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 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/system.h>
  56 #include <asm/dma.h>
  57 #include <linux/uaccess.h>
  58 #include <linux/errno.h>
  59 #include <linux/netdevice.h>
  60 #include <linux/inetdevice.h>
  61 #include <linux/igmp.h>
  62 #include <linux/etherdevice.h>
  63 #include <linux/skbuff.h>
  64 #include <net/sock.h>
  65 #include <linux/rtnetlink.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include "bonding.h"
  79 #include "bond_3ad.h"
  80 #include "bond_alb.h"
  81
  82 /*---------------------------- Module parameters ----------------------------*/
  83
  84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  85 #define BOND_LINK_MON_INTERV    0
  86 #define BOND_LINK_ARP_INTERV    0
  87
  88 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  89 static int num_grat_arp = 1;
  90 static int num_unsol_na = 1;
  91 static int miimon       = BOND_LINK_MON_INTERV;
  92 static int updelay;
  93 static int downdelay;
  94 static int use_carrier  = 1;
  95 static char *mode;
  96 static char *primary;
  97 static char *lacp_rate;
  98 static char *ad_select;
  99 static char *xmit_hash_policy;
 100 static int arp_interval = BOND_LINK_ARP_INTERV;
 101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 102 static char *arp_validate;
 103 static char *fail_over_mac;
 104 static struct bond_params bonding_defaults;
 105
 106 module_param(max_bonds, int, 0);
 107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 108 module_param(num_grat_arp, int, 0644);
 109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 110 module_param(num_unsol_na, int, 0644);
 111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 112 module_param(miimon, int, 0);
 113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 114 module_param(updelay, int, 0);
 115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 116 module_param(downdelay, int, 0);
 117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 118                             "in milliseconds");
 119 module_param(use_carrier, int, 0);
 120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 121                               "0 for off, 1 for on (default)");
 122 module_param(mode, charp, 0);
 123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 124                        "1 for active-backup, 2 for balance-xor, "
 125                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 126                        "6 for balance-alb");
 127 module_param(primary, charp, 0);
 128 MODULE_PARM_DESC(primary, "Primary network device to use");
 129 module_param(lacp_rate, charp, 0);
 130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 131                             "(slow/fast)");
 132 module_param(ad_select, charp, 0);
 133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 134 module_param(xmit_hash_policy, charp, 0);
 135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 136                                    ", 1 for layer 3+4");
 137 module_param(arp_interval, int, 0);
 138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 139 module_param_array(arp_ip_target, charp, NULL, 0);
 140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 141 module_param(arp_validate, charp, 0);
 142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 143 module_param(fail_over_mac, charp, 0);
 144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 145
 146 /*----------------------------- Global variables ----------------------------*/
 147
 148 static const char * const version =
 149         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 150
 151 LIST_HEAD(bond_dev_list);
 152
 153 #ifdef CONFIG_PROC_FS
 154 static struct proc_dir_entry *bond_proc_dir;
 155 #endif
 156
 157 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 158 static int arp_ip_count;
 159 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 160 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 161 static int lacp_fast;
 162
 163
 164 const struct bond_parm_tbl bond_lacp_tbl[] = {
 165 {       "slow",         AD_LACP_SLOW},
 166 {       "fast",         AD_LACP_FAST},
 167 {       NULL,           -1},
 168 };
 169
 170 const struct bond_parm_tbl bond_mode_tbl[] = {
 171 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 172 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 173 {       "balance-xor",          BOND_MODE_XOR},
 174 {       "broadcast",            BOND_MODE_BROADCAST},
 175 {       "802.3ad",              BOND_MODE_8023AD},
 176 {       "balance-tlb",          BOND_MODE_TLB},
 177 {       "balance-alb",          BOND_MODE_ALB},
 178 {       NULL,                   -1},
 179 };
 180
 181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 182 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 183 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 184 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 185 {       NULL,                   -1},
 186 };
 187
 188 const struct bond_parm_tbl arp_validate_tbl[] = {
 189 {       "none",                 BOND_ARP_VALIDATE_NONE},
 190 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 191 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 192 {       "all",                  BOND_ARP_VALIDATE_ALL},
 193 {       NULL,                   -1},
 194 };
 195
 196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 197 {       "none",                 BOND_FOM_NONE},
 198 {       "active",               BOND_FOM_ACTIVE},
 199 {       "follow",               BOND_FOM_FOLLOW},
 200 {       NULL,                   -1},
 201 };
 202
 203 struct bond_parm_tbl ad_select_tbl[] = {
 204 {       "stable",       BOND_AD_STABLE},
 205 {       "bandwidth",    BOND_AD_BANDWIDTH},
 206 {       "count",        BOND_AD_COUNT},
 207 {       NULL,           -1},
 208 };
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static void bond_send_gratuitous_arp(struct bonding *bond);
 213 static int bond_init(struct net_device *bond_dev);
 214 static void bond_deinit(struct net_device *bond_dev);
 215
 216 /*---------------------------- General routines -----------------------------*/
 217
 218 static const char *bond_mode_name(int mode)
 219 {
 220         static const char *names[] = {
 221                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 222                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 223                 [BOND_MODE_XOR] = "load balancing (xor)",
 224                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 225                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 226                 [BOND_MODE_TLB] = "transmit load balancing",
 227                 [BOND_MODE_ALB] = "adaptive load balancing",
 228         };
 229
 230         if (mode < 0 || mode > BOND_MODE_ALB)
 231                 return "unknown";
 232
 233         return names[mode];
 234 }
 235
 236 /*---------------------------------- VLAN -----------------------------------*/
 237
 238 /**
 239  * bond_add_vlan - add a new vlan id on bond
 240  * @bond: bond that got the notification
 241  * @vlan_id: the vlan id to add
 242  *
 243  * Returns -ENOMEM if allocation failed.
 244  */
 245 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 246 {
 247         struct vlan_entry *vlan;
 248
 249         pr_debug("bond: %s, vlan id %d\n",
 250                 (bond ? bond->dev->name : "None"), vlan_id);
 251
 252         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 253         if (!vlan)
 254                 return -ENOMEM;
 255
 256         INIT_LIST_HEAD(&vlan->vlan_list);
 257         vlan->vlan_id = vlan_id;
 258
 259         write_lock_bh(&bond->lock);
 260
 261         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 262
 263         write_unlock_bh(&bond->lock);
 264
 265         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 266
 267         return 0;
 268 }
 269
 270 /**
 271  * bond_del_vlan - delete a vlan id from bond
 272  * @bond: bond that got the notification
 273  * @vlan_id: the vlan id to delete
 274  *
 275  * returns -ENODEV if @vlan_id was not found in @bond.
 276  */
 277 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 278 {
 279         struct vlan_entry *vlan;
 280         int res = -ENODEV;
 281
 282         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 283
 284         write_lock_bh(&bond->lock);
 285
 286         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 287                 if (vlan->vlan_id == vlan_id) {
 288                         list_del(&vlan->vlan_list);
 289
 290                         if (bond_is_lb(bond))
 291                                 bond_alb_clear_vlan(bond, vlan_id);
 292
 293                         pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
 294                                 bond->dev->name);
 295
 296                         kfree(vlan);
 297
 298                         if (list_empty(&bond->vlan_list) &&
 299                             (bond->slave_cnt == 0)) {
 300                                 /* Last VLAN removed and no slaves, so
 301                                  * restore block on adding VLANs. This will
 302                                  * be removed once new slaves that are not
 303                                  * VLAN challenged will be added.
 304                                  */
 305                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 306                         }
 307
 308                         res = 0;
 309                         goto out;
 310                 }
 311         }
 312
 313         pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
 314                 bond->dev->name);
 315
 316 out:
 317         write_unlock_bh(&bond->lock);
 318         return res;
 319 }
 320
 321 /**
 322  * bond_has_challenged_slaves
 323  * @bond: the bond we're working on
 324  *
 325  * Searches the slave list. Returns 1 if a vlan challenged slave
 326  * was found, 0 otherwise.
 327  *
 328  * Assumes bond->lock is held.
 329  */
 330 static int bond_has_challenged_slaves(struct bonding *bond)
 331 {
 332         struct slave *slave;
 333         int i;
 334
 335         bond_for_each_slave(bond, slave, i) {
 336                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 337                         pr_debug("found VLAN challenged slave - %s\n",
 338                                 slave->dev->name);
 339                         return 1;
 340                 }
 341         }
 342
 343         pr_debug("no VLAN challenged slaves found\n");
 344         return 0;
 345 }
 346
 347 /**
 348  * bond_next_vlan - safely skip to the next item in the vlans list.
 349  * @bond: the bond we're working on
 350  * @curr: item we're advancing from
 351  *
 352  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 353  * or @curr->next otherwise (even if it is @curr itself again).
 354  *
 355  * Caller must hold bond->lock
 356  */
 357 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 358 {
 359         struct vlan_entry *next, *last;
 360
 361         if (list_empty(&bond->vlan_list))
 362                 return NULL;
 363
 364         if (!curr) {
 365                 next = list_entry(bond->vlan_list.next,
 366                                   struct vlan_entry, vlan_list);
 367         } else {
 368                 last = list_entry(bond->vlan_list.prev,
 369                                   struct vlan_entry, vlan_list);
 370                 if (last == curr) {
 371                         next = list_entry(bond->vlan_list.next,
 372                                           struct vlan_entry, vlan_list);
 373                 } else {
 374                         next = list_entry(curr->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 }
 377         }
 378
 379         return next;
 380 }
 381
 382 /**
 383  * bond_dev_queue_xmit - Prepare skb for xmit.
 384  *
 385  * @bond: bond device that got this skb for tx.
 386  * @skb: hw accel VLAN tagged skb to transmit
 387  * @slave_dev: slave that is supposed to xmit this skbuff
 388  *
 389  * When the bond gets an skb to transmit that is
 390  * already hardware accelerated VLAN tagged, and it
 391  * needs to relay this skb to a slave that is not
 392  * hw accel capable, the skb needs to be "unaccelerated",
 393  * i.e. strip the hwaccel tag and re-insert it as part
 394  * of the payload.
 395  */
 396 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 397                         struct net_device *slave_dev)
 398 {
 399         unsigned short uninitialized_var(vlan_id);
 400
 401         if (!list_empty(&bond->vlan_list) &&
 402             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 403             vlan_get_tag(skb, &vlan_id) == 0) {
 404                 skb->dev = slave_dev;
 405                 skb = vlan_put_tag(skb, vlan_id);
 406                 if (!skb) {
 407                         /* vlan_put_tag() frees the skb in case of error,
 408                          * so return success here so the calling functions
 409                          * won't attempt to free is again.
 410                          */
 411                         return 0;
 412                 }
 413         } else {
 414                 skb->dev = slave_dev;
 415         }
 416
 417         skb->priority = 1;
 418         dev_queue_xmit(skb);
 419
 420         return 0;
 421 }
 422
 423 /*
 424  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 425  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 426  * lock because:
 427  * a. This operation is performed in IOCTL context,
 428  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 429  * c. Holding a lock with BH disabled while directly calling a base driver
 430  *    entry point is generally a BAD idea.
 431  *
 432  * The design of synchronization/protection for this operation in the 8021q
 433  * module is good for one or more VLAN devices over a single physical device
 434  * and cannot be extended for a teaming solution like bonding, so there is a
 435  * potential race condition here where a net device from the vlan group might
 436  * be referenced (either by a base driver or the 8021q code) while it is being
 437  * removed from the system. However, it turns out we're not making matters
 438  * worse, and if it works for regular VLAN usage it will work here too.
 439 */
 440
 441 /**
 442  * bond_vlan_rx_register - Propagates registration to slaves
 443  * @bond_dev: bonding net device that got called
 444  * @grp: vlan group being registered
 445  */
 446 static void bond_vlan_rx_register(struct net_device *bond_dev,
 447                                   struct vlan_group *grp)
 448 {
 449         struct bonding *bond = netdev_priv(bond_dev);
 450         struct slave *slave;
 451         int i;
 452
 453         bond->vlgrp = grp;
 454
 455         bond_for_each_slave(bond, slave, i) {
 456                 struct net_device *slave_dev = slave->dev;
 457                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 458
 459                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 460                     slave_ops->ndo_vlan_rx_register) {
 461                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 462                 }
 463         }
 464 }
 465
 466 /**
 467  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 468  * @bond_dev: bonding net device that got called
 469  * @vid: vlan id being added
 470  */
 471 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 472 {
 473         struct bonding *bond = netdev_priv(bond_dev);
 474         struct slave *slave;
 475         int i, res;
 476
 477         bond_for_each_slave(bond, slave, i) {
 478                 struct net_device *slave_dev = slave->dev;
 479                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 480
 481                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 482                     slave_ops->ndo_vlan_rx_add_vid) {
 483                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 484                 }
 485         }
 486
 487         res = bond_add_vlan(bond, vid);
 488         if (res) {
 489                 pr_err(DRV_NAME
 490                        ": %s: Error: Failed to add vlan id %d\n",
 491                        bond_dev->name, vid);
 492         }
 493 }
 494
 495 /**
 496  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 497  * @bond_dev: bonding net device that got called
 498  * @vid: vlan id being removed
 499  */
 500 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 501 {
 502         struct bonding *bond = netdev_priv(bond_dev);
 503         struct slave *slave;
 504         struct net_device *vlan_dev;
 505         int i, res;
 506
 507         bond_for_each_slave(bond, slave, i) {
 508                 struct net_device *slave_dev = slave->dev;
 509                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 510
 511                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 512                     slave_ops->ndo_vlan_rx_kill_vid) {
 513                         /* Save and then restore vlan_dev in the grp array,
 514                          * since the slave's driver might clear it.
 515                          */
 516                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 517                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 518                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 519                 }
 520         }
 521
 522         res = bond_del_vlan(bond, vid);
 523         if (res) {
 524                 pr_err(DRV_NAME
 525                        ": %s: Error: Failed to remove vlan id %d\n",
 526                        bond_dev->name, vid);
 527         }
 528 }
 529
 530 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 531 {
 532         struct vlan_entry *vlan;
 533         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 534
 535         write_lock_bh(&bond->lock);
 536
 537         if (list_empty(&bond->vlan_list))
 538                 goto out;
 539
 540         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 541             slave_ops->ndo_vlan_rx_register)
 542                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 543
 544         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 545             !(slave_ops->ndo_vlan_rx_add_vid))
 546                 goto out;
 547
 548         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 549                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 550
 551 out:
 552         write_unlock_bh(&bond->lock);
 553 }
 554
 555 static void bond_del_vlans_from_slave(struct bonding *bond,
 556                                       struct net_device *slave_dev)
 557 {
 558         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 559         struct vlan_entry *vlan;
 560         struct net_device *vlan_dev;
 561
 562         write_lock_bh(&bond->lock);
 563
 564         if (list_empty(&bond->vlan_list))
 565                 goto out;
 566
 567         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 568             !(slave_ops->ndo_vlan_rx_kill_vid))
 569                 goto unreg;
 570
 571         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 572                 /* Save and then restore vlan_dev in the grp array,
 573                  * since the slave's driver might clear it.
 574                  */
 575                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 576                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 577                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 578         }
 579
 580 unreg:
 581         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 582             slave_ops->ndo_vlan_rx_register)
 583                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 584
 585 out:
 586         write_unlock_bh(&bond->lock);
 587 }
 588
 589 /*------------------------------- Link status -------------------------------*/
 590
 591 /*
 592  * Set the carrier state for the master according to the state of its
 593  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 594  * do special 802.3ad magic.
 595  *
 596  * Returns zero if carrier state does not change, nonzero if it does.
 597  */
 598 static int bond_set_carrier(struct bonding *bond)
 599 {
 600         struct slave *slave;
 601         int i;
 602
 603         if (bond->slave_cnt == 0)
 604                 goto down;
 605
 606         if (bond->params.mode == BOND_MODE_8023AD)
 607                 return bond_3ad_set_carrier(bond);
 608
 609         bond_for_each_slave(bond, slave, i) {
 610                 if (slave->link == BOND_LINK_UP) {
 611                         if (!netif_carrier_ok(bond->dev)) {
 612                                 netif_carrier_on(bond->dev);
 613                                 return 1;
 614                         }
 615                         return 0;
 616                 }
 617         }
 618
 619 down:
 620         if (netif_carrier_ok(bond->dev)) {
 621                 netif_carrier_off(bond->dev);
 622                 return 1;
 623         }
 624         return 0;
 625 }
 626
 627 /*
 628  * Get link speed and duplex from the slave's base driver
 629  * using ethtool. If for some reason the call fails or the
 630  * values are invalid, fake speed and duplex to 100/Full
 631  * and return error.
 632  */
 633 static int bond_update_speed_duplex(struct slave *slave)
 634 {
 635         struct net_device *slave_dev = slave->dev;
 636         struct ethtool_cmd etool;
 637         int res;
 638
 639         /* Fake speed and duplex */
 640         slave->speed = SPEED_100;
 641         slave->duplex = DUPLEX_FULL;
 642
 643         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 644                 return -1;
 645
 646         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 647         if (res < 0)
 648                 return -1;
 649
 650         switch (etool.speed) {
 651         case SPEED_10:
 652         case SPEED_100:
 653         case SPEED_1000:
 654         case SPEED_10000:
 655                 break;
 656         default:
 657                 return -1;
 658         }
 659
 660         switch (etool.duplex) {
 661         case DUPLEX_FULL:
 662         case DUPLEX_HALF:
 663                 break;
 664         default:
 665                 return -1;
 666         }
 667
 668         slave->speed = etool.speed;
 669         slave->duplex = etool.duplex;
 670
 671         return 0;
 672 }
 673
 674 /*
 675  * if <dev> supports MII link status reporting, check its link status.
 676  *
 677  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 678  * depending upon the setting of the use_carrier parameter.
 679  *
 680  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 681  * can't tell and just pretend it is), or 0, meaning that the link is
 682  * down.
 683  *
 684  * If reporting is non-zero, instead of faking link up, return -1 if
 685  * both ETHTOOL and MII ioctls fail (meaning the device does not
 686  * support them).  If use_carrier is set, return whatever it says.
 687  * It'd be nice if there was a good way to tell if a driver supports
 688  * netif_carrier, but there really isn't.
 689  */
 690 static int bond_check_dev_link(struct bonding *bond,
 691                                struct net_device *slave_dev, int reporting)
 692 {
 693         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 694         static int (*ioctl)(struct net_device *, struct ifreq *, int);
 695         struct ifreq ifr;
 696         struct mii_ioctl_data *mii;
 697
 698         if (!reporting && !netif_running(slave_dev))
 699                 return 0;
 700
 701         if (bond->params.use_carrier)
 702                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 703
 704         /* Try to get link status using Ethtool first. */
 705         if (slave_dev->ethtool_ops) {
 706                 if (slave_dev->ethtool_ops->get_link) {
 707                         u32 link;
 708
 709                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 710
 711                         return link ? BMSR_LSTATUS : 0;
 712                 }
 713         }
 714
 715         /* Ethtool can't be used, fallback to MII ioctls. */
 716         ioctl = slave_ops->ndo_do_ioctl;
 717         if (ioctl) {
 718                 /* TODO: set pointer to correct ioctl on a per team member */
 719                 /*       bases to make this more efficient. that is, once  */
 720                 /*       we determine the correct ioctl, we will always    */
 721                 /*       call it and not the others for that team          */
 722                 /*       member.                                           */
 723
 724                 /*
 725                  * We cannot assume that SIOCGMIIPHY will also read a
 726                  * register; not all network drivers (e.g., e100)
 727                  * support that.
 728                  */
 729
 730                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 731                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 732                 mii = if_mii(&ifr);
 733                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 734                         mii->reg_num = MII_BMSR;
 735                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 736                                 return mii->val_out & BMSR_LSTATUS;
 737                 }
 738         }
 739
 740         /*
 741          * If reporting, report that either there's no dev->do_ioctl,
 742          * or both SIOCGMIIREG and get_link failed (meaning that we
 743          * cannot report link status).  If not reporting, pretend
 744          * we're ok.
