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 long vlan_features = 0;
1335         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1336                                                 bond_dev->hard_header_len);
1337         int i;
1338
1339         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1340         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1341
1342         if (!bond->first_slave)
1343                 goto done;
1344
1345         features &= ~NETIF_F_ONE_FOR_ALL;
1346
1347         vlan_features = bond->first_slave->dev->vlan_features;
1348         bond_for_each_slave(bond, slave, i) {
1349                 features = netdev_increment_features(features,
1350                                                      slave->dev->features,
1351                                                      NETIF_F_ONE_FOR_ALL);
1352                 vlan_features = netdev_increment_features(vlan_features,
1353                                                         slave->dev->vlan_features,
1354                                                         NETIF_F_ONE_FOR_ALL);
1355                 if (slave->dev->hard_header_len > max_hard_header_len)
1356                         max_hard_header_len = slave->dev->hard_header_len;
1357         }
1358
1359 done:
1360         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1361         bond_dev->features = netdev_fix_features(features, NULL);
1362         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1363         bond_dev->hard_header_len = max_hard_header_len;
1364
1365         return 0;
1366 }
1367
1368 static void bond_setup_by_slave(struct net_device *bond_dev,
1369                                 struct net_device *slave_dev)
1370 {
1371         struct bonding *bond = netdev_priv(bond_dev);
1372
1373         bond_dev->header_ops        = slave_dev->header_ops;
1374
1375         bond_dev->type              = slave_dev->type;
1376         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1377         bond_dev->addr_len          = slave_dev->addr_len;
1378
1379         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1380                 slave_dev->addr_len);
1381         bond->setup_by_slave = 1;
1382 }
1383
1384 /* enslave device <slave> to bond device <master> */
1385 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1386 {
1387         struct bonding *bond = netdev_priv(bond_dev);
1388         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1389         struct slave *new_slave = NULL;
1390         struct dev_mc_list *dmi;
1391         struct sockaddr addr;
1392         int link_reporting;
1393         int old_features = bond_dev->features;
1394         int res = 0;
1395
1396         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397                 slave_ops->ndo_do_ioctl == NULL) {
1398                 pr_warning(DRV_NAME
1399                        ": %s: Warning: no link monitoring support for %s\n",
1400                        bond_dev->name, slave_dev->name);
1401         }
1402
1403         /* bond must be initialized by bond_open() before enslaving */
1404         if (!(bond_dev->flags & IFF_UP)) {
1405                 pr_warning(DRV_NAME
1406                         " %s: master_dev is not up in bond_enslave\n",
1407                         bond_dev->name);
1408         }
1409
1410         /* already enslaved */
1411         if (slave_dev->flags & IFF_SLAVE) {
1412                 pr_debug("Error, Device was already enslaved\n");
1413                 return -EBUSY;
1414         }
1415
1416         /* vlan challenged mutual exclusion */
1417         /* no need to lock since we're protected by rtnl_lock */
1418         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420                 if (!list_empty(&bond->vlan_list)) {
1421                         pr_err(DRV_NAME
1422                                ": %s: Error: cannot enslave VLAN "
1423                                "challenged slave %s on VLAN enabled "
1424                                "bond %s\n", bond_dev->name, slave_dev->name,
1425                                bond_dev->name);
1426                         return -EPERM;
1427                 } else {
1428                         pr_warning(DRV_NAME
1429                                ": %s: Warning: enslaved VLAN challenged "
1430                                "slave %s. Adding VLANs will be blocked as "
1431                                "long as %s is part of bond %s\n",
1432                                bond_dev->name, slave_dev->name, slave_dev->name,
1433                                bond_dev->name);
1434                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1435                 }
1436         } else {
1437                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438                 if (bond->slave_cnt == 0) {
1439                         /* First slave, and it is not VLAN challenged,
1440                          * so remove the block of adding VLANs over the bond.
1441                          */
1442                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1443                 }
1444         }
1445
1446         /*
1447          * Old ifenslave binaries are no longer supported.  These can
1448          * be identified with moderate accuracy by the state of the slave:
1449          * the current ifenslave will set the interface down prior to
1450          * enslaving it; the old ifenslave will not.
1451          */
1452         if ((slave_dev->flags & IFF_UP)) {
1453                 pr_err(DRV_NAME ": %s is up. "
1454                        "This may be due to an out of date ifenslave.\n",
1455                        slave_dev->name);
1456                 res = -EPERM;
1457                 goto err_undo_flags;
1458         }
1459
1460         /* set bonding device ether type by slave - bonding netdevices are
1461          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462          * there is a need to override some of the type dependent attribs/funcs.
1463          *
1464          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1466          */
1467         if (bond->slave_cnt == 0) {
1468                 if (bond_dev->type != slave_dev->type) {
1469                         dev_close(bond_dev);
1470                         pr_debug("%s: change device type from %d to %d\n",
1471                                 bond_dev->name, bond_dev->type, slave_dev->type);
1472                         if (slave_dev->type != ARPHRD_ETHER)
1473                                 bond_setup_by_slave(bond_dev, slave_dev);
1474                         else
1475                                 ether_setup(bond_dev);
1476                         dev_open(bond_dev);
1477                 }
1478         } else if (bond_dev->type != slave_dev->type) {
1479                 pr_err(DRV_NAME ": %s ether type (%d) is different "
1480                         "from other slaves (%d), can not enslave it.\n",
1481                         slave_dev->name,
1482                         slave_dev->type, bond_dev->type);
1483                         res = -EINVAL;
1484                         goto err_undo_flags;
1485         }
1486
1487         if (slave_ops->ndo_set_mac_address == NULL) {
1488                 if (bond->slave_cnt == 0) {
1489                         pr_warning(DRV_NAME
1490                                ": %s: Warning: The first slave device "
1491                                "specified does not support setting the MAC "
1492                                "address. Setting fail_over_mac to active.",
1493                                bond_dev->name);
1494                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1495                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1496                         pr_err(DRV_NAME
1497                                 ": %s: Error: The slave device specified "
1498                                 "does not support setting the MAC address, "
1499                                 "but fail_over_mac is not set to active.\n"
1500                                 , bond_dev->name);
1501                         res = -EOPNOTSUPP;
1502                         goto err_undo_flags;
1503                 }
1504         }
1505
1506         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1507         if (!new_slave) {
1508                 res = -ENOMEM;
1509                 goto err_undo_flags;
1510         }
1511
1512         /* save slave's original flags before calling
1513          * netdev_set_master and dev_open
1514          */
1515         new_slave->original_flags = slave_dev->flags;
1516
1517         /*
1518          * Save slave's original ("permanent") mac address for modes
1519          * that need it, and for restoring it upon release, and then
1520          * set it to the master's address
1521          */
1522         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1523
1524         if (!bond->params.fail_over_mac) {
1525                 /*
1526                  * Set slave to master's mac address.  The application already
1527                  * set the master's mac address to that of the first slave
1528                  */
1529                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1530                 addr.sa_family = slave_dev->type;
1531                 res = dev_set_mac_address(slave_dev, &addr);
1532                 if (res) {
1533                         pr_debug("Error %d calling set_mac_address\n", res);
1534                         goto err_free;
1535                 }
1536         }
1537
1538         res = netdev_set_master(slave_dev, bond_dev);
1539         if (res) {
1540                 pr_debug("Error %d calling netdev_set_master\n", res);
1541                 goto err_restore_mac;
1542         }
1543         /* open the slave since the application closed it */
1544         res = dev_open(slave_dev);
1545         if (res) {
1546                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1547                 goto err_unset_master;
1548         }
1549
1550         new_slave->dev = slave_dev;
1551         slave_dev->priv_flags |= IFF_BONDING;
1552
1553         if (bond_is_lb(bond)) {
1554                 /* bond_alb_init_slave() must be called before all other stages since
1555                  * it might fail and we do not want to have to undo everything
1556                  */
1557                 res = bond_alb_init_slave(bond, new_slave);
1558                 if (res)
1559                         goto err_close;
1560         }
1561
1562         /* If the mode USES_PRIMARY, then the new slave gets the
1563          * master's promisc (and mc) settings only if it becomes the
1564          * curr_active_slave, and that is taken care of later when calling
1565          * bond_change_active()
1566          */
1567         if (!USES_PRIMARY(bond->params.mode)) {
1568                 /* set promiscuity level to new slave */
1569                 if (bond_dev->flags & IFF_PROMISC) {
1570                         res = dev_set_promiscuity(slave_dev, 1);
1571                         if (res)
1572                                 goto err_close;
1573                 }
1574
1575                 /* set allmulti level to new slave */
1576                 if (bond_dev->flags & IFF_ALLMULTI) {
1577                         res = dev_set_allmulti(slave_dev, 1);
1578                         if (res)
1579                                 goto err_close;
1580                 }
1581
1582                 netif_addr_lock_bh(bond_dev);
1583                 /* upload master's mc_list to new slave */
1584                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1585                         dev_mc_add(slave_dev, dmi->dmi_addr,
1586                                    dmi->dmi_addrlen, 0);
1587                 netif_addr_unlock_bh(bond_dev);
1588         }
1589
1590         if (bond->params.mode == BOND_MODE_8023AD) {
1591                 /* add lacpdu mc addr to mc list */
1592                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1593
1594                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1595         }
1596
1597         bond_add_vlans_on_slave(bond, slave_dev);
1598
1599         write_lock_bh(&bond->lock);
1600
1601         bond_attach_slave(bond, new_slave);
1602
1603         new_slave->delay = 0;
1604         new_slave->link_failure_count = 0;
1605
1606         bond_compute_features(bond);
1607
1608         write_unlock_bh(&bond->lock);
1609
1610         read_lock(&bond->lock);
1611
1612         new_slave->last_arp_rx = jiffies;
1613
1614         if (bond->params.miimon && !bond->params.use_carrier) {
1615                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1616
1617                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1618                         /*
1619                          * miimon is set but a bonded network driver
1620                          * does not support ETHTOOL/MII and
1621                          * arp_interval is not set.  Note: if
1622                          * use_carrier is enabled, we will never go
1623                          * here (because netif_carrier is always
1624                          * supported); thus, we don't need to change
1625                          * the messages for netif_carrier.
1626                          */
1627                         pr_warning(DRV_NAME
1628                                ": %s: Warning: MII and ETHTOOL support not "
1629                                "available for interface %s, and "
1630                                "arp_interval/arp_ip_target module parameters "
1631                                "not specified, thus bonding will not detect "
1632                                "link failures! see bonding.txt for details.\n",
1633                                bond_dev->name, slave_dev->name);
1634                 } else if (link_reporting == -1) {
1635                         /* unable get link status using mii/ethtool */
1636                         pr_warning(DRV_NAME
1637                                ": %s: Warning: can't get link status from "
1638                                "interface %s; the network driver associated "
1639                                "with this interface does not support MII or "
1640                                "ETHTOOL link status reporting, thus miimon "
1641                                "has no effect on this interface.\n",
1642                                bond_dev->name, slave_dev->name);
1643                 }
1644         }
1645
1646         /* check for initial state */
1647         if (!bond->params.miimon ||
1648             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1649                 if (bond->params.updelay) {
1650                         pr_debug("Initial state of slave_dev is "
1651                                 "BOND_LINK_BACK\n");
1652                         new_slave->link  = BOND_LINK_BACK;
1653                         new_slave->delay = bond->params.updelay;
1654                 } else {
1655                         pr_debug("Initial state of slave_dev is "
1656                                 "BOND_LINK_UP\n");
1657                         new_slave->link  = BOND_LINK_UP;
1658                 }
1659                 new_slave->jiffies = jiffies;
1660         } else {
1661                 pr_debug("Initial state of slave_dev is "
1662                         "BOND_LINK_DOWN\n");
1663                 new_slave->link  = BOND_LINK_DOWN;
1664         }
1665
1666         if (bond_update_speed_duplex(new_slave) &&
1667             (new_slave->link != BOND_LINK_DOWN)) {
1668                 pr_warning(DRV_NAME
1669                        ": %s: Warning: failed to get speed and duplex from %s, "
1670                        "assumed to be 100Mb/sec and Full.\n",
1671                        bond_dev->name, new_slave->dev->name);
1672
1673                 if (bond->params.mode == BOND_MODE_8023AD) {
1674                         pr_warning(DRV_NAME
1675                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1676                                "support in base driver for proper aggregator "
1677                                "selection.\n", bond_dev->name);
1678                 }
1679         }
1680
1681         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1682                 /* if there is a primary slave, remember it */
1683                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1684                         bond->primary_slave = new_slave;
1685         }
1686
1687         write_lock_bh(&bond->curr_slave_lock);
1688
1689         switch (bond->params.mode) {
1690         case BOND_MODE_ACTIVEBACKUP:
1691                 bond_set_slave_inactive_flags(new_slave);
1692                 bond_select_active_slave(bond);
1693                 break;
1694         case BOND_MODE_8023AD:
1695                 /* in 802.3ad mode, the internal mechanism
1696                  * will activate the slaves in the selected
1697                  * aggregator
1698                  */
1699                 bond_set_slave_inactive_flags(new_slave);
1700                 /* if this is the first slave */
1701                 if (bond->slave_cnt == 1) {
1702                         SLAVE_AD_INFO(new_slave).id = 1;
1703                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1704                          * can be called only after the mac address of the bond is set
1705                          */
1706                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1707                                             bond->params.lacp_fast);
1708                 } else {
1709                         SLAVE_AD_INFO(new_slave).id =
1710                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1711                 }
1712
1713                 bond_3ad_bind_slave(new_slave);
1714                 break;
1715         case BOND_MODE_TLB:
1716         case BOND_MODE_ALB:
1717                 new_slave->state = BOND_STATE_ACTIVE;
1718                 bond_set_slave_inactive_flags(new_slave);
1719                 bond_select_active_slave(bond);
1720                 break;
1721         default:
1722                 pr_debug("This slave is always active in trunk mode\n");
1723
1724                 /* always active in trunk mode */
1725                 new_slave->state = BOND_STATE_ACTIVE;
1726
1727                 /* In trunking mode there is little meaning to curr_active_slave
1728                  * anyway (it holds no special properties of the bond device),
1729                  * so we can change it without calling change_active_interface()
1730                  */
1731                 if (!bond->curr_active_slave)
1732                         bond->curr_active_slave = new_slave;
1733
1734                 break;
1735         } /* switch(bond_mode) */
1736
1737         write_unlock_bh(&bond->curr_slave_lock);
1738
1739         bond_set_carrier(bond);
1740
1741         read_unlock(&bond->lock);
1742
1743         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1744         if (res)
1745                 goto err_close;
1746
1747         pr_info(DRV_NAME
1748                ": %s: enslaving %s as a%s interface with a%s link.\n",
1749                bond_dev->name, slave_dev->name,
1750                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1751                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1752
1753         /* enslave is successful */
1754         return 0;
1755
1756 /* Undo stages on error */
1757 err_close:
1758         dev_close(slave_dev);
1759
1760 err_unset_master:
1761         netdev_set_master(slave_dev, NULL);
1762
1763 err_restore_mac:
1764         if (!bond->params.fail_over_mac) {
1765                 /* XXX TODO - fom follow mode needs to change master's
1766                  * MAC if this slave's MAC is in use by the bond, or at
1767                  * least print a warning.
