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