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 int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 386
 387 /**
 388  * bond_dev_queue_xmit - Prepare skb for xmit.
 389  *
 390  * @bond: bond device that got this skb for tx.
 391  * @skb: hw accel VLAN tagged skb to transmit
 392  * @slave_dev: slave that is supposed to xmit this skbuff
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 395                         struct net_device *slave_dev)
 396 {
 397         skb->dev = slave_dev;
 398
 399         skb->queue_mapping = bond_queue_mapping(skb);
 400
 401         if (unlikely(netpoll_tx_running(slave_dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 struct net_device *slave_dev = slave->dev;
 439                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 440
 441                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 442                     slave_ops->ndo_vlan_rx_add_vid) {
 443                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 444                 }
 445         }
 446
 447         res = bond_add_vlan(bond, vid);
 448         if (res) {
 449                 pr_err("%s: Error: Failed to add vlan id %d\n",
 450                        bond_dev->name, vid);
 451         }
 452 }
 453
 454 /**
 455  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 456  * @bond_dev: bonding net device that got called
 457  * @vid: vlan id being removed
 458  */
 459 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 460 {
 461         struct bonding *bond = netdev_priv(bond_dev);
 462         struct slave *slave;
 463         int i, res;
 464
 465         bond_for_each_slave(bond, slave, i) {
 466                 struct net_device *slave_dev = slave->dev;
 467                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 468
 469                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 470                     slave_ops->ndo_vlan_rx_kill_vid) {
 471                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 472                 }
 473         }
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479         }
 480 }
 481
 482 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 483 {
 484         struct vlan_entry *vlan;
 485         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 486
 487         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 488             !(slave_ops->ndo_vlan_rx_add_vid))
 489                 return;
 490
 491         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 492                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 493 }
 494
 495 static void bond_del_vlans_from_slave(struct bonding *bond,
 496                                       struct net_device *slave_dev)
 497 {
 498         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 499         struct vlan_entry *vlan;
 500
 501         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 502             !(slave_ops->ndo_vlan_rx_kill_vid))
 503                 return;
 504
 505         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 506                 if (!vlan->vlan_id)
 507                         continue;
 508                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 509         }
 510 }
 511
 512 /*------------------------------- Link status -------------------------------*/
 513
 514 /*
 515  * Set the carrier state for the master according to the state of its
 516  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 517  * do special 802.3ad magic.
 518  *
 519  * Returns zero if carrier state does not change, nonzero if it does.
 520  */
 521 static int bond_set_carrier(struct bonding *bond)
 522 {
 523         struct slave *slave;
 524         int i;
 525
 526         if (bond->slave_cnt == 0)
 527                 goto down;
 528
 529         if (bond->params.mode == BOND_MODE_8023AD)
 530                 return bond_3ad_set_carrier(bond);
 531
 532         bond_for_each_slave(bond, slave, i) {
 533                 if (slave->link == BOND_LINK_UP) {
 534                         if (!netif_carrier_ok(bond->dev)) {
 535                                 netif_carrier_on(bond->dev);
 536                                 return 1;
 537                         }
 538                         return 0;
 539                 }
 540         }
 541
 542 down:
 543         if (netif_carrier_ok(bond->dev)) {
 544                 netif_carrier_off(bond->dev);
 545                 return 1;
 546         }
 547         return 0;
 548 }
 549
 550 /*
 551  * Get link speed and duplex from the slave's base driver
 552  * using ethtool. If for some reason the call fails or the
 553  * values are invalid, fake speed and duplex to 100/Full
 554  * and return error.
 555  */
 556 static int bond_update_speed_duplex(struct slave *slave)
 557 {
 558         struct net_device *slave_dev = slave->dev;
 559         struct ethtool_cmd ecmd;
 560         u32 slave_speed;
 561         int res;
 562
 563         /* Fake speed and duplex */
 564         slave->speed = SPEED_100;
 565         slave->duplex = DUPLEX_FULL;
 566
 567         res = __ethtool_get_settings(slave_dev, &ecmd);
 568         if (res < 0)
 569                 return -1;
 570
 571         slave_speed = ethtool_cmd_speed(&ecmd);
 572         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 573                 return -1;
 574
 575         switch (ecmd.duplex) {
 576         case DUPLEX_FULL:
 577         case DUPLEX_HALF:
 578                 break;
 579         default:
 580                 return -1;
 581         }
 582
 583         slave->speed = slave_speed;
 584         slave->duplex = ecmd.duplex;
 585
 586         return 0;
 587 }
 588
 589 /*
 590  * if <dev> supports MII link status reporting, check its link status.
 591  *
 592  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 593  * depending upon the setting of the use_carrier parameter.
 594  *
 595  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 596  * can't tell and just pretend it is), or 0, meaning that the link is
 597  * down.
 598  *
 599  * If reporting is non-zero, instead of faking link up, return -1 if
 600  * both ETHTOOL and MII ioctls fail (meaning the device does not
 601  * support them).  If use_carrier is set, return whatever it says.
 602  * It'd be nice if there was a good way to tell if a driver supports
 603  * netif_carrier, but there really isn't.
 604  */
 605 static int bond_check_dev_link(struct bonding *bond,
 606                                struct net_device *slave_dev, int reporting)
 607 {
 608         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 609         int (*ioctl)(struct net_device *, struct ifreq *, int);
 610         struct ifreq ifr;
 611         struct mii_ioctl_data *mii;
 612
 613         if (!reporting && !netif_running(slave_dev))
 614                 return 0;
 615
 616         if (bond->params.use_carrier)
 617                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 618
 619         /* Try to get link status using Ethtool first. */
 620         if (slave_dev->ethtool_ops) {
 621                 if (slave_dev->ethtool_ops->get_link) {
 622                         u32 link;
 623
 624                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 625
 626                         return link ? BMSR_LSTATUS : 0;
 627                 }
 628         }
 629
 630         /* Ethtool can't be used, fallback to MII ioctls. */
 631         ioctl = slave_ops->ndo_do_ioctl;
 632         if (ioctl) {
 633                 /* TODO: set pointer to correct ioctl on a per team member */
 634                 /*       bases to make this more efficient. that is, once  */
 635                 /*       we determine the correct ioctl, we will always    */
 636                 /*       call it and not the others for that team          */
 637                 /*       member.                                           */
 638
 639                 /*
 640                  * We cannot assume that SIOCGMIIPHY will also read a
 641                  * register; not all network drivers (e.g., e100)
 642                  * support that.
 643                  */
 644
 645                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 646                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 647                 mii = if_mii(&ifr);
 648                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 649                         mii->reg_num = MII_BMSR;
 650                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 651                                 return mii->val_out & BMSR_LSTATUS;
 652                 }
 653         }
 654
 655         /*
 656          * If reporting, report that either there's no dev->do_ioctl,
 657          * or both SIOCGMIIREG and get_link failed (meaning that we
 658          * cannot report link status).  If not reporting, pretend
 659          * we're ok.
 660          */
 661         return reporting ? -1 : BMSR_LSTATUS;
 662 }
 663
 664 /*----------------------------- Multicast list ------------------------------*/
 665
 666 /*
 667  * Push the promiscuity flag down to appropriate slaves
 668  */
 669 static int bond_set_promiscuity(struct bonding *bond, int inc)
 670 {
 671         int err = 0;
 672         if (USES_PRIMARY(bond->params.mode)) {
 673                 /* write lock already acquired */
 674                 if (bond->curr_active_slave) {
 675                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 676                                                   inc);
 677                 }
 678         } else {
 679                 struct slave *slave;
 680                 int i;
 681                 bond_for_each_slave(bond, slave, i) {
 682                         err = dev_set_promiscuity(slave->dev, inc);
 683                         if (err)
 684                                 return err;
 685                 }
 686         }
 687         return err;
 688 }
 689
 690 /*
 691  * Push the allmulti flag down to all slaves
 692  */
 693 static int bond_set_allmulti(struct bonding *bond, int inc)
 694 {
 695         int err = 0;
 696         if (USES_PRIMARY(bond->params.mode)) {
 697                 /* write lock already acquired */
 698                 if (bond->curr_active_slave) {
 699                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 700                                                inc);
 701                 }
 702         } else {
 703                 struct slave *slave;
 704                 int i;
 705                 bond_for_each_slave(bond, slave, i) {
 706                         err = dev_set_allmulti(slave->dev, inc);
 707                         if (err)
 708                                 return err;
 709                 }
 710         }
 711         return err;
 712 }
 713
 714 /*
 715  * Add a Multicast address to slaves
 716  * according to mode
 717  */
 718 static void bond_mc_add(struct bonding *bond, void *addr)
 719 {
 720         if (USES_PRIMARY(bond->params.mode)) {
 721                 /* write lock already acquired */
 722                 if (bond->curr_active_slave)
 723                         dev_mc_add(bond->curr_active_slave->dev, addr);
 724         } else {
 725                 struct slave *slave;
 726                 int i;
 727
 728                 bond_for_each_slave(bond, slave, i)
 729                         dev_mc_add(slave->dev, addr);
 730         }
 731 }
 732
 733 /*
 734  * Remove a multicast address from slave
 735  * according to mode
 736  */
 737 static void bond_mc_del(struct bonding *bond, void *addr)
 738 {
 739         if (USES_PRIMARY(bond->params.mode)) {
 740                 /* write lock already acquired */
 741                 if (bond->curr_active_slave)
 742                         dev_mc_del(bond->curr_active_slave->dev, addr);
 743         } else {
 744                 struct slave *slave;
 745                 int i;
 746                 bond_for_each_slave(bond, slave, i) {
 747                         dev_mc_del(slave->dev, addr);
 748                 }
 749         }
 750 }
 751
 752
 753 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 754 {
 755         struct in_device *in_dev;
 756
 757         rcu_read_lock();
 758         in_dev = __in_dev_get_rcu(dev);
 759         if (in_dev)
 760                 ip_mc_rejoin_groups(in_dev);
 761         rcu_read_unlock();
 762 }
 763
 764 /*
 765  * Retrieve the list of registered multicast addresses for the bonding
 766  * device and retransmit an IGMP JOIN request to the current active
 767  * slave.
 768  */
 769 static void bond_resend_igmp_join_requests(struct bonding *bond)
 770 {
 771         struct net_device *vlan_dev;
 772         struct vlan_entry *vlan;
 773
 774         read_lock(&bond->lock);
 775
 776         /* rejoin all groups on bond device */
 777         __bond_resend_igmp_join_requests(bond->dev);
 778
 779         /* rejoin all groups on vlan devices */
 780         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 781                 rcu_read_lock();
 782                 vlan_dev = __vlan_find_dev_deep(bond->dev,
 783                                                 vlan->vlan_id);
 784                 rcu_read_unlock();
 785                 if (vlan_dev)
 786                         __bond_resend_igmp_join_requests(vlan_dev);
 787         }
 788
 789         if (--bond->igmp_retrans > 0)
 790                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 791
 792         read_unlock(&bond->lock);
 793 }
 794
 795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 796 {
 797         struct bonding *bond = container_of(work, struct bonding,
 798                                             mcast_work.work);
 799         bond_resend_igmp_join_requests(bond);
 800 }
 801
 802 /*
 803  * flush all members of flush->mc_list from device dev->mc_list
 804  */
 805 static void bond_mc_list_flush(struct net_device *bond_dev,
 806                                struct net_device *slave_dev)
 807 {
 808         struct bonding *bond = netdev_priv(bond_dev);
 809         struct netdev_hw_addr *ha;
 810
 811         netdev_for_each_mc_addr(ha, bond_dev)
 812                 dev_mc_del(slave_dev, ha->addr);
 813
 814         if (bond->params.mode == BOND_MODE_8023AD) {
 815                 /* del lacpdu mc addr from mc list */
 816                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 817
 818                 dev_mc_del(slave_dev, lacpdu_multicast);
 819         }
 820 }
 821
 822 /*--------------------------- Active slave change ---------------------------*/
 823
 824 /*
 825  * Update the mc list and multicast-related flags for the new and
 826  * old active slaves (if any) according to the multicast mode, and
 827  * promiscuous flags unconditionally.
 828  */
 829 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 830                          struct slave *old_active)
 831 {
 832         struct netdev_hw_addr *ha;
 833
 834         if (!USES_PRIMARY(bond->params.mode))
 835                 /* nothing to do -  mc list is already up-to-date on
 836                  * all slaves
 837                  */
 838                 return;
 839
 840         if (old_active) {
 841                 if (bond->dev->flags & IFF_PROMISC)
 842                         dev_set_promiscuity(old_active->dev, -1);
 843
 844                 if (bond->dev->flags & IFF_ALLMULTI)
 845                         dev_set_allmulti(old_active->dev, -1);
 846
 847                 netdev_for_each_mc_addr(ha, bond->dev)
 848                         dev_mc_del(old_active->dev, ha->addr);
 849         }
 850
 851         if (new_active) {
 852                 /* FIXME: Signal errors upstream. */
 853                 if (bond->dev->flags & IFF_PROMISC)
 854                         dev_set_promiscuity(new_active->dev, 1);
 855
 856                 if (bond->dev->flags & IFF_ALLMULTI)
 857                         dev_set_allmulti(new_active->dev, 1);
 858
 859                 netdev_for_each_mc_addr(ha, bond->dev)
 860                         dev_mc_add(new_active->dev, ha->addr);
 861         }
 862 }
 863
 864 /*
 865  * bond_do_fail_over_mac
 866  *
 867  * Perform special MAC address swapping for fail_over_mac settings
 868  *
 869  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 870  */
 871 static void bond_do_fail_over_mac(struct bonding *bond,
 872                                   struct slave *new_active,
 873                                   struct slave *old_active)
 874         __releases(&bond->curr_slave_lock)
 875         __releases(&bond->lock)
 876         __acquires(&bond->lock)
 877         __acquires(&bond->curr_slave_lock)
 878 {
 879         u8 tmp_mac[ETH_ALEN];
 880         struct sockaddr saddr;
 881         int rv;
 882
 883         switch (bond->params.fail_over_mac) {
 884         case BOND_FOM_ACTIVE:
 885                 if (new_active)
 886                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 887                                new_active->dev->addr_len);
 888                 break;
 889         case BOND_FOM_FOLLOW:
 890                 /*
 891                  * if new_active && old_active, swap them
 892                  * if just old_active, do nothing (going to no active slave)
 893                  * if just new_active, set new_active to bond's MAC
 894                  */
 895                 if (!new_active)
 896                         return;
 897
 898                 write_unlock_bh(&bond->curr_slave_lock);
 899                 read_unlock(&bond->lock);
 900
 901                 if (old_active) {
 902                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 903                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 904                                ETH_ALEN);
 905                         saddr.sa_family = new_active->dev->type;
 906                 } else {
 907                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 908                         saddr.sa_family = bond->dev->type;
 909                 }
 910
 911                 rv = dev_set_mac_address(new_active->dev, &saddr);
 912                 if (rv) {
 913                         pr_err("%s: Error %d setting MAC of slave %s\n",
 914                                bond->dev->name, -rv, new_active->dev->name);
 915                         goto out;
 916                 }
 917
 918                 if (!old_active)
 919                         goto out;
 920
 921                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 922                 saddr.sa_family = old_active->dev->type;
 923
 924                 rv = dev_set_mac_address(old_active->dev, &saddr);
 925                 if (rv)
 926                         pr_err("%s: Error %d setting MAC of slave %s\n",
 927                                bond->dev->name, -rv, new_active->dev->name);
 928 out:
 929                 read_lock(&bond->lock);
 930                 write_lock_bh(&bond->curr_slave_lock);
 931                 break;
 932         default:
 933                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 934                        bond->dev->name, bond->params.fail_over_mac);
 935                 break;
 936         }
 937
 938 }
 939
 940 static bool bond_should_change_active(struct bonding *bond)
 941 {
 942         struct slave *prim = bond->primary_slave;
 943         struct slave *curr = bond->curr_active_slave;
 944
 945         if (!prim || !curr || curr->link != BOND_LINK_UP)
 946                 return true;
 947         if (bond->force_primary) {
 948                 bond->force_primary = false;
 949                 return true;
 950         }
 951         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 952             (prim->speed < curr->speed ||
 953              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 954                 return false;
 955         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 956                 return false;
 957         return true;
 958 }
 959
 960 /**
 961  * find_best_interface - select the best available slave to be the active one
 962  * @bond: our bonding struct
 963  *
 964  * Warning: Caller must hold curr_slave_lock for writing.
 965  */
 966 static struct slave *bond_find_best_slave(struct bonding *bond)
 967 {
 968         struct slave *new_active, *old_active;
 969         struct slave *bestslave = NULL;
 970         int mintime = bond->params.updelay;
 971         int i;
 972
 973         new_active = bond->curr_active_slave;
 974
 975         if (!new_active) { /* there were no active slaves left */
 976                 if (bond->slave_cnt > 0)   /* found one slave */
 977                         new_active = bond->first_slave;
 978                 else
 979                         return NULL; /* still no slave, return NULL */
 980         }
 981
 982         if ((bond->primary_slave) &&
 983             bond->primary_slave->link == BOND_LINK_UP &&
 984             bond_should_change_active(bond)) {
 985                 new_active = bond->primary_slave;
 986         }
 987
 988         /* remember where to stop iterating over the slaves */
 989         old_active = new_active;
 990
 991         bond_for_each_slave_from(bond, new_active, i, old_active) {
 992                 if (new_active->link == BOND_LINK_UP) {
 993                         return new_active;
 994                 } else if (new_active->link == BOND_LINK_BACK &&
 995                            IS_UP(new_active->dev)) {
 996                         /* link up, but waiting for stabilization */
 997                         if (new_active->delay < mintime) {
 998                                 mintime = new_active->delay;
 999                                 bestslave = new_active;
1000                         }
1001                 }
1002         }
1003
1004         return bestslave;
1005 }
1006
1007 static bool bond_should_notify_peers(struct bonding *bond)
1008 {
1009         struct slave *slave = bond->curr_active_slave;
1010
1011         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1012                  bond->dev->name, slave ? slave->dev->name : "NULL");
1013
1014         if (!slave || !bond->send_peer_notif ||
1015             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1016                 return false;
1017
1018         bond->send_peer_notif--;
1019         return true;
1020 }
1021
1022 /**
1023  * change_active_interface - change the active slave into the specified one
1024  * @bond: our bonding struct
1025  * @new: the new slave to make the active one
1026  *
1027  * Set the new slave to the bond's settings and unset them on the old
1028  * curr_active_slave.
