drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/seq_file.h>
  70 #include <linux/smp.h>
  71 #include <linux/if_ether.h>
  72 #include <net/arp.h>
  73 #include <linux/mii.h>
  74 #include <linux/ethtool.h>
  75 #include <linux/if_vlan.h>
  76 #include <linux/if_bonding.h>
  77 #include <linux/jiffies.h>
  78 #include <linux/preempt.h>
  79 #include <net/route.h>
  80 #include <net/net_namespace.h>
  81 #include <net/netns/generic.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89 #define BOND_LINK_MON_INTERV    0
  90 #define BOND_LINK_ARP_INTERV    0
  91
  92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  94 static int num_grat_arp = 1;
  95 static int num_unsol_na = 1;
  96 static int miimon       = BOND_LINK_MON_INTERV;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static char *ad_select;
 105 static char *xmit_hash_policy;
 106 static int arp_interval = BOND_LINK_ARP_INTERV;
 107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 108 static char *arp_validate;
 109 static char *fail_over_mac;
 110 static int all_slaves_active = 0;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113
 114 module_param(max_bonds, int, 0);
 115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 116 module_param(tx_queues, int, 0);
 117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 118 module_param(num_grat_arp, int, 0644);
 119 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 120 module_param(num_unsol_na, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 149                             "(slow/fast)");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 152 module_param(xmit_hash_policy, charp, 0);
 153 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 154                                    ", 1 for layer 3+4");
 155 module_param(arp_interval, int, 0);
 156 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 157 module_param_array(arp_ip_target, charp, NULL, 0);
 158 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 159 module_param(arp_validate, charp, 0);
 160 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 161 module_param(fail_over_mac, charp, 0);
 162 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 163 module_param(all_slaves_active, int, 0);
 164 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 165                                      "by setting active flag for all slaves.  "
 166                                      "0 for never (default), 1 for always.");
 167 module_param(resend_igmp, int, 0);
 168 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 169
 170 /*----------------------------- Global variables ----------------------------*/
 171
 172 #ifdef CONFIG_NET_POLL_CONTROLLER
 173 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 174 #endif
 175
 176 static const char * const version =
 177         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 178
 179 int bond_net_id __read_mostly;
 180
 181 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 182 static int arp_ip_count;
 183 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 184 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 185 static int lacp_fast;
 186
 187 const struct bond_parm_tbl bond_lacp_tbl[] = {
 188 {       "slow",         AD_LACP_SLOW},
 189 {       "fast",         AD_LACP_FAST},
 190 {       NULL,           -1},
 191 };
 192
 193 const struct bond_parm_tbl bond_mode_tbl[] = {
 194 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 195 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 196 {       "balance-xor",          BOND_MODE_XOR},
 197 {       "broadcast",            BOND_MODE_BROADCAST},
 198 {       "802.3ad",              BOND_MODE_8023AD},
 199 {       "balance-tlb",          BOND_MODE_TLB},
 200 {       "balance-alb",          BOND_MODE_ALB},
 201 {       NULL,                   -1},
 202 };
 203
 204 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 205 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 206 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 207 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 208 {       NULL,                   -1},
 209 };
 210
 211 const struct bond_parm_tbl arp_validate_tbl[] = {
 212 {       "none",                 BOND_ARP_VALIDATE_NONE},
 213 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 214 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 215 {       "all",                  BOND_ARP_VALIDATE_ALL},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 220 {       "none",                 BOND_FOM_NONE},
 221 {       "active",               BOND_FOM_ACTIVE},
 222 {       "follow",               BOND_FOM_FOLLOW},
 223 {       NULL,                   -1},
 224 };
 225
 226 const struct bond_parm_tbl pri_reselect_tbl[] = {
 227 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 228 {       "better",               BOND_PRI_RESELECT_BETTER},
 229 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 230 {       NULL,                   -1},
 231 };
 232
 233 struct bond_parm_tbl ad_select_tbl[] = {
 234 {       "stable",       BOND_AD_STABLE},
 235 {       "bandwidth",    BOND_AD_BANDWIDTH},
 236 {       "count",        BOND_AD_COUNT},
 237 {       NULL,           -1},
 238 };
 239
 240 /*-------------------------- Forward declarations ---------------------------*/
 241
 242 static void bond_send_gratuitous_arp(struct bonding *bond);
 243 static int bond_init(struct net_device *bond_dev);
 244 static void bond_uninit(struct net_device *bond_dev);
 245
 246 /*---------------------------- General routines -----------------------------*/
 247
 248 static const char *bond_mode_name(int mode)
 249 {
 250         static const char *names[] = {
 251                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 252                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 253                 [BOND_MODE_XOR] = "load balancing (xor)",
 254                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 255                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 256                 [BOND_MODE_TLB] = "transmit load balancing",
 257                 [BOND_MODE_ALB] = "adaptive load balancing",
 258         };
 259
 260         if (mode < 0 || mode > BOND_MODE_ALB)
 261                 return "unknown";
 262
 263         return names[mode];
 264 }
 265
 266 /*---------------------------------- VLAN -----------------------------------*/
 267
 268 /**
 269  * bond_add_vlan - add a new vlan id on bond
 270  * @bond: bond that got the notification
 271  * @vlan_id: the vlan id to add
 272  *
 273  * Returns -ENOMEM if allocation failed.