1768                  */
1769                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1770                 addr.sa_family = slave_dev->type;
1771                 dev_set_mac_address(slave_dev, &addr);
1772         }
1773
1774 err_free:
1775         kfree(new_slave);
1776
1777 err_undo_flags:
1778         bond_dev->features = old_features;
1779
1780         return res;
1781 }
1782
1783 /*
1784  * Try to release the slave device <slave> from the bond device <master>
1785  * It is legal to access curr_active_slave without a lock because all the function
1786  * is write-locked.
1787  *
1788  * The rules for slave state should be:
1789  *   for Active/Backup:
1790  *     Active stays on all backups go down
1791  *   for Bonded connections:
1792  *     The first up interface should be left on and all others downed.
1793  */
1794 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1795 {
1796         struct bonding *bond = netdev_priv(bond_dev);
1797         struct slave *slave, *oldcurrent;
1798         struct sockaddr addr;
1799
1800         /* slave is not a slave or master is not master of this slave */
1801         if (!(slave_dev->flags & IFF_SLAVE) ||
1802             (slave_dev->master != bond_dev)) {
1803                 pr_err(DRV_NAME
1804                        ": %s: Error: cannot release %s.\n",
1805                        bond_dev->name, slave_dev->name);
1806                 return -EINVAL;
1807         }
1808
1809         write_lock_bh(&bond->lock);
1810
1811         slave = bond_get_slave_by_dev(bond, slave_dev);
1812         if (!slave) {
1813                 /* not a slave of this bond */
1814                 pr_info(DRV_NAME
1815                        ": %s: %s not enslaved\n",
1816                        bond_dev->name, slave_dev->name);
1817                 write_unlock_bh(&bond->lock);
1818                 return -EINVAL;
1819         }
1820
1821         if (!bond->params.fail_over_mac) {
1822                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr)
1823                     && bond->slave_cnt > 1)
1824                         pr_warning(DRV_NAME
1825                                ": %s: Warning: the permanent HWaddr of %s - "
1826                                "%pM - is still in use by %s. "
1827                                "Set the HWaddr of %s to a different address "
1828                                "to avoid conflicts.\n",
1829                                bond_dev->name, slave_dev->name,
1830                                slave->perm_hwaddr,
1831                                bond_dev->name, slave_dev->name);
1832         }
1833
1834         /* Inform AD package of unbinding of slave. */
1835         if (bond->params.mode == BOND_MODE_8023AD) {
1836                 /* must be called before the slave is
1837                  * detached from the list
1838                  */
1839                 bond_3ad_unbind_slave(slave);
1840         }
1841
1842         pr_info(DRV_NAME
1843                ": %s: releasing %s interface %s\n",
1844                bond_dev->name,
1845                (slave->state == BOND_STATE_ACTIVE)
1846                ? "active" : "backup",
1847                slave_dev->name);
1848
1849         oldcurrent = bond->curr_active_slave;
1850
1851         bond->current_arp_slave = NULL;
1852
1853         /* release the slave from its bond */
1854         bond_detach_slave(bond, slave);
1855
1856         bond_compute_features(bond);
1857
1858         if (bond->primary_slave == slave)
1859                 bond->primary_slave = NULL;
1860
1861         if (oldcurrent == slave)
1862                 bond_change_active_slave(bond, NULL);
1863
1864         if (bond_is_lb(bond)) {
1865                 /* Must be called only after the slave has been
1866                  * detached from the list and the curr_active_slave
1867                  * has been cleared (if our_slave == old_current),
1868                  * but before a new active slave is selected.
1869                  */
1870                 write_unlock_bh(&bond->lock);
1871                 bond_alb_deinit_slave(bond, slave);
1872                 write_lock_bh(&bond->lock);
1873         }
1874
1875         if (oldcurrent == slave) {
1876                 /*
1877                  * Note that we hold RTNL over this sequence, so there
1878                  * is no concern that another slave add/remove event
1879                  * will interfere.
1880                  */
1881                 write_unlock_bh(&bond->lock);
1882                 read_lock(&bond->lock);
1883                 write_lock_bh(&bond->curr_slave_lock);
1884
1885                 bond_select_active_slave(bond);
1886
1887                 write_unlock_bh(&bond->curr_slave_lock);
1888                 read_unlock(&bond->lock);
1889                 write_lock_bh(&bond->lock);
1890         }
1891
1892         if (bond->slave_cnt == 0) {
1893                 bond_set_carrier(bond);
1894
1895                 /* if the last slave was removed, zero the mac address
1896                  * of the master so it will be set by the application
1897                  * to the mac address of the first slave
1898                  */
1899                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901                 if (list_empty(&bond->vlan_list)) {
1902                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903                 } else {
1904                         pr_warning(DRV_NAME
1905                                ": %s: Warning: clearing HW address of %s while it "
1906                                "still has VLANs.\n",
1907                                bond_dev->name, bond_dev->name);
1908                         pr_warning(DRV_NAME
1909                                ": %s: When re-adding slaves, make sure the bond's "
1910                                "HW address matches its VLANs'.\n",
1911                                bond_dev->name);
1912                 }
1913         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914                    !bond_has_challenged_slaves(bond)) {
1915                 pr_info(DRV_NAME
1916                        ": %s: last VLAN challenged slave %s "
1917                        "left bond %s. VLAN blocking is removed\n",
1918                        bond_dev->name, slave_dev->name, bond_dev->name);
1919                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920         }
1921
1922         write_unlock_bh(&bond->lock);
1923
1924         /* must do this from outside any spinlocks */
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927         bond_del_vlans_from_slave(bond, slave_dev);
1928
1929         /* If the mode USES_PRIMARY, then we should only remove its
1930          * promisc and mc settings if it was the curr_active_slave, but that was
1931          * already taken care of above when we detached the slave
1932          */
1933         if (!USES_PRIMARY(bond->params.mode)) {
1934                 /* unset promiscuity level from slave */
1935                 if (bond_dev->flags & IFF_PROMISC)
1936                         dev_set_promiscuity(slave_dev, -1);
1937
1938                 /* unset allmulti level from slave */
1939                 if (bond_dev->flags & IFF_ALLMULTI)
1940                         dev_set_allmulti(slave_dev, -1);
1941
1942                 /* flush master's mc_list from slave */
1943                 netif_addr_lock_bh(bond_dev);
1944                 bond_mc_list_flush(bond_dev, slave_dev);
1945                 netif_addr_unlock_bh(bond_dev);
1946         }
1947
1948         netdev_set_master(slave_dev, NULL);
1949
1950         /* close slave before restoring its mac address */
1951         dev_close(slave_dev);
1952
1953         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1954                 /* restore original ("permanent") mac address */
1955                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1956                 addr.sa_family = slave_dev->type;
1957                 dev_set_mac_address(slave_dev, &addr);
1958         }
1959
1960         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1961                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1962                                    IFF_SLAVE_NEEDARP);
1963
1964         kfree(slave);
1965
1966         return 0;  /* deletion OK */
1967 }
1968
1969 /*
1970 * Destroy a bonding device.
1971 * Must be under rtnl_lock when this function is called.
1972 */
1973 static void bond_uninit(struct net_device *bond_dev)
1974 {
1975         struct bonding *bond = netdev_priv(bond_dev);
1976
1977         bond_deinit(bond_dev);
1978         bond_destroy_sysfs_entry(bond);
1979
1980         if (bond->wq)
1981                 destroy_workqueue(bond->wq);
1982
1983         netif_addr_lock_bh(bond_dev);
1984         bond_mc_list_destroy(bond);
1985         netif_addr_unlock_bh(bond_dev);
1986 }
1987
1988 /*
1989 * First release a slave and than destroy the bond if no more slaves are left.
1990 * Must be under rtnl_lock when this function is called.
1991 */
1992 int  bond_release_and_destroy(struct net_device *bond_dev,
1993                               struct net_device *slave_dev)
1994 {
1995         struct bonding *bond = netdev_priv(bond_dev);
1996         int ret;
1997
1998         ret = bond_release(bond_dev, slave_dev);
1999         if ((ret == 0) && (bond->slave_cnt == 0)) {
2000                 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
2001                        bond_dev->name, bond_dev->name);
2002                 unregister_netdevice(bond_dev);
2003         }
2004         return ret;
2005 }
2006
2007 /*
2008  * This function releases all slaves.
2009  */
2010 static int bond_release_all(struct net_device *bond_dev)
2011 {
2012         struct bonding *bond = netdev_priv(bond_dev);
2013         struct slave *slave;
2014         struct net_device *slave_dev;
2015         struct sockaddr addr;
2016
2017         write_lock_bh(&bond->lock);
2018
2019         netif_carrier_off(bond_dev);
2020
2021         if (bond->slave_cnt == 0)
2022                 goto out;
2023
2024         bond->current_arp_slave = NULL;
2025         bond->primary_slave = NULL;
2026         bond_change_active_slave(bond, NULL);
2027
2028         while ((slave = bond->first_slave) != NULL) {
2029                 /* Inform AD package of unbinding of slave
2030                  * before slave is detached from the list.
2031                  */
2032                 if (bond->params.mode == BOND_MODE_8023AD)
2033                         bond_3ad_unbind_slave(slave);
2034
2035                 slave_dev = slave->dev;
2036                 bond_detach_slave(bond, slave);
2037
2038                 /* now that the slave is detached, unlock and perform
2039                  * all the undo steps that should not be called from
2040                  * within a lock.
2041                  */
2042                 write_unlock_bh(&bond->lock);
2043
2044                 if (bond_is_lb(bond)) {
2045                         /* must be called only after the slave
2046                          * has been detached from the list
2047                          */
2048                         bond_alb_deinit_slave(bond, slave);
2049                 }
2050
2051                 bond_compute_features(bond);
2052
2053                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2054                 bond_del_vlans_from_slave(bond, slave_dev);
2055
2056                 /* If the mode USES_PRIMARY, then we should only remove its
2057                  * promisc and mc settings if it was the curr_active_slave, but that was
2058                  * already taken care of above when we detached the slave
2059                  */
2060                 if (!USES_PRIMARY(bond->params.mode)) {
2061                         /* unset promiscuity level from slave */
2062                         if (bond_dev->flags & IFF_PROMISC)
2063                                 dev_set_promiscuity(slave_dev, -1);
2064
2065                         /* unset allmulti level from slave */
2066                         if (bond_dev->flags & IFF_ALLMULTI)
2067                                 dev_set_allmulti(slave_dev, -1);
2068
2069                         /* flush master's mc_list from slave */
2070                         netif_addr_lock_bh(bond_dev);
2071                         bond_mc_list_flush(bond_dev, slave_dev);
2072                         netif_addr_unlock_bh(bond_dev);
2073                 }
2074
2075                 netdev_set_master(slave_dev, NULL);
2076
2077                 /* close slave before restoring its mac address */
2078                 dev_close(slave_dev);
2079
2080                 if (!bond->params.fail_over_mac) {
2081                         /* restore original ("permanent") mac address*/
2082                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2083                         addr.sa_family = slave_dev->type;
2084                         dev_set_mac_address(slave_dev, &addr);
2085                 }
2086
2087                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2088                                            IFF_SLAVE_INACTIVE);
2089
2090                 kfree(slave);
2091
2092                 /* re-acquire the lock before getting the next slave */
2093                 write_lock_bh(&bond->lock);
2094         }
2095
2096         /* zero the mac address of the master so it will be
2097          * set by the application to the mac address of the
2098          * first slave
2099          */
2100         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2101
2102         if (list_empty(&bond->vlan_list))
2103                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2104         else {
2105                 pr_warning(DRV_NAME
2106                        ": %s: Warning: clearing HW address of %s while it "
2107                        "still has VLANs.\n",
2108                        bond_dev->name, bond_dev->name);
2109                 pr_warning(DRV_NAME
2110                        ": %s: When re-adding slaves, make sure the bond's "
2111                        "HW address matches its VLANs'.\n",
2112                        bond_dev->name);
2113         }
2114
2115         pr_info(DRV_NAME
2116                ": %s: released all slaves\n",
2117                bond_dev->name);
2118
2119 out:
2120         write_unlock_bh(&bond->lock);
2121
2122         return 0;
2123 }
2124
2125 /*
2126  * This function changes the active slave to slave <slave_dev>.
2127  * It returns -EINVAL in the following cases.
2128  *  - <slave_dev> is not found in the list.
2129  *  - There is not active slave now.
2130  *  - <slave_dev> is already active.
2131  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2132  *  - <slave_dev> is not running.
2133  * In these cases, this function does nothing.
2134  * In the other cases, current_slave pointer is changed and 0 is returned.
2135  */
2136 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2137 {
2138         struct bonding *bond = netdev_priv(bond_dev);
2139         struct slave *old_active = NULL;
2140         struct slave *new_active = NULL;
2141         int res = 0;
2142
2143         if (!USES_PRIMARY(bond->params.mode))
2144                 return -EINVAL;
2145
2146         /* Verify that master_dev is indeed the master of slave_dev */
2147         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2148                 return -EINVAL;
2149
2150         read_lock(&bond->lock);
2151
2152         read_lock(&bond->curr_slave_lock);
2153         old_active = bond->curr_active_slave;
2154         read_unlock(&bond->curr_slave_lock);
2155
2156         new_active = bond_get_slave_by_dev(bond, slave_dev);
2157
2158         /*
2159          * Changing to the current active: do nothing; return success.
2160          */
2161         if (new_active && (new_active == old_active)) {
2162                 read_unlock(&bond->lock);
2163                 return 0;
2164         }
2165
2166         if ((new_active) &&
2167             (old_active) &&
2168             (new_active->link == BOND_LINK_UP) &&
2169             IS_UP(new_active->dev)) {
2170                 write_lock_bh(&bond->curr_slave_lock);
2171                 bond_change_active_slave(bond, new_active);
2172                 write_unlock_bh(&bond->curr_slave_lock);
2173         } else
2174                 res = -EINVAL;
2175
2176         read_unlock(&bond->lock);
2177
2178         return res;
2179 }
2180
2181 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2182 {
2183         struct bonding *bond = netdev_priv(bond_dev);
2184
2185         info->bond_mode = bond->params.mode;
2186         info->miimon = bond->params.miimon;
2187
2188         read_lock(&bond->lock);
2189         info->num_slaves = bond->slave_cnt;
2190         read_unlock(&bond->lock);
2191
2192         return 0;
2193 }
2194
2195 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2196 {
2197         struct bonding *bond = netdev_priv(bond_dev);
2198         struct slave *slave;
2199         int i, res = -ENODEV;
2200
2201         read_lock(&bond->lock);
2202
2203         bond_for_each_slave(bond, slave, i) {
2204                 if (i == (int)info->slave_id) {
2205                         res = 0;
2206                         strcpy(info->slave_name, slave->dev->name);
2207                         info->link = slave->link;
2208                         info->state = slave->state;
2209                         info->link_failure_count = slave->link_failure_count;
2210                         break;
2211                 }
2212         }
2213
2214         read_unlock(&bond->lock);
2215
2216         return res;
2217 }
2218
2219 /*-------------------------------- Monitoring -------------------------------*/
2220
2221
2222 static int bond_miimon_inspect(struct bonding *bond)
2223 {
2224         struct slave *slave;
2225         int i, link_state, commit = 0;
2226         bool ignore_updelay;
2227
2228         ignore_updelay = !bond->curr_active_slave ? true : false;
2229
2230         bond_for_each_slave(bond, slave, i) {
2231                 slave->new_link = BOND_LINK_NOCHANGE;
2232
2233                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2234
2235                 switch (slave->link) {
2236                 case BOND_LINK_UP:
2237                         if (link_state)
2238                                 continue;
2239
2240                         slave->link = BOND_LINK_FAIL;
2241                         slave->delay = bond->params.downdelay;
2242                         if (slave->delay) {
2243                                 pr_info(DRV_NAME
2244                                        ": %s: link status down for %s"
2245                                        "interface %s, disabling it in %d ms.\n",
2246                                        bond->dev->name,
2247                                        (bond->params.mode ==
2248                                         BOND_MODE_ACTIVEBACKUP) ?
2249                                        ((slave->state == BOND_STATE_ACTIVE) ?
