mon_pt;
3696
3697         if (pt->type)
3698                 return;
3699
3700         pt->type = htons(ETH_P_ARP);
3701         pt->dev = bond->dev;
3702         pt->func = bond_arp_rcv;
3703         dev_add_pack(pt);
3704 }
3705
3706 void bond_unregister_arp(struct bonding *bond)
3707 {
3708         struct packet_type *pt = &bond->arp_mon_pt;
3709
3710         dev_remove_pack(pt);
3711         pt->type = 0;
3712 }
3713
3714 /*---------------------------- Hashing Policies -----------------------------*/
3715
3716 /*
3717  * Hash for the output device based upon layer 2 and layer 3 data. If
3718  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3719  */
3720 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3721 {
3722         struct ethhdr *data = (struct ethhdr *)skb->data;
3723         struct iphdr *iph = ip_hdr(skb);
3724
3725         if (skb->protocol == htons(ETH_P_IP)) {
3726                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3727                         (data->h_dest[5] ^ data->h_source[5])) % count;
3728         }
3729
3730         return (data->h_dest[5] ^ data->h_source[5]) % count;
3731 }
3732
3733 /*
3734  * Hash for the output device based upon layer 3 and layer 4 data. If
3735  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3736  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3737  */
3738 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3739 {
3740         struct ethhdr *data = (struct ethhdr *)skb->data;
3741         struct iphdr *iph = ip_hdr(skb);
3742         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3743         int layer4_xor = 0;
3744
3745         if (skb->protocol == htons(ETH_P_IP)) {
3746                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3747                     (iph->protocol == IPPROTO_TCP ||
3748                      iph->protocol == IPPROTO_UDP)) {
3749                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3750                 }
3751                 return (layer4_xor ^
3752                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3753
3754         }
3755
3756         return (data->h_dest[5] ^ data->h_source[5]) % count;
3757 }
3758
3759 /*
3760  * Hash for the output device based upon layer 2 data
3761  */
3762 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3763 {
3764         struct ethhdr *data = (struct ethhdr *)skb->data;
3765
3766         return (data->h_dest[5] ^ data->h_source[5]) % count;
3767 }
3768
3769 /*-------------------------- Device entry points ----------------------------*/
3770
3771 static int bond_open(struct net_device *bond_dev)
3772 {
3773         struct bonding *bond = netdev_priv(bond_dev);
3774
3775         bond->kill_timers = 0;
3776
3777         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3778
3779         if (bond_is_lb(bond)) {
3780                 /* bond_alb_initialize must be called before the timer
3781                  * is started.
3782                  */
3783                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3784                         /* something went wrong - fail the open operation */
3785                         return -ENOMEM;
3786                 }
3787
3788                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3789                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3790         }
3791
3792         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3793                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3794                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3795         }
3796
3797         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3798                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3799                         INIT_DELAYED_WORK(&bond->arp_work,
3800                                           bond_activebackup_arp_mon);
3801                 else
3802                         INIT_DELAYED_WORK(&bond->arp_work,
3803                                           bond_loadbalance_arp_mon);
3804
3805                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3806                 if (bond->params.arp_validate)
3807                         bond_register_arp(bond);
3808         }
3809
3810         if (bond->params.mode == BOND_MODE_8023AD) {
3811                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3812                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3813                 /* register to receive LACPDUs */
3814                 bond_register_lacpdu(bond);
3815                 bond_3ad_initiate_agg_selection(bond, 1);
3816         }
3817
3818         return 0;
3819 }
3820
3821 static int bond_close(struct net_device *bond_dev)
3822 {
3823         struct bonding *bond = netdev_priv(bond_dev);
3824
3825         if (bond->params.mode == BOND_MODE_8023AD) {
3826                 /* Unregister the receive of LACPDUs */
3827                 bond_unregister_lacpdu(bond);
3828         }
3829
3830         if (bond->params.arp_validate)
3831                 bond_unregister_arp(bond);
3832
3833         write_lock_bh(&bond->lock);
3834
3835         bond->send_grat_arp = 0;
3836         bond->send_unsol_na = 0;
3837
3838         /* signal timers not to re-arm */
3839         bond->kill_timers = 1;
3840
3841         write_unlock_bh(&bond->lock);
3842
3843         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3844                 cancel_delayed_work(&bond->mii_work);
3845         }
3846
3847         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3848                 cancel_delayed_work(&bond->arp_work);
3849         }
3850
3851         switch (bond->params.mode) {
3852         case BOND_MODE_8023AD:
3853                 cancel_delayed_work(&bond->ad_work);
3854                 break;
3855         case BOND_MODE_TLB:
3856         case BOND_MODE_ALB:
3857                 cancel_delayed_work(&bond->alb_work);
3858                 break;
3859         default:
3860                 break;
3861         }
3862
3863         if (delayed_work_pending(&bond->mcast_work))
3864                 cancel_delayed_work(&bond->mcast_work);
3865
3866         if (bond_is_lb(bond)) {
3867                 /* Must be called only after all
3868                  * slaves have been released
3869                  */
3870                 bond_alb_deinitialize(bond);
3871         }
3872
3873         return 0;
3874 }
3875
3876 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3877                                                 struct rtnl_link_stats64 *stats)
3878 {
3879         struct bonding *bond = netdev_priv(bond_dev);
3880         struct rtnl_link_stats64 temp;
3881         struct slave *slave;
3882         int i;
3883
3884         memset(stats, 0, sizeof(*stats));
3885
3886         read_lock_bh(&bond->lock);
3887
3888         bond_for_each_slave(bond, slave, i) {
3889                 const struct rtnl_link_stats64 *sstats =
3890                         dev_get_stats(slave->dev, &temp);
3891
3892                 stats->rx_packets += sstats->rx_packets;
3893                 stats->rx_bytes += sstats->rx_bytes;
3894                 stats->rx_errors += sstats->rx_errors;
3895                 stats->rx_dropped += sstats->rx_dropped;
3896
3897                 stats->tx_packets += sstats->tx_packets;
3898                 stats->tx_bytes += sstats->tx_bytes;
3899                 stats->tx_errors += sstats->tx_errors;
3900                 stats->tx_dropped += sstats->tx_dropped;
3901
3902                 stats->multicast += sstats->multicast;
3903                 stats->collisions += sstats->collisions;
3904
3905                 stats->rx_length_errors += sstats->rx_length_errors;
3906                 stats->rx_over_errors += sstats->rx_over_errors;
3907                 stats->rx_crc_errors += sstats->rx_crc_errors;
3908                 stats->rx_frame_errors += sstats->rx_frame_errors;
3909                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3910                 stats->rx_missed_errors += sstats->rx_missed_errors;
3911
3912                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3913                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3914                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3915                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3916                 stats->tx_window_errors += sstats->tx_window_errors;
3917         }
3918
3919         read_unlock_bh(&bond->lock);
3920
3921         return stats;
3922 }
3923
3924 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3925 {
3926         struct net_device *slave_dev = NULL;
3927         struct ifbond k_binfo;
3928         struct ifbond __user *u_binfo = NULL;
3929         struct ifslave k_sinfo;
3930         struct ifslave __user *u_sinfo = NULL;
3931         struct mii_ioctl_data *mii = NULL;
3932         int res = 0;
3933
3934         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3935
3936         switch (cmd) {
3937         case SIOCGMIIPHY:
3938                 mii = if_mii(ifr);
3939                 if (!mii)
3940                         return -EINVAL;
3941
3942                 mii->phy_id = 0;
3943                 /* Fall Through */
3944         case SIOCGMIIREG:
3945                 /*
3946                  * We do this again just in case we were called by SIOCGMIIREG
3947                  * instead of SIOCGMIIPHY.