 745          */
 746         return reporting ? -1 : BMSR_LSTATUS;
 747 }
 748
 749 /*----------------------------- Multicast list ------------------------------*/
 750
 751 /*
 752  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 753  */
 754 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
 755                                    const struct dev_mc_list *dmi2)
 756 {
 757         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 758                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 759 }
 760
 761 /*
 762  * returns dmi entry if found, NULL otherwise
 763  */
 764 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
 765                                                  struct dev_mc_list *mc_list)
 766 {
 767         struct dev_mc_list *idmi;
 768
 769         for (idmi = mc_list; idmi; idmi = idmi->next) {
 770                 if (bond_is_dmi_same(dmi, idmi))
 771                         return idmi;
 772         }
 773
 774         return NULL;
 775 }
 776
 777 /*
 778  * Push the promiscuity flag down to appropriate slaves
 779  */
 780 static int bond_set_promiscuity(struct bonding *bond, int inc)
 781 {
 782         int err = 0;
 783         if (USES_PRIMARY(bond->params.mode)) {
 784                 /* write lock already acquired */
 785                 if (bond->curr_active_slave) {
 786                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 787                                                   inc);
 788                 }
 789         } else {
 790                 struct slave *slave;
 791                 int i;
 792                 bond_for_each_slave(bond, slave, i) {
 793                         err = dev_set_promiscuity(slave->dev, inc);
 794                         if (err)
 795                                 return err;
 796                 }
 797         }
 798         return err;
 799 }
 800
 801 /*
 802  * Push the allmulti flag down to all slaves
 803  */
 804 static int bond_set_allmulti(struct bonding *bond, int inc)
 805 {
 806         int err = 0;
 807         if (USES_PRIMARY(bond->params.mode)) {
 808                 /* write lock already acquired */
 809                 if (bond->curr_active_slave) {
 810                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 811                                                inc);
 812                 }
 813         } else {
 814                 struct slave *slave;
 815                 int i;
 816                 bond_for_each_slave(bond, slave, i) {
 817                         err = dev_set_allmulti(slave->dev, inc);
 818                         if (err)
 819                                 return err;
 820                 }
 821         }
 822         return err;
 823 }
 824
 825 /*
 826  * Add a Multicast address to slaves
 827  * according to mode
 828  */
 829 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 830 {
 831         if (USES_PRIMARY(bond->params.mode)) {
 832                 /* write lock already acquired */
 833                 if (bond->curr_active_slave)
 834                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 835         } else {
 836                 struct slave *slave;
 837                 int i;
 838
 839                 bond_for_each_slave(bond, slave, i)
 840                         dev_mc_add(slave->dev, addr, alen, 0);
 841         }
 842 }
 843
 844 /*
 845  * Remove a multicast address from slave
 846  * according to mode
 847  */
 848 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 849 {
 850         if (USES_PRIMARY(bond->params.mode)) {
 851                 /* write lock already acquired */
 852                 if (bond->curr_active_slave)
 853                         dev_mc_delete(bond->curr_active_slave->dev, addr,
 854                                       alen, 0);
 855         } else {
 856                 struct slave *slave;
 857                 int i;
 858                 bond_for_each_slave(bond, slave, i) {
 859                         dev_mc_delete(slave->dev, addr, alen, 0);
 860                 }
 861         }
 862 }
 863
 864
 865 /*
 866  * Retrieve the list of registered multicast addresses for the bonding
 867  * device and retransmit an IGMP JOIN request to the current active
 868  * slave.
 869  */
 870 static void bond_resend_igmp_join_requests(struct bonding *bond)
 871 {
 872         struct in_device *in_dev;
 873         struct ip_mc_list *im;
 874
 875         rcu_read_lock();
 876         in_dev = __in_dev_get_rcu(bond->dev);
 877         if (in_dev) {
 878                 for (im = in_dev->mc_list; im; im = im->next)
 879                         ip_mc_rejoin_group(im);
 880         }
 881
 882         rcu_read_unlock();
 883 }
 884
 885 /*
 886  * Totally destroys the mc_list in bond
 887  */
 888 static void bond_mc_list_destroy(struct bonding *bond)
 889 {
 890         struct dev_mc_list *dmi;
 891
 892         dmi = bond->mc_list;
 893         while (dmi) {
 894                 bond->mc_list = dmi->next;
 895                 kfree(dmi);
 896                 dmi = bond->mc_list;
 897         }
 898
 899         bond->mc_list = NULL;
 900 }
 901
 902 /*
 903  * Copy all the Multicast addresses from src to the bonding device dst
 904  */
 905 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 906                              gfp_t gfp_flag)
 907 {
 908         struct dev_mc_list *dmi, *new_dmi;
 909
 910         for (dmi = mc_list; dmi; dmi = dmi->next) {
 911                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 912
 913                 if (!new_dmi) {
 914                         /* FIXME: Potential memory leak !!! */
 915                         return -ENOMEM;
 916                 }
 917
 918                 new_dmi->next = bond->mc_list;
 919                 bond->mc_list = new_dmi;
 920                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 921                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 922                 new_dmi->dmi_users = dmi->dmi_users;
 923                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 924         }
 925
 926         return 0;
 927 }
 928
 929 /*
 930  * flush all members of flush->mc_list from device dev->mc_list
 931  */
 932 static void bond_mc_list_flush(struct net_device *bond_dev,
 933                                struct net_device *slave_dev)
 934 {
 935         struct bonding *bond = netdev_priv(bond_dev);
 936         struct dev_mc_list *dmi;
 937
 938         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
 939                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 940
 941         if (bond->params.mode == BOND_MODE_8023AD) {
 942                 /* del lacpdu mc addr from mc list */
 943                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 944
 945                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 946         }
 947 }
 948
 949 /*--------------------------- Active slave change ---------------------------*/
 950
 951 /*
 952  * Update the mc list and multicast-related flags for the new and
 953  * old active slaves (if any) according to the multicast mode, and
 954  * promiscuous flags unconditionally.
 955  */
 956 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 957                          struct slave *old_active)
 958 {
 959         struct dev_mc_list *dmi;
 960
 961         if (!USES_PRIMARY(bond->params.mode))
 962                 /* nothing to do -  mc list is already up-to-date on
 963                  * all slaves
 964                  */
 965                 return;
 966
 967         if (old_active) {
 968                 if (bond->dev->flags & IFF_PROMISC)
 969                         dev_set_promiscuity(old_active->dev, -1);
 970
 971                 if (bond->dev->flags & IFF_ALLMULTI)
 972                         dev_set_allmulti(old_active->dev, -1);
 973
 974                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 975                         dev_mc_delete(old_active->dev, dmi->dmi_addr,
 976                                       dmi->dmi_addrlen, 0);
 977         }
 978
 979         if (new_active) {
 980                 /* FIXME: Signal errors upstream. */
 981                 if (bond->dev->flags & IFF_PROMISC)
 982                         dev_set_promiscuity(new_active->dev, 1);
 983
 984                 if (bond->dev->flags & IFF_ALLMULTI)
 985                         dev_set_allmulti(new_active->dev, 1);
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 988                         dev_mc_add(new_active->dev, dmi->dmi_addr,
 989                                    dmi->dmi_addrlen, 0);
 990                 bond_resend_igmp_join_requests(bond);
 991         }
 992 }
 993
 994 /*
 995  * bond_do_fail_over_mac
 996  *
 997  * Perform special MAC address swapping for fail_over_mac settings
 998  *
 999  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000  */
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002                                   struct slave *new_active,
1003                                   struct slave *old_active)
1004         __releases(&bond->curr_slave_lock)
1005         __releases(&bond->lock)
1006         __acquires(&bond->lock)
1007         __acquires(&bond->curr_slave_lock)
1008 {
1009         u8 tmp_mac[ETH_ALEN];
1010         struct sockaddr saddr;
1011         int rv;
1012
1013         switch (bond->params.fail_over_mac) {
1014         case BOND_FOM_ACTIVE:
1015                 if (new_active)
1016                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1017                                new_active->dev->addr_len);
1018                 break;
1019         case BOND_FOM_FOLLOW:
1020                 /*
1021                  * if new_active && old_active, swap them
1022                  * if just old_active, do nothing (going to no active slave)
1023                  * if just new_active, set new_active to bond's MAC
1024                  */
1025                 if (!new_active)
1026                         return;
1027
1028                 write_unlock_bh(&bond->curr_slave_lock);
1029                 read_unlock(&bond->lock);
1030
1031                 if (old_active) {
1032                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1033                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1034                                ETH_ALEN);
1035                         saddr.sa_family = new_active->dev->type;
1036                 } else {
1037                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1038                         saddr.sa_family = bond->dev->type;
1039                 }
1040
1041                 rv = dev_set_mac_address(new_active->dev, &saddr);
1042                 if (rv) {
1043                         pr_err(DRV_NAME
1044                                ": %s: Error %d setting MAC of slave %s\n",
1045                                bond->dev->name, -rv, new_active->dev->name);
1046                         goto out;
1047                 }
1048
1049                 if (!old_active)
1050                         goto out;
1051
1052                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1053                 saddr.sa_family = old_active->dev->type;
1054
1055                 rv = dev_set_mac_address(old_active->dev, &saddr);
1056                 if (rv)
1057                         pr_err(DRV_NAME
1058                                ": %s: Error %d setting MAC of slave %s\n",
1059                                bond->dev->name, -rv, new_active->dev->name);
1060 out:
1061                 read_lock(&bond->lock);
1062                 write_lock_bh(&bond->curr_slave_lock);
1063                 break;
1064         default:
1065                 pr_err(DRV_NAME
1066                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1067                        bond->dev->name, bond->params.fail_over_mac);
1068                 break;
1069         }
1070
1071 }
1072
1073
1074 /**
1075  * find_best_interface - select the best available slave to be the active one
1076  * @bond: our bonding struct
1077  *
1078  * Warning: Caller must hold curr_slave_lock for writing.
1079  */
1080 static struct slave *bond_find_best_slave(struct bonding *bond)
1081 {
1082         struct slave *new_active, *old_active;
1083         struct slave *bestslave = NULL;
1084         int mintime = bond->params.updelay;
1085         int i;
1086
1087         new_active = old_active = bond->curr_active_slave;
1088
1089         if (!new_active) { /* there were no active slaves left */
1090                 if (bond->slave_cnt > 0)   /* found one slave */
1091                         new_active = bond->first_slave;
1092                 else
1093                         return NULL; /* still no slave, return NULL */
1094         }
1095
1096         /*
1097          * first try the primary link; if arping, a link must tx/rx
1098          * traffic before it can be considered the curr_active_slave.
1099          * also, we would skip slaves between the curr_active_slave
1100          * and primary_slave that may be up and able to arp
1101          */
1102         if ((bond->primary_slave) &&
1103             (!bond->params.arp_interval) &&
1104             (IS_UP(bond->primary_slave->dev))) {
1105                 new_active = bond->primary_slave;
1106         }
1107
1108         /* remember where to stop iterating over the slaves */
1109         old_active = new_active;
1110
1111         bond_for_each_slave_from(bond, new_active, i, old_active) {
1112                 if (IS_UP(new_active->dev)) {
1113                         if (new_active->link == BOND_LINK_UP) {
1114                                 return new_active;
1115                         } else if (new_active->link == BOND_LINK_BACK) {
1116                                 /* link up, but waiting for stabilization */
1117                                 if (new_active->delay < mintime) {
1118                                         mintime = new_active->delay;
1119                                         bestslave = new_active;
1120                                 }
1121                         }
1122                 }
1123         }
1124
1125         return bestslave;
1126 }
1127
1128 /**
1129  * change_active_interface - change the active slave into the specified one
1130  * @bond: our bonding struct
1131  * @new: the new slave to make the active one
1132  *
1133  * Set the new slave to the bond's settings and unset them on the old
1134  * curr_active_slave.
1135  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1136  *
1137  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1138  * because it is apparently the best available slave we have, even though its
1139  * updelay hasn't timed out yet.
1140  *
1141  * If new_active is not NULL, caller must hold bond->lock for read and
1142  * curr_slave_lock for write_bh.
1143  */
1144 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1145 {
1146         struct slave *old_active = bond->curr_active_slave;
1147
1148         if (old_active == new_active)
1149                 return;
1150
1151         if (new_active) {
1152                 new_active->jiffies = jiffies;
1153
1154                 if (new_active->link == BOND_LINK_BACK) {
1155                         if (USES_PRIMARY(bond->params.mode)) {
1156                                 pr_info(DRV_NAME
1157                                        ": %s: making interface %s the new "
1158                                        "active one %d ms earlier.\n",
1159                                        bond->dev->name, new_active->dev->name,
1160                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1161                         }
1162
1163                         new_active->delay = 0;
1164                         new_active->link = BOND_LINK_UP;
1165
1166                         if (bond->params.mode == BOND_MODE_8023AD)
1167                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1168
1169                         if (bond_is_lb(bond))
1170                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1171                 } else {
1172                         if (USES_PRIMARY(bond->params.mode)) {
1173                                 pr_info(DRV_NAME
1174                                        ": %s: making interface %s the new "
1175                                        "active one.\n",
1176                                        bond->dev->name, new_active->dev->name);
1177                         }
1178                 }
1179         }
1180
1181         if (USES_PRIMARY(bond->params.mode))
1182                 bond_mc_swap(bond, new_active, old_active);
1183
1184         if (bond_is_lb(bond)) {
1185                 bond_alb_handle_active_change(bond, new_active);
1186                 if (old_active)
1187                         bond_set_slave_inactive_flags(old_active);
1188                 if (new_active)
1189                         bond_set_slave_active_flags(new_active);
1190         } else {
1191                 bond->curr_active_slave = new_active;
1192         }
1193
1194         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1195                 if (old_active)
1196                         bond_set_slave_inactive_flags(old_active);
1197
1198                 if (new_active) {
1199                         bond_set_slave_active_flags(new_active);
1200
1201                         if (bond->params.fail_over_mac)
1202                                 bond_do_fail_over_mac(bond, new_active,
1203                                                       old_active);
1204
1205                         bond->send_grat_arp = bond->params.num_grat_arp;
1206                         bond_send_gratuitous_arp(bond);
1207
1208                         bond->send_unsol_na = bond->params.num_unsol_na;
1209                         bond_send_unsolicited_na(bond);
1210
1211                         write_unlock_bh(&bond->curr_slave_lock);
1212                         read_unlock(&bond->lock);
1213
1214                         netdev_bonding_change(bond->dev);
1215
1216                         read_lock(&bond->lock);
1217                         write_lock_bh(&bond->curr_slave_lock);
1218                 }
1219         }
1220 }
1221
1222 /**
1223  * bond_select_active_slave - select a new active slave, if needed
1224  * @bond: our bonding struct
1225  *
1226  * This functions should be called when one of the following occurs:
1227  * - The old curr_active_slave has been released or lost its link.
1228  * - The primary_slave has got its link back.
1229  * - A slave has got its link back and there's no old curr_active_slave.
1230  *
1231  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1232  */
1233 void bond_select_active_slave(struct bonding *bond)
1234 {
1235         struct slave *best_slave;
1236         int rv;
1237
1238         best_slave = bond_find_best_slave(bond);
1239         if (best_slave != bond->curr_active_slave) {
1240                 bond_change_active_slave(bond, best_slave);
1241                 rv = bond_set_carrier(bond);
1242                 if (!rv)
1243                         return;
1244
1245                 if (netif_carrier_ok(bond->dev)) {
1246                         pr_info(DRV_NAME
1247                                ": %s: first active interface up!\n",
1248                                bond->dev->name);
1249                 } else {
1250                         pr_info(DRV_NAME ": %s: "
1251                                "now running without any active interface !\n",
1252                                bond->dev->name);
1253                 }
1254         }
1255 }
1256
1257 /*--------------------------- slave list handling ---------------------------*/
1258
1259 /*
1260  * This function attaches the slave to the end of list.
1261  *
1262  * bond->lock held for writing by caller.
1263  */
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265 {
1266         if (bond->first_slave == NULL) { /* attaching the first slave */
1267                 new_slave->next = new_slave;
1268                 new_slave->prev = new_slave;
1269                 bond->first_slave = new_slave;
1270         } else {
1271                 new_slave->next = bond->first_slave;
1272                 new_slave->prev = bond->first_slave->prev;
1273                 new_slave->next->prev = new_slave;
1274                 new_slave->prev->next = new_slave;
1275         }
1276
1277         bond->slave_cnt++;
1278 }
1279
1280 /*
1281  * This function detaches the slave from the list.
1282  * WARNING: no check is made to verify if the slave effectively
1283  * belongs to <bond>.
1284  * Nothing is freed on return, structures are just unchained.
1285  * If any slave pointer in bond was pointing to <slave>,
1286  * it should be changed by the calling function.
1287  *
1288  * bond->lock held for writing by caller.
1289  */
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291 {
1292         if (slave->next)
1293                 slave->next->prev = slave->prev;
1294
1295         if (slave->prev)
1296                 slave->prev->next = slave->next;
1297
1298         if (bond->first_slave == slave) { /* slave is the first slave */
1299                 if (bond->slave_cnt > 1) { /* there are more slave */
1300                         bond->first_slave = slave->next;
1301                 } else {
1302                         bond->first_slave = NULL; /* slave was the last one */
1303                 }
1304         }
1305
1306         slave->next = NULL;
1307         slave->prev = NULL;
1308         bond->slave_cnt--;
1309 }
1310
1311 /*---------------------------------- IOCTL ----------------------------------*/
1312
1313 static int bond_sethwaddr(struct net_device *bond_dev,
1314                           struct net_device *slave_dev)
1315 {
1316         pr_debug("bond_dev=%p\n", bond_dev);
1317         pr_debug("slave_dev=%p\n", slave_dev);
1318         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1319         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1320         return 0;
1321 }
1322
1323 #define BOND_VLAN_FEATURES \
1324         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1325          NETIF_F_HW_VLAN_FILTER)
1326
1327 /*
1328  * Compute the common dev->feature set available to all slaves.  Some
1329  * feature bits are managed elsewhere, so preserve those feature bits
1330  * on the master device.
1331  */
1332 static int bond_compute_features(struct bonding *bond)
1333 {
1334         struct slave *slave;
1335         struct net_device *bond_dev = bond->dev;
1336         unsigned long features = bond_dev->features;
1337         unsigned long vlan_features = 0;
1338         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1339                                                 bond_dev->hard_header_len);
1340         int i;
1341
1342         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1343         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1344
1345         if (!bond->first_slave)
1346                 goto done;
1347
1348         features &= ~NETIF_F_ONE_FOR_ALL;
1349
1350         vlan_features = bond->first_slave->dev->vlan_features;
1351         bond_for_each_slave(bond, slave, i) {
1352                 features = netdev_increment_features(features,
1353                                                      slave->dev->features,
1354                                                      NETIF_F_ONE_FOR_ALL);
1355                 vlan_features = netdev_increment_features(vlan_features,
1356                                                         slave->dev->vlan_features,
1357                                                         NETIF_F_ONE_FOR_ALL);
1358                 if (slave->dev->hard_header_len > max_hard_header_len)
1359                         max_hard_header_len = slave->dev->hard_header_len;
1360         }
1361
1362 done:
1363         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1364         bond_dev->features = netdev_fix_features(features, NULL);
1365         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1366         bond_dev->hard_header_len = max_hard_header_len;
1367
1368         return 0;
1369 }
1370
1371 static void bond_setup_by_slave(struct net_device *bond_dev,
1372                                 struct net_device *slave_dev)
1373 {
1374         struct bonding *bond = netdev_priv(bond_dev);
1375
1376         bond_dev->header_ops        = slave_dev->header_ops;
1377
1378         bond_dev->type              = slave_dev->type;
1379         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1380         bond_dev->addr_len          = slave_dev->addr_len;
1381
1382         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1383                 slave_dev->addr_len);
1384         bond->setup_by_slave = 1;
1385 }
1386
1387 /* enslave device <slave> to bond device <master> */
1388 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1389 {
1390         struct bonding *bond = netdev_priv(bond_dev);
1391         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1392         struct slave *new_slave = NULL;
1393         struct dev_mc_list *dmi;
1394         struct sockaddr addr;
1395         int link_reporting;
1396         int old_features = bond_dev->features;
1397         int res = 0;
1398
1399         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1400                 slave_ops->ndo_do_ioctl == NULL) {
1401                 pr_warning(DRV_NAME
1402                        ": %s: Warning: no link monitoring support for %s\n",
1403                        bond_dev->name, slave_dev->name);
1404         }
1405
1406         /* bond must be initialized by bond_open() before enslaving */
1407         if (!(bond_dev->flags & IFF_UP)) {
1408                 pr_warning(DRV_NAME
1409                         " %s: master_dev is not up in bond_enslave\n",
1410                         bond_dev->name);
1411         }
1412
1413         /* already enslaved */
1414         if (slave_dev->flags & IFF_SLAVE) {
1415                 pr_debug("Error, Device was already enslaved\n");
1416                 return -EBUSY;
1417         }
1418
1419         /* vlan challenged mutual exclusion */
1420         /* no need to lock since we're protected by rtnl_lock */
1421         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1422                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1423                 if (!list_empty(&bond->vlan_list)) {
1424                         pr_err(DRV_NAME
1425                                ": %s: Error: cannot enslave VLAN "
1426                                "challenged slave %s on VLAN enabled "
1427                                "bond %s\n", bond_dev->name, slave_dev->name,
1428                                bond_dev->name);
1429                         return -EPERM;
1430                 } else {
1431                         pr_warning(DRV_NAME
1432                                ": %s: Warning: enslaved VLAN challenged "
1433                                "slave %s. Adding VLANs will be blocked as "
1434                                "long as %s is part of bond %s\n",
1435                                bond_dev->name, slave_dev->name, slave_dev->name,
1436                                bond_dev->name);
1437                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1438                 }
1439         } else {
1440                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441                 if (bond->slave_cnt == 0) {
1442                         /* First slave, and it is not VLAN challenged,
1443                          * so remove the block of adding VLANs over the bond.
