git.oblomov.eu Git - linux-2.6/blob - 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 (bond->params.use_carrier)
 699                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 700
 701         /* Try to get link status using Ethtool first. */
 702         if (slave_dev->ethtool_ops) {
 703                 if (slave_dev->ethtool_ops->get_link) {
 704                         u32 link;
 705
 706                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 707
 708                         return link ? BMSR_LSTATUS : 0;
 709                 }
 710         }
 711
 712         /* Ethtool can't be used, fallback to MII ioctls. */
 713         ioctl = slave_ops->ndo_do_ioctl;
 714         if (ioctl) {
 715                 /* TODO: set pointer to correct ioctl on a per team member */
 716                 /*       bases to make this more efficient. that is, once  */
 717                 /*       we determine the correct ioctl, we will always    */
 718                 /*       call it and not the others for that team          */
 719                 /*       member.                                           */
 720
 721                 /*
 722                  * We cannot assume that SIOCGMIIPHY will also read a
 723                  * register; not all network drivers (e.g., e100)
 724                  * support that.
 725                  */
 726
 727                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 728                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 729                 mii = if_mii(&ifr);
 730                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 731                         mii->reg_num = MII_BMSR;
 732                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 733                                 return mii->val_out & BMSR_LSTATUS;
 734                 }
 735         }
 736
 737         /*
 738          * If reporting, report that either there's no dev->do_ioctl,
 739          * or both SIOCGMIIREG and get_link failed (meaning that we
 740          * cannot report link status).  If not reporting, pretend
 741          * we're ok.
 742          */
 743         return reporting ? -1 : BMSR_LSTATUS;
 744 }
 745
 746 /*----------------------------- Multicast list ------------------------------*/
 747
 748 /*
 749  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 750  */
 751 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
 752                                    const struct dev_mc_list *dmi2)
 753 {
 754         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 755                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 756 }
 757
 758 /*
 759  * returns dmi entry if found, NULL otherwise
 760  */
 761 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
 762                                                  struct dev_mc_list *mc_list)
 763 {
 764         struct dev_mc_list *idmi;
 765
 766         for (idmi = mc_list; idmi; idmi = idmi->next) {
 767                 if (bond_is_dmi_same(dmi, idmi))
 768                         return idmi;
 769         }
 770
 771         return NULL;
 772 }
 773
 774 /*
 775  * Push the promiscuity flag down to appropriate slaves
 776  */
 777 static int bond_set_promiscuity(struct bonding *bond, int inc)
 778 {
 779         int err = 0;
 780         if (USES_PRIMARY(bond->params.mode)) {
 781                 /* write lock already acquired */
 782                 if (bond->curr_active_slave) {
 783                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 784                                                   inc);
 785                 }
 786         } else {
 787                 struct slave *slave;
 788                 int i;
 789                 bond_for_each_slave(bond, slave, i) {
 790                         err = dev_set_promiscuity(slave->dev, inc);
 791                         if (err)
 792                                 return err;
 793                 }
 794         }
 795         return err;
 796 }
 797
 798 /*
 799  * Push the allmulti flag down to all slaves
 800  */
 801 static int bond_set_allmulti(struct bonding *bond, int inc)
 802 {
 803         int err = 0;
 804         if (USES_PRIMARY(bond->params.mode)) {
 805                 /* write lock already acquired */
 806                 if (bond->curr_active_slave) {
 807                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 808                                                inc);
 809                 }
 810         } else {
 811                 struct slave *slave;
 812                 int i;
 813                 bond_for_each_slave(bond, slave, i) {
 814                         err = dev_set_allmulti(slave->dev, inc);
 815                         if (err)
 816                                 return err;
 817                 }
 818         }
 819         return err;
 820 }
 821
 822 /*
 823  * Add a Multicast address to slaves
 824  * according to mode
 825  */
 826 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 827 {
 828         if (USES_PRIMARY(bond->params.mode)) {
 829                 /* write lock already acquired */
 830                 if (bond->curr_active_slave)
 831                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 832         } else {
 833                 struct slave *slave;
 834                 int i;
 835
 836                 bond_for_each_slave(bond, slave, i)
 837                         dev_mc_add(slave->dev, addr, alen, 0);
 838         }
 839 }
 840
 841 /*
 842  * Remove a multicast address from slave
 843  * according to mode
 844  */
 845 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 846 {
 847         if (USES_PRIMARY(bond->params.mode)) {
 848                 /* write lock already acquired */
 849                 if (bond->curr_active_slave)
 850                         dev_mc_delete(bond->curr_active_slave->dev, addr,
 851                                       alen, 0);
 852         } else {
 853                 struct slave *slave;
 854                 int i;
 855                 bond_for_each_slave(bond, slave, i) {
 856                         dev_mc_delete(slave->dev, addr, alen, 0);
 857                 }
 858         }
 859 }
 860
 861
 862 /*
 863  * Retrieve the list of registered multicast addresses for the bonding
 864  * device and retransmit an IGMP JOIN request to the current active
 865  * slave.
 866  */
 867 static void bond_resend_igmp_join_requests(struct bonding *bond)
 868 {
 869         struct in_device *in_dev;
 870         struct ip_mc_list *im;
 871
 872         rcu_read_lock();
 873         in_dev = __in_dev_get_rcu(bond->dev);
 874         if (in_dev) {
 875                 for (im = in_dev->mc_list; im; im = im->next)
 876                         ip_mc_rejoin_group(im);
 877         }
 878
 879         rcu_read_unlock();
 880 }
 881
 882 /*
 883  * Totally destroys the mc_list in bond
 884  */
 885 static void bond_mc_list_destroy(struct bonding *bond)
 886 {
 887         struct dev_mc_list *dmi;
 888
 889         dmi = bond->mc_list;
 890         while (dmi) {
 891                 bond->mc_list = dmi->next;
 892                 kfree(dmi);
 893                 dmi = bond->mc_list;
 894         }
 895
 896         bond->mc_list = NULL;
 897 }
 898
 899 /*
 900  * Copy all the Multicast addresses from src to the bonding device dst
 901  */
 902 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 903                              gfp_t gfp_flag)
 904 {
 905         struct dev_mc_list *dmi, *new_dmi;
 906
 907         for (dmi = mc_list; dmi; dmi = dmi->next) {
 908                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 909
 910                 if (!new_dmi) {
 911                         /* FIXME: Potential memory leak !!! */
 912                         return -ENOMEM;
 913                 }
 914
 915                 new_dmi->next = bond->mc_list;
 916                 bond->mc_list = new_dmi;
 917                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 918                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 919                 new_dmi->dmi_users = dmi->dmi_users;
 920                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 921         }
 922
 923         return 0;
 924 }
 925
 926 /*
 927  * flush all members of flush->mc_list from device dev->mc_list
 928  */
 929 static void bond_mc_list_flush(struct net_device *bond_dev,
 930                                struct net_device *slave_dev)
 931 {
 932         struct bonding *bond = netdev_priv(bond_dev);
 933         struct dev_mc_list *dmi;
 934
 935         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
 936                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 937
 938         if (bond->params.mode == BOND_MODE_8023AD) {
 939                 /* del lacpdu mc addr from mc list */
 940                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 941
 942                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 943         }
 944 }
 945
 946 /*--------------------------- Active slave change ---------------------------*/
 947
 948 /*
 949  * Update the mc list and multicast-related flags for the new and
 950  * old active slaves (if any) according to the multicast mode, and
 951  * promiscuous flags unconditionally.
 952  */
 953 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 954                          struct slave *old_active)
 955 {
 956         struct dev_mc_list *dmi;
 957
 958         if (!USES_PRIMARY(bond->params.mode))
 959                 /* nothing to do -  mc list is already up-to-date on
 960                  * all slaves
 961                  */
 962                 return;
 963
 964         if (old_active) {
 965                 if (bond->dev->flags & IFF_PROMISC)
 966                         dev_set_promiscuity(old_active->dev, -1);
 967
 968                 if (bond->dev->flags & IFF_ALLMULTI)
 969                         dev_set_allmulti(old_active->dev, -1);
 970
 971                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 972                         dev_mc_delete(old_active->dev, dmi->dmi_addr,
 973                                       dmi->dmi_addrlen, 0);
 974         }
 975
 976         if (new_active) {
 977                 /* FIXME: Signal errors upstream. */
 978                 if (bond->dev->flags & IFF_PROMISC)
 979                         dev_set_promiscuity(new_active->dev, 1);
 980
 981                 if (bond->dev->flags & IFF_ALLMULTI)
 982                         dev_set_allmulti(new_active->dev, 1);
 983
 984                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 985                         dev_mc_add(new_active->dev, dmi->dmi_addr,
 986                                    dmi->dmi_addrlen, 0);
 987                 bond_resend_igmp_join_requests(bond);
 988         }
 989 }
 990
 991 /*
 992  * bond_do_fail_over_mac
 993  *
 994  * Perform special MAC address swapping for fail_over_mac settings
 995  *
 996  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 997  */
 998 static void bond_do_fail_over_mac(struct bonding *bond,
 999                                   struct slave *new_active,
1000                                   struct slave *old_active)
1001         __releases(&bond->curr_slave_lock)
1002         __releases(&bond->lock)
1003         __acquires(&bond->lock)
1004         __acquires(&bond->curr_slave_lock)
1005 {
1006         u8 tmp_mac[ETH_ALEN];
1007         struct sockaddr saddr;
1008         int rv;
1009
1010         switch (bond->params.fail_over_mac) {
1011         case BOND_FOM_ACTIVE:
1012                 if (new_active)
1013                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1014                                new_active->dev->addr_len);
1015                 break;
1016         case BOND_FOM_FOLLOW:
1017                 /*
1018                  * if new_active && old_active, swap them
1019                  * if just old_active, do nothing (going to no active slave)
1020                  * if just new_active, set new_active to bond's MAC
1021                  */
1022                 if (!new_active)
1023                         return;
1024
1025                 write_unlock_bh(&bond->curr_slave_lock);
1026                 read_unlock(&bond->lock);
1027
1028                 if (old_active) {
1029                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1030                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1031                                ETH_ALEN);
1032                         saddr.sa_family = new_active->dev->type;
1033                 } else {
1034                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1035                         saddr.sa_family = bond->dev->type;
1036                 }
1037
1038                 rv = dev_set_mac_address(new_active->dev, &saddr);
1039                 if (rv) {
1040                         pr_err(DRV_NAME
1041                                ": %s: Error %d setting MAC of slave %s\n",
1042                                bond->dev->name, -rv, new_active->dev->name);
1043                         goto out;
1044                 }
1045
1046                 if (!old_active)
1047                         goto out;
1048
1049                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1050                 saddr.sa_family = old_active->dev->type;
1051
1052                 rv = dev_set_mac_address(old_active->dev, &saddr);
1053                 if (rv)
1054                         pr_err(DRV_NAME
1055                                ": %s: Error %d setting MAC of slave %s\n",
1056                                bond->dev->name, -rv, new_active->dev->name);
1057 out:
1058                 read_lock(&bond->lock);
1059                 write_lock_bh(&bond->curr_slave_lock);
1060                 break;
1061         default:
1062                 pr_err(DRV_NAME
1063                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1064                        bond->dev->name, bond->params.fail_over_mac);
1065                 break;
1066         }
1067
1068 }
1069
1070
1071 /**
1072  * find_best_interface - select the best available slave to be the active one
1073  * @bond: our bonding struct
1074  *
1075  * Warning: Caller must hold curr_slave_lock for writing.
1076  */
1077 static struct slave *bond_find_best_slave(struct bonding *bond)
1078 {
1079         struct slave *new_active, *old_active;
1080         struct slave *bestslave = NULL;
1081         int mintime = bond->params.updelay;
1082         int i;
1083
1084         new_active = old_active = bond->curr_active_slave;
1085
1086         if (!new_active) { /* there were no active slaves left */
1087                 if (bond->slave_cnt > 0)   /* found one slave */
1088                         new_active = bond->first_slave;
1089                 else
1090                         return NULL; /* still no slave, return NULL */
1091         }
1092
1093         /*
1094          * first try the primary link; if arping, a link must tx/rx
1095          * traffic before it can be considered the curr_active_slave.
1096          * also, we would skip slaves between the curr_active_slave
1097          * and primary_slave that may be up and able to arp
1098          */
1099         if ((bond->primary_slave) &&
1100             (!bond->params.arp_interval) &&
1101             (IS_UP(bond->primary_slave->dev))) {
1102                 new_active = bond->primary_slave;
1103         }
1104
1105         /* remember where to stop iterating over the slaves */
1106         old_active = new_active;
1107
1108         bond_for_each_slave_from(bond, new_active, i, old_active) {
1109                 if (IS_UP(new_active->dev)) {
1110                         if (new_active->link == BOND_LINK_UP) {
1111                                 return new_active;
1112                         } else if (new_active->link == BOND_LINK_BACK) {
1113                                 /* link up, but waiting for stabilization */
1114                                 if (new_active->delay < mintime) {
1115                                         mintime = new_active->delay;
1116                                         bestslave = new_active;
1117                                 }
1118                         }
1119                 }
1120         }
1121
1122         return bestslave;
1123 }
1124
1125 /**
1126  * change_active_interface - change the active slave into the specified one
1127  * @bond: our bonding struct
1128  * @new: the new slave to make the active one
1129  *
1130  * Set the new slave to the bond's settings and unset them on the old
1131  * curr_active_slave.
1132  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1133  *
1134  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1135  * because it is apparently the best available slave we have, even though its
1136  * updelay hasn't timed out yet.
1137  *
1138  * If new_active is not NULL, caller must hold bond->lock for read and
1139  * curr_slave_lock for write_bh.
1140  */
1141 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1142 {
1143         struct slave *old_active = bond->curr_active_slave;
1144
1145         if (old_active == new_active)
1146                 return;
1147
1148         if (new_active) {
1149                 new_active->jiffies = jiffies;
1150
1151                 if (new_active->link == BOND_LINK_BACK) {
1152                         if (USES_PRIMARY(bond->params.mode)) {
1153                                 pr_info(DRV_NAME
1154                                        ": %s: making interface %s the new "
1155                                        "active one %d ms earlier.\n",
1156                                        bond->dev->name, new_active->dev->name,
1157                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158                         }
1159
1160                         new_active->delay = 0;
1161                         new_active->link = BOND_LINK_UP;
1162
1163                         if (bond->params.mode == BOND_MODE_8023AD)
1164                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165
1166                         if (bond_is_lb(bond))
1167                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1168                 } else {
1169                         if (USES_PRIMARY(bond->params.mode)) {
1170                                 pr_info(DRV_NAME
1171                                        ": %s: making interface %s the new "
1172                                        "active one.\n",
1173                                        bond->dev->name, new_active->dev->name);
1174                         }
1175                 }
1176         }
1177
1178         if (USES_PRIMARY(bond->params.mode))
1179                 bond_mc_swap(bond, new_active, old_active);
1180
1181         if (bond_is_lb(bond)) {
1182                 bond_alb_handle_active_change(bond, new_active);
1183                 if (old_active)
1184                         bond_set_slave_inactive_flags(old_active);
1185                 if (new_active)
1186                         bond_set_slave_active_flags(new_active);
1187         } else {
1188                 bond->curr_active_slave = new_active;
1189         }
1190
1191         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1192                 if (old_active)
1193                         bond_set_slave_inactive_flags(old_active);
1194
1195                 if (new_active) {
1196                         bond_set_slave_active_flags(new_active);
1197
1198                         if (bond->params.fail_over_mac)
1199                                 bond_do_fail_over_mac(bond, new_active,
1200                                                       old_active);
1201
1202                         bond->send_grat_arp = bond->params.num_grat_arp;
1203                         bond_send_gratuitous_arp(bond);
1204
1205                         bond->send_unsol_na = bond->params.num_unsol_na;
1206                         bond_send_unsolicited_na(bond);
1207
1208                         write_unlock_bh(&bond->curr_slave_lock);
1209                         read_unlock(&bond->lock);
1210
1211                         netdev_bonding_change(bond->dev);
1212
1213                         read_lock(&bond->lock);
1214                         write_lock_bh(&bond->curr_slave_lock);
1215                 }
1216         }
1217 }
1218
1219 /**
1220  * bond_select_active_slave - select a new active slave, if needed
1221  * @bond: our bonding struct
1222  *
1223  * This functions should be called when one of the following occurs:
1224  * - The old curr_active_slave has been released or lost its link.
1225  * - The primary_slave has got its link back.
1226  * - A slave has got its link back and there's no old curr_active_slave.
1227  *
1228  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1229  */
1230 void bond_select_active_slave(struct bonding *bond)
1231 {
1232         struct slave *best_slave;
1233         int rv;
1234
1235         best_slave = bond_find_best_slave(bond);
1236         if (best_slave != bond->curr_active_slave) {
1237                 bond_change_active_slave(bond, best_slave);
1238                 rv = bond_set_carrier(bond);
1239                 if (!rv)
1240                         return;
1241
1242                 if (netif_carrier_ok(bond->dev)) {
1243                         pr_info(DRV_NAME
1244                                ": %s: first active interface up!\n",
1245                                bond->dev->name);
1246                 } else {
1247                         pr_info(DRV_NAME ": %s: "
1248                                "now running without any active interface !\n",
1249                                bond->dev->name);
1250                 }
1251         }
1252 }
1253
1254 /*--------------------------- slave list handling ---------------------------*/
1255
1256 /*
1257  * This function attaches the slave to the end of list.
1258  *
1259  * bond->lock held for writing by caller.
1260  */
1261 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1262 {
1263         if (bond->first_slave == NULL) { /* attaching the first slave */
1264                 new_slave->next = new_slave;
1265                 new_slave->prev = new_slave;
1266                 bond->first_slave = new_slave;
1267         } else {
1268                 new_slave->next = bond->first_slave;
1269                 new_slave->prev = bond->first_slave->prev;
1270                 new_slave->next->prev = new_slave;
1271                 new_slave->prev->next = new_slave;
1272         }
1273
1274         bond->slave_cnt++;
1275 }
1276
1277 /*
1278  * This function detaches the slave from the list.
