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_grat_arp = 1;
  93 static int num_unsol_na = 1;
  94 static int miimon       = BOND_LINK_MON_INTERV;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static 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(num_grat_arp, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 118 module_param(num_unsol_na, int, 0644);
 119 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 120 module_param(miimon, int, 0);
 121 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 122 module_param(updelay, int, 0);
 123 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 124 module_param(downdelay, int, 0);
 125 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 126                             "in milliseconds");
 127 module_param(use_carrier, int, 0);
 128 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 129                               "0 for off, 1 for on (default)");
 130 module_param(mode, charp, 0);
 131 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 132                        "1 for active-backup, 2 for balance-xor, "
 133                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 134                        "6 for balance-alb");
 135 module_param(primary, charp, 0);
 136 MODULE_PARM_DESC(primary, "Primary network device to use");
 137 module_param(primary_reselect, charp, 0);
 138 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 139                                    "once it comes up; "
 140                                    "0 for always (default), "
 141                                    "1 for only if speed of primary is "
 142                                    "better, "
 143                                    "2 for only on active slave "
 144                                    "failure");
 145 module_param(lacp_rate, charp, 0);
 146 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 147                             "(slow/fast)");
 148 module_param(ad_select, charp, 0);
 149 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 150 module_param(xmit_hash_policy, charp, 0);
 151 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 152                                    ", 1 for layer 3+4");
 153 module_param(arp_interval, int, 0);
 154 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 155 module_param_array(arp_ip_target, charp, NULL, 0);
 156 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 157 module_param(arp_validate, charp, 0);
 158 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 159 module_param(fail_over_mac, charp, 0);
 160 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 161 module_param(all_slaves_active, int, 0);
 162 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 163                                      "by setting active flag for all slaves.  "
 164                                      "0 for never (default), 1 for always.");
 165 module_param(resend_igmp, int, 0);
 166 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 167
 168 /*----------------------------- Global variables ----------------------------*/
 169
 170 #ifdef CONFIG_NET_POLL_CONTROLLER
 171 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 172 #endif
 173
 174 int bond_net_id __read_mostly;
 175
 176 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 177 static int arp_ip_count;
 178 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 179 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 180 static int lacp_fast;
 181
 182 const struct bond_parm_tbl bond_lacp_tbl[] = {
 183 {       "slow",         AD_LACP_SLOW},
 184 {       "fast",         AD_LACP_FAST},
 185 {       NULL,           -1},
 186 };
 187
 188 const struct bond_parm_tbl bond_mode_tbl[] = {
 189 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 190 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 191 {       "balance-xor",          BOND_MODE_XOR},
 192 {       "broadcast",            BOND_MODE_BROADCAST},
 193 {       "802.3ad",              BOND_MODE_8023AD},
 194 {       "balance-tlb",          BOND_MODE_TLB},
 195 {       "balance-alb",          BOND_MODE_ALB},
 196 {       NULL,                   -1},
 197 };
 198
 199 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 200 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 201 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 202 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 203 {       NULL,                   -1},
 204 };
 205
 206 const struct bond_parm_tbl arp_validate_tbl[] = {
 207 {       "none",                 BOND_ARP_VALIDATE_NONE},
 208 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 209 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 210 {       "all",                  BOND_ARP_VALIDATE_ALL},
 211 {       NULL,                   -1},
 212 };
 213
 214 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 215 {       "none",                 BOND_FOM_NONE},
 216 {       "active",               BOND_FOM_ACTIVE},
 217 {       "follow",               BOND_FOM_FOLLOW},
 218 {       NULL,                   -1},
 219 };
 220
 221 const struct bond_parm_tbl pri_reselect_tbl[] = {
 222 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 223 {       "better",               BOND_PRI_RESELECT_BETTER},
 224 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 225 {       NULL,                   -1},
 226 };
 227
 228 struct bond_parm_tbl ad_select_tbl[] = {
 229 {       "stable",       BOND_AD_STABLE},
 230 {       "bandwidth",    BOND_AD_BANDWIDTH},
 231 {       "count",        BOND_AD_COUNT},
 232 {       NULL,           -1},
 233 };
 234
 235 /*-------------------------- Forward declarations ---------------------------*/
 236
 237 static void bond_send_gratuitous_arp(struct bonding *bond);
 238 static int bond_init(struct net_device *bond_dev);
 239 static void bond_uninit(struct net_device *bond_dev);
 240
 241 /*---------------------------- General routines -----------------------------*/
 242
 243 const char *bond_mode_name(int mode)
 244 {
 245         static const char *names[] = {
 246                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 247                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 248                 [BOND_MODE_XOR] = "load balancing (xor)",
 249                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 250                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 251                 [BOND_MODE_TLB] = "transmit load balancing",
 252                 [BOND_MODE_ALB] = "adaptive load balancing",
 253         };
 254
 255         if (mode < 0 || mode > BOND_MODE_ALB)
 256                 return "unknown";
 257
 258         return names[mode];
 259 }
 260
 261 /*---------------------------------- VLAN -----------------------------------*/
 262
 263 /**
 264  * bond_add_vlan - add a new vlan id on bond
 265  * @bond: bond that got the notification
 266  * @vlan_id: the vlan id to add
 267  *
 268  * Returns -ENOMEM if allocation failed.
 269  */
 270 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 271 {
 272         struct vlan_entry *vlan;
 273
 274         pr_debug("bond: %s, vlan id %d\n",
 275                  (bond ? bond->dev->name : "None"), vlan_id);
 276
 277         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 278         if (!vlan)
 279                 return -ENOMEM;
 280
 281         INIT_LIST_HEAD(&vlan->vlan_list);
 282         vlan->vlan_id = vlan_id;
 283
 284         write_lock_bh(&bond->lock);
 285
 286         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 287
 288         write_unlock_bh(&bond->lock);
 289
 290         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 291
 292         return 0;
 293 }
 294
 295 /**
 296  * bond_del_vlan - delete a vlan id from bond
 297  * @bond: bond that got the notification
 298  * @vlan_id: the vlan id to delete
 299  *
 300  * returns -ENODEV if @vlan_id was not found in @bond.
 301  */
 302 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 303 {
 304         struct vlan_entry *vlan;
 305         int res = -ENODEV;
 306
 307         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 308
 309         block_netpoll_tx();
 310         write_lock_bh(&bond->lock);
 311
 312         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 313                 if (vlan->vlan_id == vlan_id) {
 314                         list_del(&vlan->vlan_list);
 315
 316                         if (bond_is_lb(bond))
 317                                 bond_alb_clear_vlan(bond, vlan_id);
 318
 319                         pr_debug("removed VLAN ID %d from bond %s\n",
 320                                  vlan_id, bond->dev->name);
 321
 322                         kfree(vlan);
 323
 324                         if (list_empty(&bond->vlan_list) &&
 325                             (bond->slave_cnt == 0)) {
 326                                 /* Last VLAN removed and no slaves, so
 327                                  * restore block on adding VLANs. This will
 328                                  * be removed once new slaves that are not
 329                                  * VLAN challenged will be added.
 330                                  */
 331                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 332                         }
 333
 334                         res = 0;
 335                         goto out;
 336                 }
 337         }
 338
 339         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 340                  vlan_id, bond->dev->name);
 341
 342 out:
 343         write_unlock_bh(&bond->lock);
 344         unblock_netpoll_tx();
 345         return res;
 346 }
 347
 348 /**
 349  * bond_has_challenged_slaves
 350  * @bond: the bond we're working on
 351  *
 352  * Searches the slave list. Returns 1 if a vlan challenged slave
 353  * was found, 0 otherwise.
 354  *
 355  * Assumes bond->lock is held.
 356  */
 357 static int bond_has_challenged_slaves(struct bonding *bond)
 358 {
 359         struct slave *slave;
 360         int i;
 361
 362         bond_for_each_slave(bond, slave, i) {
 363                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 364                         pr_debug("found VLAN challenged slave - %s\n",
 365                                  slave->dev->name);
 366                         return 1;
 367                 }
 368         }
 369
 370         pr_debug("no VLAN challenged slaves found\n");
 371         return 0;
 372 }
 373
 374 /**
 375  * bond_next_vlan - safely skip to the next item in the vlans list.
 376  * @bond: the bond we're working on
 377  * @curr: item we're advancing from
 378  *
 379  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 380  * or @curr->next otherwise (even if it is @curr itself again).
 381  *
 382  * Caller must hold bond->lock
 383  */
 384 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 385 {
 386         struct vlan_entry *next, *last;
 387
 388         if (list_empty(&bond->vlan_list))
 389                 return NULL;
 390
 391         if (!curr) {
 392                 next = list_entry(bond->vlan_list.next,
 393                                   struct vlan_entry, vlan_list);
 394         } else {
 395                 last = list_entry(bond->vlan_list.prev,
 396                                   struct vlan_entry, vlan_list);
 397                 if (last == curr) {
 398                         next = list_entry(bond->vlan_list.next,
 399                                           struct vlan_entry, vlan_list);
 400                 } else {
 401                         next = list_entry(curr->vlan_list.next,
 402                                           struct vlan_entry, vlan_list);
 403                 }
 404         }
 405
 406         return next;
 407 }
 408
 409 /**
 410  * bond_dev_queue_xmit - Prepare skb for xmit.
 411  *
 412  * @bond: bond device that got this skb for tx.
 413  * @skb: hw accel VLAN tagged skb to transmit
 414  * @slave_dev: slave that is supposed to xmit this skbuff
 415  */
 416 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 417                         struct net_device *slave_dev)
 418 {
 419         skb->dev = slave_dev;
 420         skb->priority = 1;
 421         if (unlikely(netpoll_tx_running(slave_dev)))
 422                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 423         else
 424                 dev_queue_xmit(skb);
 425
 426         return 0;
 427 }
 428
 429 /*
 430  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 431  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 432  * lock because:
 433  * a. This operation is performed in IOCTL context,
 434  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 435  * c. Holding a lock with BH disabled while directly calling a base driver
 436  *    entry point is generally a BAD idea.
 437  *
 438  * The design of synchronization/protection for this operation in the 8021q
 439  * module is good for one or more VLAN devices over a single physical device
 440  * and cannot be extended for a teaming solution like bonding, so there is a
 441  * potential race condition here where a net device from the vlan group might
 442  * be referenced (either by a base driver or the 8021q code) while it is being
 443  * removed from the system. However, it turns out we're not making matters
 444  * worse, and if it works for regular VLAN usage it will work here too.
 445 */
 446
 447 /**
 448  * bond_vlan_rx_register - Propagates registration to slaves
 449  * @bond_dev: bonding net device that got called
 450  * @grp: vlan group being registered
 451  */
 452 static void bond_vlan_rx_register(struct net_device *bond_dev,
 453                                   struct vlan_group *grp)
 454 {
 455         struct bonding *bond = netdev_priv(bond_dev);
 456         struct slave *slave;
 457         int i;
 458
 459         write_lock_bh(&bond->lock);
 460         bond->vlgrp = grp;
 461         write_unlock_bh(&bond->lock);
 462
 463         bond_for_each_slave(bond, slave, i) {
 464                 struct net_device *slave_dev = slave->dev;
 465                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 466
 467                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 468                     slave_ops->ndo_vlan_rx_register) {
 469                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 470                 }
 471         }
 472 }
 473
 474 /**
 475  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 476  * @bond_dev: bonding net device that got called
 477  * @vid: vlan id being added
 478  */
 479 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 480 {
 481         struct bonding *bond = netdev_priv(bond_dev);
 482         struct slave *slave;
 483         int i, res;
 484
 485         bond_for_each_slave(bond, slave, i) {
 486                 struct net_device *slave_dev = slave->dev;
 487                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 488
 489                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 490                     slave_ops->ndo_vlan_rx_add_vid) {
 491                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 492                 }
 493         }
 494
 495         res = bond_add_vlan(bond, vid);
 496         if (res) {
 497                 pr_err("%s: Error: Failed to add vlan id %d\n",
 498                        bond_dev->name, vid);
 499         }
 500 }
 501
 502 /**
 503  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 504  * @bond_dev: bonding net device that got called
 505  * @vid: vlan id being removed
 506  */
 507 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 508 {
 509         struct bonding *bond = netdev_priv(bond_dev);
 510         struct slave *slave;
 511         struct net_device *vlan_dev;
 512         int i, res;
 513
 514         bond_for_each_slave(bond, slave, i) {
 515                 struct net_device *slave_dev = slave->dev;
 516                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 517
 518                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 519                     slave_ops->ndo_vlan_rx_kill_vid) {
 520                         /* Save and then restore vlan_dev in the grp array,
 521                          * since the slave's driver might clear it.
 522                          */
 523                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 524                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 525                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 526                 }
 527         }
 528
 529         res = bond_del_vlan(bond, vid);
 530         if (res) {
 531                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 532                        bond_dev->name, vid);
 533         }
 534 }
 535
 536 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 537 {
 538         struct vlan_entry *vlan;
 539         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 540
 541         if (!bond->vlgrp)
 542                 return;
 543
 544         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 545             slave_ops->ndo_vlan_rx_register)
 546                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 547
 548         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 549             !(slave_ops->ndo_vlan_rx_add_vid))
 550                 return;
 551
 552         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 553                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 554 }
 555
 556 static void bond_del_vlans_from_slave(struct bonding *bond,
 557                                       struct net_device *slave_dev)
 558 {
 559         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 560         struct vlan_entry *vlan;
 561         struct net_device *vlan_dev;
 562
 563         if (!bond->vlgrp)
 564                 return;
 565
 566         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 567             !(slave_ops->ndo_vlan_rx_kill_vid))
 568                 goto unreg;
 569
 570         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 571                 if (!vlan->vlan_id)
 572                         continue;
 573                 /* Save and then restore vlan_dev in the grp array,
 574                  * since the slave's driver might clear it.
 575                  */
 576                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 577                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 578                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 579         }
 580
 581 unreg:
 582         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 583             slave_ops->ndo_vlan_rx_register)
 584                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 585 }
 586
 587 /*------------------------------- Link status -------------------------------*/
 588
 589 /*
 590  * Set the carrier state for the master according to the state of its
 591  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 592  * do special 802.3ad magic.
 593  *
 594  * Returns zero if carrier state does not change, nonzero if it does.
 595  */
 596 static int bond_set_carrier(struct bonding *bond)
 597 {
 598         struct slave *slave;
 599         int i;
 600
 601         if (bond->slave_cnt == 0)
 602                 goto down;
 603
 604         if (bond->params.mode == BOND_MODE_8023AD)
 605                 return bond_3ad_set_carrier(bond);
 606
 607         bond_for_each_slave(bond, slave, i) {
 608                 if (slave->link == BOND_LINK_UP) {
 609                         if (!netif_carrier_ok(bond->dev)) {
 610                                 netif_carrier_on(bond->dev);
 611                                 return 1;
 612                         }
 613                         return 0;
 614                 }
 615         }
 616
 617 down:
 618         if (netif_carrier_ok(bond->dev)) {
 619                 netif_carrier_off(bond->dev);
 620                 return 1;
 621         }
 622         return 0;
 623 }
 624
 625 /*
 626  * Get link speed and duplex from the slave's base driver
 627  * using ethtool. If for some reason the call fails or the
 628  * values are invalid, fake speed and duplex to 100/Full
 629  * and return error.
 630  */
 631 static int bond_update_speed_duplex(struct slave *slave)
 632 {
 633         struct net_device *slave_dev = slave->dev;
 634         struct ethtool_cmd etool;
 635         int res;
 636
 637         /* Fake speed and duplex */
 638         slave->speed = SPEED_100;
 639         slave->duplex = DUPLEX_FULL;
 640
 641         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 642                 return -1;
 643
 644         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 645         if (res < 0)
 646                 return -1;
 647
 648         switch (etool.speed) {
 649         case SPEED_10:
 650         case SPEED_100:
 651         case SPEED_1000:
 652         case SPEED_10000:
 653                 break;
 654         default:
 655                 return -1;
 656         }
 657
 658         switch (etool.duplex) {
 659         case DUPLEX_FULL:
 660         case DUPLEX_HALF:
 661                 break;
 662         default:
 663                 return -1;
 664         }
 665
 666         slave->speed = etool.speed;
 667         slave->duplex = etool.duplex;
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * if <dev> supports MII link status reporting, check its link status.
 674  *
 675  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 676  * depending upon the setting of the use_carrier parameter.
 677  *
 678  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 679  * can't tell and just pretend it is), or 0, meaning that the link is
 680  * down.
 681  *
 682  * If reporting is non-zero, instead of faking link up, return -1 if
 683  * both ETHTOOL and MII ioctls fail (meaning the device does not
 684  * support them).  If use_carrier is set, return whatever it says.
 685  * It'd be nice if there was a good way to tell if a driver supports
 686  * netif_carrier, but there really isn't.
 687  */
 688 static int bond_check_dev_link(struct bonding *bond,
 689                                struct net_device *slave_dev, int reporting)
 690 {
 691         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 692         int (*ioctl)(struct net_device *, struct ifreq *, int);
 693         struct ifreq ifr;
 694         struct mii_ioctl_data *mii;
 695
 696         if (!reporting && !netif_running(slave_dev))
 697                 return 0;
 698
 699         if (bond->params.use_carrier)
 700                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 701
 702         /* Try to get link status using Ethtool first. */
 703         if (slave_dev->ethtool_ops) {
 704                 if (slave_dev->ethtool_ops->get_link) {
 705                         u32 link;
 706
 707                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 708
 709                         return link ? BMSR_LSTATUS : 0;
 710                 }
 711         }
 712
 713         /* Ethtool can't be used, fallback to MII ioctls. */
 714         ioctl = slave_ops->ndo_do_ioctl;
 715         if (ioctl) {
 716                 /* TODO: set pointer to correct ioctl on a per team member */
 717                 /*       bases to make this more efficient. that is, once  */
 718                 /*       we determine the correct ioctl, we will always    */
 719                 /*       call it and not the others for that team          */
 720                 /*       member.                                           */
 721
 722                 /*
 723                  * We cannot assume that SIOCGMIIPHY will also read a
 724                  * register; not all network drivers (e.g., e100)
 725                  * support that.
 726                  */
 727
 728                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 729                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 730                 mii = if_mii(&ifr);
 731                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 732                         mii->reg_num = MII_BMSR;
 733                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 734                                 return mii->val_out & BMSR_LSTATUS;
 735                 }
 736         }
 737
 738         /*
 739          * If reporting, report that either there's no dev->do_ioctl,
 740          * or both SIOCGMIIREG and get_link failed (meaning that we
 741          * cannot report link status).  If not reporting, pretend
 742          * we're ok.