1444                          */
1445                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1446                 }
1447         }
1448
1449         /*
1450          * Old ifenslave binaries are no longer supported.  These can
1451          * be identified with moderate accuracy by the state of the slave:
1452          * the current ifenslave will set the interface down prior to
1453          * enslaving it; the old ifenslave will not.
1454          */
1455         if ((slave_dev->flags & IFF_UP)) {
1456                 pr_err(DRV_NAME ": %s is up. "
1457                        "This may be due to an out of date ifenslave.\n",
1458                        slave_dev->name);
1459                 res = -EPERM;
1460                 goto err_undo_flags;
1461         }
1462
1463         /* set bonding device ether type by slave - bonding netdevices are
1464          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1465          * there is a need to override some of the type dependent attribs/funcs.
1466          *
1467          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1468          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1469          */
1470         if (bond->slave_cnt == 0) {
1471                 if (bond_dev->type != slave_dev->type) {
1472                         dev_close(bond_dev);
1473                         pr_debug("%s: change device type from %d to %d\n",
1474                                 bond_dev->name, bond_dev->type, slave_dev->type);
1475                         if (slave_dev->type != ARPHRD_ETHER)
1476                                 bond_setup_by_slave(bond_dev, slave_dev);
1477                         else
1478                                 ether_setup(bond_dev);
1479                         dev_open(bond_dev);
1480                 }
1481         } else if (bond_dev->type != slave_dev->type) {
1482                 pr_err(DRV_NAME ": %s ether type (%d) is different "
1483                         "from other slaves (%d), can not enslave it.\n",
1484                         slave_dev->name,
1485                         slave_dev->type, bond_dev->type);
1486                         res = -EINVAL;
1487                         goto err_undo_flags;
1488         }
1489
1490         if (slave_ops->ndo_set_mac_address == NULL) {
1491                 if (bond->slave_cnt == 0) {
1492                         pr_warning(DRV_NAME
1493                                ": %s: Warning: The first slave device "
1494                                "specified does not support setting the MAC "
1495                                "address. Setting fail_over_mac to active.",
1496                                bond_dev->name);
1497                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1498                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1499                         pr_err(DRV_NAME
1500                                 ": %s: Error: The slave device specified "
1501                                 "does not support setting the MAC address, "
1502                                 "but fail_over_mac is not set to active.\n"
1503                                 , bond_dev->name);
1504                         res = -EOPNOTSUPP;
1505                         goto err_undo_flags;
1506                 }
1507         }
1508
1509         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1510         if (!new_slave) {
1511                 res = -ENOMEM;
1512                 goto err_undo_flags;
1513         }
1514
1515         /* save slave's original flags before calling
1516          * netdev_set_master and dev_open
1517          */
1518         new_slave->original_flags = slave_dev->flags;
1519
1520         /*
1521          * Save slave's original ("permanent") mac address for modes
1522          * that need it, and for restoring it upon release, and then
1523          * set it to the master's address
1524          */
1525         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1526
1527         if (!bond->params.fail_over_mac) {
1528                 /*
1529                  * Set slave to master's mac address.  The application already
1530                  * set the master's mac address to that of the first slave
1531                  */
1532                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1533                 addr.sa_family = slave_dev->type;
1534                 res = dev_set_mac_address(slave_dev, &addr);
1535                 if (res) {
1536                         pr_debug("Error %d calling set_mac_address\n", res);
1537                         goto err_free;
1538                 }
1539         }
1540
1541         res = netdev_set_master(slave_dev, bond_dev);
1542         if (res) {
1543                 pr_debug("Error %d calling netdev_set_master\n", res);
1544                 goto err_restore_mac;
1545         }
1546         /* open the slave since the application closed it */
1547         res = dev_open(slave_dev);
1548         if (res) {
1549                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1550                 goto err_unset_master;
1551         }
1552
1553         new_slave->dev = slave_dev;
1554         slave_dev->priv_flags |= IFF_BONDING;
1555
1556         if (bond_is_lb(bond)) {
1557                 /* bond_alb_init_slave() must be called before all other stages since
1558                  * it might fail and we do not want to have to undo everything
1559                  */
1560                 res = bond_alb_init_slave(bond, new_slave);
1561                 if (res)
1562                         goto err_close;
1563         }
1564
1565         /* If the mode USES_PRIMARY, then the new slave gets the
1566          * master's promisc (and mc) settings only if it becomes the
1567          * curr_active_slave, and that is taken care of later when calling
1568          * bond_change_active()
1569          */
1570         if (!USES_PRIMARY(bond->params.mode)) {
1571                 /* set promiscuity level to new slave */
1572                 if (bond_dev->flags & IFF_PROMISC) {
1573                         res = dev_set_promiscuity(slave_dev, 1);
1574                         if (res)
1575                                 goto err_close;
1576                 }
1577
1578                 /* set allmulti level to new slave */
1579                 if (bond_dev->flags & IFF_ALLMULTI) {
1580                         res = dev_set_allmulti(slave_dev, 1);
1581                         if (res)
1582                                 goto err_close;
1583                 }
1584
1585                 netif_addr_lock_bh(bond_dev);
1586                 /* upload master's mc_list to new slave */
1587                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1588                         dev_mc_add(slave_dev, dmi->dmi_addr,
1589                                    dmi->dmi_addrlen, 0);
1590                 netif_addr_unlock_bh(bond_dev);
1591         }
1592
1593         if (bond->params.mode == BOND_MODE_8023AD) {
1594                 /* add lacpdu mc addr to mc list */
1595                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1596
1597                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1598         }
1599
1600         bond_add_vlans_on_slave(bond, slave_dev);
1601
1602         write_lock_bh(&bond->lock);
1603
1604         bond_attach_slave(bond, new_slave);
1605
1606         new_slave->delay = 0;
1607         new_slave->link_failure_count = 0;
1608
1609         bond_compute_features(bond);
1610
1611         write_unlock_bh(&bond->lock);
1612
1613         read_lock(&bond->lock);
1614
1615         new_slave->last_arp_rx = jiffies;
1616
1617         if (bond->params.miimon && !bond->params.use_carrier) {
1618                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1619
1620                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1621                         /*
1622                          * miimon is set but a bonded network driver
1623                          * does not support ETHTOOL/MII and
1624                          * arp_interval is not set.  Note: if
1625                          * use_carrier is enabled, we will never go
1626                          * here (because netif_carrier is always
1627                          * supported); thus, we don't need to change
1628                          * the messages for netif_carrier.
1629                          */
1630                         pr_warning(DRV_NAME
1631                                ": %s: Warning: MII and ETHTOOL support not "
1632                                "available for interface %s, and "
1633                                "arp_interval/arp_ip_target module parameters "
1634                                "not specified, thus bonding will not detect "
1635                                "link failures! see bonding.txt for details.\n",
1636                                bond_dev->name, slave_dev->name);
1637                 } else if (link_reporting == -1) {
1638                         /* unable get link status using mii/ethtool */
1639                         pr_warning(DRV_NAME
1640                                ": %s: Warning: can't get link status from "
1641                                "interface %s; the network driver associated "
1642                                "with this interface does not support MII or "
1643                                "ETHTOOL link status reporting, thus miimon "
1644                                "has no effect on this interface.\n",
1645                                bond_dev->name, slave_dev->name);
1646                 }
1647         }
1648
1649         /* check for initial state */
1650         if (!bond->params.miimon ||
1651             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1652                 if (bond->params.updelay) {
1653                         pr_debug("Initial state of slave_dev is "
1654                                 "BOND_LINK_BACK\n");
1655                         new_slave->link  = BOND_LINK_BACK;
1656                         new_slave->delay = bond->params.updelay;
1657                 } else {
1658                         pr_debug("Initial state of slave_dev is "
1659                                 "BOND_LINK_UP\n");
1660                         new_slave->link  = BOND_LINK_UP;
1661                 }
1662                 new_slave->jiffies = jiffies;
1663         } else {
1664                 pr_debug("Initial state of slave_dev is "
1665                         "BOND_LINK_DOWN\n");
1666                 new_slave->link  = BOND_LINK_DOWN;
1667         }
1668
1669         if (bond_update_speed_duplex(new_slave) &&
1670             (new_slave->link != BOND_LINK_DOWN)) {
1671                 pr_warning(DRV_NAME
1672                        ": %s: Warning: failed to get speed and duplex from %s, "
1673                        "assumed to be 100Mb/sec and Full.\n",
1674                        bond_dev->name, new_slave->dev->name);
1675
1676                 if (bond->params.mode == BOND_MODE_8023AD) {
1677                         pr_warning(DRV_NAME
1678                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1679                                "support in base driver for proper aggregator "
1680                                "selection.\n", bond_dev->name);
1681                 }
1682         }
1683
1684         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1685                 /* if there is a primary slave, remember it */
1686                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1687                         bond->primary_slave = new_slave;
1688         }
1689
1690         write_lock_bh(&bond->curr_slave_lock);
1691
1692         switch (bond->params.mode) {
1693         case BOND_MODE_ACTIVEBACKUP:
1694                 bond_set_slave_inactive_flags(new_slave);
1695                 bond_select_active_slave(bond);
1696                 break;
1697         case BOND_MODE_8023AD:
1698                 /* in 802.3ad mode, the internal mechanism
1699                  * will activate the slaves in the selected
1700                  * aggregator
1701                  */
1702                 bond_set_slave_inactive_flags(new_slave);
1703                 /* if this is the first slave */
1704                 if (bond->slave_cnt == 1) {
1705                         SLAVE_AD_INFO(new_slave).id = 1;
1706                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1707                          * can be called only after the mac address of the bond is set
1708                          */
1709                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1710                                             bond->params.lacp_fast);
1711                 } else {
1712                         SLAVE_AD_INFO(new_slave).id =
1713                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1714                 }
1715
1716                 bond_3ad_bind_slave(new_slave);
1717                 break;
1718         case BOND_MODE_TLB:
1719         case BOND_MODE_ALB:
1720                 new_slave->state = BOND_STATE_ACTIVE;
1721                 bond_set_slave_inactive_flags(new_slave);
1722                 bond_select_active_slave(bond);
1723                 break;
1724         default:
1725                 pr_debug("This slave is always active in trunk mode\n");
1726
1727                 /* always active in trunk mode */
1728                 new_slave->state = BOND_STATE_ACTIVE;
1729
1730                 /* In trunking mode there is little meaning to curr_active_slave
1731                  * anyway (it holds no special properties of the bond device),
1732                  * so we can change it without calling change_active_interface()
1733                  */
1734                 if (!bond->curr_active_slave)
1735                         bond->curr_active_slave = new_slave;
1736
1737                 break;
1738         } /* switch(bond_mode) */
1739
1740         write_unlock_bh(&bond->curr_slave_lock);
1741
1742         bond_set_carrier(bond);
1743
1744         read_unlock(&bond->lock);
1745
1746         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1747         if (res)
1748                 goto err_close;
1749
1750         pr_info(DRV_NAME
1751                ": %s: enslaving %s as a%s interface with a%s link.\n",
1752                bond_dev->name, slave_dev->name,
1753                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1754                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1755
1756         /* enslave is successful */
1757         return 0;
1758
1759 /* Undo stages on error */
1760 err_close:
1761         dev_close(slave_dev);
1762
1763 err_unset_master:
1764         netdev_set_master(slave_dev, NULL);
1765
1766 err_restore_mac:
1767         if (!bond->params.fail_over_mac) {
1768                 /* XXX TODO - fom follow mode needs to change master's
1769                  * MAC if this slave's MAC is in use by the bond, or at
1770                  * least print a warning.
1771                  */
1772                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1773                 addr.sa_family = slave_dev->type;
1774                 dev_set_mac_address(slave_dev, &addr);
1775         }
1776
1777 err_free:
1778         kfree(new_slave);
1779
1780 err_undo_flags:
1781         bond_dev->features = old_features;
1782
1783         return res;
1784 }
1785
1786 /*
1787  * Try to release the slave device <slave> from the bond device <master>
1788  * It is legal to access curr_active_slave without a lock because all the function
1789  * is write-locked.
1790  *
1791  * The rules for slave state should be:
1792  *   for Active/Backup:
1793  *     Active stays on all backups go down
1794  *   for Bonded connections:
1795  *     The first up interface should be left on and all others downed.
1796  */
1797 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1798 {
1799         struct bonding *bond = netdev_priv(bond_dev);
1800         struct slave *slave, *oldcurrent;
1801         struct sockaddr addr;
1802
1803         /* slave is not a slave or master is not master of this slave */
1804         if (!(slave_dev->flags & IFF_SLAVE) ||
1805             (slave_dev->master != bond_dev)) {
1806                 pr_err(DRV_NAME
1807                        ": %s: Error: cannot release %s.\n",
1808                        bond_dev->name, slave_dev->name);
1809                 return -EINVAL;
1810         }
1811
1812         write_lock_bh(&bond->lock);
1813
1814         slave = bond_get_slave_by_dev(bond, slave_dev);
1815         if (!slave) {
1816                 /* not a slave of this bond */
1817                 pr_info(DRV_NAME
1818                        ": %s: %s not enslaved\n",
1819                        bond_dev->name, slave_dev->name);
1820                 write_unlock_bh(&bond->lock);
1821                 return -EINVAL;
1822         }
1823
1824         if (!bond->params.fail_over_mac) {
1825                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr)
1826                     && bond->slave_cnt > 1)
1827                         pr_warning(DRV_NAME
1828                                ": %s: Warning: the permanent HWaddr of %s - "
1829                                "%pM - is still in use by %s. "
1830                                "Set the HWaddr of %s to a different address "
1831                                "to avoid conflicts.\n",
1832                                bond_dev->name, slave_dev->name,
1833                                slave->perm_hwaddr,
1834                                bond_dev->name, slave_dev->name);
1835         }
1836
1837         /* Inform AD package of unbinding of slave. */
1838         if (bond->params.mode == BOND_MODE_8023AD) {
1839                 /* must be called before the slave is
1840                  * detached from the list
1841                  */
1842                 bond_3ad_unbind_slave(slave);
1843         }
1844
1845         pr_info(DRV_NAME
1846                ": %s: releasing %s interface %s\n",
1847                bond_dev->name,
1848                (slave->state == BOND_STATE_ACTIVE)
1849                ? "active" : "backup",
1850                slave_dev->name);
1851
1852         oldcurrent = bond->curr_active_slave;
1853
1854         bond->current_arp_slave = NULL;
1855
1856         /* release the slave from its bond */
1857         bond_detach_slave(bond, slave);
1858
1859         bond_compute_features(bond);
1860
1861         if (bond->primary_slave == slave)
1862                 bond->primary_slave = NULL;
1863
1864         if (oldcurrent == slave)
1865                 bond_change_active_slave(bond, NULL);
1866
1867         if (bond_is_lb(bond)) {
1868                 /* Must be called only after the slave has been
1869                  * detached from the list and the curr_active_slave
1870                  * has been cleared (if our_slave == old_current),
1871                  * but before a new active slave is selected.
1872                  */
1873                 write_unlock_bh(&bond->lock);
1874                 bond_alb_deinit_slave(bond, slave);
1875                 write_lock_bh(&bond->lock);
1876         }
1877
1878         if (oldcurrent == slave) {
1879                 /*
1880                  * Note that we hold RTNL over this sequence, so there
1881                  * is no concern that another slave add/remove event
1882                  * will interfere.
1883                  */
1884                 write_unlock_bh(&bond->lock);
1885                 read_lock(&bond->lock);
1886                 write_lock_bh(&bond->curr_slave_lock);
1887
1888                 bond_select_active_slave(bond);
1889
1890                 write_unlock_bh(&bond->curr_slave_lock);
1891                 read_unlock(&bond->lock);
1892                 write_lock_bh(&bond->lock);
1893         }
1894
1895         if (bond->slave_cnt == 0) {
1896                 bond_set_carrier(bond);
1897
1898                 /* if the last slave was removed, zero the mac address
1899                  * of the master so it will be set by the application
1900                  * to the mac address of the first slave
1901                  */
1902                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1903
1904                 if (list_empty(&bond->vlan_list)) {
1905                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1906                 } else {
1907                         pr_warning(DRV_NAME
1908                                ": %s: Warning: clearing HW address of %s while it "
1909                                "still has VLANs.\n",
1910                                bond_dev->name, bond_dev->name);
1911                         pr_warning(DRV_NAME
1912                                ": %s: When re-adding slaves, make sure the bond's "
1913                                "HW address matches its VLANs'.\n",
1914                                bond_dev->name);
1915                 }
1916         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1917                    !bond_has_challenged_slaves(bond)) {
1918                 pr_info(DRV_NAME
1919                        ": %s: last VLAN challenged slave %s "
1920                        "left bond %s. VLAN blocking is removed\n",
1921                        bond_dev->name, slave_dev->name, bond_dev->name);
1922                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1923         }
1924
1925         write_unlock_bh(&bond->lock);
1926
1927         /* must do this from outside any spinlocks */
1928         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1929
1930         bond_del_vlans_from_slave(bond, slave_dev);
1931
1932         /* If the mode USES_PRIMARY, then we should only remove its
1933          * promisc and mc settings if it was the curr_active_slave, but that was
1934          * already taken care of above when we detached the slave
1935          */
1936         if (!USES_PRIMARY(bond->params.mode)) {
1937                 /* unset promiscuity level from slave */
1938                 if (bond_dev->flags & IFF_PROMISC)
1939                         dev_set_promiscuity(slave_dev, -1);
1940
1941                 /* unset allmulti level from slave */
1942                 if (bond_dev->flags & IFF_ALLMULTI)
1943                         dev_set_allmulti(slave_dev, -1);
1944
1945                 /* flush master's mc_list from slave */
1946                 netif_addr_lock_bh(bond_dev);
1947                 bond_mc_list_flush(bond_dev, slave_dev);
1948                 netif_addr_unlock_bh(bond_dev);
1949         }
1950
1951         netdev_set_master(slave_dev, NULL);
1952
1953         /* close slave before restoring its mac address */
1954         dev_close(slave_dev);
1955
1956         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1957                 /* restore original ("permanent") mac address */
1958                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1959                 addr.sa_family = slave_dev->type;
1960                 dev_set_mac_address(slave_dev, &addr);
1961         }
1962
1963         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1964                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1965                                    IFF_SLAVE_NEEDARP);
1966
1967         kfree(slave);
1968
1969         return 0;  /* deletion OK */
1970 }
1971
1972 /*
1973 * Destroy a bonding device.
1974 * Must be under rtnl_lock when this function is called.