1029  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1030  *
1031  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1032  * because it is apparently the best available slave we have, even though its
1033  * updelay hasn't timed out yet.
1034  *
1035  * If new_active is not NULL, caller must hold bond->lock for read and
1036  * curr_slave_lock for write_bh.
1037  */
1038 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1039 {
1040         struct slave *old_active = bond->curr_active_slave;
1041
1042         if (old_active == new_active)
1043                 return;
1044
1045         if (new_active) {
1046                 new_active->jiffies = jiffies;
1047
1048                 if (new_active->link == BOND_LINK_BACK) {
1049                         if (USES_PRIMARY(bond->params.mode)) {
1050                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1051                                         bond->dev->name, new_active->dev->name,
1052                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1053                         }
1054
1055                         new_active->delay = 0;
1056                         new_active->link = BOND_LINK_UP;
1057
1058                         if (bond->params.mode == BOND_MODE_8023AD)
1059                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1060
1061                         if (bond_is_lb(bond))
1062                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1063                 } else {
1064                         if (USES_PRIMARY(bond->params.mode)) {
1065                                 pr_info("%s: making interface %s the new active one.\n",
1066                                         bond->dev->name, new_active->dev->name);
1067                         }
1068                 }
1069         }
1070
1071         if (USES_PRIMARY(bond->params.mode))
1072                 bond_mc_swap(bond, new_active, old_active);
1073
1074         if (bond_is_lb(bond)) {
1075                 bond_alb_handle_active_change(bond, new_active);
1076                 if (old_active)
1077                         bond_set_slave_inactive_flags(old_active);
1078                 if (new_active)
1079                         bond_set_slave_active_flags(new_active);
1080         } else {
1081                 bond->curr_active_slave = new_active;
1082         }
1083
1084         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1085                 if (old_active)
1086                         bond_set_slave_inactive_flags(old_active);
1087
1088                 if (new_active) {
1089                         bool should_notify_peers = false;
1090
1091                         bond_set_slave_active_flags(new_active);
1092
1093                         if (bond->params.fail_over_mac)
1094                                 bond_do_fail_over_mac(bond, new_active,
1095                                                       old_active);
1096
1097                         if (netif_running(bond->dev)) {
1098                                 bond->send_peer_notif =
1099                                         bond->params.num_peer_notif;
1100                                 should_notify_peers =
1101                                         bond_should_notify_peers(bond);
1102                         }
1103
1104                         write_unlock_bh(&bond->curr_slave_lock);
1105                         read_unlock(&bond->lock);
1106
1107                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1108                         if (should_notify_peers)
1109                                 netdev_bonding_change(bond->dev,
1110                                                       NETDEV_NOTIFY_PEERS);
1111
1112                         read_lock(&bond->lock);
1113                         write_lock_bh(&bond->curr_slave_lock);
1114                 }
1115         }
1116
1117         /* resend IGMP joins since active slave has changed or
1118          * all were sent on curr_active_slave.
1119          * resend only if bond is brought up with the affected
1120          * bonding modes and the retransmission is enabled */
1121         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1122             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1123              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1124                 bond->igmp_retrans = bond->params.resend_igmp;
1125                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1126         }
1127 }
1128
1129 /**
1130  * bond_select_active_slave - select a new active slave, if needed
1131  * @bond: our bonding struct
1132  *
1133  * This functions should be called when one of the following occurs:
1134  * - The old curr_active_slave has been released or lost its link.
1135  * - The primary_slave has got its link back.
1136  * - A slave has got its link back and there's no old curr_active_slave.
1137  *
1138  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1139  */
1140 void bond_select_active_slave(struct bonding *bond)
1141 {
1142         struct slave *best_slave;
1143         int rv;
1144
1145         best_slave = bond_find_best_slave(bond);
1146         if (best_slave != bond->curr_active_slave) {
1147                 bond_change_active_slave(bond, best_slave);
1148                 rv = bond_set_carrier(bond);
1149                 if (!rv)
1150                         return;
1151
1152                 if (netif_carrier_ok(bond->dev)) {
1153                         pr_info("%s: first active interface up!\n",
1154                                 bond->dev->name);
1155                 } else {
1156                         pr_info("%s: now running without any active interface !\n",
1157                                 bond->dev->name);
1158                 }
1159         }
1160 }
1161
1162 /*--------------------------- slave list handling ---------------------------*/
1163
1164 /*
1165  * This function attaches the slave to the end of list.
1166  *
1167  * bond->lock held for writing by caller.
1168  */
1169 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1170 {
1171         if (bond->first_slave == NULL) { /* attaching the first slave */
1172                 new_slave->next = new_slave;
1173                 new_slave->prev = new_slave;
1174                 bond->first_slave = new_slave;
1175         } else {
1176                 new_slave->next = bond->first_slave;
1177                 new_slave->prev = bond->first_slave->prev;
1178                 new_slave->next->prev = new_slave;
1179                 new_slave->prev->next = new_slave;
1180         }
1181
1182         bond->slave_cnt++;
1183 }
1184
1185 /*
1186  * This function detaches the slave from the list.
1187  * WARNING: no check is made to verify if the slave effectively
1188  * belongs to <bond>.
1189  * Nothing is freed on return, structures are just unchained.
1190  * If any slave pointer in bond was pointing to <slave>,
1191  * it should be changed by the calling function.
1192  *
1193  * bond->lock held for writing by caller.
1194  */
1195 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1196 {
1197         if (slave->next)
1198                 slave->next->prev = slave->prev;
1199
1200         if (slave->prev)
1201                 slave->prev->next = slave->next;
1202
1203         if (bond->first_slave == slave) { /* slave is the first slave */
1204                 if (bond->slave_cnt > 1) { /* there are more slave */
1205                         bond->first_slave = slave->next;
1206                 } else {
1207                         bond->first_slave = NULL; /* slave was the last one */
1208                 }
1209         }
1210
1211         slave->next = NULL;
1212         slave->prev = NULL;
1213         bond->slave_cnt--;
1214 }
1215
1216 #ifdef CONFIG_NET_POLL_CONTROLLER
1217 static inline int slave_enable_netpoll(struct slave *slave)
1218 {
1219         struct netpoll *np;
1220         int err = 0;
1221
1222         np = kzalloc(sizeof(*np), GFP_KERNEL);
1223         err = -ENOMEM;
1224         if (!np)
1225                 goto out;
1226
1227         np->dev = slave->dev;
1228         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1229         err = __netpoll_setup(np);
1230         if (err) {
1231                 kfree(np);
1232                 goto out;
1233         }
1234         slave->np = np;
1235 out:
1236         return err;
1237 }
1238 static inline void slave_disable_netpoll(struct slave *slave)
1239 {
1240         struct netpoll *np = slave->np;
1241
1242         if (!np)
1243                 return;
1244
1245         slave->np = NULL;
1246         synchronize_rcu_bh();
1247         __netpoll_cleanup(np);
1248         kfree(np);
1249 }
1250 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1251 {
1252         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1253                 return false;
1254         if (!slave_dev->netdev_ops->ndo_poll_controller)
1255                 return false;
1256         return true;
1257 }
1258
1259 static void bond_poll_controller(struct net_device *bond_dev)
1260 {
1261 }
1262
1263 static void __bond_netpoll_cleanup(struct bonding *bond)
1264 {
1265         struct slave *slave;
1266         int i;
1267
1268         bond_for_each_slave(bond, slave, i)
1269                 if (IS_UP(slave->dev))
1270                         slave_disable_netpoll(slave);
1271 }
1272 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1273 {
1274         struct bonding *bond = netdev_priv(bond_dev);
1275
1276         read_lock(&bond->lock);
1277         __bond_netpoll_cleanup(bond);
1278         read_unlock(&bond->lock);
1279 }
1280
1281 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1282 {
1283         struct bonding *bond = netdev_priv(dev);
1284         struct slave *slave;
1285         int i, err = 0;
1286
1287         read_lock(&bond->lock);
1288         bond_for_each_slave(bond, slave, i) {
1289                 err = slave_enable_netpoll(slave);
1290                 if (err) {
1291                         __bond_netpoll_cleanup(bond);
1292                         break;
1293                 }
1294         }
1295         read_unlock(&bond->lock);
1296         return err;
1297 }
1298
1299 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1300 {
1301         return bond->dev->npinfo;
1302 }
1303
1304 #else
1305 static inline int slave_enable_netpoll(struct slave *slave)
1306 {
1307         return 0;
1308 }
1309 static inline void slave_disable_netpoll(struct slave *slave)
1310 {
1311 }
1312 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1313 {
1314 }
1315 #endif
1316
1317 /*---------------------------------- IOCTL ----------------------------------*/
1318
1319 static int bond_sethwaddr(struct net_device *bond_dev,
1320                           struct net_device *slave_dev)
1321 {
1322         pr_debug("bond_dev=%p\n", bond_dev);
1323         pr_debug("slave_dev=%p\n", slave_dev);
1324         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1325         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1326         return 0;
1327 }
1328
1329 static u32 bond_fix_features(struct net_device *dev, u32 features)
1330 {
1331         struct slave *slave;
1332         struct bonding *bond = netdev_priv(dev);
1333         u32 mask;
1334         int i;
1335
1336         read_lock(&bond->lock);
1337
1338         if (!bond->first_slave) {
1339                 /* Disable adding VLANs to empty bond. But why? --mq */
1340                 features |= NETIF_F_VLAN_CHALLENGED;
1341                 goto out;
1342         }
1343
1344         mask = features;
1345         features &= ~NETIF_F_ONE_FOR_ALL;
1346         features |= NETIF_F_ALL_FOR_ALL;
1347
1348         bond_for_each_slave(bond, slave, i) {
1349                 features = netdev_increment_features(features,
1350                                                      slave->dev->features,
1351                                                      mask);
1352         }
1353
1354 out:
1355         read_unlock(&bond->lock);
1356         return features;
1357 }
1358
1359 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1360                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1361                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1362
1363 static void bond_compute_features(struct bonding *bond)
1364 {
1365         struct slave *slave;
1366         struct net_device *bond_dev = bond->dev;
1367         u32 vlan_features = BOND_VLAN_FEATURES;
1368         unsigned short max_hard_header_len = ETH_HLEN;
1369         int i;
1370
1371         read_lock(&bond->lock);
1372
1373         if (!bond->first_slave)
1374                 goto done;
1375
1376         bond_for_each_slave(bond, slave, i) {
1377                 vlan_features = netdev_increment_features(vlan_features,
1378                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1379
1380                 if (slave->dev->hard_header_len > max_hard_header_len)
1381                         max_hard_header_len = slave->dev->hard_header_len;
1382         }
1383
1384 done:
1385         bond_dev->vlan_features = vlan_features;
1386         bond_dev->hard_header_len = max_hard_header_len;
1387
1388         read_unlock(&bond->lock);
1389
1390         netdev_change_features(bond_dev);
1391 }
1392
1393 static void bond_setup_by_slave(struct net_device *bond_dev,
1394                                 struct net_device *slave_dev)
1395 {
1396         struct bonding *bond = netdev_priv(bond_dev);
1397
1398         bond_dev->header_ops        = slave_dev->header_ops;
1399
1400         bond_dev->type              = slave_dev->type;
1401         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1402         bond_dev->addr_len          = slave_dev->addr_len;
1403
1404         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1405                 slave_dev->addr_len);
1406         bond->setup_by_slave = 1;
1407 }
1408
1409 /* On bonding slaves other than the currently active slave, suppress
1410  * duplicates except for alb non-mcast/bcast.
1411  */
1412 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1413                                             struct slave *slave,
1414                                             struct bonding *bond)
1415 {
1416         if (bond_is_slave_inactive(slave)) {
1417                 if (bond->params.mode == BOND_MODE_ALB &&
1418                     skb->pkt_type != PACKET_BROADCAST &&
1419                     skb->pkt_type != PACKET_MULTICAST)
1420                         return false;
1421                 return true;
1422         }
1423         return false;
1424 }
1425
1426 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1427 {
1428         struct sk_buff *skb = *pskb;
1429         struct slave *slave;
1430         struct bonding *bond;
1431         void (*recv_probe)(struct sk_buff *, struct bonding *,
1432                                 struct slave *);
1433
1434         skb = skb_share_check(skb, GFP_ATOMIC);
1435         if (unlikely(!skb))
1436                 return RX_HANDLER_CONSUMED;
1437
1438         *pskb = skb;
1439
1440         slave = bond_slave_get_rcu(skb->dev);
1441         bond = slave->bond;
1442
1443         if (bond->params.arp_interval)
1444                 slave->dev->last_rx = jiffies;
1445
1446         recv_probe = ACCESS_ONCE(bond->recv_probe);
1447         if (recv_probe) {
1448                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1449
1450                 if (likely(nskb)) {
1451                         recv_probe(nskb, bond, slave);
1452                         dev_kfree_skb(nskb);
1453                 }
1454         }
1455
1456         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1457                 return RX_HANDLER_EXACT;
1458         }
1459
1460         skb->dev = bond->dev;
1461
1462         if (bond->params.mode == BOND_MODE_ALB &&
1463             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1464             skb->pkt_type == PACKET_HOST) {
1465
1466                 if (unlikely(skb_cow_head(skb,
1467                                           skb->data - skb_mac_header(skb)))) {
1468                         kfree_skb(skb);
1469                         return RX_HANDLER_CONSUMED;
1470                 }
1471                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1472         }
1473
1474         return RX_HANDLER_ANOTHER;
1475 }
1476
1477 /* enslave device <slave> to bond device <master> */
1478 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1479 {
1480         struct bonding *bond = netdev_priv(bond_dev);
1481         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1482         struct slave *new_slave = NULL;
1483         struct netdev_hw_addr *ha;
1484         struct sockaddr addr;
1485         int link_reporting;
1486         int res = 0;
1487
1488         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1489                 slave_ops->ndo_do_ioctl == NULL) {
1490                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1491                            bond_dev->name, slave_dev->name);
1492         }
1493
1494         /* already enslaved */
1495         if (slave_dev->flags & IFF_SLAVE) {
1496                 pr_debug("Error, Device was already enslaved\n");
1497                 return -EBUSY;
1498         }
1499
1500         /* vlan challenged mutual exclusion */
1501         /* no need to lock since we're protected by rtnl_lock */
1502         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1503                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1504                 if (bond_vlan_used(bond)) {
1505                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1506                                bond_dev->name, slave_dev->name, bond_dev->name);
1507                         return -EPERM;
1508                 } else {
1509                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1510                                    bond_dev->name, slave_dev->name,
1511                                    slave_dev->name, bond_dev->name);
1512                 }
1513         } else {
1514                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1515         }
1516
1517         /*
1518          * Old ifenslave binaries are no longer supported.  These can
1519          * be identified with moderate accuracy by the state of the slave:
1520          * the current ifenslave will set the interface down prior to
1521          * enslaving it; the old ifenslave will not.
1522          */
1523         if ((slave_dev->flags & IFF_UP)) {
1524                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1525                        slave_dev->name);
1526                 res = -EPERM;
1527                 goto err_undo_flags;
1528         }
1529
1530         /* set bonding device ether type by slave - bonding netdevices are
1531          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1532          * there is a need to override some of the type dependent attribs/funcs.
1533          *
1534          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1535          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1536          */
1537         if (bond->slave_cnt == 0) {
1538                 if (bond_dev->type != slave_dev->type) {
1539                         pr_debug("%s: change device type from %d to %d\n",
1540                                  bond_dev->name,
1541                                  bond_dev->type, slave_dev->type);
1542
1543                         res = netdev_bonding_change(bond_dev,
1544                                                     NETDEV_PRE_TYPE_CHANGE);
1545                         res = notifier_to_errno(res);
1546                         if (res) {
1547                                 pr_err("%s: refused to change device type\n",
1548                                        bond_dev->name);
1549                                 res = -EBUSY;
1550                                 goto err_undo_flags;
1551                         }
1552
1553                         /* Flush unicast and multicast addresses */
1554                         dev_uc_flush(bond_dev);
1555                         dev_mc_flush(bond_dev);
1556
1557                         if (slave_dev->type != ARPHRD_ETHER)
1558                                 bond_setup_by_slave(bond_dev, slave_dev);
1559                         else {
1560                                 ether_setup(bond_dev);
1561                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1562                         }
1563
1564                         netdev_bonding_change(bond_dev,
1565                                               NETDEV_POST_TYPE_CHANGE);
1566                 }
1567         } else if (bond_dev->type != slave_dev->type) {
1568                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1569                        slave_dev->name,
1570                        slave_dev->type, bond_dev->type);
1571                 res = -EINVAL;
1572                 goto err_undo_flags;
1573         }
1574
1575         if (slave_ops->ndo_set_mac_address == NULL) {
1576                 if (bond->slave_cnt == 0) {
1577                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1578                                    bond_dev->name);
1579                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1580                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1581                         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",
1582                                bond_dev->name);
1583                         res = -EOPNOTSUPP;
1584                         goto err_undo_flags;
1585                 }
1586         }
1587
1588         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1589
1590         /* If this is the first slave, then we need to set the master's hardware
1591          * address to be the same as the slave's. */
1592         if (is_zero_ether_addr(bond->dev->dev_addr))
1593                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1594                        slave_dev->addr_len);
1595
1596
1597         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1598         if (!new_slave) {
1599                 res = -ENOMEM;
1600                 goto err_undo_flags;
1601         }
1602
1603         /*
1604          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1605          * is set via sysfs or module option if desired.