1279  * WARNING: no check is made to verify if the slave effectively
1280  * belongs to <bond>.
1281  * Nothing is freed on return, structures are just unchained.
1282  * If any slave pointer in bond was pointing to <slave>,
1283  * it should be changed by the calling function.
1284  *
1285  * bond->lock held for writing by caller.
1286  */
1287 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1288 {
1289         if (slave->next)
1290                 slave->next->prev = slave->prev;
1291
1292         if (slave->prev)
1293                 slave->prev->next = slave->next;
1294
1295         if (bond->first_slave == slave) { /* slave is the first slave */
1296                 if (bond->slave_cnt > 1) { /* there are more slave */
1297                         bond->first_slave = slave->next;
1298                 } else {
1299                         bond->first_slave = NULL; /* slave was the last one */
1300                 }
1301         }
1302
1303         slave->next = NULL;
1304         slave->prev = NULL;
1305         bond->slave_cnt--;
1306 }
1307
1308 /*---------------------------------- IOCTL ----------------------------------*/
1309
1310 static int bond_sethwaddr(struct net_device *bond_dev,
1311                           struct net_device *slave_dev)
1312 {
1313         pr_debug("bond_dev=%p\n", bond_dev);
1314         pr_debug("slave_dev=%p\n", slave_dev);
1315         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1316         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1317         return 0;
1318 }
1319
1320 #define BOND_VLAN_FEATURES \
1321         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1322          NETIF_F_HW_VLAN_FILTER)
1323
1324 /*
1325  * Compute the common dev->feature set available to all slaves.  Some
1326  * feature bits are managed elsewhere, so preserve those feature bits
1327  * on the master device.
1328  */
1329 static int bond_compute_features(struct bonding *bond)
1330 {
1331         struct slave *slave;
1332         struct net_device *bond_dev = bond->dev;
1333         unsigned long features = bond_dev->features;
1334         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1335                                                 bond_dev->hard_header_len);
1336         int i;
1337
1338         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1339         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1340
1341         if (!bond->first_slave)
1342                 goto done;
1343
1344         features &= ~NETIF_F_ONE_FOR_ALL;
1345
1346         bond_for_each_slave(bond, slave, i) {
1347                 features = netdev_increment_features(features,
1348                                                      slave->dev->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->hard_header_len = max_hard_header_len;
1358
1359         return 0;
1360 }
1361
1362 static void bond_setup_by_slave(struct net_device *bond_dev,
1363                                 struct net_device *slave_dev)
1364 {
1365         struct bonding *bond = netdev_priv(bond_dev);
1366
1367         bond_dev->header_ops        = slave_dev->header_ops;
1368
1369         bond_dev->type              = slave_dev->type;
1370         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1371         bond_dev->addr_len          = slave_dev->addr_len;
1372
1373         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1374                 slave_dev->addr_len);
1375         bond->setup_by_slave = 1;
1376 }
1377
1378 /* enslave device <slave> to bond device <master> */
1379 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1380 {
1381         struct bonding *bond = netdev_priv(bond_dev);
1382         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1383         struct slave *new_slave = NULL;
1384         struct dev_mc_list *dmi;
1385         struct sockaddr addr;
1386         int link_reporting;
1387         int old_features = bond_dev->features;
1388         int res = 0;
1389
1390         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1391                 slave_ops->ndo_do_ioctl == NULL) {
1392                 pr_warning(DRV_NAME
1393                        ": %s: Warning: no link monitoring support for %s\n",
1394                        bond_dev->name, slave_dev->name);
1395         }
1396
1397         /* bond must be initialized by bond_open() before enslaving */
1398         if (!(bond_dev->flags & IFF_UP)) {
1399                 pr_warning(DRV_NAME
1400                         " %s: master_dev is not up in bond_enslave\n",
1401                         bond_dev->name);
1402         }
1403
1404         /* already enslaved */
1405         if (slave_dev->flags & IFF_SLAVE) {
1406                 pr_debug("Error, Device was already enslaved\n");
1407                 return -EBUSY;
1408         }
1409
1410         /* vlan challenged mutual exclusion */
1411         /* no need to lock since we're protected by rtnl_lock */
1412         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1413                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1414                 if (!list_empty(&bond->vlan_list)) {
1415                         pr_err(DRV_NAME
1416                                ": %s: Error: cannot enslave VLAN "
1417                                "challenged slave %s on VLAN enabled "
1418                                "bond %s\n", bond_dev->name, slave_dev->name,
1419                                bond_dev->name);
1420                         return -EPERM;
1421                 } else {
1422                         pr_warning(DRV_NAME
1423                                ": %s: Warning: enslaved VLAN challenged "
1424                                "slave %s. Adding VLANs will be blocked as "
1425                                "long as %s is part of bond %s\n",
1426                                bond_dev->name, slave_dev->name, slave_dev->name,
1427                                bond_dev->name);
1428                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1429                 }
1430         } else {
1431                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1432                 if (bond->slave_cnt == 0) {
1433                         /* First slave, and it is not VLAN challenged,
1434                          * so remove the block of adding VLANs over the bond.
1435                          */
1436                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1437                 }
1438         }
1439
1440         /*
1441          * Old ifenslave binaries are no longer supported.  These can
1442          * be identified with moderate accuracy by the state of the slave:
1443          * the current ifenslave will set the interface down prior to
1444          * enslaving it; the old ifenslave will not.
1445          */
1446         if ((slave_dev->flags & IFF_UP)) {
1447                 pr_err(DRV_NAME ": %s is up. "
1448                        "This may be due to an out of date ifenslave.\n",
1449                        slave_dev->name);
1450                 res = -EPERM;
1451                 goto err_undo_flags;
1452         }
1453
1454         /* set bonding device ether type by slave - bonding netdevices are
1455          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1456          * there is a need to override some of the type dependent attribs/funcs.
1457          *
1458          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1459          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1460          */
1461         if (bond->slave_cnt == 0) {
1462                 if (bond_dev->type != slave_dev->type) {
1463                         dev_close(bond_dev);
1464                         pr_debug("%s: change device type from %d to %d\n",
1465                                 bond_dev->name, bond_dev->type, slave_dev->type);
1466                         if (slave_dev->type != ARPHRD_ETHER)
1467                                 bond_setup_by_slave(bond_dev, slave_dev);
1468                         else
1469                                 ether_setup(bond_dev);
1470                         dev_open(bond_dev);
1471                 }
1472         } else if (bond_dev->type != slave_dev->type) {
1473                 pr_err(DRV_NAME ": %s ether type (%d) is different "
1474                         "from other slaves (%d), can not enslave it.\n",
1475                         slave_dev->name,
1476                         slave_dev->type, bond_dev->type);
1477                         res = -EINVAL;
1478                         goto err_undo_flags;
1479         }
1480
1481         if (slave_ops->ndo_set_mac_address == NULL) {
1482                 if (bond->slave_cnt == 0) {
1483                         pr_warning(DRV_NAME
1484                                ": %s: Warning: The first slave device "
1485                                "specified does not support setting the MAC "
1486                                "address. Setting fail_over_mac to active.",
1487                                bond_dev->name);
1488                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1489                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1490                         pr_err(DRV_NAME
1491                                 ": %s: Error: The slave device specified "
1492                                 "does not support setting the MAC address, "
1493                                 "but fail_over_mac is not set to active.\n"
1494                                 , bond_dev->name);
1495                         res = -EOPNOTSUPP;
1496                         goto err_undo_flags;
1497                 }
1498         }
1499
1500         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1501         if (!new_slave) {
1502                 res = -ENOMEM;
1503                 goto err_undo_flags;
1504         }
1505
1506         /* save slave's original flags before calling
1507          * netdev_set_master and dev_open
1508          */
1509         new_slave->original_flags = slave_dev->flags;
1510
1511         /*
1512          * Save slave's original ("permanent") mac address for modes
1513          * that need it, and for restoring it upon release, and then
1514          * set it to the master's address
1515          */
1516         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1517
1518         if (!bond->params.fail_over_mac) {
1519                 /*
1520                  * Set slave to master's mac address.  The application already
1521                  * set the master's mac address to that of the first slave
1522                  */
1523                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1524                 addr.sa_family = slave_dev->type;
1525                 res = dev_set_mac_address(slave_dev, &addr);
1526                 if (res) {
1527                         pr_debug("Error %d calling set_mac_address\n", res);
1528                         goto err_free;
1529                 }
1530         }
1531
1532         res = netdev_set_master(slave_dev, bond_dev);
1533         if (res) {
1534                 pr_debug("Error %d calling netdev_set_master\n", res);
1535                 goto err_restore_mac;
1536         }
1537         /* open the slave since the application closed it */
1538         res = dev_open(slave_dev);
1539         if (res) {
1540                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1541                 goto err_unset_master;
1542         }
1543
1544         new_slave->dev = slave_dev;
1545         slave_dev->priv_flags |= IFF_BONDING;
1546
1547         if (bond_is_lb(bond)) {
1548                 /* bond_alb_init_slave() must be called before all other stages since
1549                  * it might fail and we do not want to have to undo everything
1550                  */
1551                 res = bond_alb_init_slave(bond, new_slave);
1552                 if (res)
1553                         goto err_close;
1554         }
1555
1556         /* If the mode USES_PRIMARY, then the new slave gets the
1557          * master's promisc (and mc) settings only if it becomes the
1558          * curr_active_slave, and that is taken care of later when calling
1559          * bond_change_active()
1560          */
1561         if (!USES_PRIMARY(bond->params.mode)) {
1562                 /* set promiscuity level to new slave */
1563                 if (bond_dev->flags & IFF_PROMISC) {
1564                         res = dev_set_promiscuity(slave_dev, 1);
1565                         if (res)
1566                                 goto err_close;
1567                 }
1568
1569                 /* set allmulti level to new slave */
1570                 if (bond_dev->flags & IFF_ALLMULTI) {
1571                         res = dev_set_allmulti(slave_dev, 1);
1572                         if (res)
1573                                 goto err_close;
1574                 }
1575
1576                 netif_addr_lock_bh(bond_dev);
1577                 /* upload master's mc_list to new slave */
1578                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1579                         dev_mc_add(slave_dev, dmi->dmi_addr,
1580                                    dmi->dmi_addrlen, 0);
1581                 netif_addr_unlock_bh(bond_dev);
1582         }
1583
1584         if (bond->params.mode == BOND_MODE_8023AD) {
1585                 /* add lacpdu mc addr to mc list */
1586                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1587
1588                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1589         }
1590
1591         bond_add_vlans_on_slave(bond, slave_dev);
1592
1593         write_lock_bh(&bond->lock);
1594
1595         bond_attach_slave(bond, new_slave);
1596
1597         new_slave->delay = 0;
1598         new_slave->link_failure_count = 0;
1599
1600         bond_compute_features(bond);
1601
1602         write_unlock_bh(&bond->lock);
1603
1604         read_lock(&bond->lock);
1605
1606         new_slave->last_arp_rx = jiffies;
1607
1608         if (bond->params.miimon && !bond->params.use_carrier) {
1609                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1610
1611                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1612                         /*
1613                          * miimon is set but a bonded network driver
1614                          * does not support ETHTOOL/MII and
1615                          * arp_interval is not set.  Note: if
1616                          * use_carrier is enabled, we will never go
1617                          * here (because netif_carrier is always
1618                          * supported); thus, we don't need to change
1619                          * the messages for netif_carrier.
1620                          */
1621                         pr_warning(DRV_NAME
1622                                ": %s: Warning: MII and ETHTOOL support not "
1623                                "available for interface %s, and "
1624                                "arp_interval/arp_ip_target module parameters "
1625                                "not specified, thus bonding will not detect "
1626                                "link failures! see bonding.txt for details.\n",
1627                                bond_dev->name, slave_dev->name);
1628                 } else if (link_reporting == -1) {
1629                         /* unable get link status using mii/ethtool */
1630                         pr_warning(DRV_NAME
1631                                ": %s: Warning: can't get link status from "
1632                                "interface %s; the network driver associated "
1633                                "with this interface does not support MII or "
1634                                "ETHTOOL link status reporting, thus miimon "
1635                                "has no effect on this interface.\n",
1636                                bond_dev->name, slave_dev->name);
1637                 }
1638         }
1639
1640         /* check for initial state */
1641         if (!bond->params.miimon ||
1642             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1643                 if (bond->params.updelay) {
1644                         pr_debug("Initial state of slave_dev is "
1645                                 "BOND_LINK_BACK\n");
1646                         new_slave->link  = BOND_LINK_BACK;
1647                         new_slave->delay = bond->params.updelay;
1648                 } else {
1649                         pr_debug("Initial state of slave_dev is "
1650                                 "BOND_LINK_UP\n");
1651                         new_slave->link  = BOND_LINK_UP;
1652                 }
1653                 new_slave->jiffies = jiffies;
1654         } else {
1655                 pr_debug("Initial state of slave_dev is "
1656                         "BOND_LINK_DOWN\n");
1657                 new_slave->link  = BOND_LINK_DOWN;
1658         }
1659
1660         if (bond_update_speed_duplex(new_slave) &&
1661             (new_slave->link != BOND_LINK_DOWN)) {
1662                 pr_warning(DRV_NAME
1663                        ": %s: Warning: failed to get speed and duplex from %s, "
1664                        "assumed to be 100Mb/sec and Full.\n",
1665                        bond_dev->name, new_slave->dev->name);
1666
1667                 if (bond->params.mode == BOND_MODE_8023AD) {
1668                         pr_warning(DRV_NAME
1669                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1670                                "support in base driver for proper aggregator "
1671                                "selection.\n", bond_dev->name);
1672                 }
1673         }
1674
1675         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1676                 /* if there is a primary slave, remember it */
1677                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1678                         bond->primary_slave = new_slave;
1679         }
1680
1681         write_lock_bh(&bond->curr_slave_lock);
1682
1683         switch (bond->params.mode) {
1684         case BOND_MODE_ACTIVEBACKUP:
1685                 bond_set_slave_inactive_flags(new_slave);
1686                 bond_select_active_slave(bond);
1687                 break;
1688         case BOND_MODE_8023AD:
1689                 /* in 802.3ad mode, the internal mechanism
1690                  * will activate the slaves in the selected
1691                  * aggregator
1692                  */
1693                 bond_set_slave_inactive_flags(new_slave);
1694                 /* if this is the first slave */
1695                 if (bond->slave_cnt == 1) {
1696                         SLAVE_AD_INFO(new_slave).id = 1;
1697                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1698                          * can be called only after the mac address of the bond is set
1699                          */
1700                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1701                                             bond->params.lacp_fast);
1702                 } else {
1703                         SLAVE_AD_INFO(new_slave).id =
1704                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1705                 }
1706
1707                 bond_3ad_bind_slave(new_slave);
1708                 break;
1709         case BOND_MODE_TLB:
1710         case BOND_MODE_ALB:
1711                 new_slave->state = BOND_STATE_ACTIVE;
1712                 bond_set_slave_inactive_flags(new_slave);
1713                 bond_select_active_slave(bond);
1714                 break;
1715         default:
1716                 pr_debug("This slave is always active in trunk mode\n");
1717
1718                 /* always active in trunk mode */
1719                 new_slave->state = BOND_STATE_ACTIVE;
1720
1721                 /* In trunking mode there is little meaning to curr_active_slave
1722                  * anyway (it holds no special properties of the bond device),
1723                  * so we can change it without calling change_active_interface()
1724                  */
1725                 if (!bond->curr_active_slave)
1726                         bond->curr_active_slave = new_slave;
1727
1728                 break;
1729         } /* switch(bond_mode) */
1730
1731         write_unlock_bh(&bond->curr_slave_lock);
1732
1733         bond_set_carrier(bond);
1734
1735         read_unlock(&bond->lock);
1736
1737         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738         if (res)
1739                 goto err_close;
1740
1741         pr_info(DRV_NAME
1742                ": %s: enslaving %s as a%s interface with a%s link.\n",
1743                bond_dev->name, slave_dev->name,
1744                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1745                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1746
1747         /* enslave is successful */
1748         return 0;
1749
1750 /* Undo stages on error */
1751 err_close:
1752         dev_close(slave_dev);
1753
1754 err_unset_master:
1755         netdev_set_master(slave_dev, NULL);
1756
1757 err_restore_mac:
1758         if (!bond->params.fail_over_mac) {
1759                 /* XXX TODO - fom follow mode needs to change master's
1760                  * MAC if this slave's MAC is in use by the bond, or at
1761                  * least print a warning.
1762                  */
1763                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1764                 addr.sa_family = slave_dev->type;
1765                 dev_set_mac_address(slave_dev, &addr);
1766         }
1767
1768 err_free:
1769         kfree(new_slave);
1770
1771 err_undo_flags:
1772         bond_dev->features = old_features;
1773
1774         return res;
1775 }
1776
1777 /*
1778  * Try to release the slave device <slave> from the bond device <master>
1779  * It is legal to access curr_active_slave without a lock because all the function
1780  * is write-locked.
1781  *
1782  * The rules for slave state should be:
1783  *   for Active/Backup:
1784  *     Active stays on all backups go down
1785  *   for Bonded connections:
1786  *     The first up interface should be left on and all others downed.