1975 */
1976 static void bond_uninit(struct net_device *bond_dev)
1977 {
1978         struct bonding *bond = netdev_priv(bond_dev);
1979
1980         bond_deinit(bond_dev);
1981         bond_destroy_sysfs_entry(bond);
1982
1983         if (bond->wq)
1984                 destroy_workqueue(bond->wq);
1985
1986         netif_addr_lock_bh(bond_dev);
1987         bond_mc_list_destroy(bond);
1988         netif_addr_unlock_bh(bond_dev);
1989 }
1990
1991 /*
1992 * First release a slave and than destroy the bond if no more slaves are left.
1993 * Must be under rtnl_lock when this function is called.
1994 */
1995 int  bond_release_and_destroy(struct net_device *bond_dev,
1996                               struct net_device *slave_dev)
1997 {
1998         struct bonding *bond = netdev_priv(bond_dev);
1999         int ret;
2000
2001         ret = bond_release(bond_dev, slave_dev);
2002         if ((ret == 0) && (bond->slave_cnt == 0)) {
2003                 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
2004                        bond_dev->name, bond_dev->name);
2005                 unregister_netdevice(bond_dev);
2006         }
2007         return ret;
2008 }
2009
2010 /*
2011  * This function releases all slaves.
2012  */
2013 static int bond_release_all(struct net_device *bond_dev)
2014 {
2015         struct bonding *bond = netdev_priv(bond_dev);
2016         struct slave *slave;
2017         struct net_device *slave_dev;
2018         struct sockaddr addr;
2019
2020         write_lock_bh(&bond->lock);
2021
2022         netif_carrier_off(bond_dev);
2023
2024         if (bond->slave_cnt == 0)
2025                 goto out;
2026
2027         bond->current_arp_slave = NULL;
2028         bond->primary_slave = NULL;
2029         bond_change_active_slave(bond, NULL);
2030
2031         while ((slave = bond->first_slave) != NULL) {
2032                 /* Inform AD package of unbinding of slave
2033                  * before slave is detached from the list.
2034                  */
2035                 if (bond->params.mode == BOND_MODE_8023AD)
2036                         bond_3ad_unbind_slave(slave);
2037
2038                 slave_dev = slave->dev;
2039                 bond_detach_slave(bond, slave);
2040
2041                 /* now that the slave is detached, unlock and perform
2042                  * all the undo steps that should not be called from
2043                  * within a lock.
2044                  */
2045                 write_unlock_bh(&bond->lock);
2046
2047                 if (bond_is_lb(bond)) {
2048                         /* must be called only after the slave
2049                          * has been detached from the list
2050                          */
2051                         bond_alb_deinit_slave(bond, slave);
2052                 }
2053
2054                 bond_compute_features(bond);
2055
2056                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2057                 bond_del_vlans_from_slave(bond, slave_dev);
2058
2059                 /* If the mode USES_PRIMARY, then we should only remove its
2060                  * promisc and mc settings if it was the curr_active_slave, but that was
2061                  * already taken care of above when we detached the slave
2062                  */
2063                 if (!USES_PRIMARY(bond->params.mode)) {
2064                         /* unset promiscuity level from slave */
2065                         if (bond_dev->flags & IFF_PROMISC)
2066                                 dev_set_promiscuity(slave_dev, -1);
2067
2068                         /* unset allmulti level from slave */
2069                         if (bond_dev->flags & IFF_ALLMULTI)
2070                                 dev_set_allmulti(slave_dev, -1);
2071
2072                         /* flush master's mc_list from slave */
2073                         netif_addr_lock_bh(bond_dev);
2074                         bond_mc_list_flush(bond_dev, slave_dev);
2075                         netif_addr_unlock_bh(bond_dev);
2076                 }
2077
2078                 netdev_set_master(slave_dev, NULL);
2079
2080                 /* close slave before restoring its mac address */
2081                 dev_close(slave_dev);
2082
2083                 if (!bond->params.fail_over_mac) {
2084                         /* restore original ("permanent") mac address*/
2085                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2086                         addr.sa_family = slave_dev->type;
2087                         dev_set_mac_address(slave_dev, &addr);
2088                 }
2089
2090                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2091                                            IFF_SLAVE_INACTIVE);
2092
2093                 kfree(slave);
2094
2095                 /* re-acquire the lock before getting the next slave */
2096                 write_lock_bh(&bond->lock);
2097         }
2098
2099         /* zero the mac address of the master so it will be
2100          * set by the application to the mac address of the
2101          * first slave
2102          */
2103         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2104
2105         if (list_empty(&bond->vlan_list))
2106                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2107         else {
2108                 pr_warning(DRV_NAME
2109                        ": %s: Warning: clearing HW address of %s while it "
2110                        "still has VLANs.\n",
2111                        bond_dev->name, bond_dev->name);
2112                 pr_warning(DRV_NAME
2113                        ": %s: When re-adding slaves, make sure the bond's "
2114                        "HW address matches its VLANs'.\n",
2115                        bond_dev->name);
2116         }
2117
2118         pr_info(DRV_NAME
2119                ": %s: released all slaves\n",
2120                bond_dev->name);
2121
2122 out:
2123         write_unlock_bh(&bond->lock);
2124
2125         return 0;
2126 }
2127
2128 /*
2129  * This function changes the active slave to slave <slave_dev>.
2130  * It returns -EINVAL in the following cases.
2131  *  - <slave_dev> is not found in the list.
2132  *  - There is not active slave now.
2133  *  - <slave_dev> is already active.
2134  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2135  *  - <slave_dev> is not running.
2136  * In these cases, this function does nothing.
2137  * In the other cases, current_slave pointer is changed and 0 is returned.
2138  */
2139 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2140 {
2141         struct bonding *bond = netdev_priv(bond_dev);
2142         struct slave *old_active = NULL;
2143         struct slave *new_active = NULL;
2144         int res = 0;
2145
2146         if (!USES_PRIMARY(bond->params.mode))
2147                 return -EINVAL;
2148
2149         /* Verify that master_dev is indeed the master of slave_dev */
2150         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2151                 return -EINVAL;
2152
2153         read_lock(&bond->lock);
2154
2155         read_lock(&bond->curr_slave_lock);
2156         old_active = bond->curr_active_slave;
2157         read_unlock(&bond->curr_slave_lock);
2158
2159         new_active = bond_get_slave_by_dev(bond, slave_dev);
2160
2161         /*
2162          * Changing to the current active: do nothing; return success.
2163          */
2164         if (new_active && (new_active == old_active)) {
2165                 read_unlock(&bond->lock);
2166                 return 0;
2167         }
2168
2169         if ((new_active) &&
2170             (old_active) &&
2171             (new_active->link == BOND_LINK_UP) &&
2172             IS_UP(new_active->dev)) {
2173                 write_lock_bh(&bond->curr_slave_lock);
2174                 bond_change_active_slave(bond, new_active);
2175                 write_unlock_bh(&bond->curr_slave_lock);
2176         } else
2177                 res = -EINVAL;
2178
2179         read_unlock(&bond->lock);
2180
2181         return res;
2182 }
2183
2184 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2185 {
2186         struct bonding *bond = netdev_priv(bond_dev);
2187
2188         info->bond_mode = bond->params.mode;
2189         info->miimon = bond->params.miimon;
2190
2191         read_lock(&bond->lock);
2192         info->num_slaves = bond->slave_cnt;
2193         read_unlock(&bond->lock);
2194
2195         return 0;
2196 }
2197
2198 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2199 {
2200         struct bonding *bond = netdev_priv(bond_dev);
2201         struct slave *slave;
2202         int i, res = -ENODEV;
2203
2204         read_lock(&bond->lock);
2205
2206         bond_for_each_slave(bond, slave, i) {
2207                 if (i == (int)info->slave_id) {
2208                         res = 0;
2209                         strcpy(info->slave_name, slave->dev->name);
2210                         info->link = slave->link;
2211                         info->state = slave->state;
2212                         info->link_failure_count = slave->link_failure_count;
2213                         break;
2214                 }
2215         }
2216
2217         read_unlock(&bond->lock);
2218
2219         return res;
2220 }
2221
2222 /*-------------------------------- Monitoring -------------------------------*/
2223
2224
2225 static int bond_miimon_inspect(struct bonding *bond)
2226 {
2227         struct slave *slave;
2228         int i, link_state, commit = 0;
2229         bool ignore_updelay;
2230
2231         ignore_updelay = !bond->curr_active_slave ? true : false;
2232
2233         bond_for_each_slave(bond, slave, i) {
2234                 slave->new_link = BOND_LINK_NOCHANGE;
2235
2236                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2237
2238                 switch (slave->link) {
2239                 case BOND_LINK_UP:
2240                         if (link_state)
2241                                 continue;
2242
2243                         slave->link = BOND_LINK_FAIL;
2244                         slave->delay = bond->params.downdelay;
2245                         if (slave->delay) {
2246                                 pr_info(DRV_NAME
2247                                        ": %s: link status down for %s"
2248                                        "interface %s, disabling it in %d ms.\n",
2249                                        bond->dev->name,
2250                                        (bond->params.mode ==
2251                                         BOND_MODE_ACTIVEBACKUP) ?
2252                                        ((slave->state == BOND_STATE_ACTIVE) ?
2253                                         "active " : "backup ") : "",
2254                                        slave->dev->name,
2255                                        bond->params.downdelay * bond->params.miimon);
2256                         }
2257                         /*FALLTHRU*/
2258                 case BOND_LINK_FAIL:
2259                         if (link_state) {
2260                                 /*
2261                                  * recovered before downdelay expired
2262                                  */
2263                                 slave->link = BOND_LINK_UP;
2264                                 slave->jiffies = jiffies;
2265                                 pr_info(DRV_NAME
2266                                        ": %s: link status up again after %d "
2267                                        "ms for interface %s.\n",
2268                                        bond->dev->name,
2269                                        (bond->params.downdelay - slave->delay) *
2270                                        bond->params.miimon,
2271                                        slave->dev->name);
2272                                 continue;
2273                         }
2274
2275                         if (slave->delay <= 0) {
2276                                 slave->new_link = BOND_LINK_DOWN;
2277                                 commit++;
2278                                 continue;
2279                         }
2280
2281                         slave->delay--;
2282                         break;
2283
2284                 case BOND_LINK_DOWN:
2285                         if (!link_state)
2286                                 continue;
2287
2288                         slave->link = BOND_LINK_BACK;
2289                         slave->delay = bond->params.updelay;
2290
2291                         if (slave->delay) {
2292                                 pr_info(DRV_NAME
2293                                        ": %s: link status up for "
2294                                        "interface %s, enabling it in %d ms.\n",
2295                                        bond->dev->name, slave->dev->name,
2296                                        ignore_updelay ? 0 :
2297                                        bond->params.updelay *
2298                                        bond->params.miimon);
2299                         }
2300                         /*FALLTHRU*/
2301                 case BOND_LINK_BACK:
2302                         if (!link_state) {
2303                                 slave->link = BOND_LINK_DOWN;
2304                                 pr_info(DRV_NAME
2305                                        ": %s: link status down again after %d "
2306                                        "ms for interface %s.\n",
2307                                        bond->dev->name,
2308                                        (bond->params.updelay - slave->delay) *
2309                                        bond->params.miimon,
2310                                        slave->dev->name);
2311
2312                                 continue;
2313                         }
2314
2315                         if (ignore_updelay)
2316                                 slave->delay = 0;
2317
2318                         if (slave->delay <= 0) {
2319                                 slave->new_link = BOND_LINK_UP;
2320                                 commit++;
2321                                 ignore_updelay = false;
2322                                 continue;
2323                         }
2324
2325                         slave->delay--;
2326                         break;
2327                 }
2328         }
2329
2330         return commit;
2331 }
2332
2333 static void bond_miimon_commit(struct bonding *bond)
2334 {
2335         struct slave *slave;
2336         int i;
2337
2338         bond_for_each_slave(bond, slave, i) {
2339                 switch (slave->new_link) {
2340                 case BOND_LINK_NOCHANGE:
2341                         continue;
2342
2343                 case BOND_LINK_UP:
2344                         slave->link = BOND_LINK_UP;
2345                         slave->jiffies = jiffies;
2346
2347                         if (bond->params.mode == BOND_MODE_8023AD) {
2348                                 /* prevent it from being the active one */
2349                                 slave->state = BOND_STATE_BACKUP;
2350                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2351                                 /* make it immediately active */
2352                                 slave->state = BOND_STATE_ACTIVE;
2353                         } else if (slave != bond->primary_slave) {
2354                                 /* prevent it from being the active one */
2355                                 slave->state = BOND_STATE_BACKUP;
2356                         }
2357
2358                         pr_info(DRV_NAME
2359                                ": %s: link status definitely "
2360                                "up for interface %s.\n",
2361                                bond->dev->name, slave->dev->name);
2362
2363                         /* notify ad that the link status has changed */
2364                         if (bond->params.mode == BOND_MODE_8023AD)
2365                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2366
2367                         if (bond_is_lb(bond))
2368                                 bond_alb_handle_link_change(bond, slave,
2369                                                             BOND_LINK_UP);
2370
2371                         if (!bond->curr_active_slave ||
2372                             (slave == bond->primary_slave))
2373                                 goto do_failover;
2374
2375                         continue;
2376
2377                 case BOND_LINK_DOWN:
2378                         if (slave->link_failure_count < UINT_MAX)
2379                                 slave->link_failure_count++;
2380
2381                         slave->link = BOND_LINK_DOWN;
2382
2383                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2384                             bond->params.mode == BOND_MODE_8023AD)
2385                                 bond_set_slave_inactive_flags(slave);
2386
2387                         pr_info(DRV_NAME
2388                                ": %s: link status definitely down for "
2389                                "interface %s, disabling it\n",
2390                                bond->dev->name, slave->dev->name);
2391
2392                         if (bond->params.mode == BOND_MODE_8023AD)
2393                                 bond_3ad_handle_link_change(slave,
2394                                                             BOND_LINK_DOWN);
2395
2396                         if (bond_is_lb(bond))
2397                                 bond_alb_handle_link_change(bond, slave,
2398                                                             BOND_LINK_DOWN);
2399
2400                         if (slave == bond->curr_active_slave)
2401                                 goto do_failover;
2402
2403                         continue;
2404
2405                 default:
2406                         pr_err(DRV_NAME
2407                                ": %s: invalid new link %d on slave %s\n",
2408                                bond->dev->name, slave->new_link,
2409                                slave->dev->name);
2410                         slave->new_link = BOND_LINK_NOCHANGE;
2411
2412                         continue;
2413                 }
2414
2415 do_failover:
2416                 ASSERT_RTNL();
2417                 write_lock_bh(&bond->curr_slave_lock);
2418                 bond_select_active_slave(bond);
2419                 write_unlock_bh(&bond->curr_slave_lock);
2420         }
2421
2422         bond_set_carrier(bond);
2423 }
2424
2425 /*
2426  * bond_mii_monitor
2427  *
2428  * Really a wrapper that splits the mii monitor into two phases: an
2429  * inspection, then (if inspection indicates something needs to be done)
2430  * an acquisition of appropriate locks followed by a commit phase to
2431  * implement whatever link state changes are indicated.
2432  */
2433 void bond_mii_monitor(struct work_struct *work)
2434 {
2435         struct bonding *bond = container_of(work, struct bonding,
2436                                             mii_work.work);
2437
2438         read_lock(&bond->lock);
2439         if (bond->kill_timers)
2440                 goto out;
2441
2442         if (bond->slave_cnt == 0)
2443                 goto re_arm;
2444
2445         if (bond->send_grat_arp) {
2446                 read_lock(&bond->curr_slave_lock);
2447                 bond_send_gratuitous_arp(bond);
2448                 read_unlock(&bond->curr_slave_lock);
2449         }
2450
2451         if (bond->send_unsol_na) {
2452                 read_lock(&bond->curr_slave_lock);
2453                 bond_send_unsolicited_na(bond);
2454                 read_unlock(&bond->curr_slave_lock);
2455         }
2456
2457         if (bond_miimon_inspect(bond)) {
2458                 read_unlock(&bond->lock);
2459                 rtnl_lock();
2460                 read_lock(&bond->lock);
2461
2462                 bond_miimon_commit(bond);
2463
2464                 read_unlock(&bond->lock);
2465                 rtnl_unlock();  /* might sleep, hold no other locks */
2466                 read_lock(&bond->lock);
2467         }
2468
2469 re_arm:
2470         if (bond->params.miimon)
2471                 queue_delayed_work(bond->wq, &bond->mii_work,
2472                                    msecs_to_jiffies(bond->params.miimon));
2473 out:
2474         read_unlock(&bond->lock);
2475 }
2476
2477 static __be32 bond_glean_dev_ip(struct net_device *dev)
2478 {
2479         struct in_device *idev;
2480         struct in_ifaddr *ifa;
2481         __be32 addr = 0;
2482
2483         if (!dev)
2484                 return 0;
2485
2486         rcu_read_lock();
2487         idev = __in_dev_get_rcu(dev);
2488         if (!idev)
2489                 goto out;
2490
2491         ifa = idev->ifa_list;
2492         if (!ifa)
2493                 goto out;
2494
2495         addr = ifa->ifa_local;
2496 out:
2497         rcu_read_unlock();
2498         return addr;
2499 }
2500
2501 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2502 {
2503         struct vlan_entry *vlan;
2504
2505         if (ip == bond->master_ip)
2506                 return 1;
2507
2508         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2509                 if (ip == vlan->vlan_ip)
2510                         return 1;
2511         }
2512
2513         return 0;
2514 }
2515
2516 /*
2517  * We go to the (large) trouble of VLAN tagging ARP frames because
2518  * switches in VLAN mode (especially if ports are configured as
2519  * "native" to a VLAN) might not pass non-tagged frames.
2520  */
2521 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2522 {
2523         struct sk_buff *skb;
2524
2525         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2526                slave_dev->name, dest_ip, src_ip, vlan_id);
2527
2528         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2529                          NULL, slave_dev->dev_addr, NULL);
2530
2531         if (!skb) {
2532                 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2533                 return;
2534         }
2535         if (vlan_id) {
2536                 skb = vlan_put_tag(skb, vlan_id);
2537                 if (!skb) {
2538                         pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2539                         return;
2540                 }
2541         }
2542         arp_xmit(skb);
2543 }
2544
2545
2546 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2547 {
2548         int i, vlan_id, rv;
2549         __be32 *targets = bond->params.arp_targets;
2550         struct vlan_entry *vlan;
2551         struct net_device *vlan_dev;
2552         struct flowi fl;
2553         struct rtable *rt;
2554
2555         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2556                 if (!targets[i])
2557                         break;
2558                 pr_debug("basa: target %x\n", targets[i]);
2559                 if (list_empty(&bond->vlan_list)) {
2560                         pr_debug("basa: empty vlan: arp_send\n");
2561                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2562                                       bond->master_ip, 0);
2563                         continue;
2564                 }
2565
2566                 /*
2567                  * If VLANs are configured, we do a route lookup to
2568                  * determine which VLAN interface would be used, so we
2569                  * can tag the ARP with the proper VLAN tag.
2570                  */
2571                 memset(&fl, 0, sizeof(fl));
2572                 fl.fl4_dst = targets[i];
2573                 fl.fl4_tos = RTO_ONLINK;
2574
2575                 rv = ip_route_output_key(&init_net, &rt, &fl);
2576                 if (rv) {
2577                         if (net_ratelimit()) {
2578                                 pr_warning(DRV_NAME
2579                              ": %s: no route to arp_ip_target %pI4\n",
2580                                        bond->dev->name, &fl.fl4_dst);
2581                         }
2582                         continue;
2583                 }
2584
2585                 /*
2586                  * This target is not on a VLAN
2587                  */
2588                 if (rt->u.dst.dev == bond->dev) {
2589                         ip_rt_put(rt);
2590                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2591                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2592                                       bond->master_ip, 0);
2593                         continue;
2594                 }
2595
2596                 vlan_id = 0;
2597                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2598                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2599                         if (vlan_dev == rt->u.dst.dev) {
2600                                 vlan_id = vlan->vlan_id;
2601                                 pr_debug("basa: vlan match on %s %d\n",
2602                                        vlan_dev->name, vlan_id);
2603                                 break;
2604                         }
2605                 }
2606
2607                 if (vlan_id) {
2608                         ip_rt_put(rt);
2609                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2610                                       vlan->vlan_ip, vlan_id);
2611                         continue;
2612                 }
2613
2614                 if (net_ratelimit()) {
2615                         pr_warning(DRV_NAME
2616                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2617                                bond->dev->name, &fl.fl4_dst,
2618                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2619                 }
2620                 ip_rt_put(rt);
2621         }
2622 }
2623
2624 /*
2625  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2626  * for each VLAN above us.