2250                                         "active " : "backup ") : "",
2251                                        slave->dev->name,
2252                                        bond->params.downdelay * bond->params.miimon);
2253                         }
2254                         /*FALLTHRU*/
2255                 case BOND_LINK_FAIL:
2256                         if (link_state) {
2257                                 /*
2258                                  * recovered before downdelay expired
2259                                  */
2260                                 slave->link = BOND_LINK_UP;
2261                                 slave->jiffies = jiffies;
2262                                 pr_info(DRV_NAME
2263                                        ": %s: link status up again after %d "
2264                                        "ms for interface %s.\n",
2265                                        bond->dev->name,
2266                                        (bond->params.downdelay - slave->delay) *
2267                                        bond->params.miimon,
2268                                        slave->dev->name);
2269                                 continue;
2270                         }
2271
2272                         if (slave->delay <= 0) {
2273                                 slave->new_link = BOND_LINK_DOWN;
2274                                 commit++;
2275                                 continue;
2276                         }
2277
2278                         slave->delay--;
2279                         break;
2280
2281                 case BOND_LINK_DOWN:
2282                         if (!link_state)
2283                                 continue;
2284
2285                         slave->link = BOND_LINK_BACK;
2286                         slave->delay = bond->params.updelay;
2287
2288                         if (slave->delay) {
2289                                 pr_info(DRV_NAME
2290                                        ": %s: link status up for "
2291                                        "interface %s, enabling it in %d ms.\n",
2292                                        bond->dev->name, slave->dev->name,
2293                                        ignore_updelay ? 0 :
2294                                        bond->params.updelay *
2295                                        bond->params.miimon);
2296                         }
2297                         /*FALLTHRU*/
2298                 case BOND_LINK_BACK:
2299                         if (!link_state) {
2300                                 slave->link = BOND_LINK_DOWN;
2301                                 pr_info(DRV_NAME
2302                                        ": %s: link status down again after %d "
2303                                        "ms for interface %s.\n",
2304                                        bond->dev->name,
2305                                        (bond->params.updelay - slave->delay) *
2306                                        bond->params.miimon,
2307                                        slave->dev->name);
2308
2309                                 continue;
2310                         }
2311
2312                         if (ignore_updelay)
2313                                 slave->delay = 0;
2314
2315                         if (slave->delay <= 0) {
2316                                 slave->new_link = BOND_LINK_UP;
2317                                 commit++;
2318                                 ignore_updelay = false;
2319                                 continue;
2320                         }
2321
2322                         slave->delay--;
2323                         break;
2324                 }
2325         }
2326
2327         return commit;
2328 }
2329
2330 static void bond_miimon_commit(struct bonding *bond)
2331 {
2332         struct slave *slave;
2333         int i;
2334
2335         bond_for_each_slave(bond, slave, i) {
2336                 switch (slave->new_link) {
2337                 case BOND_LINK_NOCHANGE:
2338                         continue;
2339
2340                 case BOND_LINK_UP:
2341                         slave->link = BOND_LINK_UP;
2342                         slave->jiffies = jiffies;
2343
2344                         if (bond->params.mode == BOND_MODE_8023AD) {
2345                                 /* prevent it from being the active one */
2346                                 slave->state = BOND_STATE_BACKUP;
2347                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2348                                 /* make it immediately active */
2349                                 slave->state = BOND_STATE_ACTIVE;
2350                         } else if (slave != bond->primary_slave) {
2351                                 /* prevent it from being the active one */
2352                                 slave->state = BOND_STATE_BACKUP;
2353                         }
2354
2355                         pr_info(DRV_NAME
2356                                ": %s: link status definitely "
2357                                "up for interface %s.\n",
2358                                bond->dev->name, slave->dev->name);
2359
2360                         /* notify ad that the link status has changed */
2361                         if (bond->params.mode == BOND_MODE_8023AD)
2362                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2363
2364                         if (bond_is_lb(bond))
2365                                 bond_alb_handle_link_change(bond, slave,
2366                                                             BOND_LINK_UP);
2367
2368                         if (!bond->curr_active_slave ||
2369                             (slave == bond->primary_slave))
2370                                 goto do_failover;
2371
2372                         continue;
2373
2374                 case BOND_LINK_DOWN:
2375                         if (slave->link_failure_count < UINT_MAX)
2376                                 slave->link_failure_count++;
2377
2378                         slave->link = BOND_LINK_DOWN;
2379
2380                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2381                             bond->params.mode == BOND_MODE_8023AD)
2382                                 bond_set_slave_inactive_flags(slave);
2383
2384                         pr_info(DRV_NAME
2385                                ": %s: link status definitely down for "
2386                                "interface %s, disabling it\n",
2387                                bond->dev->name, slave->dev->name);
2388
2389                         if (bond->params.mode == BOND_MODE_8023AD)
2390                                 bond_3ad_handle_link_change(slave,
2391                                                             BOND_LINK_DOWN);
2392
2393                         if (bond_is_lb(bond))
2394                                 bond_alb_handle_link_change(bond, slave,
2395                                                             BOND_LINK_DOWN);
2396
2397                         if (slave == bond->curr_active_slave)
2398                                 goto do_failover;
2399
2400                         continue;
2401
2402                 default:
2403                         pr_err(DRV_NAME
2404                                ": %s: invalid new link %d on slave %s\n",
2405                                bond->dev->name, slave->new_link,
2406                                slave->dev->name);
2407                         slave->new_link = BOND_LINK_NOCHANGE;
2408
2409                         continue;
2410                 }
2411
2412 do_failover:
2413                 ASSERT_RTNL();
2414                 write_lock_bh(&bond->curr_slave_lock);
2415                 bond_select_active_slave(bond);
2416                 write_unlock_bh(&bond->curr_slave_lock);
2417         }
2418
2419         bond_set_carrier(bond);
2420 }
2421
2422 /*
2423  * bond_mii_monitor
2424  *
2425  * Really a wrapper that splits the mii monitor into two phases: an
2426  * inspection, then (if inspection indicates something needs to be done)
2427  * an acquisition of appropriate locks followed by a commit phase to
2428  * implement whatever link state changes are indicated.
2429  */
2430 void bond_mii_monitor(struct work_struct *work)
2431 {
2432         struct bonding *bond = container_of(work, struct bonding,
2433                                             mii_work.work);
2434
2435         read_lock(&bond->lock);
2436         if (bond->kill_timers)
2437                 goto out;
2438
2439         if (bond->slave_cnt == 0)
2440                 goto re_arm;
2441
2442         if (bond->send_grat_arp) {
2443                 read_lock(&bond->curr_slave_lock);
2444                 bond_send_gratuitous_arp(bond);
2445                 read_unlock(&bond->curr_slave_lock);
2446         }
2447
2448         if (bond->send_unsol_na) {
2449                 read_lock(&bond->curr_slave_lock);
2450                 bond_send_unsolicited_na(bond);
2451                 read_unlock(&bond->curr_slave_lock);
2452         }
2453
2454         if (bond_miimon_inspect(bond)) {
2455                 read_unlock(&bond->lock);
2456                 rtnl_lock();
2457                 read_lock(&bond->lock);
2458
2459                 bond_miimon_commit(bond);
2460
2461                 read_unlock(&bond->lock);
2462                 rtnl_unlock();  /* might sleep, hold no other locks */
2463                 read_lock(&bond->lock);
2464         }
2465
2466 re_arm:
2467         if (bond->params.miimon)
2468                 queue_delayed_work(bond->wq, &bond->mii_work,
2469                                    msecs_to_jiffies(bond->params.miimon));
2470 out:
2471         read_unlock(&bond->lock);
2472 }
2473
2474 static __be32 bond_glean_dev_ip(struct net_device *dev)
2475 {
2476         struct in_device *idev;
2477         struct in_ifaddr *ifa;
2478         __be32 addr = 0;
2479
2480         if (!dev)
2481                 return 0;
2482
2483         rcu_read_lock();
2484         idev = __in_dev_get_rcu(dev);
2485         if (!idev)
2486                 goto out;
2487
2488         ifa = idev->ifa_list;
2489         if (!ifa)
2490                 goto out;
2491
2492         addr = ifa->ifa_local;
2493 out:
2494         rcu_read_unlock();
2495         return addr;
2496 }
2497
2498 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2499 {
2500         struct vlan_entry *vlan;
2501
2502         if (ip == bond->master_ip)
2503                 return 1;
2504
2505         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2506                 if (ip == vlan->vlan_ip)
2507                         return 1;
2508         }
2509
2510         return 0;
2511 }
2512
2513 /*
2514  * We go to the (large) trouble of VLAN tagging ARP frames because
2515  * switches in VLAN mode (especially if ports are configured as
2516  * "native" to a VLAN) might not pass non-tagged frames.
2517  */
2518 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2519 {
2520         struct sk_buff *skb;
2521
2522         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2523                slave_dev->name, dest_ip, src_ip, vlan_id);
2524
2525         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2526                          NULL, slave_dev->dev_addr, NULL);
2527
2528         if (!skb) {
2529                 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2530                 return;
2531         }
2532         if (vlan_id) {
2533                 skb = vlan_put_tag(skb, vlan_id);
2534                 if (!skb) {
2535                         pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2536                         return;
2537                 }
2538         }
2539         arp_xmit(skb);
2540 }
2541
2542
2543 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2544 {
2545         int i, vlan_id, rv;
2546         __be32 *targets = bond->params.arp_targets;
2547         struct vlan_entry *vlan;
2548         struct net_device *vlan_dev;
2549         struct flowi fl;
2550         struct rtable *rt;
2551
2552         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2553                 if (!targets[i])
2554                         break;
2555                 pr_debug("basa: target %x\n", targets[i]);
2556                 if (list_empty(&bond->vlan_list)) {
2557                         pr_debug("basa: empty vlan: arp_send\n");
2558                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2559                                       bond->master_ip, 0);
2560                         continue;
2561                 }
2562
2563                 /*
2564                  * If VLANs are configured, we do a route lookup to
2565                  * determine which VLAN interface would be used, so we
2566                  * can tag the ARP with the proper VLAN tag.
2567                  */
2568                 memset(&fl, 0, sizeof(fl));
2569                 fl.fl4_dst = targets[i];
2570                 fl.fl4_tos = RTO_ONLINK;
2571
2572                 rv = ip_route_output_key(&init_net, &rt, &fl);
2573                 if (rv) {
2574                         if (net_ratelimit()) {
2575                                 pr_warning(DRV_NAME
2576                              ": %s: no route to arp_ip_target %pI4\n",
2577                                        bond->dev->name, &fl.fl4_dst);
2578                         }
2579                         continue;
2580                 }
2581
2582                 /*
2583                  * This target is not on a VLAN
2584                  */
2585                 if (rt->u.dst.dev == bond->dev) {
2586                         ip_rt_put(rt);
2587                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2588                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2589                                       bond->master_ip, 0);
2590                         continue;
2591                 }
2592
2593                 vlan_id = 0;
2594                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2595                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2596                         if (vlan_dev == rt->u.dst.dev) {
2597                                 vlan_id = vlan->vlan_id;
2598                                 pr_debug("basa: vlan match on %s %d\n",
2599                                        vlan_dev->name, vlan_id);
2600                                 break;
2601                         }
2602                 }
2603
2604                 if (vlan_id) {
2605                         ip_rt_put(rt);
2606                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2607                                       vlan->vlan_ip, vlan_id);
2608                         continue;
2609                 }
2610
2611                 if (net_ratelimit()) {
2612                         pr_warning(DRV_NAME
2613                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2614                                bond->dev->name, &fl.fl4_dst,
2615                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2616                 }
2617                 ip_rt_put(rt);
2618         }
2619 }
2620
2621 /*
2622  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2623  * for each VLAN above us.
2624  *
2625  * Caller must hold curr_slave_lock for read or better
2626  */
2627 static void bond_send_gratuitous_arp(struct bonding *bond)
2628 {
2629         struct slave *slave = bond->curr_active_slave;
2630         struct vlan_entry *vlan;
2631         struct net_device *vlan_dev;
2632
2633         pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2634                                 slave ? slave->dev->name : "NULL");
2635
2636         if (!slave || !bond->send_grat_arp ||
2637             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2638                 return;
2639
2640         bond->send_grat_arp--;
2641
2642         if (bond->master_ip) {
2643                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2644                                 bond->master_ip, 0);
2645         }
2646
2647         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2648                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2649                 if (vlan->vlan_ip) {
2650                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2651                                       vlan->vlan_ip, vlan->vlan_id);
2652                 }
2653         }
2654 }
2655
2656 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2657 {
2658         int i;
2659         __be32 *targets = bond->params.arp_targets;
2660
2661         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2662                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2663                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2664                 if (sip == targets[i]) {
2665                         if (bond_has_this_ip(bond, tip))
2666                                 slave->last_arp_rx = jiffies;
2667                         return;
2668                 }
2669         }
2670 }
2671
2672 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2673 {
2674         struct arphdr *arp;
2675         struct slave *slave;
2676         struct bonding *bond;
2677         unsigned char *arp_ptr;
2678         __be32 sip, tip;
2679
2680         if (dev_net(dev) != &init_net)
2681                 goto out;
2682
2683         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2684                 goto out;
2685
2686         bond = netdev_priv(dev);
2687         read_lock(&bond->lock);
2688
2689         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2690                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2691                 orig_dev ? orig_dev->name : "NULL");
2692
2693         slave = bond_get_slave_by_dev(bond, orig_dev);
2694         if (!slave || !slave_do_arp_validate(bond, slave))
2695                 goto out_unlock;
2696
2697         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2698                 goto out_unlock;
2699
2700         arp = arp_hdr(skb);
2701         if (arp->ar_hln != dev->addr_len ||
2702             skb->pkt_type == PACKET_OTHERHOST ||
2703             skb->pkt_type == PACKET_LOOPBACK ||
2704             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2705             arp->ar_pro != htons(ETH_P_IP) ||
2706             arp->ar_pln != 4)
2707                 goto out_unlock;
2708
2709         arp_ptr = (unsigned char *)(arp + 1);
2710         arp_ptr += dev->addr_len;
2711         memcpy(&sip, arp_ptr, 4);
2712         arp_ptr += 4 + dev->addr_len;
2713         memcpy(&tip, arp_ptr, 4);
2714
2715         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2716                 bond->dev->name, slave->dev->name, slave->state,
2717                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2718                 &sip, &tip);
2719
2720         /*
2721          * Backup slaves won't see the ARP reply, but do come through
2722          * here for each ARP probe (so we swap the sip/tip to validate
2723          * the probe).  In a "redundant switch, common router" type of
2724          * configuration, the ARP probe will (hopefully) travel from
2725          * the active, through one switch, the router, then the other
2726          * switch before reaching the backup.
2727          */
2728         if (slave->state == BOND_STATE_ACTIVE)
2729                 bond_validate_arp(bond, slave, sip, tip);
2730         else
2731                 bond_validate_arp(bond, slave, tip, sip);
2732
2733 out_unlock:
2734         read_unlock(&bond->lock);
2735 out:
2736         dev_kfree_skb(skb);
2737         return NET_RX_SUCCESS;
2738 }
2739
2740 /*
2741  * this function is called regularly to monitor each slave's link
2742  * ensuring that traffic is being sent and received when arp monitoring
2743  * is used in load-balancing mode. if the adapter has been dormant, then an
2744  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2745  * arp monitoring in active backup mode.
