drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <linux/preempt.h>
  79 #include <net/route.h>
  80 #include <net/net_namespace.h>
  81 #include <net/netns/generic.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89 #define BOND_LINK_MON_INTERV    0
  90 #define BOND_LINK_ARP_INTERV    0
  91
  92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  94 static int num_grat_arp = 1;
  95 static int num_unsol_na = 1;
  96 static int miimon       = BOND_LINK_MON_INTERV;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static char *ad_select;
 105 static char *xmit_hash_policy;
 106 static int arp_interval = BOND_LINK_ARP_INTERV;
 107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 108 static char *arp_validate;
 109 static char *fail_over_mac;
 110 static int all_slaves_active = 0;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param(num_grat_arp, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 120 module_param(num_unsol_na, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 149                             "(slow/fast)");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 152 module_param(xmit_hash_policy, charp, 0);
 153 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 154                                    ", 1 for layer 3+4");
 155 module_param(arp_interval, int, 0);
 156 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 157 module_param_array(arp_ip_target, charp, NULL, 0);
 158 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 159 module_param(arp_validate, charp, 0);
 160 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 161 module_param(fail_over_mac, charp, 0);
 162 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 163 module_param(all_slaves_active, int, 0);
 164 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 165                                      "by setting active flag for all slaves.  "
 166                                      "0 for never (default), 1 for always.");
 167 module_param(resend_igmp, int, 0);
 168 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 169
 170 /*----------------------------- Global variables ----------------------------*/
 171
 172 #ifdef CONFIG_NET_POLL_CONTROLLER
 173 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 174 #endif
 175
 176 static const char * const version =
 177         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 178
 179 int bond_net_id __read_mostly;
 180
 181 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 182 static int arp_ip_count;
 183 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 184 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 185 static int lacp_fast;
 186
 187 const struct bond_parm_tbl bond_lacp_tbl[] = {
 188 {       "slow",         AD_LACP_SLOW},
 189 {       "fast",         AD_LACP_FAST},
 190 {       NULL,           -1},
 191 };
 192
 193 const struct bond_parm_tbl bond_mode_tbl[] = {
 194 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 195 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 196 {       "balance-xor",          BOND_MODE_XOR},
 197 {       "broadcast",            BOND_MODE_BROADCAST},
 198 {       "802.3ad",              BOND_MODE_8023AD},
 199 {       "balance-tlb",          BOND_MODE_TLB},
 200 {       "balance-alb",          BOND_MODE_ALB},
 201 {       NULL,                   -1},
 202 };
 203
 204 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 205 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 206 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 207 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 208 {       NULL,                   -1},
 209 };
 210
 211 const struct bond_parm_tbl arp_validate_tbl[] = {
 212 {       "none",                 BOND_ARP_VALIDATE_NONE},
 213 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 214 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 215 {       "all",                  BOND_ARP_VALIDATE_ALL},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 220 {       "none",                 BOND_FOM_NONE},
 221 {       "active",               BOND_FOM_ACTIVE},
 222 {       "follow",               BOND_FOM_FOLLOW},
 223 {       NULL,                   -1},
 224 };
 225
 226 const struct bond_parm_tbl pri_reselect_tbl[] = {
 227 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 228 {       "better",               BOND_PRI_RESELECT_BETTER},
 229 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 230 {       NULL,                   -1},
 231 };
 232
 233 struct bond_parm_tbl ad_select_tbl[] = {
 234 {       "stable",       BOND_AD_STABLE},
 235 {       "bandwidth",    BOND_AD_BANDWIDTH},
 236 {       "count",        BOND_AD_COUNT},
 237 {       NULL,           -1},
 238 };
 239
 240 /*-------------------------- Forward declarations ---------------------------*/
 241
 242 static void bond_send_gratuitous_arp(struct bonding *bond);
 243 static int bond_init(struct net_device *bond_dev);
 244 static void bond_uninit(struct net_device *bond_dev);
 245
 246 /*---------------------------- General routines -----------------------------*/
 247
 248 static const char *bond_mode_name(int mode)
 249 {
 250         static const char *names[] = {
 251                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 252                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 253                 [BOND_MODE_XOR] = "load balancing (xor)",
 254                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 255                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 256                 [BOND_MODE_TLB] = "transmit load balancing",
 257                 [BOND_MODE_ALB] = "adaptive load balancing",
 258         };
 259
 260         if (mode < 0 || mode > BOND_MODE_ALB)
 261                 return "unknown";
 262
 263         return names[mode];
 264 }
 265
 266 /*---------------------------------- VLAN -----------------------------------*/
 267
 268 /**
 269  * bond_add_vlan - add a new vlan id on bond
 270  * @bond: bond that got the notification
 271  * @vlan_id: the vlan id to add
 272  *
 273  * Returns -ENOMEM if allocation failed.
 274  */
 275 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 276 {
 277         struct vlan_entry *vlan;
 278
 279         pr_debug("bond: %s, vlan id %d\n",
 280                  (bond ? bond->dev->name : "None"), vlan_id);
 281
 282         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 283         if (!vlan)
 284                 return -ENOMEM;
 285
 286         INIT_LIST_HEAD(&vlan->vlan_list);
 287         vlan->vlan_id = vlan_id;
 288
 289         write_lock_bh(&bond->lock);
 290
 291         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 292
 293         write_unlock_bh(&bond->lock);
 294
 295         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 296
 297         return 0;
 298 }
 299
 300 /**
 301  * bond_del_vlan - delete a vlan id from bond
 302  * @bond: bond that got the notification
 303  * @vlan_id: the vlan id to delete
 304  *
 305  * returns -ENODEV if @vlan_id was not found in @bond.
 306  */
 307 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 308 {
 309         struct vlan_entry *vlan;
 310         int res = -ENODEV;
 311
 312         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 313
 314         block_netpoll_tx();
 315         write_lock_bh(&bond->lock);
 316
 317         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 318                 if (vlan->vlan_id == vlan_id) {
 319                         list_del(&vlan->vlan_list);
 320
 321                         if (bond_is_lb(bond))
 322                                 bond_alb_clear_vlan(bond, vlan_id);
 323
 324                         pr_debug("removed VLAN ID %d from bond %s\n",
 325                                  vlan_id, bond->dev->name);
 326
 327                         kfree(vlan);
 328
 329                         if (list_empty(&bond->vlan_list) &&
 330                             (bond->slave_cnt == 0)) {
 331                                 /* Last VLAN removed and no slaves, so
 332                                  * restore block on adding VLANs. This will
 333                                  * be removed once new slaves that are not
 334                                  * VLAN challenged will be added.
 335                                  */
 336                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 337                         }
 338
 339                         res = 0;
 340                         goto out;
 341                 }
 342         }
 343
 344         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 345                  vlan_id, bond->dev->name);
 346
 347 out:
 348         write_unlock_bh(&bond->lock);
 349         unblock_netpoll_tx();
 350         return res;
 351 }
 352
 353 /**
 354  * bond_has_challenged_slaves
 355  * @bond: the bond we're working on
 356  *
 357  * Searches the slave list. Returns 1 if a vlan challenged slave
 358  * was found, 0 otherwise.
 359  *
 360  * Assumes bond->lock is held.
 361  */
 362 static int bond_has_challenged_slaves(struct bonding *bond)
 363 {
 364         struct slave *slave;
 365         int i;
 366
 367         bond_for_each_slave(bond, slave, i) {
 368                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 369                         pr_debug("found VLAN challenged slave - %s\n",
 370                                  slave->dev->name);
 371                         return 1;
 372                 }
 373         }
 374
 375         pr_debug("no VLAN challenged slaves found\n");
 376         return 0;
 377 }
 378
 379 /**
 380  * bond_next_vlan - safely skip to the next item in the vlans list.
 381  * @bond: the bond we're working on
 382  * @curr: item we're advancing from
 383  *
 384  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 385  * or @curr->next otherwise (even if it is @curr itself again).
 386  *
 387  * Caller must hold bond->lock
 388  */
 389 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 390 {
 391         struct vlan_entry *next, *last;
 392
 393         if (list_empty(&bond->vlan_list))
 394                 return NULL;
 395
 396         if (!curr) {
 397                 next = list_entry(bond->vlan_list.next,
 398                                   struct vlan_entry, vlan_list);
 399         } else {
 400                 last = list_entry(bond->vlan_list.prev,
 401                                   struct vlan_entry, vlan_list);
 402                 if (last == curr) {
 403                         next = list_entry(bond->vlan_list.next,
 404                                           struct vlan_entry, vlan_list);
 405                 } else {
 406                         next = list_entry(curr->vlan_list.next,
 407                                           struct vlan_entry, vlan_list);
 408                 }
 409         }
 410
 411         return next;
 412 }
 413
 414 /**
 415  * bond_dev_queue_xmit - Prepare skb for xmit.
 416  *
 417  * @bond: bond device that got this skb for tx.
 418  * @skb: hw accel VLAN tagged skb to transmit
 419  * @slave_dev: slave that is supposed to xmit this skbuff
 420  */
 421 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 422                         struct net_device *slave_dev)
 423 {
 424         skb->dev = slave_dev;
 425         skb->priority = 1;
 426         if (unlikely(netpoll_tx_running(slave_dev)))
 427                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 428         else
 429                 dev_queue_xmit(skb);
 430
 431         return 0;
 432 }
 433
 434 /*
 435  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 436  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 437  * lock because:
 438  * a. This operation is performed in IOCTL context,
 439  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 440  * c. Holding a lock with BH disabled while directly calling a base driver
 441  *    entry point is generally a BAD idea.
 442  *
 443  * The design of synchronization/protection for this operation in the 8021q
 444  * module is good for one or more VLAN devices over a single physical device
 445  * and cannot be extended for a teaming solution like bonding, so there is a
 446  * potential race condition here where a net device from the vlan group might
 447  * be referenced (either by a base driver or the 8021q code) while it is being
 448  * removed from the system. However, it turns out we're not making matters
 449  * worse, and if it works for regular VLAN usage it will work here too.
 450 */
 451
 452 /**
 453  * bond_vlan_rx_register - Propagates registration to slaves
 454  * @bond_dev: bonding net device that got called
 455  * @grp: vlan group being registered
 456  */
 457 static void bond_vlan_rx_register(struct net_device *bond_dev,
 458                                   struct vlan_group *grp)
 459 {
 460         struct bonding *bond = netdev_priv(bond_dev);
 461         struct slave *slave;
 462         int i;
 463
 464         write_lock_bh(&bond->lock);
 465         bond->vlgrp = grp;
 466         write_unlock_bh(&bond->lock);
 467
 468         bond_for_each_slave(bond, slave, i) {
 469                 struct net_device *slave_dev = slave->dev;
 470                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 471
 472                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 473                     slave_ops->ndo_vlan_rx_register) {
 474                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 475                 }
 476         }
 477 }
 478
 479 /**
 480  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 481  * @bond_dev: bonding net device that got called
 482  * @vid: vlan id being added
 483  */
 484 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 485 {
 486         struct bonding *bond = netdev_priv(bond_dev);
 487         struct slave *slave;
 488         int i, res;
 489
 490         bond_for_each_slave(bond, slave, i) {
 491                 struct net_device *slave_dev = slave->dev;
 492                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 493
 494                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 495                     slave_ops->ndo_vlan_rx_add_vid) {
 496                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 497                 }
 498         }
 499
 500         res = bond_add_vlan(bond, vid);
 501         if (res) {
 502                 pr_err("%s: Error: Failed to add vlan id %d\n",
 503                        bond_dev->name, vid);
 504         }
 505 }
 506
 507 /**
 508  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 509  * @bond_dev: bonding net device that got called
 510  * @vid: vlan id being removed
 511  */
 512 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 513 {
 514         struct bonding *bond = netdev_priv(bond_dev);
 515         struct slave *slave;
 516         struct net_device *vlan_dev;
 517         int i, res;
 518
 519         bond_for_each_slave(bond, slave, i) {
 520                 struct net_device *slave_dev = slave->dev;
 521                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 522
 523                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 524                     slave_ops->ndo_vlan_rx_kill_vid) {
 525                         /* Save and then restore vlan_dev in the grp array,
 526                          * since the slave's driver might clear it.
 527                          */
 528                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 529                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 530                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 531                 }
 532         }
 533
 534         res = bond_del_vlan(bond, vid);
 535         if (res) {
 536                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 537                        bond_dev->name, vid);
 538         }
 539 }
 540
 541 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 542 {
 543         struct vlan_entry *vlan;
 544         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 545
 546         if (!bond->vlgrp)
 547                 return;
 548
 549         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 550             slave_ops->ndo_vlan_rx_register)
 551                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 552
 553         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 554             !(slave_ops->ndo_vlan_rx_add_vid))
 555                 return;
 556
 557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 558                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 559 }
 560
 561 static void bond_del_vlans_from_slave(struct bonding *bond,
 562                                       struct net_device *slave_dev)
 563 {
 564         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 565         struct vlan_entry *vlan;
 566         struct net_device *vlan_dev;
 567
 568         if (!bond->vlgrp)
 569                 return;
 570
 571         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 572             !(slave_ops->ndo_vlan_rx_kill_vid))
 573                 goto unreg;
 574
 575         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 576                 if (!vlan->vlan_id)
 577                         continue;
 578                 /* Save and then restore vlan_dev in the grp array,
 579                  * since the slave's driver might clear it.
 580                  */
 581                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 582                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 583                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 584         }
 585
 586 unreg:
 587         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 588             slave_ops->ndo_vlan_rx_register)
 589                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 590 }
 591
 592 /*------------------------------- Link status -------------------------------*/
 593
 594 /*
 595  * Set the carrier state for the master according to the state of its
 596  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 597  * do special 802.3ad magic.
 598  *
 599  * Returns zero if carrier state does not change, nonzero if it does.
 600  */
 601 static int bond_set_carrier(struct bonding *bond)
 602 {
 603         struct slave *slave;
 604         int i;
 605
 606         if (bond->slave_cnt == 0)
 607                 goto down;
 608
 609         if (bond->params.mode == BOND_MODE_8023AD)
 610                 return bond_3ad_set_carrier(bond);
 611
 612         bond_for_each_slave(bond, slave, i) {
 613                 if (slave->link == BOND_LINK_UP) {
 614                         if (!netif_carrier_ok(bond->dev)) {
 615                                 netif_carrier_on(bond->dev);
 616                                 return 1;
 617                         }
 618                         return 0;
 619                 }
 620         }
 621
 622 down:
 623         if (netif_carrier_ok(bond->dev)) {
 624                 netif_carrier_off(bond->dev);
 625                 return 1;
 626         }
 627         return 0;
 628 }
 629
 630 /*
 631  * Get link speed and duplex from the slave's base driver
 632  * using ethtool. If for some reason the call fails or the
 633  * values are invalid, fake speed and duplex to 100/Full
 634  * and return error.
 635  */
 636 static int bond_update_speed_duplex(struct slave *slave)
 637 {
 638         struct net_device *slave_dev = slave->dev;
 639         struct ethtool_cmd etool;
 640         int res;
 641
 642         /* Fake speed and duplex */
 643         slave->speed = SPEED_100;
 644         slave->duplex = DUPLEX_FULL;
 645
 646         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 647                 return -1;
 648
 649         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 650         if (res < 0)
 651                 return -1;
 652
 653         switch (etool.speed) {
 654         case SPEED_10:
 655         case SPEED_100:
 656         case SPEED_1000:
 657         case SPEED_10000:
 658                 break;
 659         default:
 660                 return -1;
 661         }
 662
 663         switch (etool.duplex) {
 664         case DUPLEX_FULL:
 665         case DUPLEX_HALF:
 666                 break;
 667         default:
 668                 return -1;
 669         }
 670
 671         slave->speed = etool.speed;
 672         slave->duplex = etool.duplex;
 673
 674         return 0;
 675 }
 676
 677 /*
 678  * if <dev> supports MII link status reporting, check its link status.
 679  *
 680  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 681  * depending upon the setting of the use_carrier parameter.
 682  *
 683  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 684  * can't tell and just pretend it is), or 0, meaning that the link is
 685  * down.
 686  *
 687  * If reporting is non-zero, instead of faking link up, return -1 if
 688  * both ETHTOOL and MII ioctls fail (meaning the device does not
 689  * support them).  If use_carrier is set, return whatever it says.
 690  * It'd be nice if there was a good way to tell if a driver supports
 691  * netif_carrier, but there really isn't.
 692  */
 693 static int bond_check_dev_link(struct bonding *bond,
 694                                struct net_device *slave_dev, int reporting)
 695 {
 696         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 697         int (*ioctl)(struct net_device *, struct ifreq *, int);
 698         struct ifreq ifr;
 699         struct mii_ioctl_data *mii;
 700
 701         if (!reporting && !netif_running(slave_dev))
 702                 return 0;
 703
 704         if (bond->params.use_carrier)
 705                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 706
 707         /* Try to get link status using Ethtool first. */
 708         if (slave_dev->ethtool_ops) {
 709                 if (slave_dev->ethtool_ops->get_link) {
 710                         u32 link;
 711
 712                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 713
 714                         return link ? BMSR_LSTATUS : 0;
 715                 }
 716         }
 717
 718         /* Ethtool can't be used, fallback to MII ioctls. */
 719         ioctl = slave_ops->ndo_do_ioctl;
 720         if (ioctl) {
 721                 /* TODO: set pointer to correct ioctl on a per team member */
 722                 /*       bases to make this more efficient. that is, once  */
 723                 /*       we determine the correct ioctl, we will always    */
 724                 /*       call it and not the others for that team          */
 725                 /*       member.                                           */
 726
 727                 /*
 728                  * We cannot assume that SIOCGMIIPHY will also read a
 729                  * register; not all network drivers (e.g., e100)
 730                  * support that.
 731                  */
 732
 733                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 734                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 735                 mii = if_mii(&ifr);
 736                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 737                         mii->reg_num = MII_BMSR;
 738                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 739                                 return mii->val_out & BMSR_LSTATUS;
 740                 }
 741         }
 742
 743         /*
 744          * If reporting, report that either there's no dev->do_ioctl,
 745          * or both SIOCGMIIREG and get_link failed (meaning that we
 746          * cannot report link status).  If not reporting, pretend
 747          * we're ok.
 748          */
 749         return reporting ? -1 : BMSR_LSTATUS;
 750 }
 751
 752 /*----------------------------- Multicast list ------------------------------*/
 753
 754 /*
 755  * Push the promiscuity flag down to appropriate slaves
 756  */
 757 static int bond_set_promiscuity(struct bonding *bond, int inc)
 758 {
 759         int err = 0;
 760         if (USES_PRIMARY(bond->params.mode)) {
 761                 /* write lock already acquired */
 762                 if (bond->curr_active_slave) {
 763                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 764                                                   inc);
 765                 }
 766         } else {
 767                 struct slave *slave;
 768                 int i;
 769                 bond_for_each_slave(bond, slave, i) {
 770                         err = dev_set_promiscuity(slave->dev, inc);
 771                         if (err)
 772                                 return err;
 773                 }
 774         }
 775         return err;
 776 }
 777
 778 /*
 779  * Push the allmulti flag down to all slaves
 780  */
 781 static int bond_set_allmulti(struct bonding *bond, int inc)
 782 {
 783         int err = 0;
 784         if (USES_PRIMARY(bond->params.mode)) {
 785                 /* write lock already acquired */
 786                 if (bond->curr_active_slave) {
 787                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 788                                                inc);
 789                 }
 790         } else {
 791                 struct slave *slave;
 792                 int i;
 793                 bond_for_each_slave(bond, slave, i) {
 794                         err = dev_set_allmulti(slave->dev, inc);
 795                         if (err)
 796                                 return err;
 797                 }
 798         }
 799         return err;
 800 }
 801
 802 /*
 803  * Add a Multicast address to slaves
 804  * according to mode
 805  */
 806 static void bond_mc_add(struct bonding *bond, void *addr)
 807 {
 808         if (USES_PRIMARY(bond->params.mode)) {
 809                 /* write lock already acquired */
 810                 if (bond->curr_active_slave)
 811                         dev_mc_add(bond->curr_active_slave->dev, addr);
 812         } else {
 813                 struct slave *slave;
 814                 int i;
 815
 816                 bond_for_each_slave(bond, slave, i)
 817                         dev_mc_add(slave->dev, addr);
 818         }
 819 }
 820
 821 /*
 822  * Remove a multicast address from slave
 823  * according to mode
 824  */
 825 static void bond_mc_del(struct bonding *bond, void *addr)
 826 {
 827         if (USES_PRIMARY(bond->params.mode)) {
 828                 /* write lock already acquired */
 829                 if (bond->curr_active_slave)
 830                         dev_mc_del(bond->curr_active_slave->dev, addr);
 831         } else {
 832                 struct slave *slave;
 833                 int i;
 834                 bond_for_each_slave(bond, slave, i) {
 835                         dev_mc_del(slave->dev, addr);
 836                 }
 837         }
 838 }
 839
 840
 841 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 842 {
 843         struct in_device *in_dev;
 844
 845         rcu_read_lock();
 846         in_dev = __in_dev_get_rcu(dev);
 847         if (in_dev)
 848                 ip_mc_rejoin_groups(in_dev);
 849         rcu_read_unlock();
 850 }
 851
 852 /*
 853  * Retrieve the list of registered multicast addresses for the bonding
 854  * device and retransmit an IGMP JOIN request to the current active
 855  * slave.