 274  */
 275 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 276 {
 277         struct vlan_entry *vlan;
 278
 279         pr_debug("bond: %s, vlan id %d\n",
 280                  (bond ? bond->dev->name : "None"), vlan_id);
 281
 282         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 283         if (!vlan)
 284                 return -ENOMEM;
 285
 286         INIT_LIST_HEAD(&vlan->vlan_list);
 287         vlan->vlan_id = vlan_id;
 288
 289         write_lock_bh(&bond->lock);
 290
 291         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 292
 293         write_unlock_bh(&bond->lock);
 294
 295         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 296
 297         return 0;
 298 }
 299
 300 /**
 301  * bond_del_vlan - delete a vlan id from bond
 302  * @bond: bond that got the notification
 303  * @vlan_id: the vlan id to delete
 304  *
 305  * returns -ENODEV if @vlan_id was not found in @bond.
 306  */
 307 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 308 {
 309         struct vlan_entry *vlan;
 310         int res = -ENODEV;
 311
 312         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 313
 314         block_netpoll_tx();
 315         write_lock_bh(&bond->lock);
 316
 317         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 318                 if (vlan->vlan_id == vlan_id) {
 319                         list_del(&vlan->vlan_list);
 320
 321                         if (bond_is_lb(bond))
 322                                 bond_alb_clear_vlan(bond, vlan_id);
 323
 324                         pr_debug("removed VLAN ID %d from bond %s\n",
 325                                  vlan_id, bond->dev->name);
 326
 327                         kfree(vlan);
 328
 329                         if (list_empty(&bond->vlan_list) &&
 330                             (bond->slave_cnt == 0)) {
 331                                 /* Last VLAN removed and no slaves, so
 332                                  * restore block on adding VLANs. This will
 333                                  * be removed once new slaves that are not
 334                                  * VLAN challenged will be added.
 335                                  */
 336                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 337                         }
 338
 339                         res = 0;
 340                         goto out;
 341                 }
 342         }
 343
 344         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 345                  vlan_id, bond->dev->name);
 346
 347 out:
 348         write_unlock_bh(&bond->lock);
 349         unblock_netpoll_tx();
 350         return res;
 351 }
 352
 353 /**
 354  * bond_has_challenged_slaves
 355  * @bond: the bond we're working on
 356  *
 357  * Searches the slave list. Returns 1 if a vlan challenged slave
 358  * was found, 0 otherwise.
 359  *
 360  * Assumes bond->lock is held.
 361  */
 362 static int bond_has_challenged_slaves(struct bonding *bond)
 363 {
 364         struct slave *slave;
 365         int i;
 366
 367         bond_for_each_slave(bond, slave, i) {
 368                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 369                         pr_debug("found VLAN challenged slave - %s\n",
 370                                  slave->dev->name);
 371                         return 1;
 372                 }
 373         }
 374
 375         pr_debug("no VLAN challenged slaves found\n");
 376         return 0;
 377 }
 378
 379 /**
 380  * bond_next_vlan - safely skip to the next item in the vlans list.
 381  * @bond: the bond we're working on
 382  * @curr: item we're advancing from
 383  *
 384  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 385  * or @curr->next otherwise (even if it is @curr itself again).
 386  *
 387  * Caller must hold bond->lock
 388  */
 389 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 390 {
 391         struct vlan_entry *next, *last;
 392
 393         if (list_empty(&bond->vlan_list))
 394                 return NULL;
 395
 396         if (!curr) {
 397                 next = list_entry(bond->vlan_list.next,
 398                                   struct vlan_entry, vlan_list);
 399         } else {
 400                 last = list_entry(bond->vlan_list.prev,
 401                                   struct vlan_entry, vlan_list);
 402                 if (last == curr) {
 403                         next = list_entry(bond->vlan_list.next,
 404                                           struct vlan_entry, vlan_list);
 405                 } else {
 406                         next = list_entry(curr->vlan_list.next,
 407                                           struct vlan_entry, vlan_list);
 408                 }
 409         }
 410
 411         return next;
 412 }
 413
 414 /**
 415  * bond_dev_queue_xmit - Prepare skb for xmit.
