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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <net/route.h>
  79 #include <net/net_namespace.h>
  80 #include <net/netns/generic.h>
  81 #include "bonding.h"
  82 #include "bond_3ad.h"
  83 #include "bond_alb.h"
  84
  85 /*---------------------------- Module parameters ----------------------------*/
  86
  87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  88 #define BOND_LINK_MON_INTERV    0
  89 #define BOND_LINK_ARP_INTERV    0
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int num_grat_arp = 1;
  93 static int num_unsol_na = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static struct bond_params bonding_defaults;
 109
 110 module_param(max_bonds, int, 0);
 111 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 112 module_param(num_grat_arp, int, 0644);
 113 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 114 module_param(num_unsol_na, int, 0644);
 115 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 116 module_param(miimon, int, 0);
 117 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 118 module_param(updelay, int, 0);
 119 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 120 module_param(downdelay, int, 0);
 121 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 122                             "in milliseconds");
 123 module_param(use_carrier, int, 0);
 124 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 125                               "0 for off, 1 for on (default)");
 126 module_param(mode, charp, 0);
 127 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 128                        "1 for active-backup, 2 for balance-xor, "
 129                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 130                        "6 for balance-alb");
 131 module_param(primary, charp, 0);
 132 MODULE_PARM_DESC(primary, "Primary network device to use");
 133 module_param(primary_reselect, charp, 0);
 134 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 135                                    "once it comes up; "
 136                                    "0 for always (default), "
 137                                    "1 for only if speed of primary is "
 138                                    "better, "
 139                                    "2 for only on active slave "
 140                                    "failure");
 141 module_param(lacp_rate, charp, 0);
 142 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 143                             "(slow/fast)");
 144 module_param(ad_select, charp, 0);
 145 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 146 module_param(xmit_hash_policy, charp, 0);
 147 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 148                                    ", 1 for layer 3+4");
 149 module_param(arp_interval, int, 0);
 150 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 151 module_param_array(arp_ip_target, charp, NULL, 0);
 152 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 153 module_param(arp_validate, charp, 0);
 154 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 155 module_param(fail_over_mac, charp, 0);
 156 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 157
 158 /*----------------------------- Global variables ----------------------------*/
 159
 160 static const char * const version =
 161         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 162
 163 int bond_net_id __read_mostly;
 164
 165 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 166 static int arp_ip_count;
 167 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 168 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 169 static int lacp_fast;
 170
 171
 172 const struct bond_parm_tbl bond_lacp_tbl[] = {
 173 {       "slow",         AD_LACP_SLOW},
 174 {       "fast",         AD_LACP_FAST},
 175 {       NULL,           -1},
 176 };
 177
 178 const struct bond_parm_tbl bond_mode_tbl[] = {
 179 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 180 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 181 {       "balance-xor",          BOND_MODE_XOR},
 182 {       "broadcast",            BOND_MODE_BROADCAST},
 183 {       "802.3ad",              BOND_MODE_8023AD},
 184 {       "balance-tlb",          BOND_MODE_TLB},
 185 {       "balance-alb",          BOND_MODE_ALB},
 186 {       NULL,                   -1},
 187 };
 188
 189 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 190 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 191 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 192 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 193 {       NULL,                   -1},
 194 };
 195
 196 const struct bond_parm_tbl arp_validate_tbl[] = {
 197 {       "none",                 BOND_ARP_VALIDATE_NONE},
 198 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 199 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 200 {       "all",                  BOND_ARP_VALIDATE_ALL},
 201 {       NULL,                   -1},
 202 };
 203
 204 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 205 {       "none",                 BOND_FOM_NONE},
 206 {       "active",               BOND_FOM_ACTIVE},
 207 {       "follow",               BOND_FOM_FOLLOW},
 208 {       NULL,                   -1},
 209 };
 210
 211 const struct bond_parm_tbl pri_reselect_tbl[] = {
 212 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 213 {       "better",               BOND_PRI_RESELECT_BETTER},
 214 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 215 {       NULL,                   -1},
 216 };
 217
 218 struct bond_parm_tbl ad_select_tbl[] = {
 219 {       "stable",       BOND_AD_STABLE},
 220 {       "bandwidth",    BOND_AD_BANDWIDTH},
 221 {       "count",        BOND_AD_COUNT},
 222 {       NULL,           -1},
 223 };
 224
 225 /*-------------------------- Forward declarations ---------------------------*/
 226
 227 static void bond_send_gratuitous_arp(struct bonding *bond);
 228 static int bond_init(struct net_device *bond_dev);
 229 static void bond_uninit(struct net_device *bond_dev);
 230
 231 /*---------------------------- General routines -----------------------------*/
 232
 233 static const char *bond_mode_name(int mode)
 234 {
 235         static const char *names[] = {
 236                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 237                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 238                 [BOND_MODE_XOR] = "load balancing (xor)",
 239                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 240                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 241                 [BOND_MODE_TLB] = "transmit load balancing",
 242                 [BOND_MODE_ALB] = "adaptive load balancing",
 243         };
 244
 245         if (mode < 0 || mode > BOND_MODE_ALB)
 246                 return "unknown";
 247
 248         return names[mode];
 249 }
 250
 251 /*---------------------------------- VLAN -----------------------------------*/
 252
 253 /**
 254  * bond_add_vlan - add a new vlan id on bond
 255  * @bond: bond that got the notification
 256  * @vlan_id: the vlan id to add
 257  *
 258  * Returns -ENOMEM if allocation failed.
 259  */
 260 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 261 {
 262         struct vlan_entry *vlan;
 263
 264         pr_debug("bond: %s, vlan id %d\n",
 265                  (bond ? bond->dev->name : "None"), vlan_id);
 266
 267         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 268         if (!vlan)
 269                 return -ENOMEM;
 270
 271         INIT_LIST_HEAD(&vlan->vlan_list);
 272         vlan->vlan_id = vlan_id;
 273
 274         write_lock_bh(&bond->lock);
 275
 276         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 277
 278         write_unlock_bh(&bond->lock);
 279
 280         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 281
 282         return 0;
 283 }
 284
 285 /**
 286  * bond_del_vlan - delete a vlan id from bond
 287  * @bond: bond that got the notification
 288  * @vlan_id: the vlan id to delete
 289  *
 290  * returns -ENODEV if @vlan_id was not found in @bond.
 291  */
 292 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 293 {
 294         struct vlan_entry *vlan;
 295         int res = -ENODEV;
 296
 297         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 298
 299         write_lock_bh(&bond->lock);
 300
 301         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 302                 if (vlan->vlan_id == vlan_id) {
 303                         list_del(&vlan->vlan_list);
 304
 305                         if (bond_is_lb(bond))
 306                                 bond_alb_clear_vlan(bond, vlan_id);
 307
 308                         pr_debug("removed VLAN ID %d from bond %s\n",
 309                                  vlan_id, bond->dev->name);
 310
 311                         kfree(vlan);
 312
 313                         if (list_empty(&bond->vlan_list) &&
 314                             (bond->slave_cnt == 0)) {
 315                                 /* Last VLAN removed and no slaves, so
 316                                  * restore block on adding VLANs. This will
 317                                  * be removed once new slaves that are not
 318                                  * VLAN challenged will be added.
 319                                  */
 320                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 321                         }
 322
 323                         res = 0;
 324                         goto out;
 325                 }
 326         }
 327
 328         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 329                  vlan_id, bond->dev->name);
 330
 331 out:
 332         write_unlock_bh(&bond->lock);
 333         return res;
 334 }
 335
 336 /**
 337  * bond_has_challenged_slaves
 338  * @bond: the bond we're working on
 339  *
 340  * Searches the slave list. Returns 1 if a vlan challenged slave
 341  * was found, 0 otherwise.
 342  *
 343  * Assumes bond->lock is held.
 344  */
 345 static int bond_has_challenged_slaves(struct bonding *bond)
 346 {
 347         struct slave *slave;
 348         int i;
 349
 350         bond_for_each_slave(bond, slave, i) {
 351                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 352                         pr_debug("found VLAN challenged slave - %s\n",
 353                                  slave->dev->name);
 354                         return 1;
 355                 }
 356         }
 357
 358         pr_debug("no VLAN challenged slaves found\n");
 359         return 0;
 360 }
 361
 362 /**
 363  * bond_next_vlan - safely skip to the next item in the vlans list.
 364  * @bond: the bond we're working on
 365  * @curr: item we're advancing from
 366  *
 367  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 368  * or @curr->next otherwise (even if it is @curr itself again).
 369  *
 370  * Caller must hold bond->lock
 371  */
 372 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 373 {
 374         struct vlan_entry *next, *last;
 375
 376         if (list_empty(&bond->vlan_list))
 377                 return NULL;
 378
 379         if (!curr) {
 380                 next = list_entry(bond->vlan_list.next,
 381                                   struct vlan_entry, vlan_list);
 382         } else {
 383                 last = list_entry(bond->vlan_list.prev,
 384                                   struct vlan_entry, vlan_list);
 385                 if (last == curr) {
 386                         next = list_entry(bond->vlan_list.next,
 387                                           struct vlan_entry, vlan_list);
 388                 } else {
 389                         next = list_entry(curr->vlan_list.next,
 390                                           struct vlan_entry, vlan_list);
 391                 }
 392         }
 393
 394         return next;
 395 }
 396
 397 /**
 398  * bond_dev_queue_xmit - Prepare skb for xmit.
 399  *
 400  * @bond: bond device that got this skb for tx.
 401  * @skb: hw accel VLAN tagged skb to transmit
 402  * @slave_dev: slave that is supposed to xmit this skbuff
 403  *
 404  * When the bond gets an skb to transmit that is
 405  * already hardware accelerated VLAN tagged, and it
 406  * needs to relay this skb to a slave that is not
 407  * hw accel capable, the skb needs to be "unaccelerated",
 408  * i.e. strip the hwaccel tag and re-insert it as part
 409  * of the payload.
 410  */
 411 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 412                         struct net_device *slave_dev)
 413 {
 414         unsigned short uninitialized_var(vlan_id);
 415
 416         if (!list_empty(&bond->vlan_list) &&
 417             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 418             vlan_get_tag(skb, &vlan_id) == 0) {
 419                 skb->dev = slave_dev;
 420                 skb = vlan_put_tag(skb, vlan_id);
 421                 if (!skb) {
 422                         /* vlan_put_tag() frees the skb in case of error,
 423                          * so return success here so the calling functions
 424                          * won't attempt to free is again.
 425                          */
 426                         return 0;
 427                 }
 428         } else {
 429                 skb->dev = slave_dev;
 430         }
 431
 432         skb->priority = 1;
 433         dev_queue_xmit(skb);
 434
 435         return 0;
 436 }
 437
 438 /*
 439  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 440  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 441  * lock because:
 442  * a. This operation is performed in IOCTL context,
 443  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 444  * c. Holding a lock with BH disabled while directly calling a base driver
 445  *    entry point is generally a BAD idea.
 446  *
 447  * The design of synchronization/protection for this operation in the 8021q
 448  * module is good for one or more VLAN devices over a single physical device
 449  * and cannot be extended for a teaming solution like bonding, so there is a
 450  * potential race condition here where a net device from the vlan group might
 451  * be referenced (either by a base driver or the 8021q code) while it is being
 452  * removed from the system. However, it turns out we're not making matters
 453  * worse, and if it works for regular VLAN usage it will work here too.
 454 */
 455
 456 /**
 457  * bond_vlan_rx_register - Propagates registration to slaves
 458  * @bond_dev: bonding net device that got called
 459  * @grp: vlan group being registered
 460  */
 461 static void bond_vlan_rx_register(struct net_device *bond_dev,
 462                                   struct vlan_group *grp)
 463 {
 464         struct bonding *bond = netdev_priv(bond_dev);
 465         struct slave *slave;
 466         int i;
 467
 468         bond->vlgrp = grp;
 469
 470         bond_for_each_slave(bond, slave, i) {
 471                 struct net_device *slave_dev = slave->dev;
 472                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 473
 474                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 475                     slave_ops->ndo_vlan_rx_register) {
 476                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 477                 }
 478         }
 479 }
 480
 481 /**
 482  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 483  * @bond_dev: bonding net device that got called
 484  * @vid: vlan id being added
 485  */
 486 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 487 {
 488         struct bonding *bond = netdev_priv(bond_dev);
 489         struct slave *slave;
 490         int i, res;
 491
 492         bond_for_each_slave(bond, slave, i) {
 493                 struct net_device *slave_dev = slave->dev;
 494                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 495
 496                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 497                     slave_ops->ndo_vlan_rx_add_vid) {
 498                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 499                 }
 500         }
 501
 502         res = bond_add_vlan(bond, vid);
 503         if (res) {
 504                 pr_err("%s: Error: Failed to add vlan id %d\n",
 505                        bond_dev->name, vid);
 506         }
 507 }
 508
 509 /**
 510  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 511  * @bond_dev: bonding net device that got called
 512  * @vid: vlan id being removed
 513  */
 514 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 515 {
 516         struct bonding *bond = netdev_priv(bond_dev);
 517         struct slave *slave;
 518         struct net_device *vlan_dev;
 519         int i, res;
 520
 521         bond_for_each_slave(bond, slave, i) {
 522                 struct net_device *slave_dev = slave->dev;
 523                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 524
 525                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 526                     slave_ops->ndo_vlan_rx_kill_vid) {
 527                         /* Save and then restore vlan_dev in the grp array,
 528                          * since the slave's driver might clear it.
 529                          */
 530                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 531                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 532                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 533                 }
 534         }
 535
 536         res = bond_del_vlan(bond, vid);
 537         if (res) {
 538                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 539                        bond_dev->name, vid);
 540         }
 541 }
 542
 543 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 544 {
 545         struct vlan_entry *vlan;
 546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 547
 548         write_lock_bh(&bond->lock);
 549
 550         if (list_empty(&bond->vlan_list))
 551                 goto out;
 552
 553         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 554             slave_ops->ndo_vlan_rx_register)
 555                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 556
 557         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 558             !(slave_ops->ndo_vlan_rx_add_vid))
 559                 goto out;
 560
 561         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 562                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 563
 564 out:
 565         write_unlock_bh(&bond->lock);
 566 }
 567
 568 static void bond_del_vlans_from_slave(struct bonding *bond,
 569                                       struct net_device *slave_dev)
 570 {
 571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 572         struct vlan_entry *vlan;
 573         struct net_device *vlan_dev;
 574
 575         write_lock_bh(&bond->lock);
 576
 577         if (list_empty(&bond->vlan_list))
 578                 goto out;
 579
 580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 581             !(slave_ops->ndo_vlan_rx_kill_vid))
 582                 goto unreg;
 583
 584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 585                 /* Save and then restore vlan_dev in the grp array,
 586                  * since the slave's driver might clear it.
 587                  */
 588                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 589                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 590                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 591         }
 592
 593 unreg:
 594         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 595             slave_ops->ndo_vlan_rx_register)
 596                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 597
 598 out:
 599         write_unlock_bh(&bond->lock);
 600 }
 601
 602 /*------------------------------- Link status -------------------------------*/
 603
 604 /*
 605  * Set the carrier state for the master according to the state of its
 606  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 607  * do special 802.3ad magic.
 608  *
 609  * Returns zero if carrier state does not change, nonzero if it does.
 610  */
 611 static int bond_set_carrier(struct bonding *bond)
 612 {
 613         struct slave *slave;
 614         int i;
 615
 616         if (bond->slave_cnt == 0)
 617                 goto down;
 618
 619         if (bond->params.mode == BOND_MODE_8023AD)
 620                 return bond_3ad_set_carrier(bond);
 621
 622         bond_for_each_slave(bond, slave, i) {
 623                 if (slave->link == BOND_LINK_UP) {
 624                         if (!netif_carrier_ok(bond->dev)) {
 625                                 netif_carrier_on(bond->dev);
 626                                 return 1;
 627                         }
 628                         return 0;
 629                 }
 630         }
 631
 632 down:
 633         if (netif_carrier_ok(bond->dev)) {
 634                 netif_carrier_off(bond->dev);
 635                 return 1;
 636         }
 637         return 0;
 638 }
 639
 640 /*
 641  * Get link speed and duplex from the slave's base driver
 642  * using ethtool. If for some reason the call fails or the
 643  * values are invalid, fake speed and duplex to 100/Full
 644  * and return error.
 645  */
 646 static int bond_update_speed_duplex(struct slave *slave)
 647 {
 648         struct net_device *slave_dev = slave->dev;
 649         struct ethtool_cmd etool;
 650         int res;
 651
 652         /* Fake speed and duplex */
 653         slave->speed = SPEED_100;
 654         slave->duplex = DUPLEX_FULL;
 655
 656         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 657                 return -1;
 658
 659         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 660         if (res < 0)
 661                 return -1;
 662
 663         switch (etool.speed) {
 664         case SPEED_10:
 665         case SPEED_100:
 666         case SPEED_1000:
 667         case SPEED_10000:
 668                 break;
 669         default:
 670                 return -1;
 671         }
 672
 673         switch (etool.duplex) {
 674         case DUPLEX_FULL:
 675         case DUPLEX_HALF:
 676                 break;
 677         default:
 678                 return -1;
 679         }
 680
 681         slave->speed = etool.speed;
 682         slave->duplex = etool.duplex;
 683
 684         return 0;
 685 }
 686
 687 /*
 688  * if <dev> supports MII link status reporting, check its link status.
 689  *
 690  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 691  * depending upon the setting of the use_carrier parameter.
 692  *
 693  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 694  * can't tell and just pretend it is), or 0, meaning that the link is
 695  * down.
 696  *
 697  * If reporting is non-zero, instead of faking link up, return -1 if
 698  * both ETHTOOL and MII ioctls fail (meaning the device does not
 699  * support them).  If use_carrier is set, return whatever it says.
 700  * It'd be nice if there was a good way to tell if a driver supports
 701  * netif_carrier, but there really isn't.
 702  */
 703 static int bond_check_dev_link(struct bonding *bond,
 704                                struct net_device *slave_dev, int reporting)
 705 {
 706         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 707         int (*ioctl)(struct net_device *, struct ifreq *, int);
 708         struct ifreq ifr;
 709         struct mii_ioctl_data *mii;
 710
 711         if (!reporting && !netif_running(slave_dev))
 712                 return 0;
 713
 714         if (bond->params.use_carrier)
 715                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 716
 717         /* Try to get link status using Ethtool first. */
 718         if (slave_dev->ethtool_ops) {
 719                 if (slave_dev->ethtool_ops->get_link) {
 720                         u32 link;
 721
 722                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 723
 724                         return link ? BMSR_LSTATUS : 0;
 725                 }
 726         }
 727
 728         /* Ethtool can't be used, fallback to MII ioctls. */
 729         ioctl = slave_ops->ndo_do_ioctl;
 730         if (ioctl) {
 731                 /* TODO: set pointer to correct ioctl on a per team member */
 732                 /*       bases to make this more efficient. that is, once  */
 733                 /*       we determine the correct ioctl, we will always    */
 734                 /*       call it and not the others for that team          */
 735                 /*       member.                                           */
 736
 737                 /*
 738                  * We cannot assume that SIOCGMIIPHY will also read a
 739                  * register; not all network drivers (e.g., e100)
 740                  * support that.