 856  */
 857 static void bond_resend_igmp_join_requests(struct bonding *bond)
 858 {
 859         struct net_device *vlan_dev;
 860         struct vlan_entry *vlan;
 861
 862         read_lock(&bond->lock);
 863
 864         /* rejoin all groups on bond device */
 865         __bond_resend_igmp_join_requests(bond->dev);
 866
 867         /* rejoin all groups on vlan devices */
 868         if (bond->vlgrp) {
 869                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 870                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 871                                                          vlan->vlan_id);
 872                         if (vlan_dev)
 873                                 __bond_resend_igmp_join_requests(vlan_dev);
 874                 }
 875         }
 876
 877         if (--bond->igmp_retrans > 0)
 878                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 879
 880         read_unlock(&bond->lock);
 881 }
 882
 883 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 884 {
 885         struct bonding *bond = container_of(work, struct bonding,
 886                                                         mcast_work.work);
 887         bond_resend_igmp_join_requests(bond);
 888 }
 889
 890 /*
 891  * flush all members of flush->mc_list from device dev->mc_list
 892  */
 893 static void bond_mc_list_flush(struct net_device *bond_dev,
 894                                struct net_device *slave_dev)
 895 {
 896         struct bonding *bond = netdev_priv(bond_dev);
 897         struct netdev_hw_addr *ha;
 898
 899         netdev_for_each_mc_addr(ha, bond_dev)
 900                 dev_mc_del(slave_dev, ha->addr);
 901
 902         if (bond->params.mode == BOND_MODE_8023AD) {
 903                 /* del lacpdu mc addr from mc list */
 904                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 905
 906                 dev_mc_del(slave_dev, lacpdu_multicast);
 907         }
 908 }
 909
 910 /*--------------------------- Active slave change ---------------------------*/
 911
 912 /*
 913  * Update the mc list and multicast-related flags for the new and
 914  * old active slaves (if any) according to the multicast mode, and
 915  * promiscuous flags unconditionally.
 916  */
 917 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 918                          struct slave *old_active)
 919 {
 920         struct netdev_hw_addr *ha;
 921
 922         if (!USES_PRIMARY(bond->params.mode))
 923                 /* nothing to do -  mc list is already up-to-date on
 924                  * all slaves
 925                  */
 926                 return;
 927
 928         if (old_active) {
 929                 if (bond->dev->flags & IFF_PROMISC)
 930                         dev_set_promiscuity(old_active->dev, -1);
 931
 932                 if (bond->dev->flags & IFF_ALLMULTI)
 933                         dev_set_allmulti(old_active->dev, -1);
 934
 935                 netdev_for_each_mc_addr(ha, bond->dev)
 936                         dev_mc_del(old_active->dev, ha->addr);
 937         }
 938
 939         if (new_active) {
 940                 /* FIXME: Signal errors upstream. */
 941                 if (bond->dev->flags & IFF_PROMISC)
 942                         dev_set_promiscuity(new_active->dev, 1);
 943
 944                 if (bond->dev->flags & IFF_ALLMULTI)
 945                         dev_set_allmulti(new_active->dev, 1);
 946
 947                 netdev_for_each_mc_addr(ha, bond->dev)
 948                         dev_mc_add(new_active->dev, ha->addr);
 949         }
 950 }
 951
 952 /*
 953  * bond_do_fail_over_mac
 954  *
 955  * Perform special MAC address swapping for fail_over_mac settings
 956  *
 957  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 958  */
 959 static void bond_do_fail_over_mac(struct bonding *bond,
 960                                   struct slave *new_active,
 961                                   struct slave *old_active)
 962         __releases(&bond->curr_slave_lock)
 963         __releases(&bond->lock)
 964         __acquires(&bond->lock)
 965         __acquires(&bond->curr_slave_lock)
 966 {
 967         u8 tmp_mac[ETH_ALEN];
 968         struct sockaddr saddr;
 969         int rv;
 970
 971         switch (bond->params.fail_over_mac) {
 972         case BOND_FOM_ACTIVE:
 973                 if (new_active)
 974                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 975                                new_active->dev->addr_len);
 976                 break;
 977         case BOND_FOM_FOLLOW:
 978                 /*
 979                  * if new_active && old_active, swap them
 980                  * if just old_active, do nothing (going to no active slave)
 981                  * if just new_active, set new_active to bond's MAC
 982                  */
 983                 if (!new_active)
 984                         return;
 985
 986                 write_unlock_bh(&bond->curr_slave_lock);
 987                 read_unlock(&bond->lock);
 988
 989                 if (old_active) {
 990                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 991                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 992                                ETH_ALEN);
 993                         saddr.sa_family = new_active->dev->type;
 994                 } else {
 995                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 996                         saddr.sa_family = bond->dev->type;
 997                 }
 998
 999                 rv = dev_set_mac_address(new_active->dev, &saddr);
1000                 if (rv) {
1001                         pr_err("%s: Error %d setting MAC of slave %s\n",
1002                                bond->dev->name, -rv, new_active->dev->name);
1003                         goto out;
1004                 }
1005
1006                 if (!old_active)
1007                         goto out;
1008
1009                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1010                 saddr.sa_family = old_active->dev->type;
1011
1012                 rv = dev_set_mac_address(old_active->dev, &saddr);
1013                 if (rv)
1014                         pr_err("%s: Error %d setting MAC of slave %s\n",
1015                                bond->dev->name, -rv, new_active->dev->name);
1016 out:
1017                 read_lock(&bond->lock);
1018                 write_lock_bh(&bond->curr_slave_lock);
1019                 break;
1020         default:
1021                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1022                        bond->dev->name, bond->params.fail_over_mac);
1023                 break;
1024         }
1025
1026 }
1027
1028 static bool bond_should_change_active(struct bonding *bond)
1029 {
1030         struct slave *prim = bond->primary_slave;
1031         struct slave *curr = bond->curr_active_slave;
1032
1033         if (!prim || !curr || curr->link != BOND_LINK_UP)
1034                 return true;
1035         if (bond->force_primary) {
1036                 bond->force_primary = false;
1037                 return true;
1038         }
1039         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1040             (prim->speed < curr->speed ||
1041              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1042                 return false;
1043         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1044                 return false;
1045         return true;
1046 }
1047
1048 /**
1049  * find_best_interface - select the best available slave to be the active one
1050  * @bond: our bonding struct
1051  *
1052  * Warning: Caller must hold curr_slave_lock for writing.
1053  */
1054 static struct slave *bond_find_best_slave(struct bonding *bond)
1055 {
1056         struct slave *new_active, *old_active;
1057         struct slave *bestslave = NULL;
1058         int mintime = bond->params.updelay;
1059         int i;
1060
1061         new_active = bond->curr_active_slave;
1062
1063         if (!new_active) { /* there were no active slaves left */
1064                 if (bond->slave_cnt > 0)   /* found one slave */
1065                         new_active = bond->first_slave;
1066                 else
1067                         return NULL; /* still no slave, return NULL */
1068         }
1069
1070         if ((bond->primary_slave) &&
1071             bond->primary_slave->link == BOND_LINK_UP &&
1072             bond_should_change_active(bond)) {
1073                 new_active = bond->primary_slave;
1074         }
1075
1076         /* remember where to stop iterating over the slaves */
1077         old_active = new_active;
1078
1079         bond_for_each_slave_from(bond, new_active, i, old_active) {
1080                 if (new_active->link == BOND_LINK_UP) {
1081                         return new_active;
1082                 } else if (new_active->link == BOND_LINK_BACK &&
1083                            IS_UP(new_active->dev)) {
1084                         /* link up, but waiting for stabilization */
1085                         if (new_active->delay < mintime) {
1086                                 mintime = new_active->delay;
1087                                 bestslave = new_active;
1088                         }
1089                 }
1090         }
1091
1092         return bestslave;
1093 }
1094
1095 /**
1096  * change_active_interface - change the active slave into the specified one
1097  * @bond: our bonding struct
1098  * @new: the new slave to make the active one
1099  *
1100  * Set the new slave to the bond's settings and unset them on the old
1101  * curr_active_slave.
1102  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1103  *
1104  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1105  * because it is apparently the best available slave we have, even though its
1106  * updelay hasn't timed out yet.
1107  *
1108  * If new_active is not NULL, caller must hold bond->lock for read and
1109  * curr_slave_lock for write_bh.
1110  */
1111 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1112 {
1113         struct slave *old_active = bond->curr_active_slave;
1114
1115         if (old_active == new_active)
1116                 return;
1117
1118         if (new_active) {
1119                 new_active->jiffies = jiffies;
1120
1121                 if (new_active->link == BOND_LINK_BACK) {
1122                         if (USES_PRIMARY(bond->params.mode)) {
1123                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1124                                         bond->dev->name, new_active->dev->name,
1125                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1126                         }
1127
1128                         new_active->delay = 0;
1129                         new_active->link = BOND_LINK_UP;
1130
1131                         if (bond->params.mode == BOND_MODE_8023AD)
1132                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1133
1134                         if (bond_is_lb(bond))
1135                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1136                 } else {
1137                         if (USES_PRIMARY(bond->params.mode)) {
1138                                 pr_info("%s: making interface %s the new active one.\n",
1139                                         bond->dev->name, new_active->dev->name);
1140                         }
1141                 }
1142         }
1143
1144         if (USES_PRIMARY(bond->params.mode))
1145                 bond_mc_swap(bond, new_active, old_active);
1146
1147         if (bond_is_lb(bond)) {
1148                 bond_alb_handle_active_change(bond, new_active);
1149                 if (old_active)
1150                         bond_set_slave_inactive_flags(old_active);
1151                 if (new_active)
1152                         bond_set_slave_active_flags(new_active);
1153         } else {
1154                 bond->curr_active_slave = new_active;
1155         }
1156
1157         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1158                 if (old_active)
1159                         bond_set_slave_inactive_flags(old_active);
1160
1161                 if (new_active) {
1162                         bond_set_slave_active_flags(new_active);
1163
1164                         if (bond->params.fail_over_mac)
1165                                 bond_do_fail_over_mac(bond, new_active,
1166                                                       old_active);
1167
1168                         if (netif_running(bond->dev)) {
1169                                 bond->send_grat_arp = bond->params.num_grat_arp;
1170                                 bond_send_gratuitous_arp(bond);
1171
1172                                 bond->send_unsol_na = bond->params.num_unsol_na;
1173                                 bond_send_unsolicited_na(bond);
1174                         }
1175
1176                         write_unlock_bh(&bond->curr_slave_lock);
1177                         read_unlock(&bond->lock);
1178
1179                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1180
1181                         read_lock(&bond->lock);
1182                         write_lock_bh(&bond->curr_slave_lock);
1183                 }
1184         }
1185
1186         /* resend IGMP joins since active slave has changed or
1187          * all were sent on curr_active_slave */
1188         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1189              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1190             netif_running(bond->dev)) {
1191                 bond->igmp_retrans = bond->params.resend_igmp;
1192                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1193         }
1194 }
1195
1196 /**
1197  * bond_select_active_slave - select a new active slave, if needed
1198  * @bond: our bonding struct
1199  *
1200  * This functions should be called when one of the following occurs:
1201  * - The old curr_active_slave has been released or lost its link.
1202  * - The primary_slave has got its link back.
1203  * - A slave has got its link back and there's no old curr_active_slave.
1204  *
1205  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1206  */
1207 void bond_select_active_slave(struct bonding *bond)
1208 {
1209         struct slave *best_slave;
1210         int rv;
1211
1212         best_slave = bond_find_best_slave(bond);
1213         if (best_slave != bond->curr_active_slave) {
1214                 bond_change_active_slave(bond, best_slave);
1215                 rv = bond_set_carrier(bond);
1216                 if (!rv)
1217                         return;
1218
1219                 if (netif_carrier_ok(bond->dev)) {
1220                         pr_info("%s: first active interface up!\n",
1221                                 bond->dev->name);
1222                 } else {
1223                         pr_info("%s: now running without any active interface !\n",
1224                                 bond->dev->name);
1225                 }
1226         }
1227 }
1228
1229 /*--------------------------- slave list handling ---------------------------*/
1230
1231 /*
1232  * This function attaches the slave to the end of list.
1233  *
1234  * bond->lock held for writing by caller.
1235  */
1236 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1237 {
1238         if (bond->first_slave == NULL) { /* attaching the first slave */
1239                 new_slave->next = new_slave;
1240                 new_slave->prev = new_slave;
1241                 bond->first_slave = new_slave;
1242         } else {
1243                 new_slave->next = bond->first_slave;
1244                 new_slave->prev = bond->first_slave->prev;
1245                 new_slave->next->prev = new_slave;
1246                 new_slave->prev->next = new_slave;
1247         }
1248
1249         bond->slave_cnt++;
1250 }
1251
1252 /*
1253  * This function detaches the slave from the list.
1254  * WARNING: no check is made to verify if the slave effectively
1255  * belongs to <bond>.
1256  * Nothing is freed on return, structures are just unchained.
1257  * If any slave pointer in bond was pointing to <slave>,
1258  * it should be changed by the calling function.
1259  *
1260  * bond->lock held for writing by caller.
1261  */
1262 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1263 {
1264         if (slave->next)
1265                 slave->next->prev = slave->prev;
1266
1267         if (slave->prev)
1268                 slave->prev->next = slave->next;
1269
1270         if (bond->first_slave == slave) { /* slave is the first slave */
1271                 if (bond->slave_cnt > 1) { /* there are more slave */
1272                         bond->first_slave = slave->next;
1273                 } else {
1274                         bond->first_slave = NULL; /* slave was the last one */
1275                 }
1276         }
1277
1278         slave->next = NULL;
1279         slave->prev = NULL;
1280         bond->slave_cnt--;
1281 }
1282
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 static inline int slave_enable_netpoll(struct slave *slave)
1285 {
1286         struct netpoll *np;
1287         int err = 0;
1288
1289         np = kzalloc(sizeof(*np), GFP_KERNEL);
1290         err = -ENOMEM;
1291         if (!np)
1292                 goto out;
1293
1294         np->dev = slave->dev;
1295         err = __netpoll_setup(np);
1296         if (err) {
1297                 kfree(np);
1298                 goto out;
1299         }
1300         slave->np = np;
1301 out:
1302         return err;
1303 }
1304 static inline void slave_disable_netpoll(struct slave *slave)
1305 {
1306         struct netpoll *np = slave->np;
1307
1308         if (!np)
1309                 return;
1310
1311         slave->np = NULL;
1312         synchronize_rcu_bh();
1313         __netpoll_cleanup(np);
1314         kfree(np);
1315 }
1316 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1317 {
1318         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1319                 return false;
1320         if (!slave_dev->netdev_ops->ndo_poll_controller)
1321                 return false;
1322         return true;
1323 }
1324
1325 static void bond_poll_controller(struct net_device *bond_dev)
1326 {
1327 }
1328
1329 static void __bond_netpoll_cleanup(struct bonding *bond)
1330 {
1331         struct slave *slave;
1332         int i;
1333
1334         bond_for_each_slave(bond, slave, i)
1335                 if (IS_UP(slave->dev))
1336                         slave_disable_netpoll(slave);
1337 }
1338 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1339 {
1340         struct bonding *bond = netdev_priv(bond_dev);
1341
1342         read_lock(&bond->lock);
1343         __bond_netpoll_cleanup(bond);
1344         read_unlock(&bond->lock);
1345 }
1346
1347 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1348 {
1349         struct bonding *bond = netdev_priv(dev);
1350         struct slave *slave;
1351         int i, err = 0;
1352
1353         read_lock(&bond->lock);
1354         bond_for_each_slave(bond, slave, i) {
1355                 if (!IS_UP(slave->dev))
1356                         continue;
1357                 err = slave_enable_netpoll(slave);
1358                 if (err) {
1359                         __bond_netpoll_cleanup(bond);
1360                         break;
1361                 }
1362         }
1363         read_unlock(&bond->lock);
1364         return err;
1365 }
1366
1367 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1368 {
1369         return bond->dev->npinfo;
1370 }
1371
1372 #else
1373 static inline int slave_enable_netpoll(struct slave *slave)
1374 {
1375         return 0;
1376 }
1377 static inline void slave_disable_netpoll(struct slave *slave)
1378 {
1379 }
1380 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1381 {
1382 }
1383 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1384 {
1385         return 0;
1386 }
1387 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1388 {
1389         return NULL;
1390 }
1391 #endif
1392
1393 /*---------------------------------- IOCTL ----------------------------------*/
1394
1395 static int bond_sethwaddr(struct net_device *bond_dev,
1396                           struct net_device *slave_dev)
1397 {
1398         pr_debug("bond_dev=%p\n", bond_dev);
1399         pr_debug("slave_dev=%p\n", slave_dev);
1400         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1401         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1402         return 0;
1403 }
1404
1405 #define BOND_VLAN_FEATURES \
1406         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1407          NETIF_F_HW_VLAN_FILTER)
1408
1409 /*
1410  * Compute the common dev->feature set available to all slaves.  Some
1411  * feature bits are managed elsewhere, so preserve those feature bits
1412  * on the master device.
1413  */
1414 static int bond_compute_features(struct bonding *bond)
1415 {
1416         struct slave *slave;
1417         struct net_device *bond_dev = bond->dev;
1418         u32 features = bond_dev->features;
1419         u32 vlan_features = 0;
1420         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1421                                                 bond_dev->hard_header_len);
1422         int i;
1423
1424         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1425         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1426
1427         if (!bond->first_slave)
1428                 goto done;
1429
1430         features &= ~NETIF_F_ONE_FOR_ALL;
1431
1432         vlan_features = bond->first_slave->dev->vlan_features;
1433         bond_for_each_slave(bond, slave, i) {
1434                 features = netdev_increment_features(features,
1435                                                      slave->dev->features,
1436                                                      NETIF_F_ONE_FOR_ALL);
1437                 vlan_features = netdev_increment_features(vlan_features,
1438                                                         slave->dev->vlan_features,
1439                                                         NETIF_F_ONE_FOR_ALL);
1440                 if (slave->dev->hard_header_len > max_hard_header_len)
1441                         max_hard_header_len = slave->dev->hard_header_len;
1442         }
1443
1444 done:
1445         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1446         bond_dev->features = netdev_fix_features(bond_dev, features);
1447         bond_dev->vlan_features = netdev_fix_features(bond_dev, vlan_features);
1448         bond_dev->hard_header_len = max_hard_header_len;
1449
1450         return 0;
1451 }
1452
1453 static void bond_setup_by_slave(struct net_device *bond_dev,
1454                                 struct net_device *slave_dev)
1455 {
1456         struct bonding *bond = netdev_priv(bond_dev);
1457
1458         bond_dev->header_ops        = slave_dev->header_ops;
1459
1460         bond_dev->type              = slave_dev->type;
1461         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1462         bond_dev->addr_len          = slave_dev->addr_len;
1463
1464         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1465                 slave_dev->addr_len);
1466         bond->setup_by_slave = 1;
1467 }
1468
1469 /* On bonding slaves other than the currently active slave, suppress
1470  * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1471  * ARP on active-backup slaves with arp_validate enabled.