2746  */
2747 void bond_loadbalance_arp_mon(struct work_struct *work)
2748 {
2749         struct bonding *bond = container_of(work, struct bonding,
2750                                             arp_work.work);
2751         struct slave *slave, *oldcurrent;
2752         int do_failover = 0;
2753         int delta_in_ticks;
2754         int i;
2755
2756         read_lock(&bond->lock);
2757
2758         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2759
2760         if (bond->kill_timers)
2761                 goto out;
2762
2763         if (bond->slave_cnt == 0)
2764                 goto re_arm;
2765
2766         read_lock(&bond->curr_slave_lock);
2767         oldcurrent = bond->curr_active_slave;
2768         read_unlock(&bond->curr_slave_lock);
2769
2770         /* see if any of the previous devices are up now (i.e. they have
2771          * xmt and rcv traffic). the curr_active_slave does not come into
2772          * the picture unless it is null. also, slave->jiffies is not needed
2773          * here because we send an arp on each slave and give a slave as
2774          * long as it needs to get the tx/rx within the delta.
2775          * TODO: what about up/down delay in arp mode? it wasn't here before
2776          *       so it can wait
2777          */
2778         bond_for_each_slave(bond, slave, i) {
2779                 if (slave->link != BOND_LINK_UP) {
2780                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2781                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2782
2783                                 slave->link  = BOND_LINK_UP;
2784                                 slave->state = BOND_STATE_ACTIVE;
2785
2786                                 /* primary_slave has no meaning in round-robin
2787                                  * mode. the window of a slave being up and
2788                                  * curr_active_slave being null after enslaving
2789                                  * is closed.
2790                                  */
2791                                 if (!oldcurrent) {
2792                                         pr_info(DRV_NAME
2793                                                ": %s: link status definitely "
2794                                                "up for interface %s, ",
2795                                                bond->dev->name,
2796                                                slave->dev->name);
2797                                         do_failover = 1;
2798                                 } else {
2799                                         pr_info(DRV_NAME
2800                                                ": %s: interface %s is now up\n",
2801                                                bond->dev->name,
2802                                                slave->dev->name);
2803                                 }
2804                         }
2805                 } else {
2806                         /* slave->link == BOND_LINK_UP */
2807
2808                         /* not all switches will respond to an arp request
2809                          * when the source ip is 0, so don't take the link down
2810                          * if we don't know our ip yet
2811                          */
2812                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2813                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2814
2815                                 slave->link  = BOND_LINK_DOWN;
2816                                 slave->state = BOND_STATE_BACKUP;
2817
2818                                 if (slave->link_failure_count < UINT_MAX)
2819                                         slave->link_failure_count++;
2820
2821                                 pr_info(DRV_NAME
2822                                        ": %s: interface %s is now down.\n",
2823                                        bond->dev->name,
2824                                        slave->dev->name);
2825
2826                                 if (slave == oldcurrent)
2827                                         do_failover = 1;
2828                         }
2829                 }
2830
2831                 /* note: if switch is in round-robin mode, all links
2832                  * must tx arp to ensure all links rx an arp - otherwise
2833                  * links may oscillate or not come up at all; if switch is
2834                  * in something like xor mode, there is nothing we can
2835                  * do - all replies will be rx'ed on same link causing slaves
2836                  * to be unstable during low/no traffic periods
2837                  */
2838                 if (IS_UP(slave->dev))
2839                         bond_arp_send_all(bond, slave);
2840         }
2841
2842         if (do_failover) {
2843                 write_lock_bh(&bond->curr_slave_lock);
2844
2845                 bond_select_active_slave(bond);
2846
2847                 write_unlock_bh(&bond->curr_slave_lock);
2848         }
2849
2850 re_arm:
2851         if (bond->params.arp_interval)
2852                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2853 out:
2854         read_unlock(&bond->lock);
2855 }
2856
2857 /*
2858  * Called to inspect slaves for active-backup mode ARP monitor link state
2859  * changes.  Sets new_link in slaves to specify what action should take
2860  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2861  * to link states must be committed.
2862  *
2863  * Called with bond->lock held for read.
2864  */
2865 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2866 {
2867         struct slave *slave;
2868         int i, commit = 0;
2869
2870         bond_for_each_slave(bond, slave, i) {
2871                 slave->new_link = BOND_LINK_NOCHANGE;
2872
2873                 if (slave->link != BOND_LINK_UP) {
2874                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2875                                            delta_in_ticks)) {
2876                                 slave->new_link = BOND_LINK_UP;
2877                                 commit++;
2878                         }
2879
2880                         continue;
2881                 }
2882
2883                 /*
2884                  * Give slaves 2*delta after being enslaved or made
2885                  * active.  This avoids bouncing, as the last receive
2886                  * times need a full ARP monitor cycle to be updated.
2887                  */
2888                 if (!time_after_eq(jiffies, slave->jiffies +
2889                                    2 * delta_in_ticks))
2890                         continue;
2891
2892                 /*
2893                  * Backup slave is down if:
2894                  * - No current_arp_slave AND
2895                  * - more than 3*delta since last receive AND
2896                  * - the bond has an IP address
2897                  *
2898                  * Note: a non-null current_arp_slave indicates
2899                  * the curr_active_slave went down and we are
2900                  * searching for a new one; under this condition
2901                  * we only take the curr_active_slave down - this
2902                  * gives each slave a chance to tx/rx traffic
2903                  * before being taken out
2904                  */
2905                 if (slave->state == BOND_STATE_BACKUP &&
2906                     !bond->current_arp_slave &&
2907                     time_after(jiffies, slave_last_rx(bond, slave) +
2908                                3 * delta_in_ticks)) {
2909                         slave->new_link = BOND_LINK_DOWN;
2910                         commit++;
2911                 }
2912
2913                 /*
2914                  * Active slave is down if:
2915                  * - more than 2*delta since transmitting OR
2916                  * - (more than 2*delta since receive AND
2917                  *    the bond has an IP address)
2918                  */
2919                 if ((slave->state == BOND_STATE_ACTIVE) &&
2920                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2921                                     2 * delta_in_ticks) ||
2922                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2923                                      + 2 * delta_in_ticks)))) {
2924                         slave->new_link = BOND_LINK_DOWN;
2925                         commit++;
2926                 }
2927         }
2928
2929         read_lock(&bond->curr_slave_lock);
2930
2931         /*
2932          * Trigger a commit if the primary option setting has changed.
2933          */
2934         if (bond->primary_slave &&
2935             (bond->primary_slave != bond->curr_active_slave) &&
2936             (bond->primary_slave->link == BOND_LINK_UP))
2937                 commit++;
2938
2939         read_unlock(&bond->curr_slave_lock);
2940
2941         return commit;
2942 }
2943
2944 /*
2945  * Called to commit link state changes noted by inspection step of
2946  * active-backup mode ARP monitor.
2947  *
2948  * Called with RTNL and bond->lock for read.
2949  */
2950 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2951 {
2952         struct slave *slave;
2953         int i;
2954
2955         bond_for_each_slave(bond, slave, i) {
2956                 switch (slave->new_link) {
2957                 case BOND_LINK_NOCHANGE:
2958                         continue;
2959
2960                 case BOND_LINK_UP:
2961                         write_lock_bh(&bond->curr_slave_lock);
2962
2963                         if (!bond->curr_active_slave &&
2964                             time_before_eq(jiffies, dev_trans_start(slave->dev) +
2965                                            delta_in_ticks)) {
2966                                 slave->link = BOND_LINK_UP;
2967                                 bond_change_active_slave(bond, slave);
2968                                 bond->current_arp_slave = NULL;
2969
2970                                 pr_info(DRV_NAME
2971                                        ": %s: %s is up and now the "
2972                                        "active interface\n",
2973                                        bond->dev->name, slave->dev->name);
2974
2975                         } else if (bond->curr_active_slave != slave) {
2976                                 /* this slave has just come up but we
2977                                  * already have a current slave; this can
2978                                  * also happen if bond_enslave adds a new
2979                                  * slave that is up while we are searching
2980                                  * for a new slave
2981                                  */
2982                                 slave->link = BOND_LINK_UP;
2983                                 bond_set_slave_inactive_flags(slave);
2984                                 bond->current_arp_slave = NULL;
2985
2986                                 pr_info(DRV_NAME
2987                                        ": %s: backup interface %s is now up\n",
2988                                        bond->dev->name, slave->dev->name);
2989                         }
2990
2991                         write_unlock_bh(&bond->curr_slave_lock);
2992
2993                         break;
2994
2995                 case BOND_LINK_DOWN:
2996                         if (slave->link_failure_count < UINT_MAX)
2997                                 slave->link_failure_count++;
2998
2999                         slave->link = BOND_LINK_DOWN;
3000
3001                         if (slave == bond->curr_active_slave) {
3002                                 pr_info(DRV_NAME
3003                                        ": %s: link status down for active "
3004                                        "interface %s, disabling it\n",
3005                                        bond->dev->name, slave->dev->name);
3006
3007                                 bond_set_slave_inactive_flags(slave);
3008
3009                                 write_lock_bh(&bond->curr_slave_lock);
3010
3011                                 bond_select_active_slave(bond);
3012                                 if (bond->curr_active_slave)
3013                                         bond->curr_active_slave->jiffies =
3014                                                 jiffies;
3015
3016                                 write_unlock_bh(&bond->curr_slave_lock);
3017
3018                                 bond->current_arp_slave = NULL;
3019
3020                         } else if (slave->state == BOND_STATE_BACKUP) {
3021                                 pr_info(DRV_NAME
3022                                        ": %s: backup interface %s is now down\n",
3023                                        bond->dev->name, slave->dev->name);
3024
3025                                 bond_set_slave_inactive_flags(slave);
3026                         }
3027                         break;
3028
3029                 default:
3030                         pr_err(DRV_NAME
3031                                ": %s: impossible: new_link %d on slave %s\n",
3032                                bond->dev->name, slave->new_link,
3033                                slave->dev->name);
3034                 }
3035         }
3036
3037         /*
3038          * No race with changes to primary via sysfs, as we hold rtnl.
3039          */
3040         if (bond->primary_slave &&
3041             (bond->primary_slave != bond->curr_active_slave) &&
3042             (bond->primary_slave->link == BOND_LINK_UP)) {
3043                 write_lock_bh(&bond->curr_slave_lock);
3044                 bond_change_active_slave(bond, bond->primary_slave);
3045                 write_unlock_bh(&bond->curr_slave_lock);
3046         }
3047
3048         bond_set_carrier(bond);
3049 }
3050
3051 /*
3052  * Send ARP probes for active-backup mode ARP monitor.
3053  *
3054  * Called with bond->lock held for read.
3055  */
3056 static void bond_ab_arp_probe(struct bonding *bond)
3057 {
3058         struct slave *slave;
3059         int i;
3060
3061         read_lock(&bond->curr_slave_lock);
3062
3063         if (bond->current_arp_slave && bond->curr_active_slave)
3064                 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3065                        bond->current_arp_slave->dev->name,
3066                        bond->curr_active_slave->dev->name);
3067
3068         if (bond->curr_active_slave) {
3069                 bond_arp_send_all(bond, bond->curr_active_slave);
3070                 read_unlock(&bond->curr_slave_lock);
3071                 return;
3072         }
3073
3074         read_unlock(&bond->curr_slave_lock);
3075
3076         /* if we don't have a curr_active_slave, search for the next available
3077          * backup slave from the current_arp_slave and make it the candidate
3078          * for becoming the curr_active_slave
3079          */
3080
3081         if (!bond->current_arp_slave) {
3082                 bond->current_arp_slave = bond->first_slave;
3083                 if (!bond->current_arp_slave)
3084                         return;
3085         }
3086
3087         bond_set_slave_inactive_flags(bond->current_arp_slave);
3088
3089         /* search for next candidate */
3090         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3091                 if (IS_UP(slave->dev)) {
3092                         slave->link = BOND_LINK_BACK;
3093                         bond_set_slave_active_flags(slave);
3094                         bond_arp_send_all(bond, slave);
3095                         slave->jiffies = jiffies;
3096                         bond->current_arp_slave = slave;
3097                         break;
3098                 }
3099
3100                 /* if the link state is up at this point, we
3101                  * mark it down - this can happen if we have
3102                  * simultaneous link failures and
3103                  * reselect_active_interface doesn't make this
3104                  * one the current slave so it is still marked
3105                  * up when it is actually down
3106                  */
3107                 if (slave->link == BOND_LINK_UP) {
3108                         slave->link = BOND_LINK_DOWN;
3109                         if (slave->link_failure_count < UINT_MAX)
3110                                 slave->link_failure_count++;
3111
3112                         bond_set_slave_inactive_flags(slave);
3113
3114                         pr_info(DRV_NAME
3115                                ": %s: backup interface %s is now down.\n",
3116                                bond->dev->name, slave->dev->name);
3117                 }
3118         }
3119 }
3120
3121 void bond_activebackup_arp_mon(struct work_struct *work)
3122 {
3123         struct bonding *bond = container_of(work, struct bonding,
3124                                             arp_work.work);
3125         int delta_in_ticks;
3126
3127         read_lock(&bond->lock);
3128
3129         if (bond->kill_timers)
3130                 goto out;
3131
3132         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3133
3134         if (bond->slave_cnt == 0)
3135                 goto re_arm;
3136
3137         if (bond->send_grat_arp) {
3138                 read_lock(&bond->curr_slave_lock);
3139                 bond_send_gratuitous_arp(bond);
3140                 read_unlock(&bond->curr_slave_lock);
3141         }
3142
3143         if (bond->send_unsol_na) {
3144                 read_lock(&bond->curr_slave_lock);
3145                 bond_send_unsolicited_na(bond);
3146                 read_unlock(&bond->curr_slave_lock);
3147         }
3148
3149         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3150                 read_unlock(&bond->lock);
3151                 rtnl_lock();
3152                 read_lock(&bond->lock);
3153
3154                 bond_ab_arp_commit(bond, delta_in_ticks);
3155
3156                 read_unlock(&bond->lock);
3157                 rtnl_unlock();
3158                 read_lock(&bond->lock);
3159         }
3160
3161         bond_ab_arp_probe(bond);
3162
3163 re_arm:
3164         if (bond->params.arp_interval)
3165                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3166 out:
3167         read_unlock(&bond->lock);
3168 }
3169
3170 /*------------------------------ proc/seq_file-------------------------------*/
3171
3172 #ifdef CONFIG_PROC_FS
3173
3174 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3175         __acquires(&dev_base_lock)
3176         __acquires(&bond->lock)
3177 {
3178         struct bonding *bond = seq->private;
3179         loff_t off = 0;
3180         struct slave *slave;
3181         int i;
3182
3183         /* make sure the bond won't be taken away */
3184         read_lock(&dev_base_lock);
3185         read_lock(&bond->lock);
3186
3187         if (*pos == 0)
3188                 return SEQ_START_TOKEN;
3189
3190         bond_for_each_slave(bond, slave, i) {
3191                 if (++off == *pos)
3192                         return slave;
3193         }
3194
3195         return NULL;
3196 }
3197
3198 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3199 {
3200         struct bonding *bond = seq->private;
3201         struct slave *slave = v;
3202
3203         ++*pos;
3204         if (v == SEQ_START_TOKEN)
3205                 return bond->first_slave;
3206
3207         slave = slave->next;
3208
3209         return (slave == bond->first_slave) ? NULL : slave;
3210 }
3211
3212 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3213         __releases(&bond->lock)
3214         __releases(&dev_base_lock)
3215 {
3216         struct bonding *bond = seq->private;
3217
3218         read_unlock(&bond->lock);
3219         read_unlock(&dev_base_lock);
3220 }
3221
3222 static void bond_info_show_master(struct seq_file *seq)
3223 {
3224         struct bonding *bond = seq->private;
3225         struct slave *curr;
3226         int i;
3227
3228         read_lock(&bond->curr_slave_lock);
3229         curr = bond->curr_active_slave;
3230         read_unlock(&bond->curr_slave_lock);
3231
3232         seq_printf(seq, "Bonding Mode: %s",
3233                    bond_mode_name(bond->params.mode));
3234
3235         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3236             bond->params.fail_over_mac)
3237                 seq_printf(seq, " (fail_over_mac %s)",
3238                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3239
3240         seq_printf(seq, "\n");
3241
3242         if (bond->params.mode == BOND_MODE_XOR ||
3243                 bond->params.mode == BOND_MODE_8023AD) {
3244                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3245                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3246                         bond->params.xmit_policy);
3247         }
3248
3249         if (USES_PRIMARY(bond->params.mode)) {
3250                 seq_printf(seq, "Primary Slave: %s\n",
3251                            (bond->primary_slave) ?