 741                  */
 742
 743                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 744                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 745                 mii = if_mii(&ifr);
 746                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 747                         mii->reg_num = MII_BMSR;
 748                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 749                                 return mii->val_out & BMSR_LSTATUS;
 750                 }
 751         }
 752
 753         /*
 754          * If reporting, report that either there's no dev->do_ioctl,
 755          * or both SIOCGMIIREG and get_link failed (meaning that we
 756          * cannot report link status).  If not reporting, pretend
 757          * we're ok.
 758          */
 759         return reporting ? -1 : BMSR_LSTATUS;
 760 }
 761
 762 /*----------------------------- Multicast list ------------------------------*/
 763
 764 /*
 765  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
 766  */
 767 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
 768                                    const struct dev_mc_list *dmi2)
 769 {
 770         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
 771                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
 772 }
 773
 774 /*
 775  * returns dmi entry if found, NULL otherwise
 776  */
 777 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
 778                                                  struct dev_mc_list *mc_list)
 779 {
 780         struct dev_mc_list *idmi;
 781
 782         for (idmi = mc_list; idmi; idmi = idmi->next) {
 783                 if (bond_is_dmi_same(dmi, idmi))
 784                         return idmi;
 785         }
 786
 787         return NULL;
 788 }
 789
 790 /*
 791  * Push the promiscuity flag down to appropriate slaves
 792  */
 793 static int bond_set_promiscuity(struct bonding *bond, int inc)
 794 {
 795         int err = 0;
 796         if (USES_PRIMARY(bond->params.mode)) {
 797                 /* write lock already acquired */
 798                 if (bond->curr_active_slave) {
 799                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 800                                                   inc);
 801                 }
 802         } else {
 803                 struct slave *slave;
 804                 int i;
 805                 bond_for_each_slave(bond, slave, i) {
 806                         err = dev_set_promiscuity(slave->dev, inc);
 807                         if (err)
 808                                 return err;
 809                 }
 810         }
 811         return err;
 812 }
 813
 814 /*
 815  * Push the allmulti flag down to all slaves
 816  */
 817 static int bond_set_allmulti(struct bonding *bond, int inc)
 818 {
 819         int err = 0;
 820         if (USES_PRIMARY(bond->params.mode)) {
 821                 /* write lock already acquired */
 822                 if (bond->curr_active_slave) {
 823                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 824                                                inc);
 825                 }
 826         } else {
 827                 struct slave *slave;
 828                 int i;
 829                 bond_for_each_slave(bond, slave, i) {
 830                         err = dev_set_allmulti(slave->dev, inc);
 831                         if (err)
 832                                 return err;
 833                 }
 834         }
 835         return err;
 836 }
 837
 838 /*
 839  * Add a Multicast address to slaves
 840  * according to mode
 841  */
 842 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
 843 {
 844         if (USES_PRIMARY(bond->params.mode)) {
 845                 /* write lock already acquired */
 846                 if (bond->curr_active_slave)
 847                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
 848         } else {
 849                 struct slave *slave;
 850                 int i;
 851
 852                 bond_for_each_slave(bond, slave, i)
 853                         dev_mc_add(slave->dev, addr, alen, 0);
 854         }
 855 }
 856
 857 /*
 858  * Remove a multicast address from slave
 859  * according to mode
 860  */
 861 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
 862 {
 863         if (USES_PRIMARY(bond->params.mode)) {
 864                 /* write lock already acquired */
 865                 if (bond->curr_active_slave)
 866                         dev_mc_delete(bond->curr_active_slave->dev, addr,
 867                                       alen, 0);
 868         } else {
 869                 struct slave *slave;
 870                 int i;
 871                 bond_for_each_slave(bond, slave, i) {
 872                         dev_mc_delete(slave->dev, addr, alen, 0);
 873                 }
 874         }
 875 }
 876
 877
 878 /*
 879  * Retrieve the list of registered multicast addresses for the bonding
 880  * device and retransmit an IGMP JOIN request to the current active
 881  * slave.
 882  */
 883 static void bond_resend_igmp_join_requests(struct bonding *bond)
 884 {
 885         struct in_device *in_dev;
 886         struct ip_mc_list *im;
 887
 888         rcu_read_lock();
 889         in_dev = __in_dev_get_rcu(bond->dev);
 890         if (in_dev) {
 891                 for (im = in_dev->mc_list; im; im = im->next)
 892                         ip_mc_rejoin_group(im);
 893         }
 894
 895         rcu_read_unlock();
 896 }
 897
 898 /*
 899  * Totally destroys the mc_list in bond
 900  */
 901 static void bond_mc_list_destroy(struct bonding *bond)
 902 {
 903         struct dev_mc_list *dmi;
 904
 905         dmi = bond->mc_list;
 906         while (dmi) {
 907                 bond->mc_list = dmi->next;
 908                 kfree(dmi);
 909                 dmi = bond->mc_list;
 910         }
 911
 912         bond->mc_list = NULL;
 913 }
 914
 915 /*
 916  * Copy all the Multicast addresses from src to the bonding device dst
 917  */
 918 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
 919                              gfp_t gfp_flag)
 920 {
 921         struct dev_mc_list *dmi, *new_dmi;
 922
 923         for (dmi = mc_list; dmi; dmi = dmi->next) {
 924                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
 925
 926                 if (!new_dmi) {
 927                         /* FIXME: Potential memory leak !!! */
 928                         return -ENOMEM;
 929                 }
 930
 931                 new_dmi->next = bond->mc_list;
 932                 bond->mc_list = new_dmi;
 933                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
 934                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
 935                 new_dmi->dmi_users = dmi->dmi_users;
 936                 new_dmi->dmi_gusers = dmi->dmi_gusers;
 937         }
 938
 939         return 0;
 940 }
 941
 942 /*
 943  * flush all members of flush->mc_list from device dev->mc_list
 944  */
 945 static void bond_mc_list_flush(struct net_device *bond_dev,
 946                                struct net_device *slave_dev)
 947 {
 948         struct bonding *bond = netdev_priv(bond_dev);
 949         struct dev_mc_list *dmi;
 950
 951         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
 952                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
 953
 954         if (bond->params.mode == BOND_MODE_8023AD) {
 955                 /* del lacpdu mc addr from mc list */
 956                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 957
 958                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
 959         }
 960 }
 961
 962 /*--------------------------- Active slave change ---------------------------*/
 963
 964 /*
 965  * Update the mc list and multicast-related flags for the new and
 966  * old active slaves (if any) according to the multicast mode, and
 967  * promiscuous flags unconditionally.
 968  */
 969 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 970                          struct slave *old_active)
 971 {
 972         struct dev_mc_list *dmi;
 973
 974         if (!USES_PRIMARY(bond->params.mode))
 975                 /* nothing to do -  mc list is already up-to-date on
 976                  * all slaves
 977                  */
 978                 return;
 979
 980         if (old_active) {
 981                 if (bond->dev->flags & IFF_PROMISC)
 982                         dev_set_promiscuity(old_active->dev, -1);
 983
 984                 if (bond->dev->flags & IFF_ALLMULTI)
 985                         dev_set_allmulti(old_active->dev, -1);
 986
 987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
 988                         dev_mc_delete(old_active->dev, dmi->dmi_addr,
 989                                       dmi->dmi_addrlen, 0);
 990         }
 991
 992         if (new_active) {
 993                 /* FIXME: Signal errors upstream. */
 994                 if (bond->dev->flags & IFF_PROMISC)
 995                         dev_set_promiscuity(new_active->dev, 1);
 996
 997                 if (bond->dev->flags & IFF_ALLMULTI)
 998                         dev_set_allmulti(new_active->dev, 1);
 999
1000                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
1001                         dev_mc_add(new_active->dev, dmi->dmi_addr,
1002                                    dmi->dmi_addrlen, 0);
1003                 bond_resend_igmp_join_requests(bond);
1004         }
1005 }
1006
1007 /*
1008  * bond_do_fail_over_mac
1009  *
1010  * Perform special MAC address swapping for fail_over_mac settings
1011  *
1012  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1013  */
1014 static void bond_do_fail_over_mac(struct bonding *bond,
1015                                   struct slave *new_active,
1016                                   struct slave *old_active)
1017         __releases(&bond->curr_slave_lock)
1018         __releases(&bond->lock)
1019         __acquires(&bond->lock)
1020         __acquires(&bond->curr_slave_lock)
1021 {
1022         u8 tmp_mac[ETH_ALEN];
1023         struct sockaddr saddr;
1024         int rv;
1025
1026         switch (bond->params.fail_over_mac) {
1027         case BOND_FOM_ACTIVE:
1028                 if (new_active)
1029                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1030                                new_active->dev->addr_len);
1031                 break;
1032         case BOND_FOM_FOLLOW:
1033                 /*
1034                  * if new_active && old_active, swap them
1035                  * if just old_active, do nothing (going to no active slave)
1036                  * if just new_active, set new_active to bond's MAC
1037                  */
1038                 if (!new_active)
1039                         return;
1040
1041                 write_unlock_bh(&bond->curr_slave_lock);
1042                 read_unlock(&bond->lock);
1043
1044                 if (old_active) {
1045                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1046                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1047                                ETH_ALEN);
1048                         saddr.sa_family = new_active->dev->type;
1049                 } else {
1050                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1051                         saddr.sa_family = bond->dev->type;
1052                 }
1053
1054                 rv = dev_set_mac_address(new_active->dev, &saddr);
1055                 if (rv) {
1056                         pr_err("%s: Error %d setting MAC of slave %s\n",
1057                                bond->dev->name, -rv, new_active->dev->name);
1058                         goto out;
1059                 }
1060
1061                 if (!old_active)
1062                         goto out;
1063
1064                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1065                 saddr.sa_family = old_active->dev->type;
1066
1067                 rv = dev_set_mac_address(old_active->dev, &saddr);
1068                 if (rv)
1069                         pr_err("%s: Error %d setting MAC of slave %s\n",
1070                                bond->dev->name, -rv, new_active->dev->name);
1071 out:
1072                 read_lock(&bond->lock);
1073                 write_lock_bh(&bond->curr_slave_lock);
1074                 break;
1075         default:
1076                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1077                        bond->dev->name, bond->params.fail_over_mac);
1078                 break;
1079         }
1080
1081 }
1082
1083 static bool bond_should_change_active(struct bonding *bond)
1084 {
1085         struct slave *prim = bond->primary_slave;
1086         struct slave *curr = bond->curr_active_slave;
1087
1088         if (!prim || !curr || curr->link != BOND_LINK_UP)
1089                 return true;
1090         if (bond->force_primary) {
1091                 bond->force_primary = false;
1092                 return true;
1093         }
1094         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1095             (prim->speed < curr->speed ||
1096              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1097                 return false;
1098         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1099                 return false;
1100         return true;
1101 }
1102
1103 /**
1104  * find_best_interface - select the best available slave to be the active one
1105  * @bond: our bonding struct
1106  *
1107  * Warning: Caller must hold curr_slave_lock for writing.
1108  */
1109 static struct slave *bond_find_best_slave(struct bonding *bond)
1110 {
1111         struct slave *new_active, *old_active;
1112         struct slave *bestslave = NULL;
1113         int mintime = bond->params.updelay;
1114         int i;
1115
1116         new_active = bond->curr_active_slave;
1117
1118         if (!new_active) { /* there were no active slaves left */
1119                 if (bond->slave_cnt > 0)   /* found one slave */
1120                         new_active = bond->first_slave;
1121                 else
1122                         return NULL; /* still no slave, return NULL */
1123         }
1124
1125         if ((bond->primary_slave) &&
1126             bond->primary_slave->link == BOND_LINK_UP &&
1127             bond_should_change_active(bond)) {
1128                 new_active = bond->primary_slave;
1129         }
1130
1131         /* remember where to stop iterating over the slaves */
1132         old_active = new_active;
1133
1134         bond_for_each_slave_from(bond, new_active, i, old_active) {
1135                 if (new_active->link == BOND_LINK_UP) {
1136                         return new_active;
1137                 } else if (new_active->link == BOND_LINK_BACK &&
1138                            IS_UP(new_active->dev)) {
1139                         /* link up, but waiting for stabilization */
1140                         if (new_active->delay < mintime) {
1141                                 mintime = new_active->delay;
1142                                 bestslave = new_active;
1143                         }
1144                 }
1145         }
1146
1147         return bestslave;
1148 }
1149
1150 /**
1151  * change_active_interface - change the active slave into the specified one
1152  * @bond: our bonding struct
1153  * @new: the new slave to make the active one
1154  *
1155  * Set the new slave to the bond's settings and unset them on the old
1156  * curr_active_slave.
1157  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1158  *
1159  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1160  * because it is apparently the best available slave we have, even though its
1161  * updelay hasn't timed out yet.
1162  *
1163  * If new_active is not NULL, caller must hold bond->lock for read and
1164  * curr_slave_lock for write_bh.
1165  */
1166 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1167 {
1168         struct slave *old_active = bond->curr_active_slave;
1169
1170         if (old_active == new_active)
1171                 return;
1172
1173         if (new_active) {
1174                 new_active->jiffies = jiffies;
1175
1176                 if (new_active->link == BOND_LINK_BACK) {
1177                         if (USES_PRIMARY(bond->params.mode)) {
1178                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1179                                         bond->dev->name, new_active->dev->name,
1180                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1181                         }
1182
1183                         new_active->delay = 0;
1184                         new_active->link = BOND_LINK_UP;
1185
1186                         if (bond->params.mode == BOND_MODE_8023AD)
1187                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1188
1189                         if (bond_is_lb(bond))
1190                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1191                 } else {
1192                         if (USES_PRIMARY(bond->params.mode)) {
1193                                 pr_info("%s: making interface %s the new active one.\n",
1194                                         bond->dev->name, new_active->dev->name);
1195                         }
1196                 }
1197         }
1198
1199         if (USES_PRIMARY(bond->params.mode))
1200                 bond_mc_swap(bond, new_active, old_active);
1201
1202         if (bond_is_lb(bond)) {
1203                 bond_alb_handle_active_change(bond, new_active);
1204                 if (old_active)
1205                         bond_set_slave_inactive_flags(old_active);
1206                 if (new_active)
1207                         bond_set_slave_active_flags(new_active);
1208         } else {
1209                 bond->curr_active_slave = new_active;
1210         }
1211
1212         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1213                 if (old_active)
1214                         bond_set_slave_inactive_flags(old_active);
1215
1216                 if (new_active) {
1217                         bond_set_slave_active_flags(new_active);
1218
1219                         if (bond->params.fail_over_mac)
1220                                 bond_do_fail_over_mac(bond, new_active,
1221                                                       old_active);
1222
1223                         bond->send_grat_arp = bond->params.num_grat_arp;
1224                         bond_send_gratuitous_arp(bond);
1225
1226                         bond->send_unsol_na = bond->params.num_unsol_na;
1227                         bond_send_unsolicited_na(bond);
1228
1229                         write_unlock_bh(&bond->curr_slave_lock);
1230                         read_unlock(&bond->lock);
1231
1232                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1233
1234                         read_lock(&bond->lock);
1235                         write_lock_bh(&bond->curr_slave_lock);
1236                 }
1237         }
1238 }
1239
1240 /**
1241  * bond_select_active_slave - select a new active slave, if needed
1242  * @bond: our bonding struct
1243  *
1244  * This functions should be called when one of the following occurs:
1245  * - The old curr_active_slave has been released or lost its link.
1246  * - The primary_slave has got its link back.
1247  * - A slave has got its link back and there's no old curr_active_slave.
1248  *
1249  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1250  */
1251 void bond_select_active_slave(struct bonding *bond)
1252 {
1253         struct slave *best_slave;
1254         int rv;
1255
1256         best_slave = bond_find_best_slave(bond);
1257         if (best_slave != bond->curr_active_slave) {
1258                 bond_change_active_slave(bond, best_slave);
1259                 rv = bond_set_carrier(bond);
1260                 if (!rv)
1261                         return;
1262
1263                 if (netif_carrier_ok(bond->dev)) {
1264                         pr_info("%s: first active interface up!\n",
1265                                 bond->dev->name);
1266                 } else {
1267                         pr_info("%s: now running without any active interface !\n",
1268                                 bond->dev->name);
1269                 }
1270         }
1271 }
1272
1273 /*--------------------------- slave list handling ---------------------------*/
1274
1275 /*
1276  * This function attaches the slave to the end of list.
1277  *
1278  * bond->lock held for writing by caller.
1279  */
1280 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1281 {
1282         if (bond->first_slave == NULL) { /* attaching the first slave */
1283                 new_slave->next = new_slave;
1284                 new_slave->prev = new_slave;
1285                 bond->first_slave = new_slave;
1286         } else {
1287                 new_slave->next = bond->first_slave;
1288                 new_slave->prev = bond->first_slave->prev;
1289                 new_slave->next->prev = new_slave;
1290                 new_slave->prev->next = new_slave;
1291         }
1292
1293         bond->slave_cnt++;
1294 }
1295
1296 /*
1297  * This function detaches the slave from the list.
1298  * WARNING: no check is made to verify if the slave effectively
1299  * belongs to <bond>.
1300  * Nothing is freed on return, structures are just unchained.
1301  * If any slave pointer in bond was pointing to <slave>,
1302  * it should be changed by the calling function.
1303  *
1304  * bond->lock held for writing by caller.
1305  */
1306 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1307 {
1308         if (slave->next)
1309                 slave->next->prev = slave->prev;
1310
1311         if (slave->prev)
1312                 slave->prev->next = slave->next;
1313
1314         if (bond->first_slave == slave) { /* slave is the first slave */
1315                 if (bond->slave_cnt > 1) { /* there are more slave */
1316                         bond->first_slave = slave->next;
1317                 } else {
1318                         bond->first_slave = NULL; /* slave was the last one */
1319                 }
1320         }
1321
1322         slave->next = NULL;
1323         slave->prev = NULL;
1324         bond->slave_cnt--;
1325 }
1326
1327 /*---------------------------------- IOCTL ----------------------------------*/
1328
1329 static int bond_sethwaddr(struct net_device *bond_dev,
1330                           struct net_device *slave_dev)
1331 {
1332         pr_debug("bond_dev=%p\n", bond_dev);
1333         pr_debug("slave_dev=%p\n", slave_dev);
1334         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1335         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1336         return 0;
1337 }
1338
1339 #define BOND_VLAN_FEATURES \
1340         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1341          NETIF_F_HW_VLAN_FILTER)
1342
1343 /*
1344  * Compute the common dev->feature set available to all slaves.  Some
1345  * feature bits are managed elsewhere, so preserve those feature bits
1346  * on the master device.