2627  *
2628  * Caller must hold curr_slave_lock for read or better
2629  */
2630 static void bond_send_gratuitous_arp(struct bonding *bond)
2631 {
2632         struct slave *slave = bond->curr_active_slave;
2633         struct vlan_entry *vlan;
2634         struct net_device *vlan_dev;
2635
2636         pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2637                                 slave ? slave->dev->name : "NULL");
2638
2639         if (!slave || !bond->send_grat_arp ||
2640             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2641                 return;
2642
2643         bond->send_grat_arp--;
2644
2645         if (bond->master_ip) {
2646                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2647                                 bond->master_ip, 0);
2648         }
2649
2650         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2651                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2652                 if (vlan->vlan_ip) {
2653                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2654                                       vlan->vlan_ip, vlan->vlan_id);
2655                 }
2656         }
2657 }
2658
2659 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2660 {
2661         int i;
2662         __be32 *targets = bond->params.arp_targets;
2663
2664         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2665                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2666                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2667                 if (sip == targets[i]) {
2668                         if (bond_has_this_ip(bond, tip))
2669                                 slave->last_arp_rx = jiffies;
2670                         return;
2671                 }
2672         }
2673 }
2674
2675 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2676 {
2677         struct arphdr *arp;
2678         struct slave *slave;
2679         struct bonding *bond;
2680         unsigned char *arp_ptr;
2681         __be32 sip, tip;
2682
2683         if (dev_net(dev) != &init_net)
2684                 goto out;
2685
2686         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2687                 goto out;
2688
2689         bond = netdev_priv(dev);
2690         read_lock(&bond->lock);
2691
2692         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2693                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2694                 orig_dev ? orig_dev->name : "NULL");
2695
2696         slave = bond_get_slave_by_dev(bond, orig_dev);
2697         if (!slave || !slave_do_arp_validate(bond, slave))
2698                 goto out_unlock;
2699
2700         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2701                 goto out_unlock;
2702
2703         arp = arp_hdr(skb);
2704         if (arp->ar_hln != dev->addr_len ||
2705             skb->pkt_type == PACKET_OTHERHOST ||
2706             skb->pkt_type == PACKET_LOOPBACK ||
2707             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2708             arp->ar_pro != htons(ETH_P_IP) ||
2709             arp->ar_pln != 4)
2710                 goto out_unlock;
2711
2712         arp_ptr = (unsigned char *)(arp + 1);
2713         arp_ptr += dev->addr_len;
2714         memcpy(&sip, arp_ptr, 4);
2715         arp_ptr += 4 + dev->addr_len;
2716         memcpy(&tip, arp_ptr, 4);
2717
2718         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2719                 bond->dev->name, slave->dev->name, slave->state,
2720                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2721                 &sip, &tip);
2722
2723         /*
2724          * Backup slaves won't see the ARP reply, but do come through
2725          * here for each ARP probe (so we swap the sip/tip to validate
2726          * the probe).  In a "redundant switch, common router" type of
2727          * configuration, the ARP probe will (hopefully) travel from
2728          * the active, through one switch, the router, then the other
2729          * switch before reaching the backup.
2730          */
2731         if (slave->state == BOND_STATE_ACTIVE)
2732                 bond_validate_arp(bond, slave, sip, tip);
2733         else
2734                 bond_validate_arp(bond, slave, tip, sip);
2735
2736 out_unlock:
2737         read_unlock(&bond->lock);
2738 out:
2739         dev_kfree_skb(skb);
2740         return NET_RX_SUCCESS;
2741 }
2742
2743 /*
2744  * this function is called regularly to monitor each slave's link
2745  * ensuring that traffic is being sent and received when arp monitoring
2746  * is used in load-balancing mode. if the adapter has been dormant, then an
2747  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2748  * arp monitoring in active backup mode.
2749  */
2750 void bond_loadbalance_arp_mon(struct work_struct *work)
2751 {
2752         struct bonding *bond = container_of(work, struct bonding,
2753                                             arp_work.work);
2754         struct slave *slave, *oldcurrent;
2755         int do_failover = 0;
2756         int delta_in_ticks;
2757         int i;
2758
2759         read_lock(&bond->lock);
2760
2761         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2762
2763         if (bond->kill_timers)
2764                 goto out;
2765
2766         if (bond->slave_cnt == 0)
2767                 goto re_arm;
2768
2769         read_lock(&bond->curr_slave_lock);
2770         oldcurrent = bond->curr_active_slave;
2771         read_unlock(&bond->curr_slave_lock);
2772
2773         /* see if any of the previous devices are up now (i.e. they have
2774          * xmt and rcv traffic). the curr_active_slave does not come into
2775          * the picture unless it is null. also, slave->jiffies is not needed
2776          * here because we send an arp on each slave and give a slave as
2777          * long as it needs to get the tx/rx within the delta.
2778          * TODO: what about up/down delay in arp mode? it wasn't here before
2779          *       so it can wait
2780          */
2781         bond_for_each_slave(bond, slave, i) {
2782                 if (slave->link != BOND_LINK_UP) {
2783                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2784                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2785
2786                                 slave->link  = BOND_LINK_UP;
2787                                 slave->state = BOND_STATE_ACTIVE;
2788
2789                                 /* primary_slave has no meaning in round-robin
2790                                  * mode. the window of a slave being up and
2791                                  * curr_active_slave being null after enslaving
2792                                  * is closed.
2793                                  */
2794                                 if (!oldcurrent) {
2795                                         pr_info(DRV_NAME
2796                                                ": %s: link status definitely "
2797                                                "up for interface %s, ",
2798                                                bond->dev->name,
2799                                                slave->dev->name);
2800                                         do_failover = 1;
2801                                 } else {
2802                                         pr_info(DRV_NAME
2803                                                ": %s: interface %s is now up\n",
2804                                                bond->dev->name,
2805                                                slave->dev->name);
2806                                 }
2807                         }
2808                 } else {
2809                         /* slave->link == BOND_LINK_UP */
2810
2811                         /* not all switches will respond to an arp request
2812                          * when the source ip is 0, so don't take the link down
2813                          * if we don't know our ip yet
2814                          */
2815                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2816                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2817
2818                                 slave->link  = BOND_LINK_DOWN;
2819                                 slave->state = BOND_STATE_BACKUP;
2820
2821                                 if (slave->link_failure_count < UINT_MAX)
2822                                         slave->link_failure_count++;
2823
2824                                 pr_info(DRV_NAME
2825                                        ": %s: interface %s is now down.\n",
2826                                        bond->dev->name,
2827                                        slave->dev->name);
2828
2829                                 if (slave == oldcurrent)
2830                                         do_failover = 1;
2831                         }
2832                 }
2833
2834                 /* note: if switch is in round-robin mode, all links
2835                  * must tx arp to ensure all links rx an arp - otherwise
2836                  * links may oscillate or not come up at all; if switch is
2837                  * in something like xor mode, there is nothing we can
2838                  * do - all replies will be rx'ed on same link causing slaves
2839                  * to be unstable during low/no traffic periods
2840                  */
2841                 if (IS_UP(slave->dev))
2842                         bond_arp_send_all(bond, slave);
2843         }
2844
2845         if (do_failover) {
2846                 write_lock_bh(&bond->curr_slave_lock);
2847
2848                 bond_select_active_slave(bond);
2849
2850                 write_unlock_bh(&bond->curr_slave_lock);
2851         }
2852
2853 re_arm:
2854         if (bond->params.arp_interval)
2855                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2856 out:
2857         read_unlock(&bond->lock);
2858 }
2859
2860 /*
2861  * Called to inspect slaves for active-backup mode ARP monitor link state
2862  * changes.  Sets new_link in slaves to specify what action should take
2863  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2864  * to link states must be committed.
2865  *
2866  * Called with bond->lock held for read.
2867  */
2868 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2869 {
2870         struct slave *slave;
2871         int i, commit = 0;
2872
2873         bond_for_each_slave(bond, slave, i) {
2874                 slave->new_link = BOND_LINK_NOCHANGE;
2875
2876                 if (slave->link != BOND_LINK_UP) {
2877                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2878                                            delta_in_ticks)) {
2879                                 slave->new_link = BOND_LINK_UP;
2880                                 commit++;
2881                         }
2882
2883                         continue;
2884                 }
2885
2886                 /*
2887                  * Give slaves 2*delta after being enslaved or made
2888                  * active.  This avoids bouncing, as the last receive
2889                  * times need a full ARP monitor cycle to be updated.
2890                  */
2891                 if (!time_after_eq(jiffies, slave->jiffies +
2892                                    2 * delta_in_ticks))
2893                         continue;
2894
2895                 /*
2896                  * Backup slave is down if:
2897                  * - No current_arp_slave AND
2898                  * - more than 3*delta since last receive AND
2899                  * - the bond has an IP address
2900                  *
2901                  * Note: a non-null current_arp_slave indicates
2902                  * the curr_active_slave went down and we are
2903                  * searching for a new one; under this condition
2904                  * we only take the curr_active_slave down - this
2905                  * gives each slave a chance to tx/rx traffic
2906                  * before being taken out
2907                  */
2908                 if (slave->state == BOND_STATE_BACKUP &&
2909                     !bond->current_arp_slave &&
2910                     time_after(jiffies, slave_last_rx(bond, slave) +
2911                                3 * delta_in_ticks)) {
2912                         slave->new_link = BOND_LINK_DOWN;
2913                         commit++;
2914                 }
2915
2916                 /*
2917                  * Active slave is down if:
2918                  * - more than 2*delta since transmitting OR
2919                  * - (more than 2*delta since receive AND
2920                  *    the bond has an IP address)
2921                  */
2922                 if ((slave->state == BOND_STATE_ACTIVE) &&
2923                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2924                                     2 * delta_in_ticks) ||
2925                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2926                                      + 2 * delta_in_ticks)))) {
2927                         slave->new_link = BOND_LINK_DOWN;
2928                         commit++;
2929                 }
2930         }
2931
2932         read_lock(&bond->curr_slave_lock);
2933
2934         /*
2935          * Trigger a commit if the primary option setting has changed.
2936          */
2937         if (bond->primary_slave &&
2938             (bond->primary_slave != bond->curr_active_slave) &&
2939             (bond->primary_slave->link == BOND_LINK_UP))
2940                 commit++;
2941
2942         read_unlock(&bond->curr_slave_lock);
2943
2944         return commit;
2945 }
2946
2947 /*
2948  * Called to commit link state changes noted by inspection step of
2949  * active-backup mode ARP monitor.
2950  *
2951  * Called with RTNL and bond->lock for read.
2952  */
2953 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2954 {
2955         struct slave *slave;
2956         int i;
2957
2958         bond_for_each_slave(bond, slave, i) {
2959                 switch (slave->new_link) {
2960                 case BOND_LINK_NOCHANGE:
2961                         continue;
2962
2963                 case BOND_LINK_UP:
2964                         write_lock_bh(&bond->curr_slave_lock);
2965
2966                         if (!bond->curr_active_slave &&
2967                             time_before_eq(jiffies, dev_trans_start(slave->dev) +
2968                                            delta_in_ticks)) {
2969                                 slave->link = BOND_LINK_UP;
2970                                 bond_change_active_slave(bond, slave);
2971                                 bond->current_arp_slave = NULL;
2972
2973                                 pr_info(DRV_NAME
2974                                        ": %s: %s is up and now the "
2975                                        "active interface\n",
2976                                        bond->dev->name, slave->dev->name);
2977
2978                         } else if (bond->curr_active_slave != slave) {
2979                                 /* this slave has just come up but we
2980                                  * already have a current slave; this can
2981                                  * also happen if bond_enslave adds a new
2982                                  * slave that is up while we are searching
2983                                  * for a new slave
2984                                  */
2985                                 slave->link = BOND_LINK_UP;
2986                                 bond_set_slave_inactive_flags(slave);
2987                                 bond->current_arp_slave = NULL;
2988
2989                                 pr_info(DRV_NAME
2990                                        ": %s: backup interface %s is now up\n",
2991                                        bond->dev->name, slave->dev->name);
2992                         }
2993
2994                         write_unlock_bh(&bond->curr_slave_lock);
2995
2996                         break;
2997
2998                 case BOND_LINK_DOWN:
2999                         if (slave->link_failure_count < UINT_MAX)
3000                                 slave->link_failure_count++;
3001
3002                         slave->link = BOND_LINK_DOWN;
3003
3004                         if (slave == bond->curr_active_slave) {
3005                                 pr_info(DRV_NAME
3006                                        ": %s: link status down for active "
3007                                        "interface %s, disabling it\n",
3008                                        bond->dev->name, slave->dev->name);
3009
3010                                 bond_set_slave_inactive_flags(slave);
3011
3012                                 write_lock_bh(&bond->curr_slave_lock);
3013
3014                                 bond_select_active_slave(bond);
3015                                 if (bond->curr_active_slave)
3016                                         bond->curr_active_slave->jiffies =
3017                                                 jiffies;
3018
3019                                 write_unlock_bh(&bond->curr_slave_lock);
3020
3021                                 bond->current_arp_slave = NULL;
3022
3023                         } else if (slave->state == BOND_STATE_BACKUP) {
3024                                 pr_info(DRV_NAME
3025                                        ": %s: backup interface %s is now down\n",
3026                                        bond->dev->name, slave->dev->name);
3027
3028                                 bond_set_slave_inactive_flags(slave);
3029                         }
3030                         break;
3031
3032                 default:
3033                         pr_err(DRV_NAME
3034                                ": %s: impossible: new_link %d on slave %s\n",
3035                                bond->dev->name, slave->new_link,
3036                                slave->dev->name);
3037                 }
3038         }
3039
3040         /*
3041          * No race with changes to primary via sysfs, as we hold rtnl.
3042          */
3043         if (bond->primary_slave &&
3044             (bond->primary_slave != bond->curr_active_slave) &&
3045             (bond->primary_slave->link == BOND_LINK_UP)) {
3046                 write_lock_bh(&bond->curr_slave_lock);
3047                 bond_change_active_slave(bond, bond->primary_slave);
3048                 write_unlock_bh(&bond->curr_slave_lock);
3049         }
3050
3051         bond_set_carrier(bond);
3052 }
3053
3054 /*
3055  * Send ARP probes for active-backup mode ARP monitor.
3056  *
3057  * Called with bond->lock held for read.
3058  */
3059 static void bond_ab_arp_probe(struct bonding *bond)
3060 {
3061         struct slave *slave;
3062         int i;
3063
3064         read_lock(&bond->curr_slave_lock);
3065
3066         if (bond->current_arp_slave && bond->curr_active_slave)
3067                 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3068                        bond->current_arp_slave->dev->name,
3069                        bond->curr_active_slave->dev->name);
3070
3071         if (bond->curr_active_slave) {
3072                 bond_arp_send_all(bond, bond->curr_active_slave);
3073                 read_unlock(&bond->curr_slave_lock);
3074                 return;
3075         }
3076
3077         read_unlock(&bond->curr_slave_lock);
3078
3079         /* if we don't have a curr_active_slave, search for the next available
3080          * backup slave from the current_arp_slave and make it the candidate
3081          * for becoming the curr_active_slave
3082          */
3083
3084         if (!bond->current_arp_slave) {
3085                 bond->current_arp_slave = bond->first_slave;
3086                 if (!bond->current_arp_slave)
3087                         return;
3088         }
3089
3090         bond_set_slave_inactive_flags(bond->current_arp_slave);
3091
3092         /* search for next candidate */
3093         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3094                 if (IS_UP(slave->dev)) {
3095                         slave->link = BOND_LINK_BACK;
3096                         bond_set_slave_active_flags(slave);
3097                         bond_arp_send_all(bond, slave);
3098                         slave->jiffies = jiffies;
3099                         bond->current_arp_slave = slave;
3100                         break;
3101                 }
3102
3103                 /* if the link state is up at this point, we
3104                  * mark it down - this can happen if we have
3105                  * simultaneous link failures and
3106                  * reselect_active_interface doesn't make this
3107                  * one the current slave so it is still marked
3108                  * up when it is actually down
3109                  */
3110                 if (slave->link == BOND_LINK_UP) {
3111                         slave->link = BOND_LINK_DOWN;
3112                         if (slave->link_failure_count < UINT_MAX)
3113                                 slave->link_failure_count++;
3114
3115                         bond_set_slave_inactive_flags(slave);
3116
3117                         pr_info(DRV_NAME
3118                                ": %s: backup interface %s is now down.\n",
3119                                bond->dev->name, slave->dev->name);
3120                 }
3121         }
3122 }
3123
3124 void bond_activebackup_arp_mon(struct work_struct *work)
3125 {
3126         struct bonding *bond = container_of(work, struct bonding,
3127                                             arp_work.work);
3128         int delta_in_ticks;
3129
3130         read_lock(&bond->lock);
3131
3132         if (bond->kill_timers)
3133                 goto out;
3134
3135         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3136
3137         if (bond->slave_cnt == 0)
3138                 goto re_arm;
3139
3140         if (bond->send_grat_arp) {
3141                 read_lock(&bond->curr_slave_lock);
3142                 bond_send_gratuitous_arp(bond);
3143                 read_unlock(&bond->curr_slave_lock);
3144         }
3145
3146         if (bond->send_unsol_na) {
3147                 read_lock(&bond->curr_slave_lock);
3148                 bond_send_unsolicited_na(bond);
3149                 read_unlock(&bond->curr_slave_lock);
3150         }
3151
3152         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3153                 read_unlock(&bond->lock);
3154                 rtnl_lock();
3155                 read_lock(&bond->lock);
3156
3157                 bond_ab_arp_commit(bond, delta_in_ticks);
3158
3159                 read_unlock(&bond->lock);
3160                 rtnl_unlock();
3161                 read_lock(&bond->lock);
3162         }
3163
3164         bond_ab_arp_probe(bond);
3165
3166 re_arm:
3167         if (bond->params.arp_interval)
3168                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3169 out:
3170         read_unlock(&bond->lock);
3171 }
3172
3173 /*------------------------------ proc/seq_file-------------------------------*/
3174
3175 #ifdef CONFIG_PROC_FS
3176
3177 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3178         __acquires(&dev_base_lock)
3179         __acquires(&bond->lock)
3180 {
3181         struct bonding *bond = seq->private;
3182         loff_t off = 0;
3183         struct slave *slave;
3184         int i;
3185
3186         /* make sure the bond won't be taken away */
3187         read_lock(&dev_base_lock);
3188         read_lock(&bond->lock);
3189
3190         if (*pos == 0)
3191                 return SEQ_START_TOKEN;
3192
3193         bond_for_each_slave(bond, slave, i) {
3194                 if (++off == *pos)
3195                         return slave;
3196         }
3197
3198         return NULL;
3199 }
3200
3201 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3202 {
3203         struct bonding *bond = seq->private;
3204         struct slave *slave = v;
3205
3206         ++*pos;
3207         if (v == SEQ_START_TOKEN)
3208                 return bond->first_slave;
3209
3210         slave = slave->next;
3211
3212         return (slave == bond->first_slave) ? NULL : slave;
3213 }
3214
3215 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3216         __releases(&bond->lock)
3217         __releases(&dev_base_lock)
3218 {
3219         struct bonding *bond = seq->private;
3220
3221         read_unlock(&bond->lock);
3222         read_unlock(&dev_base_lock);
3223 }
3224
3225 static void bond_info_show_master(struct seq_file *seq)
3226 {
3227         struct bonding *bond = seq->private;
3228         struct slave *curr;
3229         int i;
3230
3231         read_lock(&bond->curr_slave_lock);
3232         curr = bond->curr_active_slave;
3233         read_unlock(&bond->curr_slave_lock);
3234
3235         seq_printf(seq, "Bonding Mode: %s",
3236                    bond_mode_name(bond->params.mode));
3237
3238         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3239             bond->params.fail_over_mac)
3240                 seq_printf(seq, " (fail_over_mac %s)",
3241                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3242
3243         seq_printf(seq, "\n");
3244
3245         if (bond->params.mode == BOND_MODE_XOR ||
3246                 bond->params.mode == BOND_MODE_8023AD) {
3247                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3248                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3249                         bond->params.xmit_policy);
3250         }
3251
3252         if (USES_PRIMARY(bond->params.mode)) {
3253                 seq_printf(seq, "Primary Slave: %s\n",
3254                            (bond->primary_slave) ?