1472  */
1473 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1474                                             struct net_device *slave_dev,
1475                                             struct net_device *bond_dev)
1476 {
1477         if (slave_dev->priv_flags & IFF_SLAVE_INACTIVE) {
1478                 if (slave_dev->priv_flags & IFF_SLAVE_NEEDARP &&
1479                     skb->protocol == __cpu_to_be16(ETH_P_ARP))
1480                         return false;
1481
1482                 if (bond_dev->priv_flags & IFF_MASTER_ALB &&
1483                     skb->pkt_type != PACKET_BROADCAST &&
1484                     skb->pkt_type != PACKET_MULTICAST)
1485                                 return false;
1486
1487                 if (bond_dev->priv_flags & IFF_MASTER_8023AD &&
1488                     skb->protocol == __cpu_to_be16(ETH_P_SLOW))
1489                         return false;
1490
1491                 return true;
1492         }
1493         return false;
1494 }
1495
1496 static struct sk_buff *bond_handle_frame(struct sk_buff *skb)
1497 {
1498         struct net_device *slave_dev;
1499         struct net_device *bond_dev;
1500
1501         skb = skb_share_check(skb, GFP_ATOMIC);
1502         if (unlikely(!skb))
1503                 return NULL;
1504         slave_dev = skb->dev;
1505         bond_dev = ACCESS_ONCE(slave_dev->master);
1506         if (unlikely(!bond_dev))
1507                 return skb;
1508
1509         if (bond_dev->priv_flags & IFF_MASTER_ARPMON)
1510                 slave_dev->last_rx = jiffies;
1511
1512         if (bond_should_deliver_exact_match(skb, slave_dev, bond_dev)) {
1513                 skb->deliver_no_wcard = 1;
1514                 return skb;
1515         }
1516
1517         skb->dev = bond_dev;
1518
1519         if (bond_dev->priv_flags & IFF_MASTER_ALB &&
1520             bond_dev->priv_flags & IFF_BRIDGE_PORT &&
1521             skb->pkt_type == PACKET_HOST) {
1522                 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
1523
1524                 memcpy(dest, bond_dev->dev_addr, ETH_ALEN);
1525         }
1526
1527         return skb;
1528 }
1529
1530 /* enslave device <slave> to bond device <master> */
1531 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1532 {
1533         struct bonding *bond = netdev_priv(bond_dev);
1534         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1535         struct slave *new_slave = NULL;
1536         struct netdev_hw_addr *ha;
1537         struct sockaddr addr;
1538         int link_reporting;
1539         int old_features = bond_dev->features;
1540         int res = 0;
1541
1542         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1543                 slave_ops->ndo_do_ioctl == NULL) {
1544                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1545                            bond_dev->name, slave_dev->name);
1546         }
1547
1548         /* bond must be initialized by bond_open() before enslaving */
1549         if (!(bond_dev->flags & IFF_UP)) {
1550                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1551                            bond_dev->name);
1552         }
1553
1554         /* already enslaved */
1555         if (slave_dev->flags & IFF_SLAVE) {
1556                 pr_debug("Error, Device was already enslaved\n");
1557                 return -EBUSY;
1558         }
1559
1560         /* vlan challenged mutual exclusion */
1561         /* no need to lock since we're protected by rtnl_lock */
1562         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1563                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1564                 if (bond->vlgrp) {
1565                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1566                                bond_dev->name, slave_dev->name, bond_dev->name);
1567                         return -EPERM;
1568                 } else {
1569                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1570                                    bond_dev->name, slave_dev->name,
1571                                    slave_dev->name, bond_dev->name);
1572                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1573                 }
1574         } else {
1575                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1576                 if (bond->slave_cnt == 0) {
1577                         /* First slave, and it is not VLAN challenged,
1578                          * so remove the block of adding VLANs over the bond.
1579                          */
1580                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1581                 }
1582         }
1583
1584         /*
1585          * Old ifenslave binaries are no longer supported.  These can
1586          * be identified with moderate accuracy by the state of the slave:
1587          * the current ifenslave will set the interface down prior to
1588          * enslaving it; the old ifenslave will not.
1589          */
1590         if ((slave_dev->flags & IFF_UP)) {
1591                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1592                        slave_dev->name);
1593                 res = -EPERM;
1594                 goto err_undo_flags;
1595         }
1596
1597         /* set bonding device ether type by slave - bonding netdevices are
1598          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1599          * there is a need to override some of the type dependent attribs/funcs.
1600          *
1601          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1602          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1603          */
1604         if (bond->slave_cnt == 0) {
1605                 if (bond_dev->type != slave_dev->type) {
1606                         pr_debug("%s: change device type from %d to %d\n",
1607                                  bond_dev->name,
1608                                  bond_dev->type, slave_dev->type);
1609
1610                         res = netdev_bonding_change(bond_dev,
1611                                                     NETDEV_PRE_TYPE_CHANGE);
1612                         res = notifier_to_errno(res);
1613                         if (res) {
1614                                 pr_err("%s: refused to change device type\n",
1615                                        bond_dev->name);
1616                                 res = -EBUSY;
1617                                 goto err_undo_flags;
1618                         }
1619
1620                         /* Flush unicast and multicast addresses */
1621                         dev_uc_flush(bond_dev);
1622                         dev_mc_flush(bond_dev);
1623
1624                         if (slave_dev->type != ARPHRD_ETHER)
1625                                 bond_setup_by_slave(bond_dev, slave_dev);
1626                         else
1627                                 ether_setup(bond_dev);
1628
1629                         netdev_bonding_change(bond_dev,
1630                                               NETDEV_POST_TYPE_CHANGE);
1631                 }
1632         } else if (bond_dev->type != slave_dev->type) {
1633                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1634                        slave_dev->name,
1635                        slave_dev->type, bond_dev->type);
1636                 res = -EINVAL;
1637                 goto err_undo_flags;
1638         }
1639
1640         if (slave_ops->ndo_set_mac_address == NULL) {
1641                 if (bond->slave_cnt == 0) {
1642                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1643                                    bond_dev->name);
1644                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1645                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1646                         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",
1647                                bond_dev->name);
1648                         res = -EOPNOTSUPP;
1649                         goto err_undo_flags;
1650                 }
1651         }
1652
1653         /* If this is the first slave, then we need to set the master's hardware
1654          * address to be the same as the slave's. */
1655         if (is_zero_ether_addr(bond->dev->dev_addr))
1656                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1657                        slave_dev->addr_len);
1658
1659
1660         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1661         if (!new_slave) {
1662                 res = -ENOMEM;
1663                 goto err_undo_flags;
1664         }
1665
1666         /*
1667          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1668          * is set via sysfs or module option if desired.
1669          */
1670         new_slave->queue_id = 0;
1671
1672         /* Save slave's original mtu and then set it to match the bond */
1673         new_slave->original_mtu = slave_dev->mtu;
1674         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1675         if (res) {
1676                 pr_debug("Error %d calling dev_set_mtu\n", res);
1677                 goto err_free;
1678         }
1679
1680         /*
1681          * Save slave's original ("permanent") mac address for modes
1682          * that need it, and for restoring it upon release, and then
1683          * set it to the master's address
1684          */
1685         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1686
1687         if (!bond->params.fail_over_mac) {
1688                 /*
1689                  * Set slave to master's mac address.  The application already
1690                  * set the master's mac address to that of the first slave
1691                  */
1692                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1693                 addr.sa_family = slave_dev->type;
1694                 res = dev_set_mac_address(slave_dev, &addr);
1695                 if (res) {
1696                         pr_debug("Error %d calling set_mac_address\n", res);
1697                         goto err_restore_mtu;
1698                 }
1699         }
1700
1701         res = netdev_set_bond_master(slave_dev, bond_dev);
1702         if (res) {
1703                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1704                 goto err_restore_mac;
1705         }
1706         res = netdev_rx_handler_register(slave_dev, bond_handle_frame, NULL);
1707         if (res) {
1708                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1709                 goto err_unset_master;
1710         }
1711
1712         /* open the slave since the application closed it */
1713         res = dev_open(slave_dev);
1714         if (res) {
1715                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1716                 goto err_unreg_rxhandler;
1717         }
1718
1719         new_slave->dev = slave_dev;
1720         slave_dev->priv_flags |= IFF_BONDING;
1721
1722         if (bond_is_lb(bond)) {
1723                 /* bond_alb_init_slave() must be called before all other stages since
1724                  * it might fail and we do not want to have to undo everything
1725                  */
1726                 res = bond_alb_init_slave(bond, new_slave);
1727                 if (res)
1728                         goto err_close;
1729         }
1730
1731         /* If the mode USES_PRIMARY, then the new slave gets the
1732          * master's promisc (and mc) settings only if it becomes the
1733          * curr_active_slave, and that is taken care of later when calling
1734          * bond_change_active()
1735          */
1736         if (!USES_PRIMARY(bond->params.mode)) {
1737                 /* set promiscuity level to new slave */
1738                 if (bond_dev->flags & IFF_PROMISC) {
1739                         res = dev_set_promiscuity(slave_dev, 1);
1740                         if (res)
1741                                 goto err_close;
1742                 }
1743
1744                 /* set allmulti level to new slave */
1745                 if (bond_dev->flags & IFF_ALLMULTI) {
1746                         res = dev_set_allmulti(slave_dev, 1);
1747                         if (res)
1748                                 goto err_close;
1749                 }
1750
1751                 netif_addr_lock_bh(bond_dev);
1752                 /* upload master's mc_list to new slave */
1753                 netdev_for_each_mc_addr(ha, bond_dev)
1754                         dev_mc_add(slave_dev, ha->addr);
1755                 netif_addr_unlock_bh(bond_dev);
1756         }
1757
1758         if (bond->params.mode == BOND_MODE_8023AD) {
1759                 /* add lacpdu mc addr to mc list */
1760                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1761
1762                 dev_mc_add(slave_dev, lacpdu_multicast);
1763         }
1764
1765         bond_add_vlans_on_slave(bond, slave_dev);
1766
1767         write_lock_bh(&bond->lock);
1768
1769         bond_attach_slave(bond, new_slave);
1770
1771         new_slave->delay = 0;
1772         new_slave->link_failure_count = 0;
1773
1774         bond_compute_features(bond);
1775
1776         write_unlock_bh(&bond->lock);
1777
1778         read_lock(&bond->lock);
1779
1780         new_slave->last_arp_rx = jiffies;
1781
1782         if (bond->params.miimon && !bond->params.use_carrier) {
1783                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1784
1785                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1786                         /*
1787                          * miimon is set but a bonded network driver
1788                          * does not support ETHTOOL/MII and
1789                          * arp_interval is not set.  Note: if
1790                          * use_carrier is enabled, we will never go
1791                          * here (because netif_carrier is always
1792                          * supported); thus, we don't need to change
1793                          * the messages for netif_carrier.
1794                          */
1795                         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",
1796                                bond_dev->name, slave_dev->name);
1797                 } else if (link_reporting == -1) {
1798                         /* unable get link status using mii/ethtool */
1799                         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",
1800                                    bond_dev->name, slave_dev->name);
1801                 }
1802         }
1803
1804         /* check for initial state */
1805         if (!bond->params.miimon ||
1806             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1807                 if (bond->params.updelay) {
1808                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1809                         new_slave->link  = BOND_LINK_BACK;
1810                         new_slave->delay = bond->params.updelay;
1811                 } else {
1812                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1813                         new_slave->link  = BOND_LINK_UP;
1814                 }
1815                 new_slave->jiffies = jiffies;
1816         } else {
1817                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1818                 new_slave->link  = BOND_LINK_DOWN;
1819         }
1820
1821         if (bond_update_speed_duplex(new_slave) &&
1822             (new_slave->link != BOND_LINK_DOWN)) {
1823                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1824                            bond_dev->name, new_slave->dev->name);
1825
1826                 if (bond->params.mode == BOND_MODE_8023AD) {
1827                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1828                                    bond_dev->name);
1829                 }
1830         }
1831
1832         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1833                 /* if there is a primary slave, remember it */
1834                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1835                         bond->primary_slave = new_slave;
1836                         bond->force_primary = true;
1837                 }
1838         }
1839
1840         write_lock_bh(&bond->curr_slave_lock);
1841
1842         switch (bond->params.mode) {
1843         case BOND_MODE_ACTIVEBACKUP:
1844                 bond_set_slave_inactive_flags(new_slave);
1845                 bond_select_active_slave(bond);
1846                 break;
1847         case BOND_MODE_8023AD:
1848                 /* in 802.3ad mode, the internal mechanism
1849                  * will activate the slaves in the selected
1850                  * aggregator
1851                  */
1852                 bond_set_slave_inactive_flags(new_slave);
1853                 /* if this is the first slave */
1854                 if (bond->slave_cnt == 1) {
1855                         SLAVE_AD_INFO(new_slave).id = 1;
1856                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1857                          * can be called only after the mac address of the bond is set
1858                          */
1859                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1860                                             bond->params.lacp_fast);
1861                 } else {
1862                         SLAVE_AD_INFO(new_slave).id =
1863                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1864                 }
1865
1866                 bond_3ad_bind_slave(new_slave);
1867                 break;
1868         case BOND_MODE_TLB:
1869         case BOND_MODE_ALB:
1870                 new_slave->state = BOND_STATE_ACTIVE;
1871                 bond_set_slave_inactive_flags(new_slave);
1872                 bond_select_active_slave(bond);
1873                 break;
1874         default:
1875                 pr_debug("This slave is always active in trunk mode\n");
1876
1877                 /* always active in trunk mode */
1878                 new_slave->state = BOND_STATE_ACTIVE;
1879
1880                 /* In trunking mode there is little meaning to curr_active_slave
1881                  * anyway (it holds no special properties of the bond device),
1882                  * so we can change it without calling change_active_interface()
1883                  */
1884                 if (!bond->curr_active_slave)
1885                         bond->curr_active_slave = new_slave;
1886
1887                 break;
1888         } /* switch(bond_mode) */
1889
1890         write_unlock_bh(&bond->curr_slave_lock);
1891
1892         bond_set_carrier(bond);
1893
1894 #ifdef CONFIG_NET_POLL_CONTROLLER
1895         slave_dev->npinfo = bond_netpoll_info(bond);
1896         if (slave_dev->npinfo) {
1897                 if (slave_enable_netpoll(new_slave)) {
1898                         read_unlock(&bond->lock);
1899                         pr_info("Error, %s: master_dev is using netpoll, "
1900                                  "but new slave device does not support netpoll.\n",
1901                                  bond_dev->name);
1902                         res = -EBUSY;
1903                         goto err_close;
1904                 }
1905         }
1906 #endif
1907
1908         read_unlock(&bond->lock);
1909
1910         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1911         if (res)
1912                 goto err_close;
1913
1914         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1915                 bond_dev->name, slave_dev->name,
1916                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1917                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1918
1919         /* enslave is successful */
1920         return 0;
1921
1922 /* Undo stages on error */
1923 err_close:
1924         dev_close(slave_dev);
1925
1926 err_unreg_rxhandler:
1927         netdev_rx_handler_unregister(slave_dev);
1928
1929 err_unset_master:
1930         netdev_set_bond_master(slave_dev, NULL);
1931
1932 err_restore_mac:
1933         if (!bond->params.fail_over_mac) {
1934                 /* XXX TODO - fom follow mode needs to change master's
1935                  * MAC if this slave's MAC is in use by the bond, or at
1936                  * least print a warning.
1937                  */
1938                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1939                 addr.sa_family = slave_dev->type;
1940                 dev_set_mac_address(slave_dev, &addr);
1941         }
1942
1943 err_restore_mtu:
1944         dev_set_mtu(slave_dev, new_slave->original_mtu);
1945
1946 err_free:
1947         kfree(new_slave);
1948
1949 err_undo_flags:
1950         bond_dev->features = old_features;
1951
1952         return res;
1953 }
1954
1955 /*
1956  * Try to release the slave device <slave> from the bond device <master>
1957  * It is legal to access curr_active_slave without a lock because all the function
1958  * is write-locked.
1959  *
1960  * The rules for slave state should be:
1961  *   for Active/Backup:
1962  *     Active stays on all backups go down
1963  *   for Bonded connections:
1964  *     The first up interface should be left on and all others downed.
1965  */
1966 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1967 {
1968         struct bonding *bond = netdev_priv(bond_dev);
1969         struct slave *slave, *oldcurrent;
1970         struct sockaddr addr;
1971
1972         /* slave is not a slave or master is not master of this slave */
1973         if (!(slave_dev->flags & IFF_SLAVE) ||
1974             (slave_dev->master != bond_dev)) {
1975                 pr_err("%s: Error: cannot release %s.\n",
1976                        bond_dev->name, slave_dev->name);
1977                 return -EINVAL;
1978         }
1979
1980         block_netpoll_tx();
1981         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1982         write_lock_bh(&bond->lock);
1983
1984         slave = bond_get_slave_by_dev(bond, slave_dev);
1985         if (!slave) {
1986                 /* not a slave of this bond */
1987                 pr_info("%s: %s not enslaved\n",
1988                         bond_dev->name, slave_dev->name);
1989                 write_unlock_bh(&bond->lock);
1990                 unblock_netpoll_tx();
1991                 return -EINVAL;
1992         }
1993
1994         if (!bond->params.fail_over_mac) {
1995                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1996                     bond->slave_cnt > 1)
1997                         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",
1998                                    bond_dev->name, slave_dev->name,
1999                                    slave->perm_hwaddr,
2000                                    bond_dev->name, slave_dev->name);
2001         }
2002
2003         /* Inform AD package of unbinding of slave. */
2004         if (bond->params.mode == BOND_MODE_8023AD) {
2005                 /* must be called before the slave is
2006                  * detached from the list
2007                  */
2008                 bond_3ad_unbind_slave(slave);
2009         }
2010
2011         pr_info("%s: releasing %s interface %s\n",
2012                 bond_dev->name,
2013                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
2014                 slave_dev->name);
2015
2016         oldcurrent = bond->curr_active_slave;
2017
2018         bond->current_arp_slave = NULL;
2019
2020         /* release the slave from its bond */
2021         bond_detach_slave(bond, slave);
2022
2023         bond_compute_features(bond);
2024
2025         if (bond->primary_slave == slave)
2026                 bond->primary_slave = NULL;
2027
2028         if (oldcurrent == slave)
2029                 bond_change_active_slave(bond, NULL);
2030
2031         if (bond_is_lb(bond)) {
2032                 /* Must be called only after the slave has been
2033                  * detached from the list and the curr_active_slave
2034                  * has been cleared (if our_slave == old_current),
2035                  * but before a new active slave is selected.
2036                  */
2037                 write_unlock_bh(&bond->lock);
2038                 bond_alb_deinit_slave(bond, slave);
2039                 write_lock_bh(&bond->lock);
2040         }
2041
2042         if (oldcurrent == slave) {
2043                 /*
2044                  * Note that we hold RTNL over this sequence, so there
2045                  * is no concern that another slave add/remove event
2046                  * will interfere.
2047                  */
2048                 write_unlock_bh(&bond->lock);
2049                 read_lock(&bond->lock);
2050                 write_lock_bh(&bond->curr_slave_lock);
2051
2052                 bond_select_active_slave(bond);
2053
2054                 write_unlock_bh(&bond->curr_slave_lock);
2055                 read_unlock(&bond->lock);
2056                 write_lock_bh(&bond->lock);
2057         }
2058
2059         if (bond->slave_cnt == 0) {
2060                 bond_set_carrier(bond);
2061
2062                 /* if the last slave was removed, zero the mac address
2063                  * of the master so it will be set by the application
2064                  * to the mac address of the first slave
2065                  */
2066                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2067
2068                 if (!bond->vlgrp) {
2069                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2070                 } else {
2071                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2072                                    bond_dev->name, bond_dev->name);
2073                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2074                                    bond_dev->name);
2075                 }
2076         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2077                    !bond_has_challenged_slaves(bond)) {
2078                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2079                         bond_dev->name, slave_dev->name, bond_dev->name);
2080                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2081         }
2082
2083         write_unlock_bh(&bond->lock);
2084         unblock_netpoll_tx();
2085
2086         /* must do this from outside any spinlocks */
2087         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2088
2089         bond_del_vlans_from_slave(bond, slave_dev);
2090
2091         /* If the mode USES_PRIMARY, then we should only remove its
2092          * promisc and mc settings if it was the curr_active_slave, but that was
2093          * already taken care of above when we detached the slave
2094          */
2095         if (!USES_PRIMARY(bond->params.mode)) {
2096                 /* unset promiscuity level from slave */
2097                 if (bond_dev->flags & IFF_PROMISC)
2098                         dev_set_promiscuity(slave_dev, -1);
2099
2100                 /* unset allmulti level from slave */
2101                 if (bond_dev->flags & IFF_ALLMULTI)
2102                         dev_set_allmulti(slave_dev, -1);
2103
2104                 /* flush master's mc_list from slave */
2105                 netif_addr_lock_bh(bond_dev);
2106                 bond_mc_list_flush(bond_dev, slave_dev);
2107                 netif_addr_unlock_bh(bond_dev);
2108         }
2109
2110         netdev_rx_handler_unregister(slave_dev);
2111         netdev_set_bond_master(slave_dev, NULL);
2112
2113         slave_disable_netpoll(slave);
2114
2115         /* close slave before restoring its mac address */
2116         dev_close(slave_dev);
2117
2118         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2119                 /* restore original ("permanent") mac address */
2120                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2121                 addr.sa_family = slave_dev->type;
2122                 dev_set_mac_address(slave_dev, &addr);
2123         }
2124
2125         dev_set_mtu(slave_dev, slave->original_mtu);
2126
2127         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2128                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2129                                    IFF_SLAVE_NEEDARP);
2130
2131         kfree(slave);
2132
2133         return 0;  /* deletion OK */
2134 }
2135
2136 /*
2137 * First release a slave and than destroy the bond if no more slaves are left.