3252                            bond->primary_slave->dev->name : "None");
3253
3254                 seq_printf(seq, "Currently Active Slave: %s\n",
3255                            (curr) ? curr->dev->name : "None");
3256         }
3257
3258         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3259                    "up" : "down");
3260         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3261         seq_printf(seq, "Up Delay (ms): %d\n",
3262                    bond->params.updelay * bond->params.miimon);
3263         seq_printf(seq, "Down Delay (ms): %d\n",
3264                    bond->params.downdelay * bond->params.miimon);
3265
3266
3267         /* ARP information */
3268         if (bond->params.arp_interval > 0) {
3269                 int printed = 0;
3270                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3271                                 bond->params.arp_interval);
3272
3273                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3274
3275                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3276                         if (!bond->params.arp_targets[i])
3277                                 break;
3278                         if (printed)
3279                                 seq_printf(seq, ",");
3280                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3281                         printed = 1;
3282                 }
3283                 seq_printf(seq, "\n");
3284         }
3285
3286         if (bond->params.mode == BOND_MODE_8023AD) {
3287                 struct ad_info ad_info;
3288
3289                 seq_puts(seq, "\n802.3ad info\n");
3290                 seq_printf(seq, "LACP rate: %s\n",
3291                            (bond->params.lacp_fast) ? "fast" : "slow");
3292                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3293                            ad_select_tbl[bond->params.ad_select].modename);
3294
3295                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3296                         seq_printf(seq, "bond %s has no active aggregator\n",
3297                                    bond->dev->name);
3298                 } else {
3299                         seq_printf(seq, "Active Aggregator Info:\n");
3300
3301                         seq_printf(seq, "\tAggregator ID: %d\n",
3302                                    ad_info.aggregator_id);
3303                         seq_printf(seq, "\tNumber of ports: %d\n",
3304                                    ad_info.ports);
3305                         seq_printf(seq, "\tActor Key: %d\n",
3306                                    ad_info.actor_key);
3307                         seq_printf(seq, "\tPartner Key: %d\n",
3308                                    ad_info.partner_key);
3309                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3310                                    ad_info.partner_system);
3311                 }
3312         }
3313 }
3314
3315 static void bond_info_show_slave(struct seq_file *seq,
3316                                  const struct slave *slave)
3317 {
3318         struct bonding *bond = seq->private;
3319
3320         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3321         seq_printf(seq, "MII Status: %s\n",
3322                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3323         seq_printf(seq, "Link Failure Count: %u\n",
3324                    slave->link_failure_count);
3325
3326         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3327
3328         if (bond->params.mode == BOND_MODE_8023AD) {
3329                 const struct aggregator *agg
3330                         = SLAVE_AD_INFO(slave).port.aggregator;
3331
3332                 if (agg)
3333                         seq_printf(seq, "Aggregator ID: %d\n",
3334                                    agg->aggregator_identifier);
3335                 else
3336                         seq_puts(seq, "Aggregator ID: N/A\n");
3337         }
3338 }
3339
3340 static int bond_info_seq_show(struct seq_file *seq, void *v)
3341 {
3342         if (v == SEQ_START_TOKEN) {
3343                 seq_printf(seq, "%s\n", version);
3344                 bond_info_show_master(seq);
3345         } else
3346                 bond_info_show_slave(seq, v);
3347
3348         return 0;
3349 }
3350
3351 static const struct seq_operations bond_info_seq_ops = {
3352         .start = bond_info_seq_start,
3353         .next  = bond_info_seq_next,
3354         .stop  = bond_info_seq_stop,
3355         .show  = bond_info_seq_show,
3356 };
3357
3358 static int bond_info_open(struct inode *inode, struct file *file)
3359 {
3360         struct seq_file *seq;
3361         struct proc_dir_entry *proc;
3362         int res;
3363
3364         res = seq_open(file, &bond_info_seq_ops);
3365         if (!res) {
3366                 /* recover the pointer buried in proc_dir_entry data */
3367                 seq = file->private_data;
3368                 proc = PDE(inode);
3369                 seq->private = proc->data;
3370         }
3371
3372         return res;
3373 }
3374
3375 static const struct file_operations bond_info_fops = {
3376         .owner   = THIS_MODULE,
3377         .open    = bond_info_open,
3378         .read    = seq_read,
3379         .llseek  = seq_lseek,
3380         .release = seq_release,
3381 };
3382
3383 static int bond_create_proc_entry(struct bonding *bond)
3384 {
3385         struct net_device *bond_dev = bond->dev;
3386
3387         if (bond_proc_dir) {
3388                 bond->proc_entry = proc_create_data(bond_dev->name,
3389                                                     S_IRUGO, bond_proc_dir,
3390                                                     &bond_info_fops, bond);
3391                 if (bond->proc_entry == NULL)
3392                         pr_warning(DRV_NAME
3393                                ": Warning: Cannot create /proc/net/%s/%s\n",
3394                                DRV_NAME, bond_dev->name);
3395                 else
3396                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3397         }
3398
3399         return 0;
3400 }
3401
3402 static void bond_remove_proc_entry(struct bonding *bond)
3403 {
3404         if (bond_proc_dir && bond->proc_entry) {
3405                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3406                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3407                 bond->proc_entry = NULL;
3408         }
3409 }
3410
3411 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3412  * Caller must hold rtnl_lock.
3413  */
3414 static void bond_create_proc_dir(void)
3415 {
3416         if (!bond_proc_dir) {
3417                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3418                 if (!bond_proc_dir)
3419                         pr_warning(DRV_NAME
3420                                 ": Warning: cannot create /proc/net/%s\n",
3421                                 DRV_NAME);
3422         }
3423 }
3424
3425 /* Destroy the bonding directory under /proc/net, if empty.
3426  * Caller must hold rtnl_lock.
3427  */
3428 static void bond_destroy_proc_dir(void)
3429 {
3430         if (bond_proc_dir) {
3431                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3432                 bond_proc_dir = NULL;
3433         }
3434 }
3435
3436 #else /* !CONFIG_PROC_FS */
3437
3438 static int bond_create_proc_entry(struct bonding *bond)
3439 {
3440 }
3441
3442 static void bond_remove_proc_entry(struct bonding *bond)
3443 {
3444 }
3445
3446 static void bond_create_proc_dir(void)
3447 {
3448 }
3449
3450 static void bond_destroy_proc_dir(void)
3451 {
3452 }
3453
3454 #endif /* CONFIG_PROC_FS */
3455
3456
3457 /*-------------------------- netdev event handling --------------------------*/
3458
3459 /*
3460  * Change device name
3461  */
3462 static int bond_event_changename(struct bonding *bond)
3463 {
3464         bond_remove_proc_entry(bond);
3465         bond_create_proc_entry(bond);
3466
3467         bond_destroy_sysfs_entry(bond);
3468         bond_create_sysfs_entry(bond);
3469
3470         return NOTIFY_DONE;
3471 }
3472
3473 static int bond_master_netdev_event(unsigned long event,
3474                                     struct net_device *bond_dev)
3475 {
3476         struct bonding *event_bond = netdev_priv(bond_dev);
3477
3478         switch (event) {
3479         case NETDEV_CHANGENAME:
3480                 return bond_event_changename(event_bond);
3481         case NETDEV_UNREGISTER:
3482                 bond_release_all(event_bond->dev);
3483                 break;
3484         default:
3485                 break;
3486         }
3487
3488         return NOTIFY_DONE;
3489 }
3490
3491 static int bond_slave_netdev_event(unsigned long event,
3492                                    struct net_device *slave_dev)
3493 {
3494         struct net_device *bond_dev = slave_dev->master;
3495         struct bonding *bond = netdev_priv(bond_dev);
3496
3497         switch (event) {
3498         case NETDEV_UNREGISTER:
3499                 if (bond_dev) {
3500                         if (bond->setup_by_slave)
3501                                 bond_release_and_destroy(bond_dev, slave_dev);
3502                         else
3503                                 bond_release(bond_dev, slave_dev);
3504                 }
3505                 break;
3506         case NETDEV_CHANGE:
3507                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3508                         struct slave *slave;
3509
3510                         slave = bond_get_slave_by_dev(bond, slave_dev);
3511                         if (slave) {
3512                                 u16 old_speed = slave->speed;
3513                                 u16 old_duplex = slave->duplex;
3514
3515                                 bond_update_speed_duplex(slave);
3516
3517                                 if (bond_is_lb(bond))
3518                                         break;
3519
3520                                 if (old_speed != slave->speed)
3521                                         bond_3ad_adapter_speed_changed(slave);
3522                                 if (old_duplex != slave->duplex)
3523                                         bond_3ad_adapter_duplex_changed(slave);
3524                         }
3525                 }
3526
3527                 break;
3528         case NETDEV_DOWN:
3529                 /*
3530                  * ... Or is it this?
3531                  */
3532                 break;
3533         case NETDEV_CHANGEMTU:
3534                 /*
3535                  * TODO: Should slaves be allowed to
3536                  * independently alter their MTU?  For
3537                  * an active-backup bond, slaves need
3538                  * not be the same type of device, so
3539                  * MTUs may vary.  For other modes,
3540                  * slaves arguably should have the
3541                  * same MTUs. To do this, we'd need to
3542                  * take over the slave's change_mtu
3543                  * function for the duration of their
3544                  * servitude.
3545                  */
3546                 break;
3547         case NETDEV_CHANGENAME:
3548                 /*
3549                  * TODO: handle changing the primary's name
3550                  */
3551                 break;
3552         case NETDEV_FEAT_CHANGE:
3553                 bond_compute_features(bond);
3554                 break;
3555         default:
3556                 break;
3557         }
3558
3559         return NOTIFY_DONE;
3560 }
3561
3562 /*
3563  * bond_netdev_event: handle netdev notifier chain events.
3564  *
3565  * This function receives events for the netdev chain.  The caller (an
3566  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3567  * locks for us to safely manipulate the slave devices (RTNL lock,
3568  * dev_probe_lock).
3569  */
3570 static int bond_netdev_event(struct notifier_block *this,
3571                              unsigned long event, void *ptr)
3572 {
3573         struct net_device *event_dev = (struct net_device *)ptr;
3574
3575         if (dev_net(event_dev) != &init_net)
3576                 return NOTIFY_DONE;
3577
3578         pr_debug("event_dev: %s, event: %lx\n",
3579                 (event_dev ? event_dev->name : "None"),
3580                 event);
3581
3582         if (!(event_dev->priv_flags & IFF_BONDING))
3583                 return NOTIFY_DONE;
3584
3585         if (event_dev->flags & IFF_MASTER) {
3586                 pr_debug("IFF_MASTER\n");
3587                 return bond_master_netdev_event(event, event_dev);
3588         }
3589
3590         if (event_dev->flags & IFF_SLAVE) {
3591                 pr_debug("IFF_SLAVE\n");
3592                 return bond_slave_netdev_event(event, event_dev);
3593         }
3594
3595         return NOTIFY_DONE;
3596 }
3597
3598 /*
3599  * bond_inetaddr_event: handle inetaddr notifier chain events.
3600  *
3601  * We keep track of device IPs primarily to use as source addresses in
3602  * ARP monitor probes (rather than spewing out broadcasts all the time).
3603  *
3604  * We track one IP for the main device (if it has one), plus one per VLAN.
3605  */
3606 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3607 {
3608         struct in_ifaddr *ifa = ptr;
3609         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3610         struct bonding *bond;
3611         struct vlan_entry *vlan;
3612
3613         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3614                 return NOTIFY_DONE;
3615
3616         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3617                 if (bond->dev == event_dev) {
3618                         switch (event) {
3619                         case NETDEV_UP:
3620                                 bond->master_ip = ifa->ifa_local;
3621                                 return NOTIFY_OK;
3622                         case NETDEV_DOWN:
3623                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3624                                 return NOTIFY_OK;
3625                         default:
3626                                 return NOTIFY_DONE;
3627                         }
3628                 }
3629
3630                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3631                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3632                         if (vlan_dev == event_dev) {
3633                                 switch (event) {
3634                                 case NETDEV_UP:
3635                                         vlan->vlan_ip = ifa->ifa_local;
3636                                         return NOTIFY_OK;
3637                                 case NETDEV_DOWN:
3638                                         vlan->vlan_ip =
3639                                                 bond_glean_dev_ip(vlan_dev);
3640                                         return NOTIFY_OK;
3641                                 default:
3642                                         return NOTIFY_DONE;
3643                                 }
3644                         }
3645                 }
3646         }
3647         return NOTIFY_DONE;
3648 }
3649
3650 static struct notifier_block bond_netdev_notifier = {
3651         .notifier_call = bond_netdev_event,
3652 };
3653
3654 static struct notifier_block bond_inetaddr_notifier = {
3655         .notifier_call = bond_inetaddr_event,
3656 };
3657
3658 /*-------------------------- Packet type handling ---------------------------*/
3659
3660 /* register to receive lacpdus on a bond */
3661 static void bond_register_lacpdu(struct bonding *bond)
3662 {
3663         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3664
3665         /* initialize packet type */
3666         pk_type->type = PKT_TYPE_LACPDU;
3667         pk_type->dev = bond->dev;
3668         pk_type->func = bond_3ad_lacpdu_recv;
3669
3670         dev_add_pack(pk_type);
3671 }
3672
3673 /* unregister to receive lacpdus on a bond */
3674 static void bond_unregister_lacpdu(struct bonding *bond)
3675 {
3676         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3677 }
3678
3679 void bond_register_arp(struct bonding *bond)
3680 {
3681         struct packet_type *pt = &bond->arp_mon_pt;
3682
3683         if (pt->type)
3684                 return;
3685
3686         pt->type = htons(ETH_P_ARP);
3687         pt->dev = bond->dev;
3688         pt->func = bond_arp_rcv;
3689         dev_add_pack(pt);
3690 }
3691
3692 void bond_unregister_arp(struct bonding *bond)
3693 {
3694         struct packet_type *pt = &bond->arp_mon_pt;
3695
3696         dev_remove_pack(pt);
3697         pt->type = 0;
3698 }
3699
3700 /*---------------------------- Hashing Policies -----------------------------*/
3701
3702 /*
3703  * Hash for the output device based upon layer 2 and layer 3 data. If
3704  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3705  */
3706 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3707                                      struct net_device *bond_dev, int count)
3708 {
3709         struct ethhdr *data = (struct ethhdr *)skb->data;
3710         struct iphdr *iph = ip_hdr(skb);
3711
3712         if (skb->protocol == htons(ETH_P_IP)) {
3713                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3714                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3715         }
3716
3717         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3718 }
3719
3720 /*
3721  * Hash for the output device based upon layer 3 and layer 4 data. If
3722  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3723  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3724  */
3725 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3726                                     struct net_device *bond_dev, int count)
3727 {
3728         struct ethhdr *data = (struct ethhdr *)skb->data;
3729         struct iphdr *iph = ip_hdr(skb);
3730         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3731         int layer4_xor = 0;
3732
3733         if (skb->protocol == htons(ETH_P_IP)) {
3734                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3735                     (iph->protocol == IPPROTO_TCP ||
3736                      iph->protocol == IPPROTO_UDP)) {
3737                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3738                 }
3739                 return (layer4_xor ^
3740                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3741
3742         }
3743
3744         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3745 }
3746
3747 /*
3748  * Hash for the output device based upon layer 2 data
3749  */
3750 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3751                                    struct net_device *bond_dev, int count)
3752 {
3753         struct ethhdr *data = (struct ethhdr *)skb->data;
3754
3755         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3756 }
3757
3758 /*-------------------------- Device entry points ----------------------------*/
3759
3760 static int bond_open(struct net_device *bond_dev)
3761 {
3762         struct bonding *bond = netdev_priv(bond_dev);
3763
3764         bond->kill_timers = 0;
3765
3766         if (bond_is_lb(bond)) {
3767                 /* bond_alb_initialize must be called before the timer
3768                  * is started.
