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/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <linux/preempt.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include <net/netns/generic.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval = BOND_LINK_ARP_INTERV;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *fail_over_mac;
 107 static int all_slaves_active = 0;
 108 static struct bond_params bonding_defaults;
 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 110
 111 module_param(max_bonds, int, 0);
 112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 113 module_param(tx_queues, int, 0);
 114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 117                                "failover event (alias of num_unsol_na)");
 118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 120                                "failover event (alias of num_grat_arp)");
 121 module_param(miimon, int, 0);
 122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 123 module_param(updelay, int, 0);
 124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 125 module_param(downdelay, int, 0);
 126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 127                             "in milliseconds");
 128 module_param(use_carrier, int, 0);
 129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 130                               "0 for off, 1 for on (default)");
 131 module_param(mode, charp, 0);
 132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 133                        "1 for active-backup, 2 for balance-xor, "
 134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 135                        "6 for balance-alb");
 136 module_param(primary, charp, 0);
 137 MODULE_PARM_DESC(primary, "Primary network device to use");
 138 module_param(primary_reselect, charp, 0);
 139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 140                                    "once it comes up; "
 141                                    "0 for always (default), "
 142                                    "1 for only if speed of primary is "
 143                                    "better, "
 144                                    "2 for only on active slave "
 145                                    "failure");
 146 module_param(lacp_rate, charp, 0);
 147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 148                             "0 for slow, 1 for fast");
 149 module_param(ad_select, charp, 0);
 150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 151                             "0 for stable (default), 1 for bandwidth, "
 152                             "2 for count");
 153 module_param(xmit_hash_policy, charp, 0);
 154 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 155                                    "0 for layer 2 (default), 1 for layer 3+4, "
 156                                    "2 for layer 2+3");
 157 module_param(arp_interval, int, 0);
 158 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 159 module_param_array(arp_ip_target, charp, NULL, 0);
 160 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 161 module_param(arp_validate, charp, 0);
 162 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 163                                "0 for none (default), 1 for active, "
 164                                "2 for backup, 3 for all");
 165 module_param(fail_over_mac, charp, 0);
 166 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 167                                 "the same MAC; 0 for none (default), "
 168                                 "1 for active, 2 for follow");
 169 module_param(all_slaves_active, int, 0);
 170 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 171                                      "by setting active flag for all slaves; "
 172                                      "0 for never (default), 1 for always.");
 173 module_param(resend_igmp, int, 0);
 174 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 175                               "link failure");
 176
 177 /*----------------------------- Global variables ----------------------------*/
 178
 179 #ifdef CONFIG_NET_POLL_CONTROLLER
 180 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 181 #endif
 182
 183 int bond_net_id __read_mostly;
 184
 185 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 186 static int arp_ip_count;
 187 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 188 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 189 static int lacp_fast;
 190
 191 const struct bond_parm_tbl bond_lacp_tbl[] = {
 192 {       "slow",         AD_LACP_SLOW},
 193 {       "fast",         AD_LACP_FAST},
 194 {       NULL,           -1},
 195 };
 196
 197 const struct bond_parm_tbl bond_mode_tbl[] = {
 198 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 199 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 200 {       "balance-xor",          BOND_MODE_XOR},
 201 {       "broadcast",            BOND_MODE_BROADCAST},
 202 {       "802.3ad",              BOND_MODE_8023AD},
 203 {       "balance-tlb",          BOND_MODE_TLB},
 204 {       "balance-alb",          BOND_MODE_ALB},
 205 {       NULL,                   -1},
 206 };
 207
 208 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 209 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 210 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 211 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl arp_validate_tbl[] = {
 216 {       "none",                 BOND_ARP_VALIDATE_NONE},
 217 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 218 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 219 {       "all",                  BOND_ARP_VALIDATE_ALL},
 220 {       NULL,                   -1},
 221 };
 222
 223 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 224 {       "none",                 BOND_FOM_NONE},
 225 {       "active",               BOND_FOM_ACTIVE},
 226 {       "follow",               BOND_FOM_FOLLOW},
 227 {       NULL,                   -1},
 228 };
 229
 230 const struct bond_parm_tbl pri_reselect_tbl[] = {
 231 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 232 {       "better",               BOND_PRI_RESELECT_BETTER},
 233 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 234 {       NULL,                   -1},
 235 };
 236
 237 struct bond_parm_tbl ad_select_tbl[] = {
 238 {       "stable",       BOND_AD_STABLE},
 239 {       "bandwidth",    BOND_AD_BANDWIDTH},
 240 {       "count",        BOND_AD_COUNT},
 241 {       NULL,           -1},
 242 };
 243
 244 /*-------------------------- Forward declarations ---------------------------*/
 245
 246 static int bond_init(struct net_device *bond_dev);
 247 static void bond_uninit(struct net_device *bond_dev);
 248
 249 /*---------------------------- General routines -----------------------------*/
 250
 251 const char *bond_mode_name(int mode)
 252 {
 253         static const char *names[] = {
 254                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 255                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 256                 [BOND_MODE_XOR] = "load balancing (xor)",
 257                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 258                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 259                 [BOND_MODE_TLB] = "transmit load balancing",
 260                 [BOND_MODE_ALB] = "adaptive load balancing",
 261         };
 262
 263         if (mode < 0 || mode > BOND_MODE_ALB)
 264                 return "unknown";
 265
 266         return names[mode];
 267 }
 268
 269 /*---------------------------------- VLAN -----------------------------------*/
 270
 271 /**
 272  * bond_add_vlan - add a new vlan id on bond
 273  * @bond: bond that got the notification
 274  * @vlan_id: the vlan id to add
 275  *
 276  * Returns -ENOMEM if allocation failed.
 277  */
 278 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 279 {
 280         struct vlan_entry *vlan;
 281
 282         pr_debug("bond: %s, vlan id %d\n",
 283                  (bond ? bond->dev->name : "None"), vlan_id);
 284
 285         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 286         if (!vlan)
 287                 return -ENOMEM;
 288
 289         INIT_LIST_HEAD(&vlan->vlan_list);
 290         vlan->vlan_id = vlan_id;
 291
 292         write_lock_bh(&bond->lock);
 293
 294         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 295
 296         write_unlock_bh(&bond->lock);
 297
 298         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 299
 300         return 0;
 301 }
 302
 303 /**
 304  * bond_del_vlan - delete a vlan id from bond
 305  * @bond: bond that got the notification
 306  * @vlan_id: the vlan id to delete
 307  *
 308  * returns -ENODEV if @vlan_id was not found in @bond.
 309  */
 310 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 311 {
 312         struct vlan_entry *vlan;
 313         int res = -ENODEV;
 314
 315         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 316
 317         block_netpoll_tx();
 318         write_lock_bh(&bond->lock);
 319
 320         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 321                 if (vlan->vlan_id == vlan_id) {
 322                         list_del(&vlan->vlan_list);
 323
 324                         if (bond_is_lb(bond))
 325                                 bond_alb_clear_vlan(bond, vlan_id);
 326
 327                         pr_debug("removed VLAN ID %d from bond %s\n",
 328                                  vlan_id, bond->dev->name);
 329
 330                         kfree(vlan);
 331
 332                         if (list_empty(&bond->vlan_list) &&
 333                             (bond->slave_cnt == 0)) {
 334                                 /* Last VLAN removed and no slaves, so
 335                                  * restore block on adding VLANs. This will
 336                                  * be removed once new slaves that are not
 337                                  * VLAN challenged will be added.
 338                                  */
 339                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 340                         }
 341
 342                         res = 0;
 343                         goto out;
 344                 }
 345         }
 346
 347         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 348                  vlan_id, bond->dev->name);
 349
 350 out:
 351         write_unlock_bh(&bond->lock);
 352         unblock_netpoll_tx();
 353         return res;
 354 }
 355
 356 /**
 357  * bond_next_vlan - safely skip to the next item in the vlans list.
 358  * @bond: the bond we're working on
 359  * @curr: item we're advancing from
 360  *
 361  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 362  * or @curr->next otherwise (even if it is @curr itself again).
 363  *
 364  * Caller must hold bond->lock
 365  */
 366 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 367 {
 368         struct vlan_entry *next, *last;
 369
 370         if (list_empty(&bond->vlan_list))
 371                 return NULL;
 372
 373         if (!curr) {
 374                 next = list_entry(bond->vlan_list.next,
 375                                   struct vlan_entry, vlan_list);
 376         } else {
 377                 last = list_entry(bond->vlan_list.prev,
 378                                   struct vlan_entry, vlan_list);
 379                 if (last == curr) {
 380                         next = list_entry(bond->vlan_list.next,
 381                                           struct vlan_entry, vlan_list);
 382                 } else {
 383                         next = list_entry(curr->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 }
 386         }
 387
 388         return next;
 389 }
 390
 391 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 392
 393 /**
 394  * bond_dev_queue_xmit - Prepare skb for xmit.
 395  *
 396  * @bond: bond device that got this skb for tx.
 397  * @skb: hw accel VLAN tagged skb to transmit
 398  * @slave_dev: slave that is supposed to xmit this skbuff
 399  */
 400 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 401                         struct net_device *slave_dev)
 402 {
 403         skb->dev = slave_dev;
 404         skb->priority = 1;
 405
 406         skb->queue_mapping = bond_queue_mapping(skb);
 407
 408         if (unlikely(netpoll_tx_running(slave_dev)))
 409                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 410         else
 411                 dev_queue_xmit(skb);
 412
 413         return 0;
 414 }
 415
 416 /*
 417  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 418  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 419  * lock because:
 420  * a. This operation is performed in IOCTL context,
 421  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 422  * c. Holding a lock with BH disabled while directly calling a base driver
 423  *    entry point is generally a BAD idea.
 424  *
 425  * The design of synchronization/protection for this operation in the 8021q
 426  * module is good for one or more VLAN devices over a single physical device
 427  * and cannot be extended for a teaming solution like bonding, so there is a
 428  * potential race condition here where a net device from the vlan group might
 429  * be referenced (either by a base driver or the 8021q code) while it is being
 430  * removed from the system. However, it turns out we're not making matters
 431  * worse, and if it works for regular VLAN usage it will work here too.
 432 */
 433
 434 /**
 435  * bond_vlan_rx_register - Propagates registration to slaves
 436  * @bond_dev: bonding net device that got called
 437  * @grp: vlan group being registered
 438  */
 439 static void bond_vlan_rx_register(struct net_device *bond_dev,
 440                                   struct vlan_group *grp)
 441 {
 442         struct bonding *bond = netdev_priv(bond_dev);
 443         struct slave *slave;
 444         int i;
 445
 446         write_lock_bh(&bond->lock);
 447         bond->vlgrp = grp;
 448         write_unlock_bh(&bond->lock);
 449
 450         bond_for_each_slave(bond, slave, i) {
 451                 struct net_device *slave_dev = slave->dev;
 452                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 453
 454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 455                     slave_ops->ndo_vlan_rx_register) {
 456                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 457                 }
 458         }
 459 }
 460
 461 /**
 462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being added
 465  */
 466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i) {
 473                 struct net_device *slave_dev = slave->dev;
 474                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 475
 476                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 477                     slave_ops->ndo_vlan_rx_add_vid) {
 478                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 479                 }
 480         }
 481
 482         res = bond_add_vlan(bond, vid);
 483         if (res) {
 484                 pr_err("%s: Error: Failed to add vlan id %d\n",
 485                        bond_dev->name, vid);
 486         }
 487 }
 488
 489 /**
 490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 491  * @bond_dev: bonding net device that got called
 492  * @vid: vlan id being removed
 493  */
 494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 495 {
 496         struct bonding *bond = netdev_priv(bond_dev);
 497         struct slave *slave;
 498         struct net_device *vlan_dev;
 499         int i, res;
 500
 501         bond_for_each_slave(bond, slave, i) {
 502                 struct net_device *slave_dev = slave->dev;
 503                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 504
 505                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 506                     slave_ops->ndo_vlan_rx_kill_vid) {
 507                         /* Save and then restore vlan_dev in the grp array,
 508                          * since the slave's driver might clear it.
 509                          */
 510                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 511                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 512                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 513                 }
 514         }
 515
 516         res = bond_del_vlan(bond, vid);
 517         if (res) {
 518                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 519                        bond_dev->name, vid);
 520         }
 521 }
 522
 523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 524 {
 525         struct vlan_entry *vlan;
 526         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 527
 528         if (!bond->vlgrp)
 529                 return;
 530
 531         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 532             slave_ops->ndo_vlan_rx_register)
 533                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 534
 535         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 536             !(slave_ops->ndo_vlan_rx_add_vid))
 537                 return;
 538
 539         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 540                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 541 }
 542
 543 static void bond_del_vlans_from_slave(struct bonding *bond,
 544                                       struct net_device *slave_dev)
 545 {
 546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 547         struct vlan_entry *vlan;
 548         struct net_device *vlan_dev;
 549
 550         if (!bond->vlgrp)
 551                 return;
 552
 553         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 554             !(slave_ops->ndo_vlan_rx_kill_vid))
 555                 goto unreg;
 556
 557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 558                 if (!vlan->vlan_id)
 559                         continue;
 560                 /* Save and then restore vlan_dev in the grp array,
 561                  * since the slave's driver might clear it.
 562                  */
 563                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 564                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 565                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 566         }
 567
 568 unreg:
 569         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 570             slave_ops->ndo_vlan_rx_register)
 571                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 572 }
 573
 574 /*------------------------------- Link status -------------------------------*/
 575
 576 /*
 577  * Set the carrier state for the master according to the state of its
 578  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 579  * do special 802.3ad magic.
 580  *
 581  * Returns zero if carrier state does not change, nonzero if it does.
 582  */
 583 static int bond_set_carrier(struct bonding *bond)
 584 {
 585         struct slave *slave;
 586         int i;
 587
 588         if (bond->slave_cnt == 0)
 589                 goto down;
 590
 591         if (bond->params.mode == BOND_MODE_8023AD)
 592                 return bond_3ad_set_carrier(bond);
 593
 594         bond_for_each_slave(bond, slave, i) {
 595                 if (slave->link == BOND_LINK_UP) {
 596                         if (!netif_carrier_ok(bond->dev)) {
 597                                 netif_carrier_on(bond->dev);
 598                                 return 1;
 599                         }
 600                         return 0;
 601                 }
 602         }
 603
 604 down:
 605         if (netif_carrier_ok(bond->dev)) {
 606                 netif_carrier_off(bond->dev);
 607                 return 1;
 608         }
 609         return 0;
 610 }
 611
 612 /*
 613  * Get link speed and duplex from the slave's base driver
 614  * using ethtool. If for some reason the call fails or the
 615  * values are invalid, fake speed and duplex to 100/Full
 616  * and return error.
 617  */
 618 static int bond_update_speed_duplex(struct slave *slave)
 619 {
 620         struct net_device *slave_dev = slave->dev;
 621         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 622         u32 slave_speed;
 623         int res;
 624
 625         /* Fake speed and duplex */
 626         slave->speed = SPEED_100;
 627         slave->duplex = DUPLEX_FULL;
 628
 629         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 630                 return -1;
 631
 632         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 633         if (res < 0)
 634                 return -1;
 635
 636         slave_speed = ethtool_cmd_speed(&etool);
 637         switch (slave_speed) {
 638         case SPEED_10:
 639         case SPEED_100:
 640         case SPEED_1000:
 641         case SPEED_10000:
 642                 break;
 643         default:
 644                 return -1;
 645         }
 646
 647         switch (etool.duplex) {
 648         case DUPLEX_FULL:
 649         case DUPLEX_HALF:
 650                 break;
 651         default:
 652                 return -1;
 653         }
 654
 655         slave->speed = slave_speed;
 656         slave->duplex = etool.duplex;
 657
 658         return 0;
 659 }
 660
 661 /*
 662  * if <dev> supports MII link status reporting, check its link status.
 663  *
 664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 665  * depending upon the setting of the use_carrier parameter.
 666  *
 667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 668  * can't tell and just pretend it is), or 0, meaning that the link is
 669  * down.
 670  *
 671  * If reporting is non-zero, instead of faking link up, return -1 if
 672  * both ETHTOOL and MII ioctls fail (meaning the device does not
 673  * support them).  If use_carrier is set, return whatever it says.
 674  * It'd be nice if there was a good way to tell if a driver supports
 675  * netif_carrier, but there really isn't.
 676  */
 677 static int bond_check_dev_link(struct bonding *bond,
 678                                struct net_device *slave_dev, int reporting)
 679 {
 680         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 681         int (*ioctl)(struct net_device *, struct ifreq *, int);
 682         struct ifreq ifr;
 683         struct mii_ioctl_data *mii;
 684
 685         if (!reporting && !netif_running(slave_dev))
 686                 return 0;
 687
 688         if (bond->params.use_carrier)
 689                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 690
 691         /* Try to get link status using Ethtool first. */
 692         if (slave_dev->ethtool_ops) {
 693                 if (slave_dev->ethtool_ops->get_link) {
 694                         u32 link;
 695
 696                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 697
 698                         return link ? BMSR_LSTATUS : 0;
 699                 }
 700         }
 701
 702         /* Ethtool can't be used, fallback to MII ioctls. */
 703         ioctl = slave_ops->ndo_do_ioctl;
 704         if (ioctl) {
 705                 /* TODO: set pointer to correct ioctl on a per team member */
 706                 /*       bases to make this more efficient. that is, once  */
 707                 /*       we determine the correct ioctl, we will always    */
 708                 /*       call it and not the others for that team          */
 709                 /*       member.                                           */
 710
 711                 /*
 712                  * We cannot assume that SIOCGMIIPHY will also read a
 713                  * register; not all network drivers (e.g., e100)
 714                  * support that.
 715                  */
 716
 717                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 718                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 719                 mii = if_mii(&ifr);
 720                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 721                         mii->reg_num = MII_BMSR;
 722                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 723                                 return mii->val_out & BMSR_LSTATUS;
 724                 }
 725         }
 726
 727         /*
 728          * If reporting, report that either there's no dev->do_ioctl,
 729          * or both SIOCGMIIREG and get_link failed (meaning that we
 730          * cannot report link status).  If not reporting, pretend
 731          * we're ok.