 416  *
 417  * @bond: bond device that got this skb for tx.
 418  * @skb: hw accel VLAN tagged skb to transmit
 419  * @slave_dev: slave that is supposed to xmit this skbuff
 420  */
 421 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 422                         struct net_device *slave_dev)
 423 {
 424         skb->dev = slave_dev;
 425         skb->priority = 1;
 426         if (unlikely(netpoll_tx_running(slave_dev)))
 427                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 428         else
 429                 dev_queue_xmit(skb);
 430
 431         return 0;
 432 }
 433
 434 /*
 435  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 436  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 437  * lock because:
 438  * a. This operation is performed in IOCTL context,
 439  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 440  * c. Holding a lock with BH disabled while directly calling a base driver
 441  *    entry point is generally a BAD idea.
 442  *
 443  * The design of synchronization/protection for this operation in the 8021q
 444  * module is good for one or more VLAN devices over a single physical device
 445  * and cannot be extended for a teaming solution like bonding, so there is a
 446  * potential race condition here where a net device from the vlan group might
 447  * be referenced (either by a base driver or the 8021q code) while it is being
 448  * removed from the system. However, it turns out we're not making matters
 449  * worse, and if it works for regular VLAN usage it will work here too.
 450 */
 451
 452 /**
 453  * bond_vlan_rx_register - Propagates registration to slaves
 454  * @bond_dev: bonding net device that got called
 455  * @grp: vlan group being registered
 456  */
 457 static void bond_vlan_rx_register(struct net_device *bond_dev,
 458                                   struct vlan_group *grp)
 459 {
 460         struct bonding *bond = netdev_priv(bond_dev);
 461         struct slave *slave;
 462         int i;
 463
 464         write_lock_bh(&bond->lock);
 465         bond->vlgrp = grp;
 466         write_unlock_bh(&bond->lock);
 467
 468         bond_for_each_slave(bond, slave, i) {
 469                 struct net_device *slave_dev = slave->dev;
 470                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 471
 472                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 473                     slave_ops->ndo_vlan_rx_register) {
 474                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 475                 }
 476         }
 477 }
 478
 479 /**
 480  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 481  * @bond_dev: bonding net device that got called
 482  * @vid: vlan id being added
 483  */
 484 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 485 {
 486         struct bonding *bond = netdev_priv(bond_dev);
 487         struct slave *slave;
 488         int i, res;
 489
 490         bond_for_each_slave(bond, slave, i) {
 491                 struct net_device *slave_dev = slave->dev;
 492                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 493
 494                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 495                     slave_ops->ndo_vlan_rx_add_vid) {
 496                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 497                 }
 498         }
 499
 500         res = bond_add_vlan(bond, vid);
 501         if (res) {
 502                 pr_err("%s: Error: Failed to add vlan id %d\n",
 503                        bond_dev->name, vid);
 504         }
 505 }
 506
 507 /**
 508  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 509  * @bond_dev: bonding net device that got called
 510  * @vid: vlan id being removed
 511  */
 512 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 513 {
 514         struct bonding *bond = netdev_priv(bond_dev);
 515         struct slave *slave;
 516         struct net_device *vlan_dev;
 517         int i, res;
 518
 519         bond_for_each_slave(bond, slave, i) {
 520                 struct net_device *slave_dev = slave->dev;
 521                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 522
 523                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 524                     slave_ops->ndo_vlan_rx_kill_vid) {
 525                         /* Save and then restore vlan_dev in the grp array,
 526                          * since the slave's driver might clear it.
 527                          */
 528                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 529                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 530                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 531                 }
 532         }
 533
 534         res = bond_del_vlan(bond, vid);
 535         if (res) {
 536                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 537                        bond_dev->name, vid);
 538         }
 539 }
 540
 541 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 542 {
 543         struct vlan_entry *vlan;
 544         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 545
 546         if (!bond->vlgrp)
 547                 return;
 548
 549         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 550             slave_ops->ndo_vlan_rx_register)
 551                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 552
 553         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 554             !(slave_ops->ndo_vlan_rx_add_vid))
 555                 return;
 556
 557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 558                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 559 }
 560
 561 static void bond_del_vlans_from_slave(struct bonding *bond,
 562                                       struct net_device *slave_dev)
 563 {
 564         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 565         struct vlan_entry *vlan;
 566         struct net_device *vlan_dev;
 567
 568         if (!bond->vlgrp)
 569                 return;
 570
 571         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 572             !(slave_ops->ndo_vlan_rx_kill_vid))
 573                 goto unreg;
 574
 575         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 576                 if (!vlan->vlan_id)
 577                         continue;
 578                 /* Save and then restore vlan_dev in the grp array,
 579                  * since the slave's driver might clear it.