3948                  */
3949                 mii = if_mii(ifr);
3950                 if (!mii)
3951                         return -EINVAL;
3952
3953
3954                 if (mii->reg_num == 1) {
3955                         struct bonding *bond = netdev_priv(bond_dev);
3956                         mii->val_out = 0;
3957                         read_lock(&bond->lock);
3958                         read_lock(&bond->curr_slave_lock);
3959                         if (netif_carrier_ok(bond->dev))
3960                                 mii->val_out = BMSR_LSTATUS;
3961
3962                         read_unlock(&bond->curr_slave_lock);
3963                         read_unlock(&bond->lock);
3964                 }
3965
3966                 return 0;
3967         case BOND_INFO_QUERY_OLD:
3968         case SIOCBONDINFOQUERY:
3969                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3970
3971                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3972                         return -EFAULT;
3973
3974                 res = bond_info_query(bond_dev, &k_binfo);
3975                 if (res == 0 &&
3976                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3977                         return -EFAULT;
3978
3979                 return res;
3980         case BOND_SLAVE_INFO_QUERY_OLD:
3981         case SIOCBONDSLAVEINFOQUERY:
3982                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3983
3984                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3985                         return -EFAULT;
3986
3987                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3988                 if (res == 0 &&
3989                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3990                         return -EFAULT;
3991
3992                 return res;
3993         default:
3994                 /* Go on */
3995                 break;
3996         }
3997
3998         if (!capable(CAP_NET_ADMIN))
3999                 return -EPERM;
4000
4001         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4002
4003         pr_debug("slave_dev=%p:\n", slave_dev);
4004
4005         if (!slave_dev)
4006                 res = -ENODEV;
4007         else {
4008                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4009                 switch (cmd) {
4010                 case BOND_ENSLAVE_OLD:
4011                 case SIOCBONDENSLAVE:
4012                         res = bond_enslave(bond_dev, slave_dev);
4013                         break;
4014                 case BOND_RELEASE_OLD:
4015                 case SIOCBONDRELEASE:
4016                         res = bond_release(bond_dev, slave_dev);
4017                         break;
4018                 case BOND_SETHWADDR_OLD:
4019                 case SIOCBONDSETHWADDR:
4020                         res = bond_sethwaddr(bond_dev, slave_dev);
4021                         break;
4022                 case BOND_CHANGE_ACTIVE_OLD:
4023                 case SIOCBONDCHANGEACTIVE:
4024                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4025                         break;
4026                 default:
4027                         res = -EOPNOTSUPP;
4028                 }
4029
4030                 dev_put(slave_dev);
4031         }
4032
4033         return res;
4034 }
4035
4036 static bool bond_addr_in_mc_list(unsigned char *addr,
4037                                  struct netdev_hw_addr_list *list,
4038                                  int addrlen)
4039 {
4040         struct netdev_hw_addr *ha;
4041
4042         netdev_hw_addr_list_for_each(ha, list)
4043                 if (!memcmp(ha->addr, addr, addrlen))
4044                         return true;
4045
4046         return false;
4047 }
4048
4049 static void bond_set_multicast_list(struct net_device *bond_dev)
4050 {
4051         struct bonding *bond = netdev_priv(bond_dev);
4052         struct netdev_hw_addr *ha;
4053         bool found;
4054
4055         /*
4056          * Do promisc before checking multicast_mode
4057          */
4058         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4059                 /*
4060                  * FIXME: Need to handle the error when one of the multi-slaves
4061                  * encounters error.
4062                  */
4063                 bond_set_promiscuity(bond, 1);
4064
4065
4066         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4067                 bond_set_promiscuity(bond, -1);
4068
4069
4070         /* set allmulti flag to slaves */
4071         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4072                 /*
4073                  * FIXME: Need to handle the error when one of the multi-slaves
4074                  * encounters error.
4075                  */
4076                 bond_set_allmulti(bond, 1);
4077
4078
4079         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4080                 bond_set_allmulti(bond, -1);
4081
4082
4083         read_lock(&bond->lock);
4084
4085         bond->flags = bond_dev->flags;
4086
4087         /* looking for addresses to add to slaves' mc list */
4088         netdev_for_each_mc_addr(ha, bond_dev) {
4089                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4090                                              bond_dev->addr_len);
4091                 if (!found)
4092                         bond_mc_add(bond, ha->addr);
4093         }
4094
4095         /* looking for addresses to delete from slaves' list */
4096         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4097                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4098                                              bond_dev->addr_len);
4099                 if (!found)
4100                         bond_mc_del(bond, ha->addr);
4101         }
4102
4103         /* save master's multicast list */
4104         __hw_addr_flush(&bond->mc_list);
4105         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4106                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4107
4108         read_unlock(&bond->lock);
4109 }
4110
4111 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4112 {
4113         struct bonding *bond = netdev_priv(dev);
4114         struct slave *slave = bond->first_slave;
4115
4116         if (slave) {
4117                 const struct net_device_ops *slave_ops
4118                         = slave->dev->netdev_ops;
4119                 if (slave_ops->ndo_neigh_setup)
4120                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4121         }
4122         return 0;
4123 }
4124
4125 /*
4126  * Change the MTU of all of a master's slaves to match the master
4127  */
4128 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4129 {
4130         struct bonding *bond = netdev_priv(bond_dev);
4131         struct slave *slave, *stop_at;
4132         int res = 0;
4133         int i;
4134
4135         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4136                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4137
4138         /* Can't hold bond->lock with bh disabled here since
4139          * some base drivers panic. On the other hand we can't
4140          * hold bond->lock without bh disabled because we'll
4141          * deadlock. The only solution is to rely on the fact
4142          * that we're under rtnl_lock here, and the slaves
4143          * list won't change. This doesn't solve the problem
4144          * of setting the slave's MTU while it is
4145          * transmitting, but the assumption is that the base
4146          * driver can handle that.