 743          */
 744         return reporting ? -1 : BMSR_LSTATUS;
 745 }
 746
 747 /*----------------------------- Multicast list ------------------------------*/
 748
 749 /*
 750  * Push the promiscuity flag down to appropriate slaves
 751  */
 752 static int bond_set_promiscuity(struct bonding *bond, int inc)
 753 {
 754         int err = 0;
 755         if (USES_PRIMARY(bond->params.mode)) {
 756                 /* write lock already acquired */
 757                 if (bond->curr_active_slave) {
 758                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 759                                                   inc);
 760                 }
 761         } else {
 762                 struct slave *slave;
 763                 int i;
 764                 bond_for_each_slave(bond, slave, i) {
 765                         err = dev_set_promiscuity(slave->dev, inc);
 766                         if (err)
 767                                 return err;
 768                 }
 769         }
 770         return err;
 771 }
 772
 773 /*
 774  * Push the allmulti flag down to all slaves
 775  */
 776 static int bond_set_allmulti(struct bonding *bond, int inc)
 777 {
 778         int err = 0;
 779         if (USES_PRIMARY(bond->params.mode)) {
 780                 /* write lock already acquired */
 781                 if (bond->curr_active_slave) {
 782                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 783                                                inc);
 784                 }
 785         } else {
 786                 struct slave *slave;
 787                 int i;
 788                 bond_for_each_slave(bond, slave, i) {
 789                         err = dev_set_allmulti(slave->dev, inc);
 790                         if (err)
 791                                 return err;
 792                 }
 793         }
 794         return err;
 795 }
 796
 797 /*
 798  * Add a Multicast address to slaves
 799  * according to mode
 800  */
 801 static void bond_mc_add(struct bonding *bond, void *addr)
 802 {
 803         if (USES_PRIMARY(bond->params.mode)) {
 804                 /* write lock already acquired */
 805                 if (bond->curr_active_slave)
 806                         dev_mc_add(bond->curr_active_slave->dev, addr);
 807         } else {
 808                 struct slave *slave;
 809                 int i;
 810
 811                 bond_for_each_slave(bond, slave, i)
 812                         dev_mc_add(slave->dev, addr);
 813         }
 814 }
 815
 816 /*
 817  * Remove a multicast address from slave
 818  * according to mode
 819  */
 820 static void bond_mc_del(struct bonding *bond, void *addr)
 821 {
 822         if (USES_PRIMARY(bond->params.mode)) {
 823                 /* write lock already acquired */
 824                 if (bond->curr_active_slave)
 825                         dev_mc_del(bond->curr_active_slave->dev, addr);
 826         } else {
 827                 struct slave *slave;
 828                 int i;
 829                 bond_for_each_slave(bond, slave, i) {
 830                         dev_mc_del(slave->dev, addr);
 831                 }
 832         }
 833 }
 834
 835
 836 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 837 {
 838         struct in_device *in_dev;
 839
 840         rcu_read_lock();
 841         in_dev = __in_dev_get_rcu(dev);
 842         if (in_dev)
 843                 ip_mc_rejoin_groups(in_dev);
 844         rcu_read_unlock();
 845 }
 846
 847 /*
 848  * Retrieve the list of registered multicast addresses for the bonding
 849  * device and retransmit an IGMP JOIN request to the current active
 850  * slave.
 851  */
 852 static void bond_resend_igmp_join_requests(struct bonding *bond)
 853 {
 854         struct net_device *vlan_dev;
 855         struct vlan_entry *vlan;
 856
 857         read_lock(&bond->lock);
 858
 859         /* rejoin all groups on bond device */
 860         __bond_resend_igmp_join_requests(bond->dev);
 861
 862         /* rejoin all groups on vlan devices */
 863         if (bond->vlgrp) {
 864                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 865                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 866                                                          vlan->vlan_id);
 867                         if (vlan_dev)
 868                                 __bond_resend_igmp_join_requests(vlan_dev);
 869                 }
 870         }
 871
 872         if (--bond->igmp_retrans > 0)
 873                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 874
 875         read_unlock(&bond->lock);
 876 }
 877
 878 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 879 {
 880         struct bonding *bond = container_of(work, struct bonding,
 881                                                         mcast_work.work);
 882         bond_resend_igmp_join_requests(bond);
 883 }
 884
 885 /*
 886  * flush all members of flush->mc_list from device dev->mc_list
 887  */
 888 static void bond_mc_list_flush(struct net_device *bond_dev,
 889                                struct net_device *slave_dev)
 890 {
 891         struct bonding *bond = netdev_priv(bond_dev);
 892         struct netdev_hw_addr *ha;
 893
 894         netdev_for_each_mc_addr(ha, bond_dev)
 895                 dev_mc_del(slave_dev, ha->addr);
 896
 897         if (bond->params.mode == BOND_MODE_8023AD) {
 898                 /* del lacpdu mc addr from mc list */
 899                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 900
 901                 dev_mc_del(slave_dev, lacpdu_multicast);
 902         }
 903 }
 904
 905 /*--------------------------- Active slave change ---------------------------*/
 906
 907 /*
 908  * Update the mc list and multicast-related flags for the new and
 909  * old active slaves (if any) according to the multicast mode, and
 910  * promiscuous flags unconditionally.
 911  */
 912 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 913                          struct slave *old_active)
 914 {
 915         struct netdev_hw_addr *ha;
 916
 917         if (!USES_PRIMARY(bond->params.mode))
 918                 /* nothing to do -  mc list is already up-to-date on
 919                  * all slaves
 920                  */
 921                 return;
 922
 923         if (old_active) {
 924                 if (bond->dev->flags & IFF_PROMISC)
 925                         dev_set_promiscuity(old_active->dev, -1);
 926
 927                 if (bond->dev->flags & IFF_ALLMULTI)
 928                         dev_set_allmulti(old_active->dev, -1);
 929
 930                 netdev_for_each_mc_addr(ha, bond->dev)
 931                         dev_mc_del(old_active->dev, ha->addr);
 932         }
 933
 934         if (new_active) {
 935                 /* FIXME: Signal errors upstream. */
 936                 if (bond->dev->flags & IFF_PROMISC)
 937                         dev_set_promiscuity(new_active->dev, 1);
 938
 939                 if (bond->dev->flags & IFF_ALLMULTI)
 940                         dev_set_allmulti(new_active->dev, 1);
 941
 942                 netdev_for_each_mc_addr(ha, bond->dev)
 943                         dev_mc_add(new_active->dev, ha->addr);
 944         }
 945 }
 946
 947 /*
 948  * bond_do_fail_over_mac
 949  *
 950  * Perform special MAC address swapping for fail_over_mac settings
 951  *
 952  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 953  */
 954 static void bond_do_fail_over_mac(struct bonding *bond,
 955                                   struct slave *new_active,
 956                                   struct slave *old_active)
 957         __releases(&bond->curr_slave_lock)
 958         __releases(&bond->lock)
 959         __acquires(&bond->lock)
 960         __acquires(&bond->curr_slave_lock)
 961 {
 962         u8 tmp_mac[ETH_ALEN];
 963         struct sockaddr saddr;
 964         int rv;
 965
 966         switch (bond->params.fail_over_mac) {
 967         case BOND_FOM_ACTIVE:
 968                 if (new_active)
 969                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 970                                new_active->dev->addr_len);
 971                 break;
 972         case BOND_FOM_FOLLOW:
 973                 /*
 974                  * if new_active && old_active, swap them
 975                  * if just old_active, do nothing (going to no active slave)
 976                  * if just new_active, set new_active to bond's MAC
 977                  */
 978                 if (!new_active)
 979                         return;
 980
 981                 write_unlock_bh(&bond->curr_slave_lock);
 982                 read_unlock(&bond->lock);
 983
 984                 if (old_active) {
 985                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 986                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 987                                ETH_ALEN);
 988                         saddr.sa_family = new_active->dev->type;
 989                 } else {
 990                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 991                         saddr.sa_family = bond->dev->type;
 992                 }
 993
 994                 rv = dev_set_mac_address(new_active->dev, &saddr);
 995                 if (rv) {
 996                         pr_err("%s: Error %d setting MAC of slave %s\n",
 997                                bond->dev->name, -rv, new_active->dev->name);
 998                         goto out;
 999                 }
1000
1001                 if (!old_active)
1002                         goto out;
1003
1004                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1005                 saddr.sa_family = old_active->dev->type;
1006
1007                 rv = dev_set_mac_address(old_active->dev, &saddr);
1008                 if (rv)
1009                         pr_err("%s: Error %d setting MAC of slave %s\n",
1010                                bond->dev->name, -rv, new_active->dev->name);
1011 out:
1012                 read_lock(&bond->lock);
1013                 write_lock_bh(&bond->curr_slave_lock);
1014                 break;
1015         default:
1016                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1017                        bond->dev->name, bond->params.fail_over_mac);
1018                 break;
1019         }
1020
1021 }
1022
1023 static bool bond_should_change_active(struct bonding *bond)
1024 {
1025         struct slave *prim = bond->primary_slave;
1026         struct slave *curr = bond->curr_active_slave;
1027
1028         if (!prim || !curr || curr->link != BOND_LINK_UP)
1029                 return true;
1030         if (bond->force_primary) {
1031                 bond->force_primary = false;
1032                 return true;
1033         }
1034         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1035             (prim->speed < curr->speed ||
1036              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1037                 return false;
1038         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1039                 return false;
1040         return true;
1041 }
1042
1043 /**
1044  * find_best_interface - select the best available slave to be the active one
1045  * @bond: our bonding struct
1046  *
1047  * Warning: Caller must hold curr_slave_lock for writing.
1048  */
1049 static struct slave *bond_find_best_slave(struct bonding *bond)
1050 {
1051         struct slave *new_active, *old_active;
1052         struct slave *bestslave = NULL;
1053         int mintime = bond->params.updelay;
1054         int i;
1055
1056         new_active = bond->curr_active_slave;
1057
1058         if (!new_active) { /* there were no active slaves left */
1059                 if (bond->slave_cnt > 0)   /* found one slave */
1060                         new_active = bond->first_slave;
1061                 else
1062                         return NULL; /* still no slave, return NULL */
1063         }
1064
1065         if ((bond->primary_slave) &&
1066             bond->primary_slave->link == BOND_LINK_UP &&
1067             bond_should_change_active(bond)) {
1068                 new_active = bond->primary_slave;
1069         }
1070
1071         /* remember where to stop iterating over the slaves */
1072         old_active = new_active;
1073
1074         bond_for_each_slave_from(bond, new_active, i, old_active) {
1075                 if (new_active->link == BOND_LINK_UP) {
1076                         return new_active;
1077                 } else if (new_active->link == BOND_LINK_BACK &&
1078                            IS_UP(new_active->dev)) {
1079                         /* link up, but waiting for stabilization */
1080                         if (new_active->delay < mintime) {
1081                                 mintime = new_active->delay;
1082                                 bestslave = new_active;
1083                         }
1084                 }
1085         }
1086
1087         return bestslave;
1088 }
1089
1090 /**
1091  * change_active_interface - change the active slave into the specified one
1092  * @bond: our bonding struct
1093  * @new: the new slave to make the active one
1094  *
1095  * Set the new slave to the bond's settings and unset them on the old
1096  * curr_active_slave.
1097  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1098  *
1099  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1100  * because it is apparently the best available slave we have, even though its
1101  * updelay hasn't timed out yet.
1102  *
1103  * If new_active is not NULL, caller must hold bond->lock for read and
1104  * curr_slave_lock for write_bh.
1105  */
1106 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1107 {
1108         struct slave *old_active = bond->curr_active_slave;
1109
1110         if (old_active == new_active)
1111                 return;
1112
1113         if (new_active) {
1114                 new_active->jiffies = jiffies;
1115
1116                 if (new_active->link == BOND_LINK_BACK) {
1117                         if (USES_PRIMARY(bond->params.mode)) {
1118                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1119                                         bond->dev->name, new_active->dev->name,
1120                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1121                         }
1122
1123                         new_active->delay = 0;
1124                         new_active->link = BOND_LINK_UP;
1125
1126                         if (bond->params.mode == BOND_MODE_8023AD)
1127                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1128
1129                         if (bond_is_lb(bond))
1130                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1131                 } else {
1132                         if (USES_PRIMARY(bond->params.mode)) {
1133                                 pr_info("%s: making interface %s the new active one.\n",
1134                                         bond->dev->name, new_active->dev->name);
1135                         }
1136                 }
1137         }
1138
1139         if (USES_PRIMARY(bond->params.mode))
1140                 bond_mc_swap(bond, new_active, old_active);
1141
1142         if (bond_is_lb(bond)) {
1143                 bond_alb_handle_active_change(bond, new_active);
1144                 if (old_active)
1145                         bond_set_slave_inactive_flags(old_active);
1146                 if (new_active)
1147                         bond_set_slave_active_flags(new_active);
1148         } else {
1149                 bond->curr_active_slave = new_active;
1150         }
1151
1152         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1153                 if (old_active)
1154                         bond_set_slave_inactive_flags(old_active);
1155
1156                 if (new_active) {
1157                         bond_set_slave_active_flags(new_active);
1158
1159                         if (bond->params.fail_over_mac)
1160                                 bond_do_fail_over_mac(bond, new_active,
1161                                                       old_active);
1162
1163                         if (netif_running(bond->dev)) {
1164                                 bond->send_grat_arp = bond->params.num_grat_arp;
1165                                 bond_send_gratuitous_arp(bond);
1166
1167                                 bond->send_unsol_na = bond->params.num_unsol_na;
1168                                 bond_send_unsolicited_na(bond);
1169                         }
1170
1171                         write_unlock_bh(&bond->curr_slave_lock);
1172                         read_unlock(&bond->lock);
1173
1174                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1175
1176                         read_lock(&bond->lock);
1177                         write_lock_bh(&bond->curr_slave_lock);
1178                 }
1179         }
1180
1181         /* resend IGMP joins since active slave has changed or
1182          * all were sent on curr_active_slave */
1183         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1184              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1185             netif_running(bond->dev)) {
1186                 bond->igmp_retrans = bond->params.resend_igmp;
1187                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1188         }
1189 }
1190
1191 /**
1192  * bond_select_active_slave - select a new active slave, if needed
1193  * @bond: our bonding struct
1194  *
1195  * This functions should be called when one of the following occurs:
1196  * - The old curr_active_slave has been released or lost its link.
1197  * - The primary_slave has got its link back.
1198  * - A slave has got its link back and there's no old curr_active_slave.
1199  *
1200  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1201  */
1202 void bond_select_active_slave(struct bonding *bond)
1203 {
1204         struct slave *best_slave;
1205         int rv;
1206
1207         best_slave = bond_find_best_slave(bond);
1208         if (best_slave != bond->curr_active_slave) {
1209                 bond_change_active_slave(bond, best_slave);
1210                 rv = bond_set_carrier(bond);
1211                 if (!rv)
1212                         return;
1213
1214                 if (netif_carrier_ok(bond->dev)) {
1215                         pr_info("%s: first active interface up!\n",
1216                                 bond->dev->name);
1217                 } else {
1218                         pr_info("%s: now running without any active interface !\n",
1219                                 bond->dev->name);
1220                 }
1221         }
1222 }
1223
1224 /*--------------------------- slave list handling ---------------------------*/
1225
1226 /*
1227  * This function attaches the slave to the end of list.
1228  *
1229  * bond->lock held for writing by caller.
1230  */
1231 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1232 {
1233         if (bond->first_slave == NULL) { /* attaching the first slave */
1234                 new_slave->next = new_slave;
1235                 new_slave->prev = new_slave;
1236                 bond->first_slave = new_slave;
1237         } else {
1238                 new_slave->next = bond->first_slave;
1239                 new_slave->prev = bond->first_slave->prev;
1240                 new_slave->next->prev = new_slave;
1241                 new_slave->prev->next = new_slave;
1242         }
1243
1244         bond->slave_cnt++;
1245 }
1246
1247 /*
1248  * This function detaches the slave from the list.
1249  * WARNING: no check is made to verify if the slave effectively
1250  * belongs to <bond>.
1251  * Nothing is freed on return, structures are just unchained.
1252  * If any slave pointer in bond was pointing to <slave>,
1253  * it should be changed by the calling function.
1254  *
1255  * bond->lock held for writing by caller.
1256  */
1257 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1258 {
1259         if (slave->next)
1260                 slave->next->prev = slave->prev;
1261
1262         if (slave->prev)
1263                 slave->prev->next = slave->next;
1264
1265         if (bond->first_slave == slave) { /* slave is the first slave */
1266                 if (bond->slave_cnt > 1) { /* there are more slave */
1267                         bond->first_slave = slave->next;
1268                 } else {
1269                         bond->first_slave = NULL; /* slave was the last one */
1270                 }
1271         }
1272
1273         slave->next = NULL;
1274         slave->prev = NULL;
1275         bond->slave_cnt--;
1276 }
1277
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 static inline int slave_enable_netpoll(struct slave *slave)
1280 {
1281         struct netpoll *np;
1282         int err = 0;
1283
1284         np = kzalloc(sizeof(*np), GFP_KERNEL);
1285         err = -ENOMEM;
1286         if (!np)
1287                 goto out;
1288
1289         np->dev = slave->dev;
1290         err = __netpoll_setup(np);
1291         if (err) {
1292                 kfree(np);
1293                 goto out;
1294         }
1295         slave->np = np;
1296 out:
1297         return err;
1298 }
1299 static inline void slave_disable_netpoll(struct slave *slave)
1300 {
1301         struct netpoll *np = slave->np;
1302
1303         if (!np)
1304                 return;
1305
1306         slave->np = NULL;
1307         synchronize_rcu_bh();
1308         __netpoll_cleanup(np);
1309         kfree(np);
1310 }
1311 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1312 {
1313         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1314                 return false;
1315         if (!slave_dev->netdev_ops->ndo_poll_controller)
1316                 return false;
1317         return true;
1318 }
1319
1320 static void bond_poll_controller(struct net_device *bond_dev)
1321 {
1322 }
1323
1324 static void __bond_netpoll_cleanup(struct bonding *bond)
1325 {
1326         struct slave *slave;
1327         int i;
1328
1329         bond_for_each_slave(bond, slave, i)
1330                 if (IS_UP(slave->dev))
1331                         slave_disable_netpoll(slave);
1332 }
1333 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1334 {
1335         struct bonding *bond = netdev_priv(bond_dev);
1336
1337         read_lock(&bond->lock);
1338         __bond_netpoll_cleanup(bond);
1339         read_unlock(&bond->lock);
1340 }
1341
1342 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1343 {
1344         struct bonding *bond = netdev_priv(dev);
1345         struct slave *slave;
1346         int i, err = 0;
1347
1348         read_lock(&bond->lock);
1349         bond_for_each_slave(bond, slave, i) {
1350                 err = slave_enable_netpoll(slave);
1351                 if (err) {
1352                         __bond_netpoll_cleanup(bond);
1353                         break;
1354                 }
1355         }
1356         read_unlock(&bond->lock);
1357         return err;
1358 }
1359
1360 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1361 {
1362         return bond->dev->npinfo;
1363 }
1364
1365 #else
1366 static inline int slave_enable_netpoll(struct slave *slave)
1367 {
1368         return 0;
1369 }
1370 static inline void slave_disable_netpoll(struct slave *slave)
1371 {
1372 }
1373 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1374 {
1375 }
1376 #endif
1377
1378 /*---------------------------------- IOCTL ----------------------------------*/
1379
1380 static int bond_sethwaddr(struct net_device *bond_dev,
1381                           struct net_device *slave_dev)
1382 {
1383         pr_debug("bond_dev=%p\n", bond_dev);
1384         pr_debug("slave_dev=%p\n", slave_dev);
1385         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1386         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1387         return 0;
1388 }
1389
1390 #define BOND_VLAN_FEATURES \
1391         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1392          NETIF_F_HW_VLAN_FILTER)
1393
1394 /*
1395  * Compute the common dev->feature set available to all slaves.  Some
1396  * feature bits are managed elsewhere, so preserve those feature bits
1397  * on the master device.