1347  */
1348 static int bond_compute_features(struct bonding *bond)
1349 {
1350         struct slave *slave;
1351         struct net_device *bond_dev = bond->dev;
1352         unsigned long features = bond_dev->features;
1353         unsigned long vlan_features = 0;
1354         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1355                                                 bond_dev->hard_header_len);
1356         int i;
1357
1358         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1359         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1360
1361         if (!bond->first_slave)
1362                 goto done;
1363
1364         features &= ~NETIF_F_ONE_FOR_ALL;
1365
1366         vlan_features = bond->first_slave->dev->vlan_features;
1367         bond_for_each_slave(bond, slave, i) {
1368                 features = netdev_increment_features(features,
1369                                                      slave->dev->features,
1370                                                      NETIF_F_ONE_FOR_ALL);
1371                 vlan_features = netdev_increment_features(vlan_features,
1372                                                         slave->dev->vlan_features,
1373                                                         NETIF_F_ONE_FOR_ALL);
1374                 if (slave->dev->hard_header_len > max_hard_header_len)
1375                         max_hard_header_len = slave->dev->hard_header_len;
1376         }
1377
1378 done:
1379         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1380         bond_dev->features = netdev_fix_features(features, NULL);
1381         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1382         bond_dev->hard_header_len = max_hard_header_len;
1383
1384         return 0;
1385 }
1386
1387 static void bond_setup_by_slave(struct net_device *bond_dev,
1388                                 struct net_device *slave_dev)
1389 {
1390         struct bonding *bond = netdev_priv(bond_dev);
1391
1392         bond_dev->header_ops        = slave_dev->header_ops;
1393
1394         bond_dev->type              = slave_dev->type;
1395         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1396         bond_dev->addr_len          = slave_dev->addr_len;
1397
1398         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1399                 slave_dev->addr_len);
1400         bond->setup_by_slave = 1;
1401 }
1402
1403 /* enslave device <slave> to bond device <master> */
1404 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1405 {
1406         struct bonding *bond = netdev_priv(bond_dev);
1407         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1408         struct slave *new_slave = NULL;
1409         struct dev_mc_list *dmi;
1410         struct sockaddr addr;
1411         int link_reporting;
1412         int old_features = bond_dev->features;
1413         int res = 0;
1414
1415         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1416                 slave_ops->ndo_do_ioctl == NULL) {
1417                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1418                            bond_dev->name, slave_dev->name);
1419         }
1420
1421         /* bond must be initialized by bond_open() before enslaving */
1422         if (!(bond_dev->flags & IFF_UP)) {
1423                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1424                            bond_dev->name);
1425         }
1426
1427         /* already enslaved */
1428         if (slave_dev->flags & IFF_SLAVE) {
1429                 pr_debug("Error, Device was already enslaved\n");
1430                 return -EBUSY;
1431         }
1432
1433         /* vlan challenged mutual exclusion */
1434         /* no need to lock since we're protected by rtnl_lock */
1435         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1436                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1437                 if (!list_empty(&bond->vlan_list)) {
1438                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1439                                bond_dev->name, slave_dev->name, bond_dev->name);
1440                         return -EPERM;
1441                 } else {
1442                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1443                                    bond_dev->name, slave_dev->name,
1444                                    slave_dev->name, bond_dev->name);
1445                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1446                 }
1447         } else {
1448                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1449                 if (bond->slave_cnt == 0) {
1450                         /* First slave, and it is not VLAN challenged,
1451                          * so remove the block of adding VLANs over the bond.
1452                          */
1453                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1454                 }
1455         }
1456
1457         /*
1458          * Old ifenslave binaries are no longer supported.  These can
1459          * be identified with moderate accuracy by the state of the slave:
1460          * the current ifenslave will set the interface down prior to
1461          * enslaving it; the old ifenslave will not.
1462          */
1463         if ((slave_dev->flags & IFF_UP)) {
1464                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1465                        slave_dev->name);
1466                 res = -EPERM;
1467                 goto err_undo_flags;
1468         }
1469
1470         /* set bonding device ether type by slave - bonding netdevices are
1471          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1472          * there is a need to override some of the type dependent attribs/funcs.
1473          *
1474          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1475          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1476          */
1477         if (bond->slave_cnt == 0) {
1478                 if (bond_dev->type != slave_dev->type) {
1479                         pr_debug("%s: change device type from %d to %d\n",
1480                                  bond_dev->name,
1481                                  bond_dev->type, slave_dev->type);
1482
1483                         netdev_bonding_change(bond_dev,
1484                                               NETDEV_PRE_TYPE_CHANGE);
1485
1486                         if (slave_dev->type != ARPHRD_ETHER)
1487                                 bond_setup_by_slave(bond_dev, slave_dev);
1488                         else
1489                                 ether_setup(bond_dev);
1490
1491                         netdev_bonding_change(bond_dev,
1492                                               NETDEV_POST_TYPE_CHANGE);
1493                 }
1494         } else if (bond_dev->type != slave_dev->type) {
1495                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1496                        slave_dev->name,
1497                        slave_dev->type, bond_dev->type);
1498                 res = -EINVAL;
1499                 goto err_undo_flags;
1500         }
1501
1502         if (slave_ops->ndo_set_mac_address == NULL) {
1503                 if (bond->slave_cnt == 0) {
1504                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1505                                    bond_dev->name);
1506                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1507                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1508                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1509                                bond_dev->name);
1510                         res = -EOPNOTSUPP;
1511                         goto err_undo_flags;
1512                 }
1513         }
1514
1515         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1516         if (!new_slave) {
1517                 res = -ENOMEM;
1518                 goto err_undo_flags;
1519         }
1520
1521         /* save slave's original flags before calling
1522          * netdev_set_master and dev_open
1523          */
1524         new_slave->original_flags = slave_dev->flags;
1525
1526         /*
1527          * Save slave's original ("permanent") mac address for modes
1528          * that need it, and for restoring it upon release, and then
1529          * set it to the master's address
1530          */
1531         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1532
1533         if (!bond->params.fail_over_mac) {
1534                 /*
1535                  * Set slave to master's mac address.  The application already
1536                  * set the master's mac address to that of the first slave
1537                  */
1538                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1539                 addr.sa_family = slave_dev->type;
1540                 res = dev_set_mac_address(slave_dev, &addr);
1541                 if (res) {
1542                         pr_debug("Error %d calling set_mac_address\n", res);
1543                         goto err_free;
1544                 }
1545         }
1546
1547         res = netdev_set_master(slave_dev, bond_dev);
1548         if (res) {
1549                 pr_debug("Error %d calling netdev_set_master\n", res);
1550                 goto err_restore_mac;
1551         }
1552         /* open the slave since the application closed it */
1553         res = dev_open(slave_dev);
1554         if (res) {
1555                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1556                 goto err_unset_master;
1557         }
1558
1559         new_slave->dev = slave_dev;
1560         slave_dev->priv_flags |= IFF_BONDING;
1561
1562         if (bond_is_lb(bond)) {
1563                 /* bond_alb_init_slave() must be called before all other stages since
1564                  * it might fail and we do not want to have to undo everything
1565                  */
1566                 res = bond_alb_init_slave(bond, new_slave);
1567                 if (res)
1568                         goto err_close;
1569         }
1570
1571         /* If the mode USES_PRIMARY, then the new slave gets the
1572          * master's promisc (and mc) settings only if it becomes the
1573          * curr_active_slave, and that is taken care of later when calling
1574          * bond_change_active()
1575          */
1576         if (!USES_PRIMARY(bond->params.mode)) {
1577                 /* set promiscuity level to new slave */
1578                 if (bond_dev->flags & IFF_PROMISC) {
1579                         res = dev_set_promiscuity(slave_dev, 1);
1580                         if (res)
1581                                 goto err_close;
1582                 }
1583
1584                 /* set allmulti level to new slave */
1585                 if (bond_dev->flags & IFF_ALLMULTI) {
1586                         res = dev_set_allmulti(slave_dev, 1);
1587                         if (res)
1588                                 goto err_close;
1589                 }
1590
1591                 netif_addr_lock_bh(bond_dev);
1592                 /* upload master's mc_list to new slave */
1593                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1594                         dev_mc_add(slave_dev, dmi->dmi_addr,
1595                                    dmi->dmi_addrlen, 0);
1596                 netif_addr_unlock_bh(bond_dev);
1597         }
1598
1599         if (bond->params.mode == BOND_MODE_8023AD) {
1600                 /* add lacpdu mc addr to mc list */
1601                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1602
1603                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1604         }
1605
1606         bond_add_vlans_on_slave(bond, slave_dev);
1607
1608         write_lock_bh(&bond->lock);
1609
1610         bond_attach_slave(bond, new_slave);
1611
1612         new_slave->delay = 0;
1613         new_slave->link_failure_count = 0;
1614
1615         bond_compute_features(bond);
1616
1617         write_unlock_bh(&bond->lock);
1618
1619         read_lock(&bond->lock);
1620
1621         new_slave->last_arp_rx = jiffies;
1622
1623         if (bond->params.miimon && !bond->params.use_carrier) {
1624                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1625
1626                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1627                         /*
1628                          * miimon is set but a bonded network driver
1629                          * does not support ETHTOOL/MII and
1630                          * arp_interval is not set.  Note: if
1631                          * use_carrier is enabled, we will never go
1632                          * here (because netif_carrier is always
1633                          * supported); thus, we don't need to change
1634                          * the messages for netif_carrier.
1635                          */
1636                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1637                                bond_dev->name, slave_dev->name);
1638                 } else if (link_reporting == -1) {
1639                         /* unable get link status using mii/ethtool */
1640                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1641                                    bond_dev->name, slave_dev->name);
1642                 }
1643         }
1644
1645         /* check for initial state */
1646         if (!bond->params.miimon ||
1647             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1648                 if (bond->params.updelay) {
1649                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1650                         new_slave->link  = BOND_LINK_BACK;
1651                         new_slave->delay = bond->params.updelay;
1652                 } else {
1653                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1654                         new_slave->link  = BOND_LINK_UP;
1655                 }
1656                 new_slave->jiffies = jiffies;
1657         } else {
1658                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1659                 new_slave->link  = BOND_LINK_DOWN;
1660         }
1661
1662         if (bond_update_speed_duplex(new_slave) &&
1663             (new_slave->link != BOND_LINK_DOWN)) {
1664                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1665                            bond_dev->name, new_slave->dev->name);
1666
1667                 if (bond->params.mode == BOND_MODE_8023AD) {
1668                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1669                                    bond_dev->name);
1670                 }
1671         }
1672
1673         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1674                 /* if there is a primary slave, remember it */
1675                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1676                         bond->primary_slave = new_slave;
1677                         bond->force_primary = true;
1678                 }
1679         }
1680
1681         write_lock_bh(&bond->curr_slave_lock);
1682
1683         switch (bond->params.mode) {
1684         case BOND_MODE_ACTIVEBACKUP:
1685                 bond_set_slave_inactive_flags(new_slave);
1686                 bond_select_active_slave(bond);
1687                 break;
1688         case BOND_MODE_8023AD:
1689                 /* in 802.3ad mode, the internal mechanism
1690                  * will activate the slaves in the selected
1691                  * aggregator
1692                  */
1693                 bond_set_slave_inactive_flags(new_slave);
1694                 /* if this is the first slave */
1695                 if (bond->slave_cnt == 1) {
1696                         SLAVE_AD_INFO(new_slave).id = 1;
1697                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1698                          * can be called only after the mac address of the bond is set
1699                          */
1700                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1701                                             bond->params.lacp_fast);
1702                 } else {
1703                         SLAVE_AD_INFO(new_slave).id =
1704                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1705                 }
1706
1707                 bond_3ad_bind_slave(new_slave);
1708                 break;
1709         case BOND_MODE_TLB:
1710         case BOND_MODE_ALB:
1711                 new_slave->state = BOND_STATE_ACTIVE;
1712                 bond_set_slave_inactive_flags(new_slave);
1713                 bond_select_active_slave(bond);
1714                 break;
1715         default:
1716                 pr_debug("This slave is always active in trunk mode\n");
1717
1718                 /* always active in trunk mode */
1719                 new_slave->state = BOND_STATE_ACTIVE;
1720
1721                 /* In trunking mode there is little meaning to curr_active_slave
1722                  * anyway (it holds no special properties of the bond device),
1723                  * so we can change it without calling change_active_interface()
1724                  */
1725                 if (!bond->curr_active_slave)
1726                         bond->curr_active_slave = new_slave;
1727
1728                 break;
1729         } /* switch(bond_mode) */
1730
1731         write_unlock_bh(&bond->curr_slave_lock);
1732
1733         bond_set_carrier(bond);
1734
1735         read_unlock(&bond->lock);
1736
1737         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738         if (res)
1739                 goto err_close;
1740
1741         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1742                 bond_dev->name, slave_dev->name,
1743                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1744                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1745
1746         /* enslave is successful */
1747         return 0;
1748
1749 /* Undo stages on error */
1750 err_close:
1751         dev_close(slave_dev);
1752
1753 err_unset_master:
1754         netdev_set_master(slave_dev, NULL);
1755
1756 err_restore_mac:
1757         if (!bond->params.fail_over_mac) {
1758                 /* XXX TODO - fom follow mode needs to change master's
1759                  * MAC if this slave's MAC is in use by the bond, or at
1760                  * least print a warning.
1761                  */
1762                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1763                 addr.sa_family = slave_dev->type;
1764                 dev_set_mac_address(slave_dev, &addr);
1765         }
1766
1767 err_free:
1768         kfree(new_slave);
1769
1770 err_undo_flags:
1771         bond_dev->features = old_features;
1772
1773         return res;
1774 }
1775
1776 /*
1777  * Try to release the slave device <slave> from the bond device <master>
1778  * It is legal to access curr_active_slave without a lock because all the function
1779  * is write-locked.
1780  *
1781  * The rules for slave state should be:
1782  *   for Active/Backup:
1783  *     Active stays on all backups go down
1784  *   for Bonded connections:
1785  *     The first up interface should be left on and all others downed.
1786  */
1787 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1788 {
1789         struct bonding *bond = netdev_priv(bond_dev);
1790         struct slave *slave, *oldcurrent;
1791         struct sockaddr addr;
1792
1793         /* slave is not a slave or master is not master of this slave */
1794         if (!(slave_dev->flags & IFF_SLAVE) ||
1795             (slave_dev->master != bond_dev)) {
1796                 pr_err("%s: Error: cannot release %s.\n",
1797                        bond_dev->name, slave_dev->name);
1798                 return -EINVAL;
1799         }
1800
1801         write_lock_bh(&bond->lock);
1802
1803         slave = bond_get_slave_by_dev(bond, slave_dev);
1804         if (!slave) {
1805                 /* not a slave of this bond */
1806                 pr_info("%s: %s not enslaved\n",
1807                         bond_dev->name, slave_dev->name);
1808                 write_unlock_bh(&bond->lock);
1809                 return -EINVAL;
1810         }
1811
1812         if (!bond->params.fail_over_mac) {
1813                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1814                     bond->slave_cnt > 1)
1815                         pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1816                                    bond_dev->name, slave_dev->name,
1817                                    slave->perm_hwaddr,
1818                                    bond_dev->name, slave_dev->name);
1819         }
1820
1821         /* Inform AD package of unbinding of slave. */
1822         if (bond->params.mode == BOND_MODE_8023AD) {
1823                 /* must be called before the slave is
1824                  * detached from the list
1825                  */
1826                 bond_3ad_unbind_slave(slave);
1827         }
1828
1829         pr_info("%s: releasing %s interface %s\n",
1830                 bond_dev->name,
1831                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1832                 slave_dev->name);
1833
1834         oldcurrent = bond->curr_active_slave;
1835
1836         bond->current_arp_slave = NULL;
1837
1838         /* release the slave from its bond */
1839         bond_detach_slave(bond, slave);
1840
1841         bond_compute_features(bond);
1842
1843         if (bond->primary_slave == slave)
1844                 bond->primary_slave = NULL;
1845
1846         if (oldcurrent == slave)
1847                 bond_change_active_slave(bond, NULL);
1848
1849         if (bond_is_lb(bond)) {
1850                 /* Must be called only after the slave has been
1851                  * detached from the list and the curr_active_slave
1852                  * has been cleared (if our_slave == old_current),
1853                  * but before a new active slave is selected.
1854                  */
1855                 write_unlock_bh(&bond->lock);
1856                 bond_alb_deinit_slave(bond, slave);
1857                 write_lock_bh(&bond->lock);
1858         }
1859
1860         if (oldcurrent == slave) {
1861                 /*
1862                  * Note that we hold RTNL over this sequence, so there
1863                  * is no concern that another slave add/remove event
1864                  * will interfere.
1865                  */
1866                 write_unlock_bh(&bond->lock);
1867                 read_lock(&bond->lock);
1868                 write_lock_bh(&bond->curr_slave_lock);
1869
1870                 bond_select_active_slave(bond);
1871
1872                 write_unlock_bh(&bond->curr_slave_lock);
1873                 read_unlock(&bond->lock);
1874                 write_lock_bh(&bond->lock);
1875         }
1876
1877         if (bond->slave_cnt == 0) {
1878                 bond_set_carrier(bond);
1879
1880                 /* if the last slave was removed, zero the mac address
1881                  * of the master so it will be set by the application
1882                  * to the mac address of the first slave
1883                  */
1884                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1885
1886                 if (list_empty(&bond->vlan_list)) {
1887                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1888                 } else {
1889                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1890                                    bond_dev->name, bond_dev->name);
1891                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1892                                    bond_dev->name);
1893                 }
1894         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1895                    !bond_has_challenged_slaves(bond)) {
1896                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1897                         bond_dev->name, slave_dev->name, bond_dev->name);
1898                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1899         }
1900
1901         write_unlock_bh(&bond->lock);
1902
1903         /* must do this from outside any spinlocks */
1904         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1905
1906         bond_del_vlans_from_slave(bond, slave_dev);
1907
1908         /* If the mode USES_PRIMARY, then we should only remove its
1909          * promisc and mc settings if it was the curr_active_slave, but that was
1910          * already taken care of above when we detached the slave
1911          */
1912         if (!USES_PRIMARY(bond->params.mode)) {
1913                 /* unset promiscuity level from slave */
1914                 if (bond_dev->flags & IFF_PROMISC)
1915                         dev_set_promiscuity(slave_dev, -1);
1916
1917                 /* unset allmulti level from slave */
1918                 if (bond_dev->flags & IFF_ALLMULTI)
1919                         dev_set_allmulti(slave_dev, -1);
1920
1921                 /* flush master's mc_list from slave */
1922                 netif_addr_lock_bh(bond_dev);
1923                 bond_mc_list_flush(bond_dev, slave_dev);
1924                 netif_addr_unlock_bh(bond_dev);
1925         }
1926
1927         netdev_set_master(slave_dev, NULL);
1928
1929         /* close slave before restoring its mac address */
1930         dev_close(slave_dev);
1931
1932         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1933                 /* restore original ("permanent") mac address */
1934                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1935                 addr.sa_family = slave_dev->type;
1936                 dev_set_mac_address(slave_dev, &addr);
1937         }
1938
1939         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1940                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1941                                    IFF_SLAVE_NEEDARP);
1942
1943         kfree(slave);
1944
1945         return 0;  /* deletion OK */
1946 }
1947
1948 /*
1949 * First release a slave and than destroy the bond if no more slaves are left.