4147          *
4148          * TODO: figure out a way to safely iterate the slaves
4149          * list, but without holding a lock around the actual
4150          * call to the base driver.
4151          */
4152
4153         bond_for_each_slave(bond, slave, i) {
4154                 pr_debug("s %p s->p %p c_m %p\n",
4155                          slave,
4156                          slave->prev,
4157                          slave->dev->netdev_ops->ndo_change_mtu);
4158
4159                 res = dev_set_mtu(slave->dev, new_mtu);
4160
4161                 if (res) {
4162                         /* If we failed to set the slave's mtu to the new value
4163                          * we must abort the operation even in ACTIVE_BACKUP
4164                          * mode, because if we allow the backup slaves to have
4165                          * different mtu values than the active slave we'll
4166                          * need to change their mtu when doing a failover. That
4167                          * means changing their mtu from timer context, which
4168                          * is probably not a good idea.
4169                          */
4170                         pr_debug("err %d %s\n", res, slave->dev->name);
4171                         goto unwind;
4172                 }
4173         }
4174
4175         bond_dev->mtu = new_mtu;
4176
4177         return 0;
4178
4179 unwind:
4180         /* unwind from head to the slave that failed */
4181         stop_at = slave;
4182         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4183                 int tmp_res;
4184
4185                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4186                 if (tmp_res) {
4187                         pr_debug("unwind err %d dev %s\n",
4188                                  tmp_res, slave->dev->name);
4189                 }
4190         }
4191
4192         return res;
4193 }
4194
4195 /*
4196  * Change HW address
4197  *
4198  * Note that many devices must be down to change the HW address, and
4199  * downing the master releases all slaves.  We can make bonds full of
4200  * bonding devices to test this, however.
4201  */
4202 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4203 {
4204         struct bonding *bond = netdev_priv(bond_dev);
4205         struct sockaddr *sa = addr, tmp_sa;
4206         struct slave *slave, *stop_at;
4207         int res = 0;
4208         int i;
4209
4210         if (bond->params.mode == BOND_MODE_ALB)
4211                 return bond_alb_set_mac_address(bond_dev, addr);
4212
4213
4214         pr_debug("bond=%p, name=%s\n",
4215                  bond, bond_dev ? bond_dev->name : "None");
4216
4217         /*
4218          * If fail_over_mac is set to active, do nothing and return
4219          * success.  Returning an error causes ifenslave to fail.
4220          */
4221         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4222                 return 0;
4223
4224         if (!is_valid_ether_addr(sa->sa_data))
4225                 return -EADDRNOTAVAIL;
4226
4227         /* Can't hold bond->lock with bh disabled here since
4228          * some base drivers panic. On the other hand we can't
4229          * hold bond->lock without bh disabled because we'll
4230          * deadlock. The only solution is to rely on the fact
4231          * that we're under rtnl_lock here, and the slaves
4232          * list won't change. This doesn't solve the problem
4233          * of setting the slave's hw address while it is
4234          * transmitting, but the assumption is that the base
4235          * driver can handle that.
4236          *
4237          * TODO: figure out a way to safely iterate the slaves
4238          * list, but without holding a lock around the actual
4239          * call to the base driver.
4240          */
4241
4242         bond_for_each_slave(bond, slave, i) {
4243                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4244                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4245
4246                 if (slave_ops->ndo_set_mac_address == NULL) {
4247                         res = -EOPNOTSUPP;
4248                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4249                         goto unwind;
4250                 }
4251
4252                 res = dev_set_mac_address(slave->dev, addr);
4253                 if (res) {
4254                         /* TODO: consider downing the slave
4255                          * and retry ?
4256                          * User should expect communications
4257                          * breakage anyway until ARP finish
4258                          * updating, so...
4259                          */
4260                         pr_debug("err %d %s\n", res, slave->dev->name);
4261                         goto unwind;
4262                 }
4263         }
4264
4265         /* success */
4266         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4267         return 0;
4268
4269 unwind:
4270         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4271         tmp_sa.sa_family = bond_dev->type;
4272
4273         /* unwind from head to the slave that failed */
4274         stop_at = slave;
4275         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4276                 int tmp_res;
4277
4278                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4279                 if (tmp_res) {
4280                         pr_debug("unwind err %d dev %s\n",
4281                                  tmp_res, slave->dev->name);
4282                 }
4283         }
4284
4285         return res;
4286 }
4287
4288 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4289 {
4290         struct bonding *bond = netdev_priv(bond_dev);
4291         struct slave *slave, *start_at;
4292         int i, slave_no, res = 1;
4293         struct iphdr *iph = ip_hdr(skb);
4294
4295         read_lock(&bond->lock);
4296
4297         if (!BOND_IS_OK(bond))
4298                 goto out;
4299         /*
4300          * Start with the curr_active_slave that joined the bond as the
4301          * default for sending IGMP traffic.  For failover purposes one
4302          * needs to maintain some consistency for the interface that will
4303          * send the join/membership reports.  The curr_active_slave found
4304          * will send all of this type of traffic.
4305          */
4306         if ((iph->protocol == IPPROTO_IGMP) &&
4307             (skb->protocol == htons(ETH_P_IP))) {
4308
4309                 read_lock(&bond->curr_slave_lock);
4310                 slave = bond->curr_active_slave;
4311                 read_unlock(&bond->curr_slave_lock);
4312
4313                 if (!slave)
4314                         goto out;
4315         } else {
4316                 /*
4317                  * Concurrent TX may collide on rr_tx_counter; we accept
4318                  * that as being rare enough not to justify using an
4319                  * atomic op here.