1787  */
1788 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1789 {
1790         struct bonding *bond = netdev_priv(bond_dev);
1791         struct slave *slave, *oldcurrent;
1792         struct sockaddr addr;
1793         int mac_addr_differ;
1794
1795         /* slave is not a slave or master is not master of this slave */
1796         if (!(slave_dev->flags & IFF_SLAVE) ||
1797             (slave_dev->master != bond_dev)) {
1798                 pr_err(DRV_NAME
1799                        ": %s: Error: cannot release %s.\n",
1800                        bond_dev->name, slave_dev->name);
1801                 return -EINVAL;
1802         }
1803
1804         write_lock_bh(&bond->lock);
1805
1806         slave = bond_get_slave_by_dev(bond, slave_dev);
1807         if (!slave) {
1808                 /* not a slave of this bond */
1809                 pr_info(DRV_NAME
1810                        ": %s: %s not enslaved\n",
1811                        bond_dev->name, slave_dev->name);
1812                 write_unlock_bh(&bond->lock);
1813                 return -EINVAL;
1814         }
1815
1816         if (!bond->params.fail_over_mac) {
1817                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1818                                          ETH_ALEN);
1819                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1820                         pr_warning(DRV_NAME
1821                                ": %s: Warning: the permanent HWaddr of %s - "
1822                                "%pM - is still in use by %s. "
1823                                "Set the HWaddr of %s to a different address "
1824                                "to avoid conflicts.\n",
1825                                bond_dev->name, slave_dev->name,
1826                                slave->perm_hwaddr,
1827                                bond_dev->name, slave_dev->name);
1828         }
1829
1830         /* Inform AD package of unbinding of slave. */
1831         if (bond->params.mode == BOND_MODE_8023AD) {
1832                 /* must be called before the slave is
1833                  * detached from the list
1834                  */
1835                 bond_3ad_unbind_slave(slave);
1836         }
1837
1838         pr_info(DRV_NAME
1839                ": %s: releasing %s interface %s\n",
1840                bond_dev->name,
1841                (slave->state == BOND_STATE_ACTIVE)
1842                ? "active" : "backup",
1843                slave_dev->name);
1844
1845         oldcurrent = bond->curr_active_slave;
1846
1847         bond->current_arp_slave = NULL;
1848
1849         /* release the slave from its bond */
1850         bond_detach_slave(bond, slave);
1851
1852         bond_compute_features(bond);
1853
1854         if (bond->primary_slave == slave)
1855                 bond->primary_slave = NULL;
1856
1857         if (oldcurrent == slave)
1858                 bond_change_active_slave(bond, NULL);
1859
1860         if (bond_is_lb(bond)) {
1861                 /* Must be called only after the slave has been
1862                  * detached from the list and the curr_active_slave
1863                  * has been cleared (if our_slave == old_current),
1864                  * but before a new active slave is selected.
1865                  */
1866                 write_unlock_bh(&bond->lock);
1867                 bond_alb_deinit_slave(bond, slave);
1868                 write_lock_bh(&bond->lock);
1869         }
1870
1871         if (oldcurrent == slave) {
1872                 /*
1873                  * Note that we hold RTNL over this sequence, so there
1874                  * is no concern that another slave add/remove event
1875                  * will interfere.
1876                  */
1877                 write_unlock_bh(&bond->lock);
1878                 read_lock(&bond->lock);
1879                 write_lock_bh(&bond->curr_slave_lock);
1880
1881                 bond_select_active_slave(bond);
1882
1883                 write_unlock_bh(&bond->curr_slave_lock);
1884                 read_unlock(&bond->lock);
1885                 write_lock_bh(&bond->lock);
1886         }
1887
1888         if (bond->slave_cnt == 0) {
1889                 bond_set_carrier(bond);
1890
1891                 /* if the last slave was removed, zero the mac address
1892                  * of the master so it will be set by the application
1893                  * to the mac address of the first slave
1894                  */
1895                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1896
1897                 if (list_empty(&bond->vlan_list)) {
1898                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1899                 } else {
1900                         pr_warning(DRV_NAME
1901                                ": %s: Warning: clearing HW address of %s while it "
1902                                "still has VLANs.\n",
1903                                bond_dev->name, bond_dev->name);
1904                         pr_warning(DRV_NAME
1905                                ": %s: When re-adding slaves, make sure the bond's "
1906                                "HW address matches its VLANs'.\n",
1907                                bond_dev->name);
1908                 }
1909         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1910                    !bond_has_challenged_slaves(bond)) {
1911                 pr_info(DRV_NAME
1912                        ": %s: last VLAN challenged slave %s "
1913                        "left bond %s. VLAN blocking is removed\n",
1914                        bond_dev->name, slave_dev->name, bond_dev->name);
1915                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1916         }
1917
1918         write_unlock_bh(&bond->lock);
1919
1920         /* must do this from outside any spinlocks */
1921         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1922
1923         bond_del_vlans_from_slave(bond, slave_dev);
1924
1925         /* If the mode USES_PRIMARY, then we should only remove its
1926          * promisc and mc settings if it was the curr_active_slave, but that was
1927          * already taken care of above when we detached the slave
1928          */
1929         if (!USES_PRIMARY(bond->params.mode)) {
1930                 /* unset promiscuity level from slave */
1931                 if (bond_dev->flags & IFF_PROMISC)
1932                         dev_set_promiscuity(slave_dev, -1);
1933
1934                 /* unset allmulti level from slave */
1935                 if (bond_dev->flags & IFF_ALLMULTI)
1936                         dev_set_allmulti(slave_dev, -1);
1937
1938                 /* flush master's mc_list from slave */
1939                 netif_addr_lock_bh(bond_dev);
1940                 bond_mc_list_flush(bond_dev, slave_dev);
1941                 netif_addr_unlock_bh(bond_dev);
1942         }
1943
1944         netdev_set_master(slave_dev, NULL);
1945
1946         /* close slave before restoring its mac address */
1947         dev_close(slave_dev);
1948
1949         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1950                 /* restore original ("permanent") mac address */
1951                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1952                 addr.sa_family = slave_dev->type;
1953                 dev_set_mac_address(slave_dev, &addr);
1954         }
1955
1956         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1957                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1958                                    IFF_SLAVE_NEEDARP);
1959
1960         kfree(slave);
1961
1962         return 0;  /* deletion OK */
1963 }
1964
1965 /*
1966 * Destroy a bonding device.
1967 * Must be under rtnl_lock when this function is called.
1968 */
1969 static void bond_uninit(struct net_device *bond_dev)
1970 {
1971         struct bonding *bond = netdev_priv(bond_dev);
1972
1973         bond_deinit(bond_dev);
1974         bond_destroy_sysfs_entry(bond);
1975
1976         if (bond->wq)
1977                 destroy_workqueue(bond->wq);
1978
1979         netif_addr_lock_bh(bond_dev);
1980         bond_mc_list_destroy(bond);
1981         netif_addr_unlock_bh(bond_dev);
1982 }
1983
1984 /*
1985 * First release a slave and than destroy the bond if no more slaves are left.
1986 * Must be under rtnl_lock when this function is called.
1987 */
1988 int  bond_release_and_destroy(struct net_device *bond_dev,
1989                               struct net_device *slave_dev)
1990 {
1991         struct bonding *bond = netdev_priv(bond_dev);
1992         int ret;
1993
1994         ret = bond_release(bond_dev, slave_dev);
1995         if ((ret == 0) && (bond->slave_cnt == 0)) {
1996                 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
1997                        bond_dev->name, bond_dev->name);
1998                 unregister_netdevice(bond_dev);
1999         }
2000         return ret;
2001 }
2002
2003 /*
2004  * This function releases all slaves.
2005  */
2006 static int bond_release_all(struct net_device *bond_dev)
2007 {
2008         struct bonding *bond = netdev_priv(bond_dev);
2009         struct slave *slave;
2010         struct net_device *slave_dev;
2011         struct sockaddr addr;
2012
2013         write_lock_bh(&bond->lock);
2014
2015         netif_carrier_off(bond_dev);
2016
2017         if (bond->slave_cnt == 0)
2018                 goto out;
2019
2020         bond->current_arp_slave = NULL;
2021         bond->primary_slave = NULL;
2022         bond_change_active_slave(bond, NULL);
2023
2024         while ((slave = bond->first_slave) != NULL) {
2025                 /* Inform AD package of unbinding of slave
2026                  * before slave is detached from the list.
2027                  */
2028                 if (bond->params.mode == BOND_MODE_8023AD)
2029                         bond_3ad_unbind_slave(slave);
2030
2031                 slave_dev = slave->dev;
2032                 bond_detach_slave(bond, slave);
2033
2034                 /* now that the slave is detached, unlock and perform
2035                  * all the undo steps that should not be called from
2036                  * within a lock.
2037                  */
2038                 write_unlock_bh(&bond->lock);
2039
2040                 if (bond_is_lb(bond)) {
2041                         /* must be called only after the slave
2042                          * has been detached from the list
2043                          */
2044                         bond_alb_deinit_slave(bond, slave);
2045                 }
2046
2047                 bond_compute_features(bond);
2048
2049                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2050                 bond_del_vlans_from_slave(bond, slave_dev);
2051
2052                 /* If the mode USES_PRIMARY, then we should only remove its
2053                  * promisc and mc settings if it was the curr_active_slave, but that was
2054                  * already taken care of above when we detached the slave
2055                  */
2056                 if (!USES_PRIMARY(bond->params.mode)) {
2057                         /* unset promiscuity level from slave */
2058                         if (bond_dev->flags & IFF_PROMISC)
2059                                 dev_set_promiscuity(slave_dev, -1);
2060
2061                         /* unset allmulti level from slave */
2062                         if (bond_dev->flags & IFF_ALLMULTI)
2063                                 dev_set_allmulti(slave_dev, -1);
2064
2065                         /* flush master's mc_list from slave */
2066                         netif_addr_lock_bh(bond_dev);
2067                         bond_mc_list_flush(bond_dev, slave_dev);
2068                         netif_addr_unlock_bh(bond_dev);
2069                 }
2070
2071                 netdev_set_master(slave_dev, NULL);
2072
2073                 /* close slave before restoring its mac address */
2074                 dev_close(slave_dev);
2075
2076                 if (!bond->params.fail_over_mac) {
2077                         /* restore original ("permanent") mac address*/
2078                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2079                         addr.sa_family = slave_dev->type;
2080                         dev_set_mac_address(slave_dev, &addr);
2081                 }
2082
2083                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2084                                            IFF_SLAVE_INACTIVE);
2085
2086                 kfree(slave);
2087
2088                 /* re-acquire the lock before getting the next slave */
2089                 write_lock_bh(&bond->lock);
2090         }
2091
2092         /* zero the mac address of the master so it will be
2093          * set by the application to the mac address of the
2094          * first slave
2095          */
2096         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2097
2098         if (list_empty(&bond->vlan_list))
2099                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2100         else {
2101                 pr_warning(DRV_NAME
2102                        ": %s: Warning: clearing HW address of %s while it "
2103                        "still has VLANs.\n",
2104                        bond_dev->name, bond_dev->name);
2105                 pr_warning(DRV_NAME
2106                        ": %s: When re-adding slaves, make sure the bond's "
2107                        "HW address matches its VLANs'.\n",
2108                        bond_dev->name);
2109         }
2110
2111         pr_info(DRV_NAME
2112                ": %s: released all slaves\n",
2113                bond_dev->name);
2114
2115 out:
2116         write_unlock_bh(&bond->lock);
2117
2118         return 0;
2119 }
2120
2121 /*
2122  * This function changes the active slave to slave <slave_dev>.
2123  * It returns -EINVAL in the following cases.
2124  *  - <slave_dev> is not found in the list.
2125  *  - There is not active slave now.
2126  *  - <slave_dev> is already active.
2127  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2128  *  - <slave_dev> is not running.
2129  * In these cases, this function does nothing.
2130  * In the other cases, current_slave pointer is changed and 0 is returned.
2131  */
2132 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2133 {
2134         struct bonding *bond = netdev_priv(bond_dev);
2135         struct slave *old_active = NULL;
2136         struct slave *new_active = NULL;
2137         int res = 0;
2138
2139         if (!USES_PRIMARY(bond->params.mode))
2140                 return -EINVAL;
2141
2142         /* Verify that master_dev is indeed the master of slave_dev */
2143         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2144                 return -EINVAL;
2145
2146         read_lock(&bond->lock);
2147
2148         read_lock(&bond->curr_slave_lock);
2149         old_active = bond->curr_active_slave;
2150         read_unlock(&bond->curr_slave_lock);
2151
2152         new_active = bond_get_slave_by_dev(bond, slave_dev);
2153
2154         /*
2155          * Changing to the current active: do nothing; return success.
2156          */
2157         if (new_active && (new_active == old_active)) {
2158                 read_unlock(&bond->lock);
2159                 return 0;
2160         }
2161
2162         if ((new_active) &&
2163             (old_active) &&
2164             (new_active->link == BOND_LINK_UP) &&
2165             IS_UP(new_active->dev)) {
2166                 write_lock_bh(&bond->curr_slave_lock);
2167                 bond_change_active_slave(bond, new_active);
2168                 write_unlock_bh(&bond->curr_slave_lock);
2169         } else
2170                 res = -EINVAL;
2171
2172         read_unlock(&bond->lock);
2173
2174         return res;
2175 }
2176
2177 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2178 {
2179         struct bonding *bond = netdev_priv(bond_dev);
2180
2181         info->bond_mode = bond->params.mode;
2182         info->miimon = bond->params.miimon;
2183
2184         read_lock(&bond->lock);
2185         info->num_slaves = bond->slave_cnt;
2186         read_unlock(&bond->lock);
2187
2188         return 0;
2189 }
2190
2191 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2192 {
2193         struct bonding *bond = netdev_priv(bond_dev);
2194         struct slave *slave;
2195         int i, res = -ENODEV;
2196
2197         read_lock(&bond->lock);
2198
2199         bond_for_each_slave(bond, slave, i) {
2200                 if (i == (int)info->slave_id) {
2201                         res = 0;
2202                         strcpy(info->slave_name, slave->dev->name);
2203                         info->link = slave->link;
2204                         info->state = slave->state;
2205                         info->link_failure_count = slave->link_failure_count;
2206                         break;
2207                 }
2208         }
2209
2210         read_unlock(&bond->lock);
2211
2212         return res;
2213 }
2214
2215 /*-------------------------------- Monitoring -------------------------------*/
2216
2217
2218 static int bond_miimon_inspect(struct bonding *bond)
2219 {
2220         struct slave *slave;
2221         int i, link_state, commit = 0;
2222         bool ignore_updelay;
2223
2224         ignore_updelay = !bond->curr_active_slave ? true : false;
2225
2226         bond_for_each_slave(bond, slave, i) {
2227                 slave->new_link = BOND_LINK_NOCHANGE;
2228
2229                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2230
2231                 switch (slave->link) {
2232                 case BOND_LINK_UP:
2233                         if (link_state)
2234                                 continue;
2235
2236                         slave->link = BOND_LINK_FAIL;
2237                         slave->delay = bond->params.downdelay;
2238                         if (slave->delay) {
2239                                 pr_info(DRV_NAME
2240                                        ": %s: link status down for %s"
2241                                        "interface %s, disabling it in %d ms.\n",
2242                                        bond->dev->name,
2243                                        (bond->params.mode ==
2244                                         BOND_MODE_ACTIVEBACKUP) ?
2245                                        ((slave->state == BOND_STATE_ACTIVE) ?