3255                            bond->primary_slave->dev->name : "None");
3256
3257                 seq_printf(seq, "Currently Active Slave: %s\n",
3258                            (curr) ? curr->dev->name : "None");
3259         }
3260
3261         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3262                    "up" : "down");
3263         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3264         seq_printf(seq, "Up Delay (ms): %d\n",
3265                    bond->params.updelay * bond->params.miimon);
3266         seq_printf(seq, "Down Delay (ms): %d\n",
3267                    bond->params.downdelay * bond->params.miimon);
3268
3269
3270         /* ARP information */
3271         if (bond->params.arp_interval > 0) {
3272                 int printed = 0;
3273                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3274                                 bond->params.arp_interval);
3275
3276                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3277
3278                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3279                         if (!bond->params.arp_targets[i])
3280                                 break;
3281                         if (printed)
3282                                 seq_printf(seq, ",");
3283                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3284                         printed = 1;
3285                 }
3286                 seq_printf(seq, "\n");
3287         }
3288
3289         if (bond->params.mode == BOND_MODE_8023AD) {
3290                 struct ad_info ad_info;
3291
3292                 seq_puts(seq, "\n802.3ad info\n");
3293                 seq_printf(seq, "LACP rate: %s\n",
3294                            (bond->params.lacp_fast) ? "fast" : "slow");
3295                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3296                            ad_select_tbl[bond->params.ad_select].modename);
3297
3298                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3299                         seq_printf(seq, "bond %s has no active aggregator\n",
3300                                    bond->dev->name);
3301                 } else {
3302                         seq_printf(seq, "Active Aggregator Info:\n");
3303
3304                         seq_printf(seq, "\tAggregator ID: %d\n",
3305                                    ad_info.aggregator_id);
3306                         seq_printf(seq, "\tNumber of ports: %d\n",
3307                                    ad_info.ports);
3308                         seq_printf(seq, "\tActor Key: %d\n",
3309                                    ad_info.actor_key);
3310                         seq_printf(seq, "\tPartner Key: %d\n",
3311                                    ad_info.partner_key);
3312                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3313                                    ad_info.partner_system);
3314                 }
3315         }
3316 }
3317
3318 static void bond_info_show_slave(struct seq_file *seq,
3319                                  const struct slave *slave)
3320 {
3321         struct bonding *bond = seq->private;
3322
3323         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3324         seq_printf(seq, "MII Status: %s\n",
3325                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3326         seq_printf(seq, "Link Failure Count: %u\n",
3327                    slave->link_failure_count);
3328
3329         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3330
3331         if (bond->params.mode == BOND_MODE_8023AD) {
3332                 const struct aggregator *agg
3333                         = SLAVE_AD_INFO(slave).port.aggregator;
3334
3335                 if (agg)
3336                         seq_printf(seq, "Aggregator ID: %d\n",
3337                                    agg->aggregator_identifier);
3338                 else
3339                         seq_puts(seq, "Aggregator ID: N/A\n");
3340         }
3341 }
3342
3343 static int bond_info_seq_show(struct seq_file *seq, void *v)
3344 {
3345         if (v == SEQ_START_TOKEN) {
3346                 seq_printf(seq, "%s\n", version);
3347                 bond_info_show_master(seq);
3348         } else
3349                 bond_info_show_slave(seq, v);
3350
3351         return 0;
3352 }
3353
3354 static const struct seq_operations bond_info_seq_ops = {
3355         .start = bond_info_seq_start,
3356         .next  = bond_info_seq_next,
3357         .stop  = bond_info_seq_stop,
3358         .show  = bond_info_seq_show,
3359 };
3360
3361 static int bond_info_open(struct inode *inode, struct file *file)
3362 {
3363         struct seq_file *seq;
3364         struct proc_dir_entry *proc;
3365         int res;
3366
3367         res = seq_open(file, &bond_info_seq_ops);
3368         if (!res) {
3369                 /* recover the pointer buried in proc_dir_entry data */
3370                 seq = file->private_data;
3371                 proc = PDE(inode);
3372                 seq->private = proc->data;
3373         }
3374
3375         return res;
3376 }
3377
3378 static const struct file_operations bond_info_fops = {
3379         .owner   = THIS_MODULE,
3380         .open    = bond_info_open,
3381         .read    = seq_read,
3382         .llseek  = seq_lseek,
3383         .release = seq_release,
3384 };
3385
3386 static int bond_create_proc_entry(struct bonding *bond)
3387 {
3388         struct net_device *bond_dev = bond->dev;
3389
3390         if (bond_proc_dir) {
3391                 bond->proc_entry = proc_create_data(bond_dev->name,
3392                                                     S_IRUGO, bond_proc_dir,
3393                                                     &bond_info_fops, bond);
3394                 if (bond->proc_entry == NULL)
3395                         pr_warning(DRV_NAME
3396                                ": Warning: Cannot create /proc/net/%s/%s\n",
3397                                DRV_NAME, bond_dev->name);
3398                 else
3399                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3400         }
3401
3402         return 0;
3403 }
3404
3405 static void bond_remove_proc_entry(struct bonding *bond)
3406 {
3407         if (bond_proc_dir && bond->proc_entry) {
3408                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3409                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3410                 bond->proc_entry = NULL;
3411         }
3412 }
3413
3414 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3415  * Caller must hold rtnl_lock.
3416  */
3417 static void bond_create_proc_dir(void)
3418 {
3419         if (!bond_proc_dir) {
3420                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3421                 if (!bond_proc_dir)
3422                         pr_warning(DRV_NAME
3423                                 ": Warning: cannot create /proc/net/%s\n",
3424                                 DRV_NAME);
3425         }
3426 }
3427
3428 /* Destroy the bonding directory under /proc/net, if empty.
3429  * Caller must hold rtnl_lock.
3430  */
3431 static void bond_destroy_proc_dir(void)
3432 {
3433         if (bond_proc_dir) {
3434                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3435                 bond_proc_dir = NULL;
3436         }
3437 }
3438
3439 #else /* !CONFIG_PROC_FS */
3440
3441 static int bond_create_proc_entry(struct bonding *bond)
3442 {
3443 }
3444
3445 static void bond_remove_proc_entry(struct bonding *bond)
3446 {
3447 }
3448
3449 static void bond_create_proc_dir(void)
3450 {
3451 }
3452
3453 static void bond_destroy_proc_dir(void)
3454 {
3455 }
3456
3457 #endif /* CONFIG_PROC_FS */
3458
3459
3460 /*-------------------------- netdev event handling --------------------------*/
3461
3462 /*
3463  * Change device name
3464  */
3465 static int bond_event_changename(struct bonding *bond)
3466 {
3467         bond_remove_proc_entry(bond);
3468         bond_create_proc_entry(bond);
3469
3470         bond_destroy_sysfs_entry(bond);
3471         bond_create_sysfs_entry(bond);
3472
3473         return NOTIFY_DONE;
3474 }
3475
3476 static int bond_master_netdev_event(unsigned long event,
3477                                     struct net_device *bond_dev)
3478 {
3479         struct bonding *event_bond = netdev_priv(bond_dev);
3480
3481         switch (event) {
3482         case NETDEV_CHANGENAME:
3483                 return bond_event_changename(event_bond);
3484         case NETDEV_UNREGISTER:
3485                 bond_release_all(event_bond->dev);
3486                 break;
3487         default:
3488                 break;
3489         }
3490
3491         return NOTIFY_DONE;
3492 }
3493
3494 static int bond_slave_netdev_event(unsigned long event,
3495                                    struct net_device *slave_dev)
3496 {
3497         struct net_device *bond_dev = slave_dev->master;
3498         struct bonding *bond = netdev_priv(bond_dev);
3499
3500         switch (event) {
3501         case NETDEV_UNREGISTER:
3502                 if (bond_dev) {
3503                         if (bond->setup_by_slave)
3504                                 bond_release_and_destroy(bond_dev, slave_dev);
3505                         else
3506                                 bond_release(bond_dev, slave_dev);
3507                 }
3508                 break;
3509         case NETDEV_CHANGE:
3510                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3511                         struct slave *slave;
3512
3513                         slave = bond_get_slave_by_dev(bond, slave_dev);
3514                         if (slave) {
3515                                 u16 old_speed = slave->speed;
3516                                 u16 old_duplex = slave->duplex;
3517
3518                                 bond_update_speed_duplex(slave);
3519
3520                                 if (bond_is_lb(bond))
3521                                         break;
3522
3523                                 if (old_speed != slave->speed)
3524                                         bond_3ad_adapter_speed_changed(slave);
3525                                 if (old_duplex != slave->duplex)
3526                                         bond_3ad_adapter_duplex_changed(slave);
3527                         }
3528                 }
3529
3530                 break;
3531         case NETDEV_DOWN:
3532                 /*
3533                  * ... Or is it this?
3534                  */
3535                 break;
3536         case NETDEV_CHANGEMTU:
3537                 /*
3538                  * TODO: Should slaves be allowed to
3539                  * independently alter their MTU?  For
3540                  * an active-backup bond, slaves need
3541                  * not be the same type of device, so
3542                  * MTUs may vary.  For other modes,
3543                  * slaves arguably should have the
3544                  * same MTUs. To do this, we'd need to
3545                  * take over the slave's change_mtu
3546                  * function for the duration of their
3547                  * servitude.
3548                  */
3549                 break;
3550         case NETDEV_CHANGENAME:
3551                 /*
3552                  * TODO: handle changing the primary's name
3553                  */
3554                 break;
3555         case NETDEV_FEAT_CHANGE:
3556                 bond_compute_features(bond);
3557                 break;
3558         default:
3559                 break;
3560         }
3561
3562         return NOTIFY_DONE;
3563 }
3564
3565 /*
3566  * bond_netdev_event: handle netdev notifier chain events.
3567  *
3568  * This function receives events for the netdev chain.  The caller (an
3569  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3570  * locks for us to safely manipulate the slave devices (RTNL lock,
3571  * dev_probe_lock).
3572  */
3573 static int bond_netdev_event(struct notifier_block *this,
3574                              unsigned long event, void *ptr)
3575 {
3576         struct net_device *event_dev = (struct net_device *)ptr;
3577
3578         if (dev_net(event_dev) != &init_net)
3579                 return NOTIFY_DONE;
3580
3581         pr_debug("event_dev: %s, event: %lx\n",
3582                 (event_dev ? event_dev->name : "None"),
3583                 event);
3584
3585         if (!(event_dev->priv_flags & IFF_BONDING))
3586                 return NOTIFY_DONE;
3587
3588         if (event_dev->flags & IFF_MASTER) {
3589                 pr_debug("IFF_MASTER\n");
3590                 return bond_master_netdev_event(event, event_dev);
3591         }
3592
3593         if (event_dev->flags & IFF_SLAVE) {
3594                 pr_debug("IFF_SLAVE\n");
3595                 return bond_slave_netdev_event(event, event_dev);
3596         }
3597
3598         return NOTIFY_DONE;
3599 }
3600
3601 /*
3602  * bond_inetaddr_event: handle inetaddr notifier chain events.
3603  *
3604  * We keep track of device IPs primarily to use as source addresses in
3605  * ARP monitor probes (rather than spewing out broadcasts all the time).
3606  *
3607  * We track one IP for the main device (if it has one), plus one per VLAN.
3608  */
3609 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3610 {
3611         struct in_ifaddr *ifa = ptr;
3612         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3613         struct bonding *bond;
3614         struct vlan_entry *vlan;
3615
3616         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3617                 return NOTIFY_DONE;
3618
3619         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3620                 if (bond->dev == event_dev) {
3621                         switch (event) {
3622                         case NETDEV_UP:
3623                                 bond->master_ip = ifa->ifa_local;
3624                                 return NOTIFY_OK;
3625                         case NETDEV_DOWN:
3626                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3627                                 return NOTIFY_OK;
3628                         default:
3629                                 return NOTIFY_DONE;
3630                         }
3631                 }
3632
3633                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3634                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3635                         if (vlan_dev == event_dev) {
3636                                 switch (event) {
3637                                 case NETDEV_UP:
3638                                         vlan->vlan_ip = ifa->ifa_local;
3639                                         return NOTIFY_OK;
3640                                 case NETDEV_DOWN:
3641                                         vlan->vlan_ip =
3642                                                 bond_glean_dev_ip(vlan_dev);
3643                                         return NOTIFY_OK;
3644                                 default:
3645                                         return NOTIFY_DONE;
3646                                 }
3647                         }
3648                 }
3649         }
3650         return NOTIFY_DONE;
3651 }
3652
3653 static struct notifier_block bond_netdev_notifier = {
3654         .notifier_call = bond_netdev_event,
3655 };
3656
3657 static struct notifier_block bond_inetaddr_notifier = {
3658         .notifier_call = bond_inetaddr_event,
3659 };
3660
3661 /*-------------------------- Packet type handling ---------------------------*/
3662
3663 /* register to receive lacpdus on a bond */
3664 static void bond_register_lacpdu(struct bonding *bond)
3665 {
3666         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3667
3668         /* initialize packet type */
3669         pk_type->type = PKT_TYPE_LACPDU;
3670         pk_type->dev = bond->dev;
3671         pk_type->func = bond_3ad_lacpdu_recv;
3672
3673         dev_add_pack(pk_type);
3674 }
3675
3676 /* unregister to receive lacpdus on a bond */
3677 static void bond_unregister_lacpdu(struct bonding *bond)
3678 {
3679         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3680 }
3681
3682 void bond_register_arp(struct bonding *bond)
3683 {
3684         struct packet_type *pt = &bond->arp_mon_pt;
3685
3686         if (pt->type)
3687                 return;
3688
3689         pt->type = htons(ETH_P_ARP);
3690         pt->dev = bond->dev;
3691         pt->func = bond_arp_rcv;
3692         dev_add_pack(pt);
3693 }
3694
3695 void bond_unregister_arp(struct bonding *bond)
3696 {
3697         struct packet_type *pt = &bond->arp_mon_pt;
3698
3699         dev_remove_pack(pt);
3700         pt->type = 0;
3701 }
3702
3703 /*---------------------------- Hashing Policies -----------------------------*/
3704
3705 /*
3706  * Hash for the output device based upon layer 2 and layer 3 data. If
3707  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3708  */
3709 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3710                                      struct net_device *bond_dev, int count)
3711 {
3712         struct ethhdr *data = (struct ethhdr *)skb->data;
3713         struct iphdr *iph = ip_hdr(skb);
3714
3715         if (skb->protocol == htons(ETH_P_IP)) {
3716                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3717                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3718         }
3719
3720         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3721 }
3722
3723 /*
3724  * Hash for the output device based upon layer 3 and layer 4 data. If
3725  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3726  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3727  */
3728 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3729                                     struct net_device *bond_dev, int count)
3730 {
3731         struct ethhdr *data = (struct ethhdr *)skb->data;
3732         struct iphdr *iph = ip_hdr(skb);
3733         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3734         int layer4_xor = 0;
3735
3736         if (skb->protocol == htons(ETH_P_IP)) {
3737                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3738                     (iph->protocol == IPPROTO_TCP ||
3739                      iph->protocol == IPPROTO_UDP)) {
3740                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3741                 }
3742                 return (layer4_xor ^
3743                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3744
3745         }
3746
3747         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3748 }
3749
3750 /*
3751  * Hash for the output device based upon layer 2 data
3752  */
3753 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3754                                    struct net_device *bond_dev, int count)
3755 {
3756         struct ethhdr *data = (struct ethhdr *)skb->data;
3757
3758         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3759 }
3760
3761 /*-------------------------- Device entry points ----------------------------*/
3762
3763 static int bond_open(struct net_device *bond_dev)
3764 {
3765         struct bonding *bond = netdev_priv(bond_dev);
3766
3767         bond->kill_timers = 0;
3768
3769         if (bond_is_lb(bond)) {
3770                 /* bond_alb_initialize must be called before the timer
3771                  * is started.
3772                  */
3773                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3774                         /* something went wrong - fail the open operation */
3775                         return -1;
3776                 }
3777
3778                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3779                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3780         }
3781
3782         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3783                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3784                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3785         }
3786
3787         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3788                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3789                         INIT_DELAYED_WORK(&bond->arp_work,
3790                                           bond_activebackup_arp_mon);
3791                 else
3792                         INIT_DELAYED_WORK(&bond->arp_work,
3793                                           bond_loadbalance_arp_mon);
3794
3795                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3796                 if (bond->params.arp_validate)
3797                         bond_register_arp(bond);
3798         }
3799
3800         if (bond->params.mode == BOND_MODE_8023AD) {
3801                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3802                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3803                 /* register to receive LACPDUs */
3804                 bond_register_lacpdu(bond);
3805                 bond_3ad_initiate_agg_selection(bond, 1);
3806         }
3807
3808         return 0;
3809 }
3810
3811 static int bond_close(struct net_device *bond_dev)
3812 {
3813         struct bonding *bond = netdev_priv(bond_dev);
3814
3815         if (bond->params.mode == BOND_MODE_8023AD) {
3816                 /* Unregister the receive of LACPDUs */
3817                 bond_unregister_lacpdu(bond);
3818         }
3819
3820         if (bond->params.arp_validate)
3821                 bond_unregister_arp(bond);
3822
3823         write_lock_bh(&bond->lock);
3824
3825         bond->send_grat_arp = 0;
3826         bond->send_unsol_na = 0;
3827
3828         /* signal timers not to re-arm */
3829         bond->kill_timers = 1;
3830
3831         write_unlock_bh(&bond->lock);
3832
3833         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3834                 cancel_delayed_work(&bond->mii_work);
3835         }
3836
3837         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3838                 cancel_delayed_work(&bond->arp_work);
3839         }
3840
3841         switch (bond->params.mode) {
3842         case BOND_MODE_8023AD:
3843                 cancel_delayed_work(&bond->ad_work);
3844                 break;
3845         case BOND_MODE_TLB:
3846         case BOND_MODE_ALB:
3847                 cancel_delayed_work(&bond->alb_work);
3848                 break;
3849         default:
3850                 break;
3851         }
3852
3853
3854         if (bond_is_lb(bond)) {
3855                 /* Must be called only after all
3856                  * slaves have been released
3857                  */
3858                 bond_alb_deinitialize(bond);
3859         }
3860
3861         return 0;
3862 }
3863
3864 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3865 {
3866         struct bonding *bond = netdev_priv(bond_dev);
3867         struct net_device_stats *stats = &bond->stats;
3868         struct net_device_stats local_stats;
3869         struct slave *slave;
3870         int i;
3871
3872         memset(&local_stats, 0, sizeof(struct net_device_stats));
3873
3874         read_lock_bh(&bond->lock);
3875
3876         bond_for_each_slave(bond, slave, i) {
3877                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3878
3879                 local_stats.rx_packets += sstats->rx_packets;
3880                 local_stats.rx_bytes += sstats->rx_bytes;
3881                 local_stats.rx_errors += sstats->rx_errors;
3882                 local_stats.rx_dropped += sstats->rx_dropped;
3883
3884                 local_stats.tx_packets += sstats->tx_packets;
3885                 local_stats.tx_bytes += sstats->tx_bytes;
3886                 local_stats.tx_errors += sstats->tx_errors;
3887                 local_stats.tx_dropped += sstats->tx_dropped;
3888
3889                 local_stats.multicast += sstats->multicast;
3890                 local_stats.collisions += sstats->collisions;
3891
3892                 local_stats.rx_length_errors += sstats->rx_length_errors;
3893                 local_stats.rx_over_errors += sstats->rx_over_errors;
3894                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3895                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3896                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3897                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3898
3899                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3900                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3901                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3902                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3903                 local_stats.tx_window_errors += sstats->tx_window_errors;
3904         }
3905
3906         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3907
3908         read_unlock_bh(&bond->lock);
3909
3910         return stats;
3911 }
3912
3913 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3914 {
3915         struct net_device *slave_dev = NULL;
3916         struct ifbond k_binfo;
3917         struct ifbond __user *u_binfo = NULL;
3918         struct ifslave k_sinfo;
3919         struct ifslave __user *u_sinfo = NULL;
3920         struct mii_ioctl_data *mii = NULL;
3921         int res = 0;
3922
3923         pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3924                 bond_dev->name, cmd);
3925
3926         switch (cmd) {
3927         case SIOCGMIIPHY:
3928                 mii = if_mii(ifr);
3929                 if (!mii)
3930                         return -EINVAL;
3931
3932                 mii->phy_id = 0;
3933                 /* Fall Through */
3934         case SIOCGMIIREG:
3935                 /*
3936                  * We do this again just in case we were called by SIOCGMIIREG
3937                  * instead of SIOCGMIIPHY.