4320                  */
4321                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4322
4323                 bond_for_each_slave(bond, slave, i) {
4324                         slave_no--;
4325                         if (slave_no < 0)
4326                                 break;
4327                 }
4328         }
4329
4330         start_at = slave;
4331         bond_for_each_slave_from(bond, slave, i, start_at) {
4332                 if (IS_UP(slave->dev) &&
4333                     (slave->link == BOND_LINK_UP) &&
4334                     (slave->state == BOND_STATE_ACTIVE)) {
4335                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4336                         break;
4337                 }
4338         }
4339
4340 out:
4341         if (res) {
4342                 /* no suitable interface, frame not sent */
4343                 dev_kfree_skb(skb);
4344         }
4345         read_unlock(&bond->lock);
4346         return NETDEV_TX_OK;
4347 }
4348
4349
4350 /*
4351  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4352  * the bond has a usable interface.
4353  */
4354 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4355 {
4356         struct bonding *bond = netdev_priv(bond_dev);
4357         int res = 1;
4358
4359         read_lock(&bond->lock);
4360         read_lock(&bond->curr_slave_lock);
4361
4362         if (!BOND_IS_OK(bond))
4363                 goto out;
4364
4365         if (!bond->curr_active_slave)
4366                 goto out;
4367
4368         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4369
4370 out:
4371         if (res)
4372                 /* no suitable interface, frame not sent */
4373                 dev_kfree_skb(skb);
4374
4375         read_unlock(&bond->curr_slave_lock);
4376         read_unlock(&bond->lock);
4377         return NETDEV_TX_OK;
4378 }
4379
4380 /*
4381  * In bond_xmit_xor() , we determine the output device by using a pre-
4382  * determined xmit_hash_policy(), If the selected device is not enabled,
4383  * find the next active slave.
4384  */
4385 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4386 {
4387         struct bonding *bond = netdev_priv(bond_dev);
4388         struct slave *slave, *start_at;
4389         int slave_no;
4390         int i;
4391         int res = 1;
4392
4393         read_lock(&bond->lock);
4394
4395         if (!BOND_IS_OK(bond))
4396                 goto out;
4397
4398         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4399
4400         bond_for_each_slave(bond, slave, i) {
4401                 slave_no--;
4402                 if (slave_no < 0)
4403                         break;
4404         }
4405
4406         start_at = slave;
4407
4408         bond_for_each_slave_from(bond, slave, i, start_at) {
4409                 if (IS_UP(slave->dev) &&
4410                     (slave->link == BOND_LINK_UP) &&
4411                     (slave->state == BOND_STATE_ACTIVE)) {
4412                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4413                         break;
4414                 }
4415         }
4416
4417 out:
4418         if (res) {
4419                 /* no suitable interface, frame not sent */
4420                 dev_kfree_skb(skb);
4421         }
4422         read_unlock(&bond->lock);
4423         return NETDEV_TX_OK;
4424 }
4425
4426 /*
4427  * in broadcast mode, we send everything to all usable interfaces.
4428  */
4429 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4430 {
4431         struct bonding *bond = netdev_priv(bond_dev);
4432         struct slave *slave, *start_at;
4433         struct net_device *tx_dev = NULL;
4434         int i;
4435         int res = 1;
4436
4437         read_lock(&bond->lock);
4438
4439         if (!BOND_IS_OK(bond))
4440                 goto out;
4441
4442         read_lock(&bond->curr_slave_lock);
4443         start_at = bond->curr_active_slave;
4444         read_unlock(&bond->curr_slave_lock);
4445
4446         if (!start_at)
4447                 goto out;
4448
4449         bond_for_each_slave_from(bond, slave, i, start_at) {
4450                 if (IS_UP(slave->dev) &&
4451                     (slave->link == BOND_LINK_UP) &&
4452                     (slave->state == BOND_STATE_ACTIVE)) {
4453                         if (tx_dev) {
4454                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4455                                 if (!skb2) {
4456                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4457                                                bond_dev->name);
4458                                         continue;
4459                                 }
4460
4461                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4462                                 if (res) {
4463                                         dev_kfree_skb(skb2);
4464                                         continue;
4465                                 }
4466                         }
4467                         tx_dev = slave->dev;
4468                 }
4469         }
4470
4471         if (tx_dev)
4472                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4473
4474 out:
4475         if (res)
4476                 /* no suitable interface, frame not sent */
4477                 dev_kfree_skb(skb);
4478
4479         /* frame sent to all suitable interfaces */
4480         read_unlock(&bond->lock);
4481         return NETDEV_TX_OK;
4482 }
4483
4484 /*------------------------- Device initialization ---------------------------*/
4485
4486 static void bond_set_xmit_hash_policy(struct bonding *bond)
4487 {
4488         switch (bond->params.xmit_policy) {
4489         case BOND_XMIT_POLICY_LAYER23:
4490                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4491                 break;
4492         case BOND_XMIT_POLICY_LAYER34:
4493                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4494                 break;
4495         case BOND_XMIT_POLICY_LAYER2:
4496         default:
4497                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4498                 break;
4499         }
4500 }
4501
4502 /*
4503  * Lookup the slave that corresponds to a qid
4504  */
4505 static inline int bond_slave_override(struct bonding *bond,
4506                                       struct sk_buff *skb)
4507 {
4508         int i, res = 1;
4509         struct slave *slave = NULL;
4510         struct slave *check_slave;
4511
4512         read_lock(&bond->lock);
4513
4514         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4515                 goto out;
4516
4517         /* Find out if any slaves have the same mapping as this skb. */
4518         bond_for_each_slave(bond, check_slave, i) {
4519                 if (check_slave->queue_id == skb->queue_mapping) {
4520                         slave = check_slave;
4521                         break;
4522                 }
4523         }
4524
4525         /* If the slave isn't UP, use default transmit policy. */
4526         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4527             (slave->link == BOND_LINK_UP)) {
4528                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4529         }
4530
4531 out:
4532         read_unlock(&bond->lock);
4533         return res;
4534 }
4535
4536 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4537 {
4538         /*
4539          * This helper function exists to help dev_pick_tx get the correct
4540          * destination queue.  Using a helper function skips the a call to
4541          * skb_tx_hash and will put the skbs in the queue we expect on their
4542          * way down to the bonding driver.