1606          */
1607         new_slave->queue_id = 0;
1608
1609         /* Save slave's original mtu and then set it to match the bond */
1610         new_slave->original_mtu = slave_dev->mtu;
1611         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1612         if (res) {
1613                 pr_debug("Error %d calling dev_set_mtu\n", res);
1614                 goto err_free;
1615         }
1616
1617         /*
1618          * Save slave's original ("permanent") mac address for modes
1619          * that need it, and for restoring it upon release, and then
1620          * set it to the master's address
1621          */
1622         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1623
1624         if (!bond->params.fail_over_mac) {
1625                 /*
1626                  * Set slave to master's mac address.  The application already
1627                  * set the master's mac address to that of the first slave
1628                  */
1629                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1630                 addr.sa_family = slave_dev->type;
1631                 res = dev_set_mac_address(slave_dev, &addr);
1632                 if (res) {
1633                         pr_debug("Error %d calling set_mac_address\n", res);
1634                         goto err_restore_mtu;
1635                 }
1636         }
1637
1638         res = netdev_set_bond_master(slave_dev, bond_dev);
1639         if (res) {
1640                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1641                 goto err_restore_mac;
1642         }
1643
1644         /* open the slave since the application closed it */
1645         res = dev_open(slave_dev);
1646         if (res) {
1647                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1648                 goto err_unset_master;
1649         }
1650
1651         new_slave->bond = bond;
1652         new_slave->dev = slave_dev;
1653         slave_dev->priv_flags |= IFF_BONDING;
1654
1655         if (bond_is_lb(bond)) {
1656                 /* bond_alb_init_slave() must be called before all other stages since
1657                  * it might fail and we do not want to have to undo everything
1658                  */
1659                 res = bond_alb_init_slave(bond, new_slave);
1660                 if (res)
1661                         goto err_close;
1662         }
1663
1664         /* If the mode USES_PRIMARY, then the new slave gets the
1665          * master's promisc (and mc) settings only if it becomes the
1666          * curr_active_slave, and that is taken care of later when calling
1667          * bond_change_active()
1668          */
1669         if (!USES_PRIMARY(bond->params.mode)) {
1670                 /* set promiscuity level to new slave */
1671                 if (bond_dev->flags & IFF_PROMISC) {
1672                         res = dev_set_promiscuity(slave_dev, 1);
1673                         if (res)
1674                                 goto err_close;
1675                 }
1676
1677                 /* set allmulti level to new slave */
1678                 if (bond_dev->flags & IFF_ALLMULTI) {
1679                         res = dev_set_allmulti(slave_dev, 1);
1680                         if (res)
1681                                 goto err_close;
1682                 }
1683
1684                 netif_addr_lock_bh(bond_dev);
1685                 /* upload master's mc_list to new slave */
1686                 netdev_for_each_mc_addr(ha, bond_dev)
1687                         dev_mc_add(slave_dev, ha->addr);
1688                 netif_addr_unlock_bh(bond_dev);
1689         }
1690
1691         if (bond->params.mode == BOND_MODE_8023AD) {
1692                 /* add lacpdu mc addr to mc list */
1693                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1694
1695                 dev_mc_add(slave_dev, lacpdu_multicast);
1696         }
1697
1698         bond_add_vlans_on_slave(bond, slave_dev);
1699
1700         write_lock_bh(&bond->lock);
1701
1702         bond_attach_slave(bond, new_slave);
1703
1704         new_slave->delay = 0;
1705         new_slave->link_failure_count = 0;
1706
1707         write_unlock_bh(&bond->lock);
1708
1709         bond_compute_features(bond);
1710
1711         read_lock(&bond->lock);
1712
1713         new_slave->last_arp_rx = jiffies;
1714
1715         if (bond->params.miimon && !bond->params.use_carrier) {
1716                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1717
1718                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1719                         /*
1720                          * miimon is set but a bonded network driver
1721                          * does not support ETHTOOL/MII and
1722                          * arp_interval is not set.  Note: if
1723                          * use_carrier is enabled, we will never go
1724                          * here (because netif_carrier is always
1725                          * supported); thus, we don't need to change
1726                          * the messages for netif_carrier.
1727                          */
1728                         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",
1729                                bond_dev->name, slave_dev->name);
1730                 } else if (link_reporting == -1) {
1731                         /* unable get link status using mii/ethtool */
1732                         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",
1733                                    bond_dev->name, slave_dev->name);
1734                 }
1735         }
1736
1737         /* check for initial state */
1738         if (!bond->params.miimon ||
1739             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1740                 if (bond->params.updelay) {
1741                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1742                         new_slave->link  = BOND_LINK_BACK;
1743                         new_slave->delay = bond->params.updelay;
1744                 } else {
1745                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1746                         new_slave->link  = BOND_LINK_UP;
1747                 }
1748                 new_slave->jiffies = jiffies;
1749         } else {
1750                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1751                 new_slave->link  = BOND_LINK_DOWN;
1752         }
1753
1754         if (bond_update_speed_duplex(new_slave) &&
1755             (new_slave->link != BOND_LINK_DOWN)) {
1756                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1757                            bond_dev->name, new_slave->dev->name);
1758
1759                 if (bond->params.mode == BOND_MODE_8023AD) {
1760                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1761                                    bond_dev->name);
1762                 }
1763         }
1764
1765         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1766                 /* if there is a primary slave, remember it */
1767                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1768                         bond->primary_slave = new_slave;
1769                         bond->force_primary = true;
1770                 }
1771         }
1772
1773         write_lock_bh(&bond->curr_slave_lock);
1774
1775         switch (bond->params.mode) {
1776         case BOND_MODE_ACTIVEBACKUP:
1777                 bond_set_slave_inactive_flags(new_slave);
1778                 bond_select_active_slave(bond);
1779                 break;
1780         case BOND_MODE_8023AD:
1781                 /* in 802.3ad mode, the internal mechanism
1782                  * will activate the slaves in the selected
1783                  * aggregator
1784                  */
1785                 bond_set_slave_inactive_flags(new_slave);
1786                 /* if this is the first slave */
1787                 if (bond->slave_cnt == 1) {
1788                         SLAVE_AD_INFO(new_slave).id = 1;
1789                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1790                          * can be called only after the mac address of the bond is set
1791                          */
1792                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1793                 } else {
1794                         SLAVE_AD_INFO(new_slave).id =
1795                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1796                 }
1797
1798                 bond_3ad_bind_slave(new_slave);
1799                 break;
1800         case BOND_MODE_TLB:
1801         case BOND_MODE_ALB:
1802                 bond_set_active_slave(new_slave);
1803                 bond_set_slave_inactive_flags(new_slave);
1804                 bond_select_active_slave(bond);
1805                 break;
1806         default:
1807                 pr_debug("This slave is always active in trunk mode\n");
1808
1809                 /* always active in trunk mode */
1810                 bond_set_active_slave(new_slave);
1811
1812                 /* In trunking mode there is little meaning to curr_active_slave
1813                  * anyway (it holds no special properties of the bond device),
1814                  * so we can change it without calling change_active_interface()
1815                  */
1816                 if (!bond->curr_active_slave)
1817                         bond->curr_active_slave = new_slave;
1818
1819                 break;
1820         } /* switch(bond_mode) */
1821
1822         write_unlock_bh(&bond->curr_slave_lock);
1823
1824         bond_set_carrier(bond);
1825
1826 #ifdef CONFIG_NET_POLL_CONTROLLER
1827         slave_dev->npinfo = bond_netpoll_info(bond);
1828         if (slave_dev->npinfo) {
1829                 if (slave_enable_netpoll(new_slave)) {
1830                         read_unlock(&bond->lock);
1831                         pr_info("Error, %s: master_dev is using netpoll, "
1832                                  "but new slave device does not support netpoll.\n",
1833                                  bond_dev->name);
1834                         res = -EBUSY;
1835                         goto err_close;
1836                 }
1837         }
1838 #endif
1839
1840         read_unlock(&bond->lock);
1841
1842         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1843         if (res)
1844                 goto err_close;
1845
1846         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1847                                          new_slave);
1848         if (res) {
1849                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1850                 goto err_dest_symlinks;
1851         }
1852
1853         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1854                 bond_dev->name, slave_dev->name,
1855                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1856                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1857
1858         /* enslave is successful */
1859         return 0;
1860
1861 /* Undo stages on error */
1862 err_dest_symlinks:
1863         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1864
1865 err_close:
1866         dev_close(slave_dev);
1867
1868 err_unset_master:
1869         netdev_set_bond_master(slave_dev, NULL);
1870
1871 err_restore_mac:
1872         if (!bond->params.fail_over_mac) {
1873                 /* XXX TODO - fom follow mode needs to change master's
1874                  * MAC if this slave's MAC is in use by the bond, or at
1875                  * least print a warning.
1876                  */
1877                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1878                 addr.sa_family = slave_dev->type;
1879                 dev_set_mac_address(slave_dev, &addr);
1880         }
1881
1882 err_restore_mtu:
1883         dev_set_mtu(slave_dev, new_slave->original_mtu);
1884
1885 err_free:
1886         kfree(new_slave);
1887
1888 err_undo_flags:
1889         bond_compute_features(bond);
1890
1891         return res;
1892 }
1893
1894 /*
1895  * Try to release the slave device <slave> from the bond device <master>
1896  * It is legal to access curr_active_slave without a lock because all the function
1897  * is write-locked.
1898  *
1899  * The rules for slave state should be:
1900  *   for Active/Backup:
1901  *     Active stays on all backups go down
1902  *   for Bonded connections:
1903  *     The first up interface should be left on and all others downed.
1904  */
1905 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1906 {
1907         struct bonding *bond = netdev_priv(bond_dev);
1908         struct slave *slave, *oldcurrent;
1909         struct sockaddr addr;
1910         u32 old_features = bond_dev->features;
1911
1912         /* slave is not a slave or master is not master of this slave */
1913         if (!(slave_dev->flags & IFF_SLAVE) ||
1914             (slave_dev->master != bond_dev)) {
1915                 pr_err("%s: Error: cannot release %s.\n",
1916                        bond_dev->name, slave_dev->name);
1917                 return -EINVAL;
1918         }
1919
1920         block_netpoll_tx();
1921         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1922         write_lock_bh(&bond->lock);
1923
1924         slave = bond_get_slave_by_dev(bond, slave_dev);
1925         if (!slave) {
1926                 /* not a slave of this bond */
1927                 pr_info("%s: %s not enslaved\n",
1928                         bond_dev->name, slave_dev->name);
1929                 write_unlock_bh(&bond->lock);
1930                 unblock_netpoll_tx();
1931                 return -EINVAL;
1932         }
1933
1934         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1935          * for this slave anymore.
1936          */
1937         netdev_rx_handler_unregister(slave_dev);
1938         write_unlock_bh(&bond->lock);
1939         synchronize_net();
1940         write_lock_bh(&bond->lock);
1941
1942         if (!bond->params.fail_over_mac) {
1943                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1944                     bond->slave_cnt > 1)
1945                         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",
1946                                    bond_dev->name, slave_dev->name,
1947                                    slave->perm_hwaddr,
1948                                    bond_dev->name, slave_dev->name);
1949         }
1950
1951         /* Inform AD package of unbinding of slave. */
1952         if (bond->params.mode == BOND_MODE_8023AD) {
1953                 /* must be called before the slave is
1954                  * detached from the list
1955                  */
1956                 bond_3ad_unbind_slave(slave);
1957         }
1958
1959         pr_info("%s: releasing %s interface %s\n",
1960                 bond_dev->name,
1961                 bond_is_active_slave(slave) ? "active" : "backup",
1962                 slave_dev->name);
1963
1964         oldcurrent = bond->curr_active_slave;
1965
1966         bond->current_arp_slave = NULL;
1967
1968         /* release the slave from its bond */
1969         bond_detach_slave(bond, slave);
1970
1971         if (bond->primary_slave == slave)
1972                 bond->primary_slave = NULL;
1973
1974         if (oldcurrent == slave)
1975                 bond_change_active_slave(bond, NULL);
1976
1977         if (bond_is_lb(bond)) {
1978                 /* Must be called only after the slave has been
1979                  * detached from the list and the curr_active_slave
1980                  * has been cleared (if our_slave == old_current),
1981                  * but before a new active slave is selected.
1982                  */
1983                 write_unlock_bh(&bond->lock);
1984                 bond_alb_deinit_slave(bond, slave);
1985                 write_lock_bh(&bond->lock);
1986         }
1987
1988         if (oldcurrent == slave) {
1989                 /*
1990                  * Note that we hold RTNL over this sequence, so there
1991                  * is no concern that another slave add/remove event
1992                  * will interfere.
1993                  */
1994                 write_unlock_bh(&bond->lock);
1995                 read_lock(&bond->lock);
1996                 write_lock_bh(&bond->curr_slave_lock);
1997
1998                 bond_select_active_slave(bond);
1999
2000                 write_unlock_bh(&bond->curr_slave_lock);
2001                 read_unlock(&bond->lock);
2002                 write_lock_bh(&bond->lock);
2003         }
2004
2005         if (bond->slave_cnt == 0) {
2006                 bond_set_carrier(bond);
2007
2008                 /* if the last slave was removed, zero the mac address
2009                  * of the master so it will be set by the application
2010                  * to the mac address of the first slave
2011                  */
2012                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2013
2014                 if (bond_vlan_used(bond)) {
2015                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2016                                    bond_dev->name, bond_dev->name);
2017                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2018                                    bond_dev->name);
2019                 }
2020         }
2021
2022         write_unlock_bh(&bond->lock);
2023         unblock_netpoll_tx();
2024
2025         bond_compute_features(bond);
2026         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2027             (old_features & NETIF_F_VLAN_CHALLENGED))
2028                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2029                         bond_dev->name, slave_dev->name, bond_dev->name);
2030
2031         /* must do this from outside any spinlocks */
2032         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2033
2034         bond_del_vlans_from_slave(bond, slave_dev);
2035
2036         /* If the mode USES_PRIMARY, then we should only remove its
2037          * promisc and mc settings if it was the curr_active_slave, but that was
2038          * already taken care of above when we detached the slave
2039          */
2040         if (!USES_PRIMARY(bond->params.mode)) {
2041                 /* unset promiscuity level from slave */
2042                 if (bond_dev->flags & IFF_PROMISC)
2043                         dev_set_promiscuity(slave_dev, -1);
2044
2045                 /* unset allmulti level from slave */
2046                 if (bond_dev->flags & IFF_ALLMULTI)
2047                         dev_set_allmulti(slave_dev, -1);
2048
2049                 /* flush master's mc_list from slave */
2050                 netif_addr_lock_bh(bond_dev);
2051                 bond_mc_list_flush(bond_dev, slave_dev);
2052                 netif_addr_unlock_bh(bond_dev);
2053         }
2054
2055         netdev_set_bond_master(slave_dev, NULL);
2056
2057         slave_disable_netpoll(slave);
2058
2059         /* close slave before restoring its mac address */
2060         dev_close(slave_dev);
2061
2062         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2063                 /* restore original ("permanent") mac address */
2064                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2065                 addr.sa_family = slave_dev->type;
2066                 dev_set_mac_address(slave_dev, &addr);
2067         }
2068
2069         dev_set_mtu(slave_dev, slave->original_mtu);
2070
2071         slave_dev->priv_flags &= ~IFF_BONDING;
2072
2073         kfree(slave);
2074
2075         return 0;  /* deletion OK */
2076 }
2077
2078 /*
2079 * First release a slave and then destroy the bond if no more slaves are left.
2080 * Must be under rtnl_lock when this function is called.
2081 */
2082 static int  bond_release_and_destroy(struct net_device *bond_dev,
2083                                      struct net_device *slave_dev)
2084 {
2085         struct bonding *bond = netdev_priv(bond_dev);
2086         int ret;
2087
2088         ret = bond_release(bond_dev, slave_dev);
2089         if ((ret == 0) && (bond->slave_cnt == 0)) {
2090                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2091                 pr_info("%s: destroying bond %s.\n",
2092                         bond_dev->name, bond_dev->name);
2093                 unregister_netdevice(bond_dev);
2094         }
2095         return ret;
2096 }
2097
2098 /*
2099  * This function releases all slaves.
2100  */
2101 static int bond_release_all(struct net_device *bond_dev)
2102 {
2103         struct bonding *bond = netdev_priv(bond_dev);
2104         struct slave *slave;
2105         struct net_device *slave_dev;
2106         struct sockaddr addr;
2107
2108         write_lock_bh(&bond->lock);
2109
2110         netif_carrier_off(bond_dev);
2111
2112         if (bond->slave_cnt == 0)
2113                 goto out;
2114
2115         bond->current_arp_slave = NULL;
2116         bond->primary_slave = NULL;
2117         bond_change_active_slave(bond, NULL);
2118
2119         while ((slave = bond->first_slave) != NULL) {
2120                 /* Inform AD package of unbinding of slave
2121                  * before slave is detached from the list.
2122                  */
2123                 if (bond->params.mode == BOND_MODE_8023AD)
2124                         bond_3ad_unbind_slave(slave);
2125
2126                 slave_dev = slave->dev;
2127                 bond_detach_slave(bond, slave);
2128
2129                 /* now that the slave is detached, unlock and perform
2130                  * all the undo steps that should not be called from
2131                  * within a lock.