 732          */
 733         return reporting ? -1 : BMSR_LSTATUS;
 734 }
 735
 736 /*----------------------------- Multicast list ------------------------------*/
 737
 738 /*
 739  * Push the promiscuity flag down to appropriate slaves
 740  */
 741 static int bond_set_promiscuity(struct bonding *bond, int inc)
 742 {
 743         int err = 0;
 744         if (USES_PRIMARY(bond->params.mode)) {
 745                 /* write lock already acquired */
 746                 if (bond->curr_active_slave) {
 747                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 748                                                   inc);
 749                 }
 750         } else {
 751                 struct slave *slave;
 752                 int i;
 753                 bond_for_each_slave(bond, slave, i) {
 754                         err = dev_set_promiscuity(slave->dev, inc);
 755                         if (err)
 756                                 return err;
 757                 }
 758         }
 759         return err;
 760 }
 761
 762 /*
 763  * Push the allmulti flag down to all slaves
 764  */
 765 static int bond_set_allmulti(struct bonding *bond, int inc)
 766 {
 767         int err = 0;
 768         if (USES_PRIMARY(bond->params.mode)) {
 769                 /* write lock already acquired */
 770                 if (bond->curr_active_slave) {
 771                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 772                                                inc);
 773                 }
 774         } else {
 775                 struct slave *slave;
 776                 int i;
 777                 bond_for_each_slave(bond, slave, i) {
 778                         err = dev_set_allmulti(slave->dev, inc);
 779                         if (err)
 780                                 return err;
 781                 }
 782         }
 783         return err;
 784 }
 785
 786 /*
 787  * Add a Multicast address to slaves
 788  * according to mode
 789  */
 790 static void bond_mc_add(struct bonding *bond, void *addr)
 791 {
 792         if (USES_PRIMARY(bond->params.mode)) {
 793                 /* write lock already acquired */
 794                 if (bond->curr_active_slave)
 795                         dev_mc_add(bond->curr_active_slave->dev, addr);
 796         } else {
 797                 struct slave *slave;
 798                 int i;
 799
 800                 bond_for_each_slave(bond, slave, i)
 801                         dev_mc_add(slave->dev, addr);
 802         }
 803 }
 804
 805 /*
 806  * Remove a multicast address from slave
 807  * according to mode
 808  */
 809 static void bond_mc_del(struct bonding *bond, void *addr)
 810 {
 811         if (USES_PRIMARY(bond->params.mode)) {
 812                 /* write lock already acquired */
 813                 if (bond->curr_active_slave)
 814                         dev_mc_del(bond->curr_active_slave->dev, addr);
 815         } else {
 816                 struct slave *slave;
 817                 int i;
 818                 bond_for_each_slave(bond, slave, i) {
 819                         dev_mc_del(slave->dev, addr);
 820                 }
 821         }
 822 }
 823
 824
 825 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 826 {
 827         struct in_device *in_dev;
 828
 829         rcu_read_lock();
 830         in_dev = __in_dev_get_rcu(dev);
 831         if (in_dev)
 832                 ip_mc_rejoin_groups(in_dev);
 833         rcu_read_unlock();
 834 }
 835
 836 /*
 837  * Retrieve the list of registered multicast addresses for the bonding
 838  * device and retransmit an IGMP JOIN request to the current active
 839  * slave.
 840  */
 841 static void bond_resend_igmp_join_requests(struct bonding *bond)
 842 {
 843         struct net_device *vlan_dev;
 844         struct vlan_entry *vlan;
 845
 846         read_lock(&bond->lock);
 847
 848         /* rejoin all groups on bond device */
 849         __bond_resend_igmp_join_requests(bond->dev);
 850
 851         /* rejoin all groups on vlan devices */
 852         if (bond->vlgrp) {
 853                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 854                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 855                                                          vlan->vlan_id);
 856                         if (vlan_dev)
 857                                 __bond_resend_igmp_join_requests(vlan_dev);
 858                 }
 859         }
 860
 861         if (--bond->igmp_retrans > 0)
 862                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 863
 864         read_unlock(&bond->lock);
 865 }
 866
 867 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 868 {
 869         struct bonding *bond = container_of(work, struct bonding,
 870                                             mcast_work.work);
 871         bond_resend_igmp_join_requests(bond);
 872 }
 873
 874 /*
 875  * flush all members of flush->mc_list from device dev->mc_list
 876  */
 877 static void bond_mc_list_flush(struct net_device *bond_dev,
 878                                struct net_device *slave_dev)
 879 {
 880         struct bonding *bond = netdev_priv(bond_dev);
 881         struct netdev_hw_addr *ha;
 882
 883         netdev_for_each_mc_addr(ha, bond_dev)
 884                 dev_mc_del(slave_dev, ha->addr);
 885
 886         if (bond->params.mode == BOND_MODE_8023AD) {
 887                 /* del lacpdu mc addr from mc list */
 888                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 889
 890                 dev_mc_del(slave_dev, lacpdu_multicast);
 891         }
 892 }
 893
 894 /*--------------------------- Active slave change ---------------------------*/
 895
 896 /*
 897  * Update the mc list and multicast-related flags for the new and
 898  * old active slaves (if any) according to the multicast mode, and
 899  * promiscuous flags unconditionally.
 900  */
 901 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 902                          struct slave *old_active)
 903 {
 904         struct netdev_hw_addr *ha;
 905
 906         if (!USES_PRIMARY(bond->params.mode))
 907                 /* nothing to do -  mc list is already up-to-date on
 908                  * all slaves
 909                  */
 910                 return;
 911
 912         if (old_active) {
 913                 if (bond->dev->flags & IFF_PROMISC)
 914                         dev_set_promiscuity(old_active->dev, -1);
 915
 916                 if (bond->dev->flags & IFF_ALLMULTI)
 917                         dev_set_allmulti(old_active->dev, -1);
 918
 919                 netdev_for_each_mc_addr(ha, bond->dev)
 920                         dev_mc_del(old_active->dev, ha->addr);
 921         }
 922
 923         if (new_active) {
 924                 /* FIXME: Signal errors upstream. */
 925                 if (bond->dev->flags & IFF_PROMISC)
 926                         dev_set_promiscuity(new_active->dev, 1);
 927
 928                 if (bond->dev->flags & IFF_ALLMULTI)
 929                         dev_set_allmulti(new_active->dev, 1);
 930
 931                 netdev_for_each_mc_addr(ha, bond->dev)
 932                         dev_mc_add(new_active->dev, ha->addr);
 933         }
 934 }
 935
 936 /*
 937  * bond_do_fail_over_mac
 938  *
 939  * Perform special MAC address swapping for fail_over_mac settings
 940  *
 941  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 942  */
 943 static void bond_do_fail_over_mac(struct bonding *bond,
 944                                   struct slave *new_active,
 945                                   struct slave *old_active)
 946         __releases(&bond->curr_slave_lock)
 947         __releases(&bond->lock)
 948         __acquires(&bond->lock)
 949         __acquires(&bond->curr_slave_lock)
 950 {
 951         u8 tmp_mac[ETH_ALEN];
 952         struct sockaddr saddr;
 953         int rv;
 954
 955         switch (bond->params.fail_over_mac) {
 956         case BOND_FOM_ACTIVE:
 957                 if (new_active)
 958                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 959                                new_active->dev->addr_len);
 960                 break;
 961         case BOND_FOM_FOLLOW:
 962                 /*
 963                  * if new_active && old_active, swap them
 964                  * if just old_active, do nothing (going to no active slave)
 965                  * if just new_active, set new_active to bond's MAC
 966                  */
 967                 if (!new_active)
 968                         return;
 969
 970                 write_unlock_bh(&bond->curr_slave_lock);
 971                 read_unlock(&bond->lock);
 972
 973                 if (old_active) {
 974                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 975                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 976                                ETH_ALEN);
 977                         saddr.sa_family = new_active->dev->type;
 978                 } else {
 979                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 980                         saddr.sa_family = bond->dev->type;
 981                 }
 982
 983                 rv = dev_set_mac_address(new_active->dev, &saddr);
 984                 if (rv) {
 985                         pr_err("%s: Error %d setting MAC of slave %s\n",
 986                                bond->dev->name, -rv, new_active->dev->name);
 987                         goto out;
 988                 }
 989
 990                 if (!old_active)
 991                         goto out;
 992
 993                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 994                 saddr.sa_family = old_active->dev->type;
 995
 996                 rv = dev_set_mac_address(old_active->dev, &saddr);
 997                 if (rv)
 998                         pr_err("%s: Error %d setting MAC of slave %s\n",
 999                                bond->dev->name, -rv, new_active->dev->name);
1000 out:
1001                 read_lock(&bond->lock);
1002                 write_lock_bh(&bond->curr_slave_lock);
1003                 break;
1004         default:
1005                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1006                        bond->dev->name, bond->params.fail_over_mac);
1007                 break;
1008         }
1009
1010 }
1011
1012 static bool bond_should_change_active(struct bonding *bond)
1013 {
1014         struct slave *prim = bond->primary_slave;
1015         struct slave *curr = bond->curr_active_slave;
1016
1017         if (!prim || !curr || curr->link != BOND_LINK_UP)
1018                 return true;
1019         if (bond->force_primary) {
1020                 bond->force_primary = false;
1021                 return true;
1022         }
1023         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1024             (prim->speed < curr->speed ||
1025              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1026                 return false;
1027         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1028                 return false;
1029         return true;
1030 }
1031
1032 /**
1033  * find_best_interface - select the best available slave to be the active one
1034  * @bond: our bonding struct
1035  *
1036  * Warning: Caller must hold curr_slave_lock for writing.
1037  */
1038 static struct slave *bond_find_best_slave(struct bonding *bond)
1039 {
1040         struct slave *new_active, *old_active;
1041         struct slave *bestslave = NULL;
1042         int mintime = bond->params.updelay;
1043         int i;
1044
1045         new_active = bond->curr_active_slave;
1046
1047         if (!new_active) { /* there were no active slaves left */
1048                 if (bond->slave_cnt > 0)   /* found one slave */
1049                         new_active = bond->first_slave;
1050                 else
1051                         return NULL; /* still no slave, return NULL */
1052         }
1053
1054         if ((bond->primary_slave) &&
1055             bond->primary_slave->link == BOND_LINK_UP &&
1056             bond_should_change_active(bond)) {
1057                 new_active = bond->primary_slave;
1058         }
1059
1060         /* remember where to stop iterating over the slaves */
1061         old_active = new_active;
1062
1063         bond_for_each_slave_from(bond, new_active, i, old_active) {
1064                 if (new_active->link == BOND_LINK_UP) {
1065                         return new_active;
1066                 } else if (new_active->link == BOND_LINK_BACK &&
1067                            IS_UP(new_active->dev)) {
1068                         /* link up, but waiting for stabilization */
1069                         if (new_active->delay < mintime) {
1070                                 mintime = new_active->delay;
1071                                 bestslave = new_active;
1072                         }
1073                 }
1074         }
1075
1076         return bestslave;
1077 }
1078
1079 static bool bond_should_notify_peers(struct bonding *bond)
1080 {
1081         struct slave *slave = bond->curr_active_slave;
1082
1083         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1084                  bond->dev->name, slave ? slave->dev->name : "NULL");
1085
1086         if (!slave || !bond->send_peer_notif ||
1087             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1088                 return false;
1089
1090         bond->send_peer_notif--;
1091         return true;
1092 }
1093
1094 /**
1095  * change_active_interface - change the active slave into the specified one
1096  * @bond: our bonding struct
1097  * @new: the new slave to make the active one
1098  *
1099  * Set the new slave to the bond's settings and unset them on the old
1100  * curr_active_slave.
1101  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1102  *
1103  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1104  * because it is apparently the best available slave we have, even though its
1105  * updelay hasn't timed out yet.
1106  *
1107  * If new_active is not NULL, caller must hold bond->lock for read and
1108  * curr_slave_lock for write_bh.
1109  */
1110 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1111 {
1112         struct slave *old_active = bond->curr_active_slave;
1113
1114         if (old_active == new_active)
1115                 return;
1116
1117         if (new_active) {
1118                 new_active->jiffies = jiffies;
1119
1120                 if (new_active->link == BOND_LINK_BACK) {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1123                                         bond->dev->name, new_active->dev->name,
1124                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1125                         }
1126
1127                         new_active->delay = 0;
1128                         new_active->link = BOND_LINK_UP;
1129
1130                         if (bond->params.mode == BOND_MODE_8023AD)
1131                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1132
1133                         if (bond_is_lb(bond))
1134                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1135                 } else {
1136                         if (USES_PRIMARY(bond->params.mode)) {
1137                                 pr_info("%s: making interface %s the new active one.\n",
1138                                         bond->dev->name, new_active->dev->name);
1139                         }
1140                 }
1141         }
1142
1143         if (USES_PRIMARY(bond->params.mode))
1144                 bond_mc_swap(bond, new_active, old_active);
1145
1146         if (bond_is_lb(bond)) {
1147                 bond_alb_handle_active_change(bond, new_active);
1148                 if (old_active)
1149                         bond_set_slave_inactive_flags(old_active);
1150                 if (new_active)
1151                         bond_set_slave_active_flags(new_active);
1152         } else {
1153                 bond->curr_active_slave = new_active;
1154         }
1155
1156         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1157                 if (old_active)
1158                         bond_set_slave_inactive_flags(old_active);
1159
1160                 if (new_active) {
1161                         bool should_notify_peers = false;
1162
1163                         bond_set_slave_active_flags(new_active);
1164
1165                         if (bond->params.fail_over_mac)
1166                                 bond_do_fail_over_mac(bond, new_active,
1167                                                       old_active);
1168
1169                         if (netif_running(bond->dev)) {
1170                                 bond->send_peer_notif =
1171                                         bond->params.num_peer_notif;
1172                                 should_notify_peers =
1173                                         bond_should_notify_peers(bond);
1174                         }
1175
1176                         write_unlock_bh(&bond->curr_slave_lock);
1177                         read_unlock(&bond->lock);
1178
1179                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1180                         if (should_notify_peers)
1181                                 netdev_bonding_change(bond->dev,
1182                                                       NETDEV_NOTIFY_PEERS);
1183
1184                         read_lock(&bond->lock);
1185                         write_lock_bh(&bond->curr_slave_lock);
1186                 }
1187         }
1188
1189         /* resend IGMP joins since active slave has changed or
1190          * all were sent on curr_active_slave.
1191          * resend only if bond is brought up with the affected
1192          * bonding modes and the retransmission is enabled */
1193         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1194             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1195              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1196                 bond->igmp_retrans = bond->params.resend_igmp;
1197                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1198         }
1199 }
1200
1201 /**
1202  * bond_select_active_slave - select a new active slave, if needed
1203  * @bond: our bonding struct
1204  *
1205  * This functions should be called when one of the following occurs:
1206  * - The old curr_active_slave has been released or lost its link.
1207  * - The primary_slave has got its link back.
1208  * - A slave has got its link back and there's no old curr_active_slave.
1209  *
1210  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1211  */
1212 void bond_select_active_slave(struct bonding *bond)
1213 {
1214         struct slave *best_slave;
1215         int rv;
1216
1217         best_slave = bond_find_best_slave(bond);
1218         if (best_slave != bond->curr_active_slave) {
1219                 bond_change_active_slave(bond, best_slave);
1220                 rv = bond_set_carrier(bond);
1221                 if (!rv)
1222                         return;
1223
1224                 if (netif_carrier_ok(bond->dev)) {
1225                         pr_info("%s: first active interface up!\n",
1226                                 bond->dev->name);
1227                 } else {
1228                         pr_info("%s: now running without any active interface !\n",
1229                                 bond->dev->name);
1230                 }
1231         }
1232 }
1233
1234 /*--------------------------- slave list handling ---------------------------*/
1235
1236 /*
1237  * This function attaches the slave to the end of list.
1238  *
1239  * bond->lock held for writing by caller.
1240  */
1241 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1242 {
1243         if (bond->first_slave == NULL) { /* attaching the first slave */
1244                 new_slave->next = new_slave;
1245                 new_slave->prev = new_slave;
1246                 bond->first_slave = new_slave;
1247         } else {
1248                 new_slave->next = bond->first_slave;
1249                 new_slave->prev = bond->first_slave->prev;
1250                 new_slave->next->prev = new_slave;
1251                 new_slave->prev->next = new_slave;
1252         }
1253
1254         bond->slave_cnt++;
1255 }
1256
1257 /*
1258  * This function detaches the slave from the list.
1259  * WARNING: no check is made to verify if the slave effectively
1260  * belongs to <bond>.
1261  * Nothing is freed on return, structures are just unchained.
1262  * If any slave pointer in bond was pointing to <slave>,
1263  * it should be changed by the calling function.
1264  *
1265  * bond->lock held for writing by caller.