3769                  */
3770                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3771                         /* something went wrong - fail the open operation */
3772                         return -1;
3773                 }
3774
3775                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3776                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3777         }
3778
3779         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3780                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3781                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3782         }
3783
3784         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3785                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3786                         INIT_DELAYED_WORK(&bond->arp_work,
3787                                           bond_activebackup_arp_mon);
3788                 else
3789                         INIT_DELAYED_WORK(&bond->arp_work,
3790                                           bond_loadbalance_arp_mon);
3791
3792                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3793                 if (bond->params.arp_validate)
3794                         bond_register_arp(bond);
3795         }
3796
3797         if (bond->params.mode == BOND_MODE_8023AD) {
3798                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3799                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3800                 /* register to receive LACPDUs */
3801                 bond_register_lacpdu(bond);
3802                 bond_3ad_initiate_agg_selection(bond, 1);
3803         }
3804
3805         return 0;
3806 }
3807
3808 static int bond_close(struct net_device *bond_dev)
3809 {
3810         struct bonding *bond = netdev_priv(bond_dev);
3811
3812         if (bond->params.mode == BOND_MODE_8023AD) {
3813                 /* Unregister the receive of LACPDUs */
3814                 bond_unregister_lacpdu(bond);
3815         }
3816
3817         if (bond->params.arp_validate)
3818                 bond_unregister_arp(bond);
3819
3820         write_lock_bh(&bond->lock);
3821
3822         bond->send_grat_arp = 0;
3823         bond->send_unsol_na = 0;
3824
3825         /* signal timers not to re-arm */
3826         bond->kill_timers = 1;
3827
3828         write_unlock_bh(&bond->lock);
3829
3830         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3831                 cancel_delayed_work(&bond->mii_work);
3832         }
3833
3834         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3835                 cancel_delayed_work(&bond->arp_work);
3836         }
3837
3838         switch (bond->params.mode) {
3839         case BOND_MODE_8023AD:
3840                 cancel_delayed_work(&bond->ad_work);
3841                 break;
3842         case BOND_MODE_TLB:
3843         case BOND_MODE_ALB:
3844                 cancel_delayed_work(&bond->alb_work);
3845                 break;
3846         default:
3847                 break;
3848         }
3849
3850
3851         if (bond_is_lb(bond)) {
3852                 /* Must be called only after all
3853                  * slaves have been released
3854                  */
3855                 bond_alb_deinitialize(bond);
3856         }
3857
3858         return 0;
3859 }
3860
3861 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3862 {
3863         struct bonding *bond = netdev_priv(bond_dev);
3864         struct net_device_stats *stats = &bond->stats;
3865         struct net_device_stats local_stats;
3866         struct slave *slave;
3867         int i;
3868
3869         memset(&local_stats, 0, sizeof(struct net_device_stats));
3870
3871         read_lock_bh(&bond->lock);
3872
3873         bond_for_each_slave(bond, slave, i) {
3874                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3875
3876                 local_stats.rx_packets += sstats->rx_packets;
3877                 local_stats.rx_bytes += sstats->rx_bytes;
3878                 local_stats.rx_errors += sstats->rx_errors;
3879                 local_stats.rx_dropped += sstats->rx_dropped;
3880
3881                 local_stats.tx_packets += sstats->tx_packets;
3882                 local_stats.tx_bytes += sstats->tx_bytes;
3883                 local_stats.tx_errors += sstats->tx_errors;
3884                 local_stats.tx_dropped += sstats->tx_dropped;
3885
3886                 local_stats.multicast += sstats->multicast;
3887                 local_stats.collisions += sstats->collisions;
3888
3889                 local_stats.rx_length_errors += sstats->rx_length_errors;
3890                 local_stats.rx_over_errors += sstats->rx_over_errors;
3891                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3892                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3893                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3894                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3895
3896                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3897                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3898                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3899                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3900                 local_stats.tx_window_errors += sstats->tx_window_errors;
3901         }
3902
3903         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3904
3905         read_unlock_bh(&bond->lock);
3906
3907         return stats;
3908 }
3909
3910 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3911 {
3912         struct net_device *slave_dev = NULL;
3913         struct ifbond k_binfo;
3914         struct ifbond __user *u_binfo = NULL;
3915         struct ifslave k_sinfo;
3916         struct ifslave __user *u_sinfo = NULL;
3917         struct mii_ioctl_data *mii = NULL;
3918         int res = 0;
3919
3920         pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3921                 bond_dev->name, cmd);
3922
3923         switch (cmd) {
3924         case SIOCGMIIPHY:
3925                 mii = if_mii(ifr);
3926                 if (!mii)
3927                         return -EINVAL;
3928
3929                 mii->phy_id = 0;
3930                 /* Fall Through */
3931         case SIOCGMIIREG:
3932                 /*
3933                  * We do this again just in case we were called by SIOCGMIIREG
3934                  * instead of SIOCGMIIPHY.
3935                  */
3936                 mii = if_mii(ifr);
3937                 if (!mii)
3938                         return -EINVAL;
3939
3940
3941                 if (mii->reg_num == 1) {
3942                         struct bonding *bond = netdev_priv(bond_dev);
3943                         mii->val_out = 0;
3944                         read_lock(&bond->lock);
3945                         read_lock(&bond->curr_slave_lock);
3946                         if (netif_carrier_ok(bond->dev))
3947                                 mii->val_out = BMSR_LSTATUS;
3948
3949                         read_unlock(&bond->curr_slave_lock);
3950                         read_unlock(&bond->lock);
3951                 }
3952
3953                 return 0;
3954         case BOND_INFO_QUERY_OLD:
3955         case SIOCBONDINFOQUERY:
3956                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3957
3958                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3959                         return -EFAULT;
3960
3961                 res = bond_info_query(bond_dev, &k_binfo);
3962                 if (res == 0 &&
3963                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3964                         return -EFAULT;
3965
3966                 return res;
3967         case BOND_SLAVE_INFO_QUERY_OLD:
3968         case SIOCBONDSLAVEINFOQUERY:
3969                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3970
3971                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3972                         return -EFAULT;
3973
3974                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3975                 if (res == 0 &&
3976                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3977                         return -EFAULT;
3978
3979                 return res;
3980         default:
3981                 /* Go on */
3982                 break;
3983         }
3984
3985         if (!capable(CAP_NET_ADMIN))
3986                 return -EPERM;
3987
3988         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3989
3990         pr_debug("slave_dev=%p: \n", slave_dev);
3991
3992         if (!slave_dev)
3993                 res = -ENODEV;
3994         else {
3995                 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
3996                 switch (cmd) {
3997                 case BOND_ENSLAVE_OLD:
3998                 case SIOCBONDENSLAVE:
3999                         res = bond_enslave(bond_dev, slave_dev);
4000                         break;
4001                 case BOND_RELEASE_OLD:
4002                 case SIOCBONDRELEASE:
4003                         res = bond_release(bond_dev, slave_dev);
4004                         break;
4005                 case BOND_SETHWADDR_OLD:
4006                 case SIOCBONDSETHWADDR:
4007                         res = bond_sethwaddr(bond_dev, slave_dev);
4008                         break;
4009                 case BOND_CHANGE_ACTIVE_OLD:
4010                 case SIOCBONDCHANGEACTIVE:
4011                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4012                         break;
4013                 default:
4014                         res = -EOPNOTSUPP;
4015                 }
4016
4017                 dev_put(slave_dev);
4018         }
4019
4020         return res;
4021 }
4022
4023 static void bond_set_multicast_list(struct net_device *bond_dev)
4024 {
4025         struct bonding *bond = netdev_priv(bond_dev);
4026         struct dev_mc_list *dmi;
4027
4028         /*
4029          * Do promisc before checking multicast_mode
4030          */
4031         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4032                 /*
4033                  * FIXME: Need to handle the error when one of the multi-slaves
4034                  * encounters error.
4035                  */
4036                 bond_set_promiscuity(bond, 1);
4037
4038
4039         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4040                 bond_set_promiscuity(bond, -1);
4041
4042
4043         /* set allmulti flag to slaves */
4044         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4045                 /*
4046                  * FIXME: Need to handle the error when one of the multi-slaves
4047                  * encounters error.
4048                  */
4049                 bond_set_allmulti(bond, 1);
4050
4051
4052         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4053                 bond_set_allmulti(bond, -1);
4054
4055
4056         read_lock(&bond->lock);
4057
4058         bond->flags = bond_dev->flags;
4059
4060         /* looking for addresses to add to slaves' mc list */
4061         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4062                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4063                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4064         }
4065
4066         /* looking for addresses to delete from slaves' list */
4067         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4068                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4069                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4070         }
4071
4072         /* save master's multicast list */
4073         bond_mc_list_destroy(bond);
4074         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4075
4076         read_unlock(&bond->lock);
4077 }
4078
4079 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4080 {
4081         struct bonding *bond = netdev_priv(dev);
4082         struct slave *slave = bond->first_slave;
4083
4084         if (slave) {
4085                 const struct net_device_ops *slave_ops
4086                         = slave->dev->netdev_ops;
4087                 if (slave_ops->ndo_neigh_setup)
4088                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4089         }
4090         return 0;
4091 }
4092
4093 /*
4094  * Change the MTU of all of a master's slaves to match the master
4095  */
4096 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4097 {
4098         struct bonding *bond = netdev_priv(bond_dev);
4099         struct slave *slave, *stop_at;
4100         int res = 0;
4101         int i;
4102
4103         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4104                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4105
4106         /* Can't hold bond->lock with bh disabled here since
4107          * some base drivers panic. On the other hand we can't
4108          * hold bond->lock without bh disabled because we'll
4109          * deadlock. The only solution is to rely on the fact
4110          * that we're under rtnl_lock here, and the slaves
4111          * list won't change. This doesn't solve the problem
4112          * of setting the slave's MTU while it is
4113          * transmitting, but the assumption is that the base
4114          * driver can handle that.
4115          *
4116          * TODO: figure out a way to safely iterate the slaves
4117          * list, but without holding a lock around the actual
4118          * call to the base driver.
4119          */
4120
4121         bond_for_each_slave(bond, slave, i) {
4122                 pr_debug("s %p s->p %p c_m %p\n", slave,
4123                         slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4124
4125                 res = dev_set_mtu(slave->dev, new_mtu);
4126
4127                 if (res) {
4128                         /* If we failed to set the slave's mtu to the new value
4129                          * we must abort the operation even in ACTIVE_BACKUP
4130                          * mode, because if we allow the backup slaves to have
4131                          * different mtu values than the active slave we'll
4132                          * need to change their mtu when doing a failover. That
4133                          * means changing their mtu from timer context, which
4134                          * is probably not a good idea.
4135                          */
4136                         pr_debug("err %d %s\n", res, slave->dev->name);
4137                         goto unwind;
4138                 }
4139         }
4140
4141         bond_dev->mtu = new_mtu;
4142
4143         return 0;
4144
4145 unwind:
4146         /* unwind from head to the slave that failed */
4147         stop_at = slave;
4148         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4149                 int tmp_res;
4150
4151                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4152                 if (tmp_res) {
4153                         pr_debug("unwind err %d dev %s\n", tmp_res,
4154                                 slave->dev->name);
4155                 }
4156         }
4157
4158         return res;
4159 }
4160
4161 /*
4162  * Change HW address
4163  *
4164  * Note that many devices must be down to change the HW address, and
4165  * downing the master releases all slaves.  We can make bonds full of
4166  * bonding devices to test this, however.
4167  */
4168 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4169 {
4170         struct bonding *bond = netdev_priv(bond_dev);
4171         struct sockaddr *sa = addr, tmp_sa;
4172         struct slave *slave, *stop_at;
4173         int res = 0;
4174         int i;
4175
4176         if (bond->params.mode == BOND_MODE_ALB)
4177                 return bond_alb_set_mac_address(bond_dev, addr);
4178
4179
4180         pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4181
4182         /*
4183          * If fail_over_mac is set to active, do nothing and return
4184          * success.  Returning an error causes ifenslave to fail.
4185          */
4186         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4187                 return 0;
4188
4189         if (!is_valid_ether_addr(sa->sa_data))
4190                 return -EADDRNOTAVAIL;
4191
4192         /* Can't hold bond->lock with bh disabled here since
4193          * some base drivers panic. On the other hand we can't
4194          * hold bond->lock without bh disabled because we'll
4195          * deadlock. The only solution is to rely on the fact
4196          * that we're under rtnl_lock here, and the slaves
4197          * list won't change. This doesn't solve the problem
4198          * of setting the slave's hw address while it is
4199          * transmitting, but the assumption is that the base
4200          * driver can handle that.
4201          *
4202          * TODO: figure out a way to safely iterate the slaves
4203          * list, but without holding a lock around the actual
4204          * call to the base driver.
4205          */
4206
4207         bond_for_each_slave(bond, slave, i) {
4208                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4209                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4210
4211                 if (slave_ops->ndo_set_mac_address == NULL) {
4212                         res = -EOPNOTSUPP;
4213                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4214                         goto unwind;
4215                 }
4216
4217                 res = dev_set_mac_address(slave->dev, addr);
4218                 if (res) {
4219                         /* TODO: consider downing the slave
4220                          * and retry ?
4221                          * User should expect communications
4222                          * breakage anyway until ARP finish
4223                          * updating, so...