4543          */
4544         return skb->queue_mapping;
4545 }
4546
4547 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4548 {
4549         struct bonding *bond = netdev_priv(dev);
4550
4551         /*
4552          * If we risk deadlock from transmitting this in the
4553          * netpoll path, tell netpoll to queue the frame for later tx
4554          */
4555         if (is_netpoll_tx_blocked(dev))
4556                 return NETDEV_TX_BUSY;
4557
4558         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4559                 if (!bond_slave_override(bond, skb))
4560                         return NETDEV_TX_OK;
4561         }
4562
4563         switch (bond->params.mode) {
4564         case BOND_MODE_ROUNDROBIN:
4565                 return bond_xmit_roundrobin(skb, dev);
4566         case BOND_MODE_ACTIVEBACKUP:
4567                 return bond_xmit_activebackup(skb, dev);
4568         case BOND_MODE_XOR:
4569                 return bond_xmit_xor(skb, dev);
4570         case BOND_MODE_BROADCAST:
4571                 return bond_xmit_broadcast(skb, dev);
4572         case BOND_MODE_8023AD:
4573                 return bond_3ad_xmit_xor(skb, dev);
4574         case BOND_MODE_ALB:
4575         case BOND_MODE_TLB:
4576                 return bond_alb_xmit(skb, dev);
4577         default:
4578                 /* Should never happen, mode already checked */
4579                 pr_err("%s: Error: Unknown bonding mode %d\n",
4580                        dev->name, bond->params.mode);
4581                 WARN_ON_ONCE(1);
4582                 dev_kfree_skb(skb);
4583                 return NETDEV_TX_OK;
4584         }
4585 }
4586
4587
4588 /*
4589  * set bond mode specific net device operations
4590  */
4591 void bond_set_mode_ops(struct bonding *bond, int mode)
4592 {
4593         struct net_device *bond_dev = bond->dev;
4594
4595         switch (mode) {
4596         case BOND_MODE_ROUNDROBIN:
4597                 break;
4598         case BOND_MODE_ACTIVEBACKUP:
4599                 break;
4600         case BOND_MODE_XOR:
4601                 bond_set_xmit_hash_policy(bond);
4602                 break;
4603         case BOND_MODE_BROADCAST:
4604                 break;
4605         case BOND_MODE_8023AD:
4606                 bond_set_master_3ad_flags(bond);
4607                 bond_set_xmit_hash_policy(bond);
4608                 break;
4609         case BOND_MODE_ALB:
4610                 bond_set_master_alb_flags(bond);
4611                 /* FALLTHRU */
4612         case BOND_MODE_TLB:
4613                 break;
4614         default:
4615                 /* Should never happen, mode already checked */
4616                 pr_err("%s: Error: Unknown bonding mode %d\n",
4617                        bond_dev->name, mode);
4618                 break;
4619         }
4620 }
4621
4622 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4623                                     struct ethtool_drvinfo *drvinfo)
4624 {
4625         strncpy(drvinfo->driver, DRV_NAME, 32);
4626         strncpy(drvinfo->version, DRV_VERSION, 32);
4627         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4628 }
4629
4630 static const struct ethtool_ops bond_ethtool_ops = {
4631         .get_drvinfo            = bond_ethtool_get_drvinfo,
4632         .get_link               = ethtool_op_get_link,
4633         .get_tx_csum            = ethtool_op_get_tx_csum,
4634         .get_sg                 = ethtool_op_get_sg,
4635         .get_tso                = ethtool_op_get_tso,
4636         .get_ufo                = ethtool_op_get_ufo,
4637         .get_flags              = ethtool_op_get_flags,
4638 };
4639
4640 static const struct net_device_ops bond_netdev_ops = {
4641         .ndo_init               = bond_init,
4642         .ndo_uninit             = bond_uninit,
4643         .ndo_open               = bond_open,
4644         .ndo_stop               = bond_close,
4645         .ndo_start_xmit         = bond_start_xmit,
4646         .ndo_select_queue       = bond_select_queue,
4647         .ndo_get_stats64        = bond_get_stats,
4648         .ndo_do_ioctl           = bond_do_ioctl,
4649         .ndo_set_multicast_list = bond_set_multicast_list,
4650         .ndo_change_mtu         = bond_change_mtu,
4651         .ndo_set_mac_address    = bond_set_mac_address,
4652         .ndo_neigh_setup        = bond_neigh_setup,
4653         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4654         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4655         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4656 #ifdef CONFIG_NET_POLL_CONTROLLER
4657         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4658         .ndo_poll_controller    = bond_poll_controller,
4659 #endif
4660 };
4661
4662 static void bond_destructor(struct net_device *bond_dev)
4663 {
4664         struct bonding *bond = netdev_priv(bond_dev);
4665         if (bond->wq)
4666                 destroy_workqueue(bond->wq);
4667         free_netdev(bond_dev);
4668 }
4669
4670 static void bond_setup(struct net_device *bond_dev)
4671 {
4672         struct bonding *bond = netdev_priv(bond_dev);
4673
4674         /* initialize rwlocks */
4675         rwlock_init(&bond->lock);
4676         rwlock_init(&bond->curr_slave_lock);
4677
4678         bond->params = bonding_defaults;
4679
4680         /* Initialize pointers */
4681         bond->dev = bond_dev;
4682         INIT_LIST_HEAD(&bond->vlan_list);
4683
4684         /* Initialize the device entry points */
4685         ether_setup(bond_dev);
4686         bond_dev->netdev_ops = &bond_netdev_ops;
4687         bond_dev->ethtool_ops = &bond_ethtool_ops;
4688         bond_set_mode_ops(bond, bond->params.mode);
4689
4690         bond_dev->destructor = bond_destructor;
4691
4692         /* Initialize the device options */
4693         bond_dev->tx_queue_len = 0;
4694         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4695         bond_dev->priv_flags |= IFF_BONDING;
4696         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4697
4698         if (bond->params.arp_interval)
4699                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4700
4701         /* At first, we block adding VLANs. That's the only way to
4702          * prevent problems that occur when adding VLANs over an
4703          * empty bond. The block will be removed once non-challenged
4704          * slaves are enslaved.
4705          */
4706         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4707
4708         /* don't acquire bond device's netif_tx_lock when
4709          * transmitting */
4710         bond_dev->features |= NETIF_F_LLTX;
4711
4712         /* By default, we declare the bond to be fully
4713          * VLAN hardware accelerated capable. Special
4714          * care is taken in the various xmit functions
4715          * when there are slaves that are not hw accel
4716          * capable
4717          */
4718         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4719                                NETIF_F_HW_VLAN_RX |
4720                                NETIF_F_HW_VLAN_FILTER);
4721
4722         /* By default, we enable GRO on bonding devices.