2246                                         "active " : "backup ") : "",
2247                                        slave->dev->name,
2248                                        bond->params.downdelay * bond->params.miimon);
2249                         }
2250                         /*FALLTHRU*/
2251                 case BOND_LINK_FAIL:
2252                         if (link_state) {
2253                                 /*
2254                                  * recovered before downdelay expired
2255                                  */
2256                                 slave->link = BOND_LINK_UP;
2257                                 slave->jiffies = jiffies;
2258                                 pr_info(DRV_NAME
2259                                        ": %s: link status up again after %d "
2260                                        "ms for interface %s.\n",
2261                                        bond->dev->name,
2262                                        (bond->params.downdelay - slave->delay) *
2263                                        bond->params.miimon,
2264                                        slave->dev->name);
2265                                 continue;
2266                         }
2267
2268                         if (slave->delay <= 0) {
2269                                 slave->new_link = BOND_LINK_DOWN;
2270                                 commit++;
2271                                 continue;
2272                         }
2273
2274                         slave->delay--;
2275                         break;
2276
2277                 case BOND_LINK_DOWN:
2278                         if (!link_state)
2279                                 continue;
2280
2281                         slave->link = BOND_LINK_BACK;
2282                         slave->delay = bond->params.updelay;
2283
2284                         if (slave->delay) {
2285                                 pr_info(DRV_NAME
2286                                        ": %s: link status up for "
2287                                        "interface %s, enabling it in %d ms.\n",
2288                                        bond->dev->name, slave->dev->name,
2289                                        ignore_updelay ? 0 :
2290                                        bond->params.updelay *
2291                                        bond->params.miimon);
2292                         }
2293                         /*FALLTHRU*/
2294                 case BOND_LINK_BACK:
2295                         if (!link_state) {
2296                                 slave->link = BOND_LINK_DOWN;
2297                                 pr_info(DRV_NAME
2298                                        ": %s: link status down again after %d "
2299                                        "ms for interface %s.\n",
2300                                        bond->dev->name,
2301                                        (bond->params.updelay - slave->delay) *
2302                                        bond->params.miimon,
2303                                        slave->dev->name);
2304
2305                                 continue;
2306                         }
2307
2308                         if (ignore_updelay)
2309                                 slave->delay = 0;
2310
2311                         if (slave->delay <= 0) {
2312                                 slave->new_link = BOND_LINK_UP;
2313                                 commit++;
2314                                 ignore_updelay = false;
2315                                 continue;
2316                         }
2317
2318                         slave->delay--;
2319                         break;
2320                 }
2321         }
2322
2323         return commit;
2324 }
2325
2326 static void bond_miimon_commit(struct bonding *bond)
2327 {
2328         struct slave *slave;
2329         int i;
2330
2331         bond_for_each_slave(bond, slave, i) {
2332                 switch (slave->new_link) {
2333                 case BOND_LINK_NOCHANGE:
2334                         continue;
2335
2336                 case BOND_LINK_UP:
2337                         slave->link = BOND_LINK_UP;
2338                         slave->jiffies = jiffies;
2339
2340                         if (bond->params.mode == BOND_MODE_8023AD) {
2341                                 /* prevent it from being the active one */
2342                                 slave->state = BOND_STATE_BACKUP;
2343                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2344                                 /* make it immediately active */
2345                                 slave->state = BOND_STATE_ACTIVE;
2346                         } else if (slave != bond->primary_slave) {
2347                                 /* prevent it from being the active one */
2348                                 slave->state = BOND_STATE_BACKUP;
2349                         }
2350
2351                         pr_info(DRV_NAME
2352                                ": %s: link status definitely "
2353                                "up for interface %s.\n",
2354                                bond->dev->name, slave->dev->name);
2355
2356                         /* notify ad that the link status has changed */
2357                         if (bond->params.mode == BOND_MODE_8023AD)
2358                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2359
2360                         if (bond_is_lb(bond))
2361                                 bond_alb_handle_link_change(bond, slave,
2362                                                             BOND_LINK_UP);
2363
2364                         if (!bond->curr_active_slave ||
2365                             (slave == bond->primary_slave))
2366                                 goto do_failover;
2367
2368                         continue;
2369
2370                 case BOND_LINK_DOWN:
2371                         if (slave->link_failure_count < UINT_MAX)
2372                                 slave->link_failure_count++;
2373
2374                         slave->link = BOND_LINK_DOWN;
2375
2376                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2377                             bond->params.mode == BOND_MODE_8023AD)
2378                                 bond_set_slave_inactive_flags(slave);
2379
2380                         pr_info(DRV_NAME
2381                                ": %s: link status definitely down for "
2382                                "interface %s, disabling it\n",
2383                                bond->dev->name, slave->dev->name);
2384
2385                         if (bond->params.mode == BOND_MODE_8023AD)
2386                                 bond_3ad_handle_link_change(slave,
2387                                                             BOND_LINK_DOWN);
2388
2389                         if (bond_is_lb(bond))
2390                                 bond_alb_handle_link_change(bond, slave,
2391                                                             BOND_LINK_DOWN);
2392
2393                         if (slave == bond->curr_active_slave)
2394                                 goto do_failover;
2395
2396                         continue;
2397
2398                 default:
2399                         pr_err(DRV_NAME
2400                                ": %s: invalid new link %d on slave %s\n",
2401                                bond->dev->name, slave->new_link,
2402                                slave->dev->name);
2403                         slave->new_link = BOND_LINK_NOCHANGE;
2404
2405                         continue;
2406                 }
2407
2408 do_failover:
2409                 ASSERT_RTNL();
2410                 write_lock_bh(&bond->curr_slave_lock);
2411                 bond_select_active_slave(bond);
2412                 write_unlock_bh(&bond->curr_slave_lock);
2413         }
2414
2415         bond_set_carrier(bond);
2416 }
2417
2418 /*
2419  * bond_mii_monitor
2420  *
2421  * Really a wrapper that splits the mii monitor into two phases: an
2422  * inspection, then (if inspection indicates something needs to be done)
2423  * an acquisition of appropriate locks followed by a commit phase to
2424  * implement whatever link state changes are indicated.
2425  */
2426 void bond_mii_monitor(struct work_struct *work)
2427 {
2428         struct bonding *bond = container_of(work, struct bonding,
2429                                             mii_work.work);
2430
2431         read_lock(&bond->lock);
2432         if (bond->kill_timers)
2433                 goto out;
2434
2435         if (bond->slave_cnt == 0)
2436                 goto re_arm;
2437
2438         if (bond->send_grat_arp) {
2439                 read_lock(&bond->curr_slave_lock);
2440                 bond_send_gratuitous_arp(bond);
2441                 read_unlock(&bond->curr_slave_lock);
2442         }
2443
2444         if (bond->send_unsol_na) {
2445                 read_lock(&bond->curr_slave_lock);
2446                 bond_send_unsolicited_na(bond);
2447                 read_unlock(&bond->curr_slave_lock);
2448         }
2449
2450         if (bond_miimon_inspect(bond)) {
2451                 read_unlock(&bond->lock);
2452                 rtnl_lock();
2453                 read_lock(&bond->lock);
2454
2455                 bond_miimon_commit(bond);
2456
2457                 read_unlock(&bond->lock);
2458                 rtnl_unlock();  /* might sleep, hold no other locks */
2459                 read_lock(&bond->lock);
2460         }
2461
2462 re_arm:
2463         if (bond->params.miimon)
2464                 queue_delayed_work(bond->wq, &bond->mii_work,
2465                                    msecs_to_jiffies(bond->params.miimon));
2466 out:
2467         read_unlock(&bond->lock);
2468 }
2469
2470 static __be32 bond_glean_dev_ip(struct net_device *dev)
2471 {
2472         struct in_device *idev;
2473         struct in_ifaddr *ifa;
2474         __be32 addr = 0;
2475
2476         if (!dev)
2477                 return 0;
2478
2479         rcu_read_lock();
2480         idev = __in_dev_get_rcu(dev);
2481         if (!idev)
2482                 goto out;
2483
2484         ifa = idev->ifa_list;
2485         if (!ifa)
2486                 goto out;
2487
2488         addr = ifa->ifa_local;
2489 out:
2490         rcu_read_unlock();
2491         return addr;
2492 }
2493
2494 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2495 {
2496         struct vlan_entry *vlan;
2497
2498         if (ip == bond->master_ip)
2499                 return 1;
2500
2501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2502                 if (ip == vlan->vlan_ip)
2503                         return 1;
2504         }
2505
2506         return 0;
2507 }
2508
2509 /*
2510  * We go to the (large) trouble of VLAN tagging ARP frames because
2511  * switches in VLAN mode (especially if ports are configured as
2512  * "native" to a VLAN) might not pass non-tagged frames.
2513  */
2514 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2515 {
2516         struct sk_buff *skb;
2517
2518         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2519                slave_dev->name, dest_ip, src_ip, vlan_id);
2520
2521         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2522                          NULL, slave_dev->dev_addr, NULL);
2523
2524         if (!skb) {
2525                 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2526                 return;
2527         }
2528         if (vlan_id) {
2529                 skb = vlan_put_tag(skb, vlan_id);
2530                 if (!skb) {
2531                         pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2532                         return;
2533                 }
2534         }
2535         arp_xmit(skb);
2536 }
2537
2538
2539 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2540 {
2541         int i, vlan_id, rv;
2542         __be32 *targets = bond->params.arp_targets;
2543         struct vlan_entry *vlan;
2544         struct net_device *vlan_dev;
2545         struct flowi fl;
2546         struct rtable *rt;
2547
2548         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2549                 if (!targets[i])
2550                         break;
2551                 pr_debug("basa: target %x\n", targets[i]);
2552                 if (list_empty(&bond->vlan_list)) {
2553                         pr_debug("basa: empty vlan: arp_send\n");
2554                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2555                                       bond->master_ip, 0);
2556                         continue;
2557                 }
2558
2559                 /*
2560                  * If VLANs are configured, we do a route lookup to
2561                  * determine which VLAN interface would be used, so we
2562                  * can tag the ARP with the proper VLAN tag.
2563                  */
2564                 memset(&fl, 0, sizeof(fl));
2565                 fl.fl4_dst = targets[i];
2566                 fl.fl4_tos = RTO_ONLINK;
2567
2568                 rv = ip_route_output_key(&init_net, &rt, &fl);
2569                 if (rv) {
2570                         if (net_ratelimit()) {
2571                                 pr_warning(DRV_NAME
2572                              ": %s: no route to arp_ip_target %pI4\n",
2573                                        bond->dev->name, &fl.fl4_dst);
2574                         }
2575                         continue;
2576                 }
2577
2578                 /*
2579                  * This target is not on a VLAN
2580                  */
2581                 if (rt->u.dst.dev == bond->dev) {
2582                         ip_rt_put(rt);
2583                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2584                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2585                                       bond->master_ip, 0);
2586                         continue;
2587                 }
2588
2589                 vlan_id = 0;
2590                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2591                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2592                         if (vlan_dev == rt->u.dst.dev) {
2593                                 vlan_id = vlan->vlan_id;
2594                                 pr_debug("basa: vlan match on %s %d\n",
2595                                        vlan_dev->name, vlan_id);
2596                                 break;
2597                         }
2598                 }
2599
2600                 if (vlan_id) {
2601                         ip_rt_put(rt);
2602                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2603                                       vlan->vlan_ip, vlan_id);
2604                         continue;
2605                 }
2606
2607                 if (net_ratelimit()) {
2608                         pr_warning(DRV_NAME
2609                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2610                                bond->dev->name, &fl.fl4_dst,
2611                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2612                 }
2613                 ip_rt_put(rt);
2614         }
2615 }
2616
2617 /*
2618  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2619  * for each VLAN above us.
2620  *
2621  * Caller must hold curr_slave_lock for read or better
2622  */
2623 static void bond_send_gratuitous_arp(struct bonding *bond)
2624 {
2625         struct slave *slave = bond->curr_active_slave;
2626         struct vlan_entry *vlan;
2627         struct net_device *vlan_dev;
2628
2629         pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2630                                 slave ? slave->dev->name : "NULL");
2631
2632         if (!slave || !bond->send_grat_arp ||
2633             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2634                 return;
2635
2636         bond->send_grat_arp--;
2637
2638         if (bond->master_ip) {
2639                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2640                                 bond->master_ip, 0);
2641         }
2642
2643         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2644                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2645                 if (vlan->vlan_ip) {
2646                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2647                                       vlan->vlan_ip, vlan->vlan_id);
2648                 }
2649         }
2650 }
2651
2652 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2653 {
2654         int i;
2655         __be32 *targets = bond->params.arp_targets;
2656
2657         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2658                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2659                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2660                 if (sip == targets[i]) {
2661                         if (bond_has_this_ip(bond, tip))
2662                                 slave->last_arp_rx = jiffies;
2663                         return;
2664                 }
2665         }
2666 }
2667
2668 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2669 {
2670         struct arphdr *arp;
2671         struct slave *slave;
2672         struct bonding *bond;
2673         unsigned char *arp_ptr;
2674         __be32 sip, tip;
2675
2676         if (dev_net(dev) != &init_net)
2677                 goto out;
2678
2679         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2680                 goto out;
2681
2682         bond = netdev_priv(dev);
2683         read_lock(&bond->lock);
2684
2685         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2686                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2687                 orig_dev ? orig_dev->name : "NULL");
2688
2689         slave = bond_get_slave_by_dev(bond, orig_dev);
2690         if (!slave || !slave_do_arp_validate(bond, slave))
2691                 goto out_unlock;
2692
2693         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2694                 goto out_unlock;
2695
2696         arp = arp_hdr(skb);
2697         if (arp->ar_hln != dev->addr_len ||
2698             skb->pkt_type == PACKET_OTHERHOST ||
2699             skb->pkt_type == PACKET_LOOPBACK ||
2700             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2701             arp->ar_pro != htons(ETH_P_IP) ||
2702             arp->ar_pln != 4)
2703                 goto out_unlock;
2704
2705         arp_ptr = (unsigned char *)(arp + 1);
2706         arp_ptr += dev->addr_len;
2707         memcpy(&sip, arp_ptr, 4);
2708         arp_ptr += 4 + dev->addr_len;
2709         memcpy(&tip, arp_ptr, 4);
2710
2711         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2712                 bond->dev->name, slave->dev->name, slave->state,
2713                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2714                 &sip, &tip);
2715
2716         /*
2717          * Backup slaves won't see the ARP reply, but do come through
2718          * here for each ARP probe (so we swap the sip/tip to validate
2719          * the probe).  In a "redundant switch, common router" type of
2720          * configuration, the ARP probe will (hopefully) travel from
2721          * the active, through one switch, the router, then the other
2722          * switch before reaching the backup.
2723          */
2724         if (slave->state == BOND_STATE_ACTIVE)
2725                 bond_validate_arp(bond, slave, sip, tip);
2726         else
2727                 bond_validate_arp(bond, slave, tip, sip);
2728
2729 out_unlock:
2730         read_unlock(&bond->lock);
2731 out:
2732         dev_kfree_skb(skb);
2733         return NET_RX_SUCCESS;
2734 }
2735
2736 /*
2737  * this function is called regularly to monitor each slave's link
2738  * ensuring that traffic is being sent and received when arp monitoring
2739  * is used in load-balancing mode. if the adapter has been dormant, then an
2740  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2741  * arp monitoring in active backup mode.
2742  */
2743 void bond_loadbalance_arp_mon(struct work_struct *work)
2744 {
2745         struct bonding *bond = container_of(work, struct bonding,
2746                                             arp_work.work);
2747         struct slave *slave, *oldcurrent;
2748         int do_failover = 0;
2749         int delta_in_ticks;
2750         int i;
2751
2752         read_lock(&bond->lock);
2753
2754         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2755
2756         if (bond->kill_timers)
2757                 goto out;
2758
2759         if (bond->slave_cnt == 0)
2760                 goto re_arm;
2761
2762         read_lock(&bond->curr_slave_lock);
2763         oldcurrent = bond->curr_active_slave;
2764         read_unlock(&bond->curr_slave_lock);
2765
2766         /* see if any of the previous devices are up now (i.e. they have
2767          * xmt and rcv traffic). the curr_active_slave does not come into
2768          * the picture unless it is null. also, slave->jiffies is not needed
2769          * here because we send an arp on each slave and give a slave as
2770          * long as it needs to get the tx/rx within the delta.
2771          * TODO: what about up/down delay in arp mode? it wasn't here before
2772          *       so it can wait
2773          */
2774         bond_for_each_slave(bond, slave, i) {
2775                 if (slave->link != BOND_LINK_UP) {
2776                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2777                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2778
2779                                 slave->link  = BOND_LINK_UP;
2780                                 slave->state = BOND_STATE_ACTIVE;
2781
2782                                 /* primary_slave has no meaning in round-robin
2783                                  * mode. the window of a slave being up and
2784                                  * curr_active_slave being null after enslaving
2785                                  * is closed.
2786                                  */
2787                                 if (!oldcurrent) {
2788                                         pr_info(DRV_NAME
2789                                                ": %s: link status definitely "
2790                                                "up for interface %s, ",
2791                                                bond->dev->name,
2792                                                slave->dev->name);
2793                                         do_failover = 1;
2794                                 } else {
2795                                         pr_info(DRV_NAME
2796                                                ": %s: interface %s is now up\n",
2797                                                bond->dev->name,
2798                                                slave->dev->name);
2799                                 }
2800                         }
2801                 } else {
2802                         /* slave->link == BOND_LINK_UP */
2803
2804                         /* not all switches will respond to an arp request
2805                          * when the source ip is 0, so don't take the link down
2806                          * if we don't know our ip yet
2807                          */
2808                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2809                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2810
2811                                 slave->link  = BOND_LINK_DOWN;
2812                                 slave->state = BOND_STATE_BACKUP;
2813
2814                                 if (slave->link_failure_count < UINT_MAX)
2815                                         slave->link_failure_count++;
2816
2817                                 pr_info(DRV_NAME
2818                                        ": %s: interface %s is now down.\n",
2819                                        bond->dev->name,
2820                                        slave->dev->name);
2821
2822                                 if (slave == oldcurrent)
2823                                         do_failover = 1;
2824                         }
2825                 }
2826
2827                 /* note: if switch is in round-robin mode, all links
2828                  * must tx arp to ensure all links rx an arp - otherwise
2829                  * links may oscillate or not come up at all; if switch is
2830                  * in something like xor mode, there is nothing we can
2831                  * do - all replies will be rx'ed on same link causing slaves
2832                  * to be unstable during low/no traffic periods
2833                  */
2834                 if (IS_UP(slave->dev))
2835                         bond_arp_send_all(bond, slave);
2836         }
2837
2838         if (do_failover) {
2839                 write_lock_bh(&bond->curr_slave_lock);
2840
2841                 bond_select_active_slave(bond);
2842
2843                 write_unlock_bh(&bond->curr_slave_lock);
2844         }
2845
2846 re_arm:
2847         if (bond->params.arp_interval)
2848                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2849 out:
2850         read_unlock(&bond->lock);
2851 }
2852
2853 /*
2854  * Called to inspect slaves for active-backup mode ARP monitor link state
2855  * changes.  Sets new_link in slaves to specify what action should take
2856  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2857  * to link states must be committed.
2858  *
2859  * Called with bond->lock held for read.
2860  */
2861 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2862 {
2863         struct slave *slave;
2864         int i, commit = 0;
2865
2866         bond_for_each_slave(bond, slave, i) {
2867                 slave->new_link = BOND_LINK_NOCHANGE;
2868
2869                 if (slave->link != BOND_LINK_UP) {
2870                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2871                                            delta_in_ticks)) {
2872                                 slave->new_link = BOND_LINK_UP;
2873                                 commit++;
2874                         }
2875
2876                         continue;
2877                 }
2878
2879                 /*
2880                  * Give slaves 2*delta after being enslaved or made
2881                  * active.  This avoids bouncing, as the last receive
2882                  * times need a full ARP monitor cycle to be updated.