4224                          */
4225                         pr_debug("err %d %s\n", res, slave->dev->name);
4226                         goto unwind;
4227                 }
4228         }
4229
4230         /* success */
4231         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4232         return 0;
4233
4234 unwind:
4235         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4236         tmp_sa.sa_family = bond_dev->type;
4237
4238         /* unwind from head to the slave that failed */
4239         stop_at = slave;
4240         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4241                 int tmp_res;
4242
4243                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4244                 if (tmp_res) {
4245                         pr_debug("unwind err %d dev %s\n", tmp_res,
4246                                 slave->dev->name);
4247                 }
4248         }
4249
4250         return res;
4251 }
4252
4253 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4254 {
4255         struct bonding *bond = netdev_priv(bond_dev);
4256         struct slave *slave, *start_at;
4257         int i, slave_no, res = 1;
4258
4259         read_lock(&bond->lock);
4260
4261         if (!BOND_IS_OK(bond))
4262                 goto out;
4263
4264         /*
4265          * Concurrent TX may collide on rr_tx_counter; we accept that
4266          * as being rare enough not to justify using an atomic op here
4267          */
4268         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4269
4270         bond_for_each_slave(bond, slave, i) {
4271                 slave_no--;
4272                 if (slave_no < 0)
4273                         break;
4274         }
4275
4276         start_at = slave;
4277         bond_for_each_slave_from(bond, slave, i, start_at) {
4278                 if (IS_UP(slave->dev) &&
4279                     (slave->link == BOND_LINK_UP) &&
4280                     (slave->state == BOND_STATE_ACTIVE)) {
4281                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4282                         break;
4283                 }
4284         }
4285
4286 out:
4287         if (res) {
4288                 /* no suitable interface, frame not sent */
4289                 dev_kfree_skb(skb);
4290         }
4291         read_unlock(&bond->lock);
4292         return NETDEV_TX_OK;
4293 }
4294
4295
4296 /*
4297  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4298  * the bond has a usable interface.
4299  */
4300 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4301 {
4302         struct bonding *bond = netdev_priv(bond_dev);
4303         int res = 1;
4304
4305         read_lock(&bond->lock);
4306         read_lock(&bond->curr_slave_lock);
4307
4308         if (!BOND_IS_OK(bond))
4309                 goto out;
4310
4311         if (!bond->curr_active_slave)
4312                 goto out;
4313
4314         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4315
4316 out:
4317         if (res)
4318                 /* no suitable interface, frame not sent */
4319                 dev_kfree_skb(skb);
4320
4321         read_unlock(&bond->curr_slave_lock);
4322         read_unlock(&bond->lock);
4323         return NETDEV_TX_OK;
4324 }
4325
4326 /*
4327  * In bond_xmit_xor() , we determine the output device by using a pre-
4328  * determined xmit_hash_policy(), If the selected device is not enabled,
4329  * find the next active slave.
4330  */
4331 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4332 {
4333         struct bonding *bond = netdev_priv(bond_dev);
4334         struct slave *slave, *start_at;
4335         int slave_no;
4336         int i;
4337         int res = 1;
4338
4339         read_lock(&bond->lock);
4340
4341         if (!BOND_IS_OK(bond))
4342                 goto out;
4343
4344         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4345
4346         bond_for_each_slave(bond, slave, i) {
4347                 slave_no--;
4348                 if (slave_no < 0)
4349                         break;
4350         }
4351
4352         start_at = slave;
4353
4354         bond_for_each_slave_from(bond, slave, i, start_at) {
4355                 if (IS_UP(slave->dev) &&
4356                     (slave->link == BOND_LINK_UP) &&
4357                     (slave->state == BOND_STATE_ACTIVE)) {
4358                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4359                         break;
4360                 }
4361         }
4362
4363 out:
4364         if (res) {
4365                 /* no suitable interface, frame not sent */
4366                 dev_kfree_skb(skb);
4367         }
4368         read_unlock(&bond->lock);
4369         return NETDEV_TX_OK;
4370 }
4371
4372 /*
4373  * in broadcast mode, we send everything to all usable interfaces.
4374  */
4375 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4376 {
4377         struct bonding *bond = netdev_priv(bond_dev);
4378         struct slave *slave, *start_at;
4379         struct net_device *tx_dev = NULL;
4380         int i;
4381         int res = 1;
4382
4383         read_lock(&bond->lock);
4384
4385         if (!BOND_IS_OK(bond))
4386                 goto out;
4387
4388         read_lock(&bond->curr_slave_lock);
4389         start_at = bond->curr_active_slave;
4390         read_unlock(&bond->curr_slave_lock);
4391
4392         if (!start_at)
4393                 goto out;
4394
4395         bond_for_each_slave_from(bond, slave, i, start_at) {
4396                 if (IS_UP(slave->dev) &&
4397                     (slave->link == BOND_LINK_UP) &&
4398                     (slave->state == BOND_STATE_ACTIVE)) {
4399                         if (tx_dev) {
4400                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4401                                 if (!skb2) {
4402                                         pr_err(DRV_NAME
4403                                                ": %s: Error: bond_xmit_broadcast(): "
4404                                                "skb_clone() failed\n",
4405                                                bond_dev->name);
4406                                         continue;
4407                                 }
4408
4409                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4410                                 if (res) {
4411                                         dev_kfree_skb(skb2);
4412                                         continue;
4413                                 }
4414                         }
4415                         tx_dev = slave->dev;
4416                 }
4417         }
4418
4419         if (tx_dev)
4420                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4421
4422 out:
4423         if (res)
4424                 /* no suitable interface, frame not sent */
4425                 dev_kfree_skb(skb);
4426
4427         /* frame sent to all suitable interfaces */
4428         read_unlock(&bond->lock);
4429         return NETDEV_TX_OK;
4430 }
4431
4432 /*------------------------- Device initialization ---------------------------*/
4433
4434 static void bond_set_xmit_hash_policy(struct bonding *bond)
4435 {
4436         switch (bond->params.xmit_policy) {
4437         case BOND_XMIT_POLICY_LAYER23:
4438                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4439                 break;
4440         case BOND_XMIT_POLICY_LAYER34:
4441                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4442                 break;
4443         case BOND_XMIT_POLICY_LAYER2:
4444         default:
4445                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4446                 break;
4447         }
4448 }
4449
4450 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4451 {
4452         const struct bonding *bond = netdev_priv(dev);
4453
4454         switch (bond->params.mode) {
4455         case BOND_MODE_ROUNDROBIN:
4456                 return bond_xmit_roundrobin(skb, dev);
4457         case BOND_MODE_ACTIVEBACKUP:
4458                 return bond_xmit_activebackup(skb, dev);
4459         case BOND_MODE_XOR:
4460                 return bond_xmit_xor(skb, dev);
4461         case BOND_MODE_BROADCAST:
4462                 return bond_xmit_broadcast(skb, dev);
4463         case BOND_MODE_8023AD:
4464                 return bond_3ad_xmit_xor(skb, dev);
4465         case BOND_MODE_ALB:
4466         case BOND_MODE_TLB:
4467                 return bond_alb_xmit(skb, dev);
4468         default:
4469                 /* Should never happen, mode already checked */
4470                 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4471                      dev->name, bond->params.mode);
4472                 WARN_ON_ONCE(1);
4473                 dev_kfree_skb(skb);
4474                 return NETDEV_TX_OK;
4475         }
4476 }
4477
4478
4479 /*
4480  * set bond mode specific net device operations
4481  */
4482 void bond_set_mode_ops(struct bonding *bond, int mode)
4483 {
4484         struct net_device *bond_dev = bond->dev;
4485
4486         switch (mode) {
4487         case BOND_MODE_ROUNDROBIN:
4488                 break;
4489         case BOND_MODE_ACTIVEBACKUP:
4490                 break;
4491         case BOND_MODE_XOR:
4492                 bond_set_xmit_hash_policy(bond);
4493                 break;
4494         case BOND_MODE_BROADCAST:
4495                 break;
4496         case BOND_MODE_8023AD:
4497                 bond_set_master_3ad_flags(bond);
4498                 bond_set_xmit_hash_policy(bond);
4499                 break;
4500         case BOND_MODE_ALB:
4501                 bond_set_master_alb_flags(bond);
4502                 /* FALLTHRU */
4503         case BOND_MODE_TLB:
4504                 break;
4505         default:
4506                 /* Should never happen, mode already checked */
4507                 pr_err(DRV_NAME
4508                        ": %s: Error: Unknown bonding mode %d\n",
4509                        bond_dev->name,
4510                        mode);
4511                 break;
4512         }
4513 }
4514
4515 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4516                                     struct ethtool_drvinfo *drvinfo)
4517 {
4518         strncpy(drvinfo->driver, DRV_NAME, 32);
4519         strncpy(drvinfo->version, DRV_VERSION, 32);
4520         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4521 }
4522
4523 static const struct ethtool_ops bond_ethtool_ops = {
4524         .get_drvinfo            = bond_ethtool_get_drvinfo,
4525         .get_link               = ethtool_op_get_link,
4526         .get_tx_csum            = ethtool_op_get_tx_csum,
4527         .get_sg                 = ethtool_op_get_sg,
4528         .get_tso                = ethtool_op_get_tso,
4529         .get_ufo                = ethtool_op_get_ufo,
4530         .get_flags              = ethtool_op_get_flags,
4531 };
4532
4533 static const struct net_device_ops bond_netdev_ops = {
4534         .ndo_init               = bond_init,
4535         .ndo_uninit             = bond_uninit,
4536         .ndo_open               = bond_open,
4537         .ndo_stop               = bond_close,
4538         .ndo_start_xmit         = bond_start_xmit,
4539         .ndo_get_stats          = bond_get_stats,
4540         .ndo_do_ioctl           = bond_do_ioctl,
4541         .ndo_set_multicast_list = bond_set_multicast_list,
4542         .ndo_change_mtu         = bond_change_mtu,
4543         .ndo_set_mac_address    = bond_set_mac_address,
4544         .ndo_neigh_setup        = bond_neigh_setup,
4545         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4546         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4547         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4548 };
4549
4550 static void bond_setup(struct net_device *bond_dev)
4551 {
4552         struct bonding *bond = netdev_priv(bond_dev);
4553
4554         /* initialize rwlocks */
4555         rwlock_init(&bond->lock);
4556         rwlock_init(&bond->curr_slave_lock);
4557
4558         bond->params = bonding_defaults;
4559
4560         /* Initialize pointers */
4561         bond->dev = bond_dev;
4562         INIT_LIST_HEAD(&bond->vlan_list);
4563
4564         /* Initialize the device entry points */
4565         ether_setup(bond_dev);
4566         bond_dev->netdev_ops = &bond_netdev_ops;
4567         bond_dev->ethtool_ops = &bond_ethtool_ops;
4568         bond_set_mode_ops(bond, bond->params.mode);
4569
4570         bond_dev->destructor = free_netdev;
4571
4572         /* Initialize the device options */
4573         bond_dev->tx_queue_len = 0;
4574         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4575         bond_dev->priv_flags |= IFF_BONDING;
4576         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4577
4578         if (bond->params.arp_interval)
4579                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4580
4581         /* At first, we block adding VLANs. That's the only way to
4582          * prevent problems that occur when adding VLANs over an
4583          * empty bond. The block will be removed once non-challenged
4584          * slaves are enslaved.
4585          */
4586         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4587
4588         /* don't acquire bond device's netif_tx_lock when
4589          * transmitting */
4590         bond_dev->features |= NETIF_F_LLTX;
4591
4592         /* By default, we declare the bond to be fully
4593          * VLAN hardware accelerated capable. Special
4594          * care is taken in the various xmit functions
4595          * when there are slaves that are not hw accel
4596          * capable
4597          */
4598         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4599                                NETIF_F_HW_VLAN_RX |
4600                                NETIF_F_HW_VLAN_FILTER);
4601
4602 }
4603
4604 static void bond_work_cancel_all(struct bonding *bond)
4605 {
4606         write_lock_bh(&bond->lock);
4607         bond->kill_timers = 1;
4608         write_unlock_bh(&bond->lock);
4609
4610         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4611                 cancel_delayed_work(&bond->mii_work);
4612
4613         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4614                 cancel_delayed_work(&bond->arp_work);
4615
4616         if (bond->params.mode == BOND_MODE_ALB &&
4617             delayed_work_pending(&bond->alb_work))
4618                 cancel_delayed_work(&bond->alb_work);
4619
4620         if (bond->params.mode == BOND_MODE_8023AD &&
4621             delayed_work_pending(&bond->ad_work))
4622                 cancel_delayed_work(&bond->ad_work);
4623 }
4624
4625 /* De-initialize device specific data.
4626  * Caller must hold rtnl_lock.
4627  */
4628 static void bond_deinit(struct net_device *bond_dev)
4629 {
4630         struct bonding *bond = netdev_priv(bond_dev);
4631
4632         list_del(&bond->bond_list);
4633
4634         bond_work_cancel_all(bond);
4635
4636         bond_remove_proc_entry(bond);
4637 }
4638
4639 /* Unregister and free all bond devices.
4640  * Caller must hold rtnl_lock.
4641  */
4642 static void bond_free_all(void)
4643 {
4644         struct bonding *bond, *nxt;
4645
4646         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4647                 struct net_device *bond_dev = bond->dev;
4648
4649                 bond_work_cancel_all(bond);
4650                 /* Release the bonded slaves */
4651                 bond_release_all(bond_dev);
4652                 unregister_netdevice(bond_dev);
4653         }
4654
4655         bond_destroy_proc_dir();
4656 }
4657
4658 /*------------------------- Module initialization ---------------------------*/
4659
4660 /*
4661  * Convert string input module parms.  Accept either the
4662  * number of the mode or its string name.  A bit complicated because
4663  * some mode names are substrings of other names, and calls from sysfs
4664  * may have whitespace in the name (trailing newlines, for example).
4665  */
4666 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4667 {
4668         int modeint = -1, i, rv;
4669         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4670
4671         for (p = (char *)buf; *p; p++)
4672                 if (!(isdigit(*p) || isspace(*p)))
4673                         break;
4674
4675         if (*p)
4676                 rv = sscanf(buf, "%20s", modestr);
4677         else
4678                 rv = sscanf(buf, "%d", &modeint);
4679
4680         if (!rv)
4681                 return -1;
4682
4683         for (i = 0; tbl[i].modename; i++) {
4684                 if (modeint == tbl[i].mode)
4685                         return tbl[i].mode;
4686                 if (strcmp(modestr, tbl[i].modename) == 0)
4687                         return tbl[i].mode;
4688         }
4689
4690         return -1;
4691 }
4692
4693 static int bond_check_params(struct bond_params *params)
4694 {
4695         int arp_validate_value, fail_over_mac_value;
4696
4697         /*
4698          * Convert string parameters.