1398  */
1399 static int bond_compute_features(struct bonding *bond)
1400 {
1401         struct slave *slave;
1402         struct net_device *bond_dev = bond->dev;
1403         u32 features = bond_dev->features;
1404         u32 vlan_features = 0;
1405         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1406                                                 bond_dev->hard_header_len);
1407         int i;
1408
1409         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1410         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1411
1412         if (!bond->first_slave)
1413                 goto done;
1414
1415         features &= ~NETIF_F_ONE_FOR_ALL;
1416
1417         vlan_features = bond->first_slave->dev->vlan_features;
1418         bond_for_each_slave(bond, slave, i) {
1419                 features = netdev_increment_features(features,
1420                                                      slave->dev->features,
1421                                                      NETIF_F_ONE_FOR_ALL);
1422                 vlan_features = netdev_increment_features(vlan_features,
1423                                                         slave->dev->vlan_features,
1424                                                         NETIF_F_ONE_FOR_ALL);
1425                 if (slave->dev->hard_header_len > max_hard_header_len)
1426                         max_hard_header_len = slave->dev->hard_header_len;
1427         }
1428
1429 done:
1430         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1431         bond_dev->features = netdev_fix_features(bond_dev, features);
1432         bond_dev->vlan_features = netdev_fix_features(bond_dev, vlan_features);
1433         bond_dev->hard_header_len = max_hard_header_len;
1434
1435         return 0;
1436 }
1437
1438 static void bond_setup_by_slave(struct net_device *bond_dev,
1439                                 struct net_device *slave_dev)
1440 {
1441         struct bonding *bond = netdev_priv(bond_dev);
1442
1443         bond_dev->header_ops        = slave_dev->header_ops;
1444
1445         bond_dev->type              = slave_dev->type;
1446         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1447         bond_dev->addr_len          = slave_dev->addr_len;
1448
1449         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1450                 slave_dev->addr_len);
1451         bond->setup_by_slave = 1;
1452 }
1453
1454 /* On bonding slaves other than the currently active slave, suppress
1455  * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
1456  * ARP on active-backup slaves with arp_validate enabled.
1457  */
1458 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1459                                             struct slave *slave,
1460                                             struct bonding *bond)
1461 {
1462         if (bond_is_slave_inactive(slave)) {
1463                 if (slave_do_arp_validate(bond, slave) &&
1464                     skb->protocol == __cpu_to_be16(ETH_P_ARP))
1465                         return false;
1466
1467                 if (bond->params.mode == BOND_MODE_ALB &&
1468                     skb->pkt_type != PACKET_BROADCAST &&
1469                     skb->pkt_type != PACKET_MULTICAST)
1470                                 return false;
1471
1472                 if (bond->params.mode == BOND_MODE_8023AD &&
1473                     skb->protocol == __cpu_to_be16(ETH_P_SLOW))
1474                         return false;
1475
1476                 return true;
1477         }
1478         return false;
1479 }
1480
1481 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1482 {
1483         struct sk_buff *skb = *pskb;
1484         struct slave *slave;
1485         struct net_device *bond_dev;
1486         struct bonding *bond;
1487
1488         slave = bond_slave_get_rcu(skb->dev);
1489         bond_dev = ACCESS_ONCE(slave->dev->master);
1490         if (unlikely(!bond_dev))
1491                 return RX_HANDLER_PASS;
1492
1493         skb = skb_share_check(skb, GFP_ATOMIC);
1494         if (unlikely(!skb))
1495                 return RX_HANDLER_CONSUMED;
1496
1497         *pskb = skb;
1498
1499         bond = netdev_priv(bond_dev);
1500
1501         if (bond->params.arp_interval)
1502                 slave->dev->last_rx = jiffies;
1503
1504         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1505                 return RX_HANDLER_EXACT;
1506         }
1507
1508         skb->dev = bond_dev;
1509
1510         if (bond->params.mode == BOND_MODE_ALB &&
1511             bond_dev->priv_flags & IFF_BRIDGE_PORT &&
1512             skb->pkt_type == PACKET_HOST) {
1513
1514                 if (unlikely(skb_cow_head(skb,
1515                                           skb->data - skb_mac_header(skb)))) {
1516                         kfree_skb(skb);
1517                         return RX_HANDLER_CONSUMED;
1518                 }
1519                 memcpy(eth_hdr(skb)->h_dest, bond_dev->dev_addr, ETH_ALEN);
1520         }
1521
1522         return RX_HANDLER_ANOTHER;
1523 }
1524
1525 /* enslave device <slave> to bond device <master> */
1526 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1527 {
1528         struct bonding *bond = netdev_priv(bond_dev);
1529         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1530         struct slave *new_slave = NULL;
1531         struct netdev_hw_addr *ha;
1532         struct sockaddr addr;
1533         int link_reporting;
1534         int old_features = bond_dev->features;
1535         int res = 0;
1536
1537         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1538                 slave_ops->ndo_do_ioctl == NULL) {
1539                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1540                            bond_dev->name, slave_dev->name);
1541         }
1542
1543         /* bond must be initialized by bond_open() before enslaving */
1544         if (!(bond_dev->flags & IFF_UP)) {
1545                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1546                            bond_dev->name);
1547         }
1548
1549         /* already enslaved */
1550         if (slave_dev->flags & IFF_SLAVE) {
1551                 pr_debug("Error, Device was already enslaved\n");
1552                 return -EBUSY;
1553         }
1554
1555         /* vlan challenged mutual exclusion */
1556         /* no need to lock since we're protected by rtnl_lock */
1557         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1558                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1559                 if (bond->vlgrp) {
1560                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1561                                bond_dev->name, slave_dev->name, bond_dev->name);
1562                         return -EPERM;
1563                 } else {
1564                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1565                                    bond_dev->name, slave_dev->name,
1566                                    slave_dev->name, bond_dev->name);
1567                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1568                 }
1569         } else {
1570                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1571                 if (bond->slave_cnt == 0) {
1572                         /* First slave, and it is not VLAN challenged,
1573                          * so remove the block of adding VLANs over the bond.
1574                          */
1575                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1576                 }
1577         }
1578
1579         /*
1580          * Old ifenslave binaries are no longer supported.  These can
1581          * be identified with moderate accuracy by the state of the slave:
1582          * the current ifenslave will set the interface down prior to
1583          * enslaving it; the old ifenslave will not.
1584          */
1585         if ((slave_dev->flags & IFF_UP)) {
1586                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1587                        slave_dev->name);
1588                 res = -EPERM;
1589                 goto err_undo_flags;
1590         }
1591
1592         /* set bonding device ether type by slave - bonding netdevices are
1593          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1594          * there is a need to override some of the type dependent attribs/funcs.
1595          *
1596          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1597          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1598          */
1599         if (bond->slave_cnt == 0) {
1600                 if (bond_dev->type != slave_dev->type) {
1601                         pr_debug("%s: change device type from %d to %d\n",
1602                                  bond_dev->name,
1603                                  bond_dev->type, slave_dev->type);
1604
1605                         res = netdev_bonding_change(bond_dev,
1606                                                     NETDEV_PRE_TYPE_CHANGE);
1607                         res = notifier_to_errno(res);
1608                         if (res) {
1609                                 pr_err("%s: refused to change device type\n",
1610                                        bond_dev->name);
1611                                 res = -EBUSY;
1612                                 goto err_undo_flags;
1613                         }
1614
1615                         /* Flush unicast and multicast addresses */
1616                         dev_uc_flush(bond_dev);
1617                         dev_mc_flush(bond_dev);
1618
1619                         if (slave_dev->type != ARPHRD_ETHER)
1620                                 bond_setup_by_slave(bond_dev, slave_dev);
1621                         else
1622                                 ether_setup(bond_dev);
1623
1624                         netdev_bonding_change(bond_dev,
1625                                               NETDEV_POST_TYPE_CHANGE);
1626                 }
1627         } else if (bond_dev->type != slave_dev->type) {
1628                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1629                        slave_dev->name,
1630                        slave_dev->type, bond_dev->type);
1631                 res = -EINVAL;
1632                 goto err_undo_flags;
1633         }
1634
1635         if (slave_ops->ndo_set_mac_address == NULL) {
1636                 if (bond->slave_cnt == 0) {
1637                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1638                                    bond_dev->name);
1639                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1640                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1641                         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",
1642                                bond_dev->name);
1643                         res = -EOPNOTSUPP;
1644                         goto err_undo_flags;
1645                 }
1646         }
1647
1648         /* If this is the first slave, then we need to set the master's hardware
1649          * address to be the same as the slave's. */
1650         if (is_zero_ether_addr(bond->dev->dev_addr))
1651                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1652                        slave_dev->addr_len);
1653
1654
1655         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1656         if (!new_slave) {
1657                 res = -ENOMEM;
1658                 goto err_undo_flags;
1659         }
1660
1661         /*
1662          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1663          * is set via sysfs or module option if desired.
1664          */
1665         new_slave->queue_id = 0;
1666
1667         /* Save slave's original mtu and then set it to match the bond */
1668         new_slave->original_mtu = slave_dev->mtu;
1669         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1670         if (res) {
1671                 pr_debug("Error %d calling dev_set_mtu\n", res);
1672                 goto err_free;
1673         }
1674
1675         /*
1676          * Save slave's original ("permanent") mac address for modes
1677          * that need it, and for restoring it upon release, and then
1678          * set it to the master's address
1679          */
1680         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1681
1682         if (!bond->params.fail_over_mac) {
1683                 /*
1684                  * Set slave to master's mac address.  The application already
1685                  * set the master's mac address to that of the first slave
1686                  */
1687                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1688                 addr.sa_family = slave_dev->type;
1689                 res = dev_set_mac_address(slave_dev, &addr);
1690                 if (res) {
1691                         pr_debug("Error %d calling set_mac_address\n", res);
1692                         goto err_restore_mtu;
1693                 }
1694         }
1695
1696         res = netdev_set_bond_master(slave_dev, bond_dev);
1697         if (res) {
1698                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1699                 goto err_restore_mac;
1700         }
1701         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1702                                          new_slave);
1703         if (res) {
1704                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1705                 goto err_unset_master;
1706         }
1707
1708         /* open the slave since the application closed it */
1709         res = dev_open(slave_dev);
1710         if (res) {
1711                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1712                 goto err_unreg_rxhandler;
1713         }
1714
1715         new_slave->dev = slave_dev;
1716         slave_dev->priv_flags |= IFF_BONDING;
1717
1718         if (bond_is_lb(bond)) {
1719                 /* bond_alb_init_slave() must be called before all other stages since
1720                  * it might fail and we do not want to have to undo everything
1721                  */
1722                 res = bond_alb_init_slave(bond, new_slave);
1723                 if (res)
1724                         goto err_close;
1725         }
1726
1727         /* If the mode USES_PRIMARY, then the new slave gets the
1728          * master's promisc (and mc) settings only if it becomes the
1729          * curr_active_slave, and that is taken care of later when calling
1730          * bond_change_active()
1731          */
1732         if (!USES_PRIMARY(bond->params.mode)) {
1733                 /* set promiscuity level to new slave */
1734                 if (bond_dev->flags & IFF_PROMISC) {
1735                         res = dev_set_promiscuity(slave_dev, 1);
1736                         if (res)
1737                                 goto err_close;
1738                 }
1739
1740                 /* set allmulti level to new slave */
1741                 if (bond_dev->flags & IFF_ALLMULTI) {
1742                         res = dev_set_allmulti(slave_dev, 1);
1743                         if (res)
1744                                 goto err_close;
1745                 }
1746
1747                 netif_addr_lock_bh(bond_dev);
1748                 /* upload master's mc_list to new slave */
1749                 netdev_for_each_mc_addr(ha, bond_dev)
1750                         dev_mc_add(slave_dev, ha->addr);
1751                 netif_addr_unlock_bh(bond_dev);
1752         }
1753
1754         if (bond->params.mode == BOND_MODE_8023AD) {
1755                 /* add lacpdu mc addr to mc list */
1756                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1757
1758                 dev_mc_add(slave_dev, lacpdu_multicast);
1759         }
1760
1761         bond_add_vlans_on_slave(bond, slave_dev);
1762
1763         write_lock_bh(&bond->lock);
1764
1765         bond_attach_slave(bond, new_slave);
1766
1767         new_slave->delay = 0;
1768         new_slave->link_failure_count = 0;
1769
1770         bond_compute_features(bond);
1771
1772         write_unlock_bh(&bond->lock);
1773
1774         read_lock(&bond->lock);
1775
1776         new_slave->last_arp_rx = jiffies;
1777
1778         if (bond->params.miimon && !bond->params.use_carrier) {
1779                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1780
1781                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1782                         /*
1783                          * miimon is set but a bonded network driver
1784                          * does not support ETHTOOL/MII and
1785                          * arp_interval is not set.  Note: if
1786                          * use_carrier is enabled, we will never go
1787                          * here (because netif_carrier is always
1788                          * supported); thus, we don't need to change
1789                          * the messages for netif_carrier.
1790                          */
1791                         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",
1792                                bond_dev->name, slave_dev->name);
1793                 } else if (link_reporting == -1) {
1794                         /* unable get link status using mii/ethtool */
1795                         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",
1796                                    bond_dev->name, slave_dev->name);
1797                 }
1798         }
1799
1800         /* check for initial state */
1801         if (!bond->params.miimon ||
1802             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1803                 if (bond->params.updelay) {
1804                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1805                         new_slave->link  = BOND_LINK_BACK;
1806                         new_slave->delay = bond->params.updelay;
1807                 } else {
1808                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1809                         new_slave->link  = BOND_LINK_UP;
1810                 }
1811                 new_slave->jiffies = jiffies;
1812         } else {
1813                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1814                 new_slave->link  = BOND_LINK_DOWN;
1815         }
1816
1817         if (bond_update_speed_duplex(new_slave) &&
1818             (new_slave->link != BOND_LINK_DOWN)) {
1819                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1820                            bond_dev->name, new_slave->dev->name);
1821
1822                 if (bond->params.mode == BOND_MODE_8023AD) {
1823                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1824                                    bond_dev->name);
1825                 }
1826         }
1827
1828         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1829                 /* if there is a primary slave, remember it */
1830                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1831                         bond->primary_slave = new_slave;
1832                         bond->force_primary = true;
1833                 }
1834         }
1835
1836         write_lock_bh(&bond->curr_slave_lock);
1837
1838         switch (bond->params.mode) {
1839         case BOND_MODE_ACTIVEBACKUP:
1840                 bond_set_slave_inactive_flags(new_slave);
1841                 bond_select_active_slave(bond);
1842                 break;
1843         case BOND_MODE_8023AD:
1844                 /* in 802.3ad mode, the internal mechanism
1845                  * will activate the slaves in the selected
1846                  * aggregator
1847                  */
1848                 bond_set_slave_inactive_flags(new_slave);
1849                 /* if this is the first slave */
1850                 if (bond->slave_cnt == 1) {
1851                         SLAVE_AD_INFO(new_slave).id = 1;
1852                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1853                          * can be called only after the mac address of the bond is set
1854                          */
1855                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1856                                             bond->params.lacp_fast);
1857                 } else {
1858                         SLAVE_AD_INFO(new_slave).id =
1859                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1860                 }
1861
1862                 bond_3ad_bind_slave(new_slave);
1863                 break;
1864         case BOND_MODE_TLB:
1865         case BOND_MODE_ALB:
1866                 bond_set_active_slave(new_slave);
1867                 bond_set_slave_inactive_flags(new_slave);
1868                 bond_select_active_slave(bond);
1869                 break;
1870         default:
1871                 pr_debug("This slave is always active in trunk mode\n");
1872
1873                 /* always active in trunk mode */
1874                 bond_set_active_slave(new_slave);
1875
1876                 /* In trunking mode there is little meaning to curr_active_slave
1877                  * anyway (it holds no special properties of the bond device),
1878                  * so we can change it without calling change_active_interface()
1879                  */
1880                 if (!bond->curr_active_slave)
1881                         bond->curr_active_slave = new_slave;
1882
1883                 break;
1884         } /* switch(bond_mode) */
1885
1886         write_unlock_bh(&bond->curr_slave_lock);
1887
1888         bond_set_carrier(bond);
1889
1890 #ifdef CONFIG_NET_POLL_CONTROLLER
1891         slave_dev->npinfo = bond_netpoll_info(bond);
1892         if (slave_dev->npinfo) {
1893                 if (slave_enable_netpoll(new_slave)) {
1894                         read_unlock(&bond->lock);
1895                         pr_info("Error, %s: master_dev is using netpoll, "
1896                                  "but new slave device does not support netpoll.\n",
1897                                  bond_dev->name);
1898                         res = -EBUSY;
1899                         goto err_close;
1900                 }
1901         }
1902 #endif
1903
1904         read_unlock(&bond->lock);
1905
1906         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1907         if (res)
1908                 goto err_close;
1909
1910         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1911                 bond_dev->name, slave_dev->name,
1912                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1913                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1914
1915         /* enslave is successful */
1916         return 0;
1917
1918 /* Undo stages on error */
1919 err_close:
1920         dev_close(slave_dev);
1921
1922 err_unreg_rxhandler:
1923         netdev_rx_handler_unregister(slave_dev);
1924         synchronize_net();
1925
1926 err_unset_master:
1927         netdev_set_bond_master(slave_dev, NULL);
1928
1929 err_restore_mac:
1930         if (!bond->params.fail_over_mac) {
1931                 /* XXX TODO - fom follow mode needs to change master's
1932                  * MAC if this slave's MAC is in use by the bond, or at
1933                  * least print a warning.
1934                  */
1935                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1936                 addr.sa_family = slave_dev->type;
1937                 dev_set_mac_address(slave_dev, &addr);
1938         }
1939
1940 err_restore_mtu:
1941         dev_set_mtu(slave_dev, new_slave->original_mtu);
1942
1943 err_free:
1944         kfree(new_slave);
1945
1946 err_undo_flags:
1947         bond_dev->features = old_features;
1948
1949         return res;
1950 }
1951
1952 /*
1953  * Try to release the slave device <slave> from the bond device <master>
1954  * It is legal to access curr_active_slave without a lock because all the function
1955  * is write-locked.
1956  *
1957  * The rules for slave state should be:
1958  *   for Active/Backup:
1959  *     Active stays on all backups go down
1960  *   for Bonded connections:
1961  *     The first up interface should be left on and all others downed.
1962  */
1963 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1964 {
1965         struct bonding *bond = netdev_priv(bond_dev);
1966         struct slave *slave, *oldcurrent;
1967         struct sockaddr addr;
1968
1969         /* slave is not a slave or master is not master of this slave */
1970         if (!(slave_dev->flags & IFF_SLAVE) ||
1971             (slave_dev->master != bond_dev)) {
1972                 pr_err("%s: Error: cannot release %s.\n",
1973                        bond_dev->name, slave_dev->name);
1974                 return -EINVAL;
1975         }
1976
1977         block_netpoll_tx();
1978         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1979         write_lock_bh(&bond->lock);
1980
1981         slave = bond_get_slave_by_dev(bond, slave_dev);
1982         if (!slave) {
1983                 /* not a slave of this bond */
1984                 pr_info("%s: %s not enslaved\n",
1985                         bond_dev->name, slave_dev->name);
1986                 write_unlock_bh(&bond->lock);
1987                 unblock_netpoll_tx();
1988                 return -EINVAL;
1989         }
1990
1991         if (!bond->params.fail_over_mac) {
1992                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1993                     bond->slave_cnt > 1)
1994                         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",
1995                                    bond_dev->name, slave_dev->name,
1996                                    slave->perm_hwaddr,
1997                                    bond_dev->name, slave_dev->name);
1998         }
1999
2000         /* Inform AD package of unbinding of slave. */
2001         if (bond->params.mode == BOND_MODE_8023AD) {
2002                 /* must be called before the slave is
2003                  * detached from the list
2004                  */
2005                 bond_3ad_unbind_slave(slave);
2006         }
2007
2008         pr_info("%s: releasing %s interface %s\n",
2009                 bond_dev->name,
2010                 bond_is_active_slave(slave) ? "active" : "backup",
2011                 slave_dev->name);
2012
2013         oldcurrent = bond->curr_active_slave;
2014
2015         bond->current_arp_slave = NULL;
2016
2017         /* release the slave from its bond */
2018         bond_detach_slave(bond, slave);
2019
2020         bond_compute_features(bond);
2021
2022         if (bond->primary_slave == slave)
2023                 bond->primary_slave = NULL;
2024
2025         if (oldcurrent == slave)
2026                 bond_change_active_slave(bond, NULL);
2027
2028         if (bond_is_lb(bond)) {
2029                 /* Must be called only after the slave has been
2030                  * detached from the list and the curr_active_slave
2031                  * has been cleared (if our_slave == old_current),
2032                  * but before a new active slave is selected.