2883                  */
2884                 if (!time_after_eq(jiffies, slave->jiffies +
2885                                    2 * delta_in_ticks))
2886                         continue;
2887
2888                 /*
2889                  * Backup slave is down if:
2890                  * - No current_arp_slave AND
2891                  * - more than 3*delta since last receive AND
2892                  * - the bond has an IP address
2893                  *
2894                  * Note: a non-null current_arp_slave indicates
2895                  * the curr_active_slave went down and we are
2896                  * searching for a new one; under this condition
2897                  * we only take the curr_active_slave down - this
2898                  * gives each slave a chance to tx/rx traffic
2899                  * before being taken out
2900                  */
2901                 if (slave->state == BOND_STATE_BACKUP &&
2902                     !bond->current_arp_slave &&
2903                     time_after(jiffies, slave_last_rx(bond, slave) +
2904                                3 * delta_in_ticks)) {
2905                         slave->new_link = BOND_LINK_DOWN;
2906                         commit++;
2907                 }
2908
2909                 /*
2910                  * Active slave is down if:
2911                  * - more than 2*delta since transmitting OR
2912                  * - (more than 2*delta since receive AND
2913                  *    the bond has an IP address)
2914                  */
2915                 if ((slave->state == BOND_STATE_ACTIVE) &&
2916                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2917                                     2 * delta_in_ticks) ||
2918                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2919                                      + 2 * delta_in_ticks)))) {
2920                         slave->new_link = BOND_LINK_DOWN;
2921                         commit++;
2922                 }
2923         }
2924
2925         read_lock(&bond->curr_slave_lock);
2926
2927         /*
2928          * Trigger a commit if the primary option setting has changed.
2929          */
2930         if (bond->primary_slave &&
2931             (bond->primary_slave != bond->curr_active_slave) &&
2932             (bond->primary_slave->link == BOND_LINK_UP))
2933                 commit++;
2934
2935         read_unlock(&bond->curr_slave_lock);
2936
2937         return commit;
2938 }
2939
2940 /*
2941  * Called to commit link state changes noted by inspection step of
2942  * active-backup mode ARP monitor.
2943  *
2944  * Called with RTNL and bond->lock for read.
2945  */
2946 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2947 {
2948         struct slave *slave;
2949         int i;
2950
2951         bond_for_each_slave(bond, slave, i) {
2952                 switch (slave->new_link) {
2953                 case BOND_LINK_NOCHANGE:
2954                         continue;
2955
2956                 case BOND_LINK_UP:
2957                         write_lock_bh(&bond->curr_slave_lock);
2958
2959                         if (!bond->curr_active_slave &&
2960                             time_before_eq(jiffies, dev_trans_start(slave->dev) +
2961                                            delta_in_ticks)) {
2962                                 slave->link = BOND_LINK_UP;
2963                                 bond_change_active_slave(bond, slave);
2964                                 bond->current_arp_slave = NULL;
2965
2966                                 pr_info(DRV_NAME
2967                                        ": %s: %s is up and now the "
2968                                        "active interface\n",
2969                                        bond->dev->name, slave->dev->name);
2970
2971                         } else if (bond->curr_active_slave != slave) {
2972                                 /* this slave has just come up but we
2973                                  * already have a current slave; this can
2974                                  * also happen if bond_enslave adds a new
2975                                  * slave that is up while we are searching
2976                                  * for a new slave
2977                                  */
2978                                 slave->link = BOND_LINK_UP;
2979                                 bond_set_slave_inactive_flags(slave);
2980                                 bond->current_arp_slave = NULL;
2981
2982                                 pr_info(DRV_NAME
2983                                        ": %s: backup interface %s is now up\n",
2984                                        bond->dev->name, slave->dev->name);
2985                         }
2986
2987                         write_unlock_bh(&bond->curr_slave_lock);
2988
2989                         break;
2990
2991                 case BOND_LINK_DOWN:
2992                         if (slave->link_failure_count < UINT_MAX)
2993                                 slave->link_failure_count++;
2994
2995                         slave->link = BOND_LINK_DOWN;
2996
2997                         if (slave == bond->curr_active_slave) {
2998                                 pr_info(DRV_NAME
2999                                        ": %s: link status down for active "
3000                                        "interface %s, disabling it\n",
3001                                        bond->dev->name, slave->dev->name);
3002
3003                                 bond_set_slave_inactive_flags(slave);
3004
3005                                 write_lock_bh(&bond->curr_slave_lock);
3006
3007                                 bond_select_active_slave(bond);
3008                                 if (bond->curr_active_slave)
3009                                         bond->curr_active_slave->jiffies =
3010                                                 jiffies;
3011
3012                                 write_unlock_bh(&bond->curr_slave_lock);
3013
3014                                 bond->current_arp_slave = NULL;
3015
3016                         } else if (slave->state == BOND_STATE_BACKUP) {
3017                                 pr_info(DRV_NAME
3018                                        ": %s: backup interface %s is now down\n",
3019                                        bond->dev->name, slave->dev->name);
3020
3021                                 bond_set_slave_inactive_flags(slave);
3022                         }
3023                         break;
3024
3025                 default:
3026                         pr_err(DRV_NAME
3027                                ": %s: impossible: new_link %d on slave %s\n",
3028                                bond->dev->name, slave->new_link,
3029                                slave->dev->name);
3030                 }
3031         }
3032
3033         /*
3034          * No race with changes to primary via sysfs, as we hold rtnl.
3035          */
3036         if (bond->primary_slave &&
3037             (bond->primary_slave != bond->curr_active_slave) &&
3038             (bond->primary_slave->link == BOND_LINK_UP)) {
3039                 write_lock_bh(&bond->curr_slave_lock);
3040                 bond_change_active_slave(bond, bond->primary_slave);
3041                 write_unlock_bh(&bond->curr_slave_lock);
3042         }
3043
3044         bond_set_carrier(bond);
3045 }
3046
3047 /*
3048  * Send ARP probes for active-backup mode ARP monitor.
3049  *
3050  * Called with bond->lock held for read.
3051  */
3052 static void bond_ab_arp_probe(struct bonding *bond)
3053 {
3054         struct slave *slave;
3055         int i;
3056
3057         read_lock(&bond->curr_slave_lock);
3058
3059         if (bond->current_arp_slave && bond->curr_active_slave)
3060                 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3061                        bond->current_arp_slave->dev->name,
3062                        bond->curr_active_slave->dev->name);
3063
3064         if (bond->curr_active_slave) {
3065                 bond_arp_send_all(bond, bond->curr_active_slave);
3066                 read_unlock(&bond->curr_slave_lock);
3067                 return;
3068         }
3069
3070         read_unlock(&bond->curr_slave_lock);
3071
3072         /* if we don't have a curr_active_slave, search for the next available
3073          * backup slave from the current_arp_slave and make it the candidate
3074          * for becoming the curr_active_slave
3075          */
3076
3077         if (!bond->current_arp_slave) {
3078                 bond->current_arp_slave = bond->first_slave;
3079                 if (!bond->current_arp_slave)
3080                         return;
3081         }
3082
3083         bond_set_slave_inactive_flags(bond->current_arp_slave);
3084
3085         /* search for next candidate */
3086         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3087                 if (IS_UP(slave->dev)) {
3088                         slave->link = BOND_LINK_BACK;
3089                         bond_set_slave_active_flags(slave);
3090                         bond_arp_send_all(bond, slave);
3091                         slave->jiffies = jiffies;
3092                         bond->current_arp_slave = slave;
3093                         break;
3094                 }
3095
3096                 /* if the link state is up at this point, we
3097                  * mark it down - this can happen if we have
3098                  * simultaneous link failures and
3099                  * reselect_active_interface doesn't make this
3100                  * one the current slave so it is still marked
3101                  * up when it is actually down
3102                  */
3103                 if (slave->link == BOND_LINK_UP) {
3104                         slave->link = BOND_LINK_DOWN;
3105                         if (slave->link_failure_count < UINT_MAX)
3106                                 slave->link_failure_count++;
3107
3108                         bond_set_slave_inactive_flags(slave);
3109
3110                         pr_info(DRV_NAME
3111                                ": %s: backup interface %s is now down.\n",
3112                                bond->dev->name, slave->dev->name);
3113                 }
3114         }
3115 }
3116
3117 void bond_activebackup_arp_mon(struct work_struct *work)
3118 {
3119         struct bonding *bond = container_of(work, struct bonding,
3120                                             arp_work.work);
3121         int delta_in_ticks;
3122
3123         read_lock(&bond->lock);
3124
3125         if (bond->kill_timers)
3126                 goto out;
3127
3128         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3129
3130         if (bond->slave_cnt == 0)
3131                 goto re_arm;
3132
3133         if (bond->send_grat_arp) {
3134                 read_lock(&bond->curr_slave_lock);
3135                 bond_send_gratuitous_arp(bond);
3136                 read_unlock(&bond->curr_slave_lock);
3137         }
3138
3139         if (bond->send_unsol_na) {
3140                 read_lock(&bond->curr_slave_lock);
3141                 bond_send_unsolicited_na(bond);
3142                 read_unlock(&bond->curr_slave_lock);
3143         }
3144
3145         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3146                 read_unlock(&bond->lock);
3147                 rtnl_lock();
3148                 read_lock(&bond->lock);
3149
3150                 bond_ab_arp_commit(bond, delta_in_ticks);
3151
3152                 read_unlock(&bond->lock);
3153                 rtnl_unlock();
3154                 read_lock(&bond->lock);
3155         }
3156
3157         bond_ab_arp_probe(bond);
3158
3159 re_arm:
3160         if (bond->params.arp_interval)
3161                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3162 out:
3163         read_unlock(&bond->lock);
3164 }
3165
3166 /*------------------------------ proc/seq_file-------------------------------*/
3167
3168 #ifdef CONFIG_PROC_FS
3169
3170 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3171         __acquires(&dev_base_lock)
3172         __acquires(&bond->lock)
3173 {
3174         struct bonding *bond = seq->private;
3175         loff_t off = 0;
3176         struct slave *slave;
3177         int i;
3178
3179         /* make sure the bond won't be taken away */
3180         read_lock(&dev_base_lock);
3181         read_lock(&bond->lock);
3182
3183         if (*pos == 0)
3184                 return SEQ_START_TOKEN;
3185
3186         bond_for_each_slave(bond, slave, i) {
3187                 if (++off == *pos)
3188                         return slave;
3189         }
3190
3191         return NULL;
3192 }
3193
3194 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3195 {
3196         struct bonding *bond = seq->private;
3197         struct slave *slave = v;
3198
3199         ++*pos;
3200         if (v == SEQ_START_TOKEN)
3201                 return bond->first_slave;
3202
3203         slave = slave->next;
3204
3205         return (slave == bond->first_slave) ? NULL : slave;
3206 }
3207
3208 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3209         __releases(&bond->lock)
3210         __releases(&dev_base_lock)
3211 {
3212         struct bonding *bond = seq->private;
3213
3214         read_unlock(&bond->lock);
3215         read_unlock(&dev_base_lock);
3216 }
3217
3218 static void bond_info_show_master(struct seq_file *seq)
3219 {
3220         struct bonding *bond = seq->private;
3221         struct slave *curr;
3222         int i;
3223
3224         read_lock(&bond->curr_slave_lock);
3225         curr = bond->curr_active_slave;
3226         read_unlock(&bond->curr_slave_lock);
3227
3228         seq_printf(seq, "Bonding Mode: %s",
3229                    bond_mode_name(bond->params.mode));
3230
3231         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3232             bond->params.fail_over_mac)
3233                 seq_printf(seq, " (fail_over_mac %s)",
3234                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3235
3236         seq_printf(seq, "\n");
3237
3238         if (bond->params.mode == BOND_MODE_XOR ||
3239                 bond->params.mode == BOND_MODE_8023AD) {
3240                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3241                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3242                         bond->params.xmit_policy);
3243         }
3244
3245         if (USES_PRIMARY(bond->params.mode)) {
3246                 seq_printf(seq, "Primary Slave: %s\n",
3247                            (bond->primary_slave) ?
3248                            bond->primary_slave->dev->name : "None");
3249
3250                 seq_printf(seq, "Currently Active Slave: %s\n",
3251                            (curr) ? curr->dev->name : "None");
3252         }
3253
3254         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3255                    "up" : "down");
3256         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3257         seq_printf(seq, "Up Delay (ms): %d\n",
3258                    bond->params.updelay * bond->params.miimon);
3259         seq_printf(seq, "Down Delay (ms): %d\n",
3260                    bond->params.downdelay * bond->params.miimon);
3261
3262
3263         /* ARP information */
3264         if (bond->params.arp_interval > 0) {
3265                 int printed = 0;
3266                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3267                                 bond->params.arp_interval);
3268
3269                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3270
3271                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3272                         if (!bond->params.arp_targets[i])
3273                                 break;
3274                         if (printed)
3275                                 seq_printf(seq, ",");
3276                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3277                         printed = 1;
3278                 }
3279                 seq_printf(seq, "\n");
3280         }
3281
3282         if (bond->params.mode == BOND_MODE_8023AD) {
3283                 struct ad_info ad_info;
3284
3285                 seq_puts(seq, "\n802.3ad info\n");
3286                 seq_printf(seq, "LACP rate: %s\n",
3287                            (bond->params.lacp_fast) ? "fast" : "slow");
3288                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3289                            ad_select_tbl[bond->params.ad_select].modename);
3290
3291                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3292                         seq_printf(seq, "bond %s has no active aggregator\n",
3293                                    bond->dev->name);
3294                 } else {
3295                         seq_printf(seq, "Active Aggregator Info:\n");
3296
3297                         seq_printf(seq, "\tAggregator ID: %d\n",
3298                                    ad_info.aggregator_id);
3299                         seq_printf(seq, "\tNumber of ports: %d\n",
3300                                    ad_info.ports);
3301                         seq_printf(seq, "\tActor Key: %d\n",
3302                                    ad_info.actor_key);
3303                         seq_printf(seq, "\tPartner Key: %d\n",
3304                                    ad_info.partner_key);
3305                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3306                                    ad_info.partner_system);
3307                 }
3308         }
3309 }
3310
3311 static void bond_info_show_slave(struct seq_file *seq,
3312                                  const struct slave *slave)
3313 {
3314         struct bonding *bond = seq->private;
3315
3316         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3317         seq_printf(seq, "MII Status: %s\n",
3318                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3319         seq_printf(seq, "Link Failure Count: %u\n",
3320                    slave->link_failure_count);
3321
3322         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3323
3324         if (bond->params.mode == BOND_MODE_8023AD) {
3325                 const struct aggregator *agg
3326                         = SLAVE_AD_INFO(slave).port.aggregator;
3327
3328                 if (agg)
3329                         seq_printf(seq, "Aggregator ID: %d\n",
3330                                    agg->aggregator_identifier);
3331                 else
3332                         seq_puts(seq, "Aggregator ID: N/A\n");
3333         }
3334 }
3335
3336 static int bond_info_seq_show(struct seq_file *seq, void *v)
3337 {
3338         if (v == SEQ_START_TOKEN) {
3339                 seq_printf(seq, "%s\n", version);
3340                 bond_info_show_master(seq);
3341         } else
3342                 bond_info_show_slave(seq, v);
3343
3344         return 0;
3345 }
3346
3347 static const struct seq_operations bond_info_seq_ops = {
3348         .start = bond_info_seq_start,
3349         .next  = bond_info_seq_next,
3350         .stop  = bond_info_seq_stop,
3351         .show  = bond_info_seq_show,
3352 };
3353
3354 static int bond_info_open(struct inode *inode, struct file *file)
3355 {
3356         struct seq_file *seq;
3357         struct proc_dir_entry *proc;
3358         int res;
3359
3360         res = seq_open(file, &bond_info_seq_ops);
3361         if (!res) {
3362                 /* recover the pointer buried in proc_dir_entry data */
3363                 seq = file->private_data;
3364                 proc = PDE(inode);
3365                 seq->private = proc->data;
3366         }
3367
3368         return res;
3369 }
3370
3371 static const struct file_operations bond_info_fops = {
3372         .owner   = THIS_MODULE,
3373         .open    = bond_info_open,
3374         .read    = seq_read,
3375         .llseek  = seq_lseek,
3376         .release = seq_release,
3377 };
3378
3379 static int bond_create_proc_entry(struct bonding *bond)
3380 {
3381         struct net_device *bond_dev = bond->dev;
3382
3383         if (bond_proc_dir) {
3384                 bond->proc_entry = proc_create_data(bond_dev->name,
3385                                                     S_IRUGO, bond_proc_dir,
3386                                                     &bond_info_fops, bond);
3387                 if (bond->proc_entry == NULL)
3388                         pr_warning(DRV_NAME
3389                                ": Warning: Cannot create /proc/net/%s/%s\n",
3390                                DRV_NAME, bond_dev->name);
3391                 else
3392                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3393         }
3394
3395         return 0;
3396 }
3397
3398 static void bond_remove_proc_entry(struct bonding *bond)
3399 {
3400         if (bond_proc_dir && bond->proc_entry) {
3401                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3402                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3403                 bond->proc_entry = NULL;
3404         }
3405 }
3406
3407 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3408  * Caller must hold rtnl_lock.
3409  */
3410 static void bond_create_proc_dir(void)
3411 {
3412         if (!bond_proc_dir) {
3413                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3414                 if (!bond_proc_dir)
3415                         pr_warning(DRV_NAME
3416                                 ": Warning: cannot create /proc/net/%s\n",
3417                                 DRV_NAME);
3418         }
3419 }
3420
3421 /* Destroy the bonding directory under /proc/net, if empty.
3422  * Caller must hold rtnl_lock.