2132                  */
2133                 write_unlock_bh(&bond->lock);
2134
2135                 /* unregister rx_handler early so bond_handle_frame wouldn't
2136                  * be called for this slave anymore.
2137                  */
2138                 netdev_rx_handler_unregister(slave_dev);
2139                 synchronize_net();
2140
2141                 if (bond_is_lb(bond)) {
2142                         /* must be called only after the slave
2143                          * has been detached from the list
2144                          */
2145                         bond_alb_deinit_slave(bond, slave);
2146                 }
2147
2148                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2149                 bond_del_vlans_from_slave(bond, slave_dev);
2150
2151                 /* If the mode USES_PRIMARY, then we should only remove its
2152                  * promisc and mc settings if it was the curr_active_slave, but that was
2153                  * already taken care of above when we detached the slave
2154                  */
2155                 if (!USES_PRIMARY(bond->params.mode)) {
2156                         /* unset promiscuity level from slave */
2157                         if (bond_dev->flags & IFF_PROMISC)
2158                                 dev_set_promiscuity(slave_dev, -1);
2159
2160                         /* unset allmulti level from slave */
2161                         if (bond_dev->flags & IFF_ALLMULTI)
2162                                 dev_set_allmulti(slave_dev, -1);
2163
2164                         /* flush master's mc_list from slave */
2165                         netif_addr_lock_bh(bond_dev);
2166                         bond_mc_list_flush(bond_dev, slave_dev);
2167                         netif_addr_unlock_bh(bond_dev);
2168                 }
2169
2170                 netdev_set_bond_master(slave_dev, NULL);
2171
2172                 slave_disable_netpoll(slave);
2173
2174                 /* close slave before restoring its mac address */
2175                 dev_close(slave_dev);
2176
2177                 if (!bond->params.fail_over_mac) {
2178                         /* restore original ("permanent") mac address*/
2179                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2180                         addr.sa_family = slave_dev->type;
2181                         dev_set_mac_address(slave_dev, &addr);
2182                 }
2183
2184                 kfree(slave);
2185
2186                 /* re-acquire the lock before getting the next slave */
2187                 write_lock_bh(&bond->lock);
2188         }
2189
2190         /* zero the mac address of the master so it will be
2191          * set by the application to the mac address of the
2192          * first slave
2193          */
2194         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2195
2196         if (bond_vlan_used(bond)) {
2197                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2198                            bond_dev->name, bond_dev->name);
2199                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2200                            bond_dev->name);
2201         }
2202
2203         pr_info("%s: released all slaves\n", bond_dev->name);
2204
2205 out:
2206         write_unlock_bh(&bond->lock);
2207
2208         bond_compute_features(bond);
2209
2210         return 0;
2211 }
2212
2213 /*
2214  * This function changes the active slave to slave <slave_dev>.
2215  * It returns -EINVAL in the following cases.
2216  *  - <slave_dev> is not found in the list.
2217  *  - There is not active slave now.
2218  *  - <slave_dev> is already active.
2219  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2220  *  - <slave_dev> is not running.
2221  * In these cases, this function does nothing.
2222  * In the other cases, current_slave pointer is changed and 0 is returned.
2223  */
2224 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2225 {
2226         struct bonding *bond = netdev_priv(bond_dev);
2227         struct slave *old_active = NULL;
2228         struct slave *new_active = NULL;
2229         int res = 0;
2230
2231         if (!USES_PRIMARY(bond->params.mode))
2232                 return -EINVAL;
2233
2234         /* Verify that master_dev is indeed the master of slave_dev */
2235         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2236                 return -EINVAL;
2237
2238         read_lock(&bond->lock);
2239
2240         read_lock(&bond->curr_slave_lock);
2241         old_active = bond->curr_active_slave;
2242         read_unlock(&bond->curr_slave_lock);
2243
2244         new_active = bond_get_slave_by_dev(bond, slave_dev);
2245
2246         /*
2247          * Changing to the current active: do nothing; return success.
2248          */
2249         if (new_active && (new_active == old_active)) {
2250                 read_unlock(&bond->lock);
2251                 return 0;
2252         }
2253
2254         if ((new_active) &&
2255             (old_active) &&
2256             (new_active->link == BOND_LINK_UP) &&
2257             IS_UP(new_active->dev)) {
2258                 block_netpoll_tx();
2259                 write_lock_bh(&bond->curr_slave_lock);
2260                 bond_change_active_slave(bond, new_active);
2261                 write_unlock_bh(&bond->curr_slave_lock);
2262                 unblock_netpoll_tx();
2263         } else
2264                 res = -EINVAL;
2265
2266         read_unlock(&bond->lock);
2267
2268         return res;
2269 }
2270
2271 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2272 {
2273         struct bonding *bond = netdev_priv(bond_dev);
2274
2275         info->bond_mode = bond->params.mode;
2276         info->miimon = bond->params.miimon;
2277
2278         read_lock(&bond->lock);
2279         info->num_slaves = bond->slave_cnt;
2280         read_unlock(&bond->lock);
2281
2282         return 0;
2283 }
2284
2285 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2286 {
2287         struct bonding *bond = netdev_priv(bond_dev);
2288         struct slave *slave;
2289         int i, res = -ENODEV;
2290
2291         read_lock(&bond->lock);
2292
2293         bond_for_each_slave(bond, slave, i) {
2294                 if (i == (int)info->slave_id) {
2295                         res = 0;
2296                         strcpy(info->slave_name, slave->dev->name);
2297                         info->link = slave->link;
2298                         info->state = bond_slave_state(slave);
2299                         info->link_failure_count = slave->link_failure_count;
2300                         break;
2301                 }
2302         }
2303
2304         read_unlock(&bond->lock);
2305
2306         return res;
2307 }
2308
2309 /*-------------------------------- Monitoring -------------------------------*/
2310
2311
2312 static int bond_miimon_inspect(struct bonding *bond)
2313 {
2314         struct slave *slave;
2315         int i, link_state, commit = 0;
2316         bool ignore_updelay;
2317
2318         ignore_updelay = !bond->curr_active_slave ? true : false;
2319
2320         bond_for_each_slave(bond, slave, i) {
2321                 slave->new_link = BOND_LINK_NOCHANGE;
2322
2323                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2324
2325                 switch (slave->link) {
2326                 case BOND_LINK_UP:
2327                         if (link_state)
2328                                 continue;
2329
2330                         slave->link = BOND_LINK_FAIL;
2331                         slave->delay = bond->params.downdelay;
2332                         if (slave->delay) {
2333                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2334                                         bond->dev->name,
2335                                         (bond->params.mode ==
2336                                          BOND_MODE_ACTIVEBACKUP) ?
2337                                         (bond_is_active_slave(slave) ?
2338                                          "active " : "backup ") : "",
2339                                         slave->dev->name,
2340                                         bond->params.downdelay * bond->params.miimon);
2341                         }
2342                         /*FALLTHRU*/
2343                 case BOND_LINK_FAIL:
2344                         if (link_state) {
2345                                 /*
2346                                  * recovered before downdelay expired
2347                                  */
2348                                 slave->link = BOND_LINK_UP;
2349                                 slave->jiffies = jiffies;
2350                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2351                                         bond->dev->name,
2352                                         (bond->params.downdelay - slave->delay) *
2353                                         bond->params.miimon,
2354                                         slave->dev->name);
2355                                 continue;
2356                         }
2357
2358                         if (slave->delay <= 0) {
2359                                 slave->new_link = BOND_LINK_DOWN;
2360                                 commit++;
2361                                 continue;
2362                         }
2363
2364                         slave->delay--;
2365                         break;
2366
2367                 case BOND_LINK_DOWN:
2368                         if (!link_state)
2369                                 continue;
2370
2371                         slave->link = BOND_LINK_BACK;
2372                         slave->delay = bond->params.updelay;
2373
2374                         if (slave->delay) {
2375                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2376                                         bond->dev->name, slave->dev->name,
2377                                         ignore_updelay ? 0 :
2378                                         bond->params.updelay *
2379                                         bond->params.miimon);
2380                         }
2381                         /*FALLTHRU*/
2382                 case BOND_LINK_BACK:
2383                         if (!link_state) {
2384                                 slave->link = BOND_LINK_DOWN;
2385                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2386                                         bond->dev->name,
2387                                         (bond->params.updelay - slave->delay) *
2388                                         bond->params.miimon,
2389                                         slave->dev->name);
2390
2391                                 continue;
2392                         }
2393
2394                         if (ignore_updelay)
2395                                 slave->delay = 0;
2396
2397                         if (slave->delay <= 0) {
2398                                 slave->new_link = BOND_LINK_UP;
2399                                 commit++;
2400                                 ignore_updelay = false;
2401                                 continue;
2402                         }
2403
2404                         slave->delay--;
2405                         break;
2406                 }
2407         }
2408
2409         return commit;
2410 }
2411
2412 static void bond_miimon_commit(struct bonding *bond)
2413 {
2414         struct slave *slave;
2415         int i;
2416
2417         bond_for_each_slave(bond, slave, i) {
2418                 switch (slave->new_link) {
2419                 case BOND_LINK_NOCHANGE:
2420                         continue;
2421
2422                 case BOND_LINK_UP:
2423                         slave->link = BOND_LINK_UP;
2424                         slave->jiffies = jiffies;
2425
2426                         if (bond->params.mode == BOND_MODE_8023AD) {
2427                                 /* prevent it from being the active one */
2428                                 bond_set_backup_slave(slave);
2429                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2430                                 /* make it immediately active */
2431                                 bond_set_active_slave(slave);
2432                         } else if (slave != bond->primary_slave) {
2433                                 /* prevent it from being the active one */
2434                                 bond_set_backup_slave(slave);
2435                         }
2436
2437                         bond_update_speed_duplex(slave);
2438
2439                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2440                                 bond->dev->name, slave->dev->name,
2441                                 slave->speed, slave->duplex ? "full" : "half");
2442
2443                         /* notify ad that the link status has changed */
2444                         if (bond->params.mode == BOND_MODE_8023AD)
2445                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2446
2447                         if (bond_is_lb(bond))
2448                                 bond_alb_handle_link_change(bond, slave,
2449                                                             BOND_LINK_UP);
2450
2451                         if (!bond->curr_active_slave ||
2452                             (slave == bond->primary_slave))
2453                                 goto do_failover;
2454
2455                         continue;
2456
2457                 case BOND_LINK_DOWN:
2458                         if (slave->link_failure_count < UINT_MAX)
2459                                 slave->link_failure_count++;
2460
2461                         slave->link = BOND_LINK_DOWN;
2462
2463                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2464                             bond->params.mode == BOND_MODE_8023AD)
2465                                 bond_set_slave_inactive_flags(slave);
2466
2467                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2468                                 bond->dev->name, slave->dev->name);
2469
2470                         if (bond->params.mode == BOND_MODE_8023AD)
2471                                 bond_3ad_handle_link_change(slave,
2472                                                             BOND_LINK_DOWN);
2473
2474                         if (bond_is_lb(bond))
2475                                 bond_alb_handle_link_change(bond, slave,
2476                                                             BOND_LINK_DOWN);
2477
2478                         if (slave == bond->curr_active_slave)
2479                                 goto do_failover;
2480
2481                         continue;
2482
2483                 default:
2484                         pr_err("%s: invalid new link %d on slave %s\n",
2485                                bond->dev->name, slave->new_link,
2486                                slave->dev->name);
2487                         slave->new_link = BOND_LINK_NOCHANGE;
2488
2489                         continue;
2490                 }
2491
2492 do_failover:
2493                 ASSERT_RTNL();
2494                 block_netpoll_tx();
2495                 write_lock_bh(&bond->curr_slave_lock);
2496                 bond_select_active_slave(bond);
2497                 write_unlock_bh(&bond->curr_slave_lock);
2498                 unblock_netpoll_tx();
2499         }
2500
2501         bond_set_carrier(bond);
2502 }
2503
2504 /*
2505  * bond_mii_monitor
2506  *
2507  * Really a wrapper that splits the mii monitor into two phases: an
2508  * inspection, then (if inspection indicates something needs to be done)
2509  * an acquisition of appropriate locks followed by a commit phase to
2510  * implement whatever link state changes are indicated.
2511  */
2512 void bond_mii_monitor(struct work_struct *work)
2513 {
2514         struct bonding *bond = container_of(work, struct bonding,
2515                                             mii_work.work);
2516         bool should_notify_peers = false;
2517         unsigned long delay;
2518
2519         read_lock(&bond->lock);
2520
2521         delay = msecs_to_jiffies(bond->params.miimon);
2522
2523         if (bond->slave_cnt == 0)
2524                 goto re_arm;
2525
2526         should_notify_peers = bond_should_notify_peers(bond);
2527
2528         if (bond_miimon_inspect(bond)) {
2529                 read_unlock(&bond->lock);
2530
2531                 /* Race avoidance with bond_close cancel of workqueue */
2532                 if (!rtnl_trylock()) {
2533                         read_lock(&bond->lock);
2534                         delay = 1;
2535                         should_notify_peers = false;
2536                         goto re_arm;
2537                 }
2538
2539                 read_lock(&bond->lock);
2540
2541                 bond_miimon_commit(bond);
2542
2543                 read_unlock(&bond->lock);
2544                 rtnl_unlock();  /* might sleep, hold no other locks */
2545                 read_lock(&bond->lock);
2546         }
2547
2548 re_arm:
2549         if (bond->params.miimon)
2550                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2551
2552         read_unlock(&bond->lock);
2553
2554         if (should_notify_peers) {
2555                 if (!rtnl_trylock()) {
2556                         read_lock(&bond->lock);
2557                         bond->send_peer_notif++;
2558                         read_unlock(&bond->lock);
2559                         return;
2560                 }
2561                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2562                 rtnl_unlock();
2563         }
2564 }
2565
2566 static __be32 bond_glean_dev_ip(struct net_device *dev)
2567 {
2568         struct in_device *idev;
2569         struct in_ifaddr *ifa;
2570         __be32 addr = 0;
2571
2572         if (!dev)
2573                 return 0;
2574
2575         rcu_read_lock();
2576         idev = __in_dev_get_rcu(dev);
2577         if (!idev)
2578                 goto out;
2579
2580         ifa = idev->ifa_list;
2581         if (!ifa)
2582                 goto out;
2583
2584         addr = ifa->ifa_local;
2585 out:
2586         rcu_read_unlock();
2587         return addr;
2588 }
2589
2590 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2591 {
2592         struct vlan_entry *vlan;
2593
2594         if (ip == bond->master_ip)
2595                 return 1;
2596
2597         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2598                 if (ip == vlan->vlan_ip)
2599                         return 1;
2600         }
2601
2602         return 0;
2603 }
2604
2605 /*
2606  * We go to the (large) trouble of VLAN tagging ARP frames because
2607  * switches in VLAN mode (especially if ports are configured as
2608  * "native" to a VLAN) might not pass non-tagged frames.
2609  */
2610 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2611 {
2612         struct sk_buff *skb;
2613
2614         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2615                  slave_dev->name, dest_ip, src_ip, vlan_id);
2616
2617         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2618                          NULL, slave_dev->dev_addr, NULL);
2619
2620         if (!skb) {
2621                 pr_err("ARP packet allocation failed\n");
2622                 return;
2623         }
2624         if (vlan_id) {
2625                 skb = vlan_put_tag(skb, vlan_id);
2626                 if (!skb) {
2627                         pr_err("failed to insert VLAN tag\n");
2628                         return;
2629                 }
2630         }
2631         arp_xmit(skb);
2632 }
2633
2634
2635 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2636 {
2637         int i, vlan_id;
2638         __be32 *targets = bond->params.arp_targets;
2639         struct vlan_entry *vlan;
2640         struct net_device *vlan_dev;
2641         struct rtable *rt;
2642
2643         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2644                 if (!targets[i])
2645                         break;
2646                 pr_debug("basa: target %x\n", targets[i]);
2647                 if (!bond_vlan_used(bond)) {
2648                         pr_debug("basa: empty vlan: arp_send\n");
2649                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2650                                       bond->master_ip, 0);
2651                         continue;
2652                 }
2653
2654                 /*
2655                  * If VLANs are configured, we do a route lookup to
2656                  * determine which VLAN interface would be used, so we
2657                  * can tag the ARP with the proper VLAN tag.