1950 * Must be under rtnl_lock when this function is called.
1951 */
1952 int  bond_release_and_destroy(struct net_device *bond_dev,
1953                               struct net_device *slave_dev)
1954 {
1955         struct bonding *bond = netdev_priv(bond_dev);
1956         int ret;
1957
1958         ret = bond_release(bond_dev, slave_dev);
1959         if ((ret == 0) && (bond->slave_cnt == 0)) {
1960                 pr_info("%s: destroying bond %s.\n",
1961                         bond_dev->name, bond_dev->name);
1962                 unregister_netdevice(bond_dev);
1963         }
1964         return ret;
1965 }
1966
1967 /*
1968  * This function releases all slaves.
1969  */
1970 static int bond_release_all(struct net_device *bond_dev)
1971 {
1972         struct bonding *bond = netdev_priv(bond_dev);
1973         struct slave *slave;
1974         struct net_device *slave_dev;
1975         struct sockaddr addr;
1976
1977         write_lock_bh(&bond->lock);
1978
1979         netif_carrier_off(bond_dev);
1980
1981         if (bond->slave_cnt == 0)
1982                 goto out;
1983
1984         bond->current_arp_slave = NULL;
1985         bond->primary_slave = NULL;
1986         bond_change_active_slave(bond, NULL);
1987
1988         while ((slave = bond->first_slave) != NULL) {
1989                 /* Inform AD package of unbinding of slave
1990                  * before slave is detached from the list.
1991                  */
1992                 if (bond->params.mode == BOND_MODE_8023AD)
1993                         bond_3ad_unbind_slave(slave);
1994
1995                 slave_dev = slave->dev;
1996                 bond_detach_slave(bond, slave);
1997
1998                 /* now that the slave is detached, unlock and perform
1999                  * all the undo steps that should not be called from
2000                  * within a lock.
2001                  */
2002                 write_unlock_bh(&bond->lock);
2003
2004                 if (bond_is_lb(bond)) {
2005                         /* must be called only after the slave
2006                          * has been detached from the list
2007                          */
2008                         bond_alb_deinit_slave(bond, slave);
2009                 }
2010
2011                 bond_compute_features(bond);
2012
2013                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2014                 bond_del_vlans_from_slave(bond, slave_dev);
2015
2016                 /* If the mode USES_PRIMARY, then we should only remove its
2017                  * promisc and mc settings if it was the curr_active_slave, but that was
2018                  * already taken care of above when we detached the slave
2019                  */
2020                 if (!USES_PRIMARY(bond->params.mode)) {
2021                         /* unset promiscuity level from slave */
2022                         if (bond_dev->flags & IFF_PROMISC)
2023                                 dev_set_promiscuity(slave_dev, -1);
2024
2025                         /* unset allmulti level from slave */
2026                         if (bond_dev->flags & IFF_ALLMULTI)
2027                                 dev_set_allmulti(slave_dev, -1);
2028
2029                         /* flush master's mc_list from slave */
2030                         netif_addr_lock_bh(bond_dev);
2031                         bond_mc_list_flush(bond_dev, slave_dev);
2032                         netif_addr_unlock_bh(bond_dev);
2033                 }
2034
2035                 netdev_set_master(slave_dev, NULL);
2036
2037                 /* close slave before restoring its mac address */
2038                 dev_close(slave_dev);
2039
2040                 if (!bond->params.fail_over_mac) {
2041                         /* restore original ("permanent") mac address*/
2042                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2043                         addr.sa_family = slave_dev->type;
2044                         dev_set_mac_address(slave_dev, &addr);
2045                 }
2046
2047                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2048                                            IFF_SLAVE_INACTIVE);
2049
2050                 kfree(slave);
2051
2052                 /* re-acquire the lock before getting the next slave */
2053                 write_lock_bh(&bond->lock);
2054         }
2055
2056         /* zero the mac address of the master so it will be
2057          * set by the application to the mac address of the
2058          * first slave
2059          */
2060         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2061
2062         if (list_empty(&bond->vlan_list))
2063                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2064         else {
2065                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2066                            bond_dev->name, bond_dev->name);
2067                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2068                            bond_dev->name);
2069         }
2070
2071         pr_info("%s: released all slaves\n", bond_dev->name);
2072
2073 out:
2074         write_unlock_bh(&bond->lock);
2075
2076         return 0;
2077 }
2078
2079 /*
2080  * This function changes the active slave to slave <slave_dev>.
2081  * It returns -EINVAL in the following cases.
2082  *  - <slave_dev> is not found in the list.
2083  *  - There is not active slave now.
2084  *  - <slave_dev> is already active.
2085  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2086  *  - <slave_dev> is not running.
2087  * In these cases, this function does nothing.
2088  * In the other cases, current_slave pointer is changed and 0 is returned.
2089  */
2090 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2091 {
2092         struct bonding *bond = netdev_priv(bond_dev);
2093         struct slave *old_active = NULL;
2094         struct slave *new_active = NULL;
2095         int res = 0;
2096
2097         if (!USES_PRIMARY(bond->params.mode))
2098                 return -EINVAL;
2099
2100         /* Verify that master_dev is indeed the master of slave_dev */
2101         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2102                 return -EINVAL;
2103
2104         read_lock(&bond->lock);
2105
2106         read_lock(&bond->curr_slave_lock);
2107         old_active = bond->curr_active_slave;
2108         read_unlock(&bond->curr_slave_lock);
2109
2110         new_active = bond_get_slave_by_dev(bond, slave_dev);
2111
2112         /*
2113          * Changing to the current active: do nothing; return success.
2114          */
2115         if (new_active && (new_active == old_active)) {
2116                 read_unlock(&bond->lock);
2117                 return 0;
2118         }
2119
2120         if ((new_active) &&
2121             (old_active) &&
2122             (new_active->link == BOND_LINK_UP) &&
2123             IS_UP(new_active->dev)) {
2124                 write_lock_bh(&bond->curr_slave_lock);
2125                 bond_change_active_slave(bond, new_active);
2126                 write_unlock_bh(&bond->curr_slave_lock);
2127         } else
2128                 res = -EINVAL;
2129
2130         read_unlock(&bond->lock);
2131
2132         return res;
2133 }
2134
2135 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2136 {
2137         struct bonding *bond = netdev_priv(bond_dev);
2138
2139         info->bond_mode = bond->params.mode;
2140         info->miimon = bond->params.miimon;
2141
2142         read_lock(&bond->lock);
2143         info->num_slaves = bond->slave_cnt;
2144         read_unlock(&bond->lock);
2145
2146         return 0;
2147 }
2148
2149 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2150 {
2151         struct bonding *bond = netdev_priv(bond_dev);
2152         struct slave *slave;
2153         int i, res = -ENODEV;
2154
2155         read_lock(&bond->lock);
2156
2157         bond_for_each_slave(bond, slave, i) {
2158                 if (i == (int)info->slave_id) {
2159                         res = 0;
2160                         strcpy(info->slave_name, slave->dev->name);
2161                         info->link = slave->link;
2162                         info->state = slave->state;
2163                         info->link_failure_count = slave->link_failure_count;
2164                         break;
2165                 }
2166         }
2167
2168         read_unlock(&bond->lock);
2169
2170         return res;
2171 }
2172
2173 /*-------------------------------- Monitoring -------------------------------*/
2174
2175
2176 static int bond_miimon_inspect(struct bonding *bond)
2177 {
2178         struct slave *slave;
2179         int i, link_state, commit = 0;
2180         bool ignore_updelay;
2181
2182         ignore_updelay = !bond->curr_active_slave ? true : false;
2183
2184         bond_for_each_slave(bond, slave, i) {
2185                 slave->new_link = BOND_LINK_NOCHANGE;
2186
2187                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2188
2189                 switch (slave->link) {
2190                 case BOND_LINK_UP:
2191                         if (link_state)
2192                                 continue;
2193
2194                         slave->link = BOND_LINK_FAIL;
2195                         slave->delay = bond->params.downdelay;
2196                         if (slave->delay) {
2197                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2198                                         bond->dev->name,
2199                                         (bond->params.mode ==
2200                                          BOND_MODE_ACTIVEBACKUP) ?
2201                                         ((slave->state == BOND_STATE_ACTIVE) ?
2202                                          "active " : "backup ") : "",
2203                                         slave->dev->name,
2204                                         bond->params.downdelay * bond->params.miimon);
2205                         }
2206                         /*FALLTHRU*/
2207                 case BOND_LINK_FAIL:
2208                         if (link_state) {
2209                                 /*
2210                                  * recovered before downdelay expired
2211                                  */
2212                                 slave->link = BOND_LINK_UP;
2213                                 slave->jiffies = jiffies;
2214                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2215                                         bond->dev->name,
2216                                         (bond->params.downdelay - slave->delay) *
2217                                         bond->params.miimon,
2218                                         slave->dev->name);
2219                                 continue;
2220                         }
2221
2222                         if (slave->delay <= 0) {
2223                                 slave->new_link = BOND_LINK_DOWN;
2224                                 commit++;
2225                                 continue;
2226                         }
2227
2228                         slave->delay--;
2229                         break;
2230
2231                 case BOND_LINK_DOWN:
2232                         if (!link_state)
2233                                 continue;
2234
2235                         slave->link = BOND_LINK_BACK;
2236                         slave->delay = bond->params.updelay;
2237
2238                         if (slave->delay) {
2239                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2240                                         bond->dev->name, slave->dev->name,
2241                                         ignore_updelay ? 0 :
2242                                         bond->params.updelay *
2243                                         bond->params.miimon);
2244                         }
2245                         /*FALLTHRU*/
2246                 case BOND_LINK_BACK:
2247                         if (!link_state) {
2248                                 slave->link = BOND_LINK_DOWN;
2249                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2250                                         bond->dev->name,
2251                                         (bond->params.updelay - slave->delay) *
2252                                         bond->params.miimon,
2253                                         slave->dev->name);
2254
2255                                 continue;
2256                         }
2257
2258                         if (ignore_updelay)
2259                                 slave->delay = 0;
2260
2261                         if (slave->delay <= 0) {
2262                                 slave->new_link = BOND_LINK_UP;
2263                                 commit++;
2264                                 ignore_updelay = false;
2265                                 continue;
2266                         }
2267
2268                         slave->delay--;
2269                         break;
2270                 }
2271         }
2272
2273         return commit;
2274 }
2275
2276 static void bond_miimon_commit(struct bonding *bond)
2277 {
2278         struct slave *slave;
2279         int i;
2280
2281         bond_for_each_slave(bond, slave, i) {
2282                 switch (slave->new_link) {
2283                 case BOND_LINK_NOCHANGE:
2284                         continue;
2285
2286                 case BOND_LINK_UP:
2287                         slave->link = BOND_LINK_UP;
2288                         slave->jiffies = jiffies;
2289
2290                         if (bond->params.mode == BOND_MODE_8023AD) {
2291                                 /* prevent it from being the active one */
2292                                 slave->state = BOND_STATE_BACKUP;
2293                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2294                                 /* make it immediately active */
2295                                 slave->state = BOND_STATE_ACTIVE;
2296                         } else if (slave != bond->primary_slave) {
2297                                 /* prevent it from being the active one */
2298                                 slave->state = BOND_STATE_BACKUP;
2299                         }
2300
2301                         pr_info("%s: link status definitely up for interface %s.\n",
2302                                 bond->dev->name, slave->dev->name);
2303
2304                         /* notify ad that the link status has changed */
2305                         if (bond->params.mode == BOND_MODE_8023AD)
2306                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2307
2308                         if (bond_is_lb(bond))
2309                                 bond_alb_handle_link_change(bond, slave,
2310                                                             BOND_LINK_UP);
2311
2312                         if (!bond->curr_active_slave ||
2313                             (slave == bond->primary_slave))
2314                                 goto do_failover;
2315
2316                         continue;
2317
2318                 case BOND_LINK_DOWN:
2319                         if (slave->link_failure_count < UINT_MAX)
2320                                 slave->link_failure_count++;
2321
2322                         slave->link = BOND_LINK_DOWN;
2323
2324                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2325                             bond->params.mode == BOND_MODE_8023AD)
2326                                 bond_set_slave_inactive_flags(slave);
2327
2328                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2329                                 bond->dev->name, slave->dev->name);
2330
2331                         if (bond->params.mode == BOND_MODE_8023AD)
2332                                 bond_3ad_handle_link_change(slave,
2333                                                             BOND_LINK_DOWN);
2334
2335                         if (bond_is_lb(bond))
2336                                 bond_alb_handle_link_change(bond, slave,
2337                                                             BOND_LINK_DOWN);
2338
2339                         if (slave == bond->curr_active_slave)
2340                                 goto do_failover;
2341
2342                         continue;
2343
2344                 default:
2345                         pr_err("%s: invalid new link %d on slave %s\n",
2346                                bond->dev->name, slave->new_link,
2347                                slave->dev->name);
2348                         slave->new_link = BOND_LINK_NOCHANGE;
2349
2350                         continue;
2351                 }
2352
2353 do_failover:
2354                 ASSERT_RTNL();
2355                 write_lock_bh(&bond->curr_slave_lock);
2356                 bond_select_active_slave(bond);
2357                 write_unlock_bh(&bond->curr_slave_lock);
2358         }
2359
2360         bond_set_carrier(bond);
2361 }
2362
2363 /*
2364  * bond_mii_monitor
2365  *
2366  * Really a wrapper that splits the mii monitor into two phases: an
2367  * inspection, then (if inspection indicates something needs to be done)
2368  * an acquisition of appropriate locks followed by a commit phase to
2369  * implement whatever link state changes are indicated.
2370  */
2371 void bond_mii_monitor(struct work_struct *work)
2372 {
2373         struct bonding *bond = container_of(work, struct bonding,
2374                                             mii_work.work);
2375
2376         read_lock(&bond->lock);
2377         if (bond->kill_timers)
2378                 goto out;
2379
2380         if (bond->slave_cnt == 0)
2381                 goto re_arm;
2382
2383         if (bond->send_grat_arp) {
2384                 read_lock(&bond->curr_slave_lock);
2385                 bond_send_gratuitous_arp(bond);
2386                 read_unlock(&bond->curr_slave_lock);
2387         }
2388
2389         if (bond->send_unsol_na) {
2390                 read_lock(&bond->curr_slave_lock);
2391                 bond_send_unsolicited_na(bond);
2392                 read_unlock(&bond->curr_slave_lock);
2393         }
2394
2395         if (bond_miimon_inspect(bond)) {
2396                 read_unlock(&bond->lock);
2397                 rtnl_lock();
2398                 read_lock(&bond->lock);
2399
2400                 bond_miimon_commit(bond);
2401
2402                 read_unlock(&bond->lock);
2403                 rtnl_unlock();  /* might sleep, hold no other locks */
2404                 read_lock(&bond->lock);
2405         }
2406
2407 re_arm:
2408         if (bond->params.miimon)
2409                 queue_delayed_work(bond->wq, &bond->mii_work,
2410                                    msecs_to_jiffies(bond->params.miimon));
2411 out:
2412         read_unlock(&bond->lock);
2413 }
2414
2415 static __be32 bond_glean_dev_ip(struct net_device *dev)
2416 {
2417         struct in_device *idev;
2418         struct in_ifaddr *ifa;
2419         __be32 addr = 0;
2420
2421         if (!dev)
2422                 return 0;
2423
2424         rcu_read_lock();
2425         idev = __in_dev_get_rcu(dev);
2426         if (!idev)
2427                 goto out;
2428
2429         ifa = idev->ifa_list;
2430         if (!ifa)
2431                 goto out;
2432
2433         addr = ifa->ifa_local;
2434 out:
2435         rcu_read_unlock();
2436         return addr;
2437 }
2438
2439 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2440 {
2441         struct vlan_entry *vlan;
2442
2443         if (ip == bond->master_ip)
2444                 return 1;
2445
2446         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2447                 if (ip == vlan->vlan_ip)
2448                         return 1;
2449         }
2450
2451         return 0;
2452 }
2453
2454 /*
2455  * We go to the (large) trouble of VLAN tagging ARP frames because
2456  * switches in VLAN mode (especially if ports are configured as
2457  * "native" to a VLAN) might not pass non-tagged frames.
2458  */
2459 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2460 {
2461         struct sk_buff *skb;
2462
2463         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2464                  slave_dev->name, dest_ip, src_ip, vlan_id);
2465
2466         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2467                          NULL, slave_dev->dev_addr, NULL);
2468
2469         if (!skb) {
2470                 pr_err("ARP packet allocation failed\n");
2471                 return;
2472         }
2473         if (vlan_id) {
2474                 skb = vlan_put_tag(skb, vlan_id);
2475                 if (!skb) {
2476                         pr_err("failed to insert VLAN tag\n");
2477                         return;
2478                 }
2479         }
2480         arp_xmit(skb);
2481 }
2482
2483
2484 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2485 {
2486         int i, vlan_id, rv;
2487         __be32 *targets = bond->params.arp_targets;
2488         struct vlan_entry *vlan;
2489         struct net_device *vlan_dev;
2490         struct flowi fl;
2491         struct rtable *rt;
2492
2493         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2494                 if (!targets[i])
2495                         break;
2496                 pr_debug("basa: target %x\n", targets[i]);
2497                 if (list_empty(&bond->vlan_list)) {
2498                         pr_debug("basa: empty vlan: arp_send\n");
2499                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2500                                       bond->master_ip, 0);
2501                         continue;
2502                 }
2503
2504                 /*
2505                  * If VLANs are configured, we do a route lookup to
2506                  * determine which VLAN interface would be used, so we
2507                  * can tag the ARP with the proper VLAN tag.
2508                  */
2509                 memset(&fl, 0, sizeof(fl));
2510                 fl.fl4_dst = targets[i];
2511                 fl.fl4_tos = RTO_ONLINK;
2512
2513                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2514                 if (rv) {
2515                         if (net_ratelimit()) {
2516                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2517                                            bond->dev->name, &fl.fl4_dst);
2518                         }
2519                         continue;
2520                 }
2521
2522                 /*
2523                  * This target is not on a VLAN
2524                  */
2525                 if (rt->u.dst.dev == bond->dev) {
2526                         ip_rt_put(rt);
2527                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2528                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2529                                       bond->master_ip, 0);
2530                         continue;
2531                 }
2532
2533                 vlan_id = 0;
2534                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2535                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2536                         if (vlan_dev == rt->u.dst.dev) {
2537                                 vlan_id = vlan->vlan_id;
2538                                 pr_debug("basa: vlan match on %s %d\n",
2539                                        vlan_dev->name, vlan_id);
2540                                 break;
2541                         }
2542                 }
2543
2544                 if (vlan_id) {
2545                         ip_rt_put(rt);
2546                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2547                                       vlan->vlan_ip, vlan_id);
2548                         continue;
2549                 }
2550
2551                 if (net_ratelimit()) {
2552                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2553                                    bond->dev->name, &fl.fl4_dst,
2554                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2555                 }
2556                 ip_rt_put(rt);
2557         }
2558 }
2559
2560 /*
2561  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2562  * for each VLAN above us.