2138 * Must be under rtnl_lock when this function is called.
2139 */
2140 static int  bond_release_and_destroy(struct net_device *bond_dev,
2141                                      struct net_device *slave_dev)
2142 {
2143         struct bonding *bond = netdev_priv(bond_dev);
2144         int ret;
2145
2146         ret = bond_release(bond_dev, slave_dev);
2147         if ((ret == 0) && (bond->slave_cnt == 0)) {
2148                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2149                 pr_info("%s: destroying bond %s.\n",
2150                         bond_dev->name, bond_dev->name);
2151                 unregister_netdevice(bond_dev);
2152         }
2153         return ret;
2154 }
2155
2156 /*
2157  * This function releases all slaves.
2158  */
2159 static int bond_release_all(struct net_device *bond_dev)
2160 {
2161         struct bonding *bond = netdev_priv(bond_dev);
2162         struct slave *slave;
2163         struct net_device *slave_dev;
2164         struct sockaddr addr;
2165
2166         write_lock_bh(&bond->lock);
2167
2168         netif_carrier_off(bond_dev);
2169
2170         if (bond->slave_cnt == 0)
2171                 goto out;
2172
2173         bond->current_arp_slave = NULL;
2174         bond->primary_slave = NULL;
2175         bond_change_active_slave(bond, NULL);
2176
2177         while ((slave = bond->first_slave) != NULL) {
2178                 /* Inform AD package of unbinding of slave
2179                  * before slave is detached from the list.
2180                  */
2181                 if (bond->params.mode == BOND_MODE_8023AD)
2182                         bond_3ad_unbind_slave(slave);
2183
2184                 slave_dev = slave->dev;
2185                 bond_detach_slave(bond, slave);
2186
2187                 /* now that the slave is detached, unlock and perform
2188                  * all the undo steps that should not be called from
2189                  * within a lock.
2190                  */
2191                 write_unlock_bh(&bond->lock);
2192
2193                 if (bond_is_lb(bond)) {
2194                         /* must be called only after the slave
2195                          * has been detached from the list
2196                          */
2197                         bond_alb_deinit_slave(bond, slave);
2198                 }
2199
2200                 bond_compute_features(bond);
2201
2202                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2203                 bond_del_vlans_from_slave(bond, slave_dev);
2204
2205                 /* If the mode USES_PRIMARY, then we should only remove its
2206                  * promisc and mc settings if it was the curr_active_slave, but that was
2207                  * already taken care of above when we detached the slave
2208                  */
2209                 if (!USES_PRIMARY(bond->params.mode)) {
2210                         /* unset promiscuity level from slave */
2211                         if (bond_dev->flags & IFF_PROMISC)
2212                                 dev_set_promiscuity(slave_dev, -1);
2213
2214                         /* unset allmulti level from slave */
2215                         if (bond_dev->flags & IFF_ALLMULTI)
2216                                 dev_set_allmulti(slave_dev, -1);
2217
2218                         /* flush master's mc_list from slave */
2219                         netif_addr_lock_bh(bond_dev);
2220                         bond_mc_list_flush(bond_dev, slave_dev);
2221                         netif_addr_unlock_bh(bond_dev);
2222                 }
2223
2224                 netdev_rx_handler_unregister(slave_dev);
2225                 netdev_set_bond_master(slave_dev, NULL);
2226
2227                 slave_disable_netpoll(slave);
2228
2229                 /* close slave before restoring its mac address */
2230                 dev_close(slave_dev);
2231
2232                 if (!bond->params.fail_over_mac) {
2233                         /* restore original ("permanent") mac address*/
2234                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2235                         addr.sa_family = slave_dev->type;
2236                         dev_set_mac_address(slave_dev, &addr);
2237                 }
2238
2239                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2240                                            IFF_SLAVE_INACTIVE);
2241
2242                 kfree(slave);
2243
2244                 /* re-acquire the lock before getting the next slave */
2245                 write_lock_bh(&bond->lock);
2246         }
2247
2248         /* zero the mac address of the master so it will be
2249          * set by the application to the mac address of the
2250          * first slave
2251          */
2252         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2253
2254         if (!bond->vlgrp) {
2255                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2256         } else {
2257                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2258                            bond_dev->name, bond_dev->name);
2259                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2260                            bond_dev->name);
2261         }
2262
2263         pr_info("%s: released all slaves\n", bond_dev->name);
2264
2265 out:
2266         write_unlock_bh(&bond->lock);
2267         return 0;
2268 }
2269
2270 /*
2271  * This function changes the active slave to slave <slave_dev>.
2272  * It returns -EINVAL in the following cases.
2273  *  - <slave_dev> is not found in the list.
2274  *  - There is not active slave now.
2275  *  - <slave_dev> is already active.
2276  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2277  *  - <slave_dev> is not running.
2278  * In these cases, this function does nothing.
2279  * In the other cases, current_slave pointer is changed and 0 is returned.
2280  */
2281 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2282 {
2283         struct bonding *bond = netdev_priv(bond_dev);
2284         struct slave *old_active = NULL;
2285         struct slave *new_active = NULL;
2286         int res = 0;
2287
2288         if (!USES_PRIMARY(bond->params.mode))
2289                 return -EINVAL;
2290
2291         /* Verify that master_dev is indeed the master of slave_dev */
2292         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2293                 return -EINVAL;
2294
2295         read_lock(&bond->lock);
2296
2297         read_lock(&bond->curr_slave_lock);
2298         old_active = bond->curr_active_slave;
2299         read_unlock(&bond->curr_slave_lock);
2300
2301         new_active = bond_get_slave_by_dev(bond, slave_dev);
2302
2303         /*
2304          * Changing to the current active: do nothing; return success.
2305          */
2306         if (new_active && (new_active == old_active)) {
2307                 read_unlock(&bond->lock);
2308                 return 0;
2309         }
2310
2311         if ((new_active) &&
2312             (old_active) &&
2313             (new_active->link == BOND_LINK_UP) &&
2314             IS_UP(new_active->dev)) {
2315                 block_netpoll_tx();
2316                 write_lock_bh(&bond->curr_slave_lock);
2317                 bond_change_active_slave(bond, new_active);
2318                 write_unlock_bh(&bond->curr_slave_lock);
2319                 unblock_netpoll_tx();
2320         } else
2321                 res = -EINVAL;
2322
2323         read_unlock(&bond->lock);
2324
2325         return res;
2326 }
2327
2328 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2329 {
2330         struct bonding *bond = netdev_priv(bond_dev);
2331
2332         info->bond_mode = bond->params.mode;
2333         info->miimon = bond->params.miimon;
2334
2335         read_lock(&bond->lock);
2336         info->num_slaves = bond->slave_cnt;
2337         read_unlock(&bond->lock);
2338
2339         return 0;
2340 }
2341
2342 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2343 {
2344         struct bonding *bond = netdev_priv(bond_dev);
2345         struct slave *slave;
2346         int i, res = -ENODEV;
2347
2348         read_lock(&bond->lock);
2349
2350         bond_for_each_slave(bond, slave, i) {
2351                 if (i == (int)info->slave_id) {
2352                         res = 0;
2353                         strcpy(info->slave_name, slave->dev->name);
2354                         info->link = slave->link;
2355                         info->state = slave->state;
2356                         info->link_failure_count = slave->link_failure_count;
2357                         break;
2358                 }
2359         }
2360
2361         read_unlock(&bond->lock);
2362
2363         return res;
2364 }
2365
2366 /*-------------------------------- Monitoring -------------------------------*/
2367
2368
2369 static int bond_miimon_inspect(struct bonding *bond)
2370 {
2371         struct slave *slave;
2372         int i, link_state, commit = 0;
2373         bool ignore_updelay;
2374
2375         ignore_updelay = !bond->curr_active_slave ? true : false;
2376
2377         bond_for_each_slave(bond, slave, i) {
2378                 slave->new_link = BOND_LINK_NOCHANGE;
2379
2380                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2381
2382                 switch (slave->link) {
2383                 case BOND_LINK_UP:
2384                         if (link_state)
2385                                 continue;
2386
2387                         slave->link = BOND_LINK_FAIL;
2388                         slave->delay = bond->params.downdelay;
2389                         if (slave->delay) {
2390                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2391                                         bond->dev->name,
2392                                         (bond->params.mode ==
2393                                          BOND_MODE_ACTIVEBACKUP) ?
2394                                         ((slave->state == BOND_STATE_ACTIVE) ?
2395                                          "active " : "backup ") : "",
2396                                         slave->dev->name,
2397                                         bond->params.downdelay * bond->params.miimon);
2398                         }
2399                         /*FALLTHRU*/
2400                 case BOND_LINK_FAIL:
2401                         if (link_state) {
2402                                 /*
2403                                  * recovered before downdelay expired
2404                                  */
2405                                 slave->link = BOND_LINK_UP;
2406                                 slave->jiffies = jiffies;
2407                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2408                                         bond->dev->name,
2409                                         (bond->params.downdelay - slave->delay) *
2410                                         bond->params.miimon,
2411                                         slave->dev->name);
2412                                 continue;
2413                         }
2414
2415                         if (slave->delay <= 0) {
2416                                 slave->new_link = BOND_LINK_DOWN;
2417                                 commit++;
2418                                 continue;
2419                         }
2420
2421                         slave->delay--;
2422                         break;
2423
2424                 case BOND_LINK_DOWN:
2425                         if (!link_state)
2426                                 continue;
2427
2428                         slave->link = BOND_LINK_BACK;
2429                         slave->delay = bond->params.updelay;
2430
2431                         if (slave->delay) {
2432                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2433                                         bond->dev->name, slave->dev->name,
2434                                         ignore_updelay ? 0 :
2435                                         bond->params.updelay *
2436                                         bond->params.miimon);
2437                         }
2438                         /*FALLTHRU*/
2439                 case BOND_LINK_BACK:
2440                         if (!link_state) {
2441                                 slave->link = BOND_LINK_DOWN;
2442                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2443                                         bond->dev->name,
2444                                         (bond->params.updelay - slave->delay) *
2445                                         bond->params.miimon,
2446                                         slave->dev->name);
2447
2448                                 continue;
2449                         }
2450
2451                         if (ignore_updelay)
2452                                 slave->delay = 0;
2453
2454                         if (slave->delay <= 0) {
2455                                 slave->new_link = BOND_LINK_UP;
2456                                 commit++;
2457                                 ignore_updelay = false;
2458                                 continue;
2459                         }
2460
2461                         slave->delay--;
2462                         break;
2463                 }
2464         }
2465
2466         return commit;
2467 }
2468
2469 static void bond_miimon_commit(struct bonding *bond)
2470 {
2471         struct slave *slave;
2472         int i;
2473
2474         bond_for_each_slave(bond, slave, i) {
2475                 switch (slave->new_link) {
2476                 case BOND_LINK_NOCHANGE:
2477                         continue;
2478
2479                 case BOND_LINK_UP:
2480                         slave->link = BOND_LINK_UP;
2481                         slave->jiffies = jiffies;
2482
2483                         if (bond->params.mode == BOND_MODE_8023AD) {
2484                                 /* prevent it from being the active one */
2485                                 slave->state = BOND_STATE_BACKUP;
2486                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2487                                 /* make it immediately active */
2488                                 slave->state = BOND_STATE_ACTIVE;
2489                         } else if (slave != bond->primary_slave) {
2490                                 /* prevent it from being the active one */
2491                                 slave->state = BOND_STATE_BACKUP;
2492                         }
2493
2494                         bond_update_speed_duplex(slave);
2495
2496                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2497                                 bond->dev->name, slave->dev->name,
2498                                 slave->speed, slave->duplex ? "full" : "half");
2499
2500                         /* notify ad that the link status has changed */
2501                         if (bond->params.mode == BOND_MODE_8023AD)
2502                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2503
2504                         if (bond_is_lb(bond))
2505                                 bond_alb_handle_link_change(bond, slave,
2506                                                             BOND_LINK_UP);
2507
2508                         if (!bond->curr_active_slave ||
2509                             (slave == bond->primary_slave))
2510                                 goto do_failover;
2511
2512                         continue;
2513
2514                 case BOND_LINK_DOWN:
2515                         if (slave->link_failure_count < UINT_MAX)
2516                                 slave->link_failure_count++;
2517
2518                         slave->link = BOND_LINK_DOWN;
2519
2520                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2521                             bond->params.mode == BOND_MODE_8023AD)
2522                                 bond_set_slave_inactive_flags(slave);
2523
2524                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2525                                 bond->dev->name, slave->dev->name);
2526
2527                         if (bond->params.mode == BOND_MODE_8023AD)
2528                                 bond_3ad_handle_link_change(slave,
2529                                                             BOND_LINK_DOWN);
2530
2531                         if (bond_is_lb(bond))
2532                                 bond_alb_handle_link_change(bond, slave,
2533                                                             BOND_LINK_DOWN);
2534
2535                         if (slave == bond->curr_active_slave)
2536                                 goto do_failover;
2537
2538                         continue;
2539
2540                 default:
2541                         pr_err("%s: invalid new link %d on slave %s\n",
2542                                bond->dev->name, slave->new_link,
2543                                slave->dev->name);
2544                         slave->new_link = BOND_LINK_NOCHANGE;
2545
2546                         continue;
2547                 }
2548
2549 do_failover:
2550                 ASSERT_RTNL();
2551                 block_netpoll_tx();
2552                 write_lock_bh(&bond->curr_slave_lock);
2553                 bond_select_active_slave(bond);
2554                 write_unlock_bh(&bond->curr_slave_lock);
2555                 unblock_netpoll_tx();
2556         }
2557
2558         bond_set_carrier(bond);
2559 }
2560
2561 /*
2562  * bond_mii_monitor
2563  *
2564  * Really a wrapper that splits the mii monitor into two phases: an
2565  * inspection, then (if inspection indicates something needs to be done)
2566  * an acquisition of appropriate locks followed by a commit phase to
2567  * implement whatever link state changes are indicated.
2568  */
2569 void bond_mii_monitor(struct work_struct *work)
2570 {
2571         struct bonding *bond = container_of(work, struct bonding,
2572                                             mii_work.work);
2573
2574         read_lock(&bond->lock);
2575         if (bond->kill_timers)
2576                 goto out;
2577
2578         if (bond->slave_cnt == 0)
2579                 goto re_arm;
2580
2581         if (bond->send_grat_arp) {
2582                 read_lock(&bond->curr_slave_lock);
2583                 bond_send_gratuitous_arp(bond);
2584                 read_unlock(&bond->curr_slave_lock);
2585         }
2586
2587         if (bond->send_unsol_na) {
2588                 read_lock(&bond->curr_slave_lock);
2589                 bond_send_unsolicited_na(bond);
2590                 read_unlock(&bond->curr_slave_lock);
2591         }
2592
2593         if (bond_miimon_inspect(bond)) {
2594                 read_unlock(&bond->lock);
2595                 rtnl_lock();
2596                 read_lock(&bond->lock);
2597
2598                 bond_miimon_commit(bond);
2599
2600                 read_unlock(&bond->lock);
2601                 rtnl_unlock();  /* might sleep, hold no other locks */
2602                 read_lock(&bond->lock);
2603         }
2604
2605 re_arm:
2606         if (bond->params.miimon)
2607                 queue_delayed_work(bond->wq, &bond->mii_work,
2608                                    msecs_to_jiffies(bond->params.miimon));
2609 out:
2610         read_unlock(&bond->lock);
2611 }
2612
2613 static __be32 bond_glean_dev_ip(struct net_device *dev)
2614 {
2615         struct in_device *idev;
2616         struct in_ifaddr *ifa;
2617         __be32 addr = 0;
2618
2619         if (!dev)
2620                 return 0;
2621
2622         rcu_read_lock();
2623         idev = __in_dev_get_rcu(dev);
2624         if (!idev)
2625                 goto out;
2626
2627         ifa = idev->ifa_list;
2628         if (!ifa)
2629                 goto out;
2630
2631         addr = ifa->ifa_local;
2632 out:
2633         rcu_read_unlock();
2634         return addr;
2635 }
2636
2637 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2638 {
2639         struct vlan_entry *vlan;
2640
2641         if (ip == bond->master_ip)
2642                 return 1;
2643
2644         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2645                 if (ip == vlan->vlan_ip)
2646                         return 1;
2647         }
2648
2649         return 0;
2650 }
2651
2652 /*
2653  * We go to the (large) trouble of VLAN tagging ARP frames because
2654  * switches in VLAN mode (especially if ports are configured as
2655  * "native" to a VLAN) might not pass non-tagged frames.
2656  */
2657 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2658 {
2659         struct sk_buff *skb;
2660
2661         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2662                  slave_dev->name, dest_ip, src_ip, vlan_id);
2663
2664         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2665                          NULL, slave_dev->dev_addr, NULL);
2666
2667         if (!skb) {
2668                 pr_err("ARP packet allocation failed\n");
2669                 return;
2670         }
2671         if (vlan_id) {
2672                 skb = vlan_put_tag(skb, vlan_id);
2673                 if (!skb) {
2674                         pr_err("failed to insert VLAN tag\n");
2675                         return;
2676                 }
2677         }
2678         arp_xmit(skb);
2679 }
2680
2681
2682 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2683 {
2684         int i, vlan_id;
2685         __be32 *targets = bond->params.arp_targets;
2686         struct vlan_entry *vlan;
2687         struct net_device *vlan_dev;
2688         struct flowi fl;
2689         struct rtable *rt;
2690
2691         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2692                 if (!targets[i])
2693                         break;
2694                 pr_debug("basa: target %x\n", targets[i]);
2695                 if (!bond->vlgrp) {
2696                         pr_debug("basa: empty vlan: arp_send\n");
2697                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2698                                       bond->master_ip, 0);
2699                         continue;
2700                 }
2701
2702                 /*
2703                  * If VLANs are configured, we do a route lookup to
2704                  * determine which VLAN interface would be used, so we
2705                  * can tag the ARP with the proper VLAN tag.
2706                  */
2707                 memset(&fl, 0, sizeof(fl));
2708                 fl.fl4_dst = targets[i];
2709                 fl.fl4_tos = RTO_ONLINK;
2710
2711                 rt = ip_route_output_key(dev_net(bond->dev), &fl);
2712                 if (IS_ERR(rt)) {
2713                         if (net_ratelimit()) {
2714                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2715                                            bond->dev->name, &fl.fl4_dst);
2716                         }
2717                         continue;
2718                 }
2719
2720                 /*
2721                  * This target is not on a VLAN
2722                  */
2723                 if (rt->dst.dev == bond->dev) {
2724                         ip_rt_put(rt);
2725                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2726                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2727                                       bond->master_ip, 0);
2728                         continue;
2729                 }
2730
2731                 vlan_id = 0;
2732                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2733                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2734                         if (vlan_dev == rt->dst.dev) {
2735                                 vlan_id = vlan->vlan_id;
2736                                 pr_debug("basa: vlan match on %s %d\n",
2737                                        vlan_dev->name, vlan_id);
2738                                 break;
2739                         }
2740                 }
2741
2742                 if (vlan_id) {
2743                         ip_rt_put(rt);
2744                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2745                                       vlan->vlan_ip, vlan_id);
2746                         continue;
2747                 }
2748
2749                 if (net_ratelimit()) {
2750                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2751                                    bond->dev->name, &fl.fl4_dst,
2752                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2753                 }
2754                 ip_rt_put(rt);
2755         }
2756 }
2757
2758 /*
2759  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2760  * for each VLAN above us.