2033                  */
2034                 write_unlock_bh(&bond->lock);
2035                 bond_alb_deinit_slave(bond, slave);
2036                 write_lock_bh(&bond->lock);
2037         }
2038
2039         if (oldcurrent == slave) {
2040                 /*
2041                  * Note that we hold RTNL over this sequence, so there
2042                  * is no concern that another slave add/remove event
2043                  * will interfere.
2044                  */
2045                 write_unlock_bh(&bond->lock);
2046                 read_lock(&bond->lock);
2047                 write_lock_bh(&bond->curr_slave_lock);
2048
2049                 bond_select_active_slave(bond);
2050
2051                 write_unlock_bh(&bond->curr_slave_lock);
2052                 read_unlock(&bond->lock);
2053                 write_lock_bh(&bond->lock);
2054         }
2055
2056         if (bond->slave_cnt == 0) {
2057                 bond_set_carrier(bond);
2058
2059                 /* if the last slave was removed, zero the mac address
2060                  * of the master so it will be set by the application
2061                  * to the mac address of the first slave
2062                  */
2063                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2064
2065                 if (!bond->vlgrp) {
2066                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2067                 } else {
2068                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2069                                    bond_dev->name, bond_dev->name);
2070                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2071                                    bond_dev->name);
2072                 }
2073         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2074                    !bond_has_challenged_slaves(bond)) {
2075                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2076                         bond_dev->name, slave_dev->name, bond_dev->name);
2077                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2078         }
2079
2080         write_unlock_bh(&bond->lock);
2081         unblock_netpoll_tx();
2082
2083         /* must do this from outside any spinlocks */
2084         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2085
2086         bond_del_vlans_from_slave(bond, slave_dev);
2087
2088         /* If the mode USES_PRIMARY, then we should only remove its
2089          * promisc and mc settings if it was the curr_active_slave, but that was
2090          * already taken care of above when we detached the slave
2091          */
2092         if (!USES_PRIMARY(bond->params.mode)) {
2093                 /* unset promiscuity level from slave */
2094                 if (bond_dev->flags & IFF_PROMISC)
2095                         dev_set_promiscuity(slave_dev, -1);
2096
2097                 /* unset allmulti level from slave */
2098                 if (bond_dev->flags & IFF_ALLMULTI)
2099                         dev_set_allmulti(slave_dev, -1);
2100
2101                 /* flush master's mc_list from slave */
2102                 netif_addr_lock_bh(bond_dev);
2103                 bond_mc_list_flush(bond_dev, slave_dev);
2104                 netif_addr_unlock_bh(bond_dev);
2105         }
2106
2107         netdev_rx_handler_unregister(slave_dev);
2108         synchronize_net();
2109         netdev_set_bond_master(slave_dev, NULL);
2110
2111         slave_disable_netpoll(slave);
2112
2113         /* close slave before restoring its mac address */
2114         dev_close(slave_dev);
2115
2116         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2117                 /* restore original ("permanent") mac address */
2118                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2119                 addr.sa_family = slave_dev->type;
2120                 dev_set_mac_address(slave_dev, &addr);
2121         }
2122
2123         dev_set_mtu(slave_dev, slave->original_mtu);
2124
2125         slave_dev->priv_flags &= ~IFF_BONDING;
2126
2127         kfree(slave);
2128
2129         return 0;  /* deletion OK */
2130 }
2131
2132 /*
2133 * First release a slave and then destroy the bond if no more slaves are left.
2134 * Must be under rtnl_lock when this function is called.
2135 */
2136 static int  bond_release_and_destroy(struct net_device *bond_dev,
2137                                      struct net_device *slave_dev)
2138 {
2139         struct bonding *bond = netdev_priv(bond_dev);
2140         int ret;
2141
2142         ret = bond_release(bond_dev, slave_dev);
2143         if ((ret == 0) && (bond->slave_cnt == 0)) {
2144                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2145                 pr_info("%s: destroying bond %s.\n",
2146                         bond_dev->name, bond_dev->name);
2147                 unregister_netdevice(bond_dev);
2148         }
2149         return ret;
2150 }
2151
2152 /*
2153  * This function releases all slaves.
2154  */
2155 static int bond_release_all(struct net_device *bond_dev)
2156 {
2157         struct bonding *bond = netdev_priv(bond_dev);
2158         struct slave *slave;
2159         struct net_device *slave_dev;
2160         struct sockaddr addr;
2161
2162         write_lock_bh(&bond->lock);
2163
2164         netif_carrier_off(bond_dev);
2165
2166         if (bond->slave_cnt == 0)
2167                 goto out;
2168
2169         bond->current_arp_slave = NULL;
2170         bond->primary_slave = NULL;
2171         bond_change_active_slave(bond, NULL);
2172
2173         while ((slave = bond->first_slave) != NULL) {
2174                 /* Inform AD package of unbinding of slave
2175                  * before slave is detached from the list.
2176                  */
2177                 if (bond->params.mode == BOND_MODE_8023AD)
2178                         bond_3ad_unbind_slave(slave);
2179
2180                 slave_dev = slave->dev;
2181                 bond_detach_slave(bond, slave);
2182
2183                 /* now that the slave is detached, unlock and perform
2184                  * all the undo steps that should not be called from
2185                  * within a lock.
2186                  */
2187                 write_unlock_bh(&bond->lock);
2188
2189                 if (bond_is_lb(bond)) {
2190                         /* must be called only after the slave
2191                          * has been detached from the list
2192                          */
2193                         bond_alb_deinit_slave(bond, slave);
2194                 }
2195
2196                 bond_compute_features(bond);
2197
2198                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2199                 bond_del_vlans_from_slave(bond, slave_dev);
2200
2201                 /* If the mode USES_PRIMARY, then we should only remove its
2202                  * promisc and mc settings if it was the curr_active_slave, but that was
2203                  * already taken care of above when we detached the slave
2204                  */
2205                 if (!USES_PRIMARY(bond->params.mode)) {
2206                         /* unset promiscuity level from slave */
2207                         if (bond_dev->flags & IFF_PROMISC)
2208                                 dev_set_promiscuity(slave_dev, -1);
2209
2210                         /* unset allmulti level from slave */
2211                         if (bond_dev->flags & IFF_ALLMULTI)
2212                                 dev_set_allmulti(slave_dev, -1);
2213
2214                         /* flush master's mc_list from slave */
2215                         netif_addr_lock_bh(bond_dev);
2216                         bond_mc_list_flush(bond_dev, slave_dev);
2217                         netif_addr_unlock_bh(bond_dev);
2218                 }
2219
2220                 netdev_rx_handler_unregister(slave_dev);
2221                 synchronize_net();
2222                 netdev_set_bond_master(slave_dev, NULL);
2223
2224                 slave_disable_netpoll(slave);
2225
2226                 /* close slave before restoring its mac address */
2227                 dev_close(slave_dev);
2228
2229                 if (!bond->params.fail_over_mac) {
2230                         /* restore original ("permanent") mac address*/
2231                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2232                         addr.sa_family = slave_dev->type;
2233                         dev_set_mac_address(slave_dev, &addr);
2234                 }
2235
2236                 kfree(slave);
2237
2238                 /* re-acquire the lock before getting the next slave */
2239                 write_lock_bh(&bond->lock);
2240         }
2241
2242         /* zero the mac address of the master so it will be
2243          * set by the application to the mac address of the
2244          * first slave
2245          */
2246         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2247
2248         if (!bond->vlgrp) {
2249                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2250         } else {
2251                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2252                            bond_dev->name, bond_dev->name);
2253                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2254                            bond_dev->name);
2255         }
2256
2257         pr_info("%s: released all slaves\n", bond_dev->name);
2258
2259 out:
2260         write_unlock_bh(&bond->lock);
2261         return 0;
2262 }
2263
2264 /*
2265  * This function changes the active slave to slave <slave_dev>.
2266  * It returns -EINVAL in the following cases.
2267  *  - <slave_dev> is not found in the list.
2268  *  - There is not active slave now.
2269  *  - <slave_dev> is already active.
2270  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2271  *  - <slave_dev> is not running.
2272  * In these cases, this function does nothing.
2273  * In the other cases, current_slave pointer is changed and 0 is returned.
2274  */
2275 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2276 {
2277         struct bonding *bond = netdev_priv(bond_dev);
2278         struct slave *old_active = NULL;
2279         struct slave *new_active = NULL;
2280         int res = 0;
2281
2282         if (!USES_PRIMARY(bond->params.mode))
2283                 return -EINVAL;
2284
2285         /* Verify that master_dev is indeed the master of slave_dev */
2286         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2287                 return -EINVAL;
2288
2289         read_lock(&bond->lock);
2290
2291         read_lock(&bond->curr_slave_lock);
2292         old_active = bond->curr_active_slave;
2293         read_unlock(&bond->curr_slave_lock);
2294
2295         new_active = bond_get_slave_by_dev(bond, slave_dev);
2296
2297         /*
2298          * Changing to the current active: do nothing; return success.
2299          */
2300         if (new_active && (new_active == old_active)) {
2301                 read_unlock(&bond->lock);
2302                 return 0;
2303         }
2304
2305         if ((new_active) &&
2306             (old_active) &&
2307             (new_active->link == BOND_LINK_UP) &&
2308             IS_UP(new_active->dev)) {
2309                 block_netpoll_tx();
2310                 write_lock_bh(&bond->curr_slave_lock);
2311                 bond_change_active_slave(bond, new_active);
2312                 write_unlock_bh(&bond->curr_slave_lock);
2313                 unblock_netpoll_tx();
2314         } else
2315                 res = -EINVAL;
2316
2317         read_unlock(&bond->lock);
2318
2319         return res;
2320 }
2321
2322 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2323 {
2324         struct bonding *bond = netdev_priv(bond_dev);
2325
2326         info->bond_mode = bond->params.mode;
2327         info->miimon = bond->params.miimon;
2328
2329         read_lock(&bond->lock);
2330         info->num_slaves = bond->slave_cnt;
2331         read_unlock(&bond->lock);
2332
2333         return 0;
2334 }
2335
2336 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2337 {
2338         struct bonding *bond = netdev_priv(bond_dev);
2339         struct slave *slave;
2340         int i, res = -ENODEV;
2341
2342         read_lock(&bond->lock);
2343
2344         bond_for_each_slave(bond, slave, i) {
2345                 if (i == (int)info->slave_id) {
2346                         res = 0;
2347                         strcpy(info->slave_name, slave->dev->name);
2348                         info->link = slave->link;
2349                         info->state = bond_slave_state(slave);
2350                         info->link_failure_count = slave->link_failure_count;
2351                         break;
2352                 }
2353         }
2354
2355         read_unlock(&bond->lock);
2356
2357         return res;
2358 }
2359
2360 /*-------------------------------- Monitoring -------------------------------*/
2361
2362
2363 static int bond_miimon_inspect(struct bonding *bond)
2364 {
2365         struct slave *slave;
2366         int i, link_state, commit = 0;
2367         bool ignore_updelay;
2368
2369         ignore_updelay = !bond->curr_active_slave ? true : false;
2370
2371         bond_for_each_slave(bond, slave, i) {
2372                 slave->new_link = BOND_LINK_NOCHANGE;
2373
2374                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2375
2376                 switch (slave->link) {
2377                 case BOND_LINK_UP:
2378                         if (link_state)
2379                                 continue;
2380
2381                         slave->link = BOND_LINK_FAIL;
2382                         slave->delay = bond->params.downdelay;
2383                         if (slave->delay) {
2384                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2385                                         bond->dev->name,
2386                                         (bond->params.mode ==
2387                                          BOND_MODE_ACTIVEBACKUP) ?
2388                                         (bond_is_active_slave(slave) ?
2389                                          "active " : "backup ") : "",
2390                                         slave->dev->name,
2391                                         bond->params.downdelay * bond->params.miimon);
2392                         }
2393                         /*FALLTHRU*/
2394                 case BOND_LINK_FAIL:
2395                         if (link_state) {
2396                                 /*
2397                                  * recovered before downdelay expired
2398                                  */
2399                                 slave->link = BOND_LINK_UP;
2400                                 slave->jiffies = jiffies;
2401                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2402                                         bond->dev->name,
2403                                         (bond->params.downdelay - slave->delay) *
2404                                         bond->params.miimon,
2405                                         slave->dev->name);
2406                                 continue;
2407                         }
2408
2409                         if (slave->delay <= 0) {
2410                                 slave->new_link = BOND_LINK_DOWN;
2411                                 commit++;
2412                                 continue;
2413                         }
2414
2415                         slave->delay--;
2416                         break;
2417
2418                 case BOND_LINK_DOWN:
2419                         if (!link_state)
2420                                 continue;
2421
2422                         slave->link = BOND_LINK_BACK;
2423                         slave->delay = bond->params.updelay;
2424
2425                         if (slave->delay) {
2426                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2427                                         bond->dev->name, slave->dev->name,
2428                                         ignore_updelay ? 0 :
2429                                         bond->params.updelay *
2430                                         bond->params.miimon);
2431                         }
2432                         /*FALLTHRU*/
2433                 case BOND_LINK_BACK:
2434                         if (!link_state) {
2435                                 slave->link = BOND_LINK_DOWN;
2436                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2437                                         bond->dev->name,
2438                                         (bond->params.updelay - slave->delay) *
2439                                         bond->params.miimon,
2440                                         slave->dev->name);
2441
2442                                 continue;
2443                         }
2444
2445                         if (ignore_updelay)
2446                                 slave->delay = 0;
2447
2448                         if (slave->delay <= 0) {
2449                                 slave->new_link = BOND_LINK_UP;
2450                                 commit++;
2451                                 ignore_updelay = false;
2452                                 continue;
2453                         }
2454
2455                         slave->delay--;
2456                         break;
2457                 }
2458         }
2459
2460         return commit;
2461 }
2462
2463 static void bond_miimon_commit(struct bonding *bond)
2464 {
2465         struct slave *slave;
2466         int i;
2467
2468         bond_for_each_slave(bond, slave, i) {
2469                 switch (slave->new_link) {
2470                 case BOND_LINK_NOCHANGE:
2471                         continue;
2472
2473                 case BOND_LINK_UP:
2474                         slave->link = BOND_LINK_UP;
2475                         slave->jiffies = jiffies;
2476
2477                         if (bond->params.mode == BOND_MODE_8023AD) {
2478                                 /* prevent it from being the active one */
2479                                 bond_set_backup_slave(slave);
2480                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2481                                 /* make it immediately active */
2482                                 bond_set_active_slave(slave);
2483                         } else if (slave != bond->primary_slave) {
2484                                 /* prevent it from being the active one */
2485                                 bond_set_backup_slave(slave);
2486                         }
2487
2488                         bond_update_speed_duplex(slave);
2489
2490                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2491                                 bond->dev->name, slave->dev->name,
2492                                 slave->speed, slave->duplex ? "full" : "half");
2493
2494                         /* notify ad that the link status has changed */
2495                         if (bond->params.mode == BOND_MODE_8023AD)
2496                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2497
2498                         if (bond_is_lb(bond))
2499                                 bond_alb_handle_link_change(bond, slave,
2500                                                             BOND_LINK_UP);
2501
2502                         if (!bond->curr_active_slave ||
2503                             (slave == bond->primary_slave))
2504                                 goto do_failover;
2505
2506                         continue;
2507
2508                 case BOND_LINK_DOWN:
2509                         if (slave->link_failure_count < UINT_MAX)
2510                                 slave->link_failure_count++;
2511
2512                         slave->link = BOND_LINK_DOWN;
2513
2514                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2515                             bond->params.mode == BOND_MODE_8023AD)
2516                                 bond_set_slave_inactive_flags(slave);
2517
2518                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2519                                 bond->dev->name, slave->dev->name);
2520
2521                         if (bond->params.mode == BOND_MODE_8023AD)
2522                                 bond_3ad_handle_link_change(slave,
2523                                                             BOND_LINK_DOWN);
2524
2525                         if (bond_is_lb(bond))
2526                                 bond_alb_handle_link_change(bond, slave,
2527                                                             BOND_LINK_DOWN);
2528
2529                         if (slave == bond->curr_active_slave)
2530                                 goto do_failover;
2531
2532                         continue;
2533
2534                 default:
2535                         pr_err("%s: invalid new link %d on slave %s\n",
2536                                bond->dev->name, slave->new_link,
2537                                slave->dev->name);
2538                         slave->new_link = BOND_LINK_NOCHANGE;
2539
2540                         continue;
2541                 }
2542
2543 do_failover:
2544                 ASSERT_RTNL();
2545                 block_netpoll_tx();
2546                 write_lock_bh(&bond->curr_slave_lock);
2547                 bond_select_active_slave(bond);
2548                 write_unlock_bh(&bond->curr_slave_lock);
2549                 unblock_netpoll_tx();
2550         }
2551
2552         bond_set_carrier(bond);
2553 }
2554
2555 /*
2556  * bond_mii_monitor
2557  *
2558  * Really a wrapper that splits the mii monitor into two phases: an
2559  * inspection, then (if inspection indicates something needs to be done)
2560  * an acquisition of appropriate locks followed by a commit phase to
2561  * implement whatever link state changes are indicated.
2562  */
2563 void bond_mii_monitor(struct work_struct *work)
2564 {
2565         struct bonding *bond = container_of(work, struct bonding,
2566                                             mii_work.work);
2567
2568         read_lock(&bond->lock);
2569         if (bond->kill_timers)
2570                 goto out;
2571
2572         if (bond->slave_cnt == 0)
2573                 goto re_arm;
2574
2575         if (bond->send_grat_arp) {
2576                 read_lock(&bond->curr_slave_lock);
2577                 bond_send_gratuitous_arp(bond);
2578                 read_unlock(&bond->curr_slave_lock);
2579         }
2580
2581         if (bond->send_unsol_na) {
2582                 read_lock(&bond->curr_slave_lock);
2583                 bond_send_unsolicited_na(bond);
2584                 read_unlock(&bond->curr_slave_lock);
2585         }
2586
2587         if (bond_miimon_inspect(bond)) {
2588                 read_unlock(&bond->lock);
2589                 rtnl_lock();
2590                 read_lock(&bond->lock);
2591
2592                 bond_miimon_commit(bond);
2593
2594                 read_unlock(&bond->lock);
2595                 rtnl_unlock();  /* might sleep, hold no other locks */
2596                 read_lock(&bond->lock);
2597         }
2598
2599 re_arm:
2600         if (bond->params.miimon)
2601                 queue_delayed_work(bond->wq, &bond->mii_work,
2602                                    msecs_to_jiffies(bond->params.miimon));
2603 out:
2604         read_unlock(&bond->lock);
2605 }
2606
2607 static __be32 bond_glean_dev_ip(struct net_device *dev)
2608 {
2609         struct in_device *idev;
2610         struct in_ifaddr *ifa;
2611         __be32 addr = 0;
2612
2613         if (!dev)
2614                 return 0;
2615
2616         rcu_read_lock();
2617         idev = __in_dev_get_rcu(dev);
2618         if (!idev)
2619                 goto out;
2620
2621         ifa = idev->ifa_list;
2622         if (!ifa)
2623                 goto out;
2624
2625         addr = ifa->ifa_local;
2626 out:
2627         rcu_read_unlock();
2628         return addr;
2629 }
2630
2631 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2632 {
2633         struct vlan_entry *vlan;
2634
2635         if (ip == bond->master_ip)
2636                 return 1;
2637
2638         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2639                 if (ip == vlan->vlan_ip)
2640                         return 1;
2641         }
2642
2643         return 0;
2644 }
2645
2646 /*
2647  * We go to the (large) trouble of VLAN tagging ARP frames because
2648  * switches in VLAN mode (especially if ports are configured as
2649  * "native" to a VLAN) might not pass non-tagged frames.