3938                  */
3939                 mii = if_mii(ifr);
3940                 if (!mii)
3941                         return -EINVAL;
3942
3943
3944                 if (mii->reg_num == 1) {
3945                         struct bonding *bond = netdev_priv(bond_dev);
3946                         mii->val_out = 0;
3947                         read_lock(&bond->lock);
3948                         read_lock(&bond->curr_slave_lock);
3949                         if (netif_carrier_ok(bond->dev))
3950                                 mii->val_out = BMSR_LSTATUS;
3951
3952                         read_unlock(&bond->curr_slave_lock);
3953                         read_unlock(&bond->lock);
3954                 }
3955
3956                 return 0;
3957         case BOND_INFO_QUERY_OLD:
3958         case SIOCBONDINFOQUERY:
3959                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3960
3961                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3962                         return -EFAULT;
3963
3964                 res = bond_info_query(bond_dev, &k_binfo);
3965                 if (res == 0 &&
3966                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3967                         return -EFAULT;
3968
3969                 return res;
3970         case BOND_SLAVE_INFO_QUERY_OLD:
3971         case SIOCBONDSLAVEINFOQUERY:
3972                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3973
3974                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3975                         return -EFAULT;
3976
3977                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3978                 if (res == 0 &&
3979                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3980                         return -EFAULT;
3981
3982                 return res;
3983         default:
3984                 /* Go on */
3985                 break;
3986         }
3987
3988         if (!capable(CAP_NET_ADMIN))
3989                 return -EPERM;
3990
3991         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3992
3993         pr_debug("slave_dev=%p: \n", slave_dev);
3994
3995         if (!slave_dev)
3996                 res = -ENODEV;
3997         else {
3998                 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
3999                 switch (cmd) {
4000                 case BOND_ENSLAVE_OLD:
4001                 case SIOCBONDENSLAVE:
4002                         res = bond_enslave(bond_dev, slave_dev);
4003                         break;
4004                 case BOND_RELEASE_OLD:
4005                 case SIOCBONDRELEASE:
4006                         res = bond_release(bond_dev, slave_dev);
4007                         break;
4008                 case BOND_SETHWADDR_OLD:
4009                 case SIOCBONDSETHWADDR:
4010                         res = bond_sethwaddr(bond_dev, slave_dev);
4011                         break;
4012                 case BOND_CHANGE_ACTIVE_OLD:
4013                 case SIOCBONDCHANGEACTIVE:
4014                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4015                         break;
4016                 default:
4017                         res = -EOPNOTSUPP;
4018                 }
4019
4020                 dev_put(slave_dev);
4021         }
4022
4023         return res;
4024 }
4025
4026 static void bond_set_multicast_list(struct net_device *bond_dev)
4027 {
4028         struct bonding *bond = netdev_priv(bond_dev);
4029         struct dev_mc_list *dmi;
4030
4031         /*
4032          * Do promisc before checking multicast_mode
4033          */
4034         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4035                 /*
4036                  * FIXME: Need to handle the error when one of the multi-slaves
4037                  * encounters error.
4038                  */
4039                 bond_set_promiscuity(bond, 1);
4040
4041
4042         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4043                 bond_set_promiscuity(bond, -1);
4044
4045
4046         /* set allmulti flag to slaves */
4047         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4048                 /*
4049                  * FIXME: Need to handle the error when one of the multi-slaves
4050                  * encounters error.
4051                  */
4052                 bond_set_allmulti(bond, 1);
4053
4054
4055         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4056                 bond_set_allmulti(bond, -1);
4057
4058
4059         read_lock(&bond->lock);
4060
4061         bond->flags = bond_dev->flags;
4062
4063         /* looking for addresses to add to slaves' mc list */
4064         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4065                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4066                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4067         }
4068
4069         /* looking for addresses to delete from slaves' list */
4070         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4071                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4072                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4073         }
4074
4075         /* save master's multicast list */
4076         bond_mc_list_destroy(bond);
4077         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4078
4079         read_unlock(&bond->lock);
4080 }
4081
4082 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4083 {
4084         struct bonding *bond = netdev_priv(dev);
4085         struct slave *slave = bond->first_slave;
4086
4087         if (slave) {
4088                 const struct net_device_ops *slave_ops
4089                         = slave->dev->netdev_ops;
4090                 if (slave_ops->ndo_neigh_setup)
4091                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4092         }
4093         return 0;
4094 }
4095
4096 /*
4097  * Change the MTU of all of a master's slaves to match the master
4098  */
4099 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4100 {
4101         struct bonding *bond = netdev_priv(bond_dev);
4102         struct slave *slave, *stop_at;
4103         int res = 0;
4104         int i;
4105
4106         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4107                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4108
4109         /* Can't hold bond->lock with bh disabled here since
4110          * some base drivers panic. On the other hand we can't
4111          * hold bond->lock without bh disabled because we'll
4112          * deadlock. The only solution is to rely on the fact
4113          * that we're under rtnl_lock here, and the slaves
4114          * list won't change. This doesn't solve the problem
4115          * of setting the slave's MTU while it is
4116          * transmitting, but the assumption is that the base
4117          * driver can handle that.
4118          *
4119          * TODO: figure out a way to safely iterate the slaves
4120          * list, but without holding a lock around the actual
4121          * call to the base driver.
4122          */
4123
4124         bond_for_each_slave(bond, slave, i) {
4125                 pr_debug("s %p s->p %p c_m %p\n", slave,
4126                         slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4127
4128                 res = dev_set_mtu(slave->dev, new_mtu);
4129
4130                 if (res) {
4131                         /* If we failed to set the slave's mtu to the new value
4132                          * we must abort the operation even in ACTIVE_BACKUP
4133                          * mode, because if we allow the backup slaves to have
4134                          * different mtu values than the active slave we'll
4135                          * need to change their mtu when doing a failover. That
4136                          * means changing their mtu from timer context, which
4137                          * is probably not a good idea.
4138                          */
4139                         pr_debug("err %d %s\n", res, slave->dev->name);
4140                         goto unwind;
4141                 }
4142         }
4143
4144         bond_dev->mtu = new_mtu;
4145
4146         return 0;
4147
4148 unwind:
4149         /* unwind from head to the slave that failed */
4150         stop_at = slave;
4151         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4152                 int tmp_res;
4153
4154                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4155                 if (tmp_res) {
4156                         pr_debug("unwind err %d dev %s\n", tmp_res,
4157                                 slave->dev->name);
4158                 }
4159         }
4160
4161         return res;
4162 }
4163
4164 /*
4165  * Change HW address
4166  *
4167  * Note that many devices must be down to change the HW address, and
4168  * downing the master releases all slaves.  We can make bonds full of
4169  * bonding devices to test this, however.
4170  */
4171 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4172 {
4173         struct bonding *bond = netdev_priv(bond_dev);
4174         struct sockaddr *sa = addr, tmp_sa;
4175         struct slave *slave, *stop_at;
4176         int res = 0;
4177         int i;
4178
4179         if (bond->params.mode == BOND_MODE_ALB)
4180                 return bond_alb_set_mac_address(bond_dev, addr);
4181
4182
4183         pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4184
4185         /*
4186          * If fail_over_mac is set to active, do nothing and return
4187          * success.  Returning an error causes ifenslave to fail.
4188          */
4189         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4190                 return 0;
4191
4192         if (!is_valid_ether_addr(sa->sa_data))
4193                 return -EADDRNOTAVAIL;
4194
4195         /* Can't hold bond->lock with bh disabled here since
4196          * some base drivers panic. On the other hand we can't
4197          * hold bond->lock without bh disabled because we'll
4198          * deadlock. The only solution is to rely on the fact
4199          * that we're under rtnl_lock here, and the slaves
4200          * list won't change. This doesn't solve the problem
4201          * of setting the slave's hw address while it is
4202          * transmitting, but the assumption is that the base
4203          * driver can handle that.
4204          *
4205          * TODO: figure out a way to safely iterate the slaves
4206          * list, but without holding a lock around the actual
4207          * call to the base driver.
4208          */
4209
4210         bond_for_each_slave(bond, slave, i) {
4211                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4212                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4213
4214                 if (slave_ops->ndo_set_mac_address == NULL) {
4215                         res = -EOPNOTSUPP;
4216                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4217                         goto unwind;
4218                 }
4219
4220                 res = dev_set_mac_address(slave->dev, addr);
4221                 if (res) {
4222                         /* TODO: consider downing the slave
4223                          * and retry ?
4224                          * User should expect communications
4225                          * breakage anyway until ARP finish
4226                          * updating, so...
4227                          */
4228                         pr_debug("err %d %s\n", res, slave->dev->name);
4229                         goto unwind;
4230                 }
4231         }
4232
4233         /* success */
4234         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4235         return 0;
4236
4237 unwind:
4238         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4239         tmp_sa.sa_family = bond_dev->type;
4240
4241         /* unwind from head to the slave that failed */
4242         stop_at = slave;
4243         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4244                 int tmp_res;
4245
4246                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4247                 if (tmp_res) {
4248                         pr_debug("unwind err %d dev %s\n", tmp_res,
4249                                 slave->dev->name);
4250                 }
4251         }
4252
4253         return res;
4254 }
4255
4256 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4257 {
4258         struct bonding *bond = netdev_priv(bond_dev);
4259         struct slave *slave, *start_at;
4260         int i, slave_no, res = 1;
4261
4262         read_lock(&bond->lock);
4263
4264         if (!BOND_IS_OK(bond))
4265                 goto out;
4266
4267         /*
4268          * Concurrent TX may collide on rr_tx_counter; we accept that
4269          * as being rare enough not to justify using an atomic op here
4270          */
4271         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4272
4273         bond_for_each_slave(bond, slave, i) {
4274                 slave_no--;
4275                 if (slave_no < 0)
4276                         break;
4277         }
4278
4279         start_at = slave;
4280         bond_for_each_slave_from(bond, slave, i, start_at) {
4281                 if (IS_UP(slave->dev) &&
4282                     (slave->link == BOND_LINK_UP) &&
4283                     (slave->state == BOND_STATE_ACTIVE)) {
4284                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4285                         break;
4286                 }
4287         }
4288
4289 out:
4290         if (res) {
4291                 /* no suitable interface, frame not sent */
4292                 dev_kfree_skb(skb);
4293         }
4294         read_unlock(&bond->lock);
4295         return NETDEV_TX_OK;
4296 }
4297
4298
4299 /*
4300  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4301  * the bond has a usable interface.
4302  */
4303 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4304 {
4305         struct bonding *bond = netdev_priv(bond_dev);
4306         int res = 1;
4307
4308         read_lock(&bond->lock);
4309         read_lock(&bond->curr_slave_lock);
4310
4311         if (!BOND_IS_OK(bond))
4312                 goto out;
4313
4314         if (!bond->curr_active_slave)
4315                 goto out;
4316
4317         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4318
4319 out:
4320         if (res)
4321                 /* no suitable interface, frame not sent */
4322                 dev_kfree_skb(skb);
4323
4324         read_unlock(&bond->curr_slave_lock);
4325         read_unlock(&bond->lock);
4326         return NETDEV_TX_OK;
4327 }
4328
4329 /*
4330  * In bond_xmit_xor() , we determine the output device by using a pre-
4331  * determined xmit_hash_policy(), If the selected device is not enabled,
4332  * find the next active slave.
4333  */
4334 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4335 {
4336         struct bonding *bond = netdev_priv(bond_dev);
4337         struct slave *slave, *start_at;
4338         int slave_no;
4339         int i;
4340         int res = 1;
4341
4342         read_lock(&bond->lock);
4343
4344         if (!BOND_IS_OK(bond))
4345                 goto out;
4346
4347         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4348
4349         bond_for_each_slave(bond, slave, i) {
4350                 slave_no--;
4351                 if (slave_no < 0)
4352                         break;
4353         }
4354
4355         start_at = slave;
4356
4357         bond_for_each_slave_from(bond, slave, i, start_at) {
4358                 if (IS_UP(slave->dev) &&
4359                     (slave->link == BOND_LINK_UP) &&
4360                     (slave->state == BOND_STATE_ACTIVE)) {
4361                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4362                         break;
4363                 }
4364         }
4365
4366 out:
4367         if (res) {
4368                 /* no suitable interface, frame not sent */
4369                 dev_kfree_skb(skb);
4370         }
4371         read_unlock(&bond->lock);
4372         return NETDEV_TX_OK;
4373 }
4374
4375 /*
4376  * in broadcast mode, we send everything to all usable interfaces.
4377  */
4378 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4379 {
4380         struct bonding *bond = netdev_priv(bond_dev);
4381         struct slave *slave, *start_at;
4382         struct net_device *tx_dev = NULL;
4383         int i;
4384         int res = 1;
4385
4386         read_lock(&bond->lock);
4387
4388         if (!BOND_IS_OK(bond))
4389                 goto out;
4390
4391         read_lock(&bond->curr_slave_lock);
4392         start_at = bond->curr_active_slave;
4393         read_unlock(&bond->curr_slave_lock);
4394
4395         if (!start_at)
4396                 goto out;
4397
4398         bond_for_each_slave_from(bond, slave, i, start_at) {
4399                 if (IS_UP(slave->dev) &&
4400                     (slave->link == BOND_LINK_UP) &&
4401                     (slave->state == BOND_STATE_ACTIVE)) {
4402                         if (tx_dev) {
4403                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4404                                 if (!skb2) {
4405                                         pr_err(DRV_NAME
4406                                                ": %s: Error: bond_xmit_broadcast(): "
4407                                                "skb_clone() failed\n",
4408                                                bond_dev->name);
4409                                         continue;
4410                                 }
4411
4412                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4413                                 if (res) {
4414                                         dev_kfree_skb(skb2);
4415                                         continue;
4416                                 }
4417                         }
4418                         tx_dev = slave->dev;
4419                 }
4420         }
4421
4422         if (tx_dev)
4423                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4424
4425 out:
4426         if (res)
4427                 /* no suitable interface, frame not sent */
4428                 dev_kfree_skb(skb);
4429
4430         /* frame sent to all suitable interfaces */
4431         read_unlock(&bond->lock);
4432         return NETDEV_TX_OK;
4433 }
4434
4435 /*------------------------- Device initialization ---------------------------*/
4436
4437 static void bond_set_xmit_hash_policy(struct bonding *bond)
4438 {
4439         switch (bond->params.xmit_policy) {
4440         case BOND_XMIT_POLICY_LAYER23:
4441                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4442                 break;
4443         case BOND_XMIT_POLICY_LAYER34:
4444                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4445                 break;
4446         case BOND_XMIT_POLICY_LAYER2:
4447         default:
4448                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4449                 break;
4450         }
4451 }
4452
4453 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4454 {
4455         const struct bonding *bond = netdev_priv(dev);
4456
4457         switch (bond->params.mode) {
4458         case BOND_MODE_ROUNDROBIN:
4459                 return bond_xmit_roundrobin(skb, dev);
4460         case BOND_MODE_ACTIVEBACKUP:
4461                 return bond_xmit_activebackup(skb, dev);
4462         case BOND_MODE_XOR:
4463                 return bond_xmit_xor(skb, dev);
4464         case BOND_MODE_BROADCAST:
4465                 return bond_xmit_broadcast(skb, dev);
4466         case BOND_MODE_8023AD:
4467                 return bond_3ad_xmit_xor(skb, dev);
4468         case BOND_MODE_ALB:
4469         case BOND_MODE_TLB:
4470                 return bond_alb_xmit(skb, dev);
4471         default:
4472                 /* Should never happen, mode already checked */
4473                 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4474                      dev->name, bond->params.mode);
4475                 WARN_ON_ONCE(1);
4476                 dev_kfree_skb(skb);
4477                 return NETDEV_TX_OK;
4478         }
4479 }
4480
4481
4482 /*
4483  * set bond mode specific net device operations
4484  */
4485 void bond_set_mode_ops(struct bonding *bond, int mode)
4486 {
4487         struct net_device *bond_dev = bond->dev;
4488
4489         switch (mode) {
4490         case BOND_MODE_ROUNDROBIN:
4491                 break;
4492         case BOND_MODE_ACTIVEBACKUP:
4493                 break;
4494         case BOND_MODE_XOR:
4495                 bond_set_xmit_hash_policy(bond);
4496                 break;
4497         case BOND_MODE_BROADCAST:
4498                 break;
4499         case BOND_MODE_8023AD:
4500                 bond_set_master_3ad_flags(bond);
4501                 bond_set_xmit_hash_policy(bond);
4502                 break;
4503         case BOND_MODE_ALB:
4504                 bond_set_master_alb_flags(bond);
4505                 /* FALLTHRU */
4506         case BOND_MODE_TLB:
4507                 break;
4508         default:
4509                 /* Should never happen, mode already checked */
4510                 pr_err(DRV_NAME
4511                        ": %s: Error: Unknown bonding mode %d\n",
4512                        bond_dev->name,
4513                        mode);
4514                 break;
4515         }
4516 }
4517
4518 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4519                                     struct ethtool_drvinfo *drvinfo)
4520 {
4521         strncpy(drvinfo->driver, DRV_NAME, 32);
4522         strncpy(drvinfo->version, DRV_VERSION, 32);
4523         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4524 }
4525
4526 static const struct ethtool_ops bond_ethtool_ops = {
4527         .get_drvinfo            = bond_ethtool_get_drvinfo,
4528         .get_link               = ethtool_op_get_link,
4529         .get_tx_csum            = ethtool_op_get_tx_csum,
4530         .get_sg                 = ethtool_op_get_sg,
4531         .get_tso                = ethtool_op_get_tso,
4532         .get_ufo                = ethtool_op_get_ufo,
4533         .get_flags              = ethtool_op_get_flags,
4534 };
4535
4536 static const struct net_device_ops bond_netdev_ops = {
4537         .ndo_init               = bond_init,
4538         .ndo_uninit             = bond_uninit,
4539         .ndo_open               = bond_open,
4540         .ndo_stop               = bond_close,
4541         .ndo_start_xmit         = bond_start_xmit,
4542         .ndo_get_stats          = bond_get_stats,
4543         .ndo_do_ioctl           = bond_do_ioctl,
4544         .ndo_set_multicast_list = bond_set_multicast_list,
4545         .ndo_change_mtu         = bond_change_mtu,
4546         .ndo_set_mac_address    = bond_set_mac_address,
4547         .ndo_neigh_setup        = bond_neigh_setup,
4548         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4549         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4550         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4551 };
4552
4553 static void bond_setup(struct net_device *bond_dev)
4554 {
4555         struct bonding *bond = netdev_priv(bond_dev);
4556
4557         /* initialize rwlocks */
4558         rwlock_init(&bond->lock);
4559         rwlock_init(&bond->curr_slave_lock);
4560
4561         bond->params = bonding_defaults;
4562
4563         /* Initialize pointers */
4564         bond->dev = bond_dev;
4565         INIT_LIST_HEAD(&bond->vlan_list);
4566
4567         /* Initialize the device entry points */
4568         ether_setup(bond_dev);
4569         bond_dev->netdev_ops = &bond_netdev_ops;
4570         bond_dev->ethtool_ops = &bond_ethtool_ops;
4571         bond_set_mode_ops(bond, bond->params.mode);
4572
4573         bond_dev->destructor = free_netdev;
4574
4575         /* Initialize the device options */
4576         bond_dev->tx_queue_len = 0;
4577         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4578         bond_dev->priv_flags |= IFF_BONDING;
4579         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4580
4581         if (bond->params.arp_interval)
4582                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4583
4584         /* At first, we block adding VLANs. That's the only way to
4585          * prevent problems that occur when adding VLANs over an
4586          * empty bond. The block will be removed once non-challenged
4587          * slaves are enslaved.