2563  *
2564  * Caller must hold curr_slave_lock for read or better
2565  */
2566 static void bond_send_gratuitous_arp(struct bonding *bond)
2567 {
2568         struct slave *slave = bond->curr_active_slave;
2569         struct vlan_entry *vlan;
2570         struct net_device *vlan_dev;
2571
2572         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2573                  bond->dev->name, slave ? slave->dev->name : "NULL");
2574
2575         if (!slave || !bond->send_grat_arp ||
2576             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2577                 return;
2578
2579         bond->send_grat_arp--;
2580
2581         if (bond->master_ip) {
2582                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2583                                 bond->master_ip, 0);
2584         }
2585
2586         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2587                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2588                 if (vlan->vlan_ip) {
2589                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2590                                       vlan->vlan_ip, vlan->vlan_id);
2591                 }
2592         }
2593 }
2594
2595 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2596 {
2597         int i;
2598         __be32 *targets = bond->params.arp_targets;
2599
2600         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2601                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2602                          &sip, &tip, i, &targets[i],
2603                          bond_has_this_ip(bond, tip));
2604                 if (sip == targets[i]) {
2605                         if (bond_has_this_ip(bond, tip))
2606                                 slave->last_arp_rx = jiffies;
2607                         return;
2608                 }
2609         }
2610 }
2611
2612 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2613 {
2614         struct arphdr *arp;
2615         struct slave *slave;
2616         struct bonding *bond;
2617         unsigned char *arp_ptr;
2618         __be32 sip, tip;
2619
2620         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2621                 /*
2622                  * When using VLANS and bonding, dev and oriv_dev may be
2623                  * incorrect if the physical interface supports VLAN
2624                  * acceleration.  With this change ARP validation now
2625                  * works for hosts only reachable on the VLAN interface.
2626                  */
2627                 dev = vlan_dev_real_dev(dev);
2628                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2629         }
2630
2631         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2632                 goto out;
2633
2634         bond = netdev_priv(dev);
2635         read_lock(&bond->lock);
2636
2637         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2638                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2639                  orig_dev ? orig_dev->name : "NULL");
2640
2641         slave = bond_get_slave_by_dev(bond, orig_dev);
2642         if (!slave || !slave_do_arp_validate(bond, slave))
2643                 goto out_unlock;
2644
2645         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2646                 goto out_unlock;
2647
2648         arp = arp_hdr(skb);
2649         if (arp->ar_hln != dev->addr_len ||
2650             skb->pkt_type == PACKET_OTHERHOST ||
2651             skb->pkt_type == PACKET_LOOPBACK ||
2652             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2653             arp->ar_pro != htons(ETH_P_IP) ||
2654             arp->ar_pln != 4)
2655                 goto out_unlock;
2656
2657         arp_ptr = (unsigned char *)(arp + 1);
2658         arp_ptr += dev->addr_len;
2659         memcpy(&sip, arp_ptr, 4);
2660         arp_ptr += 4 + dev->addr_len;
2661         memcpy(&tip, arp_ptr, 4);
2662
2663         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2664                  bond->dev->name, slave->dev->name, slave->state,
2665                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2666                  &sip, &tip);
2667
2668         /*
2669          * Backup slaves won't see the ARP reply, but do come through
2670          * here for each ARP probe (so we swap the sip/tip to validate
2671          * the probe).  In a "redundant switch, common router" type of
2672          * configuration, the ARP probe will (hopefully) travel from
2673          * the active, through one switch, the router, then the other
2674          * switch before reaching the backup.
2675          */
2676         if (slave->state == BOND_STATE_ACTIVE)
2677                 bond_validate_arp(bond, slave, sip, tip);
2678         else
2679                 bond_validate_arp(bond, slave, tip, sip);
2680
2681 out_unlock:
2682         read_unlock(&bond->lock);
2683 out:
2684         dev_kfree_skb(skb);
2685         return NET_RX_SUCCESS;
2686 }
2687
2688 /*
2689  * this function is called regularly to monitor each slave's link
2690  * ensuring that traffic is being sent and received when arp monitoring
2691  * is used in load-balancing mode. if the adapter has been dormant, then an
2692  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2693  * arp monitoring in active backup mode.
2694  */
2695 void bond_loadbalance_arp_mon(struct work_struct *work)
2696 {
2697         struct bonding *bond = container_of(work, struct bonding,
2698                                             arp_work.work);
2699         struct slave *slave, *oldcurrent;
2700         int do_failover = 0;
2701         int delta_in_ticks;
2702         int i;
2703
2704         read_lock(&bond->lock);
2705
2706         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2707
2708         if (bond->kill_timers)
2709                 goto out;
2710
2711         if (bond->slave_cnt == 0)
2712                 goto re_arm;
2713
2714         read_lock(&bond->curr_slave_lock);
2715         oldcurrent = bond->curr_active_slave;
2716         read_unlock(&bond->curr_slave_lock);
2717
2718         /* see if any of the previous devices are up now (i.e. they have
2719          * xmt and rcv traffic). the curr_active_slave does not come into
2720          * the picture unless it is null. also, slave->jiffies is not needed
2721          * here because we send an arp on each slave and give a slave as
2722          * long as it needs to get the tx/rx within the delta.
2723          * TODO: what about up/down delay in arp mode? it wasn't here before
2724          *       so it can wait
2725          */
2726         bond_for_each_slave(bond, slave, i) {
2727                 if (slave->link != BOND_LINK_UP) {
2728                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2729                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2730
2731                                 slave->link  = BOND_LINK_UP;
2732                                 slave->state = BOND_STATE_ACTIVE;
2733
2734                                 /* primary_slave has no meaning in round-robin
2735                                  * mode. the window of a slave being up and
2736                                  * curr_active_slave being null after enslaving
2737                                  * is closed.
2738                                  */
2739                                 if (!oldcurrent) {
2740                                         pr_info("%s: link status definitely up for interface %s, ",
2741                                                 bond->dev->name,
2742                                                 slave->dev->name);
2743                                         do_failover = 1;
2744                                 } else {
2745                                         pr_info("%s: interface %s is now up\n",
2746                                                 bond->dev->name,
2747                                                 slave->dev->name);
2748                                 }
2749                         }
2750                 } else {
2751                         /* slave->link == BOND_LINK_UP */
2752
2753                         /* not all switches will respond to an arp request
2754                          * when the source ip is 0, so don't take the link down
2755                          * if we don't know our ip yet
2756                          */
2757                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2758                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2759
2760                                 slave->link  = BOND_LINK_DOWN;
2761                                 slave->state = BOND_STATE_BACKUP;
2762
2763                                 if (slave->link_failure_count < UINT_MAX)
2764                                         slave->link_failure_count++;
2765
2766                                 pr_info("%s: interface %s is now down.\n",
2767                                         bond->dev->name,
2768                                         slave->dev->name);
2769
2770                                 if (slave == oldcurrent)
2771                                         do_failover = 1;
2772                         }
2773                 }
2774
2775                 /* note: if switch is in round-robin mode, all links
2776                  * must tx arp to ensure all links rx an arp - otherwise
2777                  * links may oscillate or not come up at all; if switch is
2778                  * in something like xor mode, there is nothing we can
2779                  * do - all replies will be rx'ed on same link causing slaves
2780                  * to be unstable during low/no traffic periods
2781                  */
2782                 if (IS_UP(slave->dev))
2783                         bond_arp_send_all(bond, slave);
2784         }
2785
2786         if (do_failover) {
2787                 write_lock_bh(&bond->curr_slave_lock);
2788
2789                 bond_select_active_slave(bond);
2790
2791                 write_unlock_bh(&bond->curr_slave_lock);
2792         }
2793
2794 re_arm:
2795         if (bond->params.arp_interval)
2796                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2797 out:
2798         read_unlock(&bond->lock);
2799 }
2800
2801 /*
2802  * Called to inspect slaves for active-backup mode ARP monitor link state
2803  * changes.  Sets new_link in slaves to specify what action should take
2804  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2805  * to link states must be committed.
2806  *
2807  * Called with bond->lock held for read.
2808  */
2809 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2810 {
2811         struct slave *slave;
2812         int i, commit = 0;
2813
2814         bond_for_each_slave(bond, slave, i) {
2815                 slave->new_link = BOND_LINK_NOCHANGE;
2816
2817                 if (slave->link != BOND_LINK_UP) {
2818                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2819                                            delta_in_ticks)) {
2820                                 slave->new_link = BOND_LINK_UP;
2821                                 commit++;
2822                         }
2823
2824                         continue;
2825                 }
2826
2827                 /*
2828                  * Give slaves 2*delta after being enslaved or made
2829                  * active.  This avoids bouncing, as the last receive
2830                  * times need a full ARP monitor cycle to be updated.
2831                  */
2832                 if (!time_after_eq(jiffies, slave->jiffies +
2833                                    2 * delta_in_ticks))
2834                         continue;
2835
2836                 /*
2837                  * Backup slave is down if:
2838                  * - No current_arp_slave AND
2839                  * - more than 3*delta since last receive AND
2840                  * - the bond has an IP address
2841                  *
2842                  * Note: a non-null current_arp_slave indicates
2843                  * the curr_active_slave went down and we are
2844                  * searching for a new one; under this condition
2845                  * we only take the curr_active_slave down - this
2846                  * gives each slave a chance to tx/rx traffic
2847                  * before being taken out
2848                  */
2849                 if (slave->state == BOND_STATE_BACKUP &&
2850                     !bond->current_arp_slave &&
2851                     time_after(jiffies, slave_last_rx(bond, slave) +
2852                                3 * delta_in_ticks)) {
2853                         slave->new_link = BOND_LINK_DOWN;
2854                         commit++;
2855                 }
2856
2857                 /*
2858                  * Active slave is down if:
2859                  * - more than 2*delta since transmitting OR
2860                  * - (more than 2*delta since receive AND
2861                  *    the bond has an IP address)
2862                  */
2863                 if ((slave->state == BOND_STATE_ACTIVE) &&
2864                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2865                                     2 * delta_in_ticks) ||
2866                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2867                                      + 2 * delta_in_ticks)))) {
2868                         slave->new_link = BOND_LINK_DOWN;
2869                         commit++;
2870                 }
2871         }
2872
2873         return commit;
2874 }
2875
2876 /*
2877  * Called to commit link state changes noted by inspection step of
2878  * active-backup mode ARP monitor.
2879  *
2880  * Called with RTNL and bond->lock for read.
2881  */
2882 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2883 {
2884         struct slave *slave;
2885         int i;
2886
2887         bond_for_each_slave(bond, slave, i) {
2888                 switch (slave->new_link) {
2889                 case BOND_LINK_NOCHANGE:
2890                         continue;
2891
2892                 case BOND_LINK_UP:
2893                         if ((!bond->curr_active_slave &&
2894                              time_before_eq(jiffies,
2895                                             dev_trans_start(slave->dev) +
2896                                             delta_in_ticks)) ||
2897                             bond->curr_active_slave != slave) {
2898                                 slave->link = BOND_LINK_UP;
2899                                 bond->current_arp_slave = NULL;
2900
2901                                 pr_info("%s: link status definitely up for interface %s.\n",
2902                                         bond->dev->name, slave->dev->name);
2903
2904                                 if (!bond->curr_active_slave ||
2905                                     (slave == bond->primary_slave))
2906                                         goto do_failover;
2907
2908                         }
2909
2910                         continue;
2911
2912                 case BOND_LINK_DOWN:
2913                         if (slave->link_failure_count < UINT_MAX)
2914                                 slave->link_failure_count++;
2915
2916                         slave->link = BOND_LINK_DOWN;
2917                         bond_set_slave_inactive_flags(slave);
2918
2919                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2920                                 bond->dev->name, slave->dev->name);
2921
2922                         if (slave == bond->curr_active_slave) {
2923                                 bond->current_arp_slave = NULL;
2924                                 goto do_failover;
2925                         }
2926
2927                         continue;
2928
2929                 default:
2930                         pr_err("%s: impossible: new_link %d on slave %s\n",
2931                                bond->dev->name, slave->new_link,
2932                                slave->dev->name);
2933                         continue;
2934                 }
2935
2936 do_failover:
2937                 ASSERT_RTNL();
2938                 write_lock_bh(&bond->curr_slave_lock);
2939                 bond_select_active_slave(bond);
2940                 write_unlock_bh(&bond->curr_slave_lock);
2941         }
2942
2943         bond_set_carrier(bond);
2944 }
2945
2946 /*
2947  * Send ARP probes for active-backup mode ARP monitor.
2948  *
2949  * Called with bond->lock held for read.
2950  */
2951 static void bond_ab_arp_probe(struct bonding *bond)
2952 {
2953         struct slave *slave;
2954         int i;
2955
2956         read_lock(&bond->curr_slave_lock);
2957
2958         if (bond->current_arp_slave && bond->curr_active_slave)
2959                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2960                         bond->current_arp_slave->dev->name,
2961                         bond->curr_active_slave->dev->name);
2962
2963         if (bond->curr_active_slave) {
2964                 bond_arp_send_all(bond, bond->curr_active_slave);
2965                 read_unlock(&bond->curr_slave_lock);
2966                 return;
2967         }
2968
2969         read_unlock(&bond->curr_slave_lock);
2970
2971         /* if we don't have a curr_active_slave, search for the next available
2972          * backup slave from the current_arp_slave and make it the candidate
2973          * for becoming the curr_active_slave
2974          */
2975
2976         if (!bond->current_arp_slave) {
2977                 bond->current_arp_slave = bond->first_slave;
2978                 if (!bond->current_arp_slave)
2979                         return;
2980         }
2981
2982         bond_set_slave_inactive_flags(bond->current_arp_slave);
2983
2984         /* search for next candidate */
2985         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2986                 if (IS_UP(slave->dev)) {
2987                         slave->link = BOND_LINK_BACK;
2988                         bond_set_slave_active_flags(slave);
2989                         bond_arp_send_all(bond, slave);
2990                         slave->jiffies = jiffies;
2991                         bond->current_arp_slave = slave;
2992                         break;
2993                 }
2994
2995                 /* if the link state is up at this point, we
2996                  * mark it down - this can happen if we have
2997                  * simultaneous link failures and
2998                  * reselect_active_interface doesn't make this
2999                  * one the current slave so it is still marked
3000                  * up when it is actually down
3001                  */
3002                 if (slave->link == BOND_LINK_UP) {
3003                         slave->link = BOND_LINK_DOWN;
3004                         if (slave->link_failure_count < UINT_MAX)
3005                                 slave->link_failure_count++;
3006
3007                         bond_set_slave_inactive_flags(slave);
3008
3009                         pr_info("%s: backup interface %s is now down.\n",
3010                                 bond->dev->name, slave->dev->name);
3011                 }
3012         }
3013 }
3014
3015 void bond_activebackup_arp_mon(struct work_struct *work)
3016 {
3017         struct bonding *bond = container_of(work, struct bonding,
3018                                             arp_work.work);
3019         int delta_in_ticks;
3020
3021         read_lock(&bond->lock);
3022
3023         if (bond->kill_timers)
3024                 goto out;
3025
3026         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3027
3028         if (bond->slave_cnt == 0)
3029                 goto re_arm;
3030
3031         if (bond->send_grat_arp) {
3032                 read_lock(&bond->curr_slave_lock);
3033                 bond_send_gratuitous_arp(bond);
3034                 read_unlock(&bond->curr_slave_lock);
3035         }
3036
3037         if (bond->send_unsol_na) {
3038                 read_lock(&bond->curr_slave_lock);
3039                 bond_send_unsolicited_na(bond);
3040                 read_unlock(&bond->curr_slave_lock);
3041         }
3042
3043         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3044                 read_unlock(&bond->lock);
3045                 rtnl_lock();
3046                 read_lock(&bond->lock);
3047
3048                 bond_ab_arp_commit(bond, delta_in_ticks);
3049
3050                 read_unlock(&bond->lock);
3051                 rtnl_unlock();
3052                 read_lock(&bond->lock);
3053         }
3054
3055         bond_ab_arp_probe(bond);
3056
3057 re_arm:
3058         if (bond->params.arp_interval)
3059                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3060 out:
3061         read_unlock(&bond->lock);
3062 }
3063
3064 /*------------------------------ proc/seq_file-------------------------------*/
3065
3066 #ifdef CONFIG_PROC_FS
3067
3068 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3069         __acquires(&dev_base_lock)
3070         __acquires(&bond->lock)
3071 {
3072         struct bonding *bond = seq->private;
3073         loff_t off = 0;
3074         struct slave *slave;
3075         int i;
3076
3077         /* make sure the bond won't be taken away */
3078         read_lock(&dev_base_lock);
3079         read_lock(&bond->lock);
3080
3081         if (*pos == 0)
3082                 return SEQ_START_TOKEN;
3083
3084         bond_for_each_slave(bond, slave, i) {
3085                 if (++off == *pos)
3086                         return slave;
3087         }
3088
3089         return NULL;
3090 }
3091
3092 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3093 {
3094         struct bonding *bond = seq->private;
3095         struct slave *slave = v;
3096
3097         ++*pos;
3098         if (v == SEQ_START_TOKEN)
3099                 return bond->first_slave;
3100
3101         slave = slave->next;
3102
3103         return (slave == bond->first_slave) ? NULL : slave;
3104 }
3105
3106 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3107         __releases(&bond->lock)
3108         __releases(&dev_base_lock)
3109 {
3110         struct bonding *bond = seq->private;
3111
3112         read_unlock(&bond->lock);
3113         read_unlock(&dev_base_lock);
3114 }
3115
3116 static void bond_info_show_master(struct seq_file *seq)
3117 {
3118         struct bonding *bond = seq->private;
3119         struct slave *curr;
3120         int i;
3121
3122         read_lock(&bond->curr_slave_lock);
3123         curr = bond->curr_active_slave;
3124         read_unlock(&bond->curr_slave_lock);
3125
3126         seq_printf(seq, "Bonding Mode: %s",
3127                    bond_mode_name(bond->params.mode));
3128
3129         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3130             bond->params.fail_over_mac)
3131                 seq_printf(seq, " (fail_over_mac %s)",
3132                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3133
3134         seq_printf(seq, "\n");
3135
3136         if (bond->params.mode == BOND_MODE_XOR ||
3137                 bond->params.mode == BOND_MODE_8023AD) {
3138                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3139                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3140                         bond->params.xmit_policy);
3141         }
3142
3143         if (USES_PRIMARY(bond->params.mode)) {
3144                 seq_printf(seq, "Primary Slave: %s",
3145                            (bond->primary_slave) ?