4699          */
4700         if (mode) {
4701                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4702                 if (bond_mode == -1) {
4703                         pr_err(DRV_NAME
4704                                ": Error: Invalid bonding mode \"%s\"\n",
4705                                mode == NULL ? "NULL" : mode);
4706                         return -EINVAL;
4707                 }
4708         }
4709
4710         if (xmit_hash_policy) {
4711                 if ((bond_mode != BOND_MODE_XOR) &&
4712                     (bond_mode != BOND_MODE_8023AD)) {
4713                         pr_info(DRV_NAME
4714                                ": xor_mode param is irrelevant in mode %s\n",
4715                                bond_mode_name(bond_mode));
4716                 } else {
4717                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4718                                                         xmit_hashtype_tbl);
4719                         if (xmit_hashtype == -1) {
4720                                 pr_err(DRV_NAME
4721                                        ": Error: Invalid xmit_hash_policy \"%s\"\n",
4722                                        xmit_hash_policy == NULL ? "NULL" :
4723                                        xmit_hash_policy);
4724                                 return -EINVAL;
4725                         }
4726                 }
4727         }
4728
4729         if (lacp_rate) {
4730                 if (bond_mode != BOND_MODE_8023AD) {
4731                         pr_info(DRV_NAME
4732                                ": lacp_rate param is irrelevant in mode %s\n",
4733                                bond_mode_name(bond_mode));
4734                 } else {
4735                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4736                         if (lacp_fast == -1) {
4737                                 pr_err(DRV_NAME
4738                                        ": Error: Invalid lacp rate \"%s\"\n",
4739                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4740                                 return -EINVAL;
4741                         }
4742                 }
4743         }
4744
4745         if (ad_select) {
4746                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4747                 if (params->ad_select == -1) {
4748                         pr_err(DRV_NAME
4749                                ": Error: Invalid ad_select \"%s\"\n",
4750                                ad_select == NULL ? "NULL" : ad_select);
4751                         return -EINVAL;
4752                 }
4753
4754                 if (bond_mode != BOND_MODE_8023AD) {
4755                         pr_warning(DRV_NAME
4756                                ": ad_select param only affects 802.3ad mode\n");
4757                 }
4758         } else {
4759                 params->ad_select = BOND_AD_STABLE;
4760         }
4761
4762         if (max_bonds < 0 || max_bonds > INT_MAX) {
4763                 pr_warning(DRV_NAME
4764                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4765                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4766                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4767                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4768         }
4769
4770         if (miimon < 0) {
4771                 pr_warning(DRV_NAME
4772                        ": Warning: miimon module parameter (%d), "
4773                        "not in range 0-%d, so it was reset to %d\n",
4774                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4775                 miimon = BOND_LINK_MON_INTERV;
4776         }
4777
4778         if (updelay < 0) {
4779                 pr_warning(DRV_NAME
4780                        ": Warning: updelay module parameter (%d), "
4781                        "not in range 0-%d, so it was reset to 0\n",
4782                        updelay, INT_MAX);
4783                 updelay = 0;
4784         }
4785
4786         if (downdelay < 0) {
4787                 pr_warning(DRV_NAME
4788                        ": Warning: downdelay module parameter (%d), "
4789                        "not in range 0-%d, so it was reset to 0\n",
4790                        downdelay, INT_MAX);
4791                 downdelay = 0;
4792         }
4793
4794         if ((use_carrier != 0) && (use_carrier != 1)) {
4795                 pr_warning(DRV_NAME
4796                        ": Warning: use_carrier module parameter (%d), "
4797                        "not of valid value (0/1), so it was set to 1\n",
4798                        use_carrier);
4799                 use_carrier = 1;
4800         }
4801
4802         if (num_grat_arp < 0 || num_grat_arp > 255) {
4803                 pr_warning(DRV_NAME
4804                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4805                        "was reset to 1 \n", num_grat_arp);
4806                 num_grat_arp = 1;
4807         }
4808
4809         if (num_unsol_na < 0 || num_unsol_na > 255) {
4810                 pr_warning(DRV_NAME
4811                        ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4812                        "was reset to 1 \n", num_unsol_na);
4813                 num_unsol_na = 1;
4814         }
4815
4816         /* reset values for 802.3ad */
4817         if (bond_mode == BOND_MODE_8023AD) {
4818                 if (!miimon) {
4819                         pr_warning(DRV_NAME
4820                                ": Warning: miimon must be specified, "
4821                                "otherwise bonding will not detect link "
4822                                "failure, speed and duplex which are "
4823                                "essential for 802.3ad operation\n");
4824                         pr_warning("Forcing miimon to 100msec\n");
4825                         miimon = 100;
4826                 }
4827         }
4828
4829         /* reset values for TLB/ALB */
4830         if ((bond_mode == BOND_MODE_TLB) ||
4831             (bond_mode == BOND_MODE_ALB)) {
4832                 if (!miimon) {
4833                         pr_warning(DRV_NAME
4834                                ": Warning: miimon must be specified, "
4835                                "otherwise bonding will not detect link "
4836                                "failure and link speed which are essential "
4837                                "for TLB/ALB load balancing\n");
4838                         pr_warning("Forcing miimon to 100msec\n");
4839                         miimon = 100;
4840                 }
4841         }
4842
4843         if (bond_mode == BOND_MODE_ALB) {
4844                 pr_notice(DRV_NAME
4845                        ": In ALB mode you might experience client "
4846                        "disconnections upon reconnection of a link if the "
4847                        "bonding module updelay parameter (%d msec) is "
4848                        "incompatible with the forwarding delay time of the "
4849                        "switch\n",
4850                        updelay);
4851         }
4852
4853         if (!miimon) {
4854                 if (updelay || downdelay) {
4855                         /* just warn the user the up/down delay will have
4856                          * no effect since miimon is zero...
4857                          */
4858                         pr_warning(DRV_NAME
4859                                ": Warning: miimon module parameter not set "
4860                                "and updelay (%d) or downdelay (%d) module "
4861                                "parameter is set; updelay and downdelay have "
4862                                "no effect unless miimon is set\n",
4863                                updelay, downdelay);
4864                 }
4865         } else {
4866                 /* don't allow arp monitoring */
4867                 if (arp_interval) {
4868                         pr_warning(DRV_NAME
4869                                ": Warning: miimon (%d) and arp_interval (%d) "
4870                                "can't be used simultaneously, disabling ARP "
4871                                "monitoring\n",
4872                                miimon, arp_interval);
4873                         arp_interval = 0;
4874                 }
4875
4876                 if ((updelay % miimon) != 0) {
4877                         pr_warning(DRV_NAME
4878                                ": Warning: updelay (%d) is not a multiple "
4879                                "of miimon (%d), updelay rounded to %d ms\n",
4880                                updelay, miimon, (updelay / miimon) * miimon);
4881                 }
4882
4883                 updelay /= miimon;
4884
4885                 if ((downdelay % miimon) != 0) {
4886                         pr_warning(DRV_NAME
4887                                ": Warning: downdelay (%d) is not a multiple "
4888                                "of miimon (%d), downdelay rounded to %d ms\n",
4889                                downdelay, miimon,
4890                                (downdelay / miimon) * miimon);
4891                 }
4892
4893                 downdelay /= miimon;
4894         }
4895
4896         if (arp_interval < 0) {
4897                 pr_warning(DRV_NAME
4898                        ": Warning: arp_interval module parameter (%d) "
4899                        ", not in range 0-%d, so it was reset to %d\n",
4900                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4901                 arp_interval = BOND_LINK_ARP_INTERV;
4902         }
4903
4904         for (arp_ip_count = 0;
4905              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4906              arp_ip_count++) {
4907                 /* not complete check, but should be good enough to
4908                    catch mistakes */
4909                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4910                         pr_warning(DRV_NAME
4911                                ": Warning: bad arp_ip_target module parameter "
4912                                "(%s), ARP monitoring will not be performed\n",
4913                                arp_ip_target[arp_ip_count]);
4914                         arp_interval = 0;
4915                 } else {
4916                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4917                         arp_target[arp_ip_count] = ip;
4918                 }
4919         }
4920
4921         if (arp_interval && !arp_ip_count) {
4922                 /* don't allow arping if no arp_ip_target given... */
4923                 pr_warning(DRV_NAME
4924                        ": Warning: arp_interval module parameter (%d) "
4925                        "specified without providing an arp_ip_target "
4926                        "parameter, arp_interval was reset to 0\n",
4927                        arp_interval);
4928                 arp_interval = 0;
4929         }
4930
4931         if (arp_validate) {
4932                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4933                         pr_err(DRV_NAME
4934                                ": arp_validate only supported in active-backup mode\n");
4935                         return -EINVAL;
4936                 }
4937                 if (!arp_interval) {
4938                         pr_err(DRV_NAME
4939                                ": arp_validate requires arp_interval\n");
4940                         return -EINVAL;
4941                 }
4942
4943                 arp_validate_value = bond_parse_parm(arp_validate,
4944                                                      arp_validate_tbl);
4945                 if (arp_validate_value == -1) {
4946                         pr_err(DRV_NAME
4947                                ": Error: invalid arp_validate \"%s\"\n",
4948                                arp_validate == NULL ? "NULL" : arp_validate);
4949                         return -EINVAL;
4950                 }
4951         } else
4952                 arp_validate_value = 0;
4953
4954         if (miimon) {
4955                 pr_info(DRV_NAME
4956                        ": MII link monitoring set to %d ms\n",
4957                        miimon);
4958         } else if (arp_interval) {
4959                 int i;
4960
4961                 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4962                        " validate %s, with %d target(s):",
4963                        arp_interval,
4964                        arp_validate_tbl[arp_validate_value].modename,
4965                        arp_ip_count);
4966
4967                 for (i = 0; i < arp_ip_count; i++)
4968                         pr_info(" %s", arp_ip_target[i]);
4969
4970                 pr_info("\n");
4971
4972         } else if (max_bonds) {
4973                 /* miimon and arp_interval not set, we need one so things
4974                  * work as expected, see bonding.txt for details
4975                  */
4976                 pr_warning(DRV_NAME
4977                        ": Warning: either miimon or arp_interval and "
4978                        "arp_ip_target module parameters must be specified, "
4979                        "otherwise bonding will not detect link failures! see "
4980                        "bonding.txt for details.\n");
4981         }
4982
4983         if (primary && !USES_PRIMARY(bond_mode)) {
4984                 /* currently, using a primary only makes sense
4985                  * in active backup, TLB or ALB modes
4986                  */
4987                 pr_warning(DRV_NAME
4988                        ": Warning: %s primary device specified but has no "
4989                        "effect in %s mode\n",
4990                        primary, bond_mode_name(bond_mode));
4991                 primary = NULL;
4992         }
4993
4994         if (fail_over_mac) {
4995                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4996                                                       fail_over_mac_tbl);
4997                 if (fail_over_mac_value == -1) {
4998                         pr_err(DRV_NAME
4999                                ": Error: invalid fail_over_mac \"%s\"\n",
5000                                arp_validate == NULL ? "NULL" : arp_validate);
5001                         return -EINVAL;
5002                 }
5003
5004                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5005                         pr_warning(DRV_NAME
5006                                ": Warning: fail_over_mac only affects "
5007                                "active-backup mode.\n");
5008         } else {
5009                 fail_over_mac_value = BOND_FOM_NONE;
5010         }
5011
5012         /* fill params struct with the proper values */
5013         params->mode = bond_mode;
5014         params->xmit_policy = xmit_hashtype;
5015         params->miimon = miimon;
5016         params->num_grat_arp = num_grat_arp;
5017         params->num_unsol_na = num_unsol_na;
5018         params->arp_interval = arp_interval;
5019         params->arp_validate = arp_validate_value;
5020         params->updelay = updelay;
5021         params->downdelay = downdelay;
5022         params->use_carrier = use_carrier;
5023         params->lacp_fast = lacp_fast;
5024         params->primary[0] = 0;
5025         params->fail_over_mac = fail_over_mac_value;
5026
5027         if (primary) {
5028                 strncpy(params->primary, primary, IFNAMSIZ);
5029                 params->primary[IFNAMSIZ - 1] = 0;
5030         }
5031
5032         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5033
5034         return 0;
5035 }
5036
5037 static struct lock_class_key bonding_netdev_xmit_lock_key;
5038 static struct lock_class_key bonding_netdev_addr_lock_key;
5039
5040 static void bond_set_lockdep_class_one(struct net_device *dev,
5041                                        struct netdev_queue *txq,
5042                                        void *_unused)
5043 {
5044         lockdep_set_class(&txq->_xmit_lock,
5045                           &bonding_netdev_xmit_lock_key);
5046 }
5047
5048 static void bond_set_lockdep_class(struct net_device *dev)
5049 {
5050         lockdep_set_class(&dev->addr_list_lock,
5051                           &bonding_netdev_addr_lock_key);
5052         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5053 }
5054
5055 /*
5056  * Called from registration process
5057  */
5058 static int bond_init(struct net_device *bond_dev)
5059 {
5060         struct bonding *bond = netdev_priv(bond_dev);
5061
5062         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5063
5064         bond->wq = create_singlethread_workqueue(bond_dev->name);
5065         if (!bond->wq)
5066                 return -ENOMEM;
5067
5068         bond_set_lockdep_class(bond_dev);
5069
5070         netif_carrier_off(bond_dev);
5071
5072         bond_create_proc_entry(bond);
5073         list_add_tail(&bond->bond_list, &bond_dev_list);
5074
5075         return 0;
5076 }
5077
5078 /* Create a new bond based on the specified name and bonding parameters.
5079  * If name is NULL, obtain a suitable "bond%d" name for us.
5080  * Caller must NOT hold rtnl_lock; we need to release it here before we
5081  * set up our sysfs entries.
5082  */
5083 int bond_create(const char *name)
5084 {
5085         struct net_device *bond_dev;
5086         int res;
5087
5088         rtnl_lock();
5089         /* Check to see if the bond already exists. */
5090         /* FIXME: pass netns from caller */
5091         if (name && __dev_get_by_name(&init_net, name)) {
5092                 pr_err(DRV_NAME ": cannot add bond %s; already exists\n",
5093                        name);
5094                 res = -EEXIST;
5095                 goto out_rtnl;
5096         }
5097
5098         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5099                                 bond_setup);
5100         if (!bond_dev) {
5101                 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5102                        name);
5103                 res = -ENOMEM;
5104                 goto out_rtnl;
5105         }
5106
5107         if (!name) {
5108                 res = dev_alloc_name(bond_dev, "bond%d");
5109                 if (res < 0)
5110                         goto out_netdev;
5111         }
5112
5113         res = register_netdevice(bond_dev);
5114         if (res < 0)
5115                 goto out_bond;
5116
5117         res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5118         if (res < 0)
5119                 goto out_unreg;
5120
5121         rtnl_unlock();
5122         return 0;
5123
5124 out_unreg:
5125         unregister_netdevice(bond_dev);
5126 out_bond:
5127         bond_deinit(bond_dev);
5128 out_netdev:
5129         free_netdev(bond_dev);
5130 out_rtnl:
5131         rtnl_unlock();
5132         return res;
5133 }
5134
5135 static int __init bonding_init(void)
5136 {
5137         int i;
5138         int res;
5139
5140         pr_info("%s", version);
5141
5142         res = bond_check_params(&bonding_defaults);
5143         if (res)
5144                 goto out;
5145
5146         bond_create_proc_dir();
5147
5148         for (i = 0; i < max_bonds; i++) {
5149                 res = bond_create(NULL);
5150                 if (res)
5151                         goto err;
5152         }
5153
5154         res = bond_create_sysfs();
5155         if (res)
5156                 goto err;
5157
5158         register_netdevice_notifier(&bond_netdev_notifier);
5159         register_inetaddr_notifier(&bond_inetaddr_notifier);
5160         bond_register_ipv6_notifier();
5161
5162         goto out;
5163 err:
5164         rtnl_lock();
5165         bond_free_all();
5166         rtnl_unlock();
5167 out:
5168         return res;
5169
5170 }
5171
5172 static void __exit bonding_exit(void)
5173 {
5174         unregister_netdevice_notifier(&bond_netdev_notifier);
5175         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5176         bond_unregister_ipv6_notifier();
5177
5178         bond_destroy_sysfs();
5179
5180         rtnl_lock();
5181         bond_free_all();
5182         rtnl_unlock();
5183 }
5184
5185 module_init(bonding_init);
5186 module_exit(bonding_exit);
5187 MODULE_LICENSE("GPL");
5188 MODULE_VERSION(DRV_VERSION);
5189 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5190 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");