 580                  */
 581                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 582                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 583                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 584         }
 585
 586 unreg:
 587         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 588             slave_ops->ndo_vlan_rx_register)
 589                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 590 }
 591
 592 /*------------------------------- Link status -------------------------------*/
 593
 594 /*
 595  * Set the carrier state for the master according to the state of its
 596  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 597  * do special 802.3ad magic.
 598  *
 599  * Returns zero if carrier state does not change, nonzero if it does.
 600  */
 601 static int bond_set_carrier(struct bonding *bond)
 602 {
 603         struct slave *slave;
 604         int i;
 605
 606         if (bond->slave_cnt == 0)
 607                 goto down;
 608
 609         if (bond->params.mode == BOND_MODE_8023AD)
 610                 return bond_3ad_set_carrier(bond);
 611
 612         bond_for_each_slave(bond, slave, i) {
 613                 if (slave->link == BOND_LINK_UP) {
 614                         if (!netif_carrier_ok(bond->dev)) {
 615                                 netif_carrier_on(bond->dev);
 616                                 return 1;
 617                         }
 618                         return 0;
 619                 }
 620         }
 621
 622 down:
 623         if (netif_carrier_ok(bond->dev)) {
 624                 netif_carrier_off(bond->dev);
 625                 return 1;
 626         }
 627         return 0;
 628 }
 629
 630 /*
 631  * Get link speed and duplex from the slave's base driver
 632  * using ethtool. If for some reason the call fails or the
 633  * values are invalid, fake speed and duplex to 100/Full
 634  * and return error.
 635  */
 636 static int bond_update_speed_duplex(struct slave *slave)
 637 {
 638         struct net_device *slave_dev = slave->dev;
 639         struct ethtool_cmd etool;
 640         int res;
 641
 642         /* Fake speed and duplex */
 643         slave->speed = SPEED_100;
 644         slave->duplex = DUPLEX_FULL;
 645
 646         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 647                 return -1;
 648
 649         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 650         if (res < 0)
 651                 return -1;
 652
 653         switch (etool.speed) {
 654         case SPEED_10:
 655         case SPEED_100:
 656         case SPEED_1000:
 657         case SPEED_10000:
 658                 break;
 659         default:
 660                 return -1;
 661         }
 662
 663         switch (etool.duplex) {
 664         case DUPLEX_FULL:
 665         case DUPLEX_HALF:
 666                 break;
 667         default:
 668                 return -1;
 669         }
 670
 671         slave->speed = etool.speed;
 672         slave->duplex = etool.duplex;
 673
 674         return 0;
 675 }
 676
 677 /*
 678  * if <dev> supports MII link status reporting, check its link status.
 679  *
 680  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 681  * depending upon the setting of the use_carrier parameter.
 682  *
 683  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 684  * can't tell and just pretend it is), or 0, meaning that the link is
 685  * down.
 686  *
 687  * If reporting is non-zero, instead of faking link up, return -1 if
 688  * both ETHTOOL and MII ioctls fail (meaning the device does not
 689  * support them).  If use_carrier is set, return whatever it says.
 690  * It'd be nice if there was a good way to tell if a driver supports
 691  * netif_carrier, but there really isn't.
 692  */
 693 static int bond_check_dev_link(struct bonding *bond,
 694                                struct net_device *slave_dev, int reporting)
 695 {
 696         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 697         int (*ioctl)(struct net_device *, struct ifreq *, int);
 698         struct ifreq ifr;
 699         struct mii_ioctl_data *mii;
 700
 701         if (!reporting && !netif_running(slave_dev))
 702                 return 0;
 703
 704         if (bond->params.use_carrier)
 705                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 706
 707         /* Try to get link status using Ethtool first. */
 708         if (slave_dev->ethtool_ops) {
 709                 if (slave_dev->ethtool_ops->get_link) {
 710                         u32 link;
 711
 712                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 713
 714                         return link ? BMSR_LSTATUS : 0;
 715                 }
 716         }
 717
 718         /* Ethtool can't be used, fallback to MII ioctls. */
 719         ioctl = slave_ops->ndo_do_ioctl;
 720         if (ioctl) {
 721                 /* TODO: set pointer to correct ioctl on a per team member */
 722                 /*       bases to make this more efficient. that is, once  */
 723                 /*       we determine the correct ioctl, we will always    */
 724                 /*       call it and not the others for that team          */
 725                 /*       member.                                           */
 726
 727                 /*
 728                  * We cannot assume that SIOCGMIIPHY will also read a
 729                  * register; not all network drivers (e.g., e100)
 730                  * support that.