2650  */
2651 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2652 {
2653         struct sk_buff *skb;
2654
2655         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2656                  slave_dev->name, dest_ip, src_ip, vlan_id);
2657
2658         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2659                          NULL, slave_dev->dev_addr, NULL);
2660
2661         if (!skb) {
2662                 pr_err("ARP packet allocation failed\n");
2663                 return;
2664         }
2665         if (vlan_id) {
2666                 skb = vlan_put_tag(skb, vlan_id);
2667                 if (!skb) {
2668                         pr_err("failed to insert VLAN tag\n");
2669                         return;
2670                 }
2671         }
2672         arp_xmit(skb);
2673 }
2674
2675
2676 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2677 {
2678         int i, vlan_id;
2679         __be32 *targets = bond->params.arp_targets;
2680         struct vlan_entry *vlan;
2681         struct net_device *vlan_dev;
2682         struct rtable *rt;
2683
2684         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2685                 if (!targets[i])
2686                         break;
2687                 pr_debug("basa: target %x\n", targets[i]);
2688                 if (!bond->vlgrp) {
2689                         pr_debug("basa: empty vlan: arp_send\n");
2690                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2691                                       bond->master_ip, 0);
2692                         continue;
2693                 }
2694
2695                 /*
2696                  * If VLANs are configured, we do a route lookup to
2697                  * determine which VLAN interface would be used, so we
2698                  * can tag the ARP with the proper VLAN tag.
2699                  */
2700                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2701                                      RTO_ONLINK, 0);
2702                 if (IS_ERR(rt)) {
2703                         if (net_ratelimit()) {
2704                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2705                                            bond->dev->name, &targets[i]);
2706                         }
2707                         continue;
2708                 }
2709
2710                 /*
2711                  * This target is not on a VLAN
2712                  */
2713                 if (rt->dst.dev == bond->dev) {
2714                         ip_rt_put(rt);
2715                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2716                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2717                                       bond->master_ip, 0);
2718                         continue;
2719                 }
2720
2721                 vlan_id = 0;
2722                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2723                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2724                         if (vlan_dev == rt->dst.dev) {
2725                                 vlan_id = vlan->vlan_id;
2726                                 pr_debug("basa: vlan match on %s %d\n",
2727                                        vlan_dev->name, vlan_id);
2728                                 break;
2729                         }
2730                 }
2731
2732                 if (vlan_id) {
2733                         ip_rt_put(rt);
2734                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2735                                       vlan->vlan_ip, vlan_id);
2736                         continue;
2737                 }
2738
2739                 if (net_ratelimit()) {
2740                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2741                                    bond->dev->name, &targets[i],
2742                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2743                 }
2744                 ip_rt_put(rt);
2745         }
2746 }
2747
2748 /*
2749  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2750  * for each VLAN above us.
2751  *
2752  * Caller must hold curr_slave_lock for read or better
2753  */
2754 static void bond_send_gratuitous_arp(struct bonding *bond)
2755 {
2756         struct slave *slave = bond->curr_active_slave;
2757         struct vlan_entry *vlan;
2758         struct net_device *vlan_dev;
2759
2760         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2761                  bond->dev->name, slave ? slave->dev->name : "NULL");
2762
2763         if (!slave || !bond->send_grat_arp ||
2764             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2765                 return;
2766
2767         bond->send_grat_arp--;
2768
2769         if (bond->master_ip) {
2770                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2771                                 bond->master_ip, 0);
2772         }
2773
2774         if (!bond->vlgrp)
2775                 return;
2776
2777         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2778                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2779                 if (vlan->vlan_ip) {
2780                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2781                                       vlan->vlan_ip, vlan->vlan_id);
2782                 }
2783         }
2784 }
2785
2786 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2787 {
2788         int i;
2789         __be32 *targets = bond->params.arp_targets;
2790
2791         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2792                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2793                          &sip, &tip, i, &targets[i],
2794                          bond_has_this_ip(bond, tip));
2795                 if (sip == targets[i]) {
2796                         if (bond_has_this_ip(bond, tip))
2797                                 slave->last_arp_rx = jiffies;
2798                         return;
2799                 }
2800         }
2801 }
2802
2803 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2804 {
2805         struct arphdr *arp;
2806         struct slave *slave;
2807         struct bonding *bond;
2808         unsigned char *arp_ptr;
2809         __be32 sip, tip;
2810
2811         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2812                 /*
2813                  * When using VLANS and bonding, dev and oriv_dev may be
2814                  * incorrect if the physical interface supports VLAN
2815                  * acceleration.  With this change ARP validation now
2816                  * works for hosts only reachable on the VLAN interface.
2817                  */
2818                 dev = vlan_dev_real_dev(dev);
2819                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2820         }
2821
2822         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2823                 goto out;
2824
2825         bond = netdev_priv(dev);
2826         read_lock(&bond->lock);
2827
2828         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2829                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2830                  orig_dev ? orig_dev->name : "NULL");
2831
2832         slave = bond_get_slave_by_dev(bond, orig_dev);
2833         if (!slave || !slave_do_arp_validate(bond, slave))
2834                 goto out_unlock;
2835
2836         skb = skb_share_check(skb, GFP_ATOMIC);
2837         if (!skb)
2838                 goto out_unlock;
2839
2840         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2841                 goto out_unlock;
2842
2843         arp = arp_hdr(skb);
2844         if (arp->ar_hln != dev->addr_len ||
2845             skb->pkt_type == PACKET_OTHERHOST ||
2846             skb->pkt_type == PACKET_LOOPBACK ||
2847             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2848             arp->ar_pro != htons(ETH_P_IP) ||
2849             arp->ar_pln != 4)
2850                 goto out_unlock;
2851
2852         arp_ptr = (unsigned char *)(arp + 1);
2853         arp_ptr += dev->addr_len;
2854         memcpy(&sip, arp_ptr, 4);
2855         arp_ptr += 4 + dev->addr_len;
2856         memcpy(&tip, arp_ptr, 4);
2857
2858         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2859                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2860                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2861                  &sip, &tip);
2862
2863         /*
2864          * Backup slaves won't see the ARP reply, but do come through
2865          * here for each ARP probe (so we swap the sip/tip to validate
2866          * the probe).  In a "redundant switch, common router" type of
2867          * configuration, the ARP probe will (hopefully) travel from
2868          * the active, through one switch, the router, then the other
2869          * switch before reaching the backup.
2870          */
2871         if (bond_is_active_slave(slave))
2872                 bond_validate_arp(bond, slave, sip, tip);
2873         else
2874                 bond_validate_arp(bond, slave, tip, sip);
2875
2876 out_unlock:
2877         read_unlock(&bond->lock);
2878 out:
2879         dev_kfree_skb(skb);
2880         return NET_RX_SUCCESS;
2881 }
2882
2883 /*
2884  * this function is called regularly to monitor each slave's link
2885  * ensuring that traffic is being sent and received when arp monitoring
2886  * is used in load-balancing mode. if the adapter has been dormant, then an
2887  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2888  * arp monitoring in active backup mode.
2889  */
2890 void bond_loadbalance_arp_mon(struct work_struct *work)
2891 {
2892         struct bonding *bond = container_of(work, struct bonding,
2893                                             arp_work.work);
2894         struct slave *slave, *oldcurrent;
2895         int do_failover = 0;
2896         int delta_in_ticks;
2897         int i;
2898
2899         read_lock(&bond->lock);
2900
2901         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2902
2903         if (bond->kill_timers)
2904                 goto out;
2905
2906         if (bond->slave_cnt == 0)
2907                 goto re_arm;
2908
2909         read_lock(&bond->curr_slave_lock);
2910         oldcurrent = bond->curr_active_slave;
2911         read_unlock(&bond->curr_slave_lock);
2912
2913         /* see if any of the previous devices are up now (i.e. they have
2914          * xmt and rcv traffic). the curr_active_slave does not come into
2915          * the picture unless it is null. also, slave->jiffies is not needed
2916          * here because we send an arp on each slave and give a slave as
2917          * long as it needs to get the tx/rx within the delta.
2918          * TODO: what about up/down delay in arp mode? it wasn't here before
2919          *       so it can wait
2920          */
2921         bond_for_each_slave(bond, slave, i) {
2922                 unsigned long trans_start = dev_trans_start(slave->dev);
2923
2924                 if (slave->link != BOND_LINK_UP) {
2925                         if (time_in_range(jiffies,
2926                                 trans_start - delta_in_ticks,
2927                                 trans_start + delta_in_ticks) &&
2928                             time_in_range(jiffies,
2929                                 slave->dev->last_rx - delta_in_ticks,
2930                                 slave->dev->last_rx + delta_in_ticks)) {
2931
2932                                 slave->link  = BOND_LINK_UP;
2933                                 bond_set_active_slave(slave);
2934
2935                                 /* primary_slave has no meaning in round-robin
2936                                  * mode. the window of a slave being up and
2937                                  * curr_active_slave being null after enslaving
2938                                  * is closed.
2939                                  */
2940                                 if (!oldcurrent) {
2941                                         pr_info("%s: link status definitely up for interface %s, ",
2942                                                 bond->dev->name,
2943                                                 slave->dev->name);
2944                                         do_failover = 1;
2945                                 } else {
2946                                         pr_info("%s: interface %s is now up\n",
2947                                                 bond->dev->name,
2948                                                 slave->dev->name);
2949                                 }
2950                         }
2951                 } else {
2952                         /* slave->link == BOND_LINK_UP */
2953
2954                         /* not all switches will respond to an arp request
2955                          * when the source ip is 0, so don't take the link down
2956                          * if we don't know our ip yet
2957                          */
2958                         if (!time_in_range(jiffies,
2959                                 trans_start - delta_in_ticks,
2960                                 trans_start + 2 * delta_in_ticks) ||
2961                             !time_in_range(jiffies,
2962                                 slave->dev->last_rx - delta_in_ticks,
2963                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2964
2965                                 slave->link  = BOND_LINK_DOWN;
2966                                 bond_set_backup_slave(slave);
2967
2968                                 if (slave->link_failure_count < UINT_MAX)
2969                                         slave->link_failure_count++;
2970
2971                                 pr_info("%s: interface %s is now down.\n",
2972                                         bond->dev->name,
2973                                         slave->dev->name);
2974
2975                                 if (slave == oldcurrent)
2976                                         do_failover = 1;
2977                         }
2978                 }
2979
2980                 /* note: if switch is in round-robin mode, all links
2981                  * must tx arp to ensure all links rx an arp - otherwise
2982                  * links may oscillate or not come up at all; if switch is
2983                  * in something like xor mode, there is nothing we can
2984                  * do - all replies will be rx'ed on same link causing slaves
2985                  * to be unstable during low/no traffic periods
2986                  */
2987                 if (IS_UP(slave->dev))
2988                         bond_arp_send_all(bond, slave);
2989         }
2990
2991         if (do_failover) {
2992                 block_netpoll_tx();
2993                 write_lock_bh(&bond->curr_slave_lock);
2994
2995                 bond_select_active_slave(bond);
2996
2997                 write_unlock_bh(&bond->curr_slave_lock);
2998                 unblock_netpoll_tx();
2999         }
3000
3001 re_arm:
3002         if (bond->params.arp_interval)
3003                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3004 out:
3005         read_unlock(&bond->lock);
3006 }
3007
3008 /*
3009  * Called to inspect slaves for active-backup mode ARP monitor link state
3010  * changes.  Sets new_link in slaves to specify what action should take
3011  * place for the slave.  Returns 0 if no changes are found, >0 if changes
3012  * to link states must be committed.
3013  *
3014  * Called with bond->lock held for read.
3015  */
3016 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
3017 {
3018         struct slave *slave;
3019         int i, commit = 0;
3020         unsigned long trans_start;
3021
3022         bond_for_each_slave(bond, slave, i) {
3023                 slave->new_link = BOND_LINK_NOCHANGE;
3024
3025                 if (slave->link != BOND_LINK_UP) {
3026                         if (time_in_range(jiffies,
3027                                 slave_last_rx(bond, slave) - delta_in_ticks,
3028                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
3029
3030                                 slave->new_link = BOND_LINK_UP;
3031                                 commit++;
3032                         }
3033
3034                         continue;
3035                 }
3036
3037                 /*
3038                  * Give slaves 2*delta after being enslaved or made
3039                  * active.  This avoids bouncing, as the last receive
3040                  * times need a full ARP monitor cycle to be updated.
3041                  */
3042                 if (time_in_range(jiffies,
3043                                   slave->jiffies - delta_in_ticks,
3044                                   slave->jiffies + 2 * delta_in_ticks))
3045                         continue;
3046
3047                 /*
3048                  * Backup slave is down if:
3049                  * - No current_arp_slave AND
3050                  * - more than 3*delta since last receive AND
3051                  * - the bond has an IP address
3052                  *
3053                  * Note: a non-null current_arp_slave indicates
3054                  * the curr_active_slave went down and we are
3055                  * searching for a new one; under this condition
3056                  * we only take the curr_active_slave down - this
3057                  * gives each slave a chance to tx/rx traffic
3058                  * before being taken out
3059                  */
3060                 if (!bond_is_active_slave(slave) &&
3061                     !bond->current_arp_slave &&
3062                     !time_in_range(jiffies,
3063                         slave_last_rx(bond, slave) - delta_in_ticks,
3064                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3065
3066                         slave->new_link = BOND_LINK_DOWN;
3067                         commit++;
3068                 }
3069
3070                 /*
3071                  * Active slave is down if:
3072                  * - more than 2*delta since transmitting OR
3073                  * - (more than 2*delta since receive AND
3074                  *    the bond has an IP address)
3075                  */
3076                 trans_start = dev_trans_start(slave->dev);
3077                 if (bond_is_active_slave(slave) &&
3078                     (!time_in_range(jiffies,
3079                         trans_start - delta_in_ticks,
3080                         trans_start + 2 * delta_in_ticks) ||
3081                      !time_in_range(jiffies,
3082                         slave_last_rx(bond, slave) - delta_in_ticks,
3083                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3084
3085                         slave->new_link = BOND_LINK_DOWN;
3086                         commit++;
3087                 }
3088         }
3089
3090         return commit;
3091 }
3092
3093 /*
3094  * Called to commit link state changes noted by inspection step of
3095  * active-backup mode ARP monitor.
3096  *
3097  * Called with RTNL and bond->lock for read.
3098  */
3099 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3100 {
3101         struct slave *slave;
3102         int i;
3103         unsigned long trans_start;
3104
3105         bond_for_each_slave(bond, slave, i) {
3106                 switch (slave->new_link) {
3107                 case BOND_LINK_NOCHANGE:
3108                         continue;
3109
3110                 case BOND_LINK_UP:
3111                         trans_start = dev_trans_start(slave->dev);
3112                         if ((!bond->curr_active_slave &&
3113                              time_in_range(jiffies,
3114                                            trans_start - delta_in_ticks,
3115                                            trans_start + delta_in_ticks)) ||
3116                             bond->curr_active_slave != slave) {
3117                                 slave->link = BOND_LINK_UP;
3118                                 bond->current_arp_slave = NULL;
3119
3120                                 pr_info("%s: link status definitely up for interface %s.\n",
3121                                         bond->dev->name, slave->dev->name);
3122
3123                                 if (!bond->curr_active_slave ||
3124                                     (slave == bond->primary_slave))
3125                                         goto do_failover;
3126
3127                         }
3128
3129                         continue;
3130
3131                 case BOND_LINK_DOWN:
3132                         if (slave->link_failure_count < UINT_MAX)
3133                                 slave->link_failure_count++;
3134
3135                         slave->link = BOND_LINK_DOWN;
3136                         bond_set_slave_inactive_flags(slave);
3137
3138                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3139                                 bond->dev->name, slave->dev->name);
3140
3141                         if (slave == bond->curr_active_slave) {
3142                                 bond->current_arp_slave = NULL;
3143                                 goto do_failover;
3144                         }
3145
3146                         continue;
3147
3148                 default:
3149                         pr_err("%s: impossible: new_link %d on slave %s\n",
3150                                bond->dev->name, slave->new_link,
3151                                slave->dev->name);
3152                         continue;
3153                 }
3154
3155 do_failover:
3156                 ASSERT_RTNL();
3157                 block_netpoll_tx();
3158                 write_lock_bh(&bond->curr_slave_lock);
3159                 bond_select_active_slave(bond);
3160                 write_unlock_bh(&bond->curr_slave_lock);
3161                 unblock_netpoll_tx();
3162         }
3163
3164         bond_set_carrier(bond);
3165 }
3166
3167 /*
3168  * Send ARP probes for active-backup mode ARP monitor.
3169  *
3170  * Called with bond->lock held for read.