2658                  */
2659                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2660                                      RTO_ONLINK, 0);
2661                 if (IS_ERR(rt)) {
2662                         if (net_ratelimit()) {
2663                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2664                                            bond->dev->name, &targets[i]);
2665                         }
2666                         continue;
2667                 }
2668
2669                 /*
2670                  * This target is not on a VLAN
2671                  */
2672                 if (rt->dst.dev == bond->dev) {
2673                         ip_rt_put(rt);
2674                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2675                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2676                                       bond->master_ip, 0);
2677                         continue;
2678                 }
2679
2680                 vlan_id = 0;
2681                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2682                         rcu_read_lock();
2683                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2684                                                         vlan->vlan_id);
2685                         rcu_read_unlock();
2686                         if (vlan_dev == rt->dst.dev) {
2687                                 vlan_id = vlan->vlan_id;
2688                                 pr_debug("basa: vlan match on %s %d\n",
2689                                        vlan_dev->name, vlan_id);
2690                                 break;
2691                         }
2692                 }
2693
2694                 if (vlan_id) {
2695                         ip_rt_put(rt);
2696                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2697                                       vlan->vlan_ip, vlan_id);
2698                         continue;
2699                 }
2700
2701                 if (net_ratelimit()) {
2702                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2703                                    bond->dev->name, &targets[i],
2704                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2705                 }
2706                 ip_rt_put(rt);
2707         }
2708 }
2709
2710 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2711 {
2712         int i;
2713         __be32 *targets = bond->params.arp_targets;
2714
2715         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2716                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2717                          &sip, &tip, i, &targets[i],
2718                          bond_has_this_ip(bond, tip));
2719                 if (sip == targets[i]) {
2720                         if (bond_has_this_ip(bond, tip))
2721                                 slave->last_arp_rx = jiffies;
2722                         return;
2723                 }
2724         }
2725 }
2726
2727 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2728                          struct slave *slave)
2729 {
2730         struct arphdr *arp;
2731         unsigned char *arp_ptr;
2732         __be32 sip, tip;
2733
2734         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2735                 return;
2736
2737         read_lock(&bond->lock);
2738
2739         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2740                  bond->dev->name, skb->dev->name);
2741
2742         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2743                 goto out_unlock;
2744
2745         arp = arp_hdr(skb);
2746         if (arp->ar_hln != bond->dev->addr_len ||
2747             skb->pkt_type == PACKET_OTHERHOST ||
2748             skb->pkt_type == PACKET_LOOPBACK ||
2749             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2750             arp->ar_pro != htons(ETH_P_IP) ||
2751             arp->ar_pln != 4)
2752                 goto out_unlock;
2753
2754         arp_ptr = (unsigned char *)(arp + 1);
2755         arp_ptr += bond->dev->addr_len;
2756         memcpy(&sip, arp_ptr, 4);
2757         arp_ptr += 4 + bond->dev->addr_len;
2758         memcpy(&tip, arp_ptr, 4);
2759
2760         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2761                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2762                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2763                  &sip, &tip);
2764
2765         /*
2766          * Backup slaves won't see the ARP reply, but do come through
2767          * here for each ARP probe (so we swap the sip/tip to validate
2768          * the probe).  In a "redundant switch, common router" type of
2769          * configuration, the ARP probe will (hopefully) travel from
2770          * the active, through one switch, the router, then the other
2771          * switch before reaching the backup.
2772          */
2773         if (bond_is_active_slave(slave))
2774                 bond_validate_arp(bond, slave, sip, tip);
2775         else
2776                 bond_validate_arp(bond, slave, tip, sip);
2777
2778 out_unlock:
2779         read_unlock(&bond->lock);
2780 }
2781
2782 /*
2783  * this function is called regularly to monitor each slave's link
2784  * ensuring that traffic is being sent and received when arp monitoring
2785  * is used in load-balancing mode. if the adapter has been dormant, then an
2786  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2787  * arp monitoring in active backup mode.
2788  */
2789 void bond_loadbalance_arp_mon(struct work_struct *work)
2790 {
2791         struct bonding *bond = container_of(work, struct bonding,
2792                                             arp_work.work);
2793         struct slave *slave, *oldcurrent;
2794         int do_failover = 0;
2795         int delta_in_ticks;
2796         int i;
2797
2798         read_lock(&bond->lock);
2799
2800         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2801
2802         if (bond->slave_cnt == 0)
2803                 goto re_arm;
2804
2805         read_lock(&bond->curr_slave_lock);
2806         oldcurrent = bond->curr_active_slave;
2807         read_unlock(&bond->curr_slave_lock);
2808
2809         /* see if any of the previous devices are up now (i.e. they have
2810          * xmt and rcv traffic). the curr_active_slave does not come into
2811          * the picture unless it is null. also, slave->jiffies is not needed
2812          * here because we send an arp on each slave and give a slave as
2813          * long as it needs to get the tx/rx within the delta.
2814          * TODO: what about up/down delay in arp mode? it wasn't here before
2815          *       so it can wait
2816          */
2817         bond_for_each_slave(bond, slave, i) {
2818                 unsigned long trans_start = dev_trans_start(slave->dev);
2819
2820                 if (slave->link != BOND_LINK_UP) {
2821                         if (time_in_range(jiffies,
2822                                 trans_start - delta_in_ticks,
2823                                 trans_start + delta_in_ticks) &&
2824                             time_in_range(jiffies,
2825                                 slave->dev->last_rx - delta_in_ticks,
2826                                 slave->dev->last_rx + delta_in_ticks)) {
2827
2828                                 slave->link  = BOND_LINK_UP;
2829                                 bond_set_active_slave(slave);
2830
2831                                 /* primary_slave has no meaning in round-robin
2832                                  * mode. the window of a slave being up and
2833                                  * curr_active_slave being null after enslaving
2834                                  * is closed.
2835                                  */
2836                                 if (!oldcurrent) {
2837                                         pr_info("%s: link status definitely up for interface %s, ",
2838                                                 bond->dev->name,
2839                                                 slave->dev->name);
2840                                         do_failover = 1;
2841                                 } else {
2842                                         pr_info("%s: interface %s is now up\n",
2843                                                 bond->dev->name,
2844                                                 slave->dev->name);
2845                                 }
2846                         }
2847                 } else {
2848                         /* slave->link == BOND_LINK_UP */
2849
2850                         /* not all switches will respond to an arp request
2851                          * when the source ip is 0, so don't take the link down
2852                          * if we don't know our ip yet
2853                          */
2854                         if (!time_in_range(jiffies,
2855                                 trans_start - delta_in_ticks,
2856                                 trans_start + 2 * delta_in_ticks) ||
2857                             !time_in_range(jiffies,
2858                                 slave->dev->last_rx - delta_in_ticks,
2859                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2860
2861                                 slave->link  = BOND_LINK_DOWN;
2862                                 bond_set_backup_slave(slave);
2863
2864                                 if (slave->link_failure_count < UINT_MAX)
2865                                         slave->link_failure_count++;
2866
2867                                 pr_info("%s: interface %s is now down.\n",
2868                                         bond->dev->name,
2869                                         slave->dev->name);
2870
2871                                 if (slave == oldcurrent)
2872                                         do_failover = 1;
2873                         }
2874                 }
2875
2876                 /* note: if switch is in round-robin mode, all links
2877                  * must tx arp to ensure all links rx an arp - otherwise
2878                  * links may oscillate or not come up at all; if switch is
2879                  * in something like xor mode, there is nothing we can
2880                  * do - all replies will be rx'ed on same link causing slaves
2881                  * to be unstable during low/no traffic periods
2882                  */
2883                 if (IS_UP(slave->dev))
2884                         bond_arp_send_all(bond, slave);
2885         }
2886
2887         if (do_failover) {
2888                 block_netpoll_tx();
2889                 write_lock_bh(&bond->curr_slave_lock);
2890
2891                 bond_select_active_slave(bond);
2892
2893                 write_unlock_bh(&bond->curr_slave_lock);
2894                 unblock_netpoll_tx();
2895         }
2896
2897 re_arm:
2898         if (bond->params.arp_interval)
2899                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2900
2901         read_unlock(&bond->lock);
2902 }
2903
2904 /*
2905  * Called to inspect slaves for active-backup mode ARP monitor link state
2906  * changes.  Sets new_link in slaves to specify what action should take
2907  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2908  * to link states must be committed.
2909  *
2910  * Called with bond->lock held for read.
2911  */
2912 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2913 {
2914         struct slave *slave;
2915         int i, commit = 0;
2916         unsigned long trans_start;
2917
2918         bond_for_each_slave(bond, slave, i) {
2919                 slave->new_link = BOND_LINK_NOCHANGE;
2920
2921                 if (slave->link != BOND_LINK_UP) {
2922                         if (time_in_range(jiffies,
2923                                 slave_last_rx(bond, slave) - delta_in_ticks,
2924                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2925
2926                                 slave->new_link = BOND_LINK_UP;
2927                                 commit++;
2928                         }
2929
2930                         continue;
2931                 }
2932
2933                 /*
2934                  * Give slaves 2*delta after being enslaved or made
2935                  * active.  This avoids bouncing, as the last receive
2936                  * times need a full ARP monitor cycle to be updated.
2937                  */
2938                 if (time_in_range(jiffies,
2939                                   slave->jiffies - delta_in_ticks,
2940                                   slave->jiffies + 2 * delta_in_ticks))
2941                         continue;
2942
2943                 /*
2944                  * Backup slave is down if:
2945                  * - No current_arp_slave AND
2946                  * - more than 3*delta since last receive AND
2947                  * - the bond has an IP address
2948                  *
2949                  * Note: a non-null current_arp_slave indicates
2950                  * the curr_active_slave went down and we are
2951                  * searching for a new one; under this condition
2952                  * we only take the curr_active_slave down - this
2953                  * gives each slave a chance to tx/rx traffic
2954                  * before being taken out
2955                  */
2956                 if (!bond_is_active_slave(slave) &&
2957                     !bond->current_arp_slave &&
2958                     !time_in_range(jiffies,
2959                         slave_last_rx(bond, slave) - delta_in_ticks,
2960                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2961
2962                         slave->new_link = BOND_LINK_DOWN;
2963                         commit++;
2964                 }
2965
2966                 /*
2967                  * Active slave is down if:
2968                  * - more than 2*delta since transmitting OR
2969                  * - (more than 2*delta since receive AND
2970                  *    the bond has an IP address)
2971                  */
2972                 trans_start = dev_trans_start(slave->dev);
2973                 if (bond_is_active_slave(slave) &&
2974                     (!time_in_range(jiffies,
2975                         trans_start - delta_in_ticks,
2976                         trans_start + 2 * delta_in_ticks) ||
2977                      !time_in_range(jiffies,
2978                         slave_last_rx(bond, slave) - delta_in_ticks,
2979                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2980
2981                         slave->new_link = BOND_LINK_DOWN;
2982                         commit++;
2983                 }
2984         }
2985
2986         return commit;
2987 }
2988
2989 /*
2990  * Called to commit link state changes noted by inspection step of
2991  * active-backup mode ARP monitor.
2992  *
2993  * Called with RTNL and bond->lock for read.
2994  */
2995 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2996 {
2997         struct slave *slave;
2998         int i;
2999         unsigned long trans_start;
3000
3001         bond_for_each_slave(bond, slave, i) {
3002                 switch (slave->new_link) {
3003                 case BOND_LINK_NOCHANGE:
3004                         continue;
3005
3006                 case BOND_LINK_UP:
3007                         trans_start = dev_trans_start(slave->dev);
3008                         if ((!bond->curr_active_slave &&
3009                              time_in_range(jiffies,
3010                                            trans_start - delta_in_ticks,
3011                                            trans_start + delta_in_ticks)) ||
3012                             bond->curr_active_slave != slave) {
3013                                 slave->link = BOND_LINK_UP;
3014                                 bond->current_arp_slave = NULL;
3015
3016                                 pr_info("%s: link status definitely up for interface %s.\n",
3017                                         bond->dev->name, slave->dev->name);
3018
3019                                 if (!bond->curr_active_slave ||
3020                                     (slave == bond->primary_slave))
3021                                         goto do_failover;
3022
3023                         }
3024
3025                         continue;
3026
3027                 case BOND_LINK_DOWN:
3028                         if (slave->link_failure_count < UINT_MAX)
3029                                 slave->link_failure_count++;
3030
3031                         slave->link = BOND_LINK_DOWN;
3032                         bond_set_slave_inactive_flags(slave);
3033
3034                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3035                                 bond->dev->name, slave->dev->name);
3036
3037                         if (slave == bond->curr_active_slave) {
3038                                 bond->current_arp_slave = NULL;
3039                                 goto do_failover;
3040                         }
3041
3042                         continue;
3043
3044                 default:
3045                         pr_err("%s: impossible: new_link %d on slave %s\n",
3046                                bond->dev->name, slave->new_link,
3047                                slave->dev->name);
3048                         continue;
3049                 }
3050
3051 do_failover:
3052                 ASSERT_RTNL();
3053                 block_netpoll_tx();
3054                 write_lock_bh(&bond->curr_slave_lock);
3055                 bond_select_active_slave(bond);
3056                 write_unlock_bh(&bond->curr_slave_lock);
3057                 unblock_netpoll_tx();
3058         }
3059
3060         bond_set_carrier(bond);
3061 }
3062
3063 /*
3064  * Send ARP probes for active-backup mode ARP monitor.
3065  *
3066  * Called with bond->lock held for read.
3067  */
3068 static void bond_ab_arp_probe(struct bonding *bond)
3069 {
3070         struct slave *slave;
3071         int i;
3072
3073         read_lock(&bond->curr_slave_lock);
3074
3075         if (bond->current_arp_slave && bond->curr_active_slave)
3076                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3077                         bond->current_arp_slave->dev->name,
3078                         bond->curr_active_slave->dev->name);
3079
3080         if (bond->curr_active_slave) {
3081                 bond_arp_send_all(bond, bond->curr_active_slave);
3082                 read_unlock(&bond->curr_slave_lock);
3083                 return;
3084         }
3085
3086         read_unlock(&bond->curr_slave_lock);
3087
3088         /* if we don't have a curr_active_slave, search for the next available
3089          * backup slave from the current_arp_slave and make it the candidate
3090          * for becoming the curr_active_slave
3091          */
3092
3093         if (!bond->current_arp_slave) {
3094                 bond->current_arp_slave = bond->first_slave;
3095                 if (!bond->current_arp_slave)
3096                         return;
3097         }
3098
3099         bond_set_slave_inactive_flags(bond->current_arp_slave);
3100
3101         /* search for next candidate */
3102         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3103                 if (IS_UP(slave->dev)) {
3104                         slave->link = BOND_LINK_BACK;
3105                         bond_set_slave_active_flags(slave);
3106                         bond_arp_send_all(bond, slave);
3107                         slave->jiffies = jiffies;
3108                         bond->current_arp_slave = slave;
3109                         break;
3110                 }
3111
3112                 /* if the link state is up at this point, we
3113                  * mark it down - this can happen if we have
3114                  * simultaneous link failures and
3115                  * reselect_active_interface doesn't make this
3116                  * one the current slave so it is still marked
3117                  * up when it is actually down
3118                  */
3119                 if (slave->link == BOND_LINK_UP) {
3120                         slave->link = BOND_LINK_DOWN;
3121                         if (slave->link_failure_count < UINT_MAX)
3122                                 slave->link_failure_count++;
3123
3124                         bond_set_slave_inactive_flags(slave);
3125
3126                         pr_info("%s: backup interface %s is now down.\n",
3127                                 bond->dev->name, slave->dev->name);
3128                 }
3129         }
3130 }
3131
3132 void bond_activebackup_arp_mon(struct work_struct *work)
3133 {
3134         struct bonding *bond = container_of(work, struct bonding,
3135                                             arp_work.work);
3136         bool should_notify_peers = false;
3137         int delta_in_ticks;
3138
3139         read_lock(&bond->lock);
3140
3141         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3142
3143         if (bond->slave_cnt == 0)
3144                 goto re_arm;
3145
3146         should_notify_peers = bond_should_notify_peers(bond);
3147
3148         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3149                 read_unlock(&bond->lock);
3150
3151                 /* Race avoidance with bond_close flush of workqueue */
3152                 if (!rtnl_trylock()) {
3153                         read_lock(&bond->lock);
3154                         delta_in_ticks = 1;
3155                         should_notify_peers = false;
3156                         goto re_arm;
3157                 }
3158
3159                 read_lock(&bond->lock);
3160
3161                 bond_ab_arp_commit(bond, delta_in_ticks);
3162
3163                 read_unlock(&bond->lock);
3164                 rtnl_unlock();
3165                 read_lock(&bond->lock);
3166         }
3167
3168         bond_ab_arp_probe(bond);
3169
3170 re_arm:
3171         if (bond->params.arp_interval)
3172                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3173
3174         read_unlock(&bond->lock);
3175
3176         if (should_notify_peers) {
3177                 if (!rtnl_trylock()) {
3178                         read_lock(&bond->lock);
3179                         bond->send_peer_notif++;
3180                         read_unlock(&bond->lock);
3181                         return;
3182                 }
3183                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3184                 rtnl_unlock();
3185         }
3186 }
3187
3188 /*-------------------------- netdev event handling --------------------------*/
3189
3190 /*
3191  * Change device name
3192  */
3193 static int bond_event_changename(struct bonding *bond)
3194 {
3195         bond_remove_proc_entry(bond);
3196         bond_create_proc_entry(bond);
3197
3198         bond_debug_reregister(bond);
3199
3200         return NOTIFY_DONE;
3201 }
3202
3203 static int bond_master_netdev_event(unsigned long event,
3204                                     struct net_device *bond_dev)
3205 {
3206         struct bonding *event_bond = netdev_priv(bond_dev);
3207
3208         switch (event) {
3209         case NETDEV_CHANGENAME:
3210                 return bond_event_changename(event_bond);
3211         default:
3212                 break;
3213         }
3214
3215         return NOTIFY_DONE;
3216 }
3217
3218 static int bond_slave_netdev_event(unsigned long event,
3219                                    struct net_device *slave_dev)
3220 {
3221         struct net_device *bond_dev = slave_dev->master;
3222         struct bonding *bond = netdev_priv(bond_dev);
3223
3224         switch (event) {
3225         case NETDEV_UNREGISTER:
3226                 if (bond_dev) {
3227                         if (bond->setup_by_slave)
3228                                 bond_release_and_destroy(bond_dev, slave_dev);
3229                         else
3230                                 bond_release(bond_dev, slave_dev);
3231                 }
3232                 break;
3233         case NETDEV_CHANGE:
3234                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3235                         struct slave *slave;
3236
3237                         slave = bond_get_slave_by_dev(bond, slave_dev);
3238                         if (slave) {
3239                                 u32 old_speed = slave->speed;
3240                                 u8  old_duplex = slave->duplex;
3241
3242                                 bond_update_speed_duplex(slave);
3243
3244                                 if (bond_is_lb(bond))
3245                                         break;
3246
3247                                 if (old_speed != slave->speed)
3248                                         bond_3ad_adapter_speed_changed(slave);
3249                                 if (old_duplex != slave->duplex)
3250                                         bond_3ad_adapter_duplex_changed(slave);
3251                         }
3252                 }
3253
3254                 break;
3255         case NETDEV_DOWN:
3256                 /*
3257                  * ... Or is it this?