1266  */
1267 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1268 {
1269         if (slave->next)
1270                 slave->next->prev = slave->prev;
1271
1272         if (slave->prev)
1273                 slave->prev->next = slave->next;
1274
1275         if (bond->first_slave == slave) { /* slave is the first slave */
1276                 if (bond->slave_cnt > 1) { /* there are more slave */
1277                         bond->first_slave = slave->next;
1278                 } else {
1279                         bond->first_slave = NULL; /* slave was the last one */
1280                 }
1281         }
1282
1283         slave->next = NULL;
1284         slave->prev = NULL;
1285         bond->slave_cnt--;
1286 }
1287
1288 #ifdef CONFIG_NET_POLL_CONTROLLER
1289 static inline int slave_enable_netpoll(struct slave *slave)
1290 {
1291         struct netpoll *np;
1292         int err = 0;
1293
1294         np = kzalloc(sizeof(*np), GFP_KERNEL);
1295         err = -ENOMEM;
1296         if (!np)
1297                 goto out;
1298
1299         np->dev = slave->dev;
1300         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1301         err = __netpoll_setup(np);
1302         if (err) {
1303                 kfree(np);
1304                 goto out;
1305         }
1306         slave->np = np;
1307 out:
1308         return err;
1309 }
1310 static inline void slave_disable_netpoll(struct slave *slave)
1311 {
1312         struct netpoll *np = slave->np;
1313
1314         if (!np)
1315                 return;
1316
1317         slave->np = NULL;
1318         synchronize_rcu_bh();
1319         __netpoll_cleanup(np);
1320         kfree(np);
1321 }
1322 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1323 {
1324         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1325                 return false;
1326         if (!slave_dev->netdev_ops->ndo_poll_controller)
1327                 return false;
1328         return true;
1329 }
1330
1331 static void bond_poll_controller(struct net_device *bond_dev)
1332 {
1333 }
1334
1335 static void __bond_netpoll_cleanup(struct bonding *bond)
1336 {
1337         struct slave *slave;
1338         int i;
1339
1340         bond_for_each_slave(bond, slave, i)
1341                 if (IS_UP(slave->dev))
1342                         slave_disable_netpoll(slave);
1343 }
1344 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1345 {
1346         struct bonding *bond = netdev_priv(bond_dev);
1347
1348         read_lock(&bond->lock);
1349         __bond_netpoll_cleanup(bond);
1350         read_unlock(&bond->lock);
1351 }
1352
1353 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1354 {
1355         struct bonding *bond = netdev_priv(dev);
1356         struct slave *slave;
1357         int i, err = 0;
1358
1359         read_lock(&bond->lock);
1360         bond_for_each_slave(bond, slave, i) {
1361                 err = slave_enable_netpoll(slave);
1362                 if (err) {
1363                         __bond_netpoll_cleanup(bond);
1364                         break;
1365                 }
1366         }
1367         read_unlock(&bond->lock);
1368         return err;
1369 }
1370
1371 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1372 {
1373         return bond->dev->npinfo;
1374 }
1375
1376 #else
1377 static inline int slave_enable_netpoll(struct slave *slave)
1378 {
1379         return 0;
1380 }
1381 static inline void slave_disable_netpoll(struct slave *slave)
1382 {
1383 }
1384 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1385 {
1386 }
1387 #endif
1388
1389 /*---------------------------------- IOCTL ----------------------------------*/
1390
1391 static int bond_sethwaddr(struct net_device *bond_dev,
1392                           struct net_device *slave_dev)
1393 {
1394         pr_debug("bond_dev=%p\n", bond_dev);
1395         pr_debug("slave_dev=%p\n", slave_dev);
1396         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1397         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1398         return 0;
1399 }
1400
1401 static u32 bond_fix_features(struct net_device *dev, u32 features)
1402 {
1403         struct slave *slave;
1404         struct bonding *bond = netdev_priv(dev);
1405         u32 mask;
1406         int i;
1407
1408         read_lock(&bond->lock);
1409
1410         if (!bond->first_slave) {
1411                 /* Disable adding VLANs to empty bond. But why? --mq */
1412                 features |= NETIF_F_VLAN_CHALLENGED;
1413                 goto out;
1414         }
1415
1416         mask = features;
1417         features &= ~NETIF_F_ONE_FOR_ALL;
1418         features |= NETIF_F_ALL_FOR_ALL;
1419
1420         bond_for_each_slave(bond, slave, i) {
1421                 features = netdev_increment_features(features,
1422                                                      slave->dev->features,
1423                                                      mask);
1424         }
1425
1426 out:
1427         read_unlock(&bond->lock);
1428         return features;
1429 }
1430
1431 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1432                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1433                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1434
1435 static void bond_compute_features(struct bonding *bond)
1436 {
1437         struct slave *slave;
1438         struct net_device *bond_dev = bond->dev;
1439         u32 vlan_features = BOND_VLAN_FEATURES;
1440         unsigned short max_hard_header_len = ETH_HLEN;
1441         int i;
1442
1443         read_lock(&bond->lock);
1444
1445         if (!bond->first_slave)
1446                 goto done;
1447
1448         bond_for_each_slave(bond, slave, i) {
1449                 vlan_features = netdev_increment_features(vlan_features,
1450                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1451
1452                 if (slave->dev->hard_header_len > max_hard_header_len)
1453                         max_hard_header_len = slave->dev->hard_header_len;
1454         }
1455
1456 done:
1457         bond_dev->vlan_features = vlan_features;
1458         bond_dev->hard_header_len = max_hard_header_len;
1459
1460         read_unlock(&bond->lock);
1461
1462         netdev_change_features(bond_dev);
1463 }
1464
1465 static void bond_setup_by_slave(struct net_device *bond_dev,
1466                                 struct net_device *slave_dev)
1467 {
1468         struct bonding *bond = netdev_priv(bond_dev);
1469
1470         bond_dev->header_ops        = slave_dev->header_ops;
1471
1472         bond_dev->type              = slave_dev->type;
1473         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1474         bond_dev->addr_len          = slave_dev->addr_len;
1475
1476         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1477                 slave_dev->addr_len);
1478         bond->setup_by_slave = 1;
1479 }
1480
1481 /* On bonding slaves other than the currently active slave, suppress
1482  * duplicates except for alb non-mcast/bcast.
1483  */
1484 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1485                                             struct slave *slave,
1486                                             struct bonding *bond)
1487 {
1488         if (bond_is_slave_inactive(slave)) {
1489                 if (bond->params.mode == BOND_MODE_ALB &&
1490                     skb->pkt_type != PACKET_BROADCAST &&
1491                     skb->pkt_type != PACKET_MULTICAST)
1492                         return false;
1493                 return true;
1494         }
1495         return false;
1496 }
1497
1498 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1499 {
1500         struct sk_buff *skb = *pskb;
1501         struct slave *slave;
1502         struct bonding *bond;
1503
1504         skb = skb_share_check(skb, GFP_ATOMIC);
1505         if (unlikely(!skb))
1506                 return RX_HANDLER_CONSUMED;
1507
1508         *pskb = skb;
1509
1510         slave = bond_slave_get_rcu(skb->dev);
1511         bond = slave->bond;
1512
1513         if (bond->params.arp_interval)
1514                 slave->dev->last_rx = jiffies;
1515
1516         if (bond->recv_probe) {
1517                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1518
1519                 if (likely(nskb)) {
1520                         bond->recv_probe(nskb, bond, slave);
1521                         dev_kfree_skb(nskb);
1522                 }
1523         }
1524
1525         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1526                 return RX_HANDLER_EXACT;
1527         }
1528
1529         skb->dev = bond->dev;
1530
1531         if (bond->params.mode == BOND_MODE_ALB &&
1532             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1533             skb->pkt_type == PACKET_HOST) {
1534
1535                 if (unlikely(skb_cow_head(skb,
1536                                           skb->data - skb_mac_header(skb)))) {
1537                         kfree_skb(skb);
1538                         return RX_HANDLER_CONSUMED;
1539                 }
1540                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1541         }
1542
1543         return RX_HANDLER_ANOTHER;
1544 }
1545
1546 /* enslave device <slave> to bond device <master> */
1547 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1548 {
1549         struct bonding *bond = netdev_priv(bond_dev);
1550         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1551         struct slave *new_slave = NULL;
1552         struct netdev_hw_addr *ha;
1553         struct sockaddr addr;
1554         int link_reporting;
1555         int res = 0;
1556
1557         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1558                 slave_ops->ndo_do_ioctl == NULL) {
1559                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1560                            bond_dev->name, slave_dev->name);
1561         }
1562
1563         /* already enslaved */
1564         if (slave_dev->flags & IFF_SLAVE) {
1565                 pr_debug("Error, Device was already enslaved\n");
1566                 return -EBUSY;
1567         }
1568
1569         /* vlan challenged mutual exclusion */
1570         /* no need to lock since we're protected by rtnl_lock */
1571         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1572                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1573                 if (bond->vlgrp) {
1574                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1575                                bond_dev->name, slave_dev->name, bond_dev->name);
1576                         return -EPERM;
1577                 } else {
1578                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1579                                    bond_dev->name, slave_dev->name,
1580                                    slave_dev->name, bond_dev->name);
1581                 }
1582         } else {
1583                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1584         }
1585
1586         /*
1587          * Old ifenslave binaries are no longer supported.  These can
1588          * be identified with moderate accuracy by the state of the slave:
1589          * the current ifenslave will set the interface down prior to
1590          * enslaving it; the old ifenslave will not.
1591          */
1592         if ((slave_dev->flags & IFF_UP)) {
1593                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1594                        slave_dev->name);
1595                 res = -EPERM;
1596                 goto err_undo_flags;
1597         }
1598
1599         /* set bonding device ether type by slave - bonding netdevices are
1600          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1601          * there is a need to override some of the type dependent attribs/funcs.
1602          *
1603          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1604          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1605          */
1606         if (bond->slave_cnt == 0) {
1607                 if (bond_dev->type != slave_dev->type) {
1608                         pr_debug("%s: change device type from %d to %d\n",
1609                                  bond_dev->name,
1610                                  bond_dev->type, slave_dev->type);
1611
1612                         res = netdev_bonding_change(bond_dev,
1613                                                     NETDEV_PRE_TYPE_CHANGE);
1614                         res = notifier_to_errno(res);
1615                         if (res) {
1616                                 pr_err("%s: refused to change device type\n",
1617                                        bond_dev->name);
1618                                 res = -EBUSY;
1619                                 goto err_undo_flags;
1620                         }
1621
1622                         /* Flush unicast and multicast addresses */
1623                         dev_uc_flush(bond_dev);
1624                         dev_mc_flush(bond_dev);
1625
1626                         if (slave_dev->type != ARPHRD_ETHER)
1627                                 bond_setup_by_slave(bond_dev, slave_dev);
1628                         else
1629                                 ether_setup(bond_dev);
1630
1631                         netdev_bonding_change(bond_dev,
1632                                               NETDEV_POST_TYPE_CHANGE);
1633                 }
1634         } else if (bond_dev->type != slave_dev->type) {
1635                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1636                        slave_dev->name,
1637                        slave_dev->type, bond_dev->type);
1638                 res = -EINVAL;
1639                 goto err_undo_flags;
1640         }
1641
1642         if (slave_ops->ndo_set_mac_address == NULL) {
1643                 if (bond->slave_cnt == 0) {
1644                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1645                                    bond_dev->name);
1646                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1647                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1648                         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",
1649                                bond_dev->name);
1650                         res = -EOPNOTSUPP;
1651                         goto err_undo_flags;
1652                 }
1653         }
1654
1655         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1656
1657         /* If this is the first slave, then we need to set the master's hardware
1658          * address to be the same as the slave's. */
1659         if (is_zero_ether_addr(bond->dev->dev_addr))
1660                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1661                        slave_dev->addr_len);
1662
1663
1664         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1665         if (!new_slave) {
1666                 res = -ENOMEM;
1667                 goto err_undo_flags;
1668         }
1669
1670         /*
1671          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1672          * is set via sysfs or module option if desired.
1673          */
1674         new_slave->queue_id = 0;
1675
1676         /* Save slave's original mtu and then set it to match the bond */
1677         new_slave->original_mtu = slave_dev->mtu;
1678         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1679         if (res) {
1680                 pr_debug("Error %d calling dev_set_mtu\n", res);
1681                 goto err_free;
1682         }
1683
1684         /*
1685          * Save slave's original ("permanent") mac address for modes
1686          * that need it, and for restoring it upon release, and then
1687          * set it to the master's address
1688          */
1689         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1690
1691         if (!bond->params.fail_over_mac) {
1692                 /*
1693                  * Set slave to master's mac address.  The application already
1694                  * set the master's mac address to that of the first slave
1695                  */
1696                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1697                 addr.sa_family = slave_dev->type;
1698                 res = dev_set_mac_address(slave_dev, &addr);
1699                 if (res) {
1700                         pr_debug("Error %d calling set_mac_address\n", res);
1701                         goto err_restore_mtu;
1702                 }
1703         }
1704
1705         res = netdev_set_bond_master(slave_dev, bond_dev);
1706         if (res) {
1707                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1708                 goto err_restore_mac;
1709         }
1710
1711         /* open the slave since the application closed it */
1712         res = dev_open(slave_dev);
1713         if (res) {
1714                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1715                 goto err_unset_master;
1716         }
1717
1718         new_slave->bond = bond;
1719         new_slave->dev = slave_dev;
1720         slave_dev->priv_flags |= IFF_BONDING;
1721
1722         if (bond_is_lb(bond)) {
1723                 /* bond_alb_init_slave() must be called before all other stages since
1724                  * it might fail and we do not want to have to undo everything
1725                  */
1726                 res = bond_alb_init_slave(bond, new_slave);
1727                 if (res)
1728                         goto err_close;
1729         }
1730
1731         /* If the mode USES_PRIMARY, then the new slave gets the
1732          * master's promisc (and mc) settings only if it becomes the
1733          * curr_active_slave, and that is taken care of later when calling
1734          * bond_change_active()
1735          */
1736         if (!USES_PRIMARY(bond->params.mode)) {
1737                 /* set promiscuity level to new slave */
1738                 if (bond_dev->flags & IFF_PROMISC) {
1739                         res = dev_set_promiscuity(slave_dev, 1);
1740                         if (res)
1741                                 goto err_close;
1742                 }
1743
1744                 /* set allmulti level to new slave */
1745                 if (bond_dev->flags & IFF_ALLMULTI) {
1746                         res = dev_set_allmulti(slave_dev, 1);
1747                         if (res)
1748                                 goto err_close;
1749                 }
1750
1751                 netif_addr_lock_bh(bond_dev);
1752                 /* upload master's mc_list to new slave */
1753                 netdev_for_each_mc_addr(ha, bond_dev)
1754                         dev_mc_add(slave_dev, ha->addr);
1755                 netif_addr_unlock_bh(bond_dev);
1756         }
1757
1758         if (bond->params.mode == BOND_MODE_8023AD) {
1759                 /* add lacpdu mc addr to mc list */
1760                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1761
1762                 dev_mc_add(slave_dev, lacpdu_multicast);
1763         }
1764
1765         bond_add_vlans_on_slave(bond, slave_dev);
1766
1767         write_lock_bh(&bond->lock);
1768
1769         bond_attach_slave(bond, new_slave);
1770
1771         new_slave->delay = 0;
1772         new_slave->link_failure_count = 0;
1773
1774         write_unlock_bh(&bond->lock);
1775
1776         bond_compute_features(bond);
1777
1778         read_lock(&bond->lock);
1779
1780         new_slave->last_arp_rx = jiffies;
1781
1782         if (bond->params.miimon && !bond->params.use_carrier) {
1783                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1784
1785                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1786                         /*
1787                          * miimon is set but a bonded network driver
1788                          * does not support ETHTOOL/MII and
1789                          * arp_interval is not set.  Note: if
1790                          * use_carrier is enabled, we will never go
1791                          * here (because netif_carrier is always
1792                          * supported); thus, we don't need to change
1793                          * the messages for netif_carrier.
1794                          */
1795                         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",
1796                                bond_dev->name, slave_dev->name);
1797                 } else if (link_reporting == -1) {
1798                         /* unable get link status using mii/ethtool */
1799                         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",
1800                                    bond_dev->name, slave_dev->name);
1801                 }
1802         }
1803
1804         /* check for initial state */
1805         if (!bond->params.miimon ||
1806             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1807                 if (bond->params.updelay) {
1808                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1809                         new_slave->link  = BOND_LINK_BACK;
1810                         new_slave->delay = bond->params.updelay;
1811                 } else {
1812                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1813                         new_slave->link  = BOND_LINK_UP;
1814                 }
1815                 new_slave->jiffies = jiffies;
1816         } else {
1817                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1818                 new_slave->link  = BOND_LINK_DOWN;
1819         }
1820
1821         if (bond_update_speed_duplex(new_slave) &&
1822             (new_slave->link != BOND_LINK_DOWN)) {
1823                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1824                            bond_dev->name, new_slave->dev->name);
1825
1826                 if (bond->params.mode == BOND_MODE_8023AD) {
1827                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1828                                    bond_dev->name);
1829                 }
1830         }
1831
1832         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1833                 /* if there is a primary slave, remember it */
1834                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1835                         bond->primary_slave = new_slave;
1836                         bond->force_primary = true;
1837                 }
1838         }
1839
1840         write_lock_bh(&bond->curr_slave_lock);
1841
1842         switch (bond->params.mode) {
1843         case BOND_MODE_ACTIVEBACKUP:
1844                 bond_set_slave_inactive_flags(new_slave);
1845                 bond_select_active_slave(bond);
1846                 break;
1847         case BOND_MODE_8023AD:
1848                 /* in 802.3ad mode, the internal mechanism
1849                  * will activate the slaves in the selected
1850                  * aggregator
1851                  */
1852                 bond_set_slave_inactive_flags(new_slave);
1853                 /* if this is the first slave */
1854                 if (bond->slave_cnt == 1) {
1855                         SLAVE_AD_INFO(new_slave).id = 1;
1856                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1857                          * can be called only after the mac address of the bond is set
1858                          */
1859                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1860                                             bond->params.lacp_fast);
1861                 } else {
1862                         SLAVE_AD_INFO(new_slave).id =
1863                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1864                 }
1865
1866                 bond_3ad_bind_slave(new_slave);
1867                 break;
1868         case BOND_MODE_TLB:
1869         case BOND_MODE_ALB:
1870                 bond_set_active_slave(new_slave);
1871                 bond_set_slave_inactive_flags(new_slave);
1872                 bond_select_active_slave(bond);
1873                 break;
1874         default:
1875                 pr_debug("This slave is always active in trunk mode\n");
1876
1877                 /* always active in trunk mode */
1878                 bond_set_active_slave(new_slave);
1879
1880                 /* In trunking mode there is little meaning to curr_active_slave
1881                  * anyway (it holds no special properties of the bond device),
1882                  * so we can change it without calling change_active_interface()
1883                  */
1884                 if (!bond->curr_active_slave)
1885                         bond->curr_active_slave = new_slave;
1886
1887                 break;
1888         } /* switch(bond_mode) */
1889
1890         write_unlock_bh(&bond->curr_slave_lock);
1891
1892         bond_set_carrier(bond);
1893
1894 #ifdef CONFIG_NET_POLL_CONTROLLER
1895         slave_dev->npinfo = bond_netpoll_info(bond);
1896         if (slave_dev->npinfo) {
1897                 if (slave_enable_netpoll(new_slave)) {
1898                         read_unlock(&bond->lock);
1899                         pr_info("Error, %s: master_dev is using netpoll, "
1900                                  "but new slave device does not support netpoll.\n",
1901                                  bond_dev->name);
1902                         res = -EBUSY;
1903                         goto err_close;
1904                 }
1905         }
1906 #endif
1907
1908         read_unlock(&bond->lock);
1909
1910         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1911         if (res)
1912                 goto err_close;
1913
1914         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1915                                          new_slave);
1916         if (res) {
1917                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1918                 goto err_dest_symlinks;
1919         }
1920
1921         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1922                 bond_dev->name, slave_dev->name,
1923                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1924                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1925
1926         /* enslave is successful */
1927         return 0;
1928
1929 /* Undo stages on error */
1930 err_dest_symlinks:
1931         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1932
1933 err_close:
1934         dev_close(slave_dev);
1935
1936 err_unset_master:
1937         netdev_set_bond_master(slave_dev, NULL);
1938
1939 err_restore_mac:
1940         if (!bond->params.fail_over_mac) {
1941                 /* XXX TODO - fom follow mode needs to change master's
1942                  * MAC if this slave's MAC is in use by the bond, or at
1943                  * least print a warning.
1944                  */
1945                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1946                 addr.sa_family = slave_dev->type;
1947                 dev_set_mac_address(slave_dev, &addr);
1948         }
1949
1950 err_restore_mtu:
1951         dev_set_mtu(slave_dev, new_slave->original_mtu);
1952
1953 err_free:
1954         kfree(new_slave);
1955
1956 err_undo_flags:
1957         bond_compute_features(bond);
1958
1959         return res;
1960 }
1961
1962 /*
1963  * Try to release the slave device <slave> from the bond device <master>
1964  * It is legal to access curr_active_slave without a lock because all the function
1965  * is write-locked.
1966  *
1967  * The rules for slave state should be:
1968  *   for Active/Backup:
1969  *     Active stays on all backups go down
1970  *   for Bonded connections:
1971  *     The first up interface should be left on and all others downed.