3423  */
3424 static void bond_destroy_proc_dir(void)
3425 {
3426         if (bond_proc_dir) {
3427                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3428                 bond_proc_dir = NULL;
3429         }
3430 }
3431
3432 #else /* !CONFIG_PROC_FS */
3433
3434 static int bond_create_proc_entry(struct bonding *bond)
3435 {
3436 }
3437
3438 static void bond_remove_proc_entry(struct bonding *bond)
3439 {
3440 }
3441
3442 static void bond_create_proc_dir(void)
3443 {
3444 }
3445
3446 static void bond_destroy_proc_dir(void)
3447 {
3448 }
3449
3450 #endif /* CONFIG_PROC_FS */
3451
3452
3453 /*-------------------------- netdev event handling --------------------------*/
3454
3455 /*
3456  * Change device name
3457  */
3458 static int bond_event_changename(struct bonding *bond)
3459 {
3460         bond_remove_proc_entry(bond);
3461         bond_create_proc_entry(bond);
3462
3463         bond_destroy_sysfs_entry(bond);
3464         bond_create_sysfs_entry(bond);
3465
3466         return NOTIFY_DONE;
3467 }
3468
3469 static int bond_master_netdev_event(unsigned long event,
3470                                     struct net_device *bond_dev)
3471 {
3472         struct bonding *event_bond = netdev_priv(bond_dev);
3473
3474         switch (event) {
3475         case NETDEV_CHANGENAME:
3476                 return bond_event_changename(event_bond);
3477         case NETDEV_UNREGISTER:
3478                 bond_release_all(event_bond->dev);
3479                 break;
3480         default:
3481                 break;
3482         }
3483
3484         return NOTIFY_DONE;
3485 }
3486
3487 static int bond_slave_netdev_event(unsigned long event,
3488                                    struct net_device *slave_dev)
3489 {
3490         struct net_device *bond_dev = slave_dev->master;
3491         struct bonding *bond = netdev_priv(bond_dev);
3492
3493         switch (event) {
3494         case NETDEV_UNREGISTER:
3495                 if (bond_dev) {
3496                         if (bond->setup_by_slave)
3497                                 bond_release_and_destroy(bond_dev, slave_dev);
3498                         else
3499                                 bond_release(bond_dev, slave_dev);
3500                 }
3501                 break;
3502         case NETDEV_CHANGE:
3503                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3504                         struct slave *slave;
3505
3506                         slave = bond_get_slave_by_dev(bond, slave_dev);
3507                         if (slave) {
3508                                 u16 old_speed = slave->speed;
3509                                 u16 old_duplex = slave->duplex;
3510
3511                                 bond_update_speed_duplex(slave);
3512
3513                                 if (bond_is_lb(bond))
3514                                         break;
3515
3516                                 if (old_speed != slave->speed)
3517                                         bond_3ad_adapter_speed_changed(slave);
3518                                 if (old_duplex != slave->duplex)
3519                                         bond_3ad_adapter_duplex_changed(slave);
3520                         }
3521                 }
3522
3523                 break;
3524         case NETDEV_DOWN:
3525                 /*
3526                  * ... Or is it this?
3527                  */
3528                 break;
3529         case NETDEV_CHANGEMTU:
3530                 /*
3531                  * TODO: Should slaves be allowed to
3532                  * independently alter their MTU?  For
3533                  * an active-backup bond, slaves need
3534                  * not be the same type of device, so
3535                  * MTUs may vary.  For other modes,
3536                  * slaves arguably should have the
3537                  * same MTUs. To do this, we'd need to
3538                  * take over the slave's change_mtu
3539                  * function for the duration of their
3540                  * servitude.
3541                  */
3542                 break;
3543         case NETDEV_CHANGENAME:
3544                 /*
3545                  * TODO: handle changing the primary's name
3546                  */
3547                 break;
3548         case NETDEV_FEAT_CHANGE:
3549                 bond_compute_features(bond);
3550                 break;
3551         default:
3552                 break;
3553         }
3554
3555         return NOTIFY_DONE;
3556 }
3557
3558 /*
3559  * bond_netdev_event: handle netdev notifier chain events.
3560  *
3561  * This function receives events for the netdev chain.  The caller (an
3562  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3563  * locks for us to safely manipulate the slave devices (RTNL lock,
3564  * dev_probe_lock).
3565  */
3566 static int bond_netdev_event(struct notifier_block *this,
3567                              unsigned long event, void *ptr)
3568 {
3569         struct net_device *event_dev = (struct net_device *)ptr;
3570
3571         if (dev_net(event_dev) != &init_net)
3572                 return NOTIFY_DONE;
3573
3574         pr_debug("event_dev: %s, event: %lx\n",
3575                 (event_dev ? event_dev->name : "None"),
3576                 event);
3577
3578         if (!(event_dev->priv_flags & IFF_BONDING))
3579                 return NOTIFY_DONE;
3580
3581         if (event_dev->flags & IFF_MASTER) {
3582                 pr_debug("IFF_MASTER\n");
3583                 return bond_master_netdev_event(event, event_dev);
3584         }
3585
3586         if (event_dev->flags & IFF_SLAVE) {
3587                 pr_debug("IFF_SLAVE\n");
3588                 return bond_slave_netdev_event(event, event_dev);
3589         }
3590
3591         return NOTIFY_DONE;
3592 }
3593
3594 /*
3595  * bond_inetaddr_event: handle inetaddr notifier chain events.
3596  *
3597  * We keep track of device IPs primarily to use as source addresses in
3598  * ARP monitor probes (rather than spewing out broadcasts all the time).
3599  *
3600  * We track one IP for the main device (if it has one), plus one per VLAN.
3601  */
3602 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3603 {
3604         struct in_ifaddr *ifa = ptr;
3605         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3606         struct bonding *bond;
3607         struct vlan_entry *vlan;
3608
3609         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3610                 return NOTIFY_DONE;
3611
3612         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3613                 if (bond->dev == event_dev) {
3614                         switch (event) {
3615                         case NETDEV_UP:
3616                                 bond->master_ip = ifa->ifa_local;
3617                                 return NOTIFY_OK;
3618                         case NETDEV_DOWN:
3619                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3620                                 return NOTIFY_OK;
3621                         default:
3622                                 return NOTIFY_DONE;
3623                         }
3624                 }
3625
3626                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3627                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3628                         if (vlan_dev == event_dev) {
3629                                 switch (event) {
3630                                 case NETDEV_UP:
3631                                         vlan->vlan_ip = ifa->ifa_local;
3632                                         return NOTIFY_OK;
3633                                 case NETDEV_DOWN:
3634                                         vlan->vlan_ip =
3635                                                 bond_glean_dev_ip(vlan_dev);
3636                                         return NOTIFY_OK;
3637                                 default:
3638                                         return NOTIFY_DONE;
3639                                 }
3640                         }
3641                 }
3642         }
3643         return NOTIFY_DONE;
3644 }
3645
3646 static struct notifier_block bond_netdev_notifier = {
3647         .notifier_call = bond_netdev_event,
3648 };
3649
3650 static struct notifier_block bond_inetaddr_notifier = {
3651         .notifier_call = bond_inetaddr_event,
3652 };
3653
3654 /*-------------------------- Packet type handling ---------------------------*/
3655
3656 /* register to receive lacpdus on a bond */
3657 static void bond_register_lacpdu(struct bonding *bond)
3658 {
3659         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3660
3661         /* initialize packet type */
3662         pk_type->type = PKT_TYPE_LACPDU;
3663         pk_type->dev = bond->dev;
3664         pk_type->func = bond_3ad_lacpdu_recv;
3665
3666         dev_add_pack(pk_type);
3667 }
3668
3669 /* unregister to receive lacpdus on a bond */
3670 static void bond_unregister_lacpdu(struct bonding *bond)
3671 {
3672         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3673 }
3674
3675 void bond_register_arp(struct bonding *bond)
3676 {
3677         struct packet_type *pt = &bond->arp_mon_pt;
3678
3679         if (pt->type)
3680                 return;
3681
3682         pt->type = htons(ETH_P_ARP);
3683         pt->dev = bond->dev;
3684         pt->func = bond_arp_rcv;
3685         dev_add_pack(pt);
3686 }
3687
3688 void bond_unregister_arp(struct bonding *bond)
3689 {
3690         struct packet_type *pt = &bond->arp_mon_pt;
3691
3692         dev_remove_pack(pt);
3693         pt->type = 0;
3694 }
3695
3696 /*---------------------------- Hashing Policies -----------------------------*/
3697
3698 /*
3699  * Hash for the output device based upon layer 2 and layer 3 data. If
3700  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3701  */
3702 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3703                                      struct net_device *bond_dev, int count)
3704 {
3705         struct ethhdr *data = (struct ethhdr *)skb->data;
3706         struct iphdr *iph = ip_hdr(skb);
3707
3708         if (skb->protocol == htons(ETH_P_IP)) {
3709                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3710                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3711         }
3712
3713         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3714 }
3715
3716 /*
3717  * Hash for the output device based upon layer 3 and layer 4 data. If
3718  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3719  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3720  */
3721 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3722                                     struct net_device *bond_dev, int count)
3723 {
3724         struct ethhdr *data = (struct ethhdr *)skb->data;
3725         struct iphdr *iph = ip_hdr(skb);
3726         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3727         int layer4_xor = 0;
3728
3729         if (skb->protocol == htons(ETH_P_IP)) {
3730                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3731                     (iph->protocol == IPPROTO_TCP ||
3732                      iph->protocol == IPPROTO_UDP)) {
3733                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3734                 }
3735                 return (layer4_xor ^
3736                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3737
3738         }
3739
3740         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3741 }
3742
3743 /*
3744  * Hash for the output device based upon layer 2 data
3745  */
3746 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3747                                    struct net_device *bond_dev, int count)
3748 {
3749         struct ethhdr *data = (struct ethhdr *)skb->data;
3750
3751         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3752 }
3753
3754 /*-------------------------- Device entry points ----------------------------*/
3755
3756 static int bond_open(struct net_device *bond_dev)
3757 {
3758         struct bonding *bond = netdev_priv(bond_dev);
3759
3760         bond->kill_timers = 0;
3761
3762         if (bond_is_lb(bond)) {
3763                 /* bond_alb_initialize must be called before the timer
3764                  * is started.
3765                  */
3766                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3767                         /* something went wrong - fail the open operation */
3768                         return -1;
3769                 }
3770
3771                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3772                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3773         }
3774
3775         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3776                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3777                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3778         }
3779
3780         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3781                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3782                         INIT_DELAYED_WORK(&bond->arp_work,
3783                                           bond_activebackup_arp_mon);
3784                 else
3785                         INIT_DELAYED_WORK(&bond->arp_work,
3786                                           bond_loadbalance_arp_mon);
3787
3788                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3789                 if (bond->params.arp_validate)
3790                         bond_register_arp(bond);
3791         }
3792
3793         if (bond->params.mode == BOND_MODE_8023AD) {
3794                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3795                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3796                 /* register to receive LACPDUs */
3797                 bond_register_lacpdu(bond);
3798                 bond_3ad_initiate_agg_selection(bond, 1);
3799         }
3800
3801         return 0;
3802 }
3803
3804 static int bond_close(struct net_device *bond_dev)
3805 {
3806         struct bonding *bond = netdev_priv(bond_dev);
3807
3808         if (bond->params.mode == BOND_MODE_8023AD) {
3809                 /* Unregister the receive of LACPDUs */
3810                 bond_unregister_lacpdu(bond);
3811         }
3812
3813         if (bond->params.arp_validate)
3814                 bond_unregister_arp(bond);
3815
3816         write_lock_bh(&bond->lock);
3817
3818         bond->send_grat_arp = 0;
3819         bond->send_unsol_na = 0;
3820
3821         /* signal timers not to re-arm */
3822         bond->kill_timers = 1;
3823
3824         write_unlock_bh(&bond->lock);
3825
3826         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3827                 cancel_delayed_work(&bond->mii_work);
3828         }
3829
3830         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3831                 cancel_delayed_work(&bond->arp_work);
3832         }
3833
3834         switch (bond->params.mode) {
3835         case BOND_MODE_8023AD:
3836                 cancel_delayed_work(&bond->ad_work);
3837                 break;
3838         case BOND_MODE_TLB:
3839         case BOND_MODE_ALB:
3840                 cancel_delayed_work(&bond->alb_work);
3841                 break;
3842         default:
3843                 break;
3844         }
3845
3846
3847         if (bond_is_lb(bond)) {
3848                 /* Must be called only after all
3849                  * slaves have been released
3850                  */
3851                 bond_alb_deinitialize(bond);
3852         }
3853
3854         return 0;
3855 }
3856
3857 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3858 {
3859         struct bonding *bond = netdev_priv(bond_dev);
3860         struct net_device_stats *stats = &bond->stats;
3861         struct net_device_stats local_stats;
3862         struct slave *slave;
3863         int i;
3864
3865         memset(&local_stats, 0, sizeof(struct net_device_stats));
3866
3867         read_lock_bh(&bond->lock);
3868
3869         bond_for_each_slave(bond, slave, i) {
3870                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3871
3872                 local_stats.rx_packets += sstats->rx_packets;
3873                 local_stats.rx_bytes += sstats->rx_bytes;
3874                 local_stats.rx_errors += sstats->rx_errors;
3875                 local_stats.rx_dropped += sstats->rx_dropped;
3876
3877                 local_stats.tx_packets += sstats->tx_packets;
3878                 local_stats.tx_bytes += sstats->tx_bytes;
3879                 local_stats.tx_errors += sstats->tx_errors;
3880                 local_stats.tx_dropped += sstats->tx_dropped;
3881
3882                 local_stats.multicast += sstats->multicast;
3883                 local_stats.collisions += sstats->collisions;
3884
3885                 local_stats.rx_length_errors += sstats->rx_length_errors;
3886                 local_stats.rx_over_errors += sstats->rx_over_errors;
3887                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3888                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3889                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3890                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3891
3892                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3893                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3894                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3895                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3896                 local_stats.tx_window_errors += sstats->tx_window_errors;
3897         }
3898
3899         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3900
3901         read_unlock_bh(&bond->lock);
3902
3903         return stats;
3904 }
3905
3906 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3907 {
3908         struct net_device *slave_dev = NULL;
3909         struct ifbond k_binfo;
3910         struct ifbond __user *u_binfo = NULL;
3911         struct ifslave k_sinfo;
3912         struct ifslave __user *u_sinfo = NULL;
3913         struct mii_ioctl_data *mii = NULL;
3914         int res = 0;
3915
3916         pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3917                 bond_dev->name, cmd);
3918
3919         switch (cmd) {
3920         case SIOCGMIIPHY:
3921                 mii = if_mii(ifr);
3922                 if (!mii)
3923                         return -EINVAL;
3924
3925                 mii->phy_id = 0;
3926                 /* Fall Through */
3927         case SIOCGMIIREG:
3928                 /*
3929                  * We do this again just in case we were called by SIOCGMIIREG
3930                  * instead of SIOCGMIIPHY.
3931                  */
3932                 mii = if_mii(ifr);
3933                 if (!mii)
3934                         return -EINVAL;
3935
3936
3937                 if (mii->reg_num == 1) {
3938                         struct bonding *bond = netdev_priv(bond_dev);
3939                         mii->val_out = 0;
3940                         read_lock(&bond->lock);
3941                         read_lock(&bond->curr_slave_lock);
3942                         if (netif_carrier_ok(bond->dev))
3943                                 mii->val_out = BMSR_LSTATUS;
3944
3945                         read_unlock(&bond->curr_slave_lock);
3946                         read_unlock(&bond->lock);
3947                 }
3948
3949                 return 0;
3950         case BOND_INFO_QUERY_OLD:
3951         case SIOCBONDINFOQUERY:
3952                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3953
3954                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3955                         return -EFAULT;
3956
3957                 res = bond_info_query(bond_dev, &k_binfo);
3958                 if (res == 0 &&
3959                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3960                         return -EFAULT;
3961
3962                 return res;
3963         case BOND_SLAVE_INFO_QUERY_OLD:
3964         case SIOCBONDSLAVEINFOQUERY:
3965                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3966
3967                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3968                         return -EFAULT;
3969
3970                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3971                 if (res == 0 &&
3972                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3973                         return -EFAULT;
3974
3975                 return res;
3976         default:
3977                 /* Go on */
3978                 break;
3979         }
3980
3981         if (!capable(CAP_NET_ADMIN))
3982                 return -EPERM;
3983
3984         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3985
3986         pr_debug("slave_dev=%p: \n", slave_dev);
3987
3988         if (!slave_dev)
3989                 res = -ENODEV;
3990         else {
3991                 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
3992                 switch (cmd) {
3993                 case BOND_ENSLAVE_OLD:
3994                 case SIOCBONDENSLAVE:
3995                         res = bond_enslave(bond_dev, slave_dev);
3996                         break;
3997                 case BOND_RELEASE_OLD:
3998                 case SIOCBONDRELEASE:
3999                         res = bond_release(bond_dev, slave_dev);
4000                         break;
4001                 case BOND_SETHWADDR_OLD:
4002                 case SIOCBONDSETHWADDR:
4003                         res = bond_sethwaddr(bond_dev, slave_dev);
4004                         break;
4005                 case BOND_CHANGE_ACTIVE_OLD:
4006                 case SIOCBONDCHANGEACTIVE:
4007                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4008                         break;
4009                 default:
4010                         res = -EOPNOTSUPP;
4011                 }
4012
4013                 dev_put(slave_dev);
4014         }
4015
4016         return res;
4017 }
4018
4019 static void bond_set_multicast_list(struct net_device *bond_dev)
4020 {
4021         struct bonding *bond = netdev_priv(bond_dev);
4022         struct dev_mc_list *dmi;
4023
4024         /*
4025          * Do promisc before checking multicast_mode
4026          */
4027         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4028                 /*
4029                  * FIXME: Need to handle the error when one of the multi-slaves
4030                  * encounters error.
4031                  */
4032                 bond_set_promiscuity(bond, 1);
4033
4034
4035         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4036                 bond_set_promiscuity(bond, -1);
4037
4038
4039         /* set allmulti flag to slaves */
4040         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4041                 /*
4042                  * FIXME: Need to handle the error when one of the multi-slaves
4043                  * encounters error.
