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