3146                            bond->primary_slave->dev->name : "None");
3147                 if (bond->primary_slave)
3148                         seq_printf(seq, " (primary_reselect %s)",
3149                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3150
3151                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3152                            (curr) ? curr->dev->name : "None");
3153         }
3154
3155         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3156                    "up" : "down");
3157         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3158         seq_printf(seq, "Up Delay (ms): %d\n",
3159                    bond->params.updelay * bond->params.miimon);
3160         seq_printf(seq, "Down Delay (ms): %d\n",
3161                    bond->params.downdelay * bond->params.miimon);
3162
3163
3164         /* ARP information */
3165         if (bond->params.arp_interval > 0) {
3166                 int printed = 0;
3167                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3168                                 bond->params.arp_interval);
3169
3170                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3171
3172                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3173                         if (!bond->params.arp_targets[i])
3174                                 break;
3175                         if (printed)
3176                                 seq_printf(seq, ",");
3177                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3178                         printed = 1;
3179                 }
3180                 seq_printf(seq, "\n");
3181         }
3182
3183         if (bond->params.mode == BOND_MODE_8023AD) {
3184                 struct ad_info ad_info;
3185
3186                 seq_puts(seq, "\n802.3ad info\n");
3187                 seq_printf(seq, "LACP rate: %s\n",
3188                            (bond->params.lacp_fast) ? "fast" : "slow");
3189                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3190                            ad_select_tbl[bond->params.ad_select].modename);
3191
3192                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3193                         seq_printf(seq, "bond %s has no active aggregator\n",
3194                                    bond->dev->name);
3195                 } else {
3196                         seq_printf(seq, "Active Aggregator Info:\n");
3197
3198                         seq_printf(seq, "\tAggregator ID: %d\n",
3199                                    ad_info.aggregator_id);
3200                         seq_printf(seq, "\tNumber of ports: %d\n",
3201                                    ad_info.ports);
3202                         seq_printf(seq, "\tActor Key: %d\n",
3203                                    ad_info.actor_key);
3204                         seq_printf(seq, "\tPartner Key: %d\n",
3205                                    ad_info.partner_key);
3206                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3207                                    ad_info.partner_system);
3208                 }
3209         }
3210 }
3211
3212 static void bond_info_show_slave(struct seq_file *seq,
3213                                  const struct slave *slave)
3214 {
3215         struct bonding *bond = seq->private;
3216
3217         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3218         seq_printf(seq, "MII Status: %s\n",
3219                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3220         seq_printf(seq, "Link Failure Count: %u\n",
3221                    slave->link_failure_count);
3222
3223         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3224
3225         if (bond->params.mode == BOND_MODE_8023AD) {
3226                 const struct aggregator *agg
3227                         = SLAVE_AD_INFO(slave).port.aggregator;
3228
3229                 if (agg)
3230                         seq_printf(seq, "Aggregator ID: %d\n",
3231                                    agg->aggregator_identifier);
3232                 else
3233                         seq_puts(seq, "Aggregator ID: N/A\n");
3234         }
3235 }
3236
3237 static int bond_info_seq_show(struct seq_file *seq, void *v)
3238 {
3239         if (v == SEQ_START_TOKEN) {
3240                 seq_printf(seq, "%s\n", version);
3241                 bond_info_show_master(seq);
3242         } else
3243                 bond_info_show_slave(seq, v);
3244
3245         return 0;
3246 }
3247
3248 static const struct seq_operations bond_info_seq_ops = {
3249         .start = bond_info_seq_start,
3250         .next  = bond_info_seq_next,
3251         .stop  = bond_info_seq_stop,
3252         .show  = bond_info_seq_show,
3253 };
3254
3255 static int bond_info_open(struct inode *inode, struct file *file)
3256 {
3257         struct seq_file *seq;
3258         struct proc_dir_entry *proc;
3259         int res;
3260
3261         res = seq_open(file, &bond_info_seq_ops);
3262         if (!res) {
3263                 /* recover the pointer buried in proc_dir_entry data */
3264                 seq = file->private_data;
3265                 proc = PDE(inode);
3266                 seq->private = proc->data;
3267         }
3268
3269         return res;
3270 }
3271
3272 static const struct file_operations bond_info_fops = {
3273         .owner   = THIS_MODULE,
3274         .open    = bond_info_open,
3275         .read    = seq_read,
3276         .llseek  = seq_lseek,
3277         .release = seq_release,
3278 };
3279
3280 static void bond_create_proc_entry(struct bonding *bond)
3281 {
3282         struct net_device *bond_dev = bond->dev;
3283         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3284
3285         if (bn->proc_dir) {
3286                 bond->proc_entry = proc_create_data(bond_dev->name,
3287                                                     S_IRUGO, bn->proc_dir,
3288                                                     &bond_info_fops, bond);
3289                 if (bond->proc_entry == NULL)
3290                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3291                                    DRV_NAME, bond_dev->name);
3292                 else
3293                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3294         }
3295 }
3296
3297 static void bond_remove_proc_entry(struct bonding *bond)
3298 {
3299         struct net_device *bond_dev = bond->dev;
3300         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3301
3302         if (bn->proc_dir && bond->proc_entry) {
3303                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3304                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3305                 bond->proc_entry = NULL;
3306         }
3307 }
3308
3309 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3310  * Caller must hold rtnl_lock.
3311  */
3312 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3313 {
3314         if (!bn->proc_dir) {
3315                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3316                 if (!bn->proc_dir)
3317                         pr_warning("Warning: cannot create /proc/net/%s\n",
3318                                    DRV_NAME);
3319         }
3320 }
3321
3322 /* Destroy the bonding directory under /proc/net, if empty.
3323  * Caller must hold rtnl_lock.
3324  */
3325 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3326 {
3327         if (bn->proc_dir) {
3328                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3329                 bn->proc_dir = NULL;
3330         }
3331 }
3332
3333 #else /* !CONFIG_PROC_FS */
3334
3335 static void bond_create_proc_entry(struct bonding *bond)
3336 {
3337 }
3338
3339 static void bond_remove_proc_entry(struct bonding *bond)
3340 {
3341 }
3342
3343 static inline void bond_create_proc_dir(struct bond_net *bn)
3344 {
3345 }
3346
3347 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3348 {
3349 }
3350
3351 #endif /* CONFIG_PROC_FS */
3352
3353
3354 /*-------------------------- netdev event handling --------------------------*/
3355
3356 /*
3357  * Change device name
3358  */
3359 static int bond_event_changename(struct bonding *bond)
3360 {
3361         bond_remove_proc_entry(bond);
3362         bond_create_proc_entry(bond);
3363
3364         return NOTIFY_DONE;
3365 }
3366
3367 static int bond_master_netdev_event(unsigned long event,
3368                                     struct net_device *bond_dev)
3369 {
3370         struct bonding *event_bond = netdev_priv(bond_dev);
3371
3372         switch (event) {
3373         case NETDEV_CHANGENAME:
3374                 return bond_event_changename(event_bond);
3375         default:
3376                 break;
3377         }
3378
3379         return NOTIFY_DONE;
3380 }
3381
3382 static int bond_slave_netdev_event(unsigned long event,
3383                                    struct net_device *slave_dev)
3384 {
3385         struct net_device *bond_dev = slave_dev->master;
3386         struct bonding *bond = netdev_priv(bond_dev);
3387
3388         switch (event) {
3389         case NETDEV_UNREGISTER:
3390                 if (bond_dev) {
3391                         if (bond->setup_by_slave)
3392                                 bond_release_and_destroy(bond_dev, slave_dev);
3393                         else
3394                                 bond_release(bond_dev, slave_dev);
3395                 }
3396                 break;
3397         case NETDEV_CHANGE:
3398                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3399                         struct slave *slave;
3400
3401                         slave = bond_get_slave_by_dev(bond, slave_dev);
3402                         if (slave) {
3403                                 u16 old_speed = slave->speed;
3404                                 u16 old_duplex = slave->duplex;
3405
3406                                 bond_update_speed_duplex(slave);
3407
3408                                 if (bond_is_lb(bond))
3409                                         break;
3410
3411                                 if (old_speed != slave->speed)
3412                                         bond_3ad_adapter_speed_changed(slave);
3413                                 if (old_duplex != slave->duplex)
3414                                         bond_3ad_adapter_duplex_changed(slave);
3415                         }
3416                 }
3417
3418                 break;
3419         case NETDEV_DOWN:
3420                 /*
3421                  * ... Or is it this?
3422                  */
3423                 break;
3424         case NETDEV_CHANGEMTU:
3425                 /*
3426                  * TODO: Should slaves be allowed to
3427                  * independently alter their MTU?  For
3428                  * an active-backup bond, slaves need
3429                  * not be the same type of device, so
3430                  * MTUs may vary.  For other modes,
3431                  * slaves arguably should have the
3432                  * same MTUs. To do this, we'd need to
3433                  * take over the slave's change_mtu
3434                  * function for the duration of their
3435                  * servitude.
3436                  */
3437                 break;
3438         case NETDEV_CHANGENAME:
3439                 /*
3440                  * TODO: handle changing the primary's name
3441                  */
3442                 break;
3443         case NETDEV_FEAT_CHANGE:
3444                 bond_compute_features(bond);
3445                 break;
3446         default:
3447                 break;
3448         }
3449
3450         return NOTIFY_DONE;
3451 }
3452
3453 /*
3454  * bond_netdev_event: handle netdev notifier chain events.
3455  *
3456  * This function receives events for the netdev chain.  The caller (an
3457  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3458  * locks for us to safely manipulate the slave devices (RTNL lock,
3459  * dev_probe_lock).
3460  */
3461 static int bond_netdev_event(struct notifier_block *this,
3462                              unsigned long event, void *ptr)
3463 {
3464         struct net_device *event_dev = (struct net_device *)ptr;
3465
3466         pr_debug("event_dev: %s, event: %lx\n",
3467                  event_dev ? event_dev->name : "None",
3468                  event);
3469
3470         if (!(event_dev->priv_flags & IFF_BONDING))
3471                 return NOTIFY_DONE;
3472
3473         if (event_dev->flags & IFF_MASTER) {
3474                 pr_debug("IFF_MASTER\n");
3475                 return bond_master_netdev_event(event, event_dev);
3476         }
3477
3478         if (event_dev->flags & IFF_SLAVE) {
3479                 pr_debug("IFF_SLAVE\n");
3480                 return bond_slave_netdev_event(event, event_dev);
3481         }
3482
3483         return NOTIFY_DONE;
3484 }
3485
3486 /*
3487  * bond_inetaddr_event: handle inetaddr notifier chain events.
3488  *
3489  * We keep track of device IPs primarily to use as source addresses in
3490  * ARP monitor probes (rather than spewing out broadcasts all the time).
3491  *
3492  * We track one IP for the main device (if it has one), plus one per VLAN.
3493  */
3494 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3495 {
3496         struct in_ifaddr *ifa = ptr;
3497         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3498         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3499         struct bonding *bond;
3500         struct vlan_entry *vlan;
3501
3502         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3503                 if (bond->dev == event_dev) {
3504                         switch (event) {
3505                         case NETDEV_UP:
3506                                 bond->master_ip = ifa->ifa_local;
3507                                 return NOTIFY_OK;
3508                         case NETDEV_DOWN:
3509                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3510                                 return NOTIFY_OK;
3511                         default:
3512                                 return NOTIFY_DONE;
3513                         }
3514                 }
3515
3516                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3517                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3518                         if (vlan_dev == event_dev) {
3519                                 switch (event) {
3520                                 case NETDEV_UP:
3521                                         vlan->vlan_ip = ifa->ifa_local;
3522                                         return NOTIFY_OK;
3523                                 case NETDEV_DOWN:
3524                                         vlan->vlan_ip =
3525                                                 bond_glean_dev_ip(vlan_dev);
3526                                         return NOTIFY_OK;
3527                                 default:
3528                                         return NOTIFY_DONE;
3529                                 }
3530                         }
3531                 }
3532         }
3533         return NOTIFY_DONE;
3534 }
3535
3536 static struct notifier_block bond_netdev_notifier = {
3537         .notifier_call = bond_netdev_event,
3538 };
3539
3540 static struct notifier_block bond_inetaddr_notifier = {
3541         .notifier_call = bond_inetaddr_event,
3542 };
3543
3544 /*-------------------------- Packet type handling ---------------------------*/
3545
3546 /* register to receive lacpdus on a bond */
3547 static void bond_register_lacpdu(struct bonding *bond)
3548 {
3549         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3550
3551         /* initialize packet type */
3552         pk_type->type = PKT_TYPE_LACPDU;
3553         pk_type->dev = bond->dev;
3554         pk_type->func = bond_3ad_lacpdu_recv;
3555
3556         dev_add_pack(pk_type);
3557 }
3558
3559 /* unregister to receive lacpdus on a bond */
3560 static void bond_unregister_lacpdu(struct bonding *bond)
3561 {
3562         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3563 }
3564
3565 void bond_register_arp(struct bonding *bond)
3566 {
3567         struct packet_type *pt = &bond->arp_mon_pt;
3568
3569         if (pt->type)
3570                 return;
3571
3572         pt->type = htons(ETH_P_ARP);
3573         pt->dev = bond->dev;
3574         pt->func = bond_arp_rcv;
3575         dev_add_pack(pt);
3576 }
3577
3578 void bond_unregister_arp(struct bonding *bond)
3579 {
3580         struct packet_type *pt = &bond->arp_mon_pt;
3581
3582         dev_remove_pack(pt);
3583         pt->type = 0;
3584 }
3585
3586 /*---------------------------- Hashing Policies -----------------------------*/
3587
3588 /*
3589  * Hash for the output device based upon layer 2 and layer 3 data. If
3590  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3591  */
3592 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3593 {
3594         struct ethhdr *data = (struct ethhdr *)skb->data;
3595         struct iphdr *iph = ip_hdr(skb);
3596
3597         if (skb->protocol == htons(ETH_P_IP)) {
3598                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3599                         (data->h_dest[5] ^ data->h_source[5])) % count;
3600         }
3601
3602         return (data->h_dest[5] ^ data->h_source[5]) % count;
3603 }
3604
3605 /*
3606  * Hash for the output device based upon layer 3 and layer 4 data. If
3607  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3608  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3609  */
3610 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3611 {
3612         struct ethhdr *data = (struct ethhdr *)skb->data;
3613         struct iphdr *iph = ip_hdr(skb);
3614         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3615         int layer4_xor = 0;
3616
3617         if (skb->protocol == htons(ETH_P_IP)) {
3618                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3619                     (iph->protocol == IPPROTO_TCP ||
3620                      iph->protocol == IPPROTO_UDP)) {
3621                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3622                 }
3623                 return (layer4_xor ^
3624                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3625
3626         }
3627
3628         return (data->h_dest[5] ^ data->h_source[5]) % count;
3629 }
3630
3631 /*
3632  * Hash for the output device based upon layer 2 data
3633  */
3634 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3635 {
3636         struct ethhdr *data = (struct ethhdr *)skb->data;
3637
3638         return (data->h_dest[5] ^ data->h_source[5]) % count;
3639 }
3640
3641 /*-------------------------- Device entry points ----------------------------*/
3642
3643 static int bond_open(struct net_device *bond_dev)
3644 {
3645         struct bonding *bond = netdev_priv(bond_dev);
3646
3647         bond->kill_timers = 0;
3648
3649         if (bond_is_lb(bond)) {
3650                 /* bond_alb_initialize must be called before the timer
3651                  * is started.
3652                  */
3653                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3654                         /* something went wrong - fail the open operation */
3655                         return -ENOMEM;
3656                 }
3657
3658                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3659                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3660         }
3661
3662         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3663                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3664                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3665         }
3666
3667         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3668                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3669                         INIT_DELAYED_WORK(&bond->arp_work,
3670                                           bond_activebackup_arp_mon);
3671                 else
3672                         INIT_DELAYED_WORK(&bond->arp_work,
3673                                           bond_loadbalance_arp_mon);
3674
3675                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3676                 if (bond->params.arp_validate)
3677                         bond_register_arp(bond);
3678         }
3679
3680         if (bond->params.mode == BOND_MODE_8023AD) {
3681                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3682                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3683                 /* register to receive LACPDUs */
3684                 bond_register_lacpdu(bond);
3685                 bond_3ad_initiate_agg_selection(bond, 1);
3686         }
3687
3688         return 0;
3689 }
3690
3691 static int bond_close(struct net_device *bond_dev)
3692 {
3693         struct bonding *bond = netdev_priv(bond_dev);
3694
3695         if (bond->params.mode == BOND_MODE_8023AD) {
3696                 /* Unregister the receive of LACPDUs */
3697                 bond_unregister_lacpdu(bond);
3698         }
3699
3700         if (bond->params.arp_validate)
3701                 bond_unregister_arp(bond);
3702
3703         write_lock_bh(&bond->lock);
3704
3705         bond->send_grat_arp = 0;
3706         bond->send_unsol_na = 0;
3707
3708         /* signal timers not to re-arm */
3709         bond->kill_timers = 1;
3710
3711         write_unlock_bh(&bond->lock);
3712
3713         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3714                 cancel_delayed_work(&bond->mii_work);
3715         }
3716
3717         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3718                 cancel_delayed_work(&bond->arp_work);
3719         }
3720
3721         switch (bond->params.mode) {
3722         case BOND_MODE_8023AD:
3723                 cancel_delayed_work(&bond->ad_work);
3724                 break;
3725         case BOND_MODE_TLB:
3726         case BOND_MODE_ALB:
3727                 cancel_delayed_work(&bond->alb_work);
3728                 break;
3729         default:
3730                 break;
3731         }
3732
3733
3734         if (bond_is_lb(bond)) {
3735                 /* Must be called only after all
3736                  * slaves have been released
3737                  */
3738                 bond_alb_deinitialize(bond);
3739         }
3740
3741         return 0;
3742 }
3743
3744 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3745 {
3746         struct bonding *bond = netdev_priv(bond_dev);
3747         struct net_device_stats *stats = &bond_dev->stats;
3748         struct net_device_stats local_stats;
3749         struct slave *slave;
3750         int i;
3751
3752         memset(&local_stats, 0, sizeof(struct net_device_stats));
3753
3754         read_lock_bh(&bond->lock);
3755
3756         bond_for_each_slave(bond, slave, i) {
3757                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3758
3759                 local_stats.rx_packets += sstats->rx_packets;
3760                 local_stats.rx_bytes += sstats->rx_bytes;
3761                 local_stats.rx_errors += sstats->rx_errors;
3762                 local_stats.rx_dropped += sstats->rx_dropped;
3763
3764                 local_stats.tx_packets += sstats->tx_packets;
3765                 local_stats.tx_bytes += sstats->tx_bytes;
3766                 local_stats.tx_errors += sstats->tx_errors;
3767                 local_stats.tx_dropped += sstats->tx_dropped;
3768
3769                 local_stats.multicast += sstats->multicast;
3770                 local_stats.collisions += sstats->collisions;
3771
3772                 local_stats.rx_length_errors += sstats->rx_length_errors;
3773                 local_stats.rx_over_errors += sstats->rx_over_errors;
3774                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3775                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3776                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3777                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3778
3779                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3780                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3781                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3782                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3783                 local_stats.tx_window_errors += sstats->tx_window_errors;
3784         }
3785
3786         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3787
3788         read_unlock_bh(&bond->lock);
3789
3790         return stats;
3791 }
3792
3793 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3794 {
3795         struct net_device *slave_dev = NULL;
3796         struct ifbond k_binfo;
3797         struct ifbond __user *u_binfo = NULL;
3798         struct ifslave k_sinfo;
3799         struct ifslave __user *u_sinfo = NULL;
3800         struct mii_ioctl_data *mii = NULL;
3801         int res = 0;
3802
3803         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3804
3805         switch (cmd) {
3806         case SIOCGMIIPHY:
3807                 mii = if_mii(ifr);
3808                 if (!mii)
3809                         return -EINVAL;
3810
3811                 mii->phy_id = 0;
3812                 /* Fall Through */
3813         case SIOCGMIIREG:
3814                 /*
3815                  * We do this again just in case we were called by SIOCGMIIREG
3816                  * instead of SIOCGMIIPHY.