4723          * Actual support requires lowlevel drivers are GRO ready.
4724          */
4725         bond_dev->features |= NETIF_F_GRO;
4726 }
4727
4728 static void bond_work_cancel_all(struct bonding *bond)
4729 {
4730         write_lock_bh(&bond->lock);
4731         bond->kill_timers = 1;
4732         write_unlock_bh(&bond->lock);
4733
4734         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4735                 cancel_delayed_work(&bond->mii_work);
4736
4737         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4738                 cancel_delayed_work(&bond->arp_work);
4739
4740         if (bond->params.mode == BOND_MODE_ALB &&
4741             delayed_work_pending(&bond->alb_work))
4742                 cancel_delayed_work(&bond->alb_work);
4743
4744         if (bond->params.mode == BOND_MODE_8023AD &&
4745             delayed_work_pending(&bond->ad_work))
4746                 cancel_delayed_work(&bond->ad_work);
4747
4748         if (delayed_work_pending(&bond->mcast_work))
4749                 cancel_delayed_work(&bond->mcast_work);
4750 }
4751
4752 /*
4753 * Destroy a bonding device.
4754 * Must be under rtnl_lock when this function is called.
4755 */
4756 static void bond_uninit(struct net_device *bond_dev)
4757 {
4758         struct bonding *bond = netdev_priv(bond_dev);
4759         struct vlan_entry *vlan, *tmp;
4760
4761         bond_netpoll_cleanup(bond_dev);
4762
4763         /* Release the bonded slaves */
4764         bond_release_all(bond_dev);
4765
4766         list_del(&bond->bond_list);
4767
4768         bond_work_cancel_all(bond);
4769
4770         bond_remove_proc_entry(bond);
4771
4772         bond_debug_unregister(bond);
4773
4774         __hw_addr_flush(&bond->mc_list);
4775
4776         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4777                 list_del(&vlan->vlan_list);
4778                 kfree(vlan);
4779         }
4780 }
4781
4782 /*------------------------- Module initialization ---------------------------*/
4783
4784 /*
4785  * Convert string input module parms.  Accept either the
4786  * number of the mode or its string name.  A bit complicated because
4787  * some mode names are substrings of other names, and calls from sysfs
4788  * may have whitespace in the name (trailing newlines, for example).
4789  */
4790 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4791 {
4792         int modeint = -1, i, rv;
4793         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4794
4795         for (p = (char *)buf; *p; p++)
4796                 if (!(isdigit(*p) || isspace(*p)))
4797                         break;
4798
4799         if (*p)
4800                 rv = sscanf(buf, "%20s", modestr);
4801         else
4802                 rv = sscanf(buf, "%d", &modeint);
4803
4804         if (!rv)
4805                 return -1;
4806
4807         for (i = 0; tbl[i].modename; i++) {
4808                 if (modeint == tbl[i].mode)
4809                         return tbl[i].mode;
4810                 if (strcmp(modestr, tbl[i].modename) == 0)
4811                         return tbl[i].mode;
4812         }
4813
4814         return -1;
4815 }
4816
4817 static int bond_check_params(struct bond_params *params)
4818 {
4819         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4820
4821         /*
4822          * Convert string parameters.
4823          */
4824         if (mode) {
4825                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4826                 if (bond_mode == -1) {
4827                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4828                                mode == NULL ? "NULL" : mode);
4829                         return -EINVAL;
4830                 }
4831         }
4832
4833         if (xmit_hash_policy) {
4834                 if ((bond_mode != BOND_MODE_XOR) &&
4835                     (bond_mode != BOND_MODE_8023AD)) {
4836                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4837                                bond_mode_name(bond_mode));
4838                 } else {
4839                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4840                                                         xmit_hashtype_tbl);
4841                         if (xmit_hashtype == -1) {
4842                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4843                                        xmit_hash_policy == NULL ? "NULL" :
4844                                        xmit_hash_policy);
4845                                 return -EINVAL;
4846                         }
4847                 }
4848         }
4849
4850         if (lacp_rate) {
4851                 if (bond_mode != BOND_MODE_8023AD) {
4852                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4853                                 bond_mode_name(bond_mode));
4854                 } else {
4855                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4856                         if (lacp_fast == -1) {
4857                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4858                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4859                                 return -EINVAL;
4860                         }
4861                 }
4862         }
4863
4864         if (ad_select) {
4865                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4866                 if (params->ad_select == -1) {
4867                         pr_err("Error: Invalid ad_select \"%s\"\n",
4868                                ad_select == NULL ? "NULL" : ad_select);
4869                         return -EINVAL;
4870                 }
4871
4872                 if (bond_mode != BOND_MODE_8023AD) {
4873                         pr_warning("ad_select param only affects 802.3ad mode\n");
4874                 }
4875         } else {
4876                 params->ad_select = BOND_AD_STABLE;
4877         }
4878
4879         if (max_bonds < 0) {
4880                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4881                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4882                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4883         }
4884
4885         if (miimon < 0) {
4886                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4887                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4888                 miimon = BOND_LINK_MON_INTERV;
4889         }
4890
4891         if (updelay < 0) {
4892                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4893                            updelay, INT_MAX);
4894                 updelay = 0;
4895         }
4896
4897         if (downdelay < 0) {
4898                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4899                            downdelay, INT_MAX);
4900                 downdelay = 0;
4901         }
4902
4903         if ((use_carrier != 0) && (use_carrier != 1)) {
4904                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4905                            use_carrier);
4906                 use_carrier = 1;
4907         }
4908
4909         if (num_grat_arp < 0 || num_grat_arp > 255) {
4910                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4911                            num_grat_arp);
4912                 num_grat_arp = 1;
4913         }
4914
4915         if (num_unsol_na < 0 || num_unsol_na > 255) {
4916                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4917                            num_unsol_na);
4918                 num_unsol_na = 1;
4919         }
4920
4921         /* reset values for 802.3ad */
4922         if (bond_mode == BOND_MODE_8023AD) {
4923                 if (!miimon) {
4924                         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");
4925                         pr_warning("Forcing miimon to 100msec\n");
4926                         miimon = 100;
4927                 }
4928         }
4929
4930         if (tx_queues < 1 || tx_queues > 255) {
4931                 pr_warning("Warning: tx_queues (%d) should be between "
4932                            "1 and 255, resetting to %d\n",
4933                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4934                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4935         }
4936
4937         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4938                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4939                            "not of valid value (0/1), so it was set to "
4940                            "0\n", all_slaves_active);
4941                 all_slaves_active = 0;
4942         }
4943
4944         if (resend_igmp < 0 || resend_igmp > 255) {
4945                 pr_warning("Warning: resend_igmp (%d) should be between "
4946                            "0 and 255, resetting to %d\n",
4947                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4948                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4949         }
4950
4951         /* reset values for TLB/ALB */
4952         if ((bond_mode == BOND_MODE_TLB) ||
4953             (bond_mode == BOND_MODE_ALB)) {
4954                 if (!miimon) {
4955                         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");
4956                         pr_warning("Forcing miimon to 100msec\n");
4957                         miimon = 100;
4958                 }
4959         }
4960
4961         if (bond_mode == BOND_MODE_ALB) {
4962                 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",
4963                           updelay);
4964         }
4965
4966         if (!miimon) {
4967                 if (updelay || downdelay) {
4968                         /* just warn the user the up/down delay will have
4969                          * no effect since miimon is zero...