3258                  */
3259                 break;
3260         case NETDEV_CHANGEMTU:
3261                 /*
3262                  * TODO: Should slaves be allowed to
3263                  * independently alter their MTU?  For
3264                  * an active-backup bond, slaves need
3265                  * not be the same type of device, so
3266                  * MTUs may vary.  For other modes,
3267                  * slaves arguably should have the
3268                  * same MTUs. To do this, we'd need to
3269                  * take over the slave's change_mtu
3270                  * function for the duration of their
3271                  * servitude.
3272                  */
3273                 break;
3274         case NETDEV_CHANGENAME:
3275                 /*
3276                  * TODO: handle changing the primary's name
3277                  */
3278                 break;
3279         case NETDEV_FEAT_CHANGE:
3280                 bond_compute_features(bond);
3281                 break;
3282         default:
3283                 break;
3284         }
3285
3286         return NOTIFY_DONE;
3287 }
3288
3289 /*
3290  * bond_netdev_event: handle netdev notifier chain events.
3291  *
3292  * This function receives events for the netdev chain.  The caller (an
3293  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3294  * locks for us to safely manipulate the slave devices (RTNL lock,
3295  * dev_probe_lock).
3296  */
3297 static int bond_netdev_event(struct notifier_block *this,
3298                              unsigned long event, void *ptr)
3299 {
3300         struct net_device *event_dev = (struct net_device *)ptr;
3301
3302         pr_debug("event_dev: %s, event: %lx\n",
3303                  event_dev ? event_dev->name : "None",
3304                  event);
3305
3306         if (!(event_dev->priv_flags & IFF_BONDING))
3307                 return NOTIFY_DONE;
3308
3309         if (event_dev->flags & IFF_MASTER) {
3310                 pr_debug("IFF_MASTER\n");
3311                 return bond_master_netdev_event(event, event_dev);
3312         }
3313
3314         if (event_dev->flags & IFF_SLAVE) {
3315                 pr_debug("IFF_SLAVE\n");
3316                 return bond_slave_netdev_event(event, event_dev);
3317         }
3318
3319         return NOTIFY_DONE;
3320 }
3321
3322 /*
3323  * bond_inetaddr_event: handle inetaddr notifier chain events.
3324  *
3325  * We keep track of device IPs primarily to use as source addresses in
3326  * ARP monitor probes (rather than spewing out broadcasts all the time).
3327  *
3328  * We track one IP for the main device (if it has one), plus one per VLAN.
3329  */
3330 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3331 {
3332         struct in_ifaddr *ifa = ptr;
3333         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3334         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3335         struct bonding *bond;
3336         struct vlan_entry *vlan;
3337
3338         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3339                 if (bond->dev == event_dev) {
3340                         switch (event) {
3341                         case NETDEV_UP:
3342                                 bond->master_ip = ifa->ifa_local;
3343                                 return NOTIFY_OK;
3344                         case NETDEV_DOWN:
3345                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3346                                 return NOTIFY_OK;
3347                         default:
3348                                 return NOTIFY_DONE;
3349                         }
3350                 }
3351
3352                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3353                         vlan_dev = __vlan_find_dev_deep(bond->dev,
3354                                                         vlan->vlan_id);
3355                         if (vlan_dev == event_dev) {
3356                                 switch (event) {
3357                                 case NETDEV_UP:
3358                                         vlan->vlan_ip = ifa->ifa_local;
3359                                         return NOTIFY_OK;
3360                                 case NETDEV_DOWN:
3361                                         vlan->vlan_ip =
3362                                                 bond_glean_dev_ip(vlan_dev);
3363                                         return NOTIFY_OK;
3364                                 default:
3365                                         return NOTIFY_DONE;
3366                                 }
3367                         }
3368                 }
3369         }
3370         return NOTIFY_DONE;
3371 }
3372
3373 static struct notifier_block bond_netdev_notifier = {
3374         .notifier_call = bond_netdev_event,
3375 };
3376
3377 static struct notifier_block bond_inetaddr_notifier = {
3378         .notifier_call = bond_inetaddr_event,
3379 };
3380
3381 /*---------------------------- Hashing Policies -----------------------------*/
3382
3383 /*
3384  * Hash for the output device based upon layer 2 and layer 3 data. If
3385  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3386  */
3387 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3388 {
3389         struct ethhdr *data = (struct ethhdr *)skb->data;
3390         struct iphdr *iph = ip_hdr(skb);
3391
3392         if (skb->protocol == htons(ETH_P_IP)) {
3393                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3394                         (data->h_dest[5] ^ data->h_source[5])) % count;
3395         }
3396
3397         return (data->h_dest[5] ^ data->h_source[5]) % count;
3398 }
3399
3400 /*
3401  * Hash for the output device based upon layer 3 and layer 4 data. If
3402  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3403  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3404  */
3405 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3406 {
3407         struct ethhdr *data = (struct ethhdr *)skb->data;
3408         struct iphdr *iph = ip_hdr(skb);
3409         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3410         int layer4_xor = 0;
3411
3412         if (skb->protocol == htons(ETH_P_IP)) {
3413                 if (!ip_is_fragment(iph) &&
3414                     (iph->protocol == IPPROTO_TCP ||
3415                      iph->protocol == IPPROTO_UDP)) {
3416                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3417                 }
3418                 return (layer4_xor ^
3419                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3420
3421         }
3422
3423         return (data->h_dest[5] ^ data->h_source[5]) % count;
3424 }
3425
3426 /*
3427  * Hash for the output device based upon layer 2 data
3428  */
3429 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3430 {
3431         struct ethhdr *data = (struct ethhdr *)skb->data;
3432
3433         return (data->h_dest[5] ^ data->h_source[5]) % count;
3434 }
3435
3436 /*-------------------------- Device entry points ----------------------------*/
3437
3438 static int bond_open(struct net_device *bond_dev)
3439 {
3440         struct bonding *bond = netdev_priv(bond_dev);
3441         struct slave *slave;
3442         int i;
3443
3444         /* reset slave->backup and slave->inactive */
3445         read_lock(&bond->lock);
3446         if (bond->slave_cnt > 0) {
3447                 read_lock(&bond->curr_slave_lock);
3448                 bond_for_each_slave(bond, slave, i) {
3449                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3450                                 && (slave != bond->curr_active_slave)) {
3451                                 bond_set_slave_inactive_flags(slave);
3452                         } else {
3453                                 bond_set_slave_active_flags(slave);
3454                         }
3455                 }
3456                 read_unlock(&bond->curr_slave_lock);
3457         }
3458         read_unlock(&bond->lock);
3459
3460         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3461
3462         if (bond_is_lb(bond)) {
3463                 /* bond_alb_initialize must be called before the timer
3464                  * is started.
3465                  */
3466                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3467                         /* something went wrong - fail the open operation */
3468                         return -ENOMEM;
3469                 }
3470
3471                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3472                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3473         }
3474
3475         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3476                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3477                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3478         }
3479
3480         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3481                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3482                         INIT_DELAYED_WORK(&bond->arp_work,
3483                                           bond_activebackup_arp_mon);
3484                 else
3485                         INIT_DELAYED_WORK(&bond->arp_work,
3486                                           bond_loadbalance_arp_mon);
3487
3488                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3489                 if (bond->params.arp_validate)
3490                         bond->recv_probe = bond_arp_rcv;
3491         }
3492
3493         if (bond->params.mode == BOND_MODE_8023AD) {
3494                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3495                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3496                 /* register to receive LACPDUs */
3497                 bond->recv_probe = bond_3ad_lacpdu_recv;
3498                 bond_3ad_initiate_agg_selection(bond, 1);
3499         }
3500
3501         return 0;
3502 }
3503
3504 static int bond_close(struct net_device *bond_dev)
3505 {
3506         struct bonding *bond = netdev_priv(bond_dev);
3507
3508         write_lock_bh(&bond->lock);
3509
3510         bond->send_peer_notif = 0;
3511
3512         write_unlock_bh(&bond->lock);
3513
3514         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3515                 cancel_delayed_work_sync(&bond->mii_work);
3516         }
3517
3518         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3519                 cancel_delayed_work_sync(&bond->arp_work);
3520         }
3521
3522         switch (bond->params.mode) {
3523         case BOND_MODE_8023AD:
3524                 cancel_delayed_work_sync(&bond->ad_work);
3525                 break;
3526         case BOND_MODE_TLB:
3527         case BOND_MODE_ALB:
3528                 cancel_delayed_work_sync(&bond->alb_work);
3529                 break;
3530         default:
3531                 break;
3532         }
3533
3534         if (delayed_work_pending(&bond->mcast_work))
3535                 cancel_delayed_work_sync(&bond->mcast_work);
3536
3537         if (bond_is_lb(bond)) {
3538                 /* Must be called only after all
3539                  * slaves have been released
3540                  */
3541                 bond_alb_deinitialize(bond);
3542         }
3543         bond->recv_probe = NULL;
3544
3545         return 0;
3546 }
3547
3548 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3549                                                 struct rtnl_link_stats64 *stats)
3550 {
3551         struct bonding *bond = netdev_priv(bond_dev);
3552         struct rtnl_link_stats64 temp;
3553         struct slave *slave;
3554         int i;
3555
3556         memset(stats, 0, sizeof(*stats));
3557
3558         read_lock_bh(&bond->lock);
3559
3560         bond_for_each_slave(bond, slave, i) {
3561                 const struct rtnl_link_stats64 *sstats =
3562                         dev_get_stats(slave->dev, &temp);
3563
3564                 stats->rx_packets += sstats->rx_packets;
3565                 stats->rx_bytes += sstats->rx_bytes;
3566                 stats->rx_errors += sstats->rx_errors;
3567                 stats->rx_dropped += sstats->rx_dropped;
3568
3569                 stats->tx_packets += sstats->tx_packets;
3570                 stats->tx_bytes += sstats->tx_bytes;
3571                 stats->tx_errors += sstats->tx_errors;
3572                 stats->tx_dropped += sstats->tx_dropped;
3573
3574                 stats->multicast += sstats->multicast;
3575                 stats->collisions += sstats->collisions;
3576
3577                 stats->rx_length_errors += sstats->rx_length_errors;
3578                 stats->rx_over_errors += sstats->rx_over_errors;
3579                 stats->rx_crc_errors += sstats->rx_crc_errors;
3580                 stats->rx_frame_errors += sstats->rx_frame_errors;
3581                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3582                 stats->rx_missed_errors += sstats->rx_missed_errors;
3583
3584                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3585                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3586                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3587                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3588                 stats->tx_window_errors += sstats->tx_window_errors;
3589         }
3590
3591         read_unlock_bh(&bond->lock);
3592
3593         return stats;
3594 }
3595
3596 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3597 {
3598         struct net_device *slave_dev = NULL;
3599         struct ifbond k_binfo;
3600         struct ifbond __user *u_binfo = NULL;
3601         struct ifslave k_sinfo;
3602         struct ifslave __user *u_sinfo = NULL;
3603         struct mii_ioctl_data *mii = NULL;
3604         int res = 0;
3605
3606         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3607
3608         switch (cmd) {
3609         case SIOCGMIIPHY:
3610                 mii = if_mii(ifr);
3611                 if (!mii)
3612                         return -EINVAL;
3613
3614                 mii->phy_id = 0;
3615                 /* Fall Through */
3616         case SIOCGMIIREG:
3617                 /*
3618                  * We do this again just in case we were called by SIOCGMIIREG
3619                  * instead of SIOCGMIIPHY.
3620                  */
3621                 mii = if_mii(ifr);
3622                 if (!mii)
3623                         return -EINVAL;
3624
3625
3626                 if (mii->reg_num == 1) {
3627                         struct bonding *bond = netdev_priv(bond_dev);
3628                         mii->val_out = 0;
3629                         read_lock(&bond->lock);
3630                         read_lock(&bond->curr_slave_lock);
3631                         if (netif_carrier_ok(bond->dev))
3632                                 mii->val_out = BMSR_LSTATUS;
3633
3634                         read_unlock(&bond->curr_slave_lock);
3635                         read_unlock(&bond->lock);
3636                 }
3637
3638                 return 0;
3639         case BOND_INFO_QUERY_OLD:
3640         case SIOCBONDINFOQUERY:
3641                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3642
3643                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3644                         return -EFAULT;
3645
3646                 res = bond_info_query(bond_dev, &k_binfo);
3647                 if (res == 0 &&
3648                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3649                         return -EFAULT;
3650
3651                 return res;
3652         case BOND_SLAVE_INFO_QUERY_OLD:
3653         case SIOCBONDSLAVEINFOQUERY:
3654                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3655
3656                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3657                         return -EFAULT;
3658
3659                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3660                 if (res == 0 &&
3661                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3662                         return -EFAULT;
3663
3664                 return res;
3665         default:
3666                 /* Go on */
3667                 break;
3668         }
3669
3670         if (!capable(CAP_NET_ADMIN))
3671                 return -EPERM;
3672
3673         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3674
3675         pr_debug("slave_dev=%p:\n", slave_dev);
3676
3677         if (!slave_dev)
3678                 res = -ENODEV;
3679         else {
3680                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3681                 switch (cmd) {
3682                 case BOND_ENSLAVE_OLD:
3683                 case SIOCBONDENSLAVE:
3684                         res = bond_enslave(bond_dev, slave_dev);
3685                         break;
3686                 case BOND_RELEASE_OLD:
3687                 case SIOCBONDRELEASE:
3688                         res = bond_release(bond_dev, slave_dev);
3689                         break;
3690                 case BOND_SETHWADDR_OLD:
3691                 case SIOCBONDSETHWADDR:
3692                         res = bond_sethwaddr(bond_dev, slave_dev);
3693                         break;
3694                 case BOND_CHANGE_ACTIVE_OLD:
3695                 case SIOCBONDCHANGEACTIVE:
3696                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3697                         break;
3698                 default:
3699                         res = -EOPNOTSUPP;
3700                 }
3701
3702                 dev_put(slave_dev);
3703         }
3704
3705         return res;
3706 }
3707
3708 static bool bond_addr_in_mc_list(unsigned char *addr,
3709                                  struct netdev_hw_addr_list *list,
3710                                  int addrlen)
3711 {
3712         struct netdev_hw_addr *ha;
3713
3714         netdev_hw_addr_list_for_each(ha, list)
3715                 if (!memcmp(ha->addr, addr, addrlen))
3716                         return true;
3717
3718         return false;
3719 }
3720
3721 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3722 {
3723         struct bonding *bond = netdev_priv(bond_dev);
3724
3725         if (change & IFF_PROMISC)
3726                 bond_set_promiscuity(bond,
3727                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3728
3729         if (change & IFF_ALLMULTI)
3730                 bond_set_allmulti(bond,
3731                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3732 }
3733
3734 static void bond_set_multicast_list(struct net_device *bond_dev)
3735 {
3736         struct bonding *bond = netdev_priv(bond_dev);
3737         struct netdev_hw_addr *ha;
3738         bool found;
3739
3740         read_lock(&bond->lock);
3741
3742         /* looking for addresses to add to slaves' mc list */
3743         netdev_for_each_mc_addr(ha, bond_dev) {
3744                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3745                                              bond_dev->addr_len);
3746                 if (!found)
3747                         bond_mc_add(bond, ha->addr);
3748         }
3749
3750         /* looking for addresses to delete from slaves' list */
3751         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3752                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3753                                              bond_dev->addr_len);
3754                 if (!found)
3755                         bond_mc_del(bond, ha->addr);
3756         }
3757
3758         /* save master's multicast list */
3759         __hw_addr_flush(&bond->mc_list);
3760         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3761                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3762
3763         read_unlock(&bond->lock);
3764 }
3765
3766 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3767 {
3768         struct bonding *bond = netdev_priv(dev);
3769         struct slave *slave = bond->first_slave;
3770
3771         if (slave) {
3772                 const struct net_device_ops *slave_ops
3773                         = slave->dev->netdev_ops;
3774                 if (slave_ops->ndo_neigh_setup)
3775                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3776         }
3777         return 0;
3778 }
3779
3780 /*
3781  * Change the MTU of all of a master's slaves to match the master
3782  */
3783 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3784 {
3785         struct bonding *bond = netdev_priv(bond_dev);
3786         struct slave *slave, *stop_at;
3787         int res = 0;
3788         int i;
3789
3790         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3791                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3792
3793         /* Can't hold bond->lock with bh disabled here since
3794          * some base drivers panic. On the other hand we can't
3795          * hold bond->lock without bh disabled because we'll
3796          * deadlock. The only solution is to rely on the fact
3797          * that we're under rtnl_lock here, and the slaves
3798          * list won't change. This doesn't solve the problem
3799          * of setting the slave's MTU while it is
3800          * transmitting, but the assumption is that the base
3801          * driver can handle that.
3802          *
3803          * TODO: figure out a way to safely iterate the slaves
3804          * list, but without holding a lock around the actual
3805          * call to the base driver.
3806          */
3807
3808         bond_for_each_slave(bond, slave, i) {
3809                 pr_debug("s %p s->p %p c_m %p\n",
3810                          slave,
3811                          slave->prev,
3812                          slave->dev->netdev_ops->ndo_change_mtu);
3813
3814                 res = dev_set_mtu(slave->dev, new_mtu);
3815
3816                 if (res) {
3817                         /* If we failed to set the slave's mtu to the new value
3818                          * we must abort the operation even in ACTIVE_BACKUP
3819                          * mode, because if we allow the backup slaves to have
3820                          * different mtu values than the active slave we'll
3821                          * need to change their mtu when doing a failover. That
3822                          * means changing their mtu from timer context, which
3823                          * is probably not a good idea.