1972  */
1973 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1974 {
1975         struct bonding *bond = netdev_priv(bond_dev);
1976         struct slave *slave, *oldcurrent;
1977         struct sockaddr addr;
1978         u32 old_features = bond_dev->features;
1979
1980         /* slave is not a slave or master is not master of this slave */
1981         if (!(slave_dev->flags & IFF_SLAVE) ||
1982             (slave_dev->master != bond_dev)) {
1983                 pr_err("%s: Error: cannot release %s.\n",
1984                        bond_dev->name, slave_dev->name);
1985                 return -EINVAL;
1986         }
1987
1988         block_netpoll_tx();
1989         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1990         write_lock_bh(&bond->lock);
1991
1992         slave = bond_get_slave_by_dev(bond, slave_dev);
1993         if (!slave) {
1994                 /* not a slave of this bond */
1995                 pr_info("%s: %s not enslaved\n",
1996                         bond_dev->name, slave_dev->name);
1997                 write_unlock_bh(&bond->lock);
1998                 unblock_netpoll_tx();
1999                 return -EINVAL;
2000         }
2001
2002         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2003          * for this slave anymore.
2004          */
2005         netdev_rx_handler_unregister(slave_dev);
2006         write_unlock_bh(&bond->lock);
2007         synchronize_net();
2008         write_lock_bh(&bond->lock);
2009
2010         if (!bond->params.fail_over_mac) {
2011                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
2012                     bond->slave_cnt > 1)
2013                         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",
2014                                    bond_dev->name, slave_dev->name,
2015                                    slave->perm_hwaddr,
2016                                    bond_dev->name, slave_dev->name);
2017         }
2018
2019         /* Inform AD package of unbinding of slave. */
2020         if (bond->params.mode == BOND_MODE_8023AD) {
2021                 /* must be called before the slave is
2022                  * detached from the list
2023                  */
2024                 bond_3ad_unbind_slave(slave);
2025         }
2026
2027         pr_info("%s: releasing %s interface %s\n",
2028                 bond_dev->name,
2029                 bond_is_active_slave(slave) ? "active" : "backup",
2030                 slave_dev->name);
2031
2032         oldcurrent = bond->curr_active_slave;
2033
2034         bond->current_arp_slave = NULL;
2035
2036         /* release the slave from its bond */
2037         bond_detach_slave(bond, slave);
2038
2039         if (bond->primary_slave == slave)
2040                 bond->primary_slave = NULL;
2041
2042         if (oldcurrent == slave)
2043                 bond_change_active_slave(bond, NULL);
2044
2045         if (bond_is_lb(bond)) {
2046                 /* Must be called only after the slave has been
2047                  * detached from the list and the curr_active_slave
2048                  * has been cleared (if our_slave == old_current),
2049                  * but before a new active slave is selected.
2050                  */
2051                 write_unlock_bh(&bond->lock);
2052                 bond_alb_deinit_slave(bond, slave);
2053                 write_lock_bh(&bond->lock);
2054         }
2055
2056         if (oldcurrent == slave) {
2057                 /*
2058                  * Note that we hold RTNL over this sequence, so there
2059                  * is no concern that another slave add/remove event
2060                  * will interfere.
2061                  */
2062                 write_unlock_bh(&bond->lock);
2063                 read_lock(&bond->lock);
2064                 write_lock_bh(&bond->curr_slave_lock);
2065
2066                 bond_select_active_slave(bond);
2067
2068                 write_unlock_bh(&bond->curr_slave_lock);
2069                 read_unlock(&bond->lock);
2070                 write_lock_bh(&bond->lock);
2071         }
2072
2073         if (bond->slave_cnt == 0) {
2074                 bond_set_carrier(bond);
2075
2076                 /* if the last slave was removed, zero the mac address
2077                  * of the master so it will be set by the application
2078                  * to the mac address of the first slave
2079                  */
2080                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2081
2082                 if (bond->vlgrp) {
2083                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2084                                    bond_dev->name, bond_dev->name);
2085                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2086                                    bond_dev->name);
2087                 }
2088         }
2089
2090         write_unlock_bh(&bond->lock);
2091         unblock_netpoll_tx();
2092
2093         bond_compute_features(bond);
2094         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2095             (old_features & NETIF_F_VLAN_CHALLENGED))
2096                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2097                         bond_dev->name, slave_dev->name, bond_dev->name);
2098
2099         /* must do this from outside any spinlocks */
2100         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2101
2102         bond_del_vlans_from_slave(bond, slave_dev);
2103
2104         /* If the mode USES_PRIMARY, then we should only remove its
2105          * promisc and mc settings if it was the curr_active_slave, but that was
2106          * already taken care of above when we detached the slave
2107          */
2108         if (!USES_PRIMARY(bond->params.mode)) {
2109                 /* unset promiscuity level from slave */
2110                 if (bond_dev->flags & IFF_PROMISC)
2111                         dev_set_promiscuity(slave_dev, -1);
2112
2113                 /* unset allmulti level from slave */
2114                 if (bond_dev->flags & IFF_ALLMULTI)
2115                         dev_set_allmulti(slave_dev, -1);
2116
2117                 /* flush master's mc_list from slave */
2118                 netif_addr_lock_bh(bond_dev);
2119                 bond_mc_list_flush(bond_dev, slave_dev);
2120                 netif_addr_unlock_bh(bond_dev);
2121         }
2122
2123         netdev_set_bond_master(slave_dev, NULL);
2124
2125         slave_disable_netpoll(slave);
2126
2127         /* close slave before restoring its mac address */
2128         dev_close(slave_dev);
2129
2130         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2131                 /* restore original ("permanent") mac address */
2132                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2133                 addr.sa_family = slave_dev->type;
2134                 dev_set_mac_address(slave_dev, &addr);
2135         }
2136
2137         dev_set_mtu(slave_dev, slave->original_mtu);
2138
2139         slave_dev->priv_flags &= ~IFF_BONDING;
2140
2141         kfree(slave);
2142
2143         return 0;  /* deletion OK */
2144 }
2145
2146 /*
2147 * First release a slave and then destroy the bond if no more slaves are left.
2148 * Must be under rtnl_lock when this function is called.
2149 */
2150 static int  bond_release_and_destroy(struct net_device *bond_dev,
2151                                      struct net_device *slave_dev)
2152 {
2153         struct bonding *bond = netdev_priv(bond_dev);
2154         int ret;
2155
2156         ret = bond_release(bond_dev, slave_dev);
2157         if ((ret == 0) && (bond->slave_cnt == 0)) {
2158                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2159                 pr_info("%s: destroying bond %s.\n",
2160                         bond_dev->name, bond_dev->name);
2161                 unregister_netdevice(bond_dev);
2162         }
2163         return ret;
2164 }
2165
2166 /*
2167  * This function releases all slaves.
2168  */
2169 static int bond_release_all(struct net_device *bond_dev)
2170 {
2171         struct bonding *bond = netdev_priv(bond_dev);
2172         struct slave *slave;
2173         struct net_device *slave_dev;
2174         struct sockaddr addr;
2175
2176         write_lock_bh(&bond->lock);
2177
2178         netif_carrier_off(bond_dev);
2179
2180         if (bond->slave_cnt == 0)
2181                 goto out;
2182
2183         bond->current_arp_slave = NULL;
2184         bond->primary_slave = NULL;
2185         bond_change_active_slave(bond, NULL);
2186
2187         while ((slave = bond->first_slave) != NULL) {
2188                 /* Inform AD package of unbinding of slave
2189                  * before slave is detached from the list.
2190                  */
2191                 if (bond->params.mode == BOND_MODE_8023AD)
2192                         bond_3ad_unbind_slave(slave);
2193
2194                 slave_dev = slave->dev;
2195                 bond_detach_slave(bond, slave);
2196
2197                 /* now that the slave is detached, unlock and perform
2198                  * all the undo steps that should not be called from
2199                  * within a lock.
2200                  */
2201                 write_unlock_bh(&bond->lock);
2202
2203                 /* unregister rx_handler early so bond_handle_frame wouldn't
2204                  * be called for this slave anymore.
2205                  */
2206                 netdev_rx_handler_unregister(slave_dev);
2207                 synchronize_net();
2208
2209                 if (bond_is_lb(bond)) {
2210                         /* must be called only after the slave
2211                          * has been detached from the list
2212                          */
2213                         bond_alb_deinit_slave(bond, slave);
2214                 }
2215
2216                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2217                 bond_del_vlans_from_slave(bond, slave_dev);
2218
2219                 /* If the mode USES_PRIMARY, then we should only remove its
2220                  * promisc and mc settings if it was the curr_active_slave, but that was
2221                  * already taken care of above when we detached the slave
2222                  */
2223                 if (!USES_PRIMARY(bond->params.mode)) {
2224                         /* unset promiscuity level from slave */
2225                         if (bond_dev->flags & IFF_PROMISC)
2226                                 dev_set_promiscuity(slave_dev, -1);
2227
2228                         /* unset allmulti level from slave */
2229                         if (bond_dev->flags & IFF_ALLMULTI)
2230                                 dev_set_allmulti(slave_dev, -1);
2231
2232                         /* flush master's mc_list from slave */
2233                         netif_addr_lock_bh(bond_dev);
2234                         bond_mc_list_flush(bond_dev, slave_dev);
2235                         netif_addr_unlock_bh(bond_dev);
2236                 }
2237
2238                 netdev_set_bond_master(slave_dev, NULL);
2239
2240                 slave_disable_netpoll(slave);
2241
2242                 /* close slave before restoring its mac address */
2243                 dev_close(slave_dev);
2244
2245                 if (!bond->params.fail_over_mac) {
2246                         /* restore original ("permanent") mac address*/
2247                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2248                         addr.sa_family = slave_dev->type;
2249                         dev_set_mac_address(slave_dev, &addr);
2250                 }
2251
2252                 kfree(slave);
2253
2254                 /* re-acquire the lock before getting the next slave */
2255                 write_lock_bh(&bond->lock);
2256         }
2257
2258         /* zero the mac address of the master so it will be
2259          * set by the application to the mac address of the
2260          * first slave
2261          */
2262         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2263
2264         if (bond->vlgrp) {
2265                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2266                            bond_dev->name, bond_dev->name);
2267                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2268                            bond_dev->name);
2269         }
2270
2271         pr_info("%s: released all slaves\n", bond_dev->name);
2272
2273 out:
2274         write_unlock_bh(&bond->lock);
2275
2276         bond_compute_features(bond);
2277
2278         return 0;
2279 }
2280
2281 /*
2282  * This function changes the active slave to slave <slave_dev>.
2283  * It returns -EINVAL in the following cases.
2284  *  - <slave_dev> is not found in the list.
2285  *  - There is not active slave now.
2286  *  - <slave_dev> is already active.
2287  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2288  *  - <slave_dev> is not running.
2289  * In these cases, this function does nothing.
2290  * In the other cases, current_slave pointer is changed and 0 is returned.
2291  */
2292 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2293 {
2294         struct bonding *bond = netdev_priv(bond_dev);
2295         struct slave *old_active = NULL;
2296         struct slave *new_active = NULL;
2297         int res = 0;
2298
2299         if (!USES_PRIMARY(bond->params.mode))
2300                 return -EINVAL;
2301
2302         /* Verify that master_dev is indeed the master of slave_dev */
2303         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2304                 return -EINVAL;
2305
2306         read_lock(&bond->lock);
2307
2308         read_lock(&bond->curr_slave_lock);
2309         old_active = bond->curr_active_slave;
2310         read_unlock(&bond->curr_slave_lock);
2311
2312         new_active = bond_get_slave_by_dev(bond, slave_dev);
2313
2314         /*
2315          * Changing to the current active: do nothing; return success.
2316          */
2317         if (new_active && (new_active == old_active)) {
2318                 read_unlock(&bond->lock);
2319                 return 0;
2320         }
2321
2322         if ((new_active) &&
2323             (old_active) &&
2324             (new_active->link == BOND_LINK_UP) &&
2325             IS_UP(new_active->dev)) {
2326                 block_netpoll_tx();
2327                 write_lock_bh(&bond->curr_slave_lock);
2328                 bond_change_active_slave(bond, new_active);
2329                 write_unlock_bh(&bond->curr_slave_lock);
2330                 unblock_netpoll_tx();
2331         } else
2332                 res = -EINVAL;
2333
2334         read_unlock(&bond->lock);
2335
2336         return res;
2337 }
2338
2339 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2340 {
2341         struct bonding *bond = netdev_priv(bond_dev);
2342
2343         info->bond_mode = bond->params.mode;
2344         info->miimon = bond->params.miimon;
2345
2346         read_lock(&bond->lock);
2347         info->num_slaves = bond->slave_cnt;
2348         read_unlock(&bond->lock);
2349
2350         return 0;
2351 }
2352
2353 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2354 {
2355         struct bonding *bond = netdev_priv(bond_dev);
2356         struct slave *slave;
2357         int i, res = -ENODEV;
2358
2359         read_lock(&bond->lock);
2360
2361         bond_for_each_slave(bond, slave, i) {
2362                 if (i == (int)info->slave_id) {
2363                         res = 0;
2364                         strcpy(info->slave_name, slave->dev->name);
2365                         info->link = slave->link;
2366                         info->state = bond_slave_state(slave);
2367                         info->link_failure_count = slave->link_failure_count;
2368                         break;
2369                 }
2370         }
2371
2372         read_unlock(&bond->lock);
2373
2374         return res;
2375 }
2376
2377 /*-------------------------------- Monitoring -------------------------------*/
2378
2379
2380 static int bond_miimon_inspect(struct bonding *bond)
2381 {
2382         struct slave *slave;
2383         int i, link_state, commit = 0;
2384         bool ignore_updelay;
2385
2386         ignore_updelay = !bond->curr_active_slave ? true : false;
2387
2388         bond_for_each_slave(bond, slave, i) {
2389                 slave->new_link = BOND_LINK_NOCHANGE;
2390
2391                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2392
2393                 switch (slave->link) {
2394                 case BOND_LINK_UP:
2395                         if (link_state)
2396                                 continue;
2397
2398                         slave->link = BOND_LINK_FAIL;
2399                         slave->delay = bond->params.downdelay;
2400                         if (slave->delay) {
2401                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2402                                         bond->dev->name,
2403                                         (bond->params.mode ==
2404                                          BOND_MODE_ACTIVEBACKUP) ?
2405                                         (bond_is_active_slave(slave) ?
2406                                          "active " : "backup ") : "",
2407                                         slave->dev->name,
2408                                         bond->params.downdelay * bond->params.miimon);
2409                         }
2410                         /*FALLTHRU*/
2411                 case BOND_LINK_FAIL:
2412                         if (link_state) {
2413                                 /*
2414                                  * recovered before downdelay expired
2415                                  */
2416                                 slave->link = BOND_LINK_UP;
2417                                 slave->jiffies = jiffies;
2418                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2419                                         bond->dev->name,
2420                                         (bond->params.downdelay - slave->delay) *
2421                                         bond->params.miimon,
2422                                         slave->dev->name);
2423                                 continue;
2424                         }
2425
2426                         if (slave->delay <= 0) {
2427                                 slave->new_link = BOND_LINK_DOWN;
2428                                 commit++;
2429                                 continue;
2430                         }
2431
2432                         slave->delay--;
2433                         break;
2434
2435                 case BOND_LINK_DOWN:
2436                         if (!link_state)
2437                                 continue;
2438
2439                         slave->link = BOND_LINK_BACK;
2440                         slave->delay = bond->params.updelay;
2441
2442                         if (slave->delay) {
2443                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2444                                         bond->dev->name, slave->dev->name,
2445                                         ignore_updelay ? 0 :
2446                                         bond->params.updelay *
2447                                         bond->params.miimon);
2448                         }
2449                         /*FALLTHRU*/
2450                 case BOND_LINK_BACK:
2451                         if (!link_state) {
2452                                 slave->link = BOND_LINK_DOWN;
2453                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2454                                         bond->dev->name,
2455                                         (bond->params.updelay - slave->delay) *
2456                                         bond->params.miimon,
2457                                         slave->dev->name);
2458
2459                                 continue;
2460                         }
2461
2462                         if (ignore_updelay)
2463                                 slave->delay = 0;
2464
2465                         if (slave->delay <= 0) {
2466                                 slave->new_link = BOND_LINK_UP;
2467                                 commit++;
2468                                 ignore_updelay = false;
2469                                 continue;
2470                         }
2471
2472                         slave->delay--;
2473                         break;
2474                 }
2475         }
2476
2477         return commit;
2478 }
2479
2480 static void bond_miimon_commit(struct bonding *bond)
2481 {
2482         struct slave *slave;
2483         int i;
2484
2485         bond_for_each_slave(bond, slave, i) {
2486                 switch (slave->new_link) {
2487                 case BOND_LINK_NOCHANGE:
2488                         continue;
2489
2490                 case BOND_LINK_UP:
2491                         slave->link = BOND_LINK_UP;
2492                         slave->jiffies = jiffies;
2493
2494                         if (bond->params.mode == BOND_MODE_8023AD) {
2495                                 /* prevent it from being the active one */
2496                                 bond_set_backup_slave(slave);
2497                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2498                                 /* make it immediately active */
2499                                 bond_set_active_slave(slave);
2500                         } else if (slave != bond->primary_slave) {
2501                                 /* prevent it from being the active one */
2502                                 bond_set_backup_slave(slave);
2503                         }
2504
2505                         bond_update_speed_duplex(slave);
2506
2507                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2508                                 bond->dev->name, slave->dev->name,
2509                                 slave->speed, slave->duplex ? "full" : "half");
2510
2511                         /* notify ad that the link status has changed */
2512                         if (bond->params.mode == BOND_MODE_8023AD)
2513                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2514
2515                         if (bond_is_lb(bond))
2516                                 bond_alb_handle_link_change(bond, slave,
2517                                                             BOND_LINK_UP);
2518
2519                         if (!bond->curr_active_slave ||
2520                             (slave == bond->primary_slave))
2521                                 goto do_failover;
2522
2523                         continue;
2524
2525                 case BOND_LINK_DOWN:
2526                         if (slave->link_failure_count < UINT_MAX)
2527                                 slave->link_failure_count++;
2528
2529                         slave->link = BOND_LINK_DOWN;
2530
2531                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2532                             bond->params.mode == BOND_MODE_8023AD)
2533                                 bond_set_slave_inactive_flags(slave);
2534
2535                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2536                                 bond->dev->name, slave->dev->name);
2537
2538                         if (bond->params.mode == BOND_MODE_8023AD)
2539                                 bond_3ad_handle_link_change(slave,
2540                                                             BOND_LINK_DOWN);
2541
2542                         if (bond_is_lb(bond))
2543                                 bond_alb_handle_link_change(bond, slave,
2544                                                             BOND_LINK_DOWN);
2545
2546                         if (slave == bond->curr_active_slave)
2547                                 goto do_failover;
2548
2549                         continue;
2550
2551                 default:
2552                         pr_err("%s: invalid new link %d on slave %s\n",
2553                                bond->dev->name, slave->new_link,
2554                                slave->dev->name);
2555                         slave->new_link = BOND_LINK_NOCHANGE;
2556
2557                         continue;
2558                 }
2559
2560 do_failover:
2561                 ASSERT_RTNL();
2562                 block_netpoll_tx();
2563                 write_lock_bh(&bond->curr_slave_lock);
2564                 bond_select_active_slave(bond);
2565                 write_unlock_bh(&bond->curr_slave_lock);
2566                 unblock_netpoll_tx();
2567         }
2568
2569         bond_set_carrier(bond);
2570 }
2571
2572 /*
2573  * bond_mii_monitor
2574  *
2575  * Really a wrapper that splits the mii monitor into two phases: an
2576  * inspection, then (if inspection indicates something needs to be done)
2577  * an acquisition of appropriate locks followed by a commit phase to
2578  * implement whatever link state changes are indicated.