4970                          */
4971                         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",
4972                                    updelay, downdelay);
4973                 }
4974         } else {
4975                 /* don't allow arp monitoring */
4976                 if (arp_interval) {
4977                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4978                                    miimon, arp_interval);
4979                         arp_interval = 0;
4980                 }
4981
4982                 if ((updelay % miimon) != 0) {
4983                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4984                                    updelay, miimon,
4985                                    (updelay / miimon) * miimon);
4986                 }
4987
4988                 updelay /= miimon;
4989
4990                 if ((downdelay % miimon) != 0) {
4991                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4992                                    downdelay, miimon,
4993                                    (downdelay / miimon) * miimon);
4994                 }
4995
4996                 downdelay /= miimon;
4997         }
4998
4999         if (arp_interval < 0) {
5000                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5001                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5002                 arp_interval = BOND_LINK_ARP_INTERV;
5003         }
5004
5005         for (arp_ip_count = 0;
5006              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5007              arp_ip_count++) {
5008                 /* not complete check, but should be good enough to
5009                    catch mistakes */
5010                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5011                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5012                                    arp_ip_target[arp_ip_count]);
5013                         arp_interval = 0;
5014                 } else {
5015                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5016                         arp_target[arp_ip_count] = ip;
5017                 }
5018         }
5019
5020         if (arp_interval && !arp_ip_count) {
5021                 /* don't allow arping if no arp_ip_target given... */
5022                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5023                            arp_interval);
5024                 arp_interval = 0;
5025         }
5026
5027         if (arp_validate) {
5028                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5029                         pr_err("arp_validate only supported in active-backup mode\n");
5030                         return -EINVAL;
5031                 }
5032                 if (!arp_interval) {
5033                         pr_err("arp_validate requires arp_interval\n");
5034                         return -EINVAL;
5035                 }
5036
5037                 arp_validate_value = bond_parse_parm(arp_validate,
5038                                                      arp_validate_tbl);
5039                 if (arp_validate_value == -1) {
5040                         pr_err("Error: invalid arp_validate \"%s\"\n",
5041                                arp_validate == NULL ? "NULL" : arp_validate);
5042                         return -EINVAL;
5043                 }
5044         } else
5045                 arp_validate_value = 0;
5046
5047         if (miimon) {
5048                 pr_info("MII link monitoring set to %d ms\n", miimon);
5049         } else if (arp_interval) {
5050                 int i;
5051
5052                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5053                         arp_interval,
5054                         arp_validate_tbl[arp_validate_value].modename,
5055                         arp_ip_count);
5056
5057                 for (i = 0; i < arp_ip_count; i++)
5058                         pr_info(" %s", arp_ip_target[i]);
5059
5060                 pr_info("\n");
5061
5062         } else if (max_bonds) {
5063                 /* miimon and arp_interval not set, we need one so things
5064                  * work as expected, see bonding.txt for details
5065                  */
5066                 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");
5067         }
5068
5069         if (primary && !USES_PRIMARY(bond_mode)) {
5070                 /* currently, using a primary only makes sense
5071                  * in active backup, TLB or ALB modes
5072                  */
5073                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5074                            primary, bond_mode_name(bond_mode));
5075                 primary = NULL;
5076         }
5077
5078         if (primary && primary_reselect) {
5079                 primary_reselect_value = bond_parse_parm(primary_reselect,
5080                                                          pri_reselect_tbl);
5081                 if (primary_reselect_value == -1) {
5082                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5083                                primary_reselect ==
5084                                         NULL ? "NULL" : primary_reselect);
5085                         return -EINVAL;
5086                 }
5087         } else {
5088                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5089         }
5090
5091         if (fail_over_mac) {
5092                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5093                                                       fail_over_mac_tbl);
5094                 if (fail_over_mac_value == -1) {
5095                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5096                                arp_validate == NULL ? "NULL" : arp_validate);
5097                         return -EINVAL;
5098                 }
5099
5100                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5101                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5102         } else {
5103                 fail_over_mac_value = BOND_FOM_NONE;
5104         }
5105
5106         /* fill params struct with the proper values */
5107         params->mode = bond_mode;
5108         params->xmit_policy = xmit_hashtype;
5109         params->miimon = miimon;
5110         params->num_grat_arp = num_grat_arp;
5111         params->num_unsol_na = num_unsol_na;
5112         params->arp_interval = arp_interval;
5113         params->arp_validate = arp_validate_value;
5114         params->updelay = updelay;
5115         params->downdelay = downdelay;
5116         params->use_carrier = use_carrier;
5117         params->lacp_fast = lacp_fast;
5118         params->primary[0] = 0;
5119         params->primary_reselect = primary_reselect_value;
5120         params->fail_over_mac = fail_over_mac_value;
5121         params->tx_queues = tx_queues;
5122         params->all_slaves_active = all_slaves_active;