3817                  */
3818                 mii = if_mii(ifr);
3819                 if (!mii)
3820                         return -EINVAL;
3821
3822
3823                 if (mii->reg_num == 1) {
3824                         struct bonding *bond = netdev_priv(bond_dev);
3825                         mii->val_out = 0;
3826                         read_lock(&bond->lock);
3827                         read_lock(&bond->curr_slave_lock);
3828                         if (netif_carrier_ok(bond->dev))
3829                                 mii->val_out = BMSR_LSTATUS;
3830
3831                         read_unlock(&bond->curr_slave_lock);
3832                         read_unlock(&bond->lock);
3833                 }
3834
3835                 return 0;
3836         case BOND_INFO_QUERY_OLD:
3837         case SIOCBONDINFOQUERY:
3838                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3839
3840                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3841                         return -EFAULT;
3842
3843                 res = bond_info_query(bond_dev, &k_binfo);
3844                 if (res == 0 &&
3845                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3846                         return -EFAULT;
3847
3848                 return res;
3849         case BOND_SLAVE_INFO_QUERY_OLD:
3850         case SIOCBONDSLAVEINFOQUERY:
3851                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3852
3853                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3854                         return -EFAULT;
3855
3856                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3857                 if (res == 0 &&
3858                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3859                         return -EFAULT;
3860
3861                 return res;
3862         default:
3863                 /* Go on */
3864                 break;
3865         }
3866
3867         if (!capable(CAP_NET_ADMIN))
3868                 return -EPERM;
3869
3870         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3871
3872         pr_debug("slave_dev=%p:\n", slave_dev);
3873
3874         if (!slave_dev)
3875                 res = -ENODEV;
3876         else {
3877                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3878                 switch (cmd) {
3879                 case BOND_ENSLAVE_OLD:
3880                 case SIOCBONDENSLAVE:
3881                         res = bond_enslave(bond_dev, slave_dev);
3882                         break;
3883                 case BOND_RELEASE_OLD:
3884                 case SIOCBONDRELEASE:
3885                         res = bond_release(bond_dev, slave_dev);
3886                         break;
3887                 case BOND_SETHWADDR_OLD:
3888                 case SIOCBONDSETHWADDR:
3889                         res = bond_sethwaddr(bond_dev, slave_dev);
3890                         break;
3891                 case BOND_CHANGE_ACTIVE_OLD:
3892                 case SIOCBONDCHANGEACTIVE:
3893                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3894                         break;
3895                 default:
3896                         res = -EOPNOTSUPP;
3897                 }
3898
3899                 dev_put(slave_dev);
3900         }
3901
3902         return res;
3903 }
3904
3905 static void bond_set_multicast_list(struct net_device *bond_dev)
3906 {
3907         struct bonding *bond = netdev_priv(bond_dev);
3908         struct dev_mc_list *dmi;
3909
3910         /*
3911          * Do promisc before checking multicast_mode
3912          */
3913         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3914                 /*
3915                  * FIXME: Need to handle the error when one of the multi-slaves
3916                  * encounters error.
3917                  */
3918                 bond_set_promiscuity(bond, 1);
3919
3920
3921         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3922                 bond_set_promiscuity(bond, -1);
3923
3924
3925         /* set allmulti flag to slaves */
3926         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3927                 /*
3928                  * FIXME: Need to handle the error when one of the multi-slaves
3929                  * encounters error.
3930                  */
3931                 bond_set_allmulti(bond, 1);
3932
3933
3934         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3935                 bond_set_allmulti(bond, -1);
3936
3937
3938         read_lock(&bond->lock);
3939
3940         bond->flags = bond_dev->flags;
3941
3942         /* looking for addresses to add to slaves' mc list */
3943         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3944                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
3945                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3946         }
3947
3948         /* looking for addresses to delete from slaves' list */
3949         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3950                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
3951                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3952         }
3953
3954         /* save master's multicast list */
3955         bond_mc_list_destroy(bond);
3956         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3957
3958         read_unlock(&bond->lock);
3959 }
3960
3961 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3962 {
3963         struct bonding *bond = netdev_priv(dev);
3964         struct slave *slave = bond->first_slave;
3965
3966         if (slave) {
3967                 const struct net_device_ops *slave_ops
3968                         = slave->dev->netdev_ops;
3969                 if (slave_ops->ndo_neigh_setup)
3970                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3971         }
3972         return 0;
3973 }
3974
3975 /*
3976  * Change the MTU of all of a master's slaves to match the master
3977  */
3978 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3979 {
3980         struct bonding *bond = netdev_priv(bond_dev);
3981         struct slave *slave, *stop_at;
3982         int res = 0;
3983         int i;
3984
3985         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3986                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3987
3988         /* Can't hold bond->lock with bh disabled here since
3989          * some base drivers panic. On the other hand we can't
3990          * hold bond->lock without bh disabled because we'll
3991          * deadlock. The only solution is to rely on the fact
3992          * that we're under rtnl_lock here, and the slaves
3993          * list won't change. This doesn't solve the problem
3994          * of setting the slave's MTU while it is
3995          * transmitting, but the assumption is that the base
3996          * driver can handle that.
3997          *
3998          * TODO: figure out a way to safely iterate the slaves
3999          * list, but without holding a lock around the actual
4000          * call to the base driver.
4001          */
4002
4003         bond_for_each_slave(bond, slave, i) {
4004                 pr_debug("s %p s->p %p c_m %p\n",
4005                          slave,
4006                          slave->prev,
4007                          slave->dev->netdev_ops->ndo_change_mtu);
4008
4009                 res = dev_set_mtu(slave->dev, new_mtu);
4010
4011                 if (res) {
4012                         /* If we failed to set the slave's mtu to the new value
4013                          * we must abort the operation even in ACTIVE_BACKUP
4014                          * mode, because if we allow the backup slaves to have
4015                          * different mtu values than the active slave we'll
4016                          * need to change their mtu when doing a failover. That
4017                          * means changing their mtu from timer context, which
4018                          * is probably not a good idea.
4019                          */
4020                         pr_debug("err %d %s\n", res, slave->dev->name);
4021                         goto unwind;
4022                 }
4023         }
4024
4025         bond_dev->mtu = new_mtu;
4026
4027         return 0;
4028
4029 unwind:
4030         /* unwind from head to the slave that failed */
4031         stop_at = slave;
4032         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4033                 int tmp_res;
4034
4035                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4036                 if (tmp_res) {
4037                         pr_debug("unwind err %d dev %s\n",
4038                                  tmp_res, slave->dev->name);
4039                 }
4040         }
4041
4042         return res;
4043 }
4044
4045 /*
4046  * Change HW address
4047  *
4048  * Note that many devices must be down to change the HW address, and
4049  * downing the master releases all slaves.  We can make bonds full of
4050  * bonding devices to test this, however.
4051  */
4052 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4053 {
4054         struct bonding *bond = netdev_priv(bond_dev);
4055         struct sockaddr *sa = addr, tmp_sa;
4056         struct slave *slave, *stop_at;
4057         int res = 0;
4058         int i;
4059
4060         if (bond->params.mode == BOND_MODE_ALB)
4061                 return bond_alb_set_mac_address(bond_dev, addr);
4062
4063
4064         pr_debug("bond=%p, name=%s\n",
4065                  bond, bond_dev ? bond_dev->name : "None");
4066
4067         /*
4068          * If fail_over_mac is set to active, do nothing and return
4069          * success.  Returning an error causes ifenslave to fail.
4070          */
4071         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4072                 return 0;
4073
4074         if (!is_valid_ether_addr(sa->sa_data))
4075                 return -EADDRNOTAVAIL;
4076
4077         /* Can't hold bond->lock with bh disabled here since
4078          * some base drivers panic. On the other hand we can't
4079          * hold bond->lock without bh disabled because we'll
4080          * deadlock. The only solution is to rely on the fact
4081          * that we're under rtnl_lock here, and the slaves
4082          * list won't change. This doesn't solve the problem
4083          * of setting the slave's hw address while it is
4084          * transmitting, but the assumption is that the base
4085          * driver can handle that.
4086          *
4087          * TODO: figure out a way to safely iterate the slaves
4088          * list, but without holding a lock around the actual
4089          * call to the base driver.
4090          */
4091
4092         bond_for_each_slave(bond, slave, i) {
4093                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4094                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4095
4096                 if (slave_ops->ndo_set_mac_address == NULL) {
4097                         res = -EOPNOTSUPP;
4098                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4099                         goto unwind;
4100                 }
4101
4102                 res = dev_set_mac_address(slave->dev, addr);
4103                 if (res) {
4104                         /* TODO: consider downing the slave
4105                          * and retry ?
4106                          * User should expect communications
4107                          * breakage anyway until ARP finish
4108                          * updating, so...
4109                          */
4110                         pr_debug("err %d %s\n", res, slave->dev->name);
4111                         goto unwind;
4112                 }
4113         }
4114
4115         /* success */
4116         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4117         return 0;
4118
4119 unwind:
4120         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4121         tmp_sa.sa_family = bond_dev->type;
4122
4123         /* unwind from head to the slave that failed */
4124         stop_at = slave;
4125         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4126                 int tmp_res;
4127
4128                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4129                 if (tmp_res) {
4130                         pr_debug("unwind err %d dev %s\n",
4131                                  tmp_res, slave->dev->name);
4132                 }
4133         }
4134
4135         return res;
4136 }
4137
4138 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4139 {
4140         struct bonding *bond = netdev_priv(bond_dev);
4141         struct slave *slave, *start_at;
4142         int i, slave_no, res = 1;
4143
4144         read_lock(&bond->lock);
4145
4146         if (!BOND_IS_OK(bond))
4147                 goto out;
4148
4149         /*
4150          * Concurrent TX may collide on rr_tx_counter; we accept that
4151          * as being rare enough not to justify using an atomic op here
4152          */
4153         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4154
4155         bond_for_each_slave(bond, slave, i) {
4156                 slave_no--;
4157                 if (slave_no < 0)
4158                         break;
4159         }
4160
4161         start_at = slave;
4162         bond_for_each_slave_from(bond, slave, i, start_at) {
4163                 if (IS_UP(slave->dev) &&
4164                     (slave->link == BOND_LINK_UP) &&
4165                     (slave->state == BOND_STATE_ACTIVE)) {
4166                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4167                         break;
4168                 }
4169         }
4170
4171 out:
4172         if (res) {
4173                 /* no suitable interface, frame not sent */
4174                 dev_kfree_skb(skb);
4175         }
4176         read_unlock(&bond->lock);
4177         return NETDEV_TX_OK;
4178 }
4179
4180
4181 /*
4182  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4183  * the bond has a usable interface.
4184  */
4185 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4186 {
4187         struct bonding *bond = netdev_priv(bond_dev);
4188         int res = 1;
4189
4190         read_lock(&bond->lock);
4191         read_lock(&bond->curr_slave_lock);
4192
4193         if (!BOND_IS_OK(bond))
4194                 goto out;
4195
4196         if (!bond->curr_active_slave)
4197                 goto out;
4198
4199         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4200
4201 out:
4202         if (res)
4203                 /* no suitable interface, frame not sent */
4204                 dev_kfree_skb(skb);
4205
4206         read_unlock(&bond->curr_slave_lock);
4207         read_unlock(&bond->lock);
4208         return NETDEV_TX_OK;
4209 }
4210
4211 /*
4212  * In bond_xmit_xor() , we determine the output device by using a pre-
4213  * determined xmit_hash_policy(), If the selected device is not enabled,
4214  * find the next active slave.
4215  */
4216 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4217 {
4218         struct bonding *bond = netdev_priv(bond_dev);
4219         struct slave *slave, *start_at;
4220         int slave_no;
4221         int i;
4222         int res = 1;
4223
4224         read_lock(&bond->lock);
4225
4226         if (!BOND_IS_OK(bond))
4227                 goto out;
4228
4229         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4230
4231         bond_for_each_slave(bond, slave, i) {
4232                 slave_no--;
4233                 if (slave_no < 0)
4234                         break;
4235         }
4236
4237         start_at = slave;
4238
4239         bond_for_each_slave_from(bond, slave, i, start_at) {
4240                 if (IS_UP(slave->dev) &&
4241                     (slave->link == BOND_LINK_UP) &&
4242                     (slave->state == BOND_STATE_ACTIVE)) {
4243                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4244                         break;
4245                 }
4246         }
4247
4248 out:
4249         if (res) {
4250                 /* no suitable interface, frame not sent */
4251                 dev_kfree_skb(skb);
4252         }
4253         read_unlock(&bond->lock);
4254         return NETDEV_TX_OK;
4255 }
4256
4257 /*
4258  * in broadcast mode, we send everything to all usable interfaces.
4259  */
4260 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4261 {
4262         struct bonding *bond = netdev_priv(bond_dev);
4263         struct slave *slave, *start_at;
4264         struct net_device *tx_dev = NULL;
4265         int i;
4266         int res = 1;
4267
4268         read_lock(&bond->lock);
4269
4270         if (!BOND_IS_OK(bond))
4271                 goto out;
4272
4273         read_lock(&bond->curr_slave_lock);
4274         start_at = bond->curr_active_slave;
4275         read_unlock(&bond->curr_slave_lock);
4276
4277         if (!start_at)
4278                 goto out;
4279
4280         bond_for_each_slave_from(bond, slave, i, start_at) {
4281                 if (IS_UP(slave->dev) &&
4282                     (slave->link == BOND_LINK_UP) &&
4283                     (slave->state == BOND_STATE_ACTIVE)) {
4284                         if (tx_dev) {
4285                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4286                                 if (!skb2) {
4287                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4288                                                bond_dev->name);
4289                                         continue;
4290                                 }
4291
4292                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4293                                 if (res) {
4294                                         dev_kfree_skb(skb2);
4295                                         continue;
4296                                 }
4297                         }
4298                         tx_dev = slave->dev;
4299                 }
4300         }
4301
4302         if (tx_dev)
4303                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4304
4305 out:
4306         if (res)
4307                 /* no suitable interface, frame not sent */
4308                 dev_kfree_skb(skb);
4309
4310         /* frame sent to all suitable interfaces */
4311         read_unlock(&bond->lock);
4312         return NETDEV_TX_OK;
4313 }
4314
4315 /*------------------------- Device initialization ---------------------------*/
4316
4317 static void bond_set_xmit_hash_policy(struct bonding *bond)
4318 {
4319         switch (bond->params.xmit_policy) {
4320         case BOND_XMIT_POLICY_LAYER23:
4321                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4322                 break;
4323         case BOND_XMIT_POLICY_LAYER34:
4324                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4325                 break;
4326         case BOND_XMIT_POLICY_LAYER2:
4327         default:
4328                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4329                 break;
4330         }
4331 }
4332
4333 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4334 {
4335         const struct bonding *bond = netdev_priv(dev);
4336
4337         switch (bond->params.mode) {
4338         case BOND_MODE_ROUNDROBIN:
4339                 return bond_xmit_roundrobin(skb, dev);
4340         case BOND_MODE_ACTIVEBACKUP:
4341                 return bond_xmit_activebackup(skb, dev);
4342         case BOND_MODE_XOR:
4343                 return bond_xmit_xor(skb, dev);
4344         case BOND_MODE_BROADCAST:
4345                 return bond_xmit_broadcast(skb, dev);
4346         case BOND_MODE_8023AD:
4347                 return bond_3ad_xmit_xor(skb, dev);
4348         case BOND_MODE_ALB:
4349         case BOND_MODE_TLB:
4350                 return bond_alb_xmit(skb, dev);
4351         default:
4352                 /* Should never happen, mode already checked */
4353                 pr_err("%s: Error: Unknown bonding mode %d\n",
4354                        dev->name, bond->params.mode);
4355                 WARN_ON_ONCE(1);
4356                 dev_kfree_skb(skb);
4357                 return NETDEV_TX_OK;
4358         }
4359 }
4360
4361
4362 /*
4363  * set bond mode specific net device operations
4364  */
4365 void bond_set_mode_ops(struct bonding *bond, int mode)
4366 {
4367         struct net_device *bond_dev = bond->dev;
4368
4369         switch (mode) {
4370         case BOND_MODE_ROUNDROBIN:
4371                 break;
4372         case BOND_MODE_ACTIVEBACKUP:
4373                 break;
4374         case BOND_MODE_XOR:
4375                 bond_set_xmit_hash_policy(bond);
4376                 break;
4377         case BOND_MODE_BROADCAST:
4378                 break;
4379         case BOND_MODE_8023AD:
4380                 bond_set_master_3ad_flags(bond);
4381                 bond_set_xmit_hash_policy(bond);
4382                 break;
4383         case BOND_MODE_ALB:
4384                 bond_set_master_alb_flags(bond);
4385                 /* FALLTHRU */
4386         case BOND_MODE_TLB:
4387                 break;
4388         default:
4389                 /* Should never happen, mode already checked */
4390                 pr_err("%s: Error: Unknown bonding mode %d\n",
4391                        bond_dev->name, mode);
4392                 break;
4393         }
4394 }
4395
4396 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4397                                     struct ethtool_drvinfo *drvinfo)
4398 {
4399         strncpy(drvinfo->driver, DRV_NAME, 32);
4400         strncpy(drvinfo->version, DRV_VERSION, 32);
4401         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4402 }
4403
4404 static const struct ethtool_ops bond_ethtool_ops = {
4405         .get_drvinfo            = bond_ethtool_get_drvinfo,
4406         .get_link               = ethtool_op_get_link,
4407         .get_tx_csum            = ethtool_op_get_tx_csum,
4408         .get_sg                 = ethtool_op_get_sg,
4409         .get_tso                = ethtool_op_get_tso,
4410         .get_ufo                = ethtool_op_get_ufo,
4411         .get_flags              = ethtool_op_get_flags,
4412 };
4413
4414 static const struct net_device_ops bond_netdev_ops = {
4415         .ndo_init               = bond_init,
4416         .ndo_uninit             = bond_uninit,
4417         .ndo_open               = bond_open,
4418         .ndo_stop               = bond_close,
4419         .ndo_start_xmit         = bond_start_xmit,
4420         .ndo_get_stats          = bond_get_stats,
4421         .ndo_do_ioctl           = bond_do_ioctl,
4422         .ndo_set_multicast_list = bond_set_multicast_list,
4423         .ndo_change_mtu         = bond_change_mtu,
4424         .ndo_set_mac_address    = bond_set_mac_address,
4425         .ndo_neigh_setup        = bond_neigh_setup,
4426         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4427         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4428         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4429 };
4430
4431 static void bond_setup(struct net_device *bond_dev)
4432 {
4433         struct bonding *bond = netdev_priv(bond_dev);
4434
4435         /* initialize rwlocks */
4436         rwlock_init(&bond->lock);
4437         rwlock_init(&bond->curr_slave_lock);
4438
4439         bond->params = bonding_defaults;
4440
4441         /* Initialize pointers */
4442         bond->dev = bond_dev;
4443         INIT_LIST_HEAD(&bond->vlan_list);
4444
4445         /* Initialize the device entry points */
4446         ether_setup(bond_dev);
4447         bond_dev->netdev_ops = &bond_netdev_ops;
4448         bond_dev->ethtool_ops = &bond_ethtool_ops;
4449         bond_set_mode_ops(bond, bond->params.mode);
4450
4451         bond_dev->destructor = free_netdev;
4452
4453         /* Initialize the device options */
4454         bond_dev->tx_queue_len = 0;
4455         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4456         bond_dev->priv_flags |= IFF_BONDING;
4457         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4458
4459         if (bond->params.arp_interval)
4460                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4461
4462         /* At first, we block adding VLANs. That's the only way to
4463          * prevent problems that occur when adding VLANs over an
4464          * empty bond. The block will be removed once non-challenged
4465          * slaves are enslaved.