2579  */
2580 void bond_mii_monitor(struct work_struct *work)
2581 {
2582         struct bonding *bond = container_of(work, struct bonding,
2583                                             mii_work.work);
2584         bool should_notify_peers = false;
2585
2586         read_lock(&bond->lock);
2587         if (bond->kill_timers)
2588                 goto out;
2589
2590         if (bond->slave_cnt == 0)
2591                 goto re_arm;
2592
2593         should_notify_peers = bond_should_notify_peers(bond);
2594
2595         if (bond_miimon_inspect(bond)) {
2596                 read_unlock(&bond->lock);
2597                 rtnl_lock();
2598                 read_lock(&bond->lock);
2599
2600                 bond_miimon_commit(bond);
2601
2602                 read_unlock(&bond->lock);
2603                 rtnl_unlock();  /* might sleep, hold no other locks */
2604                 read_lock(&bond->lock);
2605         }
2606
2607 re_arm:
2608         if (bond->params.miimon)
2609                 queue_delayed_work(bond->wq, &bond->mii_work,
2610                                    msecs_to_jiffies(bond->params.miimon));
2611 out:
2612         read_unlock(&bond->lock);
2613
2614         if (should_notify_peers) {
2615                 rtnl_lock();
2616                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2617                 rtnl_unlock();
2618         }
2619 }
2620
2621 static __be32 bond_glean_dev_ip(struct net_device *dev)
2622 {
2623         struct in_device *idev;
2624         struct in_ifaddr *ifa;
2625         __be32 addr = 0;
2626
2627         if (!dev)
2628                 return 0;
2629
2630         rcu_read_lock();
2631         idev = __in_dev_get_rcu(dev);
2632         if (!idev)
2633                 goto out;
2634
2635         ifa = idev->ifa_list;
2636         if (!ifa)
2637                 goto out;
2638
2639         addr = ifa->ifa_local;
2640 out:
2641         rcu_read_unlock();
2642         return addr;
2643 }
2644
2645 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2646 {
2647         struct vlan_entry *vlan;
2648
2649         if (ip == bond->master_ip)
2650                 return 1;
2651
2652         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2653                 if (ip == vlan->vlan_ip)
2654                         return 1;
2655         }
2656
2657         return 0;
2658 }
2659
2660 /*
2661  * We go to the (large) trouble of VLAN tagging ARP frames because
2662  * switches in VLAN mode (especially if ports are configured as
2663  * "native" to a VLAN) might not pass non-tagged frames.
2664  */
2665 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2666 {
2667         struct sk_buff *skb;
2668
2669         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2670                  slave_dev->name, dest_ip, src_ip, vlan_id);
2671
2672         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2673                          NULL, slave_dev->dev_addr, NULL);
2674
2675         if (!skb) {
2676                 pr_err("ARP packet allocation failed\n");
2677                 return;
2678         }
2679         if (vlan_id) {
2680                 skb = vlan_put_tag(skb, vlan_id);
2681                 if (!skb) {
2682                         pr_err("failed to insert VLAN tag\n");
2683                         return;
2684                 }
2685         }
2686         arp_xmit(skb);
2687 }
2688
2689
2690 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2691 {
2692         int i, vlan_id;
2693         __be32 *targets = bond->params.arp_targets;
2694         struct vlan_entry *vlan;
2695         struct net_device *vlan_dev;
2696         struct rtable *rt;
2697
2698         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2699                 if (!targets[i])
2700                         break;
2701                 pr_debug("basa: target %x\n", targets[i]);
2702                 if (!bond->vlgrp) {
2703                         pr_debug("basa: empty vlan: arp_send\n");
2704                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2705                                       bond->master_ip, 0);
2706                         continue;
2707                 }
2708
2709                 /*
2710                  * If VLANs are configured, we do a route lookup to
2711                  * determine which VLAN interface would be used, so we
2712                  * can tag the ARP with the proper VLAN tag.
2713                  */
2714                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2715                                      RTO_ONLINK, 0);
2716                 if (IS_ERR(rt)) {
2717                         if (net_ratelimit()) {
2718                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2719                                            bond->dev->name, &targets[i]);
2720                         }
2721                         continue;
2722                 }
2723
2724                 /*
2725                  * This target is not on a VLAN
2726                  */
2727                 if (rt->dst.dev == bond->dev) {
2728                         ip_rt_put(rt);
2729                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2730                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2731                                       bond->master_ip, 0);
2732                         continue;
2733                 }
2734
2735                 vlan_id = 0;
2736                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2737                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2738                         if (vlan_dev == rt->dst.dev) {
2739                                 vlan_id = vlan->vlan_id;
2740                                 pr_debug("basa: vlan match on %s %d\n",
2741                                        vlan_dev->name, vlan_id);
2742                                 break;
2743                         }
2744                 }
2745
2746                 if (vlan_id) {
2747                         ip_rt_put(rt);
2748                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2749                                       vlan->vlan_ip, vlan_id);
2750                         continue;
2751                 }
2752
2753                 if (net_ratelimit()) {
2754                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2755                                    bond->dev->name, &targets[i],
2756                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2757                 }
2758                 ip_rt_put(rt);
2759         }
2760 }
2761
2762 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2763 {
2764         int i;
2765         __be32 *targets = bond->params.arp_targets;
2766
2767         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2768                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2769                          &sip, &tip, i, &targets[i],
2770                          bond_has_this_ip(bond, tip));
2771                 if (sip == targets[i]) {
2772                         if (bond_has_this_ip(bond, tip))
2773                                 slave->last_arp_rx = jiffies;
2774                         return;
2775                 }
2776         }
2777 }
2778
2779 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2780                          struct slave *slave)
2781 {
2782         struct arphdr *arp;
2783         unsigned char *arp_ptr;
2784         __be32 sip, tip;
2785
2786         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2787                 return;
2788
2789         read_lock(&bond->lock);
2790
2791         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2792                  bond->dev->name, skb->dev->name);
2793
2794         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2795                 goto out_unlock;
2796
2797         arp = arp_hdr(skb);
2798         if (arp->ar_hln != bond->dev->addr_len ||
2799             skb->pkt_type == PACKET_OTHERHOST ||
2800             skb->pkt_type == PACKET_LOOPBACK ||
2801             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2802             arp->ar_pro != htons(ETH_P_IP) ||
2803             arp->ar_pln != 4)
2804                 goto out_unlock;
2805
2806         arp_ptr = (unsigned char *)(arp + 1);
2807         arp_ptr += bond->dev->addr_len;
2808         memcpy(&sip, arp_ptr, 4);
2809         arp_ptr += 4 + bond->dev->addr_len;
2810         memcpy(&tip, arp_ptr, 4);
2811
2812         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2813                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2814                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2815                  &sip, &tip);
2816
2817         /*
2818          * Backup slaves won't see the ARP reply, but do come through
2819          * here for each ARP probe (so we swap the sip/tip to validate
2820          * the probe).  In a "redundant switch, common router" type of
2821          * configuration, the ARP probe will (hopefully) travel from
2822          * the active, through one switch, the router, then the other
2823          * switch before reaching the backup.
2824          */
2825         if (bond_is_active_slave(slave))
2826                 bond_validate_arp(bond, slave, sip, tip);
2827         else
2828                 bond_validate_arp(bond, slave, tip, sip);
2829
2830 out_unlock:
2831         read_unlock(&bond->lock);
2832 }
2833
2834 /*
2835  * this function is called regularly to monitor each slave's link
2836  * ensuring that traffic is being sent and received when arp monitoring
2837  * is used in load-balancing mode. if the adapter has been dormant, then an
2838  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2839  * arp monitoring in active backup mode.
2840  */
2841 void bond_loadbalance_arp_mon(struct work_struct *work)
2842 {
2843         struct bonding *bond = container_of(work, struct bonding,
2844                                             arp_work.work);
2845         struct slave *slave, *oldcurrent;
2846         int do_failover = 0;
2847         int delta_in_ticks;
2848         int i;
2849
2850         read_lock(&bond->lock);
2851
2852         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2853
2854         if (bond->kill_timers)
2855                 goto out;
2856
2857         if (bond->slave_cnt == 0)
2858                 goto re_arm;
2859
2860         read_lock(&bond->curr_slave_lock);
2861         oldcurrent = bond->curr_active_slave;
2862         read_unlock(&bond->curr_slave_lock);
2863
2864         /* see if any of the previous devices are up now (i.e. they have
2865          * xmt and rcv traffic). the curr_active_slave does not come into
2866          * the picture unless it is null. also, slave->jiffies is not needed
2867          * here because we send an arp on each slave and give a slave as
2868          * long as it needs to get the tx/rx within the delta.
2869          * TODO: what about up/down delay in arp mode? it wasn't here before
2870          *       so it can wait
2871          */
2872         bond_for_each_slave(bond, slave, i) {
2873                 unsigned long trans_start = dev_trans_start(slave->dev);
2874
2875                 if (slave->link != BOND_LINK_UP) {
2876                         if (time_in_range(jiffies,
2877                                 trans_start - delta_in_ticks,
2878                                 trans_start + delta_in_ticks) &&
2879                             time_in_range(jiffies,
2880                                 slave->dev->last_rx - delta_in_ticks,
2881                                 slave->dev->last_rx + delta_in_ticks)) {
2882
2883                                 slave->link  = BOND_LINK_UP;
2884                                 bond_set_active_slave(slave);
2885
2886                                 /* primary_slave has no meaning in round-robin
2887                                  * mode. the window of a slave being up and
2888                                  * curr_active_slave being null after enslaving
2889                                  * is closed.
2890                                  */
2891                                 if (!oldcurrent) {
2892                                         pr_info("%s: link status definitely up for interface %s, ",
2893                                                 bond->dev->name,
2894                                                 slave->dev->name);
2895                                         do_failover = 1;
2896                                 } else {
2897                                         pr_info("%s: interface %s is now up\n",
2898                                                 bond->dev->name,
2899                                                 slave->dev->name);
2900                                 }
2901                         }
2902                 } else {
2903                         /* slave->link == BOND_LINK_UP */
2904
2905                         /* not all switches will respond to an arp request
2906                          * when the source ip is 0, so don't take the link down
2907                          * if we don't know our ip yet
2908                          */
2909                         if (!time_in_range(jiffies,
2910                                 trans_start - delta_in_ticks,
2911                                 trans_start + 2 * delta_in_ticks) ||
2912                             !time_in_range(jiffies,
2913                                 slave->dev->last_rx - delta_in_ticks,
2914                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2915
2916                                 slave->link  = BOND_LINK_DOWN;
2917                                 bond_set_backup_slave(slave);
2918
2919                                 if (slave->link_failure_count < UINT_MAX)
2920                                         slave->link_failure_count++;
2921
2922                                 pr_info("%s: interface %s is now down.\n",
2923                                         bond->dev->name,
2924                                         slave->dev->name);
2925
2926                                 if (slave == oldcurrent)
2927                                         do_failover = 1;
2928                         }
2929                 }
2930
2931                 /* note: if switch is in round-robin mode, all links
2932                  * must tx arp to ensure all links rx an arp - otherwise
2933                  * links may oscillate or not come up at all; if switch is
2934                  * in something like xor mode, there is nothing we can
2935                  * do - all replies will be rx'ed on same link causing slaves
2936                  * to be unstable during low/no traffic periods
2937                  */
2938                 if (IS_UP(slave->dev))
2939                         bond_arp_send_all(bond, slave);
2940         }
2941
2942         if (do_failover) {
2943                 block_netpoll_tx();
2944                 write_lock_bh(&bond->curr_slave_lock);
2945
2946                 bond_select_active_slave(bond);
2947
2948                 write_unlock_bh(&bond->curr_slave_lock);
2949                 unblock_netpoll_tx();
2950         }
2951
2952 re_arm:
2953         if (bond->params.arp_interval)
2954                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2955 out:
2956         read_unlock(&bond->lock);
2957 }
2958
2959 /*
2960  * Called to inspect slaves for active-backup mode ARP monitor link state
2961  * changes.  Sets new_link in slaves to specify what action should take
2962  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2963  * to link states must be committed.
2964  *
2965  * Called with bond->lock held for read.
2966  */
2967 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2968 {
2969         struct slave *slave;
2970         int i, commit = 0;
2971         unsigned long trans_start;
2972
2973         bond_for_each_slave(bond, slave, i) {
2974                 slave->new_link = BOND_LINK_NOCHANGE;
2975
2976                 if (slave->link != BOND_LINK_UP) {
2977                         if (time_in_range(jiffies,
2978                                 slave_last_rx(bond, slave) - delta_in_ticks,
2979                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2980
2981                                 slave->new_link = BOND_LINK_UP;
2982                                 commit++;
2983                         }
2984
2985                         continue;
2986                 }
2987
2988                 /*
2989                  * Give slaves 2*delta after being enslaved or made
2990                  * active.  This avoids bouncing, as the last receive
2991                  * times need a full ARP monitor cycle to be updated.
2992                  */
2993                 if (time_in_range(jiffies,
2994                                   slave->jiffies - delta_in_ticks,
2995                                   slave->jiffies + 2 * delta_in_ticks))
2996                         continue;
2997
2998                 /*
2999                  * Backup slave is down if:
3000                  * - No current_arp_slave AND
3001                  * - more than 3*delta since last receive AND
3002                  * - the bond has an IP address
3003                  *
3004                  * Note: a non-null current_arp_slave indicates
3005                  * the curr_active_slave went down and we are
3006                  * searching for a new one; under this condition
3007                  * we only take the curr_active_slave down - this
3008                  * gives each slave a chance to tx/rx traffic
3009                  * before being taken out
3010                  */
3011                 if (!bond_is_active_slave(slave) &&
3012                     !bond->current_arp_slave &&
3013                     !time_in_range(jiffies,
3014                         slave_last_rx(bond, slave) - delta_in_ticks,
3015                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3016
3017                         slave->new_link = BOND_LINK_DOWN;
3018                         commit++;
3019                 }
3020
3021                 /*
3022                  * Active slave is down if:
3023                  * - more than 2*delta since transmitting OR
3024                  * - (more than 2*delta since receive AND
3025                  *    the bond has an IP address)
3026                  */
3027                 trans_start = dev_trans_start(slave->dev);
3028                 if (bond_is_active_slave(slave) &&
3029                     (!time_in_range(jiffies,
3030                         trans_start - delta_in_ticks,
3031                         trans_start + 2 * delta_in_ticks) ||
3032                      !time_in_range(jiffies,
3033                         slave_last_rx(bond, slave) - delta_in_ticks,
3034                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3035
3036                         slave->new_link = BOND_LINK_DOWN;
3037                         commit++;
3038                 }
3039         }
3040
3041         return commit;
3042 }
3043
3044 /*
3045  * Called to commit link state changes noted by inspection step of
3046  * active-backup mode ARP monitor.
3047  *
3048  * Called with RTNL and bond->lock for read.
3049  */
3050 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3051 {
3052         struct slave *slave;
3053         int i;
3054         unsigned long trans_start;
3055
3056         bond_for_each_slave(bond, slave, i) {
3057                 switch (slave->new_link) {
3058                 case BOND_LINK_NOCHANGE:
3059                         continue;
3060
3061                 case BOND_LINK_UP:
3062                         trans_start = dev_trans_start(slave->dev);
3063                         if ((!bond->curr_active_slave &&
3064                              time_in_range(jiffies,
3065                                            trans_start - delta_in_ticks,
3066                                            trans_start + delta_in_ticks)) ||
3067                             bond->curr_active_slave != slave) {
3068                                 slave->link = BOND_LINK_UP;
3069                                 bond->current_arp_slave = NULL;
3070
3071                                 pr_info("%s: link status definitely up for interface %s.\n",
3072                                         bond->dev->name, slave->dev->name);
3073
3074                                 if (!bond->curr_active_slave ||
3075                                     (slave == bond->primary_slave))
3076                                         goto do_failover;
3077
3078                         }
3079
3080                         continue;
3081
3082                 case BOND_LINK_DOWN:
3083                         if (slave->link_failure_count < UINT_MAX)
3084                                 slave->link_failure_count++;
3085
3086                         slave->link = BOND_LINK_DOWN;
3087                         bond_set_slave_inactive_flags(slave);
3088
3089                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3090                                 bond->dev->name, slave->dev->name);
3091
3092                         if (slave == bond->curr_active_slave) {
3093                                 bond->current_arp_slave = NULL;
3094                                 goto do_failover;
3095                         }
3096
3097                         continue;
3098
3099                 default:
3100                         pr_err("%s: impossible: new_link %d on slave %s\n",
3101                                bond->dev->name, slave->new_link,
3102                                slave->dev->name);
3103                         continue;
3104                 }
3105
3106 do_failover:
3107                 ASSERT_RTNL();
3108                 block_netpoll_tx();
3109                 write_lock_bh(&bond->curr_slave_lock);
3110                 bond_select_active_slave(bond);
3111                 write_unlock_bh(&bond->curr_slave_lock);
3112                 unblock_netpoll_tx();
3113         }
3114
3115         bond_set_carrier(bond);
3116 }
3117
3118 /*
3119  * Send ARP probes for active-backup mode ARP monitor.
3120  *
3121  * Called with bond->lock held for read.