3824                          */
3825                         pr_debug("err %d %s\n", res, slave->dev->name);
3826                         goto unwind;
3827                 }
3828         }
3829
3830         bond_dev->mtu = new_mtu;
3831
3832         return 0;
3833
3834 unwind:
3835         /* unwind from head to the slave that failed */
3836         stop_at = slave;
3837         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3838                 int tmp_res;
3839
3840                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3841                 if (tmp_res) {
3842                         pr_debug("unwind err %d dev %s\n",
3843                                  tmp_res, slave->dev->name);
3844                 }
3845         }
3846
3847         return res;
3848 }
3849
3850 /*
3851  * Change HW address
3852  *
3853  * Note that many devices must be down to change the HW address, and
3854  * downing the master releases all slaves.  We can make bonds full of
3855  * bonding devices to test this, however.
3856  */
3857 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3858 {
3859         struct bonding *bond = netdev_priv(bond_dev);
3860         struct sockaddr *sa = addr, tmp_sa;
3861         struct slave *slave, *stop_at;
3862         int res = 0;
3863         int i;
3864
3865         if (bond->params.mode == BOND_MODE_ALB)
3866                 return bond_alb_set_mac_address(bond_dev, addr);
3867
3868
3869         pr_debug("bond=%p, name=%s\n",
3870                  bond, bond_dev ? bond_dev->name : "None");
3871
3872         /*
3873          * If fail_over_mac is set to active, do nothing and return
3874          * success.  Returning an error causes ifenslave to fail.
3875          */
3876         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3877                 return 0;
3878
3879         if (!is_valid_ether_addr(sa->sa_data))
3880                 return -EADDRNOTAVAIL;
3881
3882         /* Can't hold bond->lock with bh disabled here since
3883          * some base drivers panic. On the other hand we can't
3884          * hold bond->lock without bh disabled because we'll
3885          * deadlock. The only solution is to rely on the fact
3886          * that we're under rtnl_lock here, and the slaves
3887          * list won't change. This doesn't solve the problem
3888          * of setting the slave's hw address while it is
3889          * transmitting, but the assumption is that the base
3890          * driver can handle that.
3891          *
3892          * TODO: figure out a way to safely iterate the slaves
3893          * list, but without holding a lock around the actual
3894          * call to the base driver.
3895          */
3896
3897         bond_for_each_slave(bond, slave, i) {
3898                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3899                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3900
3901                 if (slave_ops->ndo_set_mac_address == NULL) {
3902                         res = -EOPNOTSUPP;
3903                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3904                         goto unwind;
3905                 }
3906
3907                 res = dev_set_mac_address(slave->dev, addr);
3908                 if (res) {
3909                         /* TODO: consider downing the slave
3910                          * and retry ?
3911                          * User should expect communications
3912                          * breakage anyway until ARP finish
3913                          * updating, so...
3914                          */
3915                         pr_debug("err %d %s\n", res, slave->dev->name);
3916                         goto unwind;
3917                 }
3918         }
3919
3920         /* success */
3921         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3922         return 0;
3923
3924 unwind:
3925         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3926         tmp_sa.sa_family = bond_dev->type;
3927
3928         /* unwind from head to the slave that failed */
3929         stop_at = slave;
3930         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3931                 int tmp_res;
3932
3933                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3934                 if (tmp_res) {
3935                         pr_debug("unwind err %d dev %s\n",
3936                                  tmp_res, slave->dev->name);
3937                 }
3938         }
3939
3940         return res;
3941 }
3942
3943 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3944 {
3945         struct bonding *bond = netdev_priv(bond_dev);
3946         struct slave *slave, *start_at;
3947         int i, slave_no, res = 1;
3948         struct iphdr *iph = ip_hdr(skb);
3949
3950         /*
3951          * Start with the curr_active_slave that joined the bond as the
3952          * default for sending IGMP traffic.  For failover purposes one
3953          * needs to maintain some consistency for the interface that will
3954          * send the join/membership reports.  The curr_active_slave found
3955          * will send all of this type of traffic.
3956          */
3957         if ((iph->protocol == IPPROTO_IGMP) &&
3958             (skb->protocol == htons(ETH_P_IP))) {
3959
3960                 read_lock(&bond->curr_slave_lock);
3961                 slave = bond->curr_active_slave;
3962                 read_unlock(&bond->curr_slave_lock);
3963
3964                 if (!slave)
3965                         goto out;
3966         } else {
3967                 /*
3968                  * Concurrent TX may collide on rr_tx_counter; we accept
3969                  * that as being rare enough not to justify using an
3970                  * atomic op here.
3971                  */
3972                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3973
3974                 bond_for_each_slave(bond, slave, i) {
3975                         slave_no--;
3976                         if (slave_no < 0)
3977                                 break;
3978                 }
3979         }
3980
3981         start_at = slave;
3982         bond_for_each_slave_from(bond, slave, i, start_at) {
3983                 if (IS_UP(slave->dev) &&
3984                     (slave->link == BOND_LINK_UP) &&
3985                     bond_is_active_slave(slave)) {
3986                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3987                         break;
3988                 }
3989         }
3990
3991 out:
3992         if (res) {
3993                 /* no suitable interface, frame not sent */
3994                 dev_kfree_skb(skb);
3995         }
3996
3997         return NETDEV_TX_OK;
3998 }
3999
4000
4001 /*
4002  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4003  * the bond has a usable interface.
4004  */
4005 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4006 {
4007         struct bonding *bond = netdev_priv(bond_dev);
4008         int res = 1;
4009
4010         read_lock(&bond->curr_slave_lock);
4011
4012         if (bond->curr_active_slave)
4013                 res = bond_dev_queue_xmit(bond, skb,
4014                         bond->curr_active_slave->dev);
4015
4016         if (res)
4017                 /* no suitable interface, frame not sent */
4018                 dev_kfree_skb(skb);
4019
4020         read_unlock(&bond->curr_slave_lock);
4021
4022         return NETDEV_TX_OK;
4023 }
4024
4025 /*
4026  * In bond_xmit_xor() , we determine the output device by using a pre-
4027  * determined xmit_hash_policy(), If the selected device is not enabled,
4028  * find the next active slave.
4029  */
4030 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4031 {
4032         struct bonding *bond = netdev_priv(bond_dev);
4033         struct slave *slave, *start_at;
4034         int slave_no;
4035         int i;
4036         int res = 1;
4037
4038         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4039
4040         bond_for_each_slave(bond, slave, i) {
4041                 slave_no--;
4042                 if (slave_no < 0)
4043                         break;
4044         }
4045
4046         start_at = slave;
4047
4048         bond_for_each_slave_from(bond, slave, i, start_at) {
4049                 if (IS_UP(slave->dev) &&
4050                     (slave->link == BOND_LINK_UP) &&
4051                     bond_is_active_slave(slave)) {
4052                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4053                         break;
4054                 }
4055         }
4056
4057         if (res) {
4058                 /* no suitable interface, frame not sent */
4059                 dev_kfree_skb(skb);
4060         }
4061
4062         return NETDEV_TX_OK;
4063 }
4064
4065 /*
4066  * in broadcast mode, we send everything to all usable interfaces.
4067  */
4068 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4069 {
4070         struct bonding *bond = netdev_priv(bond_dev);
4071         struct slave *slave, *start_at;
4072         struct net_device *tx_dev = NULL;
4073         int i;
4074         int res = 1;
4075
4076         read_lock(&bond->curr_slave_lock);
4077         start_at = bond->curr_active_slave;
4078         read_unlock(&bond->curr_slave_lock);
4079
4080         if (!start_at)
4081                 goto out;
4082
4083         bond_for_each_slave_from(bond, slave, i, start_at) {
4084                 if (IS_UP(slave->dev) &&
4085                     (slave->link == BOND_LINK_UP) &&
4086                     bond_is_active_slave(slave)) {
4087                         if (tx_dev) {
4088                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4089                                 if (!skb2) {
4090                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4091                                                bond_dev->name);
4092                                         continue;
4093                                 }
4094
4095                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4096                                 if (res) {
4097                                         dev_kfree_skb(skb2);
4098                                         continue;
4099                                 }
4100                         }
4101                         tx_dev = slave->dev;
4102                 }
4103         }
4104
4105         if (tx_dev)
4106                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4107
4108 out:
4109         if (res)
4110                 /* no suitable interface, frame not sent */
4111                 dev_kfree_skb(skb);
4112
4113         /* frame sent to all suitable interfaces */
4114         return NETDEV_TX_OK;
4115 }
4116
4117 /*------------------------- Device initialization ---------------------------*/
4118
4119 static void bond_set_xmit_hash_policy(struct bonding *bond)
4120 {
4121         switch (bond->params.xmit_policy) {
4122         case BOND_XMIT_POLICY_LAYER23:
4123                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4124                 break;
4125         case BOND_XMIT_POLICY_LAYER34:
4126                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4127                 break;
4128         case BOND_XMIT_POLICY_LAYER2:
4129         default:
4130                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4131                 break;
4132         }
4133 }
4134
4135 /*
4136  * Lookup the slave that corresponds to a qid
4137  */
4138 static inline int bond_slave_override(struct bonding *bond,
4139                                       struct sk_buff *skb)
4140 {
4141         int i, res = 1;
4142         struct slave *slave = NULL;
4143         struct slave *check_slave;
4144
4145         if (!skb->queue_mapping)
4146                 return 1;
4147
4148         /* Find out if any slaves have the same mapping as this skb. */
4149         bond_for_each_slave(bond, check_slave, i) {
4150                 if (check_slave->queue_id == skb->queue_mapping) {
4151                         slave = check_slave;
4152                         break;
4153                 }
4154         }
4155
4156         /* If the slave isn't UP, use default transmit policy. */
4157         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4158             (slave->link == BOND_LINK_UP)) {
4159                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4160         }
4161
4162         return res;
4163 }
4164
4165
4166 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4167 {
4168         /*
4169          * This helper function exists to help dev_pick_tx get the correct
4170          * destination queue.  Using a helper function skips a call to
4171          * skb_tx_hash and will put the skbs in the queue we expect on their
4172          * way down to the bonding driver.
4173          */
4174         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4175
4176         /*
4177          * Save the original txq to restore before passing to the driver
4178          */
4179         bond_queue_mapping(skb) = skb->queue_mapping;
4180
4181         if (unlikely(txq >= dev->real_num_tx_queues)) {
4182                 do {
4183                         txq -= dev->real_num_tx_queues;
4184                 } while (txq >= dev->real_num_tx_queues);
4185         }
4186         return txq;
4187 }
4188
4189 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4190 {
4191         struct bonding *bond = netdev_priv(dev);
4192
4193         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4194                 if (!bond_slave_override(bond, skb))
4195                         return NETDEV_TX_OK;
4196         }
4197
4198         switch (bond->params.mode) {
4199         case BOND_MODE_ROUNDROBIN:
4200                 return bond_xmit_roundrobin(skb, dev);
4201         case BOND_MODE_ACTIVEBACKUP:
4202                 return bond_xmit_activebackup(skb, dev);
4203         case BOND_MODE_XOR:
4204                 return bond_xmit_xor(skb, dev);
4205         case BOND_MODE_BROADCAST:
4206                 return bond_xmit_broadcast(skb, dev);
4207         case BOND_MODE_8023AD:
4208                 return bond_3ad_xmit_xor(skb, dev);
4209         case BOND_MODE_ALB:
4210         case BOND_MODE_TLB:
4211                 return bond_alb_xmit(skb, dev);
4212         default:
4213                 /* Should never happen, mode already checked */
4214                 pr_err("%s: Error: Unknown bonding mode %d\n",
4215                        dev->name, bond->params.mode);
4216                 WARN_ON_ONCE(1);
4217                 dev_kfree_skb(skb);
4218                 return NETDEV_TX_OK;
4219         }
4220 }
4221
4222 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4223 {
4224         struct bonding *bond = netdev_priv(dev);
4225         netdev_tx_t ret = NETDEV_TX_OK;
4226
4227         /*
4228          * If we risk deadlock from transmitting this in the
4229          * netpoll path, tell netpoll to queue the frame for later tx
4230          */
4231         if (is_netpoll_tx_blocked(dev))
4232                 return NETDEV_TX_BUSY;
4233
4234         read_lock(&bond->lock);
4235
4236         if (bond->slave_cnt)
4237                 ret = __bond_start_xmit(skb, dev);
4238         else
4239                 dev_kfree_skb(skb);
4240
4241         read_unlock(&bond->lock);
4242
4243         return ret;
4244 }
4245
4246 /*
4247  * set bond mode specific net device operations
4248  */
4249 void bond_set_mode_ops(struct bonding *bond, int mode)
4250 {
4251         struct net_device *bond_dev = bond->dev;
4252
4253         switch (mode) {
4254         case BOND_MODE_ROUNDROBIN:
4255                 break;
4256         case BOND_MODE_ACTIVEBACKUP:
4257                 break;
4258         case BOND_MODE_XOR:
4259                 bond_set_xmit_hash_policy(bond);
4260                 break;
4261         case BOND_MODE_BROADCAST:
4262                 break;
4263         case BOND_MODE_8023AD:
4264                 bond_set_xmit_hash_policy(bond);
4265                 break;
4266         case BOND_MODE_ALB:
4267                 /* FALLTHRU */
4268         case BOND_MODE_TLB:
4269                 break;
4270         default:
4271                 /* Should never happen, mode already checked */
4272                 pr_err("%s: Error: Unknown bonding mode %d\n",
4273                        bond_dev->name, mode);
4274                 break;
4275         }
4276 }
4277
4278 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4279                                     struct ethtool_drvinfo *drvinfo)
4280 {
4281         strncpy(drvinfo->driver, DRV_NAME, 32);
4282         strncpy(drvinfo->version, DRV_VERSION, 32);
4283         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4284 }
4285
4286 static const struct ethtool_ops bond_ethtool_ops = {
4287         .get_drvinfo            = bond_ethtool_get_drvinfo,
4288         .get_link               = ethtool_op_get_link,
4289 };
4290
4291 static const struct net_device_ops bond_netdev_ops = {
4292         .ndo_init               = bond_init,
4293         .ndo_uninit             = bond_uninit,
4294         .ndo_open               = bond_open,
4295         .ndo_stop               = bond_close,
4296         .ndo_start_xmit         = bond_start_xmit,
4297         .ndo_select_queue       = bond_select_queue,
4298         .ndo_get_stats64        = bond_get_stats,
4299         .ndo_do_ioctl           = bond_do_ioctl,
4300         .ndo_change_rx_flags    = bond_change_rx_flags,
4301         .ndo_set_rx_mode        = bond_set_multicast_list,
4302         .ndo_change_mtu         = bond_change_mtu,
4303         .ndo_set_mac_address    = bond_set_mac_address,
4304         .ndo_neigh_setup        = bond_neigh_setup,
4305         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4306         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4307 #ifdef CONFIG_NET_POLL_CONTROLLER
4308         .ndo_netpoll_setup      = bond_netpoll_setup,
4309         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4310         .ndo_poll_controller    = bond_poll_controller,
4311 #endif
4312         .ndo_add_slave          = bond_enslave,
4313         .ndo_del_slave          = bond_release,
4314         .ndo_fix_features       = bond_fix_features,
4315 };
4316
4317 static void bond_destructor(struct net_device *bond_dev)
4318 {
4319         struct bonding *bond = netdev_priv(bond_dev);
4320         if (bond->wq)
4321                 destroy_workqueue(bond->wq);
4322         free_netdev(bond_dev);
4323 }
4324
4325 static void bond_setup(struct net_device *bond_dev)
4326 {
4327         struct bonding *bond = netdev_priv(bond_dev);
4328
4329         /* initialize rwlocks */
4330         rwlock_init(&bond->lock);
4331         rwlock_init(&bond->curr_slave_lock);
4332
4333         bond->params = bonding_defaults;
4334
4335         /* Initialize pointers */
4336         bond->dev = bond_dev;
4337         INIT_LIST_HEAD(&bond->vlan_list);
4338
4339         /* Initialize the device entry points */
4340         ether_setup(bond_dev);
4341         bond_dev->netdev_ops = &bond_netdev_ops;
4342         bond_dev->ethtool_ops = &bond_ethtool_ops;
4343         bond_set_mode_ops(bond, bond->params.mode);
4344
4345         bond_dev->destructor = bond_destructor;
4346
4347         /* Initialize the device options */
4348         bond_dev->tx_queue_len = 0;
4349         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4350         bond_dev->priv_flags |= IFF_BONDING;
4351         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4352
4353         /* At first, we block adding VLANs. That's the only way to
4354          * prevent problems that occur when adding VLANs over an
4355          * empty bond. The block will be removed once non-challenged
4356          * slaves are enslaved.
4357          */
4358         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4359
4360         /* don't acquire bond device's netif_tx_lock when
4361          * transmitting */
4362         bond_dev->features |= NETIF_F_LLTX;
4363
4364         /* By default, we declare the bond to be fully
4365          * VLAN hardware accelerated capable. Special
4366          * care is taken in the various xmit functions
4367          * when there are slaves that are not hw accel
4368          * capable
4369          */
4370
4371         bond_dev->hw_features = BOND_VLAN_FEATURES |
4372                                 NETIF_F_HW_VLAN_TX |
4373                                 NETIF_F_HW_VLAN_RX |
4374                                 NETIF_F_HW_VLAN_FILTER;
4375
4376         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4377         bond_dev->features |= bond_dev->hw_features;
4378 }
4379
4380 static void bond_work_cancel_all(struct bonding *bond)
4381 {
4382         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4383                 cancel_delayed_work_sync(&bond->mii_work);
4384
4385         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4386                 cancel_delayed_work_sync(&bond->arp_work);
4387
4388         if (bond->params.mode == BOND_MODE_ALB &&
4389             delayed_work_pending(&bond->alb_work))
4390                 cancel_delayed_work_sync(&bond->alb_work);
4391
4392         if (bond->params.mode == BOND_MODE_8023AD &&
4393             delayed_work_pending(&bond->ad_work))
4394                 cancel_delayed_work_sync(&bond->ad_work);
4395
4396         if (delayed_work_pending(&bond->mcast_work))
4397                 cancel_delayed_work_sync(&bond->mcast_work);
4398 }
4399
4400 /*
4401 * Destroy a bonding device.