 731                  */
 732
 733                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 734                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 735                 mii = if_mii(&ifr);
 736                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 737                         mii->reg_num = MII_BMSR;
 738                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 739                                 return mii->val_out & BMSR_LSTATUS;
 740                 }
 741         }
 742
 743         /*
 744          * If reporting, report that either there's no dev->do_ioctl,
 745          * or both SIOCGMIIREG and get_link failed (meaning that we
 746          * cannot report link status).  If not reporting, pretend
 747          * we're ok.
 748          */
 749         return reporting ? -1 : BMSR_LSTATUS;
 750 }
 751
 752 /*----------------------------- Multicast list ------------------------------*/
 753
 754 /*
 755  * Push the promiscuity flag down to appropriate slaves
 756  */
 757 static int bond_set_promiscuity(struct bonding *bond, int inc)
 758 {
 759         int err = 0;
 760         if (USES_PRIMARY(bond->params.mode)) {
 761                 /* write lock already acquired */
 762                 if (bond->curr_active_slave) {
 763                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 764                                                   inc);
 765                 }
 766         } else {
 767                 struct slave *slave;
 768                 int i;
 769                 bond_for_each_slave(bond, slave, i) {
 770                         err = dev_set_promiscuity(slave->dev, inc);
 771                         if (err)
 772                                 return err;
 773                 }
 774         }
 775         return err;
 776 }
 777
 778 /*
 779  * Push the allmulti flag down to all slaves
 780  */
 781 static int bond_set_allmulti(struct bonding *bond, int inc)
 782 {
 783         int err = 0;
 784         if (USES_PRIMARY(bond->params.mode)) {
 785                 /* write lock already acquired */
 786                 if (bond->curr_active_slave) {
 787                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 788                                                inc);
 789                 }
 790         } else {
 791                 struct slave *slave;
 792                 int i;
 793                 bond_for_each_slave(bond, slave, i) {
 794                         err = dev_set_allmulti(slave->dev, inc);
 795                         if (err)
 796                                 return err;
 797                 }
 798         }
 799         return err;
 800 }
 801
 802 /*
 803  * Add a Multicast address to slaves
 804  * according to mode
 805  */
 806 static void bond_mc_add(struct bonding *bond, void *addr)
 807 {
 808         if (USES_PRIMARY(bond->params.mode)) {
 809                 /* write lock already acquired */
 810                 if (bond->curr_active_slave)
 811                         dev_mc_add(bond->curr_active_slave->dev, addr);
 812         } else {
 813                 struct slave *slave;
 814                 int i;
 815
 816                 bond_for_each_slave(bond, slave, i)
 817                         dev_mc_add(slave->dev, addr);
 818         }
 819 }
 820
 821 /*
 822  * Remove a multicast address from slave
 823  * according to mode
 824  */
 825 static void bond_mc_del(struct bonding *bond, void *addr)
 826 {
 827         if (USES_PRIMARY(bond->params.mode)) {
 828                 /* write lock already acquired */
 829                 if (bond->curr_active_slave)
 830                         dev_mc_del(bond->curr_active_slave->dev, addr);
 831         } else {
 832                 struct slave *slave;
 833                 int i;
 834                 bond_for_each_slave(bond, slave, i) {
 835                         dev_mc_del(slave->dev, addr);
 836                 }
 837         }
 838 }
 839
 840
 841 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 842 {
 843         struct in_device *in_dev;
 844
 845         rcu_read_lock();
 846         in_dev = __in_dev_get_rcu(dev);
 847         if (in_dev)
 848                 ip_mc_rejoin_groups(in_dev);
 849         rcu_read_unlock();
 850 }
 851
 852 /*
 853  * Retrieve the list of registered multicast addresses for the bonding
 854  * device and retransmit an IGMP JOIN request to the current active
 855  * slave.