3171  */
3172 static void bond_ab_arp_probe(struct bonding *bond)
3173 {
3174         struct slave *slave;
3175         int i;
3176
3177         read_lock(&bond->curr_slave_lock);
3178
3179         if (bond->current_arp_slave && bond->curr_active_slave)
3180                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3181                         bond->current_arp_slave->dev->name,
3182                         bond->curr_active_slave->dev->name);
3183
3184         if (bond->curr_active_slave) {
3185                 bond_arp_send_all(bond, bond->curr_active_slave);
3186                 read_unlock(&bond->curr_slave_lock);
3187                 return;
3188         }
3189
3190         read_unlock(&bond->curr_slave_lock);
3191
3192         /* if we don't have a curr_active_slave, search for the next available
3193          * backup slave from the current_arp_slave and make it the candidate
3194          * for becoming the curr_active_slave
3195          */
3196
3197         if (!bond->current_arp_slave) {
3198                 bond->current_arp_slave = bond->first_slave;
3199                 if (!bond->current_arp_slave)
3200                         return;
3201         }
3202
3203         bond_set_slave_inactive_flags(bond->current_arp_slave);
3204
3205         /* search for next candidate */
3206         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3207                 if (IS_UP(slave->dev)) {
3208                         slave->link = BOND_LINK_BACK;
3209                         bond_set_slave_active_flags(slave);
3210                         bond_arp_send_all(bond, slave);
3211                         slave->jiffies = jiffies;
3212                         bond->current_arp_slave = slave;
3213                         break;
3214                 }
3215
3216                 /* if the link state is up at this point, we
3217                  * mark it down - this can happen if we have
3218                  * simultaneous link failures and
3219                  * reselect_active_interface doesn't make this
3220                  * one the current slave so it is still marked
3221                  * up when it is actually down
3222                  */
3223                 if (slave->link == BOND_LINK_UP) {
3224                         slave->link = BOND_LINK_DOWN;
3225                         if (slave->link_failure_count < UINT_MAX)
3226                                 slave->link_failure_count++;
3227
3228                         bond_set_slave_inactive_flags(slave);
3229
3230                         pr_info("%s: backup interface %s is now down.\n",
3231                                 bond->dev->name, slave->dev->name);
3232                 }
3233         }
3234 }
3235
3236 void bond_activebackup_arp_mon(struct work_struct *work)
3237 {
3238         struct bonding *bond = container_of(work, struct bonding,
3239                                             arp_work.work);
3240         int delta_in_ticks;
3241
3242         read_lock(&bond->lock);
3243
3244         if (bond->kill_timers)
3245                 goto out;
3246
3247         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3248
3249         if (bond->slave_cnt == 0)
3250                 goto re_arm;
3251
3252         if (bond->send_grat_arp) {
3253                 read_lock(&bond->curr_slave_lock);
3254                 bond_send_gratuitous_arp(bond);
3255                 read_unlock(&bond->curr_slave_lock);
3256         }
3257
3258         if (bond->send_unsol_na) {
3259                 read_lock(&bond->curr_slave_lock);
3260                 bond_send_unsolicited_na(bond);
3261                 read_unlock(&bond->curr_slave_lock);
3262         }
3263
3264         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3265                 read_unlock(&bond->lock);
3266                 rtnl_lock();
3267                 read_lock(&bond->lock);
3268
3269                 bond_ab_arp_commit(bond, delta_in_ticks);
3270
3271                 read_unlock(&bond->lock);
3272                 rtnl_unlock();
3273                 read_lock(&bond->lock);
3274         }
3275
3276         bond_ab_arp_probe(bond);
3277
3278 re_arm:
3279         if (bond->params.arp_interval)
3280                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3281 out:
3282         read_unlock(&bond->lock);
3283 }
3284
3285 /*-------------------------- netdev event handling --------------------------*/
3286
3287 /*
3288  * Change device name
3289  */
3290 static int bond_event_changename(struct bonding *bond)
3291 {
3292         bond_remove_proc_entry(bond);
3293         bond_create_proc_entry(bond);
3294
3295         bond_debug_reregister(bond);
3296
3297         return NOTIFY_DONE;
3298 }
3299
3300 static int bond_master_netdev_event(unsigned long event,
3301                                     struct net_device *bond_dev)
3302 {
3303         struct bonding *event_bond = netdev_priv(bond_dev);
3304
3305         switch (event) {
3306         case NETDEV_CHANGENAME:
3307                 return bond_event_changename(event_bond);
3308         default:
3309                 break;
3310         }
3311
3312         return NOTIFY_DONE;
3313 }
3314
3315 static int bond_slave_netdev_event(unsigned long event,
3316                                    struct net_device *slave_dev)
3317 {
3318         struct net_device *bond_dev = slave_dev->master;
3319         struct bonding *bond = netdev_priv(bond_dev);
3320
3321         switch (event) {
3322         case NETDEV_UNREGISTER:
3323                 if (bond_dev) {
3324                         if (bond->setup_by_slave)
3325                                 bond_release_and_destroy(bond_dev, slave_dev);
3326                         else
3327                                 bond_release(bond_dev, slave_dev);
3328                 }
3329                 break;
3330         case NETDEV_CHANGE:
3331                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3332                         struct slave *slave;
3333
3334                         slave = bond_get_slave_by_dev(bond, slave_dev);
3335                         if (slave) {
3336                                 u16 old_speed = slave->speed;
3337                                 u16 old_duplex = slave->duplex;
3338
3339                                 bond_update_speed_duplex(slave);
3340
3341                                 if (bond_is_lb(bond))
3342                                         break;
3343
3344                                 if (old_speed != slave->speed)
3345                                         bond_3ad_adapter_speed_changed(slave);
3346                                 if (old_duplex != slave->duplex)
3347                                         bond_3ad_adapter_duplex_changed(slave);
3348                         }
3349                 }
3350
3351                 break;
3352         case NETDEV_DOWN:
3353                 /*
3354                  * ... Or is it this?
3355                  */
3356                 break;
3357         case NETDEV_CHANGEMTU:
3358                 /*
3359                  * TODO: Should slaves be allowed to
3360                  * independently alter their MTU?  For
3361                  * an active-backup bond, slaves need
3362                  * not be the same type of device, so
3363                  * MTUs may vary.  For other modes,
3364                  * slaves arguably should have the
3365                  * same MTUs. To do this, we'd need to
3366                  * take over the slave's change_mtu
3367                  * function for the duration of their
3368                  * servitude.
3369                  */
3370                 break;
3371         case NETDEV_CHANGENAME:
3372                 /*
3373                  * TODO: handle changing the primary's name
3374                  */
3375                 break;
3376         case NETDEV_FEAT_CHANGE:
3377                 bond_compute_features(bond);
3378                 break;
3379         default:
3380                 break;
3381         }
3382
3383         return NOTIFY_DONE;
3384 }
3385
3386 /*
3387  * bond_netdev_event: handle netdev notifier chain events.
3388  *
3389  * This function receives events for the netdev chain.  The caller (an
3390  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3391  * locks for us to safely manipulate the slave devices (RTNL lock,
3392  * dev_probe_lock).
3393  */
3394 static int bond_netdev_event(struct notifier_block *this,
3395                              unsigned long event, void *ptr)
3396 {
3397         struct net_device *event_dev = (struct net_device *)ptr;
3398
3399         pr_debug("event_dev: %s, event: %lx\n",
3400                  event_dev ? event_dev->name : "None",
3401                  event);
3402
3403         if (!(event_dev->priv_flags & IFF_BONDING))
3404                 return NOTIFY_DONE;
3405
3406         if (event_dev->flags & IFF_MASTER) {
3407                 pr_debug("IFF_MASTER\n");
3408                 return bond_master_netdev_event(event, event_dev);
3409         }
3410
3411         if (event_dev->flags & IFF_SLAVE) {
3412                 pr_debug("IFF_SLAVE\n");
3413                 return bond_slave_netdev_event(event, event_dev);
3414         }
3415
3416         return NOTIFY_DONE;
3417 }
3418
3419 /*
3420  * bond_inetaddr_event: handle inetaddr notifier chain events.
3421  *
3422  * We keep track of device IPs primarily to use as source addresses in
3423  * ARP monitor probes (rather than spewing out broadcasts all the time).
3424  *
3425  * We track one IP for the main device (if it has one), plus one per VLAN.
3426  */
3427 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3428 {
3429         struct in_ifaddr *ifa = ptr;
3430         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3431         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3432         struct bonding *bond;
3433         struct vlan_entry *vlan;
3434
3435         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3436                 if (bond->dev == event_dev) {
3437                         switch (event) {
3438                         case NETDEV_UP:
3439                                 bond->master_ip = ifa->ifa_local;
3440                                 return NOTIFY_OK;
3441                         case NETDEV_DOWN:
3442                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3443                                 return NOTIFY_OK;
3444                         default:
3445                                 return NOTIFY_DONE;
3446                         }
3447                 }
3448
3449                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3450                         if (!bond->vlgrp)
3451                                 continue;
3452                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3453                         if (vlan_dev == event_dev) {
3454                                 switch (event) {
3455                                 case NETDEV_UP:
3456                                         vlan->vlan_ip = ifa->ifa_local;
3457                                         return NOTIFY_OK;
3458                                 case NETDEV_DOWN:
3459                                         vlan->vlan_ip =
3460                                                 bond_glean_dev_ip(vlan_dev);
3461                                         return NOTIFY_OK;
3462                                 default:
3463                                         return NOTIFY_DONE;
3464                                 }
3465                         }
3466                 }
3467         }
3468         return NOTIFY_DONE;
3469 }
3470
3471 static struct notifier_block bond_netdev_notifier = {
3472         .notifier_call = bond_netdev_event,
3473 };
3474
3475 static struct notifier_block bond_inetaddr_notifier = {
3476         .notifier_call = bond_inetaddr_event,
3477 };
3478
3479 /*-------------------------- Packet type handling ---------------------------*/
3480
3481 /* register to receive lacpdus on a bond */
3482 static void bond_register_lacpdu(struct bonding *bond)
3483 {
3484         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3485
3486         /* initialize packet type */
3487         pk_type->type = PKT_TYPE_LACPDU;
3488         pk_type->dev = bond->dev;
3489         pk_type->func = bond_3ad_lacpdu_recv;
3490
3491         dev_add_pack(pk_type);
3492 }
3493
3494 /* unregister to receive lacpdus on a bond */
3495 static void bond_unregister_lacpdu(struct bonding *bond)
3496 {
3497         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3498 }
3499
3500 void bond_register_arp(struct bonding *bond)
3501 {
3502         struct packet_type *pt = &bond->arp_mon_pt;
3503
3504         if (pt->type)
3505                 return;
3506
3507         pt->type = htons(ETH_P_ARP);
3508         pt->dev = bond->dev;
3509         pt->func = bond_arp_rcv;
3510         dev_add_pack(pt);
3511 }
3512
3513 void bond_unregister_arp(struct bonding *bond)
3514 {
3515         struct packet_type *pt = &bond->arp_mon_pt;
3516
3517         dev_remove_pack(pt);
3518         pt->type = 0;
3519 }
3520
3521 /*---------------------------- Hashing Policies -----------------------------*/
3522
3523 /*
3524  * Hash for the output device based upon layer 2 and layer 3 data. If
3525  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3526  */
3527 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3528 {
3529         struct ethhdr *data = (struct ethhdr *)skb->data;
3530         struct iphdr *iph = ip_hdr(skb);
3531
3532         if (skb->protocol == htons(ETH_P_IP)) {
3533                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3534                         (data->h_dest[5] ^ data->h_source[5])) % count;
3535         }
3536
3537         return (data->h_dest[5] ^ data->h_source[5]) % count;
3538 }
3539
3540 /*
3541  * Hash for the output device based upon layer 3 and layer 4 data. If
3542  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3543  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3544  */
3545 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3546 {
3547         struct ethhdr *data = (struct ethhdr *)skb->data;
3548         struct iphdr *iph = ip_hdr(skb);
3549         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3550         int layer4_xor = 0;
3551
3552         if (skb->protocol == htons(ETH_P_IP)) {
3553                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3554                     (iph->protocol == IPPROTO_TCP ||
3555                      iph->protocol == IPPROTO_UDP)) {
3556                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3557                 }
3558                 return (layer4_xor ^
3559                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3560
3561         }
3562
3563         return (data->h_dest[5] ^ data->h_source[5]) % count;
3564 }
3565
3566 /*
3567  * Hash for the output device based upon layer 2 data
3568  */
3569 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3570 {
3571         struct ethhdr *data = (struct ethhdr *)skb->data;
3572
3573         return (data->h_dest[5] ^ data->h_source[5]) % count;
3574 }
3575
3576 /*-------------------------- Device entry points ----------------------------*/
3577
3578 static int bond_open(struct net_device *bond_dev)
3579 {
3580         struct bonding *bond = netdev_priv(bond_dev);
3581
3582         bond->kill_timers = 0;
3583
3584         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3585
3586         if (bond_is_lb(bond)) {
3587                 /* bond_alb_initialize must be called before the timer
3588                  * is started.
3589                  */
3590                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3591                         /* something went wrong - fail the open operation */
3592                         return -ENOMEM;
3593                 }
3594
3595                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3596                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3597         }
3598
3599         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3600                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3601                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3602         }
3603
3604         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3605                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3606                         INIT_DELAYED_WORK(&bond->arp_work,
3607                                           bond_activebackup_arp_mon);
3608                 else
3609                         INIT_DELAYED_WORK(&bond->arp_work,
3610                                           bond_loadbalance_arp_mon);
3611
3612                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3613                 if (bond->params.arp_validate)
3614                         bond_register_arp(bond);
3615         }
3616
3617         if (bond->params.mode == BOND_MODE_8023AD) {
3618                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3619                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3620                 /* register to receive LACPDUs */
3621                 bond_register_lacpdu(bond);
3622                 bond_3ad_initiate_agg_selection(bond, 1);
3623         }
3624
3625         return 0;
3626 }
3627
3628 static int bond_close(struct net_device *bond_dev)
3629 {
3630         struct bonding *bond = netdev_priv(bond_dev);
3631
3632         if (bond->params.mode == BOND_MODE_8023AD) {
3633                 /* Unregister the receive of LACPDUs */
3634                 bond_unregister_lacpdu(bond);
3635         }
3636
3637         if (bond->params.arp_validate)
3638                 bond_unregister_arp(bond);
3639
3640         write_lock_bh(&bond->lock);
3641
3642         bond->send_grat_arp = 0;
3643         bond->send_unsol_na = 0;
3644
3645         /* signal timers not to re-arm */
3646         bond->kill_timers = 1;
3647
3648         write_unlock_bh(&bond->lock);
3649
3650         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3651                 cancel_delayed_work(&bond->mii_work);
3652         }
3653
3654         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3655                 cancel_delayed_work(&bond->arp_work);
3656         }
3657
3658         switch (bond->params.mode) {
3659         case BOND_MODE_8023AD:
3660                 cancel_delayed_work(&bond->ad_work);
3661                 break;
3662         case BOND_MODE_TLB:
3663         case BOND_MODE_ALB:
3664                 cancel_delayed_work(&bond->alb_work);
3665                 break;
3666         default:
3667                 break;
3668         }
3669
3670         if (delayed_work_pending(&bond->mcast_work))
3671                 cancel_delayed_work(&bond->mcast_work);
3672
3673         if (bond_is_lb(bond)) {
3674                 /* Must be called only after all
3675                  * slaves have been released
3676                  */
3677                 bond_alb_deinitialize(bond);
3678         }
3679
3680         return 0;
3681 }
3682
3683 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3684                                                 struct rtnl_link_stats64 *stats)
3685 {
3686         struct bonding *bond = netdev_priv(bond_dev);
3687         struct rtnl_link_stats64 temp;
3688         struct slave *slave;
3689         int i;
3690
3691         memset(stats, 0, sizeof(*stats));
3692
3693         read_lock_bh(&bond->lock);
3694
3695         bond_for_each_slave(bond, slave, i) {
3696                 const struct rtnl_link_stats64 *sstats =
3697                         dev_get_stats(slave->dev, &temp);
3698
3699                 stats->rx_packets += sstats->rx_packets;
3700                 stats->rx_bytes += sstats->rx_bytes;
3701                 stats->rx_errors += sstats->rx_errors;
3702                 stats->rx_dropped += sstats->rx_dropped;
3703
3704                 stats->tx_packets += sstats->tx_packets;
3705                 stats->tx_bytes += sstats->tx_bytes;
3706                 stats->tx_errors += sstats->tx_errors;
3707                 stats->tx_dropped += sstats->tx_dropped;
3708
3709                 stats->multicast += sstats->multicast;
3710                 stats->collisions += sstats->collisions;
3711
3712                 stats->rx_length_errors += sstats->rx_length_errors;
3713                 stats->rx_over_errors += sstats->rx_over_errors;
3714                 stats->rx_crc_errors += sstats->rx_crc_errors;
3715                 stats->rx_frame_errors += sstats->rx_frame_errors;
3716                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3717                 stats->rx_missed_errors += sstats->rx_missed_errors;
3718
3719                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3720                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3721                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3722                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3723                 stats->tx_window_errors += sstats->tx_window_errors;
3724         }
3725
3726         read_unlock_bh(&bond->lock);
3727
3728         return stats;
3729 }
3730
3731 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3732 {
3733         struct net_device *slave_dev = NULL;
3734         struct ifbond k_binfo;
3735         struct ifbond __user *u_binfo = NULL;
3736         struct ifslave k_sinfo;
3737         struct ifslave __user *u_sinfo = NULL;
3738         struct mii_ioctl_data *mii = NULL;
3739         int res = 0;
3740
3741         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3742
3743         switch (cmd) {
3744         case SIOCGMIIPHY:
3745                 mii = if_mii(ifr);
3746                 if (!mii)
3747                         return -EINVAL;
3748
3749                 mii->phy_id = 0;
3750                 /* Fall Through */
3751         case SIOCGMIIREG:
3752                 /*
3753                  * We do this again just in case we were called by SIOCGMIIREG
3754                  * instead of SIOCGMIIPHY.
3755                  */
3756                 mii = if_mii(ifr);
3757                 if (!mii)
3758                         return -EINVAL;
3759
3760
3761                 if (mii->reg_num == 1) {
3762                         struct bonding *bond = netdev_priv(bond_dev);
3763                         mii->val_out = 0;
3764                         read_lock(&bond->lock);
3765                         read_lock(&bond->curr_slave_lock);
3766                         if (netif_carrier_ok(bond->dev))
3767                                 mii->val_out = BMSR_LSTATUS;
3768
3769                         read_unlock(&bond->curr_slave_lock);
3770                         read_unlock(&bond->lock);
3771                 }
3772
3773                 return 0;
3774         case BOND_INFO_QUERY_OLD:
3775         case SIOCBONDINFOQUERY:
3776                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3777
3778                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3779                         return -EFAULT;
3780
3781                 res = bond_info_query(bond_dev, &k_binfo);
3782                 if (res == 0 &&
3783                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3784                         return -EFAULT;
3785
3786                 return res;
3787         case BOND_SLAVE_INFO_QUERY_OLD:
3788         case SIOCBONDSLAVEINFOQUERY:
3789                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3790
3791                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3792                         return -EFAULT;
3793
3794                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3795                 if (res == 0 &&
3796                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3797                         return -EFAULT;
3798
3799                 return res;
3800         default:
3801                 /* Go on */
3802                 break;
3803         }
3804
3805         if (!capable(CAP_NET_ADMIN))
3806                 return -EPERM;
3807
3808         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3809
3810         pr_debug("slave_dev=%p:\n", slave_dev);
3811
3812         if (!slave_dev)
3813                 res = -ENODEV;
3814         else {
3815                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3816                 switch (cmd) {
3817                 case BOND_ENSLAVE_OLD:
3818                 case SIOCBONDENSLAVE:
3819                         res = bond_enslave(bond_dev, slave_dev);
3820                         break;
3821                 case BOND_RELEASE_OLD:
3822                 case SIOCBONDRELEASE:
3823                         res = bond_release(bond_dev, slave_dev);
3824                         break;
3825                 case BOND_SETHWADDR_OLD:
3826                 case SIOCBONDSETHWADDR:
3827                         res = bond_sethwaddr(bond_dev, slave_dev);
3828                         break;
3829                 case BOND_CHANGE_ACTIVE_OLD:
3830                 case SIOCBONDCHANGEACTIVE:
3831                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3832                         break;
3833                 default:
3834                         res = -EOPNOTSUPP;
3835                 }
3836
3837                 dev_put(slave_dev);
3838         }
3839
3840         return res;
3841 }
3842
3843 static bool bond_addr_in_mc_list(unsigned char *addr,
3844                                  struct netdev_hw_addr_list *list,
3845                                  int addrlen)
3846 {
3847         struct netdev_hw_addr *ha;
3848
3849         netdev_hw_addr_list_for_each(ha, list)
3850                 if (!memcmp(ha->addr, addr, addrlen))
3851                         return true;
3852
3853         return false;
3854 }
3855
3856 static void bond_set_multicast_list(struct net_device *bond_dev)
3857 {
3858         struct bonding *bond = netdev_priv(bond_dev);
3859         struct netdev_hw_addr *ha;
3860         bool found;
3861
3862         /*
3863          * Do promisc before checking multicast_mode
3864          */
3865         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3866                 /*
3867                  * FIXME: Need to handle the error when one of the multi-slaves
3868                  * encounters error.