2761  *
2762  * Caller must hold curr_slave_lock for read or better
2763  */
2764 static void bond_send_gratuitous_arp(struct bonding *bond)
2765 {
2766         struct slave *slave = bond->curr_active_slave;
2767         struct vlan_entry *vlan;
2768         struct net_device *vlan_dev;
2769
2770         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2771                  bond->dev->name, slave ? slave->dev->name : "NULL");
2772
2773         if (!slave || !bond->send_grat_arp ||
2774             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2775                 return;
2776
2777         bond->send_grat_arp--;
2778
2779         if (bond->master_ip) {
2780                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2781                                 bond->master_ip, 0);
2782         }
2783
2784         if (!bond->vlgrp)
2785                 return;
2786
2787         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2788                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2789                 if (vlan->vlan_ip) {
2790                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2791                                       vlan->vlan_ip, vlan->vlan_id);
2792                 }
2793         }
2794 }
2795
2796 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2797 {
2798         int i;
2799         __be32 *targets = bond->params.arp_targets;
2800
2801         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2802                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2803                          &sip, &tip, i, &targets[i],
2804                          bond_has_this_ip(bond, tip));
2805                 if (sip == targets[i]) {
2806                         if (bond_has_this_ip(bond, tip))
2807                                 slave->last_arp_rx = jiffies;
2808                         return;
2809                 }
2810         }
2811 }
2812
2813 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2814 {
2815         struct arphdr *arp;
2816         struct slave *slave;
2817         struct bonding *bond;
2818         unsigned char *arp_ptr;
2819         __be32 sip, tip;
2820
2821         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2822                 /*
2823                  * When using VLANS and bonding, dev and oriv_dev may be
2824                  * incorrect if the physical interface supports VLAN
2825                  * acceleration.  With this change ARP validation now
2826                  * works for hosts only reachable on the VLAN interface.
2827                  */
2828                 dev = vlan_dev_real_dev(dev);
2829                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2830         }
2831
2832         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2833                 goto out;
2834
2835         bond = netdev_priv(dev);
2836         read_lock(&bond->lock);
2837
2838         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2839                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2840                  orig_dev ? orig_dev->name : "NULL");
2841
2842         slave = bond_get_slave_by_dev(bond, orig_dev);
2843         if (!slave || !slave_do_arp_validate(bond, slave))
2844                 goto out_unlock;
2845
2846         skb = skb_share_check(skb, GFP_ATOMIC);
2847         if (!skb)
2848                 goto out_unlock;
2849
2850         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2851                 goto out_unlock;
2852
2853         arp = arp_hdr(skb);
2854         if (arp->ar_hln != dev->addr_len ||
2855             skb->pkt_type == PACKET_OTHERHOST ||
2856             skb->pkt_type == PACKET_LOOPBACK ||
2857             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2858             arp->ar_pro != htons(ETH_P_IP) ||
2859             arp->ar_pln != 4)
2860                 goto out_unlock;
2861
2862         arp_ptr = (unsigned char *)(arp + 1);
2863         arp_ptr += dev->addr_len;
2864         memcpy(&sip, arp_ptr, 4);
2865         arp_ptr += 4 + dev->addr_len;
2866         memcpy(&tip, arp_ptr, 4);
2867
2868         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2869                  bond->dev->name, slave->dev->name, slave->state,
2870                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2871                  &sip, &tip);
2872
2873         /*
2874          * Backup slaves won't see the ARP reply, but do come through
2875          * here for each ARP probe (so we swap the sip/tip to validate
2876          * the probe).  In a "redundant switch, common router" type of
2877          * configuration, the ARP probe will (hopefully) travel from
2878          * the active, through one switch, the router, then the other
2879          * switch before reaching the backup.
2880          */
2881         if (slave->state == BOND_STATE_ACTIVE)
2882                 bond_validate_arp(bond, slave, sip, tip);
2883         else
2884                 bond_validate_arp(bond, slave, tip, sip);
2885
2886 out_unlock:
2887         read_unlock(&bond->lock);
2888 out:
2889         dev_kfree_skb(skb);
2890         return NET_RX_SUCCESS;
2891 }
2892
2893 /*
2894  * this function is called regularly to monitor each slave's link
2895  * ensuring that traffic is being sent and received when arp monitoring
2896  * is used in load-balancing mode. if the adapter has been dormant, then an
2897  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2898  * arp monitoring in active backup mode.
2899  */
2900 void bond_loadbalance_arp_mon(struct work_struct *work)
2901 {
2902         struct bonding *bond = container_of(work, struct bonding,
2903                                             arp_work.work);
2904         struct slave *slave, *oldcurrent;
2905         int do_failover = 0;
2906         int delta_in_ticks;
2907         int i;
2908
2909         read_lock(&bond->lock);
2910
2911         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2912
2913         if (bond->kill_timers)
2914                 goto out;
2915
2916         if (bond->slave_cnt == 0)
2917                 goto re_arm;
2918
2919         read_lock(&bond->curr_slave_lock);
2920         oldcurrent = bond->curr_active_slave;
2921         read_unlock(&bond->curr_slave_lock);
2922
2923         /* see if any of the previous devices are up now (i.e. they have
2924          * xmt and rcv traffic). the curr_active_slave does not come into
2925          * the picture unless it is null. also, slave->jiffies is not needed
2926          * here because we send an arp on each slave and give a slave as
2927          * long as it needs to get the tx/rx within the delta.
2928          * TODO: what about up/down delay in arp mode? it wasn't here before
2929          *       so it can wait
2930          */
2931         bond_for_each_slave(bond, slave, i) {
2932                 unsigned long trans_start = dev_trans_start(slave->dev);
2933
2934                 if (slave->link != BOND_LINK_UP) {
2935                         if (time_in_range(jiffies,
2936                                 trans_start - delta_in_ticks,
2937                                 trans_start + delta_in_ticks) &&
2938                             time_in_range(jiffies,
2939                                 slave->dev->last_rx - delta_in_ticks,
2940                                 slave->dev->last_rx + delta_in_ticks)) {
2941
2942                                 slave->link  = BOND_LINK_UP;
2943                                 slave->state = BOND_STATE_ACTIVE;
2944
2945                                 /* primary_slave has no meaning in round-robin
2946                                  * mode. the window of a slave being up and
2947                                  * curr_active_slave being null after enslaving
2948                                  * is closed.
2949                                  */
2950                                 if (!oldcurrent) {
2951                                         pr_info("%s: link status definitely up for interface %s, ",
2952                                                 bond->dev->name,
2953                                                 slave->dev->name);
2954                                         do_failover = 1;
2955                                 } else {
2956                                         pr_info("%s: interface %s is now up\n",
2957                                                 bond->dev->name,
2958                                                 slave->dev->name);
2959                                 }
2960                         }
2961                 } else {
2962                         /* slave->link == BOND_LINK_UP */
2963
2964                         /* not all switches will respond to an arp request
2965                          * when the source ip is 0, so don't take the link down
2966                          * if we don't know our ip yet
2967                          */
2968                         if (!time_in_range(jiffies,
2969                                 trans_start - delta_in_ticks,
2970                                 trans_start + 2 * delta_in_ticks) ||
2971                             !time_in_range(jiffies,
2972                                 slave->dev->last_rx - delta_in_ticks,
2973                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2974
2975                                 slave->link  = BOND_LINK_DOWN;
2976                                 slave->state = BOND_STATE_BACKUP;
2977
2978                                 if (slave->link_failure_count < UINT_MAX)
2979                                         slave->link_failure_count++;
2980
2981                                 pr_info("%s: interface %s is now down.\n",
2982                                         bond->dev->name,
2983                                         slave->dev->name);
2984
2985                                 if (slave == oldcurrent)
2986                                         do_failover = 1;
2987                         }
2988                 }
2989
2990                 /* note: if switch is in round-robin mode, all links
2991                  * must tx arp to ensure all links rx an arp - otherwise
2992                  * links may oscillate or not come up at all; if switch is
2993                  * in something like xor mode, there is nothing we can
2994                  * do - all replies will be rx'ed on same link causing slaves
2995                  * to be unstable during low/no traffic periods
2996                  */
2997                 if (IS_UP(slave->dev))
2998                         bond_arp_send_all(bond, slave);
2999         }
3000
3001         if (do_failover) {
3002                 block_netpoll_tx();
3003                 write_lock_bh(&bond->curr_slave_lock);
3004
3005                 bond_select_active_slave(bond);
3006
3007                 write_unlock_bh(&bond->curr_slave_lock);
3008                 unblock_netpoll_tx();
3009         }
3010
3011 re_arm:
3012         if (bond->params.arp_interval)
3013                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3014 out:
3015         read_unlock(&bond->lock);
3016 }
3017
3018 /*
3019  * Called to inspect slaves for active-backup mode ARP monitor link state
3020  * changes.  Sets new_link in slaves to specify what action should take
3021  * place for the slave.  Returns 0 if no changes are found, >0 if changes
3022  * to link states must be committed.
3023  *
3024  * Called with bond->lock held for read.
3025  */
3026 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
3027 {
3028         struct slave *slave;
3029         int i, commit = 0;
3030         unsigned long trans_start;
3031
3032         bond_for_each_slave(bond, slave, i) {
3033                 slave->new_link = BOND_LINK_NOCHANGE;
3034
3035                 if (slave->link != BOND_LINK_UP) {
3036                         if (time_in_range(jiffies,
3037                                 slave_last_rx(bond, slave) - delta_in_ticks,
3038                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
3039
3040                                 slave->new_link = BOND_LINK_UP;
3041                                 commit++;
3042                         }
3043
3044                         continue;
3045                 }
3046
3047                 /*
3048                  * Give slaves 2*delta after being enslaved or made
3049                  * active.  This avoids bouncing, as the last receive
3050                  * times need a full ARP monitor cycle to be updated.
3051                  */
3052                 if (time_in_range(jiffies,
3053                                   slave->jiffies - delta_in_ticks,
3054                                   slave->jiffies + 2 * delta_in_ticks))
3055                         continue;
3056
3057                 /*
3058                  * Backup slave is down if:
3059                  * - No current_arp_slave AND
3060                  * - more than 3*delta since last receive AND
3061                  * - the bond has an IP address
3062                  *
3063                  * Note: a non-null current_arp_slave indicates
3064                  * the curr_active_slave went down and we are
3065                  * searching for a new one; under this condition
3066                  * we only take the curr_active_slave down - this
3067                  * gives each slave a chance to tx/rx traffic
3068                  * before being taken out
3069                  */
3070                 if (slave->state == BOND_STATE_BACKUP &&
3071                     !bond->current_arp_slave &&
3072                     !time_in_range(jiffies,
3073                         slave_last_rx(bond, slave) - delta_in_ticks,
3074                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3075
3076                         slave->new_link = BOND_LINK_DOWN;
3077                         commit++;
3078                 }
3079
3080                 /*
3081                  * Active slave is down if:
3082                  * - more than 2*delta since transmitting OR
3083                  * - (more than 2*delta since receive AND
3084                  *    the bond has an IP address)
3085                  */
3086                 trans_start = dev_trans_start(slave->dev);
3087                 if ((slave->state == BOND_STATE_ACTIVE) &&
3088                     (!time_in_range(jiffies,
3089                         trans_start - delta_in_ticks,
3090                         trans_start + 2 * delta_in_ticks) ||
3091                      !time_in_range(jiffies,
3092                         slave_last_rx(bond, slave) - delta_in_ticks,
3093                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3094
3095                         slave->new_link = BOND_LINK_DOWN;
3096                         commit++;
3097                 }
3098         }
3099
3100         return commit;
3101 }
3102
3103 /*
3104  * Called to commit link state changes noted by inspection step of
3105  * active-backup mode ARP monitor.
3106  *
3107  * Called with RTNL and bond->lock for read.
3108  */
3109 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3110 {
3111         struct slave *slave;
3112         int i;
3113         unsigned long trans_start;
3114
3115         bond_for_each_slave(bond, slave, i) {
3116                 switch (slave->new_link) {
3117                 case BOND_LINK_NOCHANGE:
3118                         continue;
3119
3120                 case BOND_LINK_UP:
3121                         trans_start = dev_trans_start(slave->dev);
3122                         if ((!bond->curr_active_slave &&
3123                              time_in_range(jiffies,
3124                                            trans_start - delta_in_ticks,
3125                                            trans_start + delta_in_ticks)) ||
3126                             bond->curr_active_slave != slave) {
3127                                 slave->link = BOND_LINK_UP;
3128                                 bond->current_arp_slave = NULL;
3129
3130                                 pr_info("%s: link status definitely up for interface %s.\n",
3131                                         bond->dev->name, slave->dev->name);
3132
3133                                 if (!bond->curr_active_slave ||
3134                                     (slave == bond->primary_slave))
3135                                         goto do_failover;
3136
3137                         }
3138
3139                         continue;
3140
3141                 case BOND_LINK_DOWN:
3142                         if (slave->link_failure_count < UINT_MAX)
3143                                 slave->link_failure_count++;
3144
3145                         slave->link = BOND_LINK_DOWN;
3146                         bond_set_slave_inactive_flags(slave);
3147
3148                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3149                                 bond->dev->name, slave->dev->name);
3150
3151                         if (slave == bond->curr_active_slave) {
3152                                 bond->current_arp_slave = NULL;
3153                                 goto do_failover;
3154                         }
3155
3156                         continue;
3157
3158                 default:
3159                         pr_err("%s: impossible: new_link %d on slave %s\n",
3160                                bond->dev->name, slave->new_link,
3161                                slave->dev->name);
3162                         continue;
3163                 }
3164
3165 do_failover:
3166                 ASSERT_RTNL();
3167                 block_netpoll_tx();
3168                 write_lock_bh(&bond->curr_slave_lock);
3169                 bond_select_active_slave(bond);
3170                 write_unlock_bh(&bond->curr_slave_lock);
3171                 unblock_netpoll_tx();
3172         }
3173
3174         bond_set_carrier(bond);
3175 }
3176
3177 /*
3178  * Send ARP probes for active-backup mode ARP monitor.
3179  *
3180  * Called with bond->lock held for read.
3181  */
3182 static void bond_ab_arp_probe(struct bonding *bond)
3183 {
3184         struct slave *slave;
3185         int i;
3186
3187         read_lock(&bond->curr_slave_lock);
3188
3189         if (bond->current_arp_slave && bond->curr_active_slave)
3190                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3191                         bond->current_arp_slave->dev->name,
3192                         bond->curr_active_slave->dev->name);
3193
3194         if (bond->curr_active_slave) {
3195                 bond_arp_send_all(bond, bond->curr_active_slave);
3196                 read_unlock(&bond->curr_slave_lock);
3197                 return;
3198         }
3199
3200         read_unlock(&bond->curr_slave_lock);
3201
3202         /* if we don't have a curr_active_slave, search for the next available
3203          * backup slave from the current_arp_slave and make it the candidate
3204          * for becoming the curr_active_slave
3205          */
3206
3207         if (!bond->current_arp_slave) {
3208                 bond->current_arp_slave = bond->first_slave;
3209                 if (!bond->current_arp_slave)
3210                         return;
3211         }
3212
3213         bond_set_slave_inactive_flags(bond->current_arp_slave);
3214
3215         /* search for next candidate */
3216         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3217                 if (IS_UP(slave->dev)) {
3218                         slave->link = BOND_LINK_BACK;
3219                         bond_set_slave_active_flags(slave);
3220                         bond_arp_send_all(bond, slave);
3221                         slave->jiffies = jiffies;
3222                         bond->current_arp_slave = slave;
3223                         break;
3224                 }
3225
3226                 /* if the link state is up at this point, we
3227                  * mark it down - this can happen if we have
3228                  * simultaneous link failures and
3229                  * reselect_active_interface doesn't make this
3230                  * one the current slave so it is still marked
3231                  * up when it is actually down
3232                  */
3233                 if (slave->link == BOND_LINK_UP) {
3234                         slave->link = BOND_LINK_DOWN;
3235                         if (slave->link_failure_count < UINT_MAX)
3236                                 slave->link_failure_count++;
3237
3238                         bond_set_slave_inactive_flags(slave);
3239
3240                         pr_info("%s: backup interface %s is now down.\n",
3241                                 bond->dev->name, slave->dev->name);
3242                 }
3243         }
3244 }
3245
3246 void bond_activebackup_arp_mon(struct work_struct *work)
3247 {
3248         struct bonding *bond = container_of(work, struct bonding,
3249                                             arp_work.work);
3250         int delta_in_ticks;
3251
3252         read_lock(&bond->lock);
3253
3254         if (bond->kill_timers)
3255                 goto out;
3256
3257         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3258
3259         if (bond->slave_cnt == 0)
3260                 goto re_arm;
3261
3262         if (bond->send_grat_arp) {
3263                 read_lock(&bond->curr_slave_lock);
3264                 bond_send_gratuitous_arp(bond);
3265                 read_unlock(&bond->curr_slave_lock);
3266         }
3267
3268         if (bond->send_unsol_na) {
3269                 read_lock(&bond->curr_slave_lock);
3270                 bond_send_unsolicited_na(bond);
3271                 read_unlock(&bond->curr_slave_lock);
3272         }
3273
3274         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3275                 read_unlock(&bond->lock);
3276                 rtnl_lock();
3277                 read_lock(&bond->lock);
3278
3279                 bond_ab_arp_commit(bond, delta_in_ticks);
3280
3281                 read_unlock(&bond->lock);
3282                 rtnl_unlock();
3283                 read_lock(&bond->lock);
3284         }
3285
3286         bond_ab_arp_probe(bond);
3287
3288 re_arm:
3289         if (bond->params.arp_interval)
3290                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3291 out:
3292         read_unlock(&bond->lock);
3293 }
3294
3295 /*------------------------------ proc/seq_file-------------------------------*/
3296
3297 #ifdef CONFIG_PROC_FS
3298
3299 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3300         __acquires(RCU)
3301         __acquires(&bond->lock)
3302 {
3303         struct bonding *bond = seq->private;
3304         loff_t off = 0;
3305         struct slave *slave;
3306         int i;
3307
3308         /* make sure the bond won't be taken away */
3309         rcu_read_lock();
3310         read_lock(&bond->lock);
3311
3312         if (*pos == 0)
3313                 return SEQ_START_TOKEN;
3314
3315         bond_for_each_slave(bond, slave, i) {
3316                 if (++off == *pos)
3317                         return slave;
3318         }
3319
3320         return NULL;
3321 }
3322
3323 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3324 {
3325         struct bonding *bond = seq->private;
3326         struct slave *slave = v;
3327
3328         ++*pos;
3329         if (v == SEQ_START_TOKEN)
3330                 return bond->first_slave;
3331
3332         slave = slave->next;
3333
3334         return (slave == bond->first_slave) ? NULL : slave;
3335 }
3336
3337 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3338         __releases(&bond->lock)
3339         __releases(RCU)
3340 {
3341         struct bonding *bond = seq->private;
3342
3343         read_unlock(&bond->lock);
3344         rcu_read_unlock();
3345 }
3346
3347 static void bond_info_show_master(struct seq_file *seq)
3348 {
3349         struct bonding *bond = seq->private;
3350         struct slave *curr;
3351         int i;
3352
3353         read_lock(&bond->curr_slave_lock);
3354         curr = bond->curr_active_slave;
3355         read_unlock(&bond->curr_slave_lock);
3356
3357         seq_printf(seq, "Bonding Mode: %s",
3358                    bond_mode_name(bond->params.mode));
3359
3360         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3361             bond->params.fail_over_mac)
3362                 seq_printf(seq, " (fail_over_mac %s)",
3363                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3364
3365         seq_printf(seq, "\n");
3366
3367         if (bond->params.mode == BOND_MODE_XOR ||
3368                 bond->params.mode == BOND_MODE_8023AD) {
3369                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3370                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3371                         bond->params.xmit_policy);
3372         }
3373
3374         if (USES_PRIMARY(bond->params.mode)) {
3375                 seq_printf(seq, "Primary Slave: %s",
3376                            (bond->primary_slave) ?