5123         params->resend_igmp = resend_igmp;
5124
5125         if (primary) {
5126                 strncpy(params->primary, primary, IFNAMSIZ);
5127                 params->primary[IFNAMSIZ - 1] = 0;
5128         }
5129
5130         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5131
5132         return 0;
5133 }
5134
5135 static struct lock_class_key bonding_netdev_xmit_lock_key;
5136 static struct lock_class_key bonding_netdev_addr_lock_key;
5137
5138 static void bond_set_lockdep_class_one(struct net_device *dev,
5139                                        struct netdev_queue *txq,
5140                                        void *_unused)
5141 {
5142         lockdep_set_class(&txq->_xmit_lock,
5143                           &bonding_netdev_xmit_lock_key);
5144 }
5145
5146 static void bond_set_lockdep_class(struct net_device *dev)
5147 {
5148         lockdep_set_class(&dev->addr_list_lock,
5149                           &bonding_netdev_addr_lock_key);
5150         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5151 }
5152
5153 /*
5154  * Called from registration process
5155  */
5156 static int bond_init(struct net_device *bond_dev)
5157 {
5158         struct bonding *bond = netdev_priv(bond_dev);
5159         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5160
5161         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5162
5163         bond->wq = create_singlethread_workqueue(bond_dev->name);
5164         if (!bond->wq)
5165                 return -ENOMEM;
5166
5167         bond_set_lockdep_class(bond_dev);
5168
5169         netif_carrier_off(bond_dev);
5170
5171         bond_create_proc_entry(bond);
5172         list_add_tail(&bond->bond_list, &bn->dev_list);
5173
5174         bond_prepare_sysfs_group(bond);
5175
5176         bond_debug_register(bond);
5177
5178         __hw_addr_init(&bond->mc_list);
5179         return 0;
5180 }
5181
5182 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5183 {
5184         if (tb[IFLA_ADDRESS]) {
5185                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5186                         return -EINVAL;
5187                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5188                         return -EADDRNOTAVAIL;
5189         }
5190         return 0;
5191 }
5192
5193 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5194         .kind           = "bond",
5195         .priv_size      = sizeof(struct bonding),
5196         .setup          = bond_setup,
5197         .validate       = bond_validate,
5198 };
5199
5200 /* Create a new bond based on the specified name and bonding parameters.
5201  * If name is NULL, obtain a suitable "bond%d" name for us.
5202  * Caller must NOT hold rtnl_lock; we need to release it here before we
5203  * set up our sysfs entries.
5204  */
5205 int bond_create(struct net *net, const char *name)
5206 {
5207         struct net_device *bond_dev;
5208         int res;
5209
5210         rtnl_lock();
5211
5212         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5213                                 bond_setup, tx_queues);
5214         if (!bond_dev) {
5215                 pr_err("%s: eek! can't alloc netdev!\n", name);
5216                 rtnl_unlock();
5217                 return -ENOMEM;
5218         }
5219
5220         dev_net_set(bond_dev, net);
5221         bond_dev->rtnl_link_ops = &bond_link_ops;
5222
5223         if (!name) {
5224                 res = dev_alloc_name(bond_dev, "bond%d");
5225                 if (res < 0)
5226                         goto out;
5227         } else {
5228                 /*
5229                  * If we're given a name to register
5230                  * we need to ensure that its not already
5231                  * registered
5232                  */
5233                 res = -EEXIST;
5234                 if (__dev_get_by_name(net, name) != NULL)
5235                         goto out;
5236         }
5237
5238         res = register_netdevice(bond_dev);
5239
5240 out:
5241         rtnl_unlock();
5242         if (res < 0)
5243                 bond_destructor(bond_dev);
5244         return res;
5245 }
5246
5247 static int __net_init bond_net_init(struct net *net)
5248 {
5249         struct bond_net *bn = net_generic(net, bond_net_id);
5250
5251         bn->net = net;
5252         INIT_LIST_HEAD(&bn->dev_list);
5253
5254         bond_create_proc_dir(bn);
5255
5256         return 0;
5257 }
5258
5259 static void __net_exit bond_net_exit(struct net *net)
5260 {
5261         struct bond_net *bn = net_generic(net, bond_net_id);
5262
5263         bond_destroy_proc_dir(bn);
5264 }
5265
5266 static struct pernet_operations bond_net_ops = {
5267         .init = bond_net_init,
5268         .exit = bond_net_exit,
5269         .id   = &bond_net_id,
5270         .size = sizeof(struct bond_net),
5271 };
5272
5273 static int __init bonding_init(void)
5274 {
5275         int i;
5276         int res;
5277
5278         pr_info("%s", version);
5279
5280         res = bond_check_params(&bonding_defaults);
5281         if (res)
5282                 goto out;
5283
5284         res = register_pernet_subsys(&bond_net_ops);
5285         if (res)
5286                 goto out;
5287
5288         res = rtnl_link_register(&bond_link_ops);
5289         if (res)
5290                 goto err_link;
5291
5292         bond_create_debugfs();
5293
5294         for (i = 0; i < max_bonds; i++) {
5295                 res = bond_create(&init_net, NULL);
5296                 if (res)
5297                         goto err;
5298         }
5299
5300         res = bond_create_sysfs();
5301         if (res)
5302                 goto err;
5303
5304         register_netdevice_notifier(&bond_netdev_notifier);
5305         register_inetaddr_notifier(&bond_inetaddr_notifier);
5306         bond_register_ipv6_notifier();
5307 out:
5308         return res;
5309 err:
5310         rtnl_link_unregister(&bond_link_ops);
5311 err_link:
5312         unregister_pernet_subsys(&bond_net_ops);
5313         goto out;
5314
5315 }
5316
5317 static void __exit bonding_exit(void)
5318 {
5319         unregister_netdevice_notifier(&bond_netdev_notifier);
5320         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5321         bond_unregister_ipv6_notifier();
5322
5323         bond_destroy_sysfs();
5324         bond_destroy_debugfs();
5325
5326         rtnl_link_unregister(&bond_link_ops);
5327         unregister_pernet_subsys(&bond_net_ops);
5328
5329 #ifdef CONFIG_NET_POLL_CONTROLLER
5330         /*
5331          * Make sure we don't have an imbalance on our netpoll blocking
5332          */
5333         WARN_ON(atomic_read(&netpoll_block_tx));
5334 #endif
5335 }
5336
5337 module_init(bonding_init);
5338 module_exit(bonding_exit);
5339 MODULE_LICENSE("GPL");
5340 MODULE_VERSION(DRV_VERSION);
5341 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5342 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5343 MODULE_ALIAS_RTNL_LINK("bond");