4044                  */
4045                 bond_set_allmulti(bond, 1);
4046
4047
4048         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4049                 bond_set_allmulti(bond, -1);
4050
4051
4052         read_lock(&bond->lock);
4053
4054         bond->flags = bond_dev->flags;
4055
4056         /* looking for addresses to add to slaves' mc list */
4057         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4058                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4059                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4060         }
4061
4062         /* looking for addresses to delete from slaves' list */
4063         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4064                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4065                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4066         }
4067
4068         /* save master's multicast list */
4069         bond_mc_list_destroy(bond);
4070         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4071
4072         read_unlock(&bond->lock);
4073 }
4074
4075 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4076 {
4077         struct bonding *bond = netdev_priv(dev);
4078         struct slave *slave = bond->first_slave;
4079
4080         if (slave) {
4081                 const struct net_device_ops *slave_ops
4082                         = slave->dev->netdev_ops;
4083                 if (slave_ops->ndo_neigh_setup)
4084                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4085         }
4086         return 0;
4087 }
4088
4089 /*
4090  * Change the MTU of all of a master's slaves to match the master
4091  */
4092 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4093 {
4094         struct bonding *bond = netdev_priv(bond_dev);
4095         struct slave *slave, *stop_at;
4096         int res = 0;
4097         int i;
4098
4099         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4100                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4101
4102         /* Can't hold bond->lock with bh disabled here since
4103          * some base drivers panic. On the other hand we can't
4104          * hold bond->lock without bh disabled because we'll
4105          * deadlock. The only solution is to rely on the fact
4106          * that we're under rtnl_lock here, and the slaves
4107          * list won't change. This doesn't solve the problem
4108          * of setting the slave's MTU while it is
4109          * transmitting, but the assumption is that the base
4110          * driver can handle that.
4111          *
4112          * TODO: figure out a way to safely iterate the slaves
4113          * list, but without holding a lock around the actual
4114          * call to the base driver.
4115          */
4116
4117         bond_for_each_slave(bond, slave, i) {
4118                 pr_debug("s %p s->p %p c_m %p\n", slave,
4119                         slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4120
4121                 res = dev_set_mtu(slave->dev, new_mtu);
4122
4123                 if (res) {
4124                         /* If we failed to set the slave's mtu to the new value
4125                          * we must abort the operation even in ACTIVE_BACKUP
4126                          * mode, because if we allow the backup slaves to have
4127                          * different mtu values than the active slave we'll
4128                          * need to change their mtu when doing a failover. That
4129                          * means changing their mtu from timer context, which
4130                          * is probably not a good idea.
4131                          */
4132                         pr_debug("err %d %s\n", res, slave->dev->name);
4133                         goto unwind;
4134                 }
4135         }
4136
4137         bond_dev->mtu = new_mtu;
4138
4139         return 0;
4140
4141 unwind:
4142         /* unwind from head to the slave that failed */
4143         stop_at = slave;
4144         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4145                 int tmp_res;
4146
4147                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4148                 if (tmp_res) {
4149                         pr_debug("unwind err %d dev %s\n", tmp_res,
4150                                 slave->dev->name);
4151                 }
4152         }
4153
4154         return res;
4155 }
4156
4157 /*
4158  * Change HW address
4159  *
4160  * Note that many devices must be down to change the HW address, and
4161  * downing the master releases all slaves.  We can make bonds full of
4162  * bonding devices to test this, however.
4163  */
4164 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4165 {
4166         struct bonding *bond = netdev_priv(bond_dev);
4167         struct sockaddr *sa = addr, tmp_sa;
4168         struct slave *slave, *stop_at;
4169         int res = 0;
4170         int i;
4171
4172         if (bond->params.mode == BOND_MODE_ALB)
4173                 return bond_alb_set_mac_address(bond_dev, addr);
4174
4175
4176         pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4177
4178         /*
4179          * If fail_over_mac is set to active, do nothing and return
4180          * success.  Returning an error causes ifenslave to fail.
4181          */
4182         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4183                 return 0;
4184
4185         if (!is_valid_ether_addr(sa->sa_data))
4186                 return -EADDRNOTAVAIL;
4187
4188         /* Can't hold bond->lock with bh disabled here since
4189          * some base drivers panic. On the other hand we can't
4190          * hold bond->lock without bh disabled because we'll
4191          * deadlock. The only solution is to rely on the fact
4192          * that we're under rtnl_lock here, and the slaves
4193          * list won't change. This doesn't solve the problem
4194          * of setting the slave's hw address while it is
4195          * transmitting, but the assumption is that the base
4196          * driver can handle that.
4197          *
4198          * TODO: figure out a way to safely iterate the slaves
4199          * list, but without holding a lock around the actual
4200          * call to the base driver.
4201          */
4202
4203         bond_for_each_slave(bond, slave, i) {
4204                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4205                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4206
4207                 if (slave_ops->ndo_set_mac_address == NULL) {
4208                         res = -EOPNOTSUPP;
4209                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4210                         goto unwind;
4211                 }
4212
4213                 res = dev_set_mac_address(slave->dev, addr);
4214                 if (res) {
4215                         /* TODO: consider downing the slave
4216                          * and retry ?
4217                          * User should expect communications
4218                          * breakage anyway until ARP finish
4219                          * updating, so...
4220                          */
4221                         pr_debug("err %d %s\n", res, slave->dev->name);
4222                         goto unwind;
4223                 }
4224         }
4225
4226         /* success */
4227         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4228         return 0;
4229
4230 unwind:
4231         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4232         tmp_sa.sa_family = bond_dev->type;
4233
4234         /* unwind from head to the slave that failed */
4235         stop_at = slave;
4236         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4237                 int tmp_res;
4238
4239                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4240                 if (tmp_res) {
4241                         pr_debug("unwind err %d dev %s\n", tmp_res,
4242                                 slave->dev->name);
4243                 }
4244         }
4245
4246         return res;
4247 }
4248
4249 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4250 {
4251         struct bonding *bond = netdev_priv(bond_dev);
4252         struct slave *slave, *start_at;
4253         int i, slave_no, res = 1;
4254
4255         read_lock(&bond->lock);
4256
4257         if (!BOND_IS_OK(bond))
4258                 goto out;
4259
4260         /*
4261          * Concurrent TX may collide on rr_tx_counter; we accept that
4262          * as being rare enough not to justify using an atomic op here
4263          */
4264         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4265
4266         bond_for_each_slave(bond, slave, i) {
4267                 slave_no--;
4268                 if (slave_no < 0)
4269                         break;
4270         }
4271
4272         start_at = slave;
4273         bond_for_each_slave_from(bond, slave, i, start_at) {
4274                 if (IS_UP(slave->dev) &&
4275                     (slave->link == BOND_LINK_UP) &&
4276                     (slave->state == BOND_STATE_ACTIVE)) {
4277                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4278                         break;
4279                 }
4280         }
4281
4282 out:
4283         if (res) {
4284                 /* no suitable interface, frame not sent */
4285                 dev_kfree_skb(skb);
4286         }
4287         read_unlock(&bond->lock);
4288         return 0;
4289 }
4290
4291
4292 /*
4293  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4294  * the bond has a usable interface.
4295  */
4296 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4297 {
4298         struct bonding *bond = netdev_priv(bond_dev);
4299         int res = 1;
4300
4301         read_lock(&bond->lock);
4302         read_lock(&bond->curr_slave_lock);
4303
4304         if (!BOND_IS_OK(bond))
4305                 goto out;
4306
4307         if (!bond->curr_active_slave)
4308                 goto out;
4309
4310         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4311
4312 out:
4313         if (res)
4314                 /* no suitable interface, frame not sent */
4315                 dev_kfree_skb(skb);
4316
4317         read_unlock(&bond->curr_slave_lock);
4318         read_unlock(&bond->lock);
4319         return 0;
4320 }
4321
4322 /*
4323  * In bond_xmit_xor() , we determine the output device by using a pre-
4324  * determined xmit_hash_policy(), If the selected device is not enabled,
4325  * find the next active slave.
4326  */
4327 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4328 {
4329         struct bonding *bond = netdev_priv(bond_dev);
4330         struct slave *slave, *start_at;
4331         int slave_no;
4332         int i;
4333         int res = 1;
4334
4335         read_lock(&bond->lock);
4336
4337         if (!BOND_IS_OK(bond))
4338                 goto out;
4339
4340         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4341
4342         bond_for_each_slave(bond, slave, i) {
4343                 slave_no--;
4344                 if (slave_no < 0)
4345                         break;
4346         }
4347
4348         start_at = slave;
4349
4350         bond_for_each_slave_from(bond, slave, i, start_at) {
4351                 if (IS_UP(slave->dev) &&
4352                     (slave->link == BOND_LINK_UP) &&
4353                     (slave->state == BOND_STATE_ACTIVE)) {
4354                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4355                         break;
4356                 }
4357         }
4358
4359 out:
4360         if (res) {
4361                 /* no suitable interface, frame not sent */
4362                 dev_kfree_skb(skb);
4363         }
4364         read_unlock(&bond->lock);
4365         return 0;
4366 }
4367
4368 /*
4369  * in broadcast mode, we send everything to all usable interfaces.
4370  */
4371 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4372 {
4373         struct bonding *bond = netdev_priv(bond_dev);
4374         struct slave *slave, *start_at;
4375         struct net_device *tx_dev = NULL;
4376         int i;
4377         int res = 1;
4378
4379         read_lock(&bond->lock);
4380
4381         if (!BOND_IS_OK(bond))
4382                 goto out;
4383
4384         read_lock(&bond->curr_slave_lock);
4385         start_at = bond->curr_active_slave;
4386         read_unlock(&bond->curr_slave_lock);
4387
4388         if (!start_at)
4389                 goto out;
4390
4391         bond_for_each_slave_from(bond, slave, i, start_at) {
4392                 if (IS_UP(slave->dev) &&
4393                     (slave->link == BOND_LINK_UP) &&
4394                     (slave->state == BOND_STATE_ACTIVE)) {
4395                         if (tx_dev) {
4396                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4397                                 if (!skb2) {
4398                                         pr_err(DRV_NAME
4399                                                ": %s: Error: bond_xmit_broadcast(): "
4400                                                "skb_clone() failed\n",
4401                                                bond_dev->name);
4402                                         continue;
4403                                 }
4404
4405                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4406                                 if (res) {
4407                                         dev_kfree_skb(skb2);
4408                                         continue;
4409                                 }
4410                         }
4411                         tx_dev = slave->dev;
4412                 }
4413         }
4414
4415         if (tx_dev)
4416                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4417
4418 out:
4419         if (res)
4420                 /* no suitable interface, frame not sent */
4421                 dev_kfree_skb(skb);
4422
4423         /* frame sent to all suitable interfaces */
4424         read_unlock(&bond->lock);
4425         return 0;
4426 }
4427
4428 /*------------------------- Device initialization ---------------------------*/
4429
4430 static void bond_set_xmit_hash_policy(struct bonding *bond)
4431 {
4432         switch (bond->params.xmit_policy) {
4433         case BOND_XMIT_POLICY_LAYER23:
4434                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4435                 break;
4436         case BOND_XMIT_POLICY_LAYER34:
4437                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4438                 break;
4439         case BOND_XMIT_POLICY_LAYER2:
4440         default:
4441                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4442                 break;
4443         }
4444 }
4445
4446 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4447 {
4448         const struct bonding *bond = netdev_priv(dev);
4449
4450         switch (bond->params.mode) {
4451         case BOND_MODE_ROUNDROBIN:
4452                 return bond_xmit_roundrobin(skb, dev);
4453         case BOND_MODE_ACTIVEBACKUP:
4454                 return bond_xmit_activebackup(skb, dev);
4455         case BOND_MODE_XOR:
4456                 return bond_xmit_xor(skb, dev);
4457         case BOND_MODE_BROADCAST:
4458                 return bond_xmit_broadcast(skb, dev);
4459         case BOND_MODE_8023AD:
4460                 return bond_3ad_xmit_xor(skb, dev);
4461         case BOND_MODE_ALB:
4462         case BOND_MODE_TLB:
4463                 return bond_alb_xmit(skb, dev);
4464         default:
4465                 /* Should never happen, mode already checked */
4466                 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4467                      dev->name, bond->params.mode);
4468                 WARN_ON_ONCE(1);
4469                 dev_kfree_skb(skb);
4470                 return NETDEV_TX_OK;
4471         }
4472 }
4473
4474
4475 /*
4476  * set bond mode specific net device operations
4477  */
4478 void bond_set_mode_ops(struct bonding *bond, int mode)
4479 {
4480         struct net_device *bond_dev = bond->dev;
4481
4482         switch (mode) {
4483         case BOND_MODE_ROUNDROBIN:
4484                 break;
4485         case BOND_MODE_ACTIVEBACKUP:
4486                 break;
4487         case BOND_MODE_XOR:
4488                 bond_set_xmit_hash_policy(bond);
4489                 break;
4490         case BOND_MODE_BROADCAST:
4491                 break;
4492         case BOND_MODE_8023AD:
4493                 bond_set_master_3ad_flags(bond);
4494                 bond_set_xmit_hash_policy(bond);
4495                 break;
4496         case BOND_MODE_ALB:
4497                 bond_set_master_alb_flags(bond);
4498                 /* FALLTHRU */
4499         case BOND_MODE_TLB:
4500                 break;
4501         default:
4502                 /* Should never happen, mode already checked */
4503                 pr_err(DRV_NAME
4504                        ": %s: Error: Unknown bonding mode %d\n",
4505                        bond_dev->name,
4506                        mode);
4507                 break;
4508         }
4509 }
4510
4511 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4512                                     struct ethtool_drvinfo *drvinfo)
4513 {
4514         strncpy(drvinfo->driver, DRV_NAME, 32);
4515         strncpy(drvinfo->version, DRV_VERSION, 32);
4516         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4517 }
4518
4519 static const struct ethtool_ops bond_ethtool_ops = {
4520         .get_drvinfo            = bond_ethtool_get_drvinfo,
4521         .get_link               = ethtool_op_get_link,
4522         .get_tx_csum            = ethtool_op_get_tx_csum,
4523         .get_sg                 = ethtool_op_get_sg,
4524         .get_tso                = ethtool_op_get_tso,
4525         .get_ufo                = ethtool_op_get_ufo,
4526         .get_flags              = ethtool_op_get_flags,
4527 };
4528
4529 static const struct net_device_ops bond_netdev_ops = {
4530         .ndo_init               = bond_init,
4531         .ndo_uninit             = bond_uninit,
4532         .ndo_open               = bond_open,
4533         .ndo_stop               = bond_close,
4534         .ndo_start_xmit         = bond_start_xmit,
4535         .ndo_get_stats          = bond_get_stats,
4536         .ndo_do_ioctl           = bond_do_ioctl,
4537         .ndo_set_multicast_list = bond_set_multicast_list,
4538         .ndo_change_mtu         = bond_change_mtu,
4539         .ndo_set_mac_address    = bond_set_mac_address,
4540         .ndo_neigh_setup        = bond_neigh_setup,
4541         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4542         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4543         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4544 };
4545
4546 static void bond_setup(struct net_device *bond_dev)
4547 {
4548         struct bonding *bond = netdev_priv(bond_dev);
4549
4550         /* initialize rwlocks */
4551         rwlock_init(&bond->lock);
4552         rwlock_init(&bond->curr_slave_lock);
4553
4554         bond->params = bonding_defaults;
4555
4556         /* Initialize pointers */
4557         bond->dev = bond_dev;
4558         INIT_LIST_HEAD(&bond->vlan_list);
4559
4560         /* Initialize the device entry points */
4561         ether_setup(bond_dev);
4562         bond_dev->netdev_ops = &bond_netdev_ops;
4563         bond_dev->ethtool_ops = &bond_ethtool_ops;
4564         bond_set_mode_ops(bond, bond->params.mode);
4565
4566         bond_dev->destructor = free_netdev;
4567
4568         /* Initialize the device options */
4569         bond_dev->tx_queue_len = 0;
4570         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4571         bond_dev->priv_flags |= IFF_BONDING;
4572         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4573
4574         if (bond->params.arp_interval)
4575                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4576
4577         /* At first, we block adding VLANs. That's the only way to
4578          * prevent problems that occur when adding VLANs over an
4579          * empty bond. The block will be removed once non-challenged
4580          * slaves are enslaved.
4581          */
4582         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4583
4584         /* don't acquire bond device's netif_tx_lock when
4585          * transmitting */
4586         bond_dev->features |= NETIF_F_LLTX;
4587
4588         /* By default, we declare the bond to be fully
4589          * VLAN hardware accelerated capable. Special
4590          * care is taken in the various xmit functions
4591          * when there are slaves that are not hw accel
4592          * capable
4593          */
4594         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4595                                NETIF_F_HW_VLAN_RX |
4596                                NETIF_F_HW_VLAN_FILTER);
4597
4598 }
4599
4600 static void bond_work_cancel_all(struct bonding *bond)
4601 {
4602         write_lock_bh(&bond->lock);
4603         bond->kill_timers = 1;
4604         write_unlock_bh(&bond->lock);
4605
4606         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4607                 cancel_delayed_work(&bond->mii_work);
4608
4609         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4610                 cancel_delayed_work(&bond->arp_work);
4611
4612         if (bond->params.mode == BOND_MODE_ALB &&
4613             delayed_work_pending(&bond->alb_work))
4614                 cancel_delayed_work(&bond->alb_work);
4615
4616         if (bond->params.mode == BOND_MODE_8023AD &&
4617             delayed_work_pending(&bond->ad_work))
4618                 cancel_delayed_work(&bond->ad_work);
4619 }
4620
4621 /* De-initialize device specific data.
4622  * Caller must hold rtnl_lock.
4623  */
4624 static void bond_deinit(struct net_device *bond_dev)
4625 {
4626         struct bonding *bond = netdev_priv(bond_dev);
4627
4628         list_del(&bond->bond_list);
4629
4630         bond_work_cancel_all(bond);
4631
4632         bond_remove_proc_entry(bond);
4633 }
4634
4635 /* Unregister and free all bond devices.
4636  * Caller must hold rtnl_lock.