3377                            bond->primary_slave->dev->name : "None");
3378                 if (bond->primary_slave)
3379                         seq_printf(seq, " (primary_reselect %s)",
3380                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3381
3382                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3383                            (curr) ? curr->dev->name : "None");
3384         }
3385
3386         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3387                    "up" : "down");
3388         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3389         seq_printf(seq, "Up Delay (ms): %d\n",
3390                    bond->params.updelay * bond->params.miimon);
3391         seq_printf(seq, "Down Delay (ms): %d\n",
3392                    bond->params.downdelay * bond->params.miimon);
3393
3394
3395         /* ARP information */
3396         if (bond->params.arp_interval > 0) {
3397                 int printed = 0;
3398                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3399                                 bond->params.arp_interval);
3400
3401                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3402
3403                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3404                         if (!bond->params.arp_targets[i])
3405                                 break;
3406                         if (printed)
3407                                 seq_printf(seq, ",");
3408                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3409                         printed = 1;
3410                 }
3411                 seq_printf(seq, "\n");
3412         }
3413
3414         if (bond->params.mode == BOND_MODE_8023AD) {
3415                 struct ad_info ad_info;
3416
3417                 seq_puts(seq, "\n802.3ad info\n");
3418                 seq_printf(seq, "LACP rate: %s\n",
3419                            (bond->params.lacp_fast) ? "fast" : "slow");
3420                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3421                            ad_select_tbl[bond->params.ad_select].modename);
3422
3423                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3424                         seq_printf(seq, "bond %s has no active aggregator\n",
3425                                    bond->dev->name);
3426                 } else {
3427                         seq_printf(seq, "Active Aggregator Info:\n");
3428
3429                         seq_printf(seq, "\tAggregator ID: %d\n",
3430                                    ad_info.aggregator_id);
3431                         seq_printf(seq, "\tNumber of ports: %d\n",
3432                                    ad_info.ports);
3433                         seq_printf(seq, "\tActor Key: %d\n",
3434                                    ad_info.actor_key);
3435                         seq_printf(seq, "\tPartner Key: %d\n",
3436                                    ad_info.partner_key);
3437                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3438                                    ad_info.partner_system);
3439                 }
3440         }
3441 }
3442
3443 static void bond_info_show_slave(struct seq_file *seq,
3444                                  const struct slave *slave)
3445 {
3446         struct bonding *bond = seq->private;
3447
3448         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3449         seq_printf(seq, "MII Status: %s\n",
3450                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3451         seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3452         seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3453         seq_printf(seq, "Link Failure Count: %u\n",
3454                    slave->link_failure_count);
3455
3456         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3457
3458         if (bond->params.mode == BOND_MODE_8023AD) {
3459                 const struct aggregator *agg
3460                         = SLAVE_AD_INFO(slave).port.aggregator;
3461
3462                 if (agg)
3463                         seq_printf(seq, "Aggregator ID: %d\n",
3464                                    agg->aggregator_identifier);
3465                 else
3466                         seq_puts(seq, "Aggregator ID: N/A\n");
3467         }
3468         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3469 }
3470
3471 static int bond_info_seq_show(struct seq_file *seq, void *v)
3472 {
3473         if (v == SEQ_START_TOKEN) {
3474                 seq_printf(seq, "%s\n", version);
3475                 bond_info_show_master(seq);
3476         } else
3477                 bond_info_show_slave(seq, v);
3478
3479         return 0;
3480 }
3481
3482 static const struct seq_operations bond_info_seq_ops = {
3483         .start = bond_info_seq_start,
3484         .next  = bond_info_seq_next,
3485         .stop  = bond_info_seq_stop,
3486         .show  = bond_info_seq_show,
3487 };
3488
3489 static int bond_info_open(struct inode *inode, struct file *file)
3490 {
3491         struct seq_file *seq;
3492         struct proc_dir_entry *proc;
3493         int res;
3494
3495         res = seq_open(file, &bond_info_seq_ops);
3496         if (!res) {
3497                 /* recover the pointer buried in proc_dir_entry data */
3498                 seq = file->private_data;
3499                 proc = PDE(inode);
3500                 seq->private = proc->data;
3501         }
3502
3503         return res;
3504 }
3505
3506 static const struct file_operations bond_info_fops = {
3507         .owner   = THIS_MODULE,
3508         .open    = bond_info_open,
3509         .read    = seq_read,
3510         .llseek  = seq_lseek,
3511         .release = seq_release,
3512 };
3513
3514 static void bond_create_proc_entry(struct bonding *bond)
3515 {
3516         struct net_device *bond_dev = bond->dev;
3517         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3518
3519         if (bn->proc_dir) {
3520                 bond->proc_entry = proc_create_data(bond_dev->name,
3521                                                     S_IRUGO, bn->proc_dir,
3522                                                     &bond_info_fops, bond);
3523                 if (bond->proc_entry == NULL)
3524                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3525                                    DRV_NAME, bond_dev->name);
3526                 else
3527                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3528         }
3529 }
3530
3531 static void bond_remove_proc_entry(struct bonding *bond)
3532 {
3533         struct net_device *bond_dev = bond->dev;
3534         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3535
3536         if (bn->proc_dir && bond->proc_entry) {
3537                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3538                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3539                 bond->proc_entry = NULL;
3540         }
3541 }
3542
3543 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3544  * Caller must hold rtnl_lock.
3545  */
3546 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3547 {
3548         if (!bn->proc_dir) {
3549                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3550                 if (!bn->proc_dir)
3551                         pr_warning("Warning: cannot create /proc/net/%s\n",
3552                                    DRV_NAME);
3553         }
3554 }
3555
3556 /* Destroy the bonding directory under /proc/net, if empty.
3557  * Caller must hold rtnl_lock.
3558  */
3559 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3560 {
3561         if (bn->proc_dir) {
3562                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3563                 bn->proc_dir = NULL;
3564         }
3565 }
3566
3567 #else /* !CONFIG_PROC_FS */
3568
3569 static void bond_create_proc_entry(struct bonding *bond)
3570 {
3571 }
3572
3573 static void bond_remove_proc_entry(struct bonding *bond)
3574 {
3575 }
3576
3577 static inline void bond_create_proc_dir(struct bond_net *bn)
3578 {
3579 }
3580
3581 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3582 {
3583 }
3584
3585 #endif /* CONFIG_PROC_FS */
3586
3587
3588 /*-------------------------- netdev event handling --------------------------*/
3589
3590 /*
3591  * Change device name
3592  */
3593 static int bond_event_changename(struct bonding *bond)
3594 {
3595         bond_remove_proc_entry(bond);
3596         bond_create_proc_entry(bond);
3597
3598         bond_debug_reregister(bond);
3599
3600         return NOTIFY_DONE;
3601 }
3602
3603 static int bond_master_netdev_event(unsigned long event,
3604                                     struct net_device *bond_dev)
3605 {
3606         struct bonding *event_bond = netdev_priv(bond_dev);
3607
3608         switch (event) {
3609         case NETDEV_CHANGENAME:
3610                 return bond_event_changename(event_bond);
3611         default:
3612                 break;
3613         }
3614
3615         return NOTIFY_DONE;
3616 }
3617
3618 static int bond_slave_netdev_event(unsigned long event,
3619                                    struct net_device *slave_dev)
3620 {
3621         struct net_device *bond_dev = slave_dev->master;
3622         struct bonding *bond = netdev_priv(bond_dev);
3623
3624         switch (event) {
3625         case NETDEV_UNREGISTER:
3626                 if (bond_dev) {
3627                         if (bond->setup_by_slave)
3628                                 bond_release_and_destroy(bond_dev, slave_dev);
3629                         else
3630                                 bond_release(bond_dev, slave_dev);
3631                 }
3632                 break;
3633         case NETDEV_CHANGE:
3634                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3635                         struct slave *slave;
3636
3637                         slave = bond_get_slave_by_dev(bond, slave_dev);
3638                         if (slave) {
3639                                 u16 old_speed = slave->speed;
3640                                 u16 old_duplex = slave->duplex;
3641
3642                                 bond_update_speed_duplex(slave);
3643
3644                                 if (bond_is_lb(bond))
3645                                         break;
3646
3647                                 if (old_speed != slave->speed)
3648                                         bond_3ad_adapter_speed_changed(slave);
3649                                 if (old_duplex != slave->duplex)
3650                                         bond_3ad_adapter_duplex_changed(slave);
3651                         }
3652                 }
3653
3654                 break;
3655         case NETDEV_DOWN:
3656                 /*
3657                  * ... Or is it this?
3658                  */
3659                 break;
3660         case NETDEV_CHANGEMTU:
3661                 /*
3662                  * TODO: Should slaves be allowed to
3663                  * independently alter their MTU?  For
3664                  * an active-backup bond, slaves need
3665                  * not be the same type of device, so
3666                  * MTUs may vary.  For other modes,
3667                  * slaves arguably should have the
3668                  * same MTUs. To do this, we'd need to
3669                  * take over the slave's change_mtu
3670                  * function for the duration of their
3671                  * servitude.
3672                  */
3673                 break;
3674         case NETDEV_CHANGENAME:
3675                 /*
3676                  * TODO: handle changing the primary's name
3677                  */
3678                 break;
3679         case NETDEV_FEAT_CHANGE:
3680                 bond_compute_features(bond);
3681                 break;
3682         default:
3683                 break;
3684         }
3685
3686         return NOTIFY_DONE;
3687 }
3688
3689 /*
3690  * bond_netdev_event: handle netdev notifier chain events.
3691  *
3692  * This function receives events for the netdev chain.  The caller (an
3693  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3694  * locks for us to safely manipulate the slave devices (RTNL lock,
3695  * dev_probe_lock).
3696  */
3697 static int bond_netdev_event(struct notifier_block *this,
3698                              unsigned long event, void *ptr)
3699 {
3700         struct net_device *event_dev = (struct net_device *)ptr;
3701
3702         pr_debug("event_dev: %s, event: %lx\n",
3703                  event_dev ? event_dev->name : "None",
3704                  event);
3705
3706         if (!(event_dev->priv_flags & IFF_BONDING))
3707                 return NOTIFY_DONE;
3708
3709         if (event_dev->flags & IFF_MASTER) {
3710                 pr_debug("IFF_MASTER\n");
3711                 return bond_master_netdev_event(event, event_dev);
3712         }
3713
3714         if (event_dev->flags & IFF_SLAVE) {
3715                 pr_debug("IFF_SLAVE\n");
3716                 return bond_slave_netdev_event(event, event_dev);
3717         }
3718
3719         return NOTIFY_DONE;
3720 }
3721
3722 /*
3723  * bond_inetaddr_event: handle inetaddr notifier chain events.
3724  *
3725  * We keep track of device IPs primarily to use as source addresses in
3726  * ARP monitor probes (rather than spewing out broadcasts all the time).
3727  *
3728  * We track one IP for the main device (if it has one), plus one per VLAN.
3729  */
3730 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3731 {
3732         struct in_ifaddr *ifa = ptr;
3733         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3734         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3735         struct bonding *bond;
3736         struct vlan_entry *vlan;
3737
3738         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3739                 if (bond->dev == event_dev) {
3740                         switch (event) {
3741                         case NETDEV_UP:
3742                                 bond->master_ip = ifa->ifa_local;
3743                                 return NOTIFY_OK;
3744                         case NETDEV_DOWN:
3745                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3746                                 return NOTIFY_OK;
3747                         default:
3748                                 return NOTIFY_DONE;
3749                         }
3750                 }
3751
3752                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3753                         if (!bond->vlgrp)
3754                                 continue;
3755                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3756                         if (vlan_dev == event_dev) {
3757                                 switch (event) {
3758                                 case NETDEV_UP:
3759                                         vlan->vlan_ip = ifa->ifa_local;
3760                                         return NOTIFY_OK;
3761                                 case NETDEV_DOWN:
3762                                         vlan->vlan_ip =
3763                                                 bond_glean_dev_ip(vlan_dev);
3764                                         return NOTIFY_OK;
3765                                 default:
3766                                         return NOTIFY_DONE;
3767                                 }
3768                         }
3769                 }
3770         }
3771         return NOTIFY_DONE;
3772 }
3773
3774 static struct notifier_block bond_netdev_notifier = {
3775         .notifier_call = bond_netdev_event,
3776 };
3777
3778 static struct notifier_block bond_inetaddr_notifier = {
3779         .notifier_call = bond_inetaddr_event,
3780 };
3781
3782 /*-------------------------- Packet type handling ---------------------------*/
3783
3784 /* register to receive lacpdus on a bond */
3785 static void bond_register_lacpdu(struct bonding *bond)
3786 {
3787         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3788
3789         /* initialize packet type */
3790         pk_type->type = PKT_TYPE_LACPDU;
3791         pk_type->dev = bond->dev;
3792         pk_type->func = bond_3ad_lacpdu_recv;
3793
3794         dev_add_pack(pk_type);
3795 }
3796
3797 /* unregister to receive lacpdus on a bond */
3798 static void bond_unregister_lacpdu(struct bonding *bond)
3799 {
3800         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3801 }
3802
3803 void bond_register_arp(struct bonding *bond)
3804 {
3805         struct packet_type *pt = &bond->arp_mon_pt;
3806
3807         if (pt->type)
3808                 return;
3809
3810         pt->type = htons(ETH_P_ARP);
3811         pt->dev = bond->dev;
3812         pt->func = bond_arp_rcv;
3813         dev_add_pack(pt);
3814 }
3815
3816 void bond_unregister_arp(struct bonding *bond)
3817 {
3818         struct packet_type *pt = &bond->arp_mon_pt;
3819
3820         dev_remove_pack(pt);
3821         pt->type = 0;
3822 }
3823
3824 /*---------------------------- Hashing Policies -----------------------------*/
3825
3826 /*
3827  * Hash for the output device based upon layer 2 and layer 3 data. If
3828  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3829  */
3830 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3831 {
3832         struct ethhdr *data = (struct ethhdr *)skb->data;
3833         struct iphdr *iph = ip_hdr(skb);
3834
3835         if (skb->protocol == htons(ETH_P_IP)) {
3836                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3837                         (data->h_dest[5] ^ data->h_source[5])) % count;
3838         }
3839
3840         return (data->h_dest[5] ^ data->h_source[5]) % count;
3841 }
3842
3843 /*
3844  * Hash for the output device based upon layer 3 and layer 4 data. If
3845  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3846  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3847  */
3848 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3849 {
3850         struct ethhdr *data = (struct ethhdr *)skb->data;
3851         struct iphdr *iph = ip_hdr(skb);
3852         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3853         int layer4_xor = 0;
3854
3855         if (skb->protocol == htons(ETH_P_IP)) {
3856                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3857                     (iph->protocol == IPPROTO_TCP ||
3858                      iph->protocol == IPPROTO_UDP)) {
3859                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3860                 }
3861                 return (layer4_xor ^
3862                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3863
3864         }
3865
3866         return (data->h_dest[5] ^ data->h_source[5]) % count;
3867 }
3868
3869 /*
3870  * Hash for the output device based upon layer 2 data
3871  */
3872 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3873 {
3874         struct ethhdr *data = (struct ethhdr *)skb->data;
3875
3876         return (data->h_dest[5] ^ data->h_source[5]) % count;
3877 }
3878
3879 /*-------------------------- Device entry points ----------------------------*/
3880
3881 static int bond_open(struct net_device *bond_dev)
3882 {
3883         struct bonding *bond = netdev_priv(bond_dev);
3884
3885         bond->kill_timers = 0;
3886
3887         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3888
3889         if (bond_is_lb(bond)) {
3890                 /* bond_alb_initialize must be called before the timer
3891                  * is started.
3892                  */
3893                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3894                         /* something went wrong - fail the open operation */
3895                         return -ENOMEM;
3896                 }
3897
3898                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3899                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3900         }
3901
3902         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3903                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3904                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3905         }
3906
3907         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3908                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3909                         INIT_DELAYED_WORK(&bond->arp_work,
3910                                           bond_activebackup_arp_mon);
3911                 else
3912                         INIT_DELAYED_WORK(&bond->arp_work,
3913                                           bond_loadbalance_arp_mon);
3914
3915                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3916                 if (bond->params.arp_validate)
3917                         bond_register_arp(bond);
3918         }
3919
3920         if (bond->params.mode == BOND_MODE_8023AD) {
3921                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3922                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3923                 /* register to receive LACPDUs */
3924                 bond_register_lacpdu(bond);
3925                 bond_3ad_initiate_agg_selection(bond, 1);
3926         }
3927
3928         return 0;
3929 }
3930
3931 static int bond_close(struct net_device *bond_dev)
3932 {
3933         struct bonding *bond = netdev_priv(bond_dev);
3934
3935         if (bond->params.mode == BOND_MODE_8023AD) {
3936                 /* Unregister the receive of LACPDUs */
3937                 bond_unregister_lacpdu(bond);
3938         }
3939
3940         if (bond->params.arp_validate)
3941                 bond_unregister_arp(bond);
3942
3943         write_lock_bh(&bond->lock);
3944
3945         bond->send_grat_arp = 0;
3946         bond->send_unsol_na = 0;
3947
3948         /* signal timers not to re-arm */
3949         bond->kill_timers = 1;
3950
3951         write_unlock_bh(&bond->lock);
3952
3953         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3954                 cancel_delayed_work(&bond->mii_work);
3955         }
3956
3957         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3958                 cancel_delayed_work(&bond->arp_work);
3959         }
3960
3961         switch (bond->params.mode) {
3962         case BOND_MODE_8023AD:
3963                 cancel_delayed_work(&bond->ad_work);
3964                 break;
3965         case BOND_MODE_TLB:
3966         case BOND_MODE_ALB:
3967                 cancel_delayed_work(&bond->alb_work);
3968                 break;
3969         default:
3970                 break;
3971         }
3972
3973         if (delayed_work_pending(&bond->mcast_work))
3974                 cancel_delayed_work(&bond->mcast_work);
3975
3976         if (bond_is_lb(bond)) {
3977                 /* Must be called only after all
3978                  * slaves have been released
3979                  */
3980                 bond_alb_deinitialize(bond);
3981         }
3982
3983         return 0;
3984 }
3985
3986 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3987                                                 struct rtnl_link_stats64 *stats)
3988 {
3989         struct bonding *bond = netdev_priv(bond_dev);
3990         struct rtnl_link_stats64 temp;
3991         struct slave *slave;
3992         int i;
3993
3994         memset(stats, 0, sizeof(*stats));
3995
3996         read_lock_bh(&bond->lock);
3997
3998         bond_for_each_slave(bond, slave, i) {
3999                 const struct rtnl_link_stats64 *sstats =
4000                         dev_get_stats(slave->dev, &temp);
4001
4002                 stats->rx_packets += sstats->rx_packets;
4003                 stats->rx_bytes += sstats->rx_bytes;
4004                 stats->rx_errors += sstats->rx_errors;
4005                 stats->rx_dropped += sstats->rx_dropped;
4006
4007                 stats->tx_packets += sstats->tx_packets;
4008                 stats->tx_bytes += sstats->tx_bytes;
4009                 stats->tx_errors += sstats->tx_errors;
4010                 stats->tx_dropped += sstats->tx_dropped;
4011
4012                 stats->multicast += sstats->multicast;
4013                 stats->collisions += sstats->collisions;
4014
4015                 stats->rx_length_errors += sstats->rx_length_errors;
4016                 stats->rx_over_errors += sstats->rx_over_errors;
4017                 stats->rx_crc_errors += sstats->rx_crc_errors;
4018                 stats->rx_frame_errors += sstats->rx_frame_errors;
4019                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
4020                 stats->rx_missed_errors += sstats->rx_missed_errors;
4021
4022                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
4023                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
4024                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
4025                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
4026                 stats->tx_window_errors += sstats->tx_window_errors;
4027         }
4028
4029         read_unlock_bh(&bond->lock);
4030
4031         return stats;
4032 }
4033
4034 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4035 {
4036         struct net_device *slave_dev = NULL;
4037         struct ifbond k_binfo;
4038         struct ifbond __user *u_binfo = NULL;
4039         struct ifslave k_sinfo;
4040         struct ifslave __user *u_sinfo = NULL;
4041         struct mii_ioctl_data *mii = NULL;
4042         int res = 0;
4043
4044         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
4045
4046         switch (cmd) {
4047         case SIOCGMIIPHY:
4048                 mii = if_mii(ifr);
4049                 if (!mii)
4050                         return -EINVAL;
4051
4052                 mii->phy_id = 0;
4053                 /* Fall Through */
4054         case SIOCGMIIREG:
4055                 /*
4056                  * We do this again just in case we were called by SIOCGMIIREG
4057                  * instead of SIOCGMIIPHY.