3122  */
3123 static void bond_ab_arp_probe(struct bonding *bond)
3124 {
3125         struct slave *slave;
3126         int i;
3127
3128         read_lock(&bond->curr_slave_lock);
3129
3130         if (bond->current_arp_slave && bond->curr_active_slave)
3131                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3132                         bond->current_arp_slave->dev->name,
3133                         bond->curr_active_slave->dev->name);
3134
3135         if (bond->curr_active_slave) {
3136                 bond_arp_send_all(bond, bond->curr_active_slave);
3137                 read_unlock(&bond->curr_slave_lock);
3138                 return;
3139         }
3140
3141         read_unlock(&bond->curr_slave_lock);
3142
3143         /* if we don't have a curr_active_slave, search for the next available
3144          * backup slave from the current_arp_slave and make it the candidate
3145          * for becoming the curr_active_slave
3146          */
3147
3148         if (!bond->current_arp_slave) {
3149                 bond->current_arp_slave = bond->first_slave;
3150                 if (!bond->current_arp_slave)
3151                         return;
3152         }
3153
3154         bond_set_slave_inactive_flags(bond->current_arp_slave);
3155
3156         /* search for next candidate */
3157         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3158                 if (IS_UP(slave->dev)) {
3159                         slave->link = BOND_LINK_BACK;
3160                         bond_set_slave_active_flags(slave);
3161                         bond_arp_send_all(bond, slave);
3162                         slave->jiffies = jiffies;
3163                         bond->current_arp_slave = slave;
3164                         break;
3165                 }
3166
3167                 /* if the link state is up at this point, we
3168                  * mark it down - this can happen if we have
3169                  * simultaneous link failures and
3170                  * reselect_active_interface doesn't make this
3171                  * one the current slave so it is still marked
3172                  * up when it is actually down
3173                  */
3174                 if (slave->link == BOND_LINK_UP) {
3175                         slave->link = BOND_LINK_DOWN;
3176                         if (slave->link_failure_count < UINT_MAX)
3177                                 slave->link_failure_count++;
3178
3179                         bond_set_slave_inactive_flags(slave);
3180
3181                         pr_info("%s: backup interface %s is now down.\n",
3182                                 bond->dev->name, slave->dev->name);
3183                 }
3184         }
3185 }
3186
3187 void bond_activebackup_arp_mon(struct work_struct *work)
3188 {
3189         struct bonding *bond = container_of(work, struct bonding,
3190                                             arp_work.work);
3191         bool should_notify_peers = false;
3192         int delta_in_ticks;
3193
3194         read_lock(&bond->lock);
3195
3196         if (bond->kill_timers)
3197                 goto out;
3198
3199         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3200
3201         if (bond->slave_cnt == 0)
3202                 goto re_arm;
3203
3204         should_notify_peers = bond_should_notify_peers(bond);
3205
3206         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3207                 read_unlock(&bond->lock);
3208                 rtnl_lock();
3209                 read_lock(&bond->lock);
3210
3211                 bond_ab_arp_commit(bond, delta_in_ticks);
3212
3213                 read_unlock(&bond->lock);
3214                 rtnl_unlock();
3215                 read_lock(&bond->lock);
3216         }
3217
3218         bond_ab_arp_probe(bond);
3219
3220 re_arm:
3221         if (bond->params.arp_interval)
3222                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3223 out:
3224         read_unlock(&bond->lock);
3225
3226         if (should_notify_peers) {
3227                 rtnl_lock();
3228                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3229                 rtnl_unlock();
3230         }
3231 }
3232
3233 /*-------------------------- netdev event handling --------------------------*/
3234
3235 /*
3236  * Change device name
3237  */
3238 static int bond_event_changename(struct bonding *bond)
3239 {
3240         bond_remove_proc_entry(bond);
3241         bond_create_proc_entry(bond);
3242
3243         bond_debug_reregister(bond);
3244
3245         return NOTIFY_DONE;
3246 }
3247
3248 static int bond_master_netdev_event(unsigned long event,
3249                                     struct net_device *bond_dev)
3250 {
3251         struct bonding *event_bond = netdev_priv(bond_dev);
3252
3253         switch (event) {
3254         case NETDEV_CHANGENAME:
3255                 return bond_event_changename(event_bond);
3256         default:
3257                 break;
3258         }
3259
3260         return NOTIFY_DONE;
3261 }
3262
3263 static int bond_slave_netdev_event(unsigned long event,
3264                                    struct net_device *slave_dev)
3265 {
3266         struct net_device *bond_dev = slave_dev->master;
3267         struct bonding *bond = netdev_priv(bond_dev);
3268
3269         switch (event) {
3270         case NETDEV_UNREGISTER:
3271                 if (bond_dev) {
3272                         if (bond->setup_by_slave)
3273                                 bond_release_and_destroy(bond_dev, slave_dev);
3274                         else
3275                                 bond_release(bond_dev, slave_dev);
3276                 }
3277                 break;
3278         case NETDEV_CHANGE:
3279                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3280                         struct slave *slave;
3281
3282                         slave = bond_get_slave_by_dev(bond, slave_dev);
3283                         if (slave) {
3284                                 u32 old_speed = slave->speed;
3285                                 u8  old_duplex = slave->duplex;
3286
3287                                 bond_update_speed_duplex(slave);
3288
3289                                 if (bond_is_lb(bond))
3290                                         break;
3291
3292                                 if (old_speed != slave->speed)
3293                                         bond_3ad_adapter_speed_changed(slave);
3294                                 if (old_duplex != slave->duplex)
3295                                         bond_3ad_adapter_duplex_changed(slave);
3296                         }
3297                 }
3298
3299                 break;
3300         case NETDEV_DOWN:
3301                 /*
3302                  * ... Or is it this?
3303                  */
3304                 break;
3305         case NETDEV_CHANGEMTU:
3306                 /*
3307                  * TODO: Should slaves be allowed to
3308                  * independently alter their MTU?  For
3309                  * an active-backup bond, slaves need
3310                  * not be the same type of device, so
3311                  * MTUs may vary.  For other modes,
3312                  * slaves arguably should have the
3313                  * same MTUs. To do this, we'd need to
3314                  * take over the slave's change_mtu
3315                  * function for the duration of their
3316                  * servitude.
3317                  */
3318                 break;
3319         case NETDEV_CHANGENAME:
3320                 /*
3321                  * TODO: handle changing the primary's name
3322                  */
3323                 break;
3324         case NETDEV_FEAT_CHANGE:
3325                 bond_compute_features(bond);
3326                 break;
3327         default:
3328                 break;
3329         }
3330
3331         return NOTIFY_DONE;
3332 }
3333
3334 /*
3335  * bond_netdev_event: handle netdev notifier chain events.
3336  *
3337  * This function receives events for the netdev chain.  The caller (an
3338  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3339  * locks for us to safely manipulate the slave devices (RTNL lock,
3340  * dev_probe_lock).
3341  */
3342 static int bond_netdev_event(struct notifier_block *this,
3343                              unsigned long event, void *ptr)
3344 {
3345         struct net_device *event_dev = (struct net_device *)ptr;
3346
3347         pr_debug("event_dev: %s, event: %lx\n",
3348                  event_dev ? event_dev->name : "None",
3349                  event);
3350
3351         if (!(event_dev->priv_flags & IFF_BONDING))
3352                 return NOTIFY_DONE;
3353
3354         if (event_dev->flags & IFF_MASTER) {
3355                 pr_debug("IFF_MASTER\n");
3356                 return bond_master_netdev_event(event, event_dev);
3357         }
3358
3359         if (event_dev->flags & IFF_SLAVE) {
3360                 pr_debug("IFF_SLAVE\n");
3361                 return bond_slave_netdev_event(event, event_dev);
3362         }
3363
3364         return NOTIFY_DONE;
3365 }
3366
3367 /*
3368  * bond_inetaddr_event: handle inetaddr notifier chain events.
3369  *
3370  * We keep track of device IPs primarily to use as source addresses in
3371  * ARP monitor probes (rather than spewing out broadcasts all the time).
3372  *
3373  * We track one IP for the main device (if it has one), plus one per VLAN.
3374  */
3375 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3376 {
3377         struct in_ifaddr *ifa = ptr;
3378         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3379         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3380         struct bonding *bond;
3381         struct vlan_entry *vlan;
3382
3383         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3384                 if (bond->dev == event_dev) {
3385                         switch (event) {
3386                         case NETDEV_UP:
3387                                 bond->master_ip = ifa->ifa_local;
3388                                 return NOTIFY_OK;
3389                         case NETDEV_DOWN:
3390                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3391                                 return NOTIFY_OK;
3392                         default:
3393                                 return NOTIFY_DONE;
3394                         }
3395                 }
3396
3397                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3398                         if (!bond->vlgrp)
3399                                 continue;
3400                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3401                         if (vlan_dev == event_dev) {
3402                                 switch (event) {
3403                                 case NETDEV_UP:
3404                                         vlan->vlan_ip = ifa->ifa_local;
3405                                         return NOTIFY_OK;
3406                                 case NETDEV_DOWN:
3407                                         vlan->vlan_ip =
3408                                                 bond_glean_dev_ip(vlan_dev);
3409                                         return NOTIFY_OK;
3410                                 default:
3411                                         return NOTIFY_DONE;
3412                                 }
3413                         }
3414                 }
3415         }
3416         return NOTIFY_DONE;
3417 }
3418
3419 static struct notifier_block bond_netdev_notifier = {
3420         .notifier_call = bond_netdev_event,
3421 };
3422
3423 static struct notifier_block bond_inetaddr_notifier = {
3424         .notifier_call = bond_inetaddr_event,
3425 };
3426
3427 /*---------------------------- Hashing Policies -----------------------------*/
3428
3429 /*
3430  * Hash for the output device based upon layer 2 and layer 3 data. If
3431  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3432  */
3433 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3434 {
3435         struct ethhdr *data = (struct ethhdr *)skb->data;
3436         struct iphdr *iph = ip_hdr(skb);
3437
3438         if (skb->protocol == htons(ETH_P_IP)) {
3439                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3440                         (data->h_dest[5] ^ data->h_source[5])) % count;
3441         }
3442
3443         return (data->h_dest[5] ^ data->h_source[5]) % count;
3444 }
3445
3446 /*
3447  * Hash for the output device based upon layer 3 and layer 4 data. If
3448  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3449  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3450  */
3451 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3452 {
3453         struct ethhdr *data = (struct ethhdr *)skb->data;
3454         struct iphdr *iph = ip_hdr(skb);
3455         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3456         int layer4_xor = 0;
3457
3458         if (skb->protocol == htons(ETH_P_IP)) {
3459                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3460                     (iph->protocol == IPPROTO_TCP ||
3461                      iph->protocol == IPPROTO_UDP)) {
3462                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3463                 }
3464                 return (layer4_xor ^
3465                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3466
3467         }
3468
3469         return (data->h_dest[5] ^ data->h_source[5]) % count;
3470 }
3471
3472 /*
3473  * Hash for the output device based upon layer 2 data
3474  */
3475 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3476 {
3477         struct ethhdr *data = (struct ethhdr *)skb->data;
3478
3479         return (data->h_dest[5] ^ data->h_source[5]) % count;
3480 }
3481
3482 /*-------------------------- Device entry points ----------------------------*/
3483
3484 static int bond_open(struct net_device *bond_dev)
3485 {
3486         struct bonding *bond = netdev_priv(bond_dev);
3487
3488         bond->kill_timers = 0;
3489
3490         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3491
3492         if (bond_is_lb(bond)) {
3493                 /* bond_alb_initialize must be called before the timer
3494                  * is started.
3495                  */
3496                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3497                         /* something went wrong - fail the open operation */
3498                         return -ENOMEM;
3499                 }
3500
3501                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3502                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3503         }
3504
3505         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3506                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3507                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3508         }
3509
3510         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3511                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3512                         INIT_DELAYED_WORK(&bond->arp_work,
3513                                           bond_activebackup_arp_mon);
3514                 else
3515                         INIT_DELAYED_WORK(&bond->arp_work,
3516                                           bond_loadbalance_arp_mon);
3517
3518                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3519                 if (bond->params.arp_validate)
3520                         bond->recv_probe = bond_arp_rcv;
3521         }
3522
3523         if (bond->params.mode == BOND_MODE_8023AD) {
3524                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3525                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3526                 /* register to receive LACPDUs */
3527                 bond->recv_probe = bond_3ad_lacpdu_recv;
3528                 bond_3ad_initiate_agg_selection(bond, 1);
3529         }
3530
3531         return 0;
3532 }
3533
3534 static int bond_close(struct net_device *bond_dev)
3535 {
3536         struct bonding *bond = netdev_priv(bond_dev);
3537
3538         write_lock_bh(&bond->lock);
3539
3540         bond->send_peer_notif = 0;
3541
3542         /* signal timers not to re-arm */
3543         bond->kill_timers = 1;
3544
3545         write_unlock_bh(&bond->lock);
3546
3547         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3548                 cancel_delayed_work(&bond->mii_work);
3549         }
3550
3551         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3552                 cancel_delayed_work(&bond->arp_work);
3553         }
3554
3555         switch (bond->params.mode) {
3556         case BOND_MODE_8023AD:
3557                 cancel_delayed_work(&bond->ad_work);
3558                 break;
3559         case BOND_MODE_TLB:
3560         case BOND_MODE_ALB:
3561                 cancel_delayed_work(&bond->alb_work);
3562                 break;
3563         default:
3564                 break;
3565         }
3566
3567         if (delayed_work_pending(&bond->mcast_work))
3568                 cancel_delayed_work(&bond->mcast_work);
3569
3570         if (bond_is_lb(bond)) {
3571                 /* Must be called only after all
3572                  * slaves have been released
3573                  */
3574                 bond_alb_deinitialize(bond);
3575         }
3576         bond->recv_probe = NULL;
3577
3578         return 0;
3579 }
3580
3581 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3582                                                 struct rtnl_link_stats64 *stats)
3583 {
3584         struct bonding *bond = netdev_priv(bond_dev);
3585         struct rtnl_link_stats64 temp;
3586         struct slave *slave;
3587         int i;
3588
3589         memset(stats, 0, sizeof(*stats));
3590
3591         read_lock_bh(&bond->lock);
3592
3593         bond_for_each_slave(bond, slave, i) {
3594                 const struct rtnl_link_stats64 *sstats =
3595                         dev_get_stats(slave->dev, &temp);
3596
3597                 stats->rx_packets += sstats->rx_packets;
3598                 stats->rx_bytes += sstats->rx_bytes;
3599                 stats->rx_errors += sstats->rx_errors;
3600                 stats->rx_dropped += sstats->rx_dropped;
3601
3602                 stats->tx_packets += sstats->tx_packets;
3603                 stats->tx_bytes += sstats->tx_bytes;
3604                 stats->tx_errors += sstats->tx_errors;
3605                 stats->tx_dropped += sstats->tx_dropped;
3606
3607                 stats->multicast += sstats->multicast;
3608                 stats->collisions += sstats->collisions;
3609
3610                 stats->rx_length_errors += sstats->rx_length_errors;
3611                 stats->rx_over_errors += sstats->rx_over_errors;
3612                 stats->rx_crc_errors += sstats->rx_crc_errors;
3613                 stats->rx_frame_errors += sstats->rx_frame_errors;
3614                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3615                 stats->rx_missed_errors += sstats->rx_missed_errors;
3616
3617                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3618                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3619                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3620                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3621                 stats->tx_window_errors += sstats->tx_window_errors;
3622         }
3623
3624         read_unlock_bh(&bond->lock);
3625
3626         return stats;
3627 }
3628
3629 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3630 {
3631         struct net_device *slave_dev = NULL;
3632         struct ifbond k_binfo;
3633         struct ifbond __user *u_binfo = NULL;
3634         struct ifslave k_sinfo;
3635         struct ifslave __user *u_sinfo = NULL;
3636         struct mii_ioctl_data *mii = NULL;
3637         int res = 0;
3638
3639         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3640
3641         switch (cmd) {
3642         case SIOCGMIIPHY:
3643                 mii = if_mii(ifr);
3644                 if (!mii)
3645                         return -EINVAL;
3646
3647                 mii->phy_id = 0;
3648                 /* Fall Through */
3649         case SIOCGMIIREG:
3650                 /*
3651                  * We do this again just in case we were called by SIOCGMIIREG
3652                  * instead of SIOCGMIIPHY.
3653                  */
3654                 mii = if_mii(ifr);
3655                 if (!mii)
3656                         return -EINVAL;
3657
3658
3659                 if (mii->reg_num == 1) {
3660                         struct bonding *bond = netdev_priv(bond_dev);
3661                         mii->val_out = 0;
3662                         read_lock(&bond->lock);
3663                         read_lock(&bond->curr_slave_lock);
3664                         if (netif_carrier_ok(bond->dev))
3665                                 mii->val_out = BMSR_LSTATUS;
3666
3667                         read_unlock(&bond->curr_slave_lock);
3668                         read_unlock(&bond->lock);
3669                 }
3670
3671                 return 0;
3672         case BOND_INFO_QUERY_OLD:
3673         case SIOCBONDINFOQUERY:
3674                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3675
3676                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3677                         return -EFAULT;
3678
3679                 res = bond_info_query(bond_dev, &k_binfo);
3680                 if (res == 0 &&
3681                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3682                         return -EFAULT;
3683
3684                 return res;
3685         case BOND_SLAVE_INFO_QUERY_OLD:
3686         case SIOCBONDSLAVEINFOQUERY:
3687                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3688
3689                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3690                         return -EFAULT;
3691
3692                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3693                 if (res == 0 &&
3694                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3695                         return -EFAULT;
3696
3697                 return res;
3698         default:
3699                 /* Go on */
3700                 break;
3701         }
3702
3703         if (!capable(CAP_NET_ADMIN))
3704                 return -EPERM;
3705
3706         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3707
3708         pr_debug("slave_dev=%p:\n", slave_dev);
3709
3710         if (!slave_dev)
3711                 res = -ENODEV;
3712         else {
3713                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3714                 switch (cmd) {
3715                 case BOND_ENSLAVE_OLD:
3716                 case SIOCBONDENSLAVE:
3717                         res = bond_enslave(bond_dev, slave_dev);
3718                         break;
3719                 case BOND_RELEASE_OLD:
3720                 case SIOCBONDRELEASE:
3721                         res = bond_release(bond_dev, slave_dev);
3722                         break;
3723                 case BOND_SETHWADDR_OLD:
3724                 case SIOCBONDSETHWADDR:
3725                         res = bond_sethwaddr(bond_dev, slave_dev);
3726                         break;
3727                 case BOND_CHANGE_ACTIVE_OLD:
3728                 case SIOCBONDCHANGEACTIVE:
3729                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3730                         break;
3731                 default:
3732                         res = -EOPNOTSUPP;
3733                 }
3734
3735                 dev_put(slave_dev);
3736         }
3737
3738         return res;
3739 }
3740
3741 static bool bond_addr_in_mc_list(unsigned char *addr,
3742                                  struct netdev_hw_addr_list *list,
3743                                  int addrlen)
3744 {
3745         struct netdev_hw_addr *ha;
3746
3747         netdev_hw_addr_list_for_each(ha, list)
3748                 if (!memcmp(ha->addr, addr, addrlen))
3749                         return true;
3750
3751         return false;
3752 }
3753
3754 static void bond_set_multicast_list(struct net_device *bond_dev)
3755 {
3756         struct bonding *bond = netdev_priv(bond_dev);
3757         struct netdev_hw_addr *ha;
3758         bool found;
3759
3760         /*
3761          * Do promisc before checking multicast_mode
3762          */
3763         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3764                 /*
3765                  * FIXME: Need to handle the error when one of the multi-slaves
3766                  * encounters error.
3767                  */
3768                 bond_set_promiscuity(bond, 1);
3769
3770
3771         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3772                 bond_set_promiscuity(bond, -1);
3773
3774
3775         /* set allmulti flag to slaves */
3776         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3777                 /*
3778                  * FIXME: Need to handle the error when one of the multi-slaves
3779                  * encounters error.