4588          */
4589         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4590
4591         /* don't acquire bond device's netif_tx_lock when
4592          * transmitting */
4593         bond_dev->features |= NETIF_F_LLTX;
4594
4595         /* By default, we declare the bond to be fully
4596          * VLAN hardware accelerated capable. Special
4597          * care is taken in the various xmit functions
4598          * when there are slaves that are not hw accel
4599          * capable
4600          */
4601         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4602                                NETIF_F_HW_VLAN_RX |
4603                                NETIF_F_HW_VLAN_FILTER);
4604
4605 }
4606
4607 static void bond_work_cancel_all(struct bonding *bond)
4608 {
4609         write_lock_bh(&bond->lock);
4610         bond->kill_timers = 1;
4611         write_unlock_bh(&bond->lock);
4612
4613         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4614                 cancel_delayed_work(&bond->mii_work);
4615
4616         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4617                 cancel_delayed_work(&bond->arp_work);
4618
4619         if (bond->params.mode == BOND_MODE_ALB &&
4620             delayed_work_pending(&bond->alb_work))
4621                 cancel_delayed_work(&bond->alb_work);
4622
4623         if (bond->params.mode == BOND_MODE_8023AD &&
4624             delayed_work_pending(&bond->ad_work))
4625                 cancel_delayed_work(&bond->ad_work);
4626 }
4627
4628 /* De-initialize device specific data.
4629  * Caller must hold rtnl_lock.
4630  */
4631 static void bond_deinit(struct net_device *bond_dev)
4632 {
4633         struct bonding *bond = netdev_priv(bond_dev);
4634
4635         list_del(&bond->bond_list);
4636
4637         bond_work_cancel_all(bond);
4638
4639         bond_remove_proc_entry(bond);
4640 }
4641
4642 /* Unregister and free all bond devices.
4643  * Caller must hold rtnl_lock.
4644  */
4645 static void bond_free_all(void)
4646 {
4647         struct bonding *bond, *nxt;
4648
4649         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4650                 struct net_device *bond_dev = bond->dev;
4651
4652                 bond_work_cancel_all(bond);
4653                 /* Release the bonded slaves */
4654                 bond_release_all(bond_dev);
4655                 unregister_netdevice(bond_dev);
4656         }
4657
4658         bond_destroy_proc_dir();
4659 }
4660
4661 /*------------------------- Module initialization ---------------------------*/
4662
4663 /*
4664  * Convert string input module parms.  Accept either the
4665  * number of the mode or its string name.  A bit complicated because
4666  * some mode names are substrings of other names, and calls from sysfs
4667  * may have whitespace in the name (trailing newlines, for example).
4668  */
4669 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4670 {
4671         int modeint = -1, i, rv;
4672         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4673
4674         for (p = (char *)buf; *p; p++)
4675                 if (!(isdigit(*p) || isspace(*p)))
4676                         break;
4677
4678         if (*p)
4679                 rv = sscanf(buf, "%20s", modestr);
4680         else
4681                 rv = sscanf(buf, "%d", &modeint);
4682
4683         if (!rv)
4684                 return -1;
4685
4686         for (i = 0; tbl[i].modename; i++) {
4687                 if (modeint == tbl[i].mode)
4688                         return tbl[i].mode;
4689                 if (strcmp(modestr, tbl[i].modename) == 0)
4690                         return tbl[i].mode;
4691         }
4692
4693         return -1;
4694 }
4695
4696 static int bond_check_params(struct bond_params *params)
4697 {
4698         int arp_validate_value, fail_over_mac_value;
4699
4700         /*
4701          * Convert string parameters.
4702          */
4703         if (mode) {
4704                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4705                 if (bond_mode == -1) {
4706                         pr_err(DRV_NAME
4707                                ": Error: Invalid bonding mode \"%s\"\n",
4708                                mode == NULL ? "NULL" : mode);
4709                         return -EINVAL;
4710                 }
4711         }
4712
4713         if (xmit_hash_policy) {
4714                 if ((bond_mode != BOND_MODE_XOR) &&
4715                     (bond_mode != BOND_MODE_8023AD)) {
4716                         pr_info(DRV_NAME
4717                                ": xor_mode param is irrelevant in mode %s\n",
4718                                bond_mode_name(bond_mode));
4719                 } else {
4720                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4721                                                         xmit_hashtype_tbl);
4722                         if (xmit_hashtype == -1) {
4723                                 pr_err(DRV_NAME
4724                                        ": Error: Invalid xmit_hash_policy \"%s\"\n",
4725                                        xmit_hash_policy == NULL ? "NULL" :
4726                                        xmit_hash_policy);
4727                                 return -EINVAL;
4728                         }
4729                 }
4730         }
4731
4732         if (lacp_rate) {
4733                 if (bond_mode != BOND_MODE_8023AD) {
4734                         pr_info(DRV_NAME
4735                                ": lacp_rate param is irrelevant in mode %s\n",
4736                                bond_mode_name(bond_mode));
4737                 } else {
4738                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4739                         if (lacp_fast == -1) {
4740                                 pr_err(DRV_NAME
4741                                        ": Error: Invalid lacp rate \"%s\"\n",
4742                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4743                                 return -EINVAL;
4744                         }
4745                 }
4746         }
4747
4748         if (ad_select) {
4749                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4750                 if (params->ad_select == -1) {
4751                         pr_err(DRV_NAME
4752                                ": Error: Invalid ad_select \"%s\"\n",
4753                                ad_select == NULL ? "NULL" : ad_select);
4754                         return -EINVAL;
4755                 }
4756
4757                 if (bond_mode != BOND_MODE_8023AD) {
4758                         pr_warning(DRV_NAME
4759                                ": ad_select param only affects 802.3ad mode\n");
4760                 }
4761         } else {
4762                 params->ad_select = BOND_AD_STABLE;
4763         }
4764
4765         if (max_bonds < 0) {
4766                 pr_warning(DRV_NAME
4767                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4768                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4769                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4770                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4771         }
4772
4773         if (miimon < 0) {
4774                 pr_warning(DRV_NAME
4775                        ": Warning: miimon module parameter (%d), "
4776                        "not in range 0-%d, so it was reset to %d\n",
4777                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4778                 miimon = BOND_LINK_MON_INTERV;
4779         }
4780
4781         if (updelay < 0) {
4782                 pr_warning(DRV_NAME
4783                        ": Warning: updelay module parameter (%d), "
4784                        "not in range 0-%d, so it was reset to 0\n",
4785                        updelay, INT_MAX);
4786                 updelay = 0;
4787         }
4788
4789         if (downdelay < 0) {
4790                 pr_warning(DRV_NAME
4791                        ": Warning: downdelay module parameter (%d), "
4792                        "not in range 0-%d, so it was reset to 0\n",
4793                        downdelay, INT_MAX);
4794                 downdelay = 0;
4795         }
4796
4797         if ((use_carrier != 0) && (use_carrier != 1)) {
4798                 pr_warning(DRV_NAME
4799                        ": Warning: use_carrier module parameter (%d), "
4800                        "not of valid value (0/1), so it was set to 1\n",
4801                        use_carrier);
4802                 use_carrier = 1;
4803         }
4804
4805         if (num_grat_arp < 0 || num_grat_arp > 255) {
4806                 pr_warning(DRV_NAME
4807                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4808                        "was reset to 1 \n", num_grat_arp);
4809                 num_grat_arp = 1;
4810         }
4811
4812         if (num_unsol_na < 0 || num_unsol_na > 255) {
4813                 pr_warning(DRV_NAME
4814                        ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4815                        "was reset to 1 \n", num_unsol_na);
4816                 num_unsol_na = 1;
4817         }
4818
4819         /* reset values for 802.3ad */
4820         if (bond_mode == BOND_MODE_8023AD) {
4821                 if (!miimon) {
4822                         pr_warning(DRV_NAME
4823                                ": Warning: miimon must be specified, "
4824                                "otherwise bonding will not detect link "
4825                                "failure, speed and duplex which are "
4826                                "essential for 802.3ad operation\n");
4827                         pr_warning("Forcing miimon to 100msec\n");
4828                         miimon = 100;
4829                 }
4830         }
4831
4832         /* reset values for TLB/ALB */
4833         if ((bond_mode == BOND_MODE_TLB) ||
4834             (bond_mode == BOND_MODE_ALB)) {
4835                 if (!miimon) {
4836                         pr_warning(DRV_NAME
4837                                ": Warning: miimon must be specified, "
4838                                "otherwise bonding will not detect link "
4839                                "failure and link speed which are essential "
4840                                "for TLB/ALB load balancing\n");
4841                         pr_warning("Forcing miimon to 100msec\n");
4842                         miimon = 100;
4843                 }
4844         }
4845
4846         if (bond_mode == BOND_MODE_ALB) {
4847                 pr_notice(DRV_NAME
4848                        ": In ALB mode you might experience client "
4849                        "disconnections upon reconnection of a link if the "
4850                        "bonding module updelay parameter (%d msec) is "
4851                        "incompatible with the forwarding delay time of the "
4852                        "switch\n",
4853                        updelay);
4854         }
4855
4856         if (!miimon) {
4857                 if (updelay || downdelay) {
4858                         /* just warn the user the up/down delay will have
4859                          * no effect since miimon is zero...
4860                          */
4861                         pr_warning(DRV_NAME
4862                                ": Warning: miimon module parameter not set "
4863                                "and updelay (%d) or downdelay (%d) module "
4864                                "parameter is set; updelay and downdelay have "
4865                                "no effect unless miimon is set\n",
4866                                updelay, downdelay);
4867                 }
4868         } else {
4869                 /* don't allow arp monitoring */
4870                 if (arp_interval) {
4871                         pr_warning(DRV_NAME
4872                                ": Warning: miimon (%d) and arp_interval (%d) "
4873                                "can't be used simultaneously, disabling ARP "
4874                                "monitoring\n",
4875                                miimon, arp_interval);
4876                         arp_interval = 0;
4877                 }
4878
4879                 if ((updelay % miimon) != 0) {
4880                         pr_warning(DRV_NAME
4881                                ": Warning: updelay (%d) is not a multiple "
4882                                "of miimon (%d), updelay rounded to %d ms\n",
4883                                updelay, miimon, (updelay / miimon) * miimon);
4884                 }
4885
4886                 updelay /= miimon;
4887
4888                 if ((downdelay % miimon) != 0) {
4889                         pr_warning(DRV_NAME
4890                                ": Warning: downdelay (%d) is not a multiple "
4891                                "of miimon (%d), downdelay rounded to %d ms\n",
4892                                downdelay, miimon,
4893                                (downdelay / miimon) * miimon);
4894                 }
4895
4896                 downdelay /= miimon;
4897         }
4898
4899         if (arp_interval < 0) {
4900                 pr_warning(DRV_NAME
4901                        ": Warning: arp_interval module parameter (%d) "
4902                        ", not in range 0-%d, so it was reset to %d\n",
4903                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4904                 arp_interval = BOND_LINK_ARP_INTERV;
4905         }
4906
4907         for (arp_ip_count = 0;
4908              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4909              arp_ip_count++) {
4910                 /* not complete check, but should be good enough to
4911                    catch mistakes */
4912                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4913                         pr_warning(DRV_NAME
4914                                ": Warning: bad arp_ip_target module parameter "
4915                                "(%s), ARP monitoring will not be performed\n",
4916                                arp_ip_target[arp_ip_count]);
4917                         arp_interval = 0;
4918                 } else {
4919                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4920                         arp_target[arp_ip_count] = ip;
4921                 }
4922         }
4923
4924         if (arp_interval && !arp_ip_count) {
4925                 /* don't allow arping if no arp_ip_target given... */
4926                 pr_warning(DRV_NAME
4927                        ": Warning: arp_interval module parameter (%d) "
4928                        "specified without providing an arp_ip_target "
4929                        "parameter, arp_interval was reset to 0\n",
4930                        arp_interval);
4931                 arp_interval = 0;
4932         }
4933
4934         if (arp_validate) {
4935                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4936                         pr_err(DRV_NAME
4937                                ": arp_validate only supported in active-backup mode\n");
4938                         return -EINVAL;
4939                 }
4940                 if (!arp_interval) {
4941                         pr_err(DRV_NAME
4942                                ": arp_validate requires arp_interval\n");
4943                         return -EINVAL;
4944                 }
4945
4946                 arp_validate_value = bond_parse_parm(arp_validate,
4947                                                      arp_validate_tbl);
4948                 if (arp_validate_value == -1) {
4949                         pr_err(DRV_NAME
4950                                ": Error: invalid arp_validate \"%s\"\n",
4951                                arp_validate == NULL ? "NULL" : arp_validate);
4952                         return -EINVAL;
4953                 }
4954         } else
4955                 arp_validate_value = 0;
4956
4957         if (miimon) {
4958                 pr_info(DRV_NAME
4959                        ": MII link monitoring set to %d ms\n",
4960                        miimon);
4961         } else if (arp_interval) {
4962                 int i;
4963
4964                 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4965                        " validate %s, with %d target(s):",
4966                        arp_interval,
4967                        arp_validate_tbl[arp_validate_value].modename,
4968                        arp_ip_count);
4969
4970                 for (i = 0; i < arp_ip_count; i++)
4971                         pr_info(" %s", arp_ip_target[i]);
4972
4973                 pr_info("\n");
4974
4975         } else if (max_bonds) {
4976                 /* miimon and arp_interval not set, we need one so things
4977                  * work as expected, see bonding.txt for details
4978                  */
4979                 pr_warning(DRV_NAME
4980                        ": Warning: either miimon or arp_interval and "
4981                        "arp_ip_target module parameters must be specified, "
4982                        "otherwise bonding will not detect link failures! see "
4983                        "bonding.txt for details.\n");
4984         }
4985
4986         if (primary && !USES_PRIMARY(bond_mode)) {
4987                 /* currently, using a primary only makes sense
4988                  * in active backup, TLB or ALB modes
4989                  */
4990                 pr_warning(DRV_NAME
4991                        ": Warning: %s primary device specified but has no "
4992                        "effect in %s mode\n",
4993                        primary, bond_mode_name(bond_mode));
4994                 primary = NULL;
4995         }
4996
4997         if (fail_over_mac) {
4998                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4999                                                       fail_over_mac_tbl);
5000                 if (fail_over_mac_value == -1) {
5001                         pr_err(DRV_NAME
5002                                ": Error: invalid fail_over_mac \"%s\"\n",
5003                                arp_validate == NULL ? "NULL" : arp_validate);
5004                         return -EINVAL;
5005                 }
5006
5007                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5008                         pr_warning(DRV_NAME
5009                                ": Warning: fail_over_mac only affects "
5010                                "active-backup mode.\n");
5011         } else {
5012                 fail_over_mac_value = BOND_FOM_NONE;
5013         }
5014
5015         /* fill params struct with the proper values */
5016         params->mode = bond_mode;
5017         params->xmit_policy = xmit_hashtype;
5018         params->miimon = miimon;
5019         params->num_grat_arp = num_grat_arp;
5020         params->num_unsol_na = num_unsol_na;
5021         params->arp_interval = arp_interval;
5022         params->arp_validate = arp_validate_value;
5023         params->updelay = updelay;
5024         params->downdelay = downdelay;
5025         params->use_carrier = use_carrier;
5026         params->lacp_fast = lacp_fast;
5027         params->primary[0] = 0;
5028         params->fail_over_mac = fail_over_mac_value;
5029
5030         if (primary) {
5031                 strncpy(params->primary, primary, IFNAMSIZ);
5032                 params->primary[IFNAMSIZ - 1] = 0;
5033         }
5034
5035         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5036
5037         return 0;
5038 }
5039
5040 static struct lock_class_key bonding_netdev_xmit_lock_key;
5041 static struct lock_class_key bonding_netdev_addr_lock_key;
5042
5043 static void bond_set_lockdep_class_one(struct net_device *dev,
5044                                        struct netdev_queue *txq,
5045                                        void *_unused)
5046 {
5047         lockdep_set_class(&txq->_xmit_lock,
5048                           &bonding_netdev_xmit_lock_key);
5049 }
5050
5051 static void bond_set_lockdep_class(struct net_device *dev)
5052 {
5053         lockdep_set_class(&dev->addr_list_lock,
5054                           &bonding_netdev_addr_lock_key);
5055         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5056 }
5057
5058 /*
5059  * Called from registration process
5060  */
5061 static int bond_init(struct net_device *bond_dev)
5062 {
5063         struct bonding *bond = netdev_priv(bond_dev);
5064
5065         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5066
5067         bond->wq = create_singlethread_workqueue(bond_dev->name);
5068         if (!bond->wq)
5069                 return -ENOMEM;
5070
5071         bond_set_lockdep_class(bond_dev);
5072
5073         netif_carrier_off(bond_dev);
5074
5075         bond_create_proc_entry(bond);
5076         list_add_tail(&bond->bond_list, &bond_dev_list);
5077
5078         return 0;
5079 }
5080
5081 /* Create a new bond based on the specified name and bonding parameters.
5082  * If name is NULL, obtain a suitable "bond%d" name for us.
5083  * Caller must NOT hold rtnl_lock; we need to release it here before we
5084  * set up our sysfs entries.
5085  */
5086 int bond_create(const char *name)
5087 {
5088         struct net_device *bond_dev;
5089         int res;
5090
5091         rtnl_lock();
5092         /* Check to see if the bond already exists. */
5093         /* FIXME: pass netns from caller */
5094         if (name && __dev_get_by_name(&init_net, name)) {
5095                 pr_err(DRV_NAME ": cannot add bond %s; already exists\n",
5096                        name);
5097                 res = -EEXIST;
5098                 goto out_rtnl;
5099         }
5100
5101         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5102                                 bond_setup);
5103         if (!bond_dev) {
5104                 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5105                        name);
5106                 res = -ENOMEM;
5107                 goto out_rtnl;
5108         }
5109
5110         if (!name) {
5111                 res = dev_alloc_name(bond_dev, "bond%d");
5112                 if (res < 0)
5113                         goto out_netdev;
5114         }
5115
5116         res = register_netdevice(bond_dev);
5117         if (res < 0)
5118                 goto out_bond;
5119
5120         res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5121         if (res < 0)
5122                 goto out_unreg;
5123
5124         rtnl_unlock();
5125         return 0;
5126
5127 out_unreg:
5128         unregister_netdevice(bond_dev);
5129 out_bond:
5130         bond_deinit(bond_dev);
5131 out_netdev:
5132         free_netdev(bond_dev);
5133 out_rtnl:
5134         rtnl_unlock();
5135         return res;
5136 }
5137
5138 static int __init bonding_init(void)
5139 {
5140         int i;
5141         int res;
5142
5143         pr_info("%s", version);
5144
5145         res = bond_check_params(&bonding_defaults);
5146         if (res)
5147                 goto out;
5148
5149         bond_create_proc_dir();
5150
5151         for (i = 0; i < max_bonds; i++) {
5152                 res = bond_create(NULL);
5153                 if (res)
5154                         goto err;
5155         }
5156
5157         res = bond_create_sysfs();
5158         if (res)
5159                 goto err;
5160
5161         register_netdevice_notifier(&bond_netdev_notifier);
5162         register_inetaddr_notifier(&bond_inetaddr_notifier);
5163         bond_register_ipv6_notifier();
5164
5165         goto out;
5166 err:
5167         rtnl_lock();
5168         bond_free_all();
5169         rtnl_unlock();
5170 out:
5171         return res;
5172
5173 }
5174
5175 static void __exit bonding_exit(void)
5176 {
5177         unregister_netdevice_notifier(&bond_netdev_notifier);
5178         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5179         bond_unregister_ipv6_notifier();
5180
5181         bond_destroy_sysfs();
5182
5183         rtnl_lock();
5184         bond_free_all();
5185         rtnl_unlock();
5186 }
5187
5188 module_init(bonding_init);
5189 module_exit(bonding_exit);
5190 MODULE_LICENSE("GPL");
5191 MODULE_VERSION(DRV_VERSION);
5192 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5193 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");