4402 * Must be under rtnl_lock when this function is called.
4403 */
4404 static void bond_uninit(struct net_device *bond_dev)
4405 {
4406         struct bonding *bond = netdev_priv(bond_dev);
4407         struct vlan_entry *vlan, *tmp;
4408
4409         bond_netpoll_cleanup(bond_dev);
4410
4411         /* Release the bonded slaves */
4412         bond_release_all(bond_dev);
4413
4414         list_del(&bond->bond_list);
4415
4416         bond_work_cancel_all(bond);
4417
4418         bond_remove_proc_entry(bond);
4419
4420         bond_debug_unregister(bond);
4421
4422         __hw_addr_flush(&bond->mc_list);
4423
4424         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4425                 list_del(&vlan->vlan_list);
4426                 kfree(vlan);
4427         }
4428 }
4429
4430 /*------------------------- Module initialization ---------------------------*/
4431
4432 /*
4433  * Convert string input module parms.  Accept either the
4434  * number of the mode or its string name.  A bit complicated because
4435  * some mode names are substrings of other names, and calls from sysfs
4436  * may have whitespace in the name (trailing newlines, for example).
4437  */
4438 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4439 {
4440         int modeint = -1, i, rv;
4441         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4442
4443         for (p = (char *)buf; *p; p++)
4444                 if (!(isdigit(*p) || isspace(*p)))
4445                         break;
4446
4447         if (*p)
4448                 rv = sscanf(buf, "%20s", modestr);
4449         else
4450                 rv = sscanf(buf, "%d", &modeint);
4451
4452         if (!rv)
4453                 return -1;
4454
4455         for (i = 0; tbl[i].modename; i++) {
4456                 if (modeint == tbl[i].mode)
4457                         return tbl[i].mode;
4458                 if (strcmp(modestr, tbl[i].modename) == 0)
4459                         return tbl[i].mode;
4460         }
4461
4462         return -1;
4463 }
4464
4465 static int bond_check_params(struct bond_params *params)
4466 {
4467         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4468
4469         /*
4470          * Convert string parameters.
4471          */
4472         if (mode) {
4473                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4474                 if (bond_mode == -1) {
4475                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4476                                mode == NULL ? "NULL" : mode);
4477                         return -EINVAL;
4478                 }
4479         }
4480
4481         if (xmit_hash_policy) {
4482                 if ((bond_mode != BOND_MODE_XOR) &&
4483                     (bond_mode != BOND_MODE_8023AD)) {
4484                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4485                                bond_mode_name(bond_mode));
4486                 } else {
4487                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4488                                                         xmit_hashtype_tbl);
4489                         if (xmit_hashtype == -1) {
4490                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4491                                        xmit_hash_policy == NULL ? "NULL" :
4492                                        xmit_hash_policy);
4493                                 return -EINVAL;
4494                         }
4495                 }
4496         }
4497
4498         if (lacp_rate) {
4499                 if (bond_mode != BOND_MODE_8023AD) {
4500                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4501                                 bond_mode_name(bond_mode));
4502                 } else {
4503                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4504                         if (lacp_fast == -1) {
4505                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4506                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4507                                 return -EINVAL;
4508                         }
4509                 }
4510         }
4511
4512         if (ad_select) {
4513                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4514                 if (params->ad_select == -1) {
4515                         pr_err("Error: Invalid ad_select \"%s\"\n",
4516                                ad_select == NULL ? "NULL" : ad_select);
4517                         return -EINVAL;
4518                 }
4519
4520                 if (bond_mode != BOND_MODE_8023AD) {
4521                         pr_warning("ad_select param only affects 802.3ad mode\n");
4522                 }
4523         } else {
4524                 params->ad_select = BOND_AD_STABLE;
4525         }
4526
4527         if (max_bonds < 0) {
4528                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4529                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4530                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4531         }
4532
4533         if (miimon < 0) {
4534                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4535                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4536                 miimon = BOND_LINK_MON_INTERV;
4537         }
4538
4539         if (updelay < 0) {
4540                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4541                            updelay, INT_MAX);
4542                 updelay = 0;
4543         }
4544
4545         if (downdelay < 0) {
4546                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547                            downdelay, INT_MAX);
4548                 downdelay = 0;
4549         }
4550
4551         if ((use_carrier != 0) && (use_carrier != 1)) {
4552                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4553                            use_carrier);
4554                 use_carrier = 1;
4555         }
4556
4557         if (num_peer_notif < 0 || num_peer_notif > 255) {
4558                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4559                            num_peer_notif);
4560                 num_peer_notif = 1;
4561         }
4562
4563         /* reset values for 802.3ad */
4564         if (bond_mode == BOND_MODE_8023AD) {
4565                 if (!miimon) {
4566                         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");
4567                         pr_warning("Forcing miimon to 100msec\n");
4568                         miimon = 100;
4569                 }
4570         }
4571
4572         if (tx_queues < 1 || tx_queues > 255) {
4573                 pr_warning("Warning: tx_queues (%d) should be between "
4574                            "1 and 255, resetting to %d\n",
4575                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4576                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4577         }
4578
4579         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4580                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4581                            "not of valid value (0/1), so it was set to "
4582                            "0\n", all_slaves_active);
4583                 all_slaves_active = 0;
4584         }
4585
4586         if (resend_igmp < 0 || resend_igmp > 255) {
4587                 pr_warning("Warning: resend_igmp (%d) should be between "
4588                            "0 and 255, resetting to %d\n",
4589                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4590                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4591         }
4592
4593         /* reset values for TLB/ALB */
4594         if ((bond_mode == BOND_MODE_TLB) ||
4595             (bond_mode == BOND_MODE_ALB)) {
4596                 if (!miimon) {
4597                         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");
4598                         pr_warning("Forcing miimon to 100msec\n");
4599                         miimon = 100;
4600                 }
4601         }
4602
4603         if (bond_mode == BOND_MODE_ALB) {
4604                 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",
4605                           updelay);
4606         }
4607
4608         if (!miimon) {
4609                 if (updelay || downdelay) {
4610                         /* just warn the user the up/down delay will have
4611                          * no effect since miimon is zero...
4612                          */
4613                         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",
4614                                    updelay, downdelay);
4615                 }
4616         } else {
4617                 /* don't allow arp monitoring */
4618                 if (arp_interval) {
4619                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4620                                    miimon, arp_interval);
4621                         arp_interval = 0;
4622                 }
4623
4624                 if ((updelay % miimon) != 0) {
4625                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4626                                    updelay, miimon,
4627                                    (updelay / miimon) * miimon);
4628                 }
4629
4630                 updelay /= miimon;
4631
4632                 if ((downdelay % miimon) != 0) {
4633                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4634                                    downdelay, miimon,
4635                                    (downdelay / miimon) * miimon);
4636                 }
4637
4638                 downdelay /= miimon;
4639         }
4640
4641         if (arp_interval < 0) {
4642                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4643                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4644                 arp_interval = BOND_LINK_ARP_INTERV;
4645         }
4646
4647         for (arp_ip_count = 0;
4648              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4649              arp_ip_count++) {
4650                 /* not complete check, but should be good enough to
4651                    catch mistakes */
4652                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4653                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4654                                    arp_ip_target[arp_ip_count]);
4655                         arp_interval = 0;
4656                 } else {
4657                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4658                         arp_target[arp_ip_count] = ip;
4659                 }
4660         }
4661
4662         if (arp_interval && !arp_ip_count) {
4663                 /* don't allow arping if no arp_ip_target given... */
4664                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4665                            arp_interval);
4666                 arp_interval = 0;
4667         }
4668
4669         if (arp_validate) {
4670                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4671                         pr_err("arp_validate only supported in active-backup mode\n");
4672                         return -EINVAL;
4673                 }
4674                 if (!arp_interval) {
4675                         pr_err("arp_validate requires arp_interval\n");
4676                         return -EINVAL;
4677                 }
4678
4679                 arp_validate_value = bond_parse_parm(arp_validate,
4680                                                      arp_validate_tbl);
4681                 if (arp_validate_value == -1) {
4682                         pr_err("Error: invalid arp_validate \"%s\"\n",
4683                                arp_validate == NULL ? "NULL" : arp_validate);
4684                         return -EINVAL;
4685                 }
4686         } else
4687                 arp_validate_value = 0;
4688
4689         if (miimon) {
4690                 pr_info("MII link monitoring set to %d ms\n", miimon);
4691         } else if (arp_interval) {
4692                 int i;
4693
4694                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4695                         arp_interval,
4696                         arp_validate_tbl[arp_validate_value].modename,
4697                         arp_ip_count);
4698
4699                 for (i = 0; i < arp_ip_count; i++)
4700                         pr_info(" %s", arp_ip_target[i]);
4701
4702                 pr_info("\n");
4703
4704         } else if (max_bonds) {
4705                 /* miimon and arp_interval not set, we need one so things
4706                  * work as expected, see bonding.txt for details
4707                  */
4708                 pr_debug("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");
4709         }
4710
4711         if (primary && !USES_PRIMARY(bond_mode)) {
4712                 /* currently, using a primary only makes sense
4713                  * in active backup, TLB or ALB modes
4714                  */
4715                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4716                            primary, bond_mode_name(bond_mode));
4717                 primary = NULL;
4718         }
4719
4720         if (primary && primary_reselect) {
4721                 primary_reselect_value = bond_parse_parm(primary_reselect,
4722                                                          pri_reselect_tbl);
4723                 if (primary_reselect_value == -1) {
4724                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4725                                primary_reselect ==
4726                                         NULL ? "NULL" : primary_reselect);
4727                         return -EINVAL;
4728                 }
4729         } else {
4730                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4731         }
4732
4733         if (fail_over_mac) {
4734                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4735                                                       fail_over_mac_tbl);
4736                 if (fail_over_mac_value == -1) {
4737                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4738                                arp_validate == NULL ? "NULL" : arp_validate);
4739                         return -EINVAL;
4740                 }
4741
4742                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4743                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4744         } else {
4745                 fail_over_mac_value = BOND_FOM_NONE;
4746         }
4747
4748         /* fill params struct with the proper values */
4749         params->mode = bond_mode;
4750         params->xmit_policy = xmit_hashtype;
4751         params->miimon = miimon;
4752         params->num_peer_notif = num_peer_notif;
4753         params->arp_interval = arp_interval;
4754         params->arp_validate = arp_validate_value;
4755         params->updelay = updelay;
4756         params->downdelay = downdelay;
4757         params->use_carrier = use_carrier;
4758         params->lacp_fast = lacp_fast;
4759         params->primary[0] = 0;
4760         params->primary_reselect = primary_reselect_value;
4761         params->fail_over_mac = fail_over_mac_value;
4762         params->tx_queues = tx_queues;
4763         params->all_slaves_active = all_slaves_active;
4764         params->resend_igmp = resend_igmp;
4765         params->min_links = min_links;
4766
4767         if (primary) {
4768                 strncpy(params->primary, primary, IFNAMSIZ);
4769                 params->primary[IFNAMSIZ - 1] = 0;
4770         }
4771
4772         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4773
4774         return 0;
4775 }
4776
4777 static struct lock_class_key bonding_netdev_xmit_lock_key;
4778 static struct lock_class_key bonding_netdev_addr_lock_key;
4779
4780 static void bond_set_lockdep_class_one(struct net_device *dev,
4781                                        struct netdev_queue *txq,
4782                                        void *_unused)
4783 {
4784         lockdep_set_class(&txq->_xmit_lock,
4785                           &bonding_netdev_xmit_lock_key);
4786 }
4787
4788 static void bond_set_lockdep_class(struct net_device *dev)
4789 {
4790         lockdep_set_class(&dev->addr_list_lock,
4791                           &bonding_netdev_addr_lock_key);
4792         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4793 }
4794
4795 /*
4796  * Called from registration process
4797  */
4798 static int bond_init(struct net_device *bond_dev)
4799 {
4800         struct bonding *bond = netdev_priv(bond_dev);
4801         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4802         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4803
4804         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4805
4806         /*
4807          * Initialize locks that may be required during
4808          * en/deslave operations.  All of the bond_open work
4809          * (of which this is part) should really be moved to
4810          * a phase prior to dev_open
4811          */
4812         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4813         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4814
4815         bond->wq = create_singlethread_workqueue(bond_dev->name);
4816         if (!bond->wq)
4817                 return -ENOMEM;
4818
4819         bond_set_lockdep_class(bond_dev);
4820
4821         bond_create_proc_entry(bond);
4822         list_add_tail(&bond->bond_list, &bn->dev_list);
4823
4824         bond_prepare_sysfs_group(bond);
4825
4826         bond_debug_register(bond);
4827
4828         __hw_addr_init(&bond->mc_list);
4829         return 0;
4830 }
4831
4832 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4833 {
4834         if (tb[IFLA_ADDRESS]) {
4835                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4836                         return -EINVAL;
4837                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4838                         return -EADDRNOTAVAIL;
4839         }
4840         return 0;
4841 }
4842
4843 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4844                               unsigned int *num_queues,
4845                               unsigned int *real_num_queues)
4846 {
4847         *num_queues = tx_queues;
4848         return 0;
4849 }
4850
4851 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4852         .kind           = "bond",
4853         .priv_size      = sizeof(struct bonding),
4854         .setup          = bond_setup,
4855         .validate       = bond_validate,
4856         .get_tx_queues  = bond_get_tx_queues,
4857 };
4858
4859 /* Create a new bond based on the specified name and bonding parameters.
4860  * If name is NULL, obtain a suitable "bond%d" name for us.
4861  * Caller must NOT hold rtnl_lock; we need to release it here before we
4862  * set up our sysfs entries.
4863  */
4864 int bond_create(struct net *net, const char *name)
4865 {
4866         struct net_device *bond_dev;
4867         int res;
4868
4869         rtnl_lock();
4870
4871         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4872                                    name ? name : "bond%d",
4873                                    bond_setup, tx_queues);
4874         if (!bond_dev) {
4875                 pr_err("%s: eek! can't alloc netdev!\n", name);
4876                 rtnl_unlock();
4877                 return -ENOMEM;
4878         }
4879
4880         dev_net_set(bond_dev, net);
4881         bond_dev->rtnl_link_ops = &bond_link_ops;
4882
4883         res = register_netdevice(bond_dev);
4884
4885         netif_carrier_off(bond_dev);
4886
4887         rtnl_unlock();
4888         if (res < 0)
4889                 bond_destructor(bond_dev);
4890         return res;
4891 }
4892
4893 static int __net_init bond_net_init(struct net *net)
4894 {
4895         struct bond_net *bn = net_generic(net, bond_net_id);
4896
4897         bn->net = net;
4898         INIT_LIST_HEAD(&bn->dev_list);
4899
4900         bond_create_proc_dir(bn);
4901         bond_create_sysfs(bn);
4902
4903         return 0;
4904 }
4905
4906 static void __net_exit bond_net_exit(struct net *net)
4907 {
4908         struct bond_net *bn = net_generic(net, bond_net_id);
4909
4910         bond_destroy_sysfs(bn);
4911         bond_destroy_proc_dir(bn);
4912 }
4913
4914 static struct pernet_operations bond_net_ops = {
4915         .init = bond_net_init,
4916         .exit = bond_net_exit,
4917         .id   = &bond_net_id,
4918         .size = sizeof(struct bond_net),
4919 };
4920
4921 static int __init bonding_init(void)
4922 {
4923         int i;
4924         int res;
4925
4926         pr_info("%s", bond_version);
4927
4928         res = bond_check_params(&bonding_defaults);
4929         if (res)
4930                 goto out;
4931
4932         res = register_pernet_subsys(&bond_net_ops);
4933         if (res)
4934                 goto out;
4935
4936         res = rtnl_link_register(&bond_link_ops);
4937         if (res)
4938                 goto err_link;
4939
4940         bond_create_debugfs();
4941
4942         for (i = 0; i < max_bonds; i++) {
4943                 res = bond_create(&init_net, NULL);
4944                 if (res)
4945                         goto err;
4946         }
4947
4948         register_netdevice_notifier(&bond_netdev_notifier);
4949         register_inetaddr_notifier(&bond_inetaddr_notifier);
4950 out:
4951         return res;
4952 err:
4953         rtnl_link_unregister(&bond_link_ops);
4954 err_link:
4955         unregister_pernet_subsys(&bond_net_ops);
4956         goto out;
4957
4958 }
4959
4960 static void __exit bonding_exit(void)
4961 {
4962         unregister_netdevice_notifier(&bond_netdev_notifier);
4963         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4964
4965         bond_destroy_debugfs();
4966
4967         rtnl_link_unregister(&bond_link_ops);
4968         unregister_pernet_subsys(&bond_net_ops);
4969
4970 #ifdef CONFIG_NET_POLL_CONTROLLER
4971         /*
4972          * Make sure we don't have an imbalance on our netpoll blocking
4973          */
4974         WARN_ON(atomic_read(&netpoll_block_tx));
4975 #endif
4976 }
4977
4978 module_init(bonding_init);
4979 module_exit(bonding_exit);
4980 MODULE_LICENSE("GPL");
4981 MODULE_VERSION(DRV_VERSION);
4982 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4983 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4984 MODULE_ALIAS_RTNL_LINK("bond");