3780                  */
3781                 bond_set_allmulti(bond, 1);
3782
3783
3784         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3785                 bond_set_allmulti(bond, -1);
3786
3787
3788         read_lock(&bond->lock);
3789
3790         bond->flags = bond_dev->flags;
3791
3792         /* looking for addresses to add to slaves' mc list */
3793         netdev_for_each_mc_addr(ha, bond_dev) {
3794                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3795                                              bond_dev->addr_len);
3796                 if (!found)
3797                         bond_mc_add(bond, ha->addr);
3798         }
3799
3800         /* looking for addresses to delete from slaves' list */
3801         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3802                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3803                                              bond_dev->addr_len);
3804                 if (!found)
3805                         bond_mc_del(bond, ha->addr);
3806         }
3807
3808         /* save master's multicast list */
3809         __hw_addr_flush(&bond->mc_list);
3810         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3811                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3812
3813         read_unlock(&bond->lock);
3814 }
3815
3816 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3817 {
3818         struct bonding *bond = netdev_priv(dev);
3819         struct slave *slave = bond->first_slave;
3820
3821         if (slave) {
3822                 const struct net_device_ops *slave_ops
3823                         = slave->dev->netdev_ops;
3824                 if (slave_ops->ndo_neigh_setup)
3825                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3826         }
3827         return 0;
3828 }
3829
3830 /*
3831  * Change the MTU of all of a master's slaves to match the master
3832  */
3833 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3834 {
3835         struct bonding *bond = netdev_priv(bond_dev);
3836         struct slave *slave, *stop_at;
3837         int res = 0;
3838         int i;
3839
3840         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3841                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3842
3843         /* Can't hold bond->lock with bh disabled here since
3844          * some base drivers panic. On the other hand we can't
3845          * hold bond->lock without bh disabled because we'll
3846          * deadlock. The only solution is to rely on the fact
3847          * that we're under rtnl_lock here, and the slaves
3848          * list won't change. This doesn't solve the problem
3849          * of setting the slave's MTU while it is
3850          * transmitting, but the assumption is that the base
3851          * driver can handle that.
3852          *
3853          * TODO: figure out a way to safely iterate the slaves
3854          * list, but without holding a lock around the actual
3855          * call to the base driver.
3856          */
3857
3858         bond_for_each_slave(bond, slave, i) {
3859                 pr_debug("s %p s->p %p c_m %p\n",
3860                          slave,
3861                          slave->prev,
3862                          slave->dev->netdev_ops->ndo_change_mtu);
3863
3864                 res = dev_set_mtu(slave->dev, new_mtu);
3865
3866                 if (res) {
3867                         /* If we failed to set the slave's mtu to the new value
3868                          * we must abort the operation even in ACTIVE_BACKUP
3869                          * mode, because if we allow the backup slaves to have
3870                          * different mtu values than the active slave we'll
3871                          * need to change their mtu when doing a failover. That
3872                          * means changing their mtu from timer context, which
3873                          * is probably not a good idea.
3874                          */
3875                         pr_debug("err %d %s\n", res, slave->dev->name);
3876                         goto unwind;
3877                 }
3878         }
3879
3880         bond_dev->mtu = new_mtu;
3881
3882         return 0;
3883
3884 unwind:
3885         /* unwind from head to the slave that failed */
3886         stop_at = slave;
3887         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3888                 int tmp_res;
3889
3890                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3891                 if (tmp_res) {
3892                         pr_debug("unwind err %d dev %s\n",
3893                                  tmp_res, slave->dev->name);
3894                 }
3895         }
3896
3897         return res;
3898 }
3899
3900 /*
3901  * Change HW address
3902  *
3903  * Note that many devices must be down to change the HW address, and
3904  * downing the master releases all slaves.  We can make bonds full of
3905  * bonding devices to test this, however.
3906  */
3907 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3908 {
3909         struct bonding *bond = netdev_priv(bond_dev);
3910         struct sockaddr *sa = addr, tmp_sa;
3911         struct slave *slave, *stop_at;
3912         int res = 0;
3913         int i;
3914
3915         if (bond->params.mode == BOND_MODE_ALB)
3916                 return bond_alb_set_mac_address(bond_dev, addr);
3917
3918
3919         pr_debug("bond=%p, name=%s\n",
3920                  bond, bond_dev ? bond_dev->name : "None");
3921
3922         /*
3923          * If fail_over_mac is set to active, do nothing and return
3924          * success.  Returning an error causes ifenslave to fail.
3925          */
3926         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3927                 return 0;
3928
3929         if (!is_valid_ether_addr(sa->sa_data))
3930                 return -EADDRNOTAVAIL;
3931
3932         /* Can't hold bond->lock with bh disabled here since
3933          * some base drivers panic. On the other hand we can't
3934          * hold bond->lock without bh disabled because we'll
3935          * deadlock. The only solution is to rely on the fact
3936          * that we're under rtnl_lock here, and the slaves
3937          * list won't change. This doesn't solve the problem
3938          * of setting the slave's hw address while it is
3939          * transmitting, but the assumption is that the base
3940          * driver can handle that.
3941          *
3942          * TODO: figure out a way to safely iterate the slaves
3943          * list, but without holding a lock around the actual
3944          * call to the base driver.
3945          */
3946
3947         bond_for_each_slave(bond, slave, i) {
3948                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3949                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3950
3951                 if (slave_ops->ndo_set_mac_address == NULL) {
3952                         res = -EOPNOTSUPP;
3953                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3954                         goto unwind;
3955                 }
3956
3957                 res = dev_set_mac_address(slave->dev, addr);
3958                 if (res) {
3959                         /* TODO: consider downing the slave
3960                          * and retry ?
3961                          * User should expect communications
3962                          * breakage anyway until ARP finish
3963                          * updating, so...
3964                          */
3965                         pr_debug("err %d %s\n", res, slave->dev->name);
3966                         goto unwind;
3967                 }
3968         }
3969
3970         /* success */
3971         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3972         return 0;
3973
3974 unwind:
3975         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3976         tmp_sa.sa_family = bond_dev->type;
3977
3978         /* unwind from head to the slave that failed */
3979         stop_at = slave;
3980         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3981                 int tmp_res;
3982
3983                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3984                 if (tmp_res) {
3985                         pr_debug("unwind err %d dev %s\n",
3986                                  tmp_res, slave->dev->name);
3987                 }
3988         }
3989
3990         return res;
3991 }
3992
3993 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3994 {
3995         struct bonding *bond = netdev_priv(bond_dev);
3996         struct slave *slave, *start_at;
3997         int i, slave_no, res = 1;
3998         struct iphdr *iph = ip_hdr(skb);
3999
4000         /*
4001          * Start with the curr_active_slave that joined the bond as the
4002          * default for sending IGMP traffic.  For failover purposes one
4003          * needs to maintain some consistency for the interface that will
4004          * send the join/membership reports.  The curr_active_slave found
4005          * will send all of this type of traffic.
4006          */
4007         if ((iph->protocol == IPPROTO_IGMP) &&
4008             (skb->protocol == htons(ETH_P_IP))) {
4009
4010                 read_lock(&bond->curr_slave_lock);
4011                 slave = bond->curr_active_slave;
4012                 read_unlock(&bond->curr_slave_lock);
4013
4014                 if (!slave)
4015                         goto out;
4016         } else {
4017                 /*
4018                  * Concurrent TX may collide on rr_tx_counter; we accept
4019                  * that as being rare enough not to justify using an
4020                  * atomic op here.
4021                  */
4022                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4023
4024                 bond_for_each_slave(bond, slave, i) {
4025                         slave_no--;
4026                         if (slave_no < 0)
4027                                 break;
4028                 }
4029         }
4030
4031         start_at = slave;
4032         bond_for_each_slave_from(bond, slave, i, start_at) {
4033                 if (IS_UP(slave->dev) &&
4034                     (slave->link == BOND_LINK_UP) &&
4035                     bond_is_active_slave(slave)) {
4036                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4037                         break;
4038                 }
4039         }
4040
4041 out:
4042         if (res) {
4043                 /* no suitable interface, frame not sent */
4044                 dev_kfree_skb(skb);
4045         }
4046
4047         return NETDEV_TX_OK;
4048 }
4049
4050
4051 /*
4052  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4053  * the bond has a usable interface.
4054  */
4055 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4056 {
4057         struct bonding *bond = netdev_priv(bond_dev);
4058         int res = 1;
4059
4060         read_lock(&bond->curr_slave_lock);
4061
4062         if (bond->curr_active_slave)
4063                 res = bond_dev_queue_xmit(bond, skb,
4064                         bond->curr_active_slave->dev);
4065
4066         if (res)
4067                 /* no suitable interface, frame not sent */
4068                 dev_kfree_skb(skb);
4069
4070         read_unlock(&bond->curr_slave_lock);
4071
4072         return NETDEV_TX_OK;
4073 }
4074
4075 /*
4076  * In bond_xmit_xor() , we determine the output device by using a pre-
4077  * determined xmit_hash_policy(), If the selected device is not enabled,
4078  * find the next active slave.
4079  */
4080 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4081 {
4082         struct bonding *bond = netdev_priv(bond_dev);
4083         struct slave *slave, *start_at;
4084         int slave_no;
4085         int i;
4086         int res = 1;
4087
4088         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4089
4090         bond_for_each_slave(bond, slave, i) {
4091                 slave_no--;
4092                 if (slave_no < 0)
4093                         break;
4094         }
4095
4096         start_at = slave;
4097
4098         bond_for_each_slave_from(bond, slave, i, start_at) {
4099                 if (IS_UP(slave->dev) &&
4100                     (slave->link == BOND_LINK_UP) &&
4101                     bond_is_active_slave(slave)) {
4102                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4103                         break;
4104                 }
4105         }
4106
4107         if (res) {
4108                 /* no suitable interface, frame not sent */
4109                 dev_kfree_skb(skb);
4110         }
4111
4112         return NETDEV_TX_OK;
4113 }
4114
4115 /*
4116  * in broadcast mode, we send everything to all usable interfaces.
4117  */
4118 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4119 {
4120         struct bonding *bond = netdev_priv(bond_dev);
4121         struct slave *slave, *start_at;
4122         struct net_device *tx_dev = NULL;
4123         int i;
4124         int res = 1;
4125
4126         read_lock(&bond->curr_slave_lock);
4127         start_at = bond->curr_active_slave;
4128         read_unlock(&bond->curr_slave_lock);
4129
4130         if (!start_at)
4131                 goto out;
4132
4133         bond_for_each_slave_from(bond, slave, i, start_at) {
4134                 if (IS_UP(slave->dev) &&
4135                     (slave->link == BOND_LINK_UP) &&
4136                     bond_is_active_slave(slave)) {
4137                         if (tx_dev) {
4138                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4139                                 if (!skb2) {
4140                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4141                                                bond_dev->name);
4142                                         continue;
4143                                 }
4144
4145                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4146                                 if (res) {
4147                                         dev_kfree_skb(skb2);
4148                                         continue;
4149                                 }
4150                         }
4151                         tx_dev = slave->dev;
4152                 }
4153         }
4154
4155         if (tx_dev)
4156                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4157
4158 out:
4159         if (res)
4160                 /* no suitable interface, frame not sent */
4161                 dev_kfree_skb(skb);
4162
4163         /* frame sent to all suitable interfaces */
4164         return NETDEV_TX_OK;
4165 }
4166
4167 /*------------------------- Device initialization ---------------------------*/
4168
4169 static void bond_set_xmit_hash_policy(struct bonding *bond)
4170 {
4171         switch (bond->params.xmit_policy) {
4172         case BOND_XMIT_POLICY_LAYER23:
4173                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4174                 break;
4175         case BOND_XMIT_POLICY_LAYER34:
4176                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4177                 break;
4178         case BOND_XMIT_POLICY_LAYER2:
4179         default:
4180                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4181                 break;
4182         }
4183 }
4184
4185 /*
4186  * Lookup the slave that corresponds to a qid
4187  */
4188 static inline int bond_slave_override(struct bonding *bond,
4189                                       struct sk_buff *skb)
4190 {
4191         int i, res = 1;
4192         struct slave *slave = NULL;
4193         struct slave *check_slave;
4194
4195         if (!skb->queue_mapping)
4196                 return 1;
4197
4198         /* Find out if any slaves have the same mapping as this skb. */
4199         bond_for_each_slave(bond, check_slave, i) {
4200                 if (check_slave->queue_id == skb->queue_mapping) {
4201                         slave = check_slave;
4202                         break;
4203                 }
4204         }
4205
4206         /* If the slave isn't UP, use default transmit policy. */
4207         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4208             (slave->link == BOND_LINK_UP)) {
4209                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4210         }
4211
4212         return res;
4213 }
4214
4215
4216 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4217 {
4218         /*
4219          * This helper function exists to help dev_pick_tx get the correct
4220          * destination queue.  Using a helper function skips a call to
4221          * skb_tx_hash and will put the skbs in the queue we expect on their
4222          * way down to the bonding driver.
4223          */
4224         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4225
4226         /*
4227          * Save the original txq to restore before passing to the driver
4228          */
4229         bond_queue_mapping(skb) = skb->queue_mapping;
4230
4231         if (unlikely(txq >= dev->real_num_tx_queues)) {
4232                 do {
4233                         txq -= dev->real_num_tx_queues;
4234                 } while (txq >= dev->real_num_tx_queues);
4235         }
4236         return txq;
4237 }
4238
4239 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4240 {
4241         struct bonding *bond = netdev_priv(dev);
4242
4243         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4244                 if (!bond_slave_override(bond, skb))
4245                         return NETDEV_TX_OK;
4246         }
4247
4248         switch (bond->params.mode) {
4249         case BOND_MODE_ROUNDROBIN:
4250                 return bond_xmit_roundrobin(skb, dev);
4251         case BOND_MODE_ACTIVEBACKUP:
4252                 return bond_xmit_activebackup(skb, dev);
4253         case BOND_MODE_XOR:
4254                 return bond_xmit_xor(skb, dev);
4255         case BOND_MODE_BROADCAST:
4256                 return bond_xmit_broadcast(skb, dev);
4257         case BOND_MODE_8023AD:
4258                 return bond_3ad_xmit_xor(skb, dev);
4259         case BOND_MODE_ALB:
4260         case BOND_MODE_TLB:
4261                 return bond_alb_xmit(skb, dev);
4262         default:
4263                 /* Should never happen, mode already checked */
4264                 pr_err("%s: Error: Unknown bonding mode %d\n",
4265                        dev->name, bond->params.mode);
4266                 WARN_ON_ONCE(1);
4267                 dev_kfree_skb(skb);
4268                 return NETDEV_TX_OK;
4269         }
4270 }
4271
4272 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4273 {
4274         struct bonding *bond = netdev_priv(dev);
4275         netdev_tx_t ret = NETDEV_TX_OK;
4276
4277         /*
4278          * If we risk deadlock from transmitting this in the
4279          * netpoll path, tell netpoll to queue the frame for later tx
4280          */
4281         if (is_netpoll_tx_blocked(dev))
4282                 return NETDEV_TX_BUSY;
4283
4284         read_lock(&bond->lock);
4285
4286         if (bond->slave_cnt)
4287                 ret = __bond_start_xmit(skb, dev);
4288         else
4289                 dev_kfree_skb(skb);
4290
4291         read_unlock(&bond->lock);
4292
4293         return ret;
4294 }
4295
4296 /*
4297  * set bond mode specific net device operations
4298  */
4299 void bond_set_mode_ops(struct bonding *bond, int mode)
4300 {
4301         struct net_device *bond_dev = bond->dev;
4302
4303         switch (mode) {
4304         case BOND_MODE_ROUNDROBIN:
4305                 break;
4306         case BOND_MODE_ACTIVEBACKUP:
4307                 break;
4308         case BOND_MODE_XOR:
4309                 bond_set_xmit_hash_policy(bond);
4310                 break;
4311         case BOND_MODE_BROADCAST:
4312                 break;
4313         case BOND_MODE_8023AD:
4314                 bond_set_xmit_hash_policy(bond);
4315                 break;
4316         case BOND_MODE_ALB:
4317                 /* FALLTHRU */
4318         case BOND_MODE_TLB:
4319                 break;
4320         default:
4321                 /* Should never happen, mode already checked */
4322                 pr_err("%s: Error: Unknown bonding mode %d\n",
4323                        bond_dev->name, mode);
4324                 break;
4325         }
4326 }
4327
4328 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4329                                     struct ethtool_drvinfo *drvinfo)
4330 {
4331         strncpy(drvinfo->driver, DRV_NAME, 32);
4332         strncpy(drvinfo->version, DRV_VERSION, 32);
4333         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4334 }
4335
4336 static const struct ethtool_ops bond_ethtool_ops = {
4337         .get_drvinfo            = bond_ethtool_get_drvinfo,
4338         .get_link               = ethtool_op_get_link,
4339 };
4340
4341 static const struct net_device_ops bond_netdev_ops = {
4342         .ndo_init               = bond_init,
4343         .ndo_uninit             = bond_uninit,
4344         .ndo_open               = bond_open,
4345         .ndo_stop               = bond_close,
4346         .ndo_start_xmit         = bond_start_xmit,
4347         .ndo_select_queue       = bond_select_queue,
4348         .ndo_get_stats64        = bond_get_stats,
4349         .ndo_do_ioctl           = bond_do_ioctl,
4350         .ndo_set_multicast_list = bond_set_multicast_list,
4351         .ndo_change_mtu         = bond_change_mtu,
4352         .ndo_set_mac_address    = bond_set_mac_address,
4353         .ndo_neigh_setup        = bond_neigh_setup,
4354         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4355         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4356         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4357 #ifdef CONFIG_NET_POLL_CONTROLLER
4358         .ndo_netpoll_setup      = bond_netpoll_setup,
4359         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4360         .ndo_poll_controller    = bond_poll_controller,
4361 #endif
4362         .ndo_add_slave          = bond_enslave,
4363         .ndo_del_slave          = bond_release,
4364         .ndo_fix_features       = bond_fix_features,
4365 };
4366
4367 static void bond_destructor(struct net_device *bond_dev)
4368 {
4369         struct bonding *bond = netdev_priv(bond_dev);
4370         if (bond->wq)
4371                 destroy_workqueue(bond->wq);
4372         free_netdev(bond_dev);
4373 }
4374
4375 static void bond_setup(struct net_device *bond_dev)
4376 {
4377         struct bonding *bond = netdev_priv(bond_dev);
4378
4379         /* initialize rwlocks */
4380         rwlock_init(&bond->lock);
4381         rwlock_init(&bond->curr_slave_lock);
4382
4383         bond->params = bonding_defaults;
4384
4385         /* Initialize pointers */
4386         bond->dev = bond_dev;
4387         INIT_LIST_HEAD(&bond->vlan_list);
4388
4389         /* Initialize the device entry points */
4390         ether_setup(bond_dev);
4391         bond_dev->netdev_ops = &bond_netdev_ops;
4392         bond_dev->ethtool_ops = &bond_ethtool_ops;
4393         bond_set_mode_ops(bond, bond->params.mode);
4394
4395         bond_dev->destructor = bond_destructor;
4396
4397         /* Initialize the device options */
4398         bond_dev->tx_queue_len = 0;
4399         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4400         bond_dev->priv_flags |= IFF_BONDING;
4401         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4402
4403         /* At first, we block adding VLANs. That's the only way to
4404          * prevent problems that occur when adding VLANs over an
4405          * empty bond. The block will be removed once non-challenged
4406          * slaves are enslaved.