 856  */
 857 static void bond_resend_igmp_join_requests(struct bonding *bond)
 858 {
 859         struct net_device *vlan_dev;
 860         struct vlan_entry *vlan;
 861
 862         read_lock(&bond->lock);
 863
 864         /* rejoin all groups on bond device */
 865         __bond_resend_igmp_join_requests(bond->dev);
 866
 867         /* rejoin all groups on vlan devices */
 868         if (bond->vlgrp) {
 869                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 870                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 871                                                          vlan->vlan_id);
 872                         if (vlan_dev)
 873                                 __bond_resend_igmp_join_requests(vlan_dev);
 874                 }
 875         }
 876
 877         if (--bond->igmp_retrans > 0)
 878                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 879
 880         read_unlock(&bond->lock);
 881 }
 882
 883 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 884 {
 885         struct bonding *bond = container_of(work, struct bonding,
 886                                                         mcast_work.work);
 887         bond_resend_igmp_join_requests(bond);
 888 }
 889
 890 /*
 891  * flush all members of flush->mc_list from device dev->mc_list
 892  */
 893 static void bond_mc_list_flush(struct net_device *bond_dev,
 894                                struct net_device *slave_dev)
 895 {
 896         struct bonding *bond = netdev_priv(bond_dev);
 897         struct netdev_hw_addr *ha;
 898
 899         netdev_for_each_mc_addr(ha, bond_dev)
 900                 dev_mc_del(slave_dev, ha->addr);
 901
 902         if (bond->params.mode == BOND_MODE_8023AD) {
 903                 /* del lacpdu mc addr from mc list */
 904                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 905
 906                 dev_mc_del(slave_dev, lacpdu_multicast);
 907         }
 908 }
 909
 910 /*--------------------------- Active slave change ---------------------------*/
 911
 912 /*
 913  * Update the mc list and multicast-related flags for the new and
 914  * old active slaves (if any) according to the multicast mode, and
 915  * promiscuous flags unconditionally.
 916  */
 917 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 918                          struct slave *old_active)
 919 {
 920         struct netdev_hw_addr *ha;
 921
 922         if (!USES_PRIMARY(bond->params.mode))
 923                 /* nothing to do -  mc list is already up-to-date on
 924                  * all slaves
 925                  */
 926                 return;
 927
 928         if (old_active) {
 929                 if (bond->dev->flags & IFF_PROMISC)
 930                         dev_set_promiscuity(old_active->dev, -1);
 931
 932                 if (bond->dev->flags & IFF_ALLMULTI)
 933                         dev_set_allmulti(old_active->dev, -1);
 934
 935                 netdev_for_each_mc_addr(ha, bond->dev)
 936                         dev_mc_del(old_active->dev, ha->addr);
 937         }
 938
 939         if (new_active) {
 940                 /* FIXME: Signal errors upstream. */
 941                 if (bond->dev->flags & IFF_PROMISC)
 942                         dev_set_promiscuity(new_active->dev, 1);
 943
 944                 if (bond->dev->flags & IFF_ALLMULTI)
 945                         dev_set_allmulti(new_active->dev, 1);
 946
 947                 netdev_for_each_mc_addr(ha, bond->dev)
 948                         dev_mc_add(new_active->dev, ha->addr);
 949         }
 950 }
 951
 952 /*
 953  * bond_do_fail_over_mac
 954  *
 955  * Perform special MAC address swapping for fail_over_mac settings
 956  *
 957  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 958  */
 959 static void bond_do_fail_over_mac(struct bonding *bond,
 960                                   struct slave *new_active,
 961                                   struct slave *old_active)
 962         __releases(&bond->curr_slave_lock)
 963         __releases(&bond->lock)
 964         __acquires(&bond->lock)
 965         __acquires(&bond->curr_slave_lock)
 966 {
 967         u8 tmp_mac[ETH_ALEN];
 968         struct sockaddr saddr;
 969         int rv;
 970
 971         switch (bond->params.fail_over_mac) {
 972         case BOND_FOM_ACTIVE:
 973                 if (new_active)
 974                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 975                                new_active->dev->addr_len);
 976                 break;
 977         case BOND_FOM_FOLLOW:
 978                 /*
 979                  * if new_active && old_active, swap them
 980                  * if just old_active, do nothing (going to no active slave)
 981                  * if just new_active, set new_active to bond's MAC
 982                  */
 983                 if (!new_active)
 984                         return;
 985
 986                 write_unlock_bh(&bond->curr_slave_lock);