4058                  */
4059                 mii = if_mii(ifr);
4060                 if (!mii)
4061                         return -EINVAL;
4062
4063
4064                 if (mii->reg_num == 1) {
4065                         struct bonding *bond = netdev_priv(bond_dev);
4066                         mii->val_out = 0;
4067                         read_lock(&bond->lock);
4068                         read_lock(&bond->curr_slave_lock);
4069                         if (netif_carrier_ok(bond->dev))
4070                                 mii->val_out = BMSR_LSTATUS;
4071
4072                         read_unlock(&bond->curr_slave_lock);
4073                         read_unlock(&bond->lock);
4074                 }
4075
4076                 return 0;
4077         case BOND_INFO_QUERY_OLD:
4078         case SIOCBONDINFOQUERY:
4079                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4080
4081                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4082                         return -EFAULT;
4083
4084                 res = bond_info_query(bond_dev, &k_binfo);
4085                 if (res == 0 &&
4086                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4087                         return -EFAULT;
4088
4089                 return res;
4090         case BOND_SLAVE_INFO_QUERY_OLD:
4091         case SIOCBONDSLAVEINFOQUERY:
4092                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4093
4094                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4095                         return -EFAULT;
4096
4097                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4098                 if (res == 0 &&
4099                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4100                         return -EFAULT;
4101
4102                 return res;
4103         default:
4104                 /* Go on */
4105                 break;
4106         }
4107
4108         if (!capable(CAP_NET_ADMIN))
4109                 return -EPERM;
4110
4111         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4112
4113         pr_debug("slave_dev=%p:\n", slave_dev);
4114
4115         if (!slave_dev)
4116                 res = -ENODEV;
4117         else {
4118                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4119                 switch (cmd) {
4120                 case BOND_ENSLAVE_OLD:
4121                 case SIOCBONDENSLAVE:
4122                         res = bond_enslave(bond_dev, slave_dev);
4123                         break;
4124                 case BOND_RELEASE_OLD:
4125                 case SIOCBONDRELEASE:
4126                         res = bond_release(bond_dev, slave_dev);
4127                         break;
4128                 case BOND_SETHWADDR_OLD:
4129                 case SIOCBONDSETHWADDR:
4130                         res = bond_sethwaddr(bond_dev, slave_dev);
4131                         break;
4132                 case BOND_CHANGE_ACTIVE_OLD:
4133                 case SIOCBONDCHANGEACTIVE:
4134                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4135                         break;
4136                 default:
4137                         res = -EOPNOTSUPP;
4138                 }
4139
4140                 dev_put(slave_dev);
4141         }
4142
4143         return res;
4144 }
4145
4146 static bool bond_addr_in_mc_list(unsigned char *addr,
4147                                  struct netdev_hw_addr_list *list,
4148                                  int addrlen)
4149 {
4150         struct netdev_hw_addr *ha;
4151
4152         netdev_hw_addr_list_for_each(ha, list)
4153                 if (!memcmp(ha->addr, addr, addrlen))
4154                         return true;
4155
4156         return false;
4157 }
4158
4159 static void bond_set_multicast_list(struct net_device *bond_dev)
4160 {
4161         struct bonding *bond = netdev_priv(bond_dev);
4162         struct netdev_hw_addr *ha;
4163         bool found;
4164
4165         /*
4166          * Do promisc before checking multicast_mode
4167          */
4168         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4169                 /*
4170                  * FIXME: Need to handle the error when one of the multi-slaves
4171                  * encounters error.
4172                  */
4173                 bond_set_promiscuity(bond, 1);
4174
4175
4176         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4177                 bond_set_promiscuity(bond, -1);
4178
4179
4180         /* set allmulti flag to slaves */
4181         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4182                 /*
4183                  * FIXME: Need to handle the error when one of the multi-slaves
4184                  * encounters error.
4185                  */
4186                 bond_set_allmulti(bond, 1);
4187
4188
4189         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4190                 bond_set_allmulti(bond, -1);
4191
4192
4193         read_lock(&bond->lock);
4194
4195         bond->flags = bond_dev->flags;
4196
4197         /* looking for addresses to add to slaves' mc list */
4198         netdev_for_each_mc_addr(ha, bond_dev) {
4199                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4200                                              bond_dev->addr_len);
4201                 if (!found)
4202                         bond_mc_add(bond, ha->addr);
4203         }
4204
4205         /* looking for addresses to delete from slaves' list */
4206         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4207                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4208                                              bond_dev->addr_len);
4209                 if (!found)
4210                         bond_mc_del(bond, ha->addr);
4211         }
4212
4213         /* save master's multicast list */
4214         __hw_addr_flush(&bond->mc_list);
4215         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4216                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4217
4218         read_unlock(&bond->lock);
4219 }
4220
4221 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4222 {
4223         struct bonding *bond = netdev_priv(dev);
4224         struct slave *slave = bond->first_slave;
4225
4226         if (slave) {
4227                 const struct net_device_ops *slave_ops
4228                         = slave->dev->netdev_ops;
4229                 if (slave_ops->ndo_neigh_setup)
4230                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4231         }
4232         return 0;
4233 }
4234
4235 /*
4236  * Change the MTU of all of a master's slaves to match the master
4237  */
4238 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4239 {
4240         struct bonding *bond = netdev_priv(bond_dev);
4241         struct slave *slave, *stop_at;
4242         int res = 0;
4243         int i;
4244
4245         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4246                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4247
4248         /* Can't hold bond->lock with bh disabled here since
4249          * some base drivers panic. On the other hand we can't
4250          * hold bond->lock without bh disabled because we'll
4251          * deadlock. The only solution is to rely on the fact
4252          * that we're under rtnl_lock here, and the slaves
4253          * list won't change. This doesn't solve the problem
4254          * of setting the slave's MTU while it is
4255          * transmitting, but the assumption is that the base
4256          * driver can handle that.
4257          *
4258          * TODO: figure out a way to safely iterate the slaves
4259          * list, but without holding a lock around the actual
4260          * call to the base driver.
4261          */
4262
4263         bond_for_each_slave(bond, slave, i) {
4264                 pr_debug("s %p s->p %p c_m %p\n",
4265                          slave,
4266                          slave->prev,
4267                          slave->dev->netdev_ops->ndo_change_mtu);
4268
4269                 res = dev_set_mtu(slave->dev, new_mtu);
4270
4271                 if (res) {
4272                         /* If we failed to set the slave's mtu to the new value
4273                          * we must abort the operation even in ACTIVE_BACKUP
4274                          * mode, because if we allow the backup slaves to have
4275                          * different mtu values than the active slave we'll
4276                          * need to change their mtu when doing a failover. That
4277                          * means changing their mtu from timer context, which
4278                          * is probably not a good idea.
4279                          */
4280                         pr_debug("err %d %s\n", res, slave->dev->name);
4281                         goto unwind;
4282                 }
4283         }
4284
4285         bond_dev->mtu = new_mtu;
4286
4287         return 0;
4288
4289 unwind:
4290         /* unwind from head to the slave that failed */
4291         stop_at = slave;
4292         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4293                 int tmp_res;
4294
4295                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4296                 if (tmp_res) {
4297                         pr_debug("unwind err %d dev %s\n",
4298                                  tmp_res, slave->dev->name);
4299                 }
4300         }
4301
4302         return res;
4303 }
4304
4305 /*
4306  * Change HW address
4307  *
4308  * Note that many devices must be down to change the HW address, and
4309  * downing the master releases all slaves.  We can make bonds full of
4310  * bonding devices to test this, however.
4311  */
4312 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4313 {
4314         struct bonding *bond = netdev_priv(bond_dev);
4315         struct sockaddr *sa = addr, tmp_sa;
4316         struct slave *slave, *stop_at;
4317         int res = 0;
4318         int i;
4319
4320         if (bond->params.mode == BOND_MODE_ALB)
4321                 return bond_alb_set_mac_address(bond_dev, addr);
4322
4323
4324         pr_debug("bond=%p, name=%s\n",
4325                  bond, bond_dev ? bond_dev->name : "None");
4326
4327         /*
4328          * If fail_over_mac is set to active, do nothing and return
4329          * success.  Returning an error causes ifenslave to fail.
4330          */
4331         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4332                 return 0;
4333
4334         if (!is_valid_ether_addr(sa->sa_data))
4335                 return -EADDRNOTAVAIL;
4336
4337         /* Can't hold bond->lock with bh disabled here since
4338          * some base drivers panic. On the other hand we can't
4339          * hold bond->lock without bh disabled because we'll
4340          * deadlock. The only solution is to rely on the fact
4341          * that we're under rtnl_lock here, and the slaves
4342          * list won't change. This doesn't solve the problem
4343          * of setting the slave's hw address while it is
4344          * transmitting, but the assumption is that the base
4345          * driver can handle that.
4346          *
4347          * TODO: figure out a way to safely iterate the slaves
4348          * list, but without holding a lock around the actual
4349          * call to the base driver.
4350          */
4351
4352         bond_for_each_slave(bond, slave, i) {
4353                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4354                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4355
4356                 if (slave_ops->ndo_set_mac_address == NULL) {
4357                         res = -EOPNOTSUPP;
4358                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4359                         goto unwind;
4360                 }
4361
4362                 res = dev_set_mac_address(slave->dev, addr);
4363                 if (res) {
4364                         /* TODO: consider downing the slave
4365                          * and retry ?
4366                          * User should expect communications
4367                          * breakage anyway until ARP finish
4368                          * updating, so...
4369                          */
4370                         pr_debug("err %d %s\n", res, slave->dev->name);
4371                         goto unwind;
4372                 }
4373         }
4374
4375         /* success */
4376         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4377         return 0;
4378
4379 unwind:
4380         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4381         tmp_sa.sa_family = bond_dev->type;
4382
4383         /* unwind from head to the slave that failed */
4384         stop_at = slave;
4385         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4386                 int tmp_res;
4387
4388                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4389                 if (tmp_res) {
4390                         pr_debug("unwind err %d dev %s\n",
4391                                  tmp_res, slave->dev->name);
4392                 }
4393         }
4394
4395         return res;
4396 }
4397
4398 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4399 {
4400         struct bonding *bond = netdev_priv(bond_dev);
4401         struct slave *slave, *start_at;
4402         int i, slave_no, res = 1;
4403         struct iphdr *iph = ip_hdr(skb);
4404
4405         read_lock(&bond->lock);
4406
4407         if (!BOND_IS_OK(bond))
4408                 goto out;
4409         /*
4410          * Start with the curr_active_slave that joined the bond as the
4411          * default for sending IGMP traffic.  For failover purposes one
4412          * needs to maintain some consistency for the interface that will
4413          * send the join/membership reports.  The curr_active_slave found
4414          * will send all of this type of traffic.
4415          */
4416         if ((iph->protocol == IPPROTO_IGMP) &&
4417             (skb->protocol == htons(ETH_P_IP))) {
4418
4419                 read_lock(&bond->curr_slave_lock);
4420                 slave = bond->curr_active_slave;
4421                 read_unlock(&bond->curr_slave_lock);
4422
4423                 if (!slave)
4424                         goto out;
4425         } else {
4426                 /*
4427                  * Concurrent TX may collide on rr_tx_counter; we accept
4428                  * that as being rare enough not to justify using an
4429                  * atomic op here.
4430                  */
4431                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4432
4433                 bond_for_each_slave(bond, slave, i) {
4434                         slave_no--;
4435                         if (slave_no < 0)
4436                                 break;
4437                 }
4438         }
4439
4440         start_at = slave;
4441         bond_for_each_slave_from(bond, slave, i, start_at) {
4442                 if (IS_UP(slave->dev) &&
4443                     (slave->link == BOND_LINK_UP) &&
4444                     (slave->state == BOND_STATE_ACTIVE)) {
4445                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4446                         break;
4447                 }
4448         }
4449
4450 out:
4451         if (res) {
4452                 /* no suitable interface, frame not sent */
4453                 dev_kfree_skb(skb);
4454         }
4455         read_unlock(&bond->lock);
4456         return NETDEV_TX_OK;
4457 }
4458
4459
4460 /*
4461  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4462  * the bond has a usable interface.
4463  */
4464 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4465 {
4466         struct bonding *bond = netdev_priv(bond_dev);
4467         int res = 1;
4468
4469         read_lock(&bond->lock);
4470         read_lock(&bond->curr_slave_lock);
4471
4472         if (!BOND_IS_OK(bond))
4473                 goto out;
4474
4475         if (!bond->curr_active_slave)
4476                 goto out;
4477
4478         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4479
4480 out:
4481         if (res)
4482                 /* no suitable interface, frame not sent */
4483                 dev_kfree_skb(skb);
4484
4485         read_unlock(&bond->curr_slave_lock);
4486         read_unlock(&bond->lock);
4487         return NETDEV_TX_OK;
4488 }
4489
4490 /*
4491  * In bond_xmit_xor() , we determine the output device by using a pre-
4492  * determined xmit_hash_policy(), If the selected device is not enabled,
4493  * find the next active slave.
4494  */
4495 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4496 {
4497         struct bonding *bond = netdev_priv(bond_dev);
4498         struct slave *slave, *start_at;
4499         int slave_no;
4500         int i;
4501         int res = 1;
4502
4503         read_lock(&bond->lock);
4504
4505         if (!BOND_IS_OK(bond))
4506                 goto out;
4507
4508         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4509
4510         bond_for_each_slave(bond, slave, i) {
4511                 slave_no--;
4512                 if (slave_no < 0)
4513                         break;
4514         }
4515
4516         start_at = slave;
4517
4518         bond_for_each_slave_from(bond, slave, i, start_at) {
4519                 if (IS_UP(slave->dev) &&
4520                     (slave->link == BOND_LINK_UP) &&
4521                     (slave->state == BOND_STATE_ACTIVE)) {
4522                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4523                         break;
4524                 }
4525         }
4526
4527 out:
4528         if (res) {
4529                 /* no suitable interface, frame not sent */
4530                 dev_kfree_skb(skb);
4531         }
4532         read_unlock(&bond->lock);
4533         return NETDEV_TX_OK;
4534 }
4535
4536 /*
4537  * in broadcast mode, we send everything to all usable interfaces.
4538  */
4539 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4540 {
4541         struct bonding *bond = netdev_priv(bond_dev);
4542         struct slave *slave, *start_at;
4543         struct net_device *tx_dev = NULL;
4544         int i;
4545         int res = 1;
4546
4547         read_lock(&bond->lock);
4548
4549         if (!BOND_IS_OK(bond))
4550                 goto out;
4551
4552         read_lock(&bond->curr_slave_lock);
4553         start_at = bond->curr_active_slave;
4554         read_unlock(&bond->curr_slave_lock);
4555
4556         if (!start_at)
4557                 goto out;
4558
4559         bond_for_each_slave_from(bond, slave, i, start_at) {
4560                 if (IS_UP(slave->dev) &&
4561                     (slave->link == BOND_LINK_UP) &&
4562                     (slave->state == BOND_STATE_ACTIVE)) {
4563                         if (tx_dev) {
4564                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4565                                 if (!skb2) {
4566                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4567                                                bond_dev->name);
4568                                         continue;
4569                                 }
4570
4571                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4572                                 if (res) {
4573                                         dev_kfree_skb(skb2);
4574                                         continue;
4575                                 }
4576                         }
4577                         tx_dev = slave->dev;
4578                 }
4579         }
4580
4581         if (tx_dev)
4582                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4583
4584 out:
4585         if (res)
4586                 /* no suitable interface, frame not sent */
4587                 dev_kfree_skb(skb);
4588
4589         /* frame sent to all suitable interfaces */
4590         read_unlock(&bond->lock);
4591         return NETDEV_TX_OK;
4592 }
4593
4594 /*------------------------- Device initialization ---------------------------*/
4595
4596 static void bond_set_xmit_hash_policy(struct bonding *bond)
4597 {
4598         switch (bond->params.xmit_policy) {
4599         case BOND_XMIT_POLICY_LAYER23:
4600                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4601                 break;
4602         case BOND_XMIT_POLICY_LAYER34:
4603                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4604                 break;
4605         case BOND_XMIT_POLICY_LAYER2:
4606         default:
4607                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4608                 break;
4609         }
4610 }
4611
4612 /*
4613  * Lookup the slave that corresponds to a qid
4614  */
4615 static inline int bond_slave_override(struct bonding *bond,
4616                                       struct sk_buff *skb)
4617 {
4618         int i, res = 1;
4619         struct slave *slave = NULL;
4620         struct slave *check_slave;
4621
4622         read_lock(&bond->lock);
4623
4624         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4625                 goto out;
4626
4627         /* Find out if any slaves have the same mapping as this skb. */
4628         bond_for_each_slave(bond, check_slave, i) {
4629                 if (check_slave->queue_id == skb->queue_mapping) {
4630                         slave = check_slave;
4631                         break;
4632                 }
4633         }
4634
4635         /* If the slave isn't UP, use default transmit policy. */
4636         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4637             (slave->link == BOND_LINK_UP)) {
4638                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4639         }
4640
4641 out:
4642         read_unlock(&bond->lock);
4643         return res;
4644 }
4645
4646 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4647 {
4648         /*
4649          * This helper function exists to help dev_pick_tx get the correct
4650          * destination queue.  Using a helper function skips the a call to
4651          * skb_tx_hash and will put the skbs in the queue we expect on their
4652          * way down to the bonding driver.
4653          */
4654         return skb->queue_mapping;
4655 }
4656
4657 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4658 {
4659         struct bonding *bond = netdev_priv(dev);
4660
4661         /*
4662          * If we risk deadlock from transmitting this in the
4663          * netpoll path, tell netpoll to queue the frame for later tx
4664          */
4665         if (is_netpoll_tx_blocked(dev))
4666                 return NETDEV_TX_BUSY;
4667
4668         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4669                 if (!bond_slave_override(bond, skb))
4670                         return NETDEV_TX_OK;
4671         }
4672
4673         switch (bond->params.mode) {
4674         case BOND_MODE_ROUNDROBIN:
4675                 return bond_xmit_roundrobin(skb, dev);
4676         case BOND_MODE_ACTIVEBACKUP:
4677                 return bond_xmit_activebackup(skb, dev);
4678         case BOND_MODE_XOR:
4679                 return bond_xmit_xor(skb, dev);
4680         case BOND_MODE_BROADCAST:
4681                 return bond_xmit_broadcast(skb, dev);
4682         case BOND_MODE_8023AD:
4683                 return bond_3ad_xmit_xor(skb, dev);
4684         case BOND_MODE_ALB:
4685         case BOND_MODE_TLB:
4686                 return bond_alb_xmit(skb, dev);
4687         default:
4688                 /* Should never happen, mode already checked */
4689                 pr_err("%s: Error: Unknown bonding mode %d\n",
4690                        dev->name, bond->params.mode);
4691                 WARN_ON_ONCE(1);
4692                 dev_kfree_skb(skb);
4693                 return NETDEV_TX_OK;
4694         }
4695 }
4696
4697
4698 /*
4699  * set bond mode specific net device operations
4700  */
4701 void bond_set_mode_ops(struct bonding *bond, int mode)
4702 {
4703         struct net_device *bond_dev = bond->dev;
4704
4705         switch (mode) {
4706         case BOND_MODE_ROUNDROBIN:
4707                 break;
4708         case BOND_MODE_ACTIVEBACKUP:
4709                 break;
4710         case BOND_MODE_XOR:
4711                 bond_set_xmit_hash_policy(bond);
4712                 break;
4713         case BOND_MODE_BROADCAST:
4714                 break;
4715         case BOND_MODE_8023AD:
4716                 bond_set_master_3ad_flags(bond);
4717                 bond_set_xmit_hash_policy(bond);
4718                 break;
4719         case BOND_MODE_ALB:
4720                 bond_set_master_alb_flags(bond);
4721                 /* FALLTHRU */
4722         case BOND_MODE_TLB:
4723                 break;
4724         default:
4725                 /* Should never happen, mode already checked */
4726                 pr_err("%s: Error: Unknown bonding mode %d\n",
4727                        bond_dev->name, mode);
4728                 break;
4729         }
4730 }
4731
4732 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4733                                     struct ethtool_drvinfo *drvinfo)
4734 {
4735         strncpy(drvinfo->driver, DRV_NAME, 32);
4736         strncpy(drvinfo->version, DRV_VERSION, 32);
4737         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4738 }
4739
4740 static const struct ethtool_ops bond_ethtool_ops = {
4741         .get_drvinfo            = bond_ethtool_get_drvinfo,
4742         .get_link               = ethtool_op_get_link,
4743         .get_tx_csum            = ethtool_op_get_tx_csum,
4744         .get_sg                 = ethtool_op_get_sg,
4745         .get_tso                = ethtool_op_get_tso,
4746         .get_ufo                = ethtool_op_get_ufo,
4747         .get_flags              = ethtool_op_get_flags,
4748 };
4749
4750 static const struct net_device_ops bond_netdev_ops = {
4751         .ndo_init               = bond_init,
4752         .ndo_uninit             = bond_uninit,
4753         .ndo_open               = bond_open,
4754         .ndo_stop               = bond_close,
4755         .ndo_start_xmit         = bond_start_xmit,
4756         .ndo_select_queue       = bond_select_queue,
4757         .ndo_get_stats64        = bond_get_stats,
4758         .ndo_do_ioctl           = bond_do_ioctl,
4759         .ndo_set_multicast_list = bond_set_multicast_list,
4760         .ndo_change_mtu         = bond_change_mtu,
4761         .ndo_set_mac_address    = bond_set_mac_address,
4762         .ndo_neigh_setup        = bond_neigh_setup,
4763         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4764         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4765         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4766 #ifdef CONFIG_NET_POLL_CONTROLLER
4767         .ndo_netpoll_setup      = bond_netpoll_setup,
4768         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4769         .ndo_poll_controller    = bond_poll_controller,
4770 #endif
4771         .ndo_add_slave          = bond_enslave,
4772         .ndo_del_slave          = bond_release,
4773 };
4774
4775 static void bond_destructor(struct net_device *bond_dev)
4776 {
4777         struct bonding *bond = netdev_priv(bond_dev);
4778         if (bond->wq)
4779                 destroy_workqueue(bond->wq);
4780         free_netdev(bond_dev);
4781 }
4782
4783 static void bond_setup(struct net_device *bond_dev)
4784 {
4785         struct bonding *bond = netdev_priv(bond_dev);
4786
4787         /* initialize rwlocks */
4788         rwlock_init(&bond->lock);
4789         rwlock_init(&bond->curr_slave_lock);
4790
4791         bond->params = bonding_defaults;
4792
4793         /* Initialize pointers */
4794         bond->dev = bond_dev;
4795         INIT_LIST_HEAD(&bond->vlan_list);
4796
4797         /* Initialize the device entry points */
4798         ether_setup(bond_dev);
4799         bond_dev->netdev_ops = &bond_netdev_ops;
4800         bond_dev->ethtool_ops = &bond_ethtool_ops;
4801         bond_set_mode_ops(bond, bond->params.mode);
4802
4803         bond_dev->destructor = bond_destructor;
4804
4805         /* Initialize the device options */
4806         bond_dev->tx_queue_len = 0;
4807         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4808         bond_dev->priv_flags |= IFF_BONDING;
4809         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4810
4811         if (bond->params.arp_interval)
4812                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4813
4814         /* At first, we block adding VLANs. That's the only way to
4815          * prevent problems that occur when adding VLANs over an
4816          * empty bond. The block will be removed once non-challenged
4817          * slaves are enslaved.