3869                  */
3870                 bond_set_promiscuity(bond, 1);
3871
3872
3873         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3874                 bond_set_promiscuity(bond, -1);
3875
3876
3877         /* set allmulti flag to slaves */
3878         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3879                 /*
3880                  * FIXME: Need to handle the error when one of the multi-slaves
3881                  * encounters error.
3882                  */
3883                 bond_set_allmulti(bond, 1);
3884
3885
3886         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3887                 bond_set_allmulti(bond, -1);
3888
3889
3890         read_lock(&bond->lock);
3891
3892         bond->flags = bond_dev->flags;
3893
3894         /* looking for addresses to add to slaves' mc list */
3895         netdev_for_each_mc_addr(ha, bond_dev) {
3896                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3897                                              bond_dev->addr_len);
3898                 if (!found)
3899                         bond_mc_add(bond, ha->addr);
3900         }
3901
3902         /* looking for addresses to delete from slaves' list */
3903         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3904                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3905                                              bond_dev->addr_len);
3906                 if (!found)
3907                         bond_mc_del(bond, ha->addr);
3908         }
3909
3910         /* save master's multicast list */
3911         __hw_addr_flush(&bond->mc_list);
3912         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3913                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3914
3915         read_unlock(&bond->lock);
3916 }
3917
3918 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3919 {
3920         struct bonding *bond = netdev_priv(dev);
3921         struct slave *slave = bond->first_slave;
3922
3923         if (slave) {
3924                 const struct net_device_ops *slave_ops
3925                         = slave->dev->netdev_ops;
3926                 if (slave_ops->ndo_neigh_setup)
3927                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3928         }
3929         return 0;
3930 }
3931
3932 /*
3933  * Change the MTU of all of a master's slaves to match the master
3934  */
3935 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3936 {
3937         struct bonding *bond = netdev_priv(bond_dev);
3938         struct slave *slave, *stop_at;
3939         int res = 0;
3940         int i;
3941
3942         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3943                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3944
3945         /* Can't hold bond->lock with bh disabled here since
3946          * some base drivers panic. On the other hand we can't
3947          * hold bond->lock without bh disabled because we'll
3948          * deadlock. The only solution is to rely on the fact
3949          * that we're under rtnl_lock here, and the slaves
3950          * list won't change. This doesn't solve the problem
3951          * of setting the slave's MTU while it is
3952          * transmitting, but the assumption is that the base
3953          * driver can handle that.
3954          *
3955          * TODO: figure out a way to safely iterate the slaves
3956          * list, but without holding a lock around the actual
3957          * call to the base driver.
3958          */
3959
3960         bond_for_each_slave(bond, slave, i) {
3961                 pr_debug("s %p s->p %p c_m %p\n",
3962                          slave,
3963                          slave->prev,
3964                          slave->dev->netdev_ops->ndo_change_mtu);
3965
3966                 res = dev_set_mtu(slave->dev, new_mtu);
3967
3968                 if (res) {
3969                         /* If we failed to set the slave's mtu to the new value
3970                          * we must abort the operation even in ACTIVE_BACKUP
3971                          * mode, because if we allow the backup slaves to have
3972                          * different mtu values than the active slave we'll
3973                          * need to change their mtu when doing a failover. That
3974                          * means changing their mtu from timer context, which
3975                          * is probably not a good idea.
3976                          */
3977                         pr_debug("err %d %s\n", res, slave->dev->name);
3978                         goto unwind;
3979                 }
3980         }
3981
3982         bond_dev->mtu = new_mtu;
3983
3984         return 0;
3985
3986 unwind:
3987         /* unwind from head to the slave that failed */
3988         stop_at = slave;
3989         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3990                 int tmp_res;
3991
3992                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3993                 if (tmp_res) {
3994                         pr_debug("unwind err %d dev %s\n",
3995                                  tmp_res, slave->dev->name);
3996                 }
3997         }
3998
3999         return res;
4000 }
4001
4002 /*
4003  * Change HW address
4004  *
4005  * Note that many devices must be down to change the HW address, and
4006  * downing the master releases all slaves.  We can make bonds full of
4007  * bonding devices to test this, however.
4008  */
4009 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4010 {
4011         struct bonding *bond = netdev_priv(bond_dev);
4012         struct sockaddr *sa = addr, tmp_sa;
4013         struct slave *slave, *stop_at;
4014         int res = 0;
4015         int i;
4016
4017         if (bond->params.mode == BOND_MODE_ALB)
4018                 return bond_alb_set_mac_address(bond_dev, addr);
4019
4020
4021         pr_debug("bond=%p, name=%s\n",
4022                  bond, bond_dev ? bond_dev->name : "None");
4023
4024         /*
4025          * If fail_over_mac is set to active, do nothing and return
4026          * success.  Returning an error causes ifenslave to fail.
4027          */
4028         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4029                 return 0;
4030
4031         if (!is_valid_ether_addr(sa->sa_data))
4032                 return -EADDRNOTAVAIL;
4033
4034         /* Can't hold bond->lock with bh disabled here since
4035          * some base drivers panic. On the other hand we can't
4036          * hold bond->lock without bh disabled because we'll
4037          * deadlock. The only solution is to rely on the fact
4038          * that we're under rtnl_lock here, and the slaves
4039          * list won't change. This doesn't solve the problem
4040          * of setting the slave's hw address while it is
4041          * transmitting, but the assumption is that the base
4042          * driver can handle that.
4043          *
4044          * TODO: figure out a way to safely iterate the slaves
4045          * list, but without holding a lock around the actual
4046          * call to the base driver.
4047          */
4048
4049         bond_for_each_slave(bond, slave, i) {
4050                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4051                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4052
4053                 if (slave_ops->ndo_set_mac_address == NULL) {
4054                         res = -EOPNOTSUPP;
4055                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4056                         goto unwind;
4057                 }
4058
4059                 res = dev_set_mac_address(slave->dev, addr);
4060                 if (res) {
4061                         /* TODO: consider downing the slave
4062                          * and retry ?
4063                          * User should expect communications
4064                          * breakage anyway until ARP finish
4065                          * updating, so...
4066                          */
4067                         pr_debug("err %d %s\n", res, slave->dev->name);
4068                         goto unwind;
4069                 }
4070         }
4071
4072         /* success */
4073         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4074         return 0;
4075
4076 unwind:
4077         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4078         tmp_sa.sa_family = bond_dev->type;
4079
4080         /* unwind from head to the slave that failed */
4081         stop_at = slave;
4082         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4083                 int tmp_res;
4084
4085                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4086                 if (tmp_res) {
4087                         pr_debug("unwind err %d dev %s\n",
4088                                  tmp_res, slave->dev->name);
4089                 }
4090         }
4091
4092         return res;
4093 }
4094
4095 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4096 {
4097         struct bonding *bond = netdev_priv(bond_dev);
4098         struct slave *slave, *start_at;
4099         int i, slave_no, res = 1;
4100         struct iphdr *iph = ip_hdr(skb);
4101
4102         read_lock(&bond->lock);
4103
4104         if (!BOND_IS_OK(bond))
4105                 goto out;
4106         /*
4107          * Start with the curr_active_slave that joined the bond as the
4108          * default for sending IGMP traffic.  For failover purposes one
4109          * needs to maintain some consistency for the interface that will
4110          * send the join/membership reports.  The curr_active_slave found
4111          * will send all of this type of traffic.
4112          */
4113         if ((iph->protocol == IPPROTO_IGMP) &&
4114             (skb->protocol == htons(ETH_P_IP))) {
4115
4116                 read_lock(&bond->curr_slave_lock);
4117                 slave = bond->curr_active_slave;
4118                 read_unlock(&bond->curr_slave_lock);
4119
4120                 if (!slave)
4121                         goto out;
4122         } else {
4123                 /*
4124                  * Concurrent TX may collide on rr_tx_counter; we accept
4125                  * that as being rare enough not to justify using an
4126                  * atomic op here.
4127                  */
4128                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4129
4130                 bond_for_each_slave(bond, slave, i) {
4131                         slave_no--;
4132                         if (slave_no < 0)
4133                                 break;
4134                 }
4135         }
4136
4137         start_at = slave;
4138         bond_for_each_slave_from(bond, slave, i, start_at) {
4139                 if (IS_UP(slave->dev) &&
4140                     (slave->link == BOND_LINK_UP) &&
4141                     bond_is_active_slave(slave)) {
4142                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4143                         break;
4144                 }
4145         }
4146
4147 out:
4148         if (res) {
4149                 /* no suitable interface, frame not sent */
4150                 dev_kfree_skb(skb);
4151         }
4152         read_unlock(&bond->lock);
4153         return NETDEV_TX_OK;
4154 }
4155
4156
4157 /*
4158  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4159  * the bond has a usable interface.
4160  */
4161 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4162 {
4163         struct bonding *bond = netdev_priv(bond_dev);
4164         int res = 1;
4165
4166         read_lock(&bond->lock);
4167         read_lock(&bond->curr_slave_lock);
4168
4169         if (!BOND_IS_OK(bond))
4170                 goto out;
4171
4172         if (!bond->curr_active_slave)
4173                 goto out;
4174
4175         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4176
4177 out:
4178         if (res)
4179                 /* no suitable interface, frame not sent */
4180                 dev_kfree_skb(skb);
4181
4182         read_unlock(&bond->curr_slave_lock);
4183         read_unlock(&bond->lock);
4184         return NETDEV_TX_OK;
4185 }
4186
4187 /*
4188  * In bond_xmit_xor() , we determine the output device by using a pre-
4189  * determined xmit_hash_policy(), If the selected device is not enabled,
4190  * find the next active slave.
4191  */
4192 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4193 {
4194         struct bonding *bond = netdev_priv(bond_dev);
4195         struct slave *slave, *start_at;
4196         int slave_no;
4197         int i;
4198         int res = 1;
4199
4200         read_lock(&bond->lock);
4201
4202         if (!BOND_IS_OK(bond))
4203                 goto out;
4204
4205         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4206
4207         bond_for_each_slave(bond, slave, i) {
4208                 slave_no--;
4209                 if (slave_no < 0)
4210                         break;
4211         }
4212
4213         start_at = slave;
4214
4215         bond_for_each_slave_from(bond, slave, i, start_at) {
4216                 if (IS_UP(slave->dev) &&
4217                     (slave->link == BOND_LINK_UP) &&
4218                     bond_is_active_slave(slave)) {
4219                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4220                         break;
4221                 }
4222         }
4223
4224 out:
4225         if (res) {
4226                 /* no suitable interface, frame not sent */
4227                 dev_kfree_skb(skb);
4228         }
4229         read_unlock(&bond->lock);
4230         return NETDEV_TX_OK;
4231 }
4232
4233 /*
4234  * in broadcast mode, we send everything to all usable interfaces.
4235  */
4236 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4237 {
4238         struct bonding *bond = netdev_priv(bond_dev);
4239         struct slave *slave, *start_at;
4240         struct net_device *tx_dev = NULL;
4241         int i;
4242         int res = 1;
4243
4244         read_lock(&bond->lock);
4245
4246         if (!BOND_IS_OK(bond))
4247                 goto out;
4248
4249         read_lock(&bond->curr_slave_lock);
4250         start_at = bond->curr_active_slave;
4251         read_unlock(&bond->curr_slave_lock);
4252
4253         if (!start_at)
4254                 goto out;
4255
4256         bond_for_each_slave_from(bond, slave, i, start_at) {
4257                 if (IS_UP(slave->dev) &&
4258                     (slave->link == BOND_LINK_UP) &&
4259                     bond_is_active_slave(slave)) {
4260                         if (tx_dev) {
4261                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4262                                 if (!skb2) {
4263                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4264                                                bond_dev->name);
4265                                         continue;
4266                                 }
4267
4268                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4269                                 if (res) {
4270                                         dev_kfree_skb(skb2);
4271                                         continue;
4272                                 }
4273                         }
4274                         tx_dev = slave->dev;
4275                 }
4276         }
4277
4278         if (tx_dev)
4279                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4280
4281 out:
4282         if (res)
4283                 /* no suitable interface, frame not sent */
4284                 dev_kfree_skb(skb);
4285
4286         /* frame sent to all suitable interfaces */
4287         read_unlock(&bond->lock);
4288         return NETDEV_TX_OK;
4289 }
4290
4291 /*------------------------- Device initialization ---------------------------*/
4292
4293 static void bond_set_xmit_hash_policy(struct bonding *bond)
4294 {
4295         switch (bond->params.xmit_policy) {
4296         case BOND_XMIT_POLICY_LAYER23:
4297                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4298                 break;
4299         case BOND_XMIT_POLICY_LAYER34:
4300                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4301                 break;
4302         case BOND_XMIT_POLICY_LAYER2:
4303         default:
4304                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4305                 break;
4306         }
4307 }
4308
4309 /*
4310  * Lookup the slave that corresponds to a qid
4311  */
4312 static inline int bond_slave_override(struct bonding *bond,
4313                                       struct sk_buff *skb)
4314 {
4315         int i, res = 1;
4316         struct slave *slave = NULL;
4317         struct slave *check_slave;
4318
4319         read_lock(&bond->lock);
4320
4321         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4322                 goto out;
4323
4324         /* Find out if any slaves have the same mapping as this skb. */
4325         bond_for_each_slave(bond, check_slave, i) {
4326                 if (check_slave->queue_id == skb->queue_mapping) {
4327                         slave = check_slave;
4328                         break;
4329                 }
4330         }
4331
4332         /* If the slave isn't UP, use default transmit policy. */
4333         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4334             (slave->link == BOND_LINK_UP)) {
4335                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4336         }
4337
4338 out:
4339         read_unlock(&bond->lock);
4340         return res;
4341 }
4342
4343 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4344 {
4345         /*
4346          * This helper function exists to help dev_pick_tx get the correct
4347          * destination queue.  Using a helper function skips a call to
4348          * skb_tx_hash and will put the skbs in the queue we expect on their
4349          * way down to the bonding driver.
4350          */
4351         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4352
4353         if (unlikely(txq >= dev->real_num_tx_queues)) {
4354                 do
4355                         txq -= dev->real_num_tx_queues;
4356                 while (txq >= dev->real_num_tx_queues);
4357         }
4358         return txq;
4359 }
4360
4361 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4362 {
4363         struct bonding *bond = netdev_priv(dev);
4364
4365         /*
4366          * If we risk deadlock from transmitting this in the
4367          * netpoll path, tell netpoll to queue the frame for later tx
4368          */
4369         if (is_netpoll_tx_blocked(dev))
4370                 return NETDEV_TX_BUSY;
4371
4372         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4373                 if (!bond_slave_override(bond, skb))
4374                         return NETDEV_TX_OK;
4375         }
4376
4377         switch (bond->params.mode) {
4378         case BOND_MODE_ROUNDROBIN:
4379                 return bond_xmit_roundrobin(skb, dev);
4380         case BOND_MODE_ACTIVEBACKUP:
4381                 return bond_xmit_activebackup(skb, dev);
4382         case BOND_MODE_XOR:
4383                 return bond_xmit_xor(skb, dev);
4384         case BOND_MODE_BROADCAST:
4385                 return bond_xmit_broadcast(skb, dev);
4386         case BOND_MODE_8023AD:
4387                 return bond_3ad_xmit_xor(skb, dev);
4388         case BOND_MODE_ALB:
4389         case BOND_MODE_TLB:
4390                 return bond_alb_xmit(skb, dev);
4391         default:
4392                 /* Should never happen, mode already checked */
4393                 pr_err("%s: Error: Unknown bonding mode %d\n",
4394                        dev->name, bond->params.mode);
4395                 WARN_ON_ONCE(1);
4396                 dev_kfree_skb(skb);
4397                 return NETDEV_TX_OK;
4398         }
4399 }
4400
4401
4402 /*
4403  * set bond mode specific net device operations
4404  */
4405 void bond_set_mode_ops(struct bonding *bond, int mode)
4406 {
4407         struct net_device *bond_dev = bond->dev;
4408
4409         switch (mode) {
4410         case BOND_MODE_ROUNDROBIN:
4411                 break;
4412         case BOND_MODE_ACTIVEBACKUP:
4413                 break;
4414         case BOND_MODE_XOR:
4415                 bond_set_xmit_hash_policy(bond);
4416                 break;
4417         case BOND_MODE_BROADCAST:
4418                 break;
4419         case BOND_MODE_8023AD:
4420                 bond_set_xmit_hash_policy(bond);
4421                 break;
4422         case BOND_MODE_ALB:
4423                 /* FALLTHRU */
4424         case BOND_MODE_TLB:
4425                 break;
4426         default:
4427                 /* Should never happen, mode already checked */
4428                 pr_err("%s: Error: Unknown bonding mode %d\n",
4429                        bond_dev->name, mode);
4430                 break;
4431         }
4432 }
4433
4434 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4435                                     struct ethtool_drvinfo *drvinfo)
4436 {
4437         strncpy(drvinfo->driver, DRV_NAME, 32);
4438         strncpy(drvinfo->version, DRV_VERSION, 32);
4439         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4440 }
4441
4442 static const struct ethtool_ops bond_ethtool_ops = {
4443         .get_drvinfo            = bond_ethtool_get_drvinfo,
4444         .get_link               = ethtool_op_get_link,
4445         .get_tx_csum            = ethtool_op_get_tx_csum,
4446         .get_sg                 = ethtool_op_get_sg,
4447         .get_tso                = ethtool_op_get_tso,
4448         .get_ufo                = ethtool_op_get_ufo,
4449         .get_flags              = ethtool_op_get_flags,
4450 };
4451
4452 static const struct net_device_ops bond_netdev_ops = {
4453         .ndo_init               = bond_init,
4454         .ndo_uninit             = bond_uninit,
4455         .ndo_open               = bond_open,
4456         .ndo_stop               = bond_close,
4457         .ndo_start_xmit         = bond_start_xmit,
4458         .ndo_select_queue       = bond_select_queue,
4459         .ndo_get_stats64        = bond_get_stats,
4460         .ndo_do_ioctl           = bond_do_ioctl,
4461         .ndo_set_multicast_list = bond_set_multicast_list,
4462         .ndo_change_mtu         = bond_change_mtu,
4463         .ndo_set_mac_address    = bond_set_mac_address,
4464         .ndo_neigh_setup        = bond_neigh_setup,
4465         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4466         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4467         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4468 #ifdef CONFIG_NET_POLL_CONTROLLER
4469         .ndo_netpoll_setup      = bond_netpoll_setup,
4470         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4471         .ndo_poll_controller    = bond_poll_controller,
4472 #endif
4473         .ndo_add_slave          = bond_enslave,
4474         .ndo_del_slave          = bond_release,
4475 };
4476
4477 static void bond_destructor(struct net_device *bond_dev)
4478 {
4479         struct bonding *bond = netdev_priv(bond_dev);
4480         if (bond->wq)
4481                 destroy_workqueue(bond->wq);
4482         free_netdev(bond_dev);
4483 }
4484
4485 static void bond_setup(struct net_device *bond_dev)
4486 {
4487         struct bonding *bond = netdev_priv(bond_dev);
4488
4489         /* initialize rwlocks */
4490         rwlock_init(&bond->lock);
4491         rwlock_init(&bond->curr_slave_lock);
4492
4493         bond->params = bonding_defaults;
4494
4495         /* Initialize pointers */
4496         bond->dev = bond_dev;
4497         INIT_LIST_HEAD(&bond->vlan_list);
4498
4499         /* Initialize the device entry points */
4500         ether_setup(bond_dev);
4501         bond_dev->netdev_ops = &bond_netdev_ops;
4502         bond_dev->ethtool_ops = &bond_ethtool_ops;
4503         bond_set_mode_ops(bond, bond->params.mode);
4504
4505         bond_dev->destructor = bond_destructor;
4506
4507         /* Initialize the device options */
4508         bond_dev->tx_queue_len = 0;
4509         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4510         bond_dev->priv_flags |= IFF_BONDING;
4511         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4512
4513         /* At first, we block adding VLANs. That's the only way to
4514          * prevent problems that occur when adding VLANs over an
4515          * empty bond. The block will be removed once non-challenged
4516          * slaves are enslaved.