4407          */
4408         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4409
4410         /* don't acquire bond device's netif_tx_lock when
4411          * transmitting */
4412         bond_dev->features |= NETIF_F_LLTX;
4413
4414         /* By default, we declare the bond to be fully
4415          * VLAN hardware accelerated capable. Special
4416          * care is taken in the various xmit functions
4417          * when there are slaves that are not hw accel
4418          * capable
4419          */
4420
4421         bond_dev->hw_features = BOND_VLAN_FEATURES |
4422                                 NETIF_F_HW_VLAN_TX |
4423                                 NETIF_F_HW_VLAN_RX |
4424                                 NETIF_F_HW_VLAN_FILTER;
4425
4426         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4427         bond_dev->features |= bond_dev->hw_features;
4428 }
4429
4430 static void bond_work_cancel_all(struct bonding *bond)
4431 {
4432         write_lock_bh(&bond->lock);
4433         bond->kill_timers = 1;
4434         write_unlock_bh(&bond->lock);
4435
4436         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4437                 cancel_delayed_work(&bond->mii_work);
4438
4439         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4440                 cancel_delayed_work(&bond->arp_work);
4441
4442         if (bond->params.mode == BOND_MODE_ALB &&
4443             delayed_work_pending(&bond->alb_work))
4444                 cancel_delayed_work(&bond->alb_work);
4445
4446         if (bond->params.mode == BOND_MODE_8023AD &&
4447             delayed_work_pending(&bond->ad_work))
4448                 cancel_delayed_work(&bond->ad_work);
4449
4450         if (delayed_work_pending(&bond->mcast_work))
4451                 cancel_delayed_work(&bond->mcast_work);
4452 }
4453
4454 /*
4455 * Destroy a bonding device.
4456 * Must be under rtnl_lock when this function is called.
4457 */
4458 static void bond_uninit(struct net_device *bond_dev)
4459 {
4460         struct bonding *bond = netdev_priv(bond_dev);
4461         struct vlan_entry *vlan, *tmp;
4462
4463         bond_netpoll_cleanup(bond_dev);
4464
4465         /* Release the bonded slaves */
4466         bond_release_all(bond_dev);
4467
4468         list_del(&bond->bond_list);
4469
4470         bond_work_cancel_all(bond);
4471
4472         bond_remove_proc_entry(bond);
4473
4474         bond_debug_unregister(bond);
4475
4476         __hw_addr_flush(&bond->mc_list);
4477
4478         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4479                 list_del(&vlan->vlan_list);
4480                 kfree(vlan);
4481         }
4482 }
4483
4484 /*------------------------- Module initialization ---------------------------*/
4485
4486 /*
4487  * Convert string input module parms.  Accept either the
4488  * number of the mode or its string name.  A bit complicated because
4489  * some mode names are substrings of other names, and calls from sysfs
4490  * may have whitespace in the name (trailing newlines, for example).
4491  */
4492 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4493 {
4494         int modeint = -1, i, rv;
4495         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4496
4497         for (p = (char *)buf; *p; p++)
4498                 if (!(isdigit(*p) || isspace(*p)))
4499                         break;
4500
4501         if (*p)
4502                 rv = sscanf(buf, "%20s", modestr);
4503         else
4504                 rv = sscanf(buf, "%d", &modeint);
4505
4506         if (!rv)
4507                 return -1;
4508
4509         for (i = 0; tbl[i].modename; i++) {
4510                 if (modeint == tbl[i].mode)
4511                         return tbl[i].mode;
4512                 if (strcmp(modestr, tbl[i].modename) == 0)
4513                         return tbl[i].mode;
4514         }
4515
4516         return -1;
4517 }
4518
4519 static int bond_check_params(struct bond_params *params)
4520 {
4521         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4522
4523         /*
4524          * Convert string parameters.
4525          */
4526         if (mode) {
4527                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4528                 if (bond_mode == -1) {
4529                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4530                                mode == NULL ? "NULL" : mode);
4531                         return -EINVAL;
4532                 }
4533         }
4534
4535         if (xmit_hash_policy) {
4536                 if ((bond_mode != BOND_MODE_XOR) &&
4537                     (bond_mode != BOND_MODE_8023AD)) {
4538                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4539                                bond_mode_name(bond_mode));
4540                 } else {
4541                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4542                                                         xmit_hashtype_tbl);
4543                         if (xmit_hashtype == -1) {
4544                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4545                                        xmit_hash_policy == NULL ? "NULL" :
4546                                        xmit_hash_policy);
4547                                 return -EINVAL;
4548                         }
4549                 }
4550         }
4551
4552         if (lacp_rate) {
4553                 if (bond_mode != BOND_MODE_8023AD) {
4554                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4555                                 bond_mode_name(bond_mode));
4556                 } else {
4557                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4558                         if (lacp_fast == -1) {
4559                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4560                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4561                                 return -EINVAL;
4562                         }
4563                 }
4564         }
4565
4566         if (ad_select) {
4567                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4568                 if (params->ad_select == -1) {
4569                         pr_err("Error: Invalid ad_select \"%s\"\n",
4570                                ad_select == NULL ? "NULL" : ad_select);
4571                         return -EINVAL;
4572                 }
4573
4574                 if (bond_mode != BOND_MODE_8023AD) {
4575                         pr_warning("ad_select param only affects 802.3ad mode\n");
4576                 }
4577         } else {
4578                 params->ad_select = BOND_AD_STABLE;
4579         }
4580
4581         if (max_bonds < 0) {
4582                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4583                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4584                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4585         }
4586
4587         if (miimon < 0) {
4588                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4589                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4590                 miimon = BOND_LINK_MON_INTERV;
4591         }
4592
4593         if (updelay < 0) {
4594                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4595                            updelay, INT_MAX);
4596                 updelay = 0;
4597         }
4598
4599         if (downdelay < 0) {
4600                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4601                            downdelay, INT_MAX);
4602                 downdelay = 0;
4603         }
4604
4605         if ((use_carrier != 0) && (use_carrier != 1)) {
4606                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4607                            use_carrier);
4608                 use_carrier = 1;
4609         }
4610
4611         if (num_peer_notif < 0 || num_peer_notif > 255) {
4612                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4613                            num_peer_notif);
4614                 num_peer_notif = 1;
4615         }
4616
4617         /* reset values for 802.3ad */
4618         if (bond_mode == BOND_MODE_8023AD) {
4619                 if (!miimon) {
4620                         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");
4621                         pr_warning("Forcing miimon to 100msec\n");
4622                         miimon = 100;
4623                 }
4624         }
4625
4626         if (tx_queues < 1 || tx_queues > 255) {
4627                 pr_warning("Warning: tx_queues (%d) should be between "
4628                            "1 and 255, resetting to %d\n",
4629                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4630                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4631         }
4632
4633         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4634                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4635                            "not of valid value (0/1), so it was set to "
4636                            "0\n", all_slaves_active);
4637                 all_slaves_active = 0;
4638         }
4639
4640         if (resend_igmp < 0 || resend_igmp > 255) {
4641                 pr_warning("Warning: resend_igmp (%d) should be between "
4642                            "0 and 255, resetting to %d\n",
4643                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4644                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4645         }
4646
4647         /* reset values for TLB/ALB */
4648         if ((bond_mode == BOND_MODE_TLB) ||
4649             (bond_mode == BOND_MODE_ALB)) {
4650                 if (!miimon) {
4651                         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");
4652                         pr_warning("Forcing miimon to 100msec\n");
4653                         miimon = 100;
4654                 }
4655         }
4656
4657         if (bond_mode == BOND_MODE_ALB) {
4658                 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",
4659                           updelay);
4660         }
4661
4662         if (!miimon) {
4663                 if (updelay || downdelay) {
4664                         /* just warn the user the up/down delay will have
4665                          * no effect since miimon is zero...
4666                          */
4667                         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",
4668                                    updelay, downdelay);
4669                 }
4670         } else {
4671                 /* don't allow arp monitoring */
4672                 if (arp_interval) {
4673                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4674                                    miimon, arp_interval);
4675                         arp_interval = 0;
4676                 }
4677
4678                 if ((updelay % miimon) != 0) {
4679                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4680                                    updelay, miimon,
4681                                    (updelay / miimon) * miimon);
4682                 }
4683
4684                 updelay /= miimon;
4685
4686                 if ((downdelay % miimon) != 0) {
4687                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4688                                    downdelay, miimon,
4689                                    (downdelay / miimon) * miimon);
4690                 }
4691
4692                 downdelay /= miimon;
4693         }
4694
4695         if (arp_interval < 0) {
4696                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4697                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4698                 arp_interval = BOND_LINK_ARP_INTERV;
4699         }
4700
4701         for (arp_ip_count = 0;
4702              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4703              arp_ip_count++) {
4704                 /* not complete check, but should be good enough to
4705                    catch mistakes */
4706                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4707                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4708                                    arp_ip_target[arp_ip_count]);
4709                         arp_interval = 0;
4710                 } else {
4711                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4712                         arp_target[arp_ip_count] = ip;
4713                 }
4714         }
4715
4716         if (arp_interval && !arp_ip_count) {
4717                 /* don't allow arping if no arp_ip_target given... */
4718                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4719                            arp_interval);
4720                 arp_interval = 0;
4721         }
4722
4723         if (arp_validate) {
4724                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4725                         pr_err("arp_validate only supported in active-backup mode\n");
4726                         return -EINVAL;
4727                 }
4728                 if (!arp_interval) {
4729                         pr_err("arp_validate requires arp_interval\n");
4730                         return -EINVAL;
4731                 }
4732
4733                 arp_validate_value = bond_parse_parm(arp_validate,
4734                                                      arp_validate_tbl);
4735                 if (arp_validate_value == -1) {
4736                         pr_err("Error: invalid arp_validate \"%s\"\n",
4737                                arp_validate == NULL ? "NULL" : arp_validate);
4738                         return -EINVAL;
4739                 }
4740         } else
4741                 arp_validate_value = 0;
4742
4743         if (miimon) {
4744                 pr_info("MII link monitoring set to %d ms\n", miimon);
4745         } else if (arp_interval) {
4746                 int i;
4747
4748                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4749                         arp_interval,
4750                         arp_validate_tbl[arp_validate_value].modename,
4751                         arp_ip_count);
4752
4753                 for (i = 0; i < arp_ip_count; i++)
4754                         pr_info(" %s", arp_ip_target[i]);
4755
4756                 pr_info("\n");
4757
4758         } else if (max_bonds) {
4759                 /* miimon and arp_interval not set, we need one so things
4760                  * work as expected, see bonding.txt for details
4761                  */
4762                 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");
4763         }
4764
4765         if (primary && !USES_PRIMARY(bond_mode)) {
4766                 /* currently, using a primary only makes sense
4767                  * in active backup, TLB or ALB modes
4768                  */
4769                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4770                            primary, bond_mode_name(bond_mode));
4771                 primary = NULL;
4772         }
4773
4774         if (primary && primary_reselect) {
4775                 primary_reselect_value = bond_parse_parm(primary_reselect,
4776                                                          pri_reselect_tbl);
4777                 if (primary_reselect_value == -1) {
4778                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4779                                primary_reselect ==
4780                                         NULL ? "NULL" : primary_reselect);
4781                         return -EINVAL;
4782                 }
4783         } else {
4784                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4785         }
4786
4787         if (fail_over_mac) {
4788                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4789                                                       fail_over_mac_tbl);
4790                 if (fail_over_mac_value == -1) {
4791                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4792                                arp_validate == NULL ? "NULL" : arp_validate);
4793                         return -EINVAL;
4794                 }
4795
4796                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4797                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4798         } else {
4799                 fail_over_mac_value = BOND_FOM_NONE;
4800         }
4801
4802         /* fill params struct with the proper values */
4803         params->mode = bond_mode;
4804         params->xmit_policy = xmit_hashtype;
4805         params->miimon = miimon;
4806         params->num_peer_notif = num_peer_notif;
4807         params->arp_interval = arp_interval;
4808         params->arp_validate = arp_validate_value;
4809         params->updelay = updelay;
4810         params->downdelay = downdelay;
4811         params->use_carrier = use_carrier;
4812         params->lacp_fast = lacp_fast;
4813         params->primary[0] = 0;
4814         params->primary_reselect = primary_reselect_value;
4815         params->fail_over_mac = fail_over_mac_value;
4816         params->tx_queues = tx_queues;
4817         params->all_slaves_active = all_slaves_active;
4818         params->resend_igmp = resend_igmp;
4819
4820         if (primary) {
4821                 strncpy(params->primary, primary, IFNAMSIZ);
4822                 params->primary[IFNAMSIZ - 1] = 0;
4823         }
4824
4825         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4826
4827         return 0;
4828 }
4829
4830 static struct lock_class_key bonding_netdev_xmit_lock_key;
4831 static struct lock_class_key bonding_netdev_addr_lock_key;
4832
4833 static void bond_set_lockdep_class_one(struct net_device *dev,
4834                                        struct netdev_queue *txq,
4835                                        void *_unused)
4836 {
4837         lockdep_set_class(&txq->_xmit_lock,
4838                           &bonding_netdev_xmit_lock_key);
4839 }
4840
4841 static void bond_set_lockdep_class(struct net_device *dev)
4842 {
4843         lockdep_set_class(&dev->addr_list_lock,
4844                           &bonding_netdev_addr_lock_key);
4845         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4846 }
4847
4848 /*
4849  * Called from registration process
4850  */
4851 static int bond_init(struct net_device *bond_dev)
4852 {
4853         struct bonding *bond = netdev_priv(bond_dev);
4854         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4855         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4856
4857         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4858
4859         /*
4860          * Initialize locks that may be required during
4861          * en/deslave operations.  All of the bond_open work
4862          * (of which this is part) should really be moved to
4863          * a phase prior to dev_open
4864          */
4865         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4866         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4867
4868         bond->wq = create_singlethread_workqueue(bond_dev->name);
4869         if (!bond->wq)
4870                 return -ENOMEM;
4871
4872         bond_set_lockdep_class(bond_dev);
4873
4874         bond_create_proc_entry(bond);
4875         list_add_tail(&bond->bond_list, &bn->dev_list);
4876
4877         bond_prepare_sysfs_group(bond);
4878
4879         bond_debug_register(bond);
4880
4881         __hw_addr_init(&bond->mc_list);
4882         return 0;
4883 }
4884
4885 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4886 {
4887         if (tb[IFLA_ADDRESS]) {
4888                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4889                         return -EINVAL;
4890                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4891                         return -EADDRNOTAVAIL;
4892         }
4893         return 0;
4894 }
4895
4896 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4897         .kind           = "bond",
4898         .priv_size      = sizeof(struct bonding),
4899         .setup          = bond_setup,
4900         .validate       = bond_validate,
4901 };
4902
4903 /* Create a new bond based on the specified name and bonding parameters.
4904  * If name is NULL, obtain a suitable "bond%d" name for us.
4905  * Caller must NOT hold rtnl_lock; we need to release it here before we
4906  * set up our sysfs entries.
4907  */
4908 int bond_create(struct net *net, const char *name)
4909 {
4910         struct net_device *bond_dev;
4911         int res;
4912
4913         rtnl_lock();
4914
4915         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4916                                    name ? name : "bond%d",
4917                                    bond_setup, tx_queues);
4918         if (!bond_dev) {
4919                 pr_err("%s: eek! can't alloc netdev!\n", name);
4920                 rtnl_unlock();
4921                 return -ENOMEM;
4922         }
4923
4924         dev_net_set(bond_dev, net);
4925         bond_dev->rtnl_link_ops = &bond_link_ops;
4926
4927         res = register_netdevice(bond_dev);
4928
4929         netif_carrier_off(bond_dev);
4930
4931         rtnl_unlock();
4932         if (res < 0)
4933                 bond_destructor(bond_dev);
4934         return res;
4935 }
4936
4937 static int __net_init bond_net_init(struct net *net)
4938 {
4939         struct bond_net *bn = net_generic(net, bond_net_id);
4940
4941         bn->net = net;
4942         INIT_LIST_HEAD(&bn->dev_list);
4943
4944         bond_create_proc_dir(bn);
4945
4946         return 0;
4947 }
4948
4949 static void __net_exit bond_net_exit(struct net *net)
4950 {
4951         struct bond_net *bn = net_generic(net, bond_net_id);
4952
4953         bond_destroy_proc_dir(bn);
4954 }
4955
4956 static struct pernet_operations bond_net_ops = {
4957         .init = bond_net_init,
4958         .exit = bond_net_exit,
4959         .id   = &bond_net_id,
4960         .size = sizeof(struct bond_net),
4961 };
4962
4963 static int __init bonding_init(void)
4964 {
4965         int i;
4966         int res;
4967
4968         pr_info("%s", bond_version);
4969
4970         res = bond_check_params(&bonding_defaults);
4971         if (res)
4972                 goto out;
4973
4974         res = register_pernet_subsys(&bond_net_ops);
4975         if (res)
4976                 goto out;
4977
4978         res = rtnl_link_register(&bond_link_ops);
4979         if (res)
4980                 goto err_link;
4981
4982         bond_create_debugfs();
4983
4984         for (i = 0; i < max_bonds; i++) {
4985                 res = bond_create(&init_net, NULL);
4986                 if (res)
4987                         goto err;
4988         }
4989
4990         res = bond_create_sysfs();
4991         if (res)
4992                 goto err;
4993
4994         register_netdevice_notifier(&bond_netdev_notifier);
4995         register_inetaddr_notifier(&bond_inetaddr_notifier);
4996 out:
4997         return res;
4998 err:
4999         rtnl_link_unregister(&bond_link_ops);
5000 err_link:
5001         unregister_pernet_subsys(&bond_net_ops);
5002         goto out;
5003
5004 }
5005
5006 static void __exit bonding_exit(void)
5007 {
5008         unregister_netdevice_notifier(&bond_netdev_notifier);
5009         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5010
5011         bond_destroy_sysfs();
5012         bond_destroy_debugfs();
5013
5014         rtnl_link_unregister(&bond_link_ops);
5015         unregister_pernet_subsys(&bond_net_ops);
5016
5017 #ifdef CONFIG_NET_POLL_CONTROLLER
5018         /*
5019          * Make sure we don't have an imbalance on our netpoll blocking
5020          */
5021         WARN_ON(atomic_read(&netpoll_block_tx));
5022 #endif
5023 }
5024
5025 module_init(bonding_init);
5026 module_exit(bonding_exit);
5027 MODULE_LICENSE("GPL");
5028 MODULE_VERSION(DRV_VERSION);
5029 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5030 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5031 MODULE_ALIAS_RTNL_LINK("bond");