 987                 read_unlock(&bond->lock);
 988
 989                 if (old_active) {
 990                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 991                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 992                                ETH_ALEN);
 993                         saddr.sa_family = new_active->dev->type;
 994                 } else {
 995                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 996                         saddr.sa_family = bond->dev->type;
 997                 }
 998
 999                 rv = dev_set_mac_address(new_active->dev, &saddr);
1000                 if (rv) {
1001                         pr_err("%s: Error %d setting MAC of slave %s\n",
1002                                bond->dev->name, -rv, new_active->dev->name);
1003                         goto out;
1004                 }
1005
1006                 if (!old_active)
1007                         goto out;
1008
1009                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1010                 saddr.sa_family = old_active->dev->type;
1011
1012                 rv = dev_set_mac_address(old_active->dev, &saddr);
1013                 if (rv)
1014                         pr_err("%s: Error %d setting MAC of slave %s\n",
1015                                bond->dev->name, -rv, new_active->dev->name);
1016 out:
1017                 read_lock(&bond->lock);
1018                 write_lock_bh(&bond->curr_slave_lock);
1019                 break;
1020         default:
1021                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1022                        bond->dev->name, bond->params.fail_over_mac);
1023                 break;
1024         }
1025
1026 }
1027
1028 static bool bond_should_change_active(struct bonding *bond)
1029 {
1030         struct slave *prim = bond->primary_slave;
1031         struct slave *curr = bond->curr_active_slave;
1032
1033         if (!prim || !curr || curr->link != BOND_LINK_UP)
1034                 return true;
1035         if (bond->force_primary) {
1036                 bond->force_primary = false;
1037                 return true;
1038         }
1039         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1040             (prim->speed < curr->speed ||
1041              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1042                 return false;
1043         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1044                 return false;
1045         return true;
1046 }
1047
1048 /**
1049  * find_best_interface - select the best available slave to be the active one
1050  * @bond: our bonding struct
1051  *
1052  * Warning: Caller must hold curr_slave_lock for writing.
1053  */
1054 static struct slave *bond_find_best_slave(struct bonding *bond)
1055 {
1056         struct slave *new_active, *old_active;
1057         struct slave *bestslave = NULL;
1058         int mintime = bond->params.updelay;
1059         int i;
1060
1061         new_active = bond->curr_active_slave;
1062
1063         if (!new_active) { /* there were no active slaves left */
1064                 if (bond->slave_cnt > 0)   /* found one slave */
1065                         new_active = bond->first_slave;
1066                 else
1067                         return NULL; /* still no slave, return NULL */
1068         }
1069
1070         if ((bond->primary_slave) &&
1071             bond->primary_slave->link == BOND_LINK_UP &&
1072             bond_should_change_active(bond)) {
1073                 new_active = bond->primary_slave;
1074         }
1075
1076         /* remember where to stop iterating over the slaves */
1077         old_active = new_active;
1078
1079         bond_for_each_slave_from(bond, new_active, i, old_active) {
1080                 if (new_active->link == BOND_LINK_UP) {
1081                         return new_active;
1082                 } else if (new_active->link == BOND_LINK_BACK &&
1083                            IS_UP(new_active->dev)) {
1084                         /* link up, but waiting for stabilization */
1085                         if (new_active->delay < mintime) {
1086                                 mintime = new_active->delay;
1087                                 bestslave = new_active;
1088                         }
1089                 }
1090         }
1091
1092         return bestslave;
1093 }
1094
1095 /**
1096  * change_active_interface - change the active slave into the specified one
1097  * @bond: our bonding struct
1098  * @new: the new slave to make the active one
1099  *
1100  * Set the new slave to the bond's settings and unset them on the old
1101  * curr_active_slave.
1102  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1103  *
1104  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1105  * because it is apparently the best available slave we have, even though its
1106  * updelay hasn't timed out yet.
1107  *
1108  * If new_active is not NULL, caller must hold bond->lock for read and
1109  * curr_slave_lock for write_bh.