4466          */
4467         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4468
4469         /* don't acquire bond device's netif_tx_lock when
4470          * transmitting */
4471         bond_dev->features |= NETIF_F_LLTX;
4472
4473         /* By default, we declare the bond to be fully
4474          * VLAN hardware accelerated capable. Special
4475          * care is taken in the various xmit functions
4476          * when there are slaves that are not hw accel
4477          * capable
4478          */
4479         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4480                                NETIF_F_HW_VLAN_RX |
4481                                NETIF_F_HW_VLAN_FILTER);
4482
4483 }
4484
4485 static void bond_work_cancel_all(struct bonding *bond)
4486 {
4487         write_lock_bh(&bond->lock);
4488         bond->kill_timers = 1;
4489         write_unlock_bh(&bond->lock);
4490
4491         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4492                 cancel_delayed_work(&bond->mii_work);
4493
4494         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4495                 cancel_delayed_work(&bond->arp_work);
4496
4497         if (bond->params.mode == BOND_MODE_ALB &&
4498             delayed_work_pending(&bond->alb_work))
4499                 cancel_delayed_work(&bond->alb_work);
4500
4501         if (bond->params.mode == BOND_MODE_8023AD &&
4502             delayed_work_pending(&bond->ad_work))
4503                 cancel_delayed_work(&bond->ad_work);
4504 }
4505
4506 /*
4507 * Destroy a bonding device.
4508 * Must be under rtnl_lock when this function is called.
4509 */
4510 static void bond_uninit(struct net_device *bond_dev)
4511 {
4512         struct bonding *bond = netdev_priv(bond_dev);
4513
4514         /* Release the bonded slaves */
4515         bond_release_all(bond_dev);
4516
4517         list_del(&bond->bond_list);
4518
4519         bond_work_cancel_all(bond);
4520
4521         bond_remove_proc_entry(bond);
4522
4523         if (bond->wq)
4524                 destroy_workqueue(bond->wq);
4525
4526         netif_addr_lock_bh(bond_dev);
4527         bond_mc_list_destroy(bond);
4528         netif_addr_unlock_bh(bond_dev);
4529 }
4530
4531 /*------------------------- Module initialization ---------------------------*/
4532
4533 /*
4534  * Convert string input module parms.  Accept either the
4535  * number of the mode or its string name.  A bit complicated because
4536  * some mode names are substrings of other names, and calls from sysfs
4537  * may have whitespace in the name (trailing newlines, for example).
4538  */
4539 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4540 {
4541         int modeint = -1, i, rv;
4542         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4543
4544         for (p = (char *)buf; *p; p++)
4545                 if (!(isdigit(*p) || isspace(*p)))
4546                         break;
4547
4548         if (*p)
4549                 rv = sscanf(buf, "%20s", modestr);
4550         else
4551                 rv = sscanf(buf, "%d", &modeint);
4552
4553         if (!rv)
4554                 return -1;
4555
4556         for (i = 0; tbl[i].modename; i++) {
4557                 if (modeint == tbl[i].mode)
4558                         return tbl[i].mode;
4559                 if (strcmp(modestr, tbl[i].modename) == 0)
4560                         return tbl[i].mode;
4561         }
4562
4563         return -1;
4564 }
4565
4566 static int bond_check_params(struct bond_params *params)
4567 {
4568         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4569
4570         /*
4571          * Convert string parameters.
4572          */
4573         if (mode) {
4574                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4575                 if (bond_mode == -1) {
4576                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4577                                mode == NULL ? "NULL" : mode);
4578                         return -EINVAL;
4579                 }
4580         }
4581
4582         if (xmit_hash_policy) {
4583                 if ((bond_mode != BOND_MODE_XOR) &&
4584                     (bond_mode != BOND_MODE_8023AD)) {
4585                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4586                                bond_mode_name(bond_mode));
4587                 } else {
4588                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4589                                                         xmit_hashtype_tbl);
4590                         if (xmit_hashtype == -1) {
4591                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4592                                        xmit_hash_policy == NULL ? "NULL" :
4593                                        xmit_hash_policy);
4594                                 return -EINVAL;
4595                         }
4596                 }
4597         }
4598
4599         if (lacp_rate) {
4600                 if (bond_mode != BOND_MODE_8023AD) {
4601                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4602                                 bond_mode_name(bond_mode));
4603                 } else {
4604                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4605                         if (lacp_fast == -1) {
4606                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4607                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4608                                 return -EINVAL;
4609                         }
4610                 }
4611         }
4612
4613         if (ad_select) {
4614                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4615                 if (params->ad_select == -1) {
4616                         pr_err("Error: Invalid ad_select \"%s\"\n",
4617                                ad_select == NULL ? "NULL" : ad_select);
4618                         return -EINVAL;
4619                 }
4620
4621                 if (bond_mode != BOND_MODE_8023AD) {
4622                         pr_warning("ad_select param only affects 802.3ad mode\n");
4623                 }
4624         } else {
4625                 params->ad_select = BOND_AD_STABLE;
4626         }
4627
4628         if (max_bonds < 0) {
4629                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4630                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4631                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4632         }
4633
4634         if (miimon < 0) {
4635                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4636                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4637                 miimon = BOND_LINK_MON_INTERV;
4638         }
4639
4640         if (updelay < 0) {
4641                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4642                            updelay, INT_MAX);
4643                 updelay = 0;
4644         }
4645
4646         if (downdelay < 0) {
4647                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4648                            downdelay, INT_MAX);
4649                 downdelay = 0;
4650         }
4651
4652         if ((use_carrier != 0) && (use_carrier != 1)) {
4653                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4654                            use_carrier);
4655                 use_carrier = 1;
4656         }
4657
4658         if (num_grat_arp < 0 || num_grat_arp > 255) {
4659                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1 \n",
4660                            num_grat_arp);
4661                 num_grat_arp = 1;
4662         }
4663
4664         if (num_unsol_na < 0 || num_unsol_na > 255) {
4665                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1 \n",
4666                            num_unsol_na);
4667                 num_unsol_na = 1;
4668         }
4669
4670         /* reset values for 802.3ad */
4671         if (bond_mode == BOND_MODE_8023AD) {
4672                 if (!miimon) {
4673                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4674                         pr_warning("Forcing miimon to 100msec\n");
4675                         miimon = 100;
4676                 }
4677         }
4678
4679         /* reset values for TLB/ALB */
4680         if ((bond_mode == BOND_MODE_TLB) ||
4681             (bond_mode == BOND_MODE_ALB)) {
4682                 if (!miimon) {
4683                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4684                         pr_warning("Forcing miimon to 100msec\n");
4685                         miimon = 100;
4686                 }
4687         }
4688
4689         if (bond_mode == BOND_MODE_ALB) {
4690                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4691                           updelay);
4692         }
4693
4694         if (!miimon) {
4695                 if (updelay || downdelay) {
4696                         /* just warn the user the up/down delay will have
4697                          * no effect since miimon is zero...
4698                          */
4699                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4700                                    updelay, downdelay);
4701                 }
4702         } else {
4703                 /* don't allow arp monitoring */
4704                 if (arp_interval) {
4705                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4706                                    miimon, arp_interval);
4707                         arp_interval = 0;
4708                 }
4709
4710                 if ((updelay % miimon) != 0) {
4711                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4712                                    updelay, miimon,
4713                                    (updelay / miimon) * miimon);
4714                 }
4715
4716                 updelay /= miimon;
4717
4718                 if ((downdelay % miimon) != 0) {
4719                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4720                                    downdelay, miimon,
4721                                    (downdelay / miimon) * miimon);
4722                 }
4723
4724                 downdelay /= miimon;
4725         }
4726
4727         if (arp_interval < 0) {
4728                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4729                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4730                 arp_interval = BOND_LINK_ARP_INTERV;
4731         }
4732
4733         for (arp_ip_count = 0;
4734              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4735              arp_ip_count++) {
4736                 /* not complete check, but should be good enough to
4737                    catch mistakes */
4738                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4739                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4740                                    arp_ip_target[arp_ip_count]);
4741                         arp_interval = 0;
4742                 } else {
4743                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4744                         arp_target[arp_ip_count] = ip;
4745                 }
4746         }
4747
4748         if (arp_interval && !arp_ip_count) {
4749                 /* don't allow arping if no arp_ip_target given... */
4750                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4751                            arp_interval);
4752                 arp_interval = 0;
4753         }
4754
4755         if (arp_validate) {
4756                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4757                         pr_err("arp_validate only supported in active-backup mode\n");
4758                         return -EINVAL;
4759                 }
4760                 if (!arp_interval) {
4761                         pr_err("arp_validate requires arp_interval\n");
4762                         return -EINVAL;
4763                 }
4764
4765                 arp_validate_value = bond_parse_parm(arp_validate,
4766                                                      arp_validate_tbl);
4767                 if (arp_validate_value == -1) {
4768                         pr_err("Error: invalid arp_validate \"%s\"\n",
4769                                arp_validate == NULL ? "NULL" : arp_validate);
4770                         return -EINVAL;
4771                 }
4772         } else
4773                 arp_validate_value = 0;
4774
4775         if (miimon) {
4776                 pr_info("MII link monitoring set to %d ms\n", miimon);
4777         } else if (arp_interval) {
4778                 int i;
4779
4780                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4781                         arp_interval,
4782                         arp_validate_tbl[arp_validate_value].modename,
4783                         arp_ip_count);
4784
4785                 for (i = 0; i < arp_ip_count; i++)
4786                         pr_info(" %s", arp_ip_target[i]);
4787
4788                 pr_info("\n");
4789
4790         } else if (max_bonds) {
4791                 /* miimon and arp_interval not set, we need one so things
4792                  * work as expected, see bonding.txt for details
4793                  */
4794                 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4795         }
4796
4797         if (primary && !USES_PRIMARY(bond_mode)) {
4798                 /* currently, using a primary only makes sense
4799                  * in active backup, TLB or ALB modes
4800                  */
4801                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4802                            primary, bond_mode_name(bond_mode));
4803                 primary = NULL;
4804         }
4805
4806         if (primary && primary_reselect) {
4807                 primary_reselect_value = bond_parse_parm(primary_reselect,
4808                                                          pri_reselect_tbl);
4809                 if (primary_reselect_value == -1) {
4810                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4811                                primary_reselect ==
4812                                         NULL ? "NULL" : primary_reselect);
4813                         return -EINVAL;
4814                 }
4815         } else {
4816                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4817         }
4818
4819         if (fail_over_mac) {
4820                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4821                                                       fail_over_mac_tbl);
4822                 if (fail_over_mac_value == -1) {
4823                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4824                                arp_validate == NULL ? "NULL" : arp_validate);
4825                         return -EINVAL;
4826                 }
4827
4828                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4829                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4830         } else {
4831                 fail_over_mac_value = BOND_FOM_NONE;
4832         }
4833
4834         /* fill params struct with the proper values */
4835         params->mode = bond_mode;
4836         params->xmit_policy = xmit_hashtype;
4837         params->miimon = miimon;
4838         params->num_grat_arp = num_grat_arp;
4839         params->num_unsol_na = num_unsol_na;
4840         params->arp_interval = arp_interval;
4841         params->arp_validate = arp_validate_value;
4842         params->updelay = updelay;
4843         params->downdelay = downdelay;
4844         params->use_carrier = use_carrier;
4845         params->lacp_fast = lacp_fast;
4846         params->primary[0] = 0;
4847         params->primary_reselect = primary_reselect_value;
4848         params->fail_over_mac = fail_over_mac_value;
4849
4850         if (primary) {
4851                 strncpy(params->primary, primary, IFNAMSIZ);
4852                 params->primary[IFNAMSIZ - 1] = 0;
4853         }
4854
4855         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4856
4857         return 0;
4858 }
4859
4860 static struct lock_class_key bonding_netdev_xmit_lock_key;
4861 static struct lock_class_key bonding_netdev_addr_lock_key;
4862
4863 static void bond_set_lockdep_class_one(struct net_device *dev,
4864                                        struct netdev_queue *txq,
4865                                        void *_unused)
4866 {
4867         lockdep_set_class(&txq->_xmit_lock,
4868                           &bonding_netdev_xmit_lock_key);
4869 }
4870
4871 static void bond_set_lockdep_class(struct net_device *dev)
4872 {
4873         lockdep_set_class(&dev->addr_list_lock,
4874                           &bonding_netdev_addr_lock_key);
4875         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4876 }
4877
4878 /*
4879  * Called from registration process
4880  */
4881 static int bond_init(struct net_device *bond_dev)
4882 {
4883         struct bonding *bond = netdev_priv(bond_dev);
4884         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4885
4886         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4887
4888         bond->wq = create_singlethread_workqueue(bond_dev->name);
4889         if (!bond->wq)
4890                 return -ENOMEM;
4891
4892         bond_set_lockdep_class(bond_dev);
4893
4894         netif_carrier_off(bond_dev);
4895
4896         bond_create_proc_entry(bond);
4897         list_add_tail(&bond->bond_list, &bn->dev_list);
4898
4899         bond_prepare_sysfs_group(bond);
4900         return 0;
4901 }
4902
4903 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4904 {
4905         if (tb[IFLA_ADDRESS]) {
4906                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4907                         return -EINVAL;
4908                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4909                         return -EADDRNOTAVAIL;
4910         }
4911         return 0;
4912 }
4913
4914 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4915         .kind           = "bond",
4916         .priv_size      = sizeof(struct bonding),
4917         .setup          = bond_setup,
4918         .validate       = bond_validate,
4919 };
4920
4921 /* Create a new bond based on the specified name and bonding parameters.
4922  * If name is NULL, obtain a suitable "bond%d" name for us.
4923  * Caller must NOT hold rtnl_lock; we need to release it here before we
4924  * set up our sysfs entries.
4925  */
4926 int bond_create(struct net *net, const char *name)
4927 {
4928         struct net_device *bond_dev;
4929         int res;
4930
4931         rtnl_lock();
4932
4933         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4934                                 bond_setup);
4935         if (!bond_dev) {
4936                 pr_err("%s: eek! can't alloc netdev!\n", name);
4937                 res = -ENOMEM;
4938                 goto out;
4939         }
4940
4941         dev_net_set(bond_dev, net);
4942         bond_dev->rtnl_link_ops = &bond_link_ops;
4943
4944         if (!name) {
4945                 res = dev_alloc_name(bond_dev, "bond%d");
4946                 if (res < 0)
4947                         goto out_netdev;
4948         }
4949
4950         res = register_netdevice(bond_dev);
4951         if (res < 0)
4952                 goto out_netdev;
4953
4954 out:
4955         rtnl_unlock();
4956         return res;
4957 out_netdev:
4958         free_netdev(bond_dev);
4959         goto out;
4960 }
4961
4962 static int __net_init bond_net_init(struct net *net)
4963 {
4964         struct bond_net *bn = net_generic(net, bond_net_id);
4965
4966         bn->net = net;
4967         INIT_LIST_HEAD(&bn->dev_list);
4968
4969         bond_create_proc_dir(bn);
4970
4971         return 0;
4972 }
4973
4974 static void __net_exit bond_net_exit(struct net *net)
4975 {
4976         struct bond_net *bn = net_generic(net, bond_net_id);
4977
4978         bond_destroy_proc_dir(bn);
4979 }
4980
4981 static struct pernet_operations bond_net_ops = {
4982         .init = bond_net_init,
4983         .exit = bond_net_exit,
4984         .id   = &bond_net_id,
4985         .size = sizeof(struct bond_net),
4986 };
4987
4988 static int __init bonding_init(void)
4989 {
4990         int i;
4991         int res;
4992
4993         pr_info("%s", version);
4994
4995         res = bond_check_params(&bonding_defaults);
4996         if (res)
4997                 goto out;
4998
4999         res = register_pernet_subsys(&bond_net_ops);
5000         if (res)
5001                 goto out;
5002
5003         res = rtnl_link_register(&bond_link_ops);
5004         if (res)
5005                 goto err_link;
5006
5007         for (i = 0; i < max_bonds; i++) {
5008                 res = bond_create(&init_net, NULL);
5009                 if (res)
5010                         goto err;
5011         }
5012
5013         res = bond_create_sysfs();
5014         if (res)
5015                 goto err;
5016
5017         register_netdevice_notifier(&bond_netdev_notifier);
5018         register_inetaddr_notifier(&bond_inetaddr_notifier);
5019         bond_register_ipv6_notifier();
5020 out:
5021         return res;
5022 err:
5023         rtnl_link_unregister(&bond_link_ops);
5024 err_link:
5025         unregister_pernet_subsys(&bond_net_ops);
5026         goto out;
5027
5028 }
5029
5030 static void __exit bonding_exit(void)
5031 {
5032         unregister_netdevice_notifier(&bond_netdev_notifier);
5033         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5034         bond_unregister_ipv6_notifier();
5035
5036         bond_destroy_sysfs();
5037
5038         rtnl_link_unregister(&bond_link_ops);
5039         unregister_pernet_subsys(&bond_net_ops);
5040 }
5041
5042 module_init(bonding_init);
5043 module_exit(bonding_exit);
5044 MODULE_LICENSE("GPL");
5045 MODULE_VERSION(DRV_VERSION);
5046 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5047 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5048 MODULE_ALIAS_RTNL_LINK("bond");