4517          */
4518         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4519
4520         /* don't acquire bond device's netif_tx_lock when
4521          * transmitting */
4522         bond_dev->features |= NETIF_F_LLTX;
4523
4524         /* By default, we declare the bond to be fully
4525          * VLAN hardware accelerated capable. Special
4526          * care is taken in the various xmit functions
4527          * when there are slaves that are not hw accel
4528          * capable
4529          */
4530         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4531                                NETIF_F_HW_VLAN_RX |
4532                                NETIF_F_HW_VLAN_FILTER);
4533
4534         /* By default, we enable GRO on bonding devices.
4535          * Actual support requires lowlevel drivers are GRO ready.
4536          */
4537         bond_dev->features |= NETIF_F_GRO;
4538 }
4539
4540 static void bond_work_cancel_all(struct bonding *bond)
4541 {
4542         write_lock_bh(&bond->lock);
4543         bond->kill_timers = 1;
4544         write_unlock_bh(&bond->lock);
4545
4546         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4547                 cancel_delayed_work(&bond->mii_work);
4548
4549         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4550                 cancel_delayed_work(&bond->arp_work);
4551
4552         if (bond->params.mode == BOND_MODE_ALB &&
4553             delayed_work_pending(&bond->alb_work))
4554                 cancel_delayed_work(&bond->alb_work);
4555
4556         if (bond->params.mode == BOND_MODE_8023AD &&
4557             delayed_work_pending(&bond->ad_work))
4558                 cancel_delayed_work(&bond->ad_work);
4559
4560         if (delayed_work_pending(&bond->mcast_work))
4561                 cancel_delayed_work(&bond->mcast_work);
4562 }
4563
4564 /*
4565 * Destroy a bonding device.
4566 * Must be under rtnl_lock when this function is called.
4567 */
4568 static void bond_uninit(struct net_device *bond_dev)
4569 {
4570         struct bonding *bond = netdev_priv(bond_dev);
4571         struct vlan_entry *vlan, *tmp;
4572
4573         bond_netpoll_cleanup(bond_dev);
4574
4575         /* Release the bonded slaves */
4576         bond_release_all(bond_dev);
4577
4578         list_del(&bond->bond_list);
4579
4580         bond_work_cancel_all(bond);
4581
4582         bond_remove_proc_entry(bond);
4583
4584         bond_debug_unregister(bond);
4585
4586         __hw_addr_flush(&bond->mc_list);
4587
4588         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4589                 list_del(&vlan->vlan_list);
4590                 kfree(vlan);
4591         }
4592 }
4593
4594 /*------------------------- Module initialization ---------------------------*/
4595
4596 /*
4597  * Convert string input module parms.  Accept either the
4598  * number of the mode or its string name.  A bit complicated because
4599  * some mode names are substrings of other names, and calls from sysfs
4600  * may have whitespace in the name (trailing newlines, for example).
4601  */
4602 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4603 {
4604         int modeint = -1, i, rv;
4605         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4606
4607         for (p = (char *)buf; *p; p++)
4608                 if (!(isdigit(*p) || isspace(*p)))
4609                         break;
4610
4611         if (*p)
4612                 rv = sscanf(buf, "%20s", modestr);
4613         else
4614                 rv = sscanf(buf, "%d", &modeint);
4615
4616         if (!rv)
4617                 return -1;
4618
4619         for (i = 0; tbl[i].modename; i++) {
4620                 if (modeint == tbl[i].mode)
4621                         return tbl[i].mode;
4622                 if (strcmp(modestr, tbl[i].modename) == 0)
4623                         return tbl[i].mode;
4624         }
4625
4626         return -1;
4627 }
4628
4629 static int bond_check_params(struct bond_params *params)
4630 {
4631         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4632
4633         /*
4634          * Convert string parameters.
4635          */
4636         if (mode) {
4637                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4638                 if (bond_mode == -1) {
4639                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4640                                mode == NULL ? "NULL" : mode);
4641                         return -EINVAL;
4642                 }
4643         }
4644
4645         if (xmit_hash_policy) {
4646                 if ((bond_mode != BOND_MODE_XOR) &&
4647                     (bond_mode != BOND_MODE_8023AD)) {
4648                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4649                                bond_mode_name(bond_mode));
4650                 } else {
4651                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4652                                                         xmit_hashtype_tbl);
4653                         if (xmit_hashtype == -1) {
4654                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4655                                        xmit_hash_policy == NULL ? "NULL" :
4656                                        xmit_hash_policy);
4657                                 return -EINVAL;
4658                         }
4659                 }
4660         }
4661
4662         if (lacp_rate) {
4663                 if (bond_mode != BOND_MODE_8023AD) {
4664                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4665                                 bond_mode_name(bond_mode));
4666                 } else {
4667                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4668                         if (lacp_fast == -1) {
4669                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4670                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4671                                 return -EINVAL;
4672                         }
4673                 }
4674         }
4675
4676         if (ad_select) {
4677                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4678                 if (params->ad_select == -1) {
4679                         pr_err("Error: Invalid ad_select \"%s\"\n",
4680                                ad_select == NULL ? "NULL" : ad_select);
4681                         return -EINVAL;
4682                 }
4683
4684                 if (bond_mode != BOND_MODE_8023AD) {
4685                         pr_warning("ad_select param only affects 802.3ad mode\n");
4686                 }
4687         } else {
4688                 params->ad_select = BOND_AD_STABLE;
4689         }
4690
4691         if (max_bonds < 0) {
4692                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4693                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4694                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4695         }
4696
4697         if (miimon < 0) {
4698                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4699                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4700                 miimon = BOND_LINK_MON_INTERV;
4701         }
4702
4703         if (updelay < 0) {
4704                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4705                            updelay, INT_MAX);
4706                 updelay = 0;
4707         }
4708
4709         if (downdelay < 0) {
4710                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4711                            downdelay, INT_MAX);
4712                 downdelay = 0;
4713         }
4714
4715         if ((use_carrier != 0) && (use_carrier != 1)) {
4716                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4717                            use_carrier);
4718                 use_carrier = 1;
4719         }
4720
4721         if (num_grat_arp < 0 || num_grat_arp > 255) {
4722                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4723                            num_grat_arp);
4724                 num_grat_arp = 1;
4725         }
4726
4727         if (num_unsol_na < 0 || num_unsol_na > 255) {
4728                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4729                            num_unsol_na);
4730                 num_unsol_na = 1;
4731         }
4732
4733         /* reset values for 802.3ad */
4734         if (bond_mode == BOND_MODE_8023AD) {
4735                 if (!miimon) {
4736                         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");
4737                         pr_warning("Forcing miimon to 100msec\n");
4738                         miimon = 100;
4739                 }
4740         }
4741
4742         if (tx_queues < 1 || tx_queues > 255) {
4743                 pr_warning("Warning: tx_queues (%d) should be between "
4744                            "1 and 255, resetting to %d\n",
4745                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4746                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4747         }
4748
4749         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4750                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4751                            "not of valid value (0/1), so it was set to "
4752                            "0\n", all_slaves_active);
4753                 all_slaves_active = 0;
4754         }
4755
4756         if (resend_igmp < 0 || resend_igmp > 255) {
4757                 pr_warning("Warning: resend_igmp (%d) should be between "
4758                            "0 and 255, resetting to %d\n",
4759                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4760                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4761         }
4762
4763         /* reset values for TLB/ALB */
4764         if ((bond_mode == BOND_MODE_TLB) ||
4765             (bond_mode == BOND_MODE_ALB)) {
4766                 if (!miimon) {
4767                         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");
4768                         pr_warning("Forcing miimon to 100msec\n");
4769                         miimon = 100;
4770                 }
4771         }
4772
4773         if (bond_mode == BOND_MODE_ALB) {
4774                 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",
4775                           updelay);
4776         }
4777
4778         if (!miimon) {
4779                 if (updelay || downdelay) {
4780                         /* just warn the user the up/down delay will have
4781                          * no effect since miimon is zero...
4782                          */
4783                         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",
4784                                    updelay, downdelay);
4785                 }
4786         } else {
4787                 /* don't allow arp monitoring */
4788                 if (arp_interval) {
4789                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4790                                    miimon, arp_interval);
4791                         arp_interval = 0;
4792                 }
4793
4794                 if ((updelay % miimon) != 0) {
4795                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4796                                    updelay, miimon,
4797                                    (updelay / miimon) * miimon);
4798                 }
4799
4800                 updelay /= miimon;
4801
4802                 if ((downdelay % miimon) != 0) {
4803                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4804                                    downdelay, miimon,
4805                                    (downdelay / miimon) * miimon);
4806                 }
4807
4808                 downdelay /= miimon;
4809         }
4810
4811         if (arp_interval < 0) {
4812                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4813                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4814                 arp_interval = BOND_LINK_ARP_INTERV;
4815         }
4816
4817         for (arp_ip_count = 0;
4818              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4819              arp_ip_count++) {
4820                 /* not complete check, but should be good enough to
4821                    catch mistakes */
4822                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4823                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4824                                    arp_ip_target[arp_ip_count]);
4825                         arp_interval = 0;
4826                 } else {
4827                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4828                         arp_target[arp_ip_count] = ip;
4829                 }
4830         }
4831
4832         if (arp_interval && !arp_ip_count) {
4833                 /* don't allow arping if no arp_ip_target given... */
4834                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4835                            arp_interval);
4836                 arp_interval = 0;
4837         }
4838
4839         if (arp_validate) {
4840                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4841                         pr_err("arp_validate only supported in active-backup mode\n");
4842                         return -EINVAL;
4843                 }
4844                 if (!arp_interval) {
4845                         pr_err("arp_validate requires arp_interval\n");
4846                         return -EINVAL;
4847                 }
4848
4849                 arp_validate_value = bond_parse_parm(arp_validate,
4850                                                      arp_validate_tbl);
4851                 if (arp_validate_value == -1) {
4852                         pr_err("Error: invalid arp_validate \"%s\"\n",
4853                                arp_validate == NULL ? "NULL" : arp_validate);
4854                         return -EINVAL;
4855                 }
4856         } else
4857                 arp_validate_value = 0;
4858
4859         if (miimon) {
4860                 pr_info("MII link monitoring set to %d ms\n", miimon);
4861         } else if (arp_interval) {
4862                 int i;
4863
4864                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4865                         arp_interval,
4866                         arp_validate_tbl[arp_validate_value].modename,
4867                         arp_ip_count);
4868
4869                 for (i = 0; i < arp_ip_count; i++)
4870                         pr_info(" %s", arp_ip_target[i]);
4871
4872                 pr_info("\n");
4873
4874         } else if (max_bonds) {
4875                 /* miimon and arp_interval not set, we need one so things
4876                  * work as expected, see bonding.txt for details
4877                  */
4878                 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4879         }
4880
4881         if (primary && !USES_PRIMARY(bond_mode)) {
4882                 /* currently, using a primary only makes sense
4883                  * in active backup, TLB or ALB modes
4884                  */
4885                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4886                            primary, bond_mode_name(bond_mode));
4887                 primary = NULL;
4888         }
4889
4890         if (primary && primary_reselect) {
4891                 primary_reselect_value = bond_parse_parm(primary_reselect,
4892                                                          pri_reselect_tbl);
4893                 if (primary_reselect_value == -1) {
4894                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4895                                primary_reselect ==
4896                                         NULL ? "NULL" : primary_reselect);
4897                         return -EINVAL;
4898                 }
4899         } else {
4900                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4901         }
4902
4903         if (fail_over_mac) {
4904                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4905                                                       fail_over_mac_tbl);
4906                 if (fail_over_mac_value == -1) {
4907                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4908                                arp_validate == NULL ? "NULL" : arp_validate);
4909                         return -EINVAL;
4910                 }
4911
4912                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4913                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4914         } else {
4915                 fail_over_mac_value = BOND_FOM_NONE;
4916         }
4917
4918         /* fill params struct with the proper values */
4919         params->mode = bond_mode;
4920         params->xmit_policy = xmit_hashtype;
4921         params->miimon = miimon;
4922         params->num_grat_arp = num_grat_arp;
4923         params->num_unsol_na = num_unsol_na;
4924         params->arp_interval = arp_interval;
4925         params->arp_validate = arp_validate_value;
4926         params->updelay = updelay;
4927         params->downdelay = downdelay;
4928         params->use_carrier = use_carrier;
4929         params->lacp_fast = lacp_fast;
4930         params->primary[0] = 0;
4931         params->primary_reselect = primary_reselect_value;
4932         params->fail_over_mac = fail_over_mac_value;
4933         params->tx_queues = tx_queues;
4934         params->all_slaves_active = all_slaves_active;
4935         params->resend_igmp = resend_igmp;
4936
4937         if (primary) {
4938                 strncpy(params->primary, primary, IFNAMSIZ);
4939                 params->primary[IFNAMSIZ - 1] = 0;
4940         }
4941
4942         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4943
4944         return 0;
4945 }
4946
4947 static struct lock_class_key bonding_netdev_xmit_lock_key;
4948 static struct lock_class_key bonding_netdev_addr_lock_key;
4949
4950 static void bond_set_lockdep_class_one(struct net_device *dev,
4951                                        struct netdev_queue *txq,
4952                                        void *_unused)
4953 {
4954         lockdep_set_class(&txq->_xmit_lock,
4955                           &bonding_netdev_xmit_lock_key);
4956 }
4957
4958 static void bond_set_lockdep_class(struct net_device *dev)
4959 {
4960         lockdep_set_class(&dev->addr_list_lock,
4961                           &bonding_netdev_addr_lock_key);
4962         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4963 }
4964
4965 /*
4966  * Called from registration process
4967  */
4968 static int bond_init(struct net_device *bond_dev)
4969 {
4970         struct bonding *bond = netdev_priv(bond_dev);
4971         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4972
4973         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4974
4975         bond->wq = create_singlethread_workqueue(bond_dev->name);
4976         if (!bond->wq)
4977                 return -ENOMEM;
4978
4979         bond_set_lockdep_class(bond_dev);
4980
4981         bond_create_proc_entry(bond);
4982         list_add_tail(&bond->bond_list, &bn->dev_list);
4983
4984         bond_prepare_sysfs_group(bond);
4985
4986         bond_debug_register(bond);
4987
4988         __hw_addr_init(&bond->mc_list);
4989         return 0;
4990 }
4991
4992 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4993 {
4994         if (tb[IFLA_ADDRESS]) {
4995                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4996                         return -EINVAL;
4997                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4998                         return -EADDRNOTAVAIL;
4999         }
5000         return 0;
5001 }
5002
5003 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5004         .kind           = "bond",
5005         .priv_size      = sizeof(struct bonding),
5006         .setup          = bond_setup,
5007         .validate       = bond_validate,
5008 };
5009
5010 /* Create a new bond based on the specified name and bonding parameters.
5011  * If name is NULL, obtain a suitable "bond%d" name for us.
5012  * Caller must NOT hold rtnl_lock; we need to release it here before we
5013  * set up our sysfs entries.
5014  */
5015 int bond_create(struct net *net, const char *name)
5016 {
5017         struct net_device *bond_dev;
5018         int res;
5019
5020         rtnl_lock();
5021
5022         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5023                                 bond_setup, tx_queues);
5024         if (!bond_dev) {
5025                 pr_err("%s: eek! can't alloc netdev!\n", name);
5026                 rtnl_unlock();
5027                 return -ENOMEM;
5028         }
5029
5030         dev_net_set(bond_dev, net);
5031         bond_dev->rtnl_link_ops = &bond_link_ops;
5032
5033         if (!name) {
5034                 res = dev_alloc_name(bond_dev, "bond%d");
5035                 if (res < 0)
5036                         goto out;
5037         } else {
5038                 /*
5039                  * If we're given a name to register
5040                  * we need to ensure that its not already
5041                  * registered
5042                  */
5043                 res = -EEXIST;
5044                 if (__dev_get_by_name(net, name) != NULL)
5045                         goto out;
5046         }
5047
5048         res = register_netdevice(bond_dev);
5049
5050         netif_carrier_off(bond_dev);
5051
5052 out:
5053         rtnl_unlock();
5054         if (res < 0)
5055                 bond_destructor(bond_dev);
5056         return res;
5057 }
5058
5059 static int __net_init bond_net_init(struct net *net)
5060 {
5061         struct bond_net *bn = net_generic(net, bond_net_id);
5062
5063         bn->net = net;
5064         INIT_LIST_HEAD(&bn->dev_list);
5065
5066         bond_create_proc_dir(bn);
5067
5068         return 0;
5069 }
5070
5071 static void __net_exit bond_net_exit(struct net *net)
5072 {
5073         struct bond_net *bn = net_generic(net, bond_net_id);
5074
5075         bond_destroy_proc_dir(bn);
5076 }
5077
5078 static struct pernet_operations bond_net_ops = {
5079         .init = bond_net_init,
5080         .exit = bond_net_exit,
5081         .id   = &bond_net_id,
5082         .size = sizeof(struct bond_net),
5083 };
5084
5085 static int __init bonding_init(void)
5086 {
5087         int i;
5088         int res;
5089
5090         pr_info("%s", bond_version);
5091
5092         res = bond_check_params(&bonding_defaults);
5093         if (res)
5094                 goto out;
5095
5096         res = register_pernet_subsys(&bond_net_ops);
5097         if (res)
5098                 goto out;
5099
5100         res = rtnl_link_register(&bond_link_ops);
5101         if (res)
5102                 goto err_link;
5103
5104         bond_create_debugfs();
5105
5106         for (i = 0; i < max_bonds; i++) {
5107                 res = bond_create(&init_net, NULL);
5108                 if (res)
5109                         goto err;
5110         }
5111
5112         res = bond_create_sysfs();
5113         if (res)
5114                 goto err;
5115
5116         register_netdevice_notifier(&bond_netdev_notifier);
5117         register_inetaddr_notifier(&bond_inetaddr_notifier);
5118         bond_register_ipv6_notifier();
5119 out:
5120         return res;
5121 err:
5122         rtnl_link_unregister(&bond_link_ops);
5123 err_link:
5124         unregister_pernet_subsys(&bond_net_ops);
5125         goto out;
5126
5127 }
5128
5129 static void __exit bonding_exit(void)
5130 {
5131         unregister_netdevice_notifier(&bond_netdev_notifier);
5132         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5133         bond_unregister_ipv6_notifier();
5134
5135         bond_destroy_sysfs();
5136         bond_destroy_debugfs();
5137
5138         rtnl_link_unregister(&bond_link_ops);
5139         unregister_pernet_subsys(&bond_net_ops);
5140
5141 #ifdef CONFIG_NET_POLL_CONTROLLER
5142         /*
5143          * Make sure we don't have an imbalance on our netpoll blocking
5144          */
5145         WARN_ON(atomic_read(&netpoll_block_tx));
5146 #endif
5147 }
5148
5149 module_init(bonding_init);
5150 module_exit(bonding_exit);
5151 MODULE_LICENSE("GPL");
5152 MODULE_VERSION(DRV_VERSION);
5153 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5154 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5155 MODULE_ALIAS_RTNL_LINK("bond");