4637  */
4638 static void bond_free_all(void)
4639 {
4640         struct bonding *bond, *nxt;
4641
4642         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4643                 struct net_device *bond_dev = bond->dev;
4644
4645                 bond_work_cancel_all(bond);
4646                 /* Release the bonded slaves */
4647                 bond_release_all(bond_dev);
4648                 unregister_netdevice(bond_dev);
4649         }
4650
4651         bond_destroy_proc_dir();
4652 }
4653
4654 /*------------------------- Module initialization ---------------------------*/
4655
4656 /*
4657  * Convert string input module parms.  Accept either the
4658  * number of the mode or its string name.  A bit complicated because
4659  * some mode names are substrings of other names, and calls from sysfs
4660  * may have whitespace in the name (trailing newlines, for example).
4661  */
4662 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4663 {
4664         int modeint = -1, i, rv;
4665         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4666
4667         for (p = (char *)buf; *p; p++)
4668                 if (!(isdigit(*p) || isspace(*p)))
4669                         break;
4670
4671         if (*p)
4672                 rv = sscanf(buf, "%20s", modestr);
4673         else
4674                 rv = sscanf(buf, "%d", &modeint);
4675
4676         if (!rv)
4677                 return -1;
4678
4679         for (i = 0; tbl[i].modename; i++) {
4680                 if (modeint == tbl[i].mode)
4681                         return tbl[i].mode;
4682                 if (strcmp(modestr, tbl[i].modename) == 0)
4683                         return tbl[i].mode;
4684         }
4685
4686         return -1;
4687 }
4688
4689 static int bond_check_params(struct bond_params *params)
4690 {
4691         int arp_validate_value, fail_over_mac_value;
4692
4693         /*
4694          * Convert string parameters.
4695          */
4696         if (mode) {
4697                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4698                 if (bond_mode == -1) {
4699                         pr_err(DRV_NAME
4700                                ": Error: Invalid bonding mode \"%s\"\n",
4701                                mode == NULL ? "NULL" : mode);
4702                         return -EINVAL;
4703                 }
4704         }
4705
4706         if (xmit_hash_policy) {
4707                 if ((bond_mode != BOND_MODE_XOR) &&
4708                     (bond_mode != BOND_MODE_8023AD)) {
4709                         pr_info(DRV_NAME
4710                                ": xor_mode param is irrelevant in mode %s\n",
4711                                bond_mode_name(bond_mode));
4712                 } else {
4713                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4714                                                         xmit_hashtype_tbl);
4715                         if (xmit_hashtype == -1) {
4716                                 pr_err(DRV_NAME
4717                                        ": Error: Invalid xmit_hash_policy \"%s\"\n",
4718                                        xmit_hash_policy == NULL ? "NULL" :
4719                                        xmit_hash_policy);
4720                                 return -EINVAL;
4721                         }
4722                 }
4723         }
4724
4725         if (lacp_rate) {
4726                 if (bond_mode != BOND_MODE_8023AD) {
4727                         pr_info(DRV_NAME
4728                                ": lacp_rate param is irrelevant in mode %s\n",
4729                                bond_mode_name(bond_mode));
4730                 } else {
4731                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4732                         if (lacp_fast == -1) {
4733                                 pr_err(DRV_NAME
4734                                        ": Error: Invalid lacp rate \"%s\"\n",
4735                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4736                                 return -EINVAL;
4737                         }
4738                 }
4739         }
4740
4741         if (ad_select) {
4742                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4743                 if (params->ad_select == -1) {
4744                         pr_err(DRV_NAME
4745                                ": Error: Invalid ad_select \"%s\"\n",
4746                                ad_select == NULL ? "NULL" : ad_select);
4747                         return -EINVAL;
4748                 }
4749
4750                 if (bond_mode != BOND_MODE_8023AD) {
4751                         pr_warning(DRV_NAME
4752                                ": ad_select param only affects 802.3ad mode\n");
4753                 }
4754         } else {
4755                 params->ad_select = BOND_AD_STABLE;
4756         }
4757
4758         if (max_bonds < 0 || max_bonds > INT_MAX) {
4759                 pr_warning(DRV_NAME
4760                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4761                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4762                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4763                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4764         }
4765
4766         if (miimon < 0) {
4767                 pr_warning(DRV_NAME
4768                        ": Warning: miimon module parameter (%d), "
4769                        "not in range 0-%d, so it was reset to %d\n",
4770                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4771                 miimon = BOND_LINK_MON_INTERV;
4772         }
4773
4774         if (updelay < 0) {
4775                 pr_warning(DRV_NAME
4776                        ": Warning: updelay module parameter (%d), "
4777                        "not in range 0-%d, so it was reset to 0\n",
4778                        updelay, INT_MAX);
4779                 updelay = 0;
4780         }
4781
4782         if (downdelay < 0) {
4783                 pr_warning(DRV_NAME
4784                        ": Warning: downdelay module parameter (%d), "
4785                        "not in range 0-%d, so it was reset to 0\n",
4786                        downdelay, INT_MAX);
4787                 downdelay = 0;
4788         }
4789
4790         if ((use_carrier != 0) && (use_carrier != 1)) {
4791                 pr_warning(DRV_NAME
4792                        ": Warning: use_carrier module parameter (%d), "
4793                        "not of valid value (0/1), so it was set to 1\n",
4794                        use_carrier);
4795                 use_carrier = 1;
4796         }
4797
4798         if (num_grat_arp < 0 || num_grat_arp > 255) {
4799                 pr_warning(DRV_NAME
4800                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4801                        "was reset to 1 \n", num_grat_arp);
4802                 num_grat_arp = 1;
4803         }
4804
4805         if (num_unsol_na < 0 || num_unsol_na > 255) {
4806                 pr_warning(DRV_NAME
4807                        ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4808                        "was reset to 1 \n", num_unsol_na);
4809                 num_unsol_na = 1;
4810         }
4811
4812         /* reset values for 802.3ad */
4813         if (bond_mode == BOND_MODE_8023AD) {
4814                 if (!miimon) {
4815                         pr_warning(DRV_NAME
4816                                ": Warning: miimon must be specified, "
4817                                "otherwise bonding will not detect link "
4818                                "failure, speed and duplex which are "
4819                                "essential for 802.3ad operation\n");
4820                         pr_warning("Forcing miimon to 100msec\n");
4821                         miimon = 100;
4822                 }
4823         }
4824
4825         /* reset values for TLB/ALB */
4826         if ((bond_mode == BOND_MODE_TLB) ||
4827             (bond_mode == BOND_MODE_ALB)) {
4828                 if (!miimon) {
4829                         pr_warning(DRV_NAME
4830                                ": Warning: miimon must be specified, "
4831                                "otherwise bonding will not detect link "
4832                                "failure and link speed which are essential "
4833                                "for TLB/ALB load balancing\n");
4834                         pr_warning("Forcing miimon to 100msec\n");
4835                         miimon = 100;
4836                 }
4837         }
4838
4839         if (bond_mode == BOND_MODE_ALB) {
4840                 printk(KERN_NOTICE DRV_NAME
4841                        ": In ALB mode you might experience client "
4842                        "disconnections upon reconnection of a link if the "
4843                        "bonding module updelay parameter (%d msec) is "
4844                        "incompatible with the forwarding delay time of the "
4845                        "switch\n",
4846                        updelay);
4847         }
4848
4849         if (!miimon) {
4850                 if (updelay || downdelay) {
4851                         /* just warn the user the up/down delay will have
4852                          * no effect since miimon is zero...
4853                          */
4854                         pr_warning(DRV_NAME
4855                                ": Warning: miimon module parameter not set "
4856                                "and updelay (%d) or downdelay (%d) module "
4857                                "parameter is set; updelay and downdelay have "
4858                                "no effect unless miimon is set\n",
4859                                updelay, downdelay);
4860                 }
4861         } else {
4862                 /* don't allow arp monitoring */
4863                 if (arp_interval) {
4864                         pr_warning(DRV_NAME
4865                                ": Warning: miimon (%d) and arp_interval (%d) "
4866                                "can't be used simultaneously, disabling ARP "
4867                                "monitoring\n",
4868                                miimon, arp_interval);
4869                         arp_interval = 0;
4870                 }
4871
4872                 if ((updelay % miimon) != 0) {
4873                         pr_warning(DRV_NAME
4874                                ": Warning: updelay (%d) is not a multiple "
4875                                "of miimon (%d), updelay rounded to %d ms\n",
4876                                updelay, miimon, (updelay / miimon) * miimon);
4877                 }
4878
4879                 updelay /= miimon;
4880
4881                 if ((downdelay % miimon) != 0) {
4882                         pr_warning(DRV_NAME
4883                                ": Warning: downdelay (%d) is not a multiple "
4884                                "of miimon (%d), downdelay rounded to %d ms\n",
4885                                downdelay, miimon,
4886                                (downdelay / miimon) * miimon);
4887                 }
4888
4889                 downdelay /= miimon;
4890         }
4891
4892         if (arp_interval < 0) {
4893                 pr_warning(DRV_NAME
4894                        ": Warning: arp_interval module parameter (%d) "
4895                        ", not in range 0-%d, so it was reset to %d\n",
4896                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4897                 arp_interval = BOND_LINK_ARP_INTERV;
4898         }
4899
4900         for (arp_ip_count = 0;
4901              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4902              arp_ip_count++) {
4903                 /* not complete check, but should be good enough to
4904                    catch mistakes */
4905                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4906                         pr_warning(DRV_NAME
4907                                ": Warning: bad arp_ip_target module parameter "
4908                                "(%s), ARP monitoring will not be performed\n",
4909                                arp_ip_target[arp_ip_count]);
4910                         arp_interval = 0;
4911                 } else {
4912                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4913                         arp_target[arp_ip_count] = ip;
4914                 }
4915         }
4916
4917         if (arp_interval && !arp_ip_count) {
4918                 /* don't allow arping if no arp_ip_target given... */
4919                 pr_warning(DRV_NAME
4920                        ": Warning: arp_interval module parameter (%d) "
4921                        "specified without providing an arp_ip_target "
4922                        "parameter, arp_interval was reset to 0\n",
4923                        arp_interval);
4924                 arp_interval = 0;
4925         }
4926
4927         if (arp_validate) {
4928                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4929                         pr_err(DRV_NAME
4930                                ": arp_validate only supported in active-backup mode\n");
4931                         return -EINVAL;
4932                 }
4933                 if (!arp_interval) {
4934                         pr_err(DRV_NAME
4935                                ": arp_validate requires arp_interval\n");
4936                         return -EINVAL;
4937                 }
4938
4939                 arp_validate_value = bond_parse_parm(arp_validate,
4940                                                      arp_validate_tbl);
4941                 if (arp_validate_value == -1) {
4942                         pr_err(DRV_NAME
4943                                ": Error: invalid arp_validate \"%s\"\n",
4944                                arp_validate == NULL ? "NULL" : arp_validate);
4945                         return -EINVAL;
4946                 }
4947         } else
4948                 arp_validate_value = 0;
4949
4950         if (miimon) {
4951                 pr_info(DRV_NAME
4952                        ": MII link monitoring set to %d ms\n",
4953                        miimon);
4954         } else if (arp_interval) {
4955                 int i;
4956
4957                 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4958                        " validate %s, with %d target(s):",
4959                        arp_interval,
4960                        arp_validate_tbl[arp_validate_value].modename,
4961                        arp_ip_count);
4962
4963                 for (i = 0; i < arp_ip_count; i++)
4964                         printk(" %s", arp_ip_target[i]);
4965
4966                 printk("\n");
4967
4968         } else if (max_bonds) {
4969                 /* miimon and arp_interval not set, we need one so things
4970                  * work as expected, see bonding.txt for details
4971                  */
4972                 pr_warning(DRV_NAME
4973                        ": Warning: either miimon or arp_interval and "
4974                        "arp_ip_target module parameters must be specified, "
4975                        "otherwise bonding will not detect link failures! see "
4976                        "bonding.txt for details.\n");
4977         }
4978
4979         if (primary && !USES_PRIMARY(bond_mode)) {
4980                 /* currently, using a primary only makes sense
4981                  * in active backup, TLB or ALB modes
4982                  */
4983                 pr_warning(DRV_NAME
4984                        ": Warning: %s primary device specified but has no "
4985                        "effect in %s mode\n",
4986                        primary, bond_mode_name(bond_mode));
4987                 primary = NULL;
4988         }
4989
4990         if (fail_over_mac) {
4991                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4992                                                       fail_over_mac_tbl);
4993                 if (fail_over_mac_value == -1) {
4994                         pr_err(DRV_NAME
4995                                ": Error: invalid fail_over_mac \"%s\"\n",
4996                                arp_validate == NULL ? "NULL" : arp_validate);
4997                         return -EINVAL;
4998                 }
4999
5000                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5001                         pr_warning(DRV_NAME
5002                                ": Warning: fail_over_mac only affects "
5003                                "active-backup mode.\n");
5004         } else {
5005                 fail_over_mac_value = BOND_FOM_NONE;
5006         }
5007
5008         /* fill params struct with the proper values */
5009         params->mode = bond_mode;
5010         params->xmit_policy = xmit_hashtype;
5011         params->miimon = miimon;
5012         params->num_grat_arp = num_grat_arp;
5013         params->num_unsol_na = num_unsol_na;
5014         params->arp_interval = arp_interval;
5015         params->arp_validate = arp_validate_value;
5016         params->updelay = updelay;
5017         params->downdelay = downdelay;
5018         params->use_carrier = use_carrier;
5019         params->lacp_fast = lacp_fast;
5020         params->primary[0] = 0;
5021         params->fail_over_mac = fail_over_mac_value;
5022
5023         if (primary) {
5024                 strncpy(params->primary, primary, IFNAMSIZ);
5025                 params->primary[IFNAMSIZ - 1] = 0;
5026         }
5027
5028         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5029
5030         return 0;
5031 }
5032
5033 static struct lock_class_key bonding_netdev_xmit_lock_key;
5034 static struct lock_class_key bonding_netdev_addr_lock_key;
5035
5036 static void bond_set_lockdep_class_one(struct net_device *dev,
5037                                        struct netdev_queue *txq,
5038                                        void *_unused)
5039 {
5040         lockdep_set_class(&txq->_xmit_lock,
5041                           &bonding_netdev_xmit_lock_key);
5042 }
5043
5044 static void bond_set_lockdep_class(struct net_device *dev)
5045 {
5046         lockdep_set_class(&dev->addr_list_lock,
5047                           &bonding_netdev_addr_lock_key);
5048         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5049 }
5050
5051 /*
5052  * Called from registration process
5053  */
5054 static int bond_init(struct net_device *bond_dev)
5055 {
5056         struct bonding *bond = netdev_priv(bond_dev);
5057
5058         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5059
5060         bond->wq = create_singlethread_workqueue(bond_dev->name);
5061         if (!bond->wq)
5062                 return -ENOMEM;
5063
5064         bond_set_lockdep_class(bond_dev);
5065
5066         netif_carrier_off(bond_dev);
5067
5068         bond_create_proc_entry(bond);
5069         list_add_tail(&bond->bond_list, &bond_dev_list);
5070
5071         return 0;
5072 }
5073
5074 /* Create a new bond based on the specified name and bonding parameters.
5075  * If name is NULL, obtain a suitable "bond%d" name for us.
5076  * Caller must NOT hold rtnl_lock; we need to release it here before we
5077  * set up our sysfs entries.
5078  */
5079 int bond_create(const char *name)
5080 {
5081         struct net_device *bond_dev;
5082         int res;
5083
5084         rtnl_lock();
5085         /* Check to see if the bond already exists. */
5086         /* FIXME: pass netns from caller */
5087         if (name && __dev_get_by_name(&init_net, name)) {
5088                 pr_err(DRV_NAME ": cannot add bond %s; already exists\n",
5089                        name);
5090                 res = -EEXIST;
5091                 goto out_rtnl;
5092         }
5093
5094         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5095                                 bond_setup);
5096         if (!bond_dev) {
5097                 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5098                        name);
5099                 res = -ENOMEM;
5100                 goto out_rtnl;
5101         }
5102
5103         if (!name) {
5104                 res = dev_alloc_name(bond_dev, "bond%d");
5105                 if (res < 0)
5106                         goto out_netdev;
5107         }
5108
5109         res = register_netdevice(bond_dev);
5110         if (res < 0)
5111                 goto out_bond;
5112
5113         res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5114         if (res < 0)
5115                 goto out_unreg;
5116
5117         rtnl_unlock();
5118         return 0;
5119
5120 out_unreg:
5121         unregister_netdevice(bond_dev);
5122 out_bond:
5123         bond_deinit(bond_dev);
5124 out_netdev:
5125         free_netdev(bond_dev);
5126 out_rtnl:
5127         rtnl_unlock();
5128         return res;
5129 }
5130
5131 static int __init bonding_init(void)
5132 {
5133         int i;
5134         int res;
5135
5136         pr_info("%s", version);
5137
5138         res = bond_check_params(&bonding_defaults);
5139         if (res)
5140                 goto out;
5141
5142         bond_create_proc_dir();
5143
5144         for (i = 0; i < max_bonds; i++) {
5145                 res = bond_create(NULL);
5146                 if (res)
5147                         goto err;
5148         }
5149
5150         res = bond_create_sysfs();
5151         if (res)
5152                 goto err;
5153
5154         register_netdevice_notifier(&bond_netdev_notifier);
5155         register_inetaddr_notifier(&bond_inetaddr_notifier);
5156         bond_register_ipv6_notifier();
5157
5158         goto out;
5159 err:
5160         rtnl_lock();
5161         bond_free_all();
5162         rtnl_unlock();
5163 out:
5164         return res;
5165
5166 }
5167
5168 static void __exit bonding_exit(void)
5169 {
5170         unregister_netdevice_notifier(&bond_netdev_notifier);
5171         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5172         bond_unregister_ipv6_notifier();
5173
5174         bond_destroy_sysfs();
5175
5176         rtnl_lock();
5177         bond_free_all();
5178         rtnl_unlock();
5179 }
5180
5181 module_init(bonding_init);
5182 module_exit(bonding_exit);
5183 MODULE_LICENSE("GPL");
5184 MODULE_VERSION(DRV_VERSION);
5185 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5186 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");