4818          */
4819         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4820
4821         /* don't acquire bond device's netif_tx_lock when
4822          * transmitting */
4823         bond_dev->features |= NETIF_F_LLTX;
4824
4825         /* By default, we declare the bond to be fully
4826          * VLAN hardware accelerated capable. Special
4827          * care is taken in the various xmit functions
4828          * when there are slaves that are not hw accel
4829          * capable
4830          */
4831         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4832                                NETIF_F_HW_VLAN_RX |
4833                                NETIF_F_HW_VLAN_FILTER);
4834
4835         /* By default, we enable GRO on bonding devices.
4836          * Actual support requires lowlevel drivers are GRO ready.
4837          */
4838         bond_dev->features |= NETIF_F_GRO;
4839 }
4840
4841 static void bond_work_cancel_all(struct bonding *bond)
4842 {
4843         write_lock_bh(&bond->lock);
4844         bond->kill_timers = 1;
4845         write_unlock_bh(&bond->lock);
4846
4847         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4848                 cancel_delayed_work(&bond->mii_work);
4849
4850         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4851                 cancel_delayed_work(&bond->arp_work);
4852
4853         if (bond->params.mode == BOND_MODE_ALB &&
4854             delayed_work_pending(&bond->alb_work))
4855                 cancel_delayed_work(&bond->alb_work);
4856
4857         if (bond->params.mode == BOND_MODE_8023AD &&
4858             delayed_work_pending(&bond->ad_work))
4859                 cancel_delayed_work(&bond->ad_work);
4860
4861         if (delayed_work_pending(&bond->mcast_work))
4862                 cancel_delayed_work(&bond->mcast_work);
4863 }
4864
4865 /*
4866 * Destroy a bonding device.
4867 * Must be under rtnl_lock when this function is called.
4868 */
4869 static void bond_uninit(struct net_device *bond_dev)
4870 {
4871         struct bonding *bond = netdev_priv(bond_dev);
4872         struct vlan_entry *vlan, *tmp;
4873
4874         bond_netpoll_cleanup(bond_dev);
4875
4876         /* Release the bonded slaves */
4877         bond_release_all(bond_dev);
4878
4879         list_del(&bond->bond_list);
4880
4881         bond_work_cancel_all(bond);
4882
4883         bond_remove_proc_entry(bond);
4884
4885         bond_debug_unregister(bond);
4886
4887         __hw_addr_flush(&bond->mc_list);
4888
4889         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4890                 list_del(&vlan->vlan_list);
4891                 kfree(vlan);
4892         }
4893 }
4894
4895 /*------------------------- Module initialization ---------------------------*/
4896
4897 /*
4898  * Convert string input module parms.  Accept either the
4899  * number of the mode or its string name.  A bit complicated because
4900  * some mode names are substrings of other names, and calls from sysfs
4901  * may have whitespace in the name (trailing newlines, for example).
4902  */
4903 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4904 {
4905         int modeint = -1, i, rv;
4906         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4907
4908         for (p = (char *)buf; *p; p++)
4909                 if (!(isdigit(*p) || isspace(*p)))
4910                         break;
4911
4912         if (*p)
4913                 rv = sscanf(buf, "%20s", modestr);
4914         else
4915                 rv = sscanf(buf, "%d", &modeint);
4916
4917         if (!rv)
4918                 return -1;
4919
4920         for (i = 0; tbl[i].modename; i++) {
4921                 if (modeint == tbl[i].mode)
4922                         return tbl[i].mode;
4923                 if (strcmp(modestr, tbl[i].modename) == 0)
4924                         return tbl[i].mode;
4925         }
4926
4927         return -1;
4928 }
4929
4930 static int bond_check_params(struct bond_params *params)
4931 {
4932         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4933
4934         /*
4935          * Convert string parameters.
4936          */
4937         if (mode) {
4938                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4939                 if (bond_mode == -1) {
4940                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4941                                mode == NULL ? "NULL" : mode);
4942                         return -EINVAL;
4943                 }
4944         }
4945
4946         if (xmit_hash_policy) {
4947                 if ((bond_mode != BOND_MODE_XOR) &&
4948                     (bond_mode != BOND_MODE_8023AD)) {
4949                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4950                                bond_mode_name(bond_mode));
4951                 } else {
4952                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4953                                                         xmit_hashtype_tbl);
4954                         if (xmit_hashtype == -1) {
4955                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4956                                        xmit_hash_policy == NULL ? "NULL" :
4957                                        xmit_hash_policy);
4958                                 return -EINVAL;
4959                         }
4960                 }
4961         }
4962
4963         if (lacp_rate) {
4964                 if (bond_mode != BOND_MODE_8023AD) {
4965                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4966                                 bond_mode_name(bond_mode));
4967                 } else {
4968                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4969                         if (lacp_fast == -1) {
4970                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4971                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4972                                 return -EINVAL;
4973                         }
4974                 }
4975         }
4976
4977         if (ad_select) {
4978                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4979                 if (params->ad_select == -1) {
4980                         pr_err("Error: Invalid ad_select \"%s\"\n",
4981                                ad_select == NULL ? "NULL" : ad_select);
4982                         return -EINVAL;
4983                 }
4984
4985                 if (bond_mode != BOND_MODE_8023AD) {
4986                         pr_warning("ad_select param only affects 802.3ad mode\n");
4987                 }
4988         } else {
4989                 params->ad_select = BOND_AD_STABLE;
4990         }
4991
4992         if (max_bonds < 0) {
4993                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4994                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4995                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4996         }
4997
4998         if (miimon < 0) {
4999                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
5000                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
5001                 miimon = BOND_LINK_MON_INTERV;
5002         }
5003
5004         if (updelay < 0) {
5005                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5006                            updelay, INT_MAX);
5007                 updelay = 0;
5008         }
5009
5010         if (downdelay < 0) {
5011                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5012                            downdelay, INT_MAX);
5013                 downdelay = 0;
5014         }
5015
5016         if ((use_carrier != 0) && (use_carrier != 1)) {
5017                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5018                            use_carrier);
5019                 use_carrier = 1;
5020         }
5021
5022         if (num_grat_arp < 0 || num_grat_arp > 255) {
5023                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
5024                            num_grat_arp);
5025                 num_grat_arp = 1;
5026         }
5027
5028         if (num_unsol_na < 0 || num_unsol_na > 255) {
5029                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5030                            num_unsol_na);
5031                 num_unsol_na = 1;
5032         }
5033
5034         /* reset values for 802.3ad */
5035         if (bond_mode == BOND_MODE_8023AD) {
5036                 if (!miimon) {
5037                         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");
5038                         pr_warning("Forcing miimon to 100msec\n");
5039                         miimon = 100;
5040                 }
5041         }
5042
5043         if (tx_queues < 1 || tx_queues > 255) {
5044                 pr_warning("Warning: tx_queues (%d) should be between "
5045                            "1 and 255, resetting to %d\n",
5046                            tx_queues, BOND_DEFAULT_TX_QUEUES);
5047                 tx_queues = BOND_DEFAULT_TX_QUEUES;
5048         }
5049
5050         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5051                 pr_warning("Warning: all_slaves_active module parameter (%d), "
5052                            "not of valid value (0/1), so it was set to "
5053                            "0\n", all_slaves_active);
5054                 all_slaves_active = 0;
5055         }
5056
5057         if (resend_igmp < 0 || resend_igmp > 255) {
5058                 pr_warning("Warning: resend_igmp (%d) should be between "
5059                            "0 and 255, resetting to %d\n",
5060                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5061                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5062         }
5063
5064         /* reset values for TLB/ALB */
5065         if ((bond_mode == BOND_MODE_TLB) ||
5066             (bond_mode == BOND_MODE_ALB)) {
5067                 if (!miimon) {
5068                         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");
5069                         pr_warning("Forcing miimon to 100msec\n");
5070                         miimon = 100;
5071                 }
5072         }
5073
5074         if (bond_mode == BOND_MODE_ALB) {
5075                 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",
5076                           updelay);
5077         }
5078
5079         if (!miimon) {
5080                 if (updelay || downdelay) {
5081                         /* just warn the user the up/down delay will have
5082                          * no effect since miimon is zero...
5083                          */
5084                         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",
5085                                    updelay, downdelay);
5086                 }
5087         } else {
5088                 /* don't allow arp monitoring */
5089                 if (arp_interval) {
5090                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5091                                    miimon, arp_interval);
5092                         arp_interval = 0;
5093                 }
5094
5095                 if ((updelay % miimon) != 0) {
5096                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5097                                    updelay, miimon,
5098                                    (updelay / miimon) * miimon);
5099                 }
5100
5101                 updelay /= miimon;
5102
5103                 if ((downdelay % miimon) != 0) {
5104                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5105                                    downdelay, miimon,
5106                                    (downdelay / miimon) * miimon);
5107                 }
5108
5109                 downdelay /= miimon;
5110         }
5111
5112         if (arp_interval < 0) {
5113                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5114                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5115                 arp_interval = BOND_LINK_ARP_INTERV;
5116         }
5117
5118         for (arp_ip_count = 0;
5119              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5120              arp_ip_count++) {
5121                 /* not complete check, but should be good enough to
5122                    catch mistakes */
5123                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5124                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5125                                    arp_ip_target[arp_ip_count]);
5126                         arp_interval = 0;
5127                 } else {
5128                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5129                         arp_target[arp_ip_count] = ip;
5130                 }
5131         }
5132
5133         if (arp_interval && !arp_ip_count) {
5134                 /* don't allow arping if no arp_ip_target given... */
5135                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5136                            arp_interval);
5137                 arp_interval = 0;
5138         }
5139
5140         if (arp_validate) {
5141                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5142                         pr_err("arp_validate only supported in active-backup mode\n");
5143                         return -EINVAL;
5144                 }
5145                 if (!arp_interval) {
5146                         pr_err("arp_validate requires arp_interval\n");
5147                         return -EINVAL;
5148                 }
5149
5150                 arp_validate_value = bond_parse_parm(arp_validate,
5151                                                      arp_validate_tbl);
5152                 if (arp_validate_value == -1) {
5153                         pr_err("Error: invalid arp_validate \"%s\"\n",
5154                                arp_validate == NULL ? "NULL" : arp_validate);
5155                         return -EINVAL;
5156                 }
5157         } else
5158                 arp_validate_value = 0;
5159
5160         if (miimon) {
5161                 pr_info("MII link monitoring set to %d ms\n", miimon);
5162         } else if (arp_interval) {
5163                 int i;
5164
5165                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5166                         arp_interval,
5167                         arp_validate_tbl[arp_validate_value].modename,
5168                         arp_ip_count);
5169
5170                 for (i = 0; i < arp_ip_count; i++)
5171                         pr_info(" %s", arp_ip_target[i]);
5172
5173                 pr_info("\n");
5174
5175         } else if (max_bonds) {
5176                 /* miimon and arp_interval not set, we need one so things
5177                  * work as expected, see bonding.txt for details
5178                  */
5179                 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");
5180         }
5181
5182         if (primary && !USES_PRIMARY(bond_mode)) {
5183                 /* currently, using a primary only makes sense
5184                  * in active backup, TLB or ALB modes
5185                  */
5186                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5187                            primary, bond_mode_name(bond_mode));
5188                 primary = NULL;
5189         }
5190
5191         if (primary && primary_reselect) {
5192                 primary_reselect_value = bond_parse_parm(primary_reselect,
5193                                                          pri_reselect_tbl);
5194                 if (primary_reselect_value == -1) {
5195                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5196                                primary_reselect ==
5197                                         NULL ? "NULL" : primary_reselect);
5198                         return -EINVAL;
5199                 }
5200         } else {
5201                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5202         }
5203
5204         if (fail_over_mac) {
5205                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5206                                                       fail_over_mac_tbl);
5207                 if (fail_over_mac_value == -1) {
5208                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5209                                arp_validate == NULL ? "NULL" : arp_validate);
5210                         return -EINVAL;
5211                 }
5212
5213                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5214                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5215         } else {
5216                 fail_over_mac_value = BOND_FOM_NONE;
5217         }
5218
5219         /* fill params struct with the proper values */
5220         params->mode = bond_mode;
5221         params->xmit_policy = xmit_hashtype;
5222         params->miimon = miimon;
5223         params->num_grat_arp = num_grat_arp;
5224         params->num_unsol_na = num_unsol_na;
5225         params->arp_interval = arp_interval;
5226         params->arp_validate = arp_validate_value;
5227         params->updelay = updelay;
5228         params->downdelay = downdelay;
5229         params->use_carrier = use_carrier;
5230         params->lacp_fast = lacp_fast;
5231         params->primary[0] = 0;
5232         params->primary_reselect = primary_reselect_value;
5233         params->fail_over_mac = fail_over_mac_value;
5234         params->tx_queues = tx_queues;
5235         params->all_slaves_active = all_slaves_active;
5236         params->resend_igmp = resend_igmp;
5237
5238         if (primary) {
5239                 strncpy(params->primary, primary, IFNAMSIZ);
5240                 params->primary[IFNAMSIZ - 1] = 0;
5241         }
5242
5243         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5244
5245         return 0;
5246 }
5247
5248 static struct lock_class_key bonding_netdev_xmit_lock_key;
5249 static struct lock_class_key bonding_netdev_addr_lock_key;
5250
5251 static void bond_set_lockdep_class_one(struct net_device *dev,
5252                                        struct netdev_queue *txq,
5253                                        void *_unused)
5254 {
5255         lockdep_set_class(&txq->_xmit_lock,
5256                           &bonding_netdev_xmit_lock_key);
5257 }
5258
5259 static void bond_set_lockdep_class(struct net_device *dev)
5260 {
5261         lockdep_set_class(&dev->addr_list_lock,
5262                           &bonding_netdev_addr_lock_key);
5263         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5264 }
5265
5266 /*
5267  * Called from registration process
5268  */
5269 static int bond_init(struct net_device *bond_dev)
5270 {
5271         struct bonding *bond = netdev_priv(bond_dev);
5272         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5273
5274         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5275
5276         bond->wq = create_singlethread_workqueue(bond_dev->name);
5277         if (!bond->wq)
5278                 return -ENOMEM;
5279
5280         bond_set_lockdep_class(bond_dev);
5281
5282         netif_carrier_off(bond_dev);
5283
5284         bond_create_proc_entry(bond);
5285         list_add_tail(&bond->bond_list, &bn->dev_list);
5286
5287         bond_prepare_sysfs_group(bond);
5288
5289         bond_debug_register(bond);
5290
5291         __hw_addr_init(&bond->mc_list);
5292         return 0;
5293 }
5294
5295 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5296 {
5297         if (tb[IFLA_ADDRESS]) {
5298                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5299                         return -EINVAL;
5300                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5301                         return -EADDRNOTAVAIL;
5302         }
5303         return 0;
5304 }
5305
5306 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5307         .kind           = "bond",
5308         .priv_size      = sizeof(struct bonding),
5309         .setup          = bond_setup,
5310         .validate       = bond_validate,
5311 };
5312
5313 /* Create a new bond based on the specified name and bonding parameters.
5314  * If name is NULL, obtain a suitable "bond%d" name for us.
5315  * Caller must NOT hold rtnl_lock; we need to release it here before we
5316  * set up our sysfs entries.
5317  */
5318 int bond_create(struct net *net, const char *name)
5319 {
5320         struct net_device *bond_dev;
5321         int res;
5322
5323         rtnl_lock();
5324
5325         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5326                                 bond_setup, tx_queues);
5327         if (!bond_dev) {
5328                 pr_err("%s: eek! can't alloc netdev!\n", name);
5329                 rtnl_unlock();
5330                 return -ENOMEM;
5331         }
5332
5333         dev_net_set(bond_dev, net);
5334         bond_dev->rtnl_link_ops = &bond_link_ops;
5335
5336         if (!name) {
5337                 res = dev_alloc_name(bond_dev, "bond%d");
5338                 if (res < 0)
5339                         goto out;
5340         } else {
5341                 /*
5342                  * If we're given a name to register
5343                  * we need to ensure that its not already
5344                  * registered
5345                  */
5346                 res = -EEXIST;
5347                 if (__dev_get_by_name(net, name) != NULL)
5348                         goto out;
5349         }
5350
5351         res = register_netdevice(bond_dev);
5352
5353 out:
5354         rtnl_unlock();
5355         if (res < 0)
5356                 bond_destructor(bond_dev);
5357         return res;
5358 }
5359
5360 static int __net_init bond_net_init(struct net *net)
5361 {
5362         struct bond_net *bn = net_generic(net, bond_net_id);
5363
5364         bn->net = net;
5365         INIT_LIST_HEAD(&bn->dev_list);
5366
5367         bond_create_proc_dir(bn);
5368
5369         return 0;
5370 }
5371
5372 static void __net_exit bond_net_exit(struct net *net)
5373 {
5374         struct bond_net *bn = net_generic(net, bond_net_id);
5375
5376         bond_destroy_proc_dir(bn);
5377 }
5378
5379 static struct pernet_operations bond_net_ops = {
5380         .init = bond_net_init,
5381         .exit = bond_net_exit,
5382         .id   = &bond_net_id,
5383         .size = sizeof(struct bond_net),
5384 };
5385
5386 static int __init bonding_init(void)
5387 {
5388         int i;
5389         int res;
5390
5391         pr_info("%s", version);
5392
5393         res = bond_check_params(&bonding_defaults);
5394         if (res)
5395                 goto out;
5396
5397         res = register_pernet_subsys(&bond_net_ops);
5398         if (res)
5399                 goto out;
5400
5401         res = rtnl_link_register(&bond_link_ops);
5402         if (res)
5403                 goto err_link;
5404
5405         bond_create_debugfs();
5406
5407         for (i = 0; i < max_bonds; i++) {
5408                 res = bond_create(&init_net, NULL);
5409                 if (res)
5410                         goto err;
5411         }
5412
5413         res = bond_create_sysfs();
5414         if (res)
5415                 goto err;
5416
5417         register_netdevice_notifier(&bond_netdev_notifier);
5418         register_inetaddr_notifier(&bond_inetaddr_notifier);
5419         bond_register_ipv6_notifier();
5420 out:
5421         return res;
5422 err:
5423         rtnl_link_unregister(&bond_link_ops);
5424 err_link:
5425         unregister_pernet_subsys(&bond_net_ops);
5426         goto out;
5427
5428 }
5429
5430 static void __exit bonding_exit(void)
5431 {
5432         unregister_netdevice_notifier(&bond_netdev_notifier);
5433         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5434         bond_unregister_ipv6_notifier();
5435
5436         bond_destroy_sysfs();
5437         bond_destroy_debugfs();
5438
5439         rtnl_link_unregister(&bond_link_ops);
5440         unregister_pernet_subsys(&bond_net_ops);
5441
5442 #ifdef CONFIG_NET_POLL_CONTROLLER
5443         /*
5444          * Make sure we don't have an imbalance on our netpoll blocking
5445          */
5446         WARN_ON(atomic_read(&netpoll_block_tx));
5447 #endif
5448 }
5449
5450 module_init(bonding_init);
5451 module_exit(bonding_exit);
5452 MODULE_LICENSE("GPL");
5453 MODULE_VERSION(DRV_VERSION);
5454 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5455 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5456 MODULE_ALIAS_RTNL_LINK("bond");