1110  */
1111 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1112 {
1113         struct slave *old_active = bond->curr_active_slave;
1114
1115         if (old_active == new_active)
1116                 return;
1117
1118         if (new_active) {
1119                 new_active->jiffies = jiffies;
1120
1121                 if (new_active->link == BOND_LINK_BACK) {
1122                         if (USES_PRIMARY(bond->params.mode)) {
1123                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1124                                         bond->dev->name, new_active->dev->name,
1125                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1126                         }
1127
1128                         new_active->delay = 0;
1129                         new_active->link = BOND_LINK_UP;
1130
1131                         if (bond->params.mode == BOND_MODE_8023AD)
1132                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1133
1134                         if (bond_is_lb(bond))
1135                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1136                 } else {
1137                         if (USES_PRIMARY(bond->params.mode)) {
1138                                 pr_info("%s: making interface %s the new active one.\n",
1139                                         bond->dev->name, new_active->dev->name);
1140                         }
1141                 }
1142         }
1143
1144         if (USES_PRIMARY(bond->params.mode))
1145                 bond_mc_swap(bond, new_active, old_active);
1146
1147         if (bond_is_lb(bond)) {
1148                 bond_alb_handle_active_change(bond, new_active);
1149                 if (old_active)
1150                         bond_set_slave_inactive_flags(old_active);
1151                 if (new_active)
1152                         bond_set_slave_active_flags(new_active);
1153         } else {
1154                 bond->curr_active_slave = new_active;
1155         }
1156
1157         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1158                 if (old_active)
1159                         bond_set_slave_inactive_flags(old_active);
1160
1161                 if (new_active) {
1162                         bond_set_slave_active_flags(new_active);
1163
1164                         if (bond->params.fail_over_mac)
1165                                 bond_do_fail_over_mac(bond, new_active,
1166                                                       old_active);
1167
1168                         if (netif_running(bond->dev)) {
1169                                 bond->send_grat_arp = bond->params.num_grat_arp;
1170                                 bond_send_gratuitous_arp(bond);
1171
1172                                 bond->send_unsol_na = bond->params.num_unsol_na;
1173                                 bond_send_unsolicited_na(bond);
1174                         }
1175
1176                         write_unlock_bh(&bond->curr_slave_lock);
1177                         read_unlock(&bond->lock);
1178
1179                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1180
1181                         read_lock(&bond->lock);
1182                         write_lock_bh(&bond->curr_slave_lock);
1183                 }
1184         }
1185
1186         /* resend IGMP joins since active slave has changed or
1187          * all were sent on curr_active_slave */
1188         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1189              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1190             netif_running(bond->dev)) {
1191                 bond->igmp_retrans = bond->params.resend_igmp;
1192                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1193         }
1194 }
1195
1196 /**
1197  * bond_select_active_slave - select a new active slave, if needed
1198  * @bond: our bonding struct
1199  *
1200  * This functions should be called when one of the following occurs:
1201  * - The old curr_active_slave has been released or lost its link.
1202  * - The primary_slave has got its link back.
1203  * - A slave has got its link back and there's no old curr_active_slave.
1204  *
1205  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1206  */
1207 void bond_select_active_slave(struct bonding *bond)
1208 {
1209         struct slave *best_slave;
1210         int rv;
1211
1212         best_slave = bond_find_best_slave(bond);
1213         if (best_slave != bond->curr_active_slave) {
1214                 bond_change_active_slave(bond, best_slave);
1215                 rv = bond_set_carrier(bond);
1216                 if (!rv)
1217                         return;
1218
1219                 if (netif_carrier_ok(bond->dev)) {
1220                         pr_info("%s: first active interface up!\n",
1221                                 bond->dev->name);
1222                 } else {
1223                         pr_info("%s: now running without any active interface !\n",
1224                                 bond->dev->name);
1225                 }
1226         }
1227 }
1228
1229 /*--------------------------- slave list handling ---------------------------*/
1230
1231 /*
1232  * This function attaches the slave to the end of list.
1233  *
1234  * bond->lock held for writing by caller.
1235  */
1236 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1237 {
1238         if (bond->first_slave == NULL) { /* attaching the first slave */
1239                 new_slave->next = new_slave;
1240                 new_slave->prev = new_slave;
1241                 bond->first_slave = new_slave;
1242         } else {
1243                 new_slave->next = bond->first_slave;
1244                 new_slave->prev = bond->first_slave->prev;
1245                 new_slave->next->prev = new_slave;
1246                 new_slave->prev->next = new_slave;
1247         }
1248
1249         bond->slave_cnt++;
1250 }
1251
1252 /*
1253  * This function detaches the slave from the list.
1254  * WARNING: no check is made to verify if the slave effectively
1255  * belongs to <bond>.
1256  * Nothing is freed on return, structures are just unchained.
1257  * If any slave pointer in bond was pointing to <slave>,
1258  * it should be changed by the calling function.
1259  *
1260  * bond->lock held for writing by caller.
1261  */
1262 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1263 {
1264         if (slave->next)
1265                 slave->next->prev = slave->prev;
1266
1267         if (slave->prev)
1268                 slave->prev->next = slave->next;
1269
1270         if (bond->first_slave == slave) { /* slave is the first slave */
1271                 if (bond->slave_cnt > 1) { /* there are more slave */
1272                         bond->first_slave = slave->next;
1273                 } else {
1274                         bond->first_slave = NULL; /* slave was the last one */
1275                 }
1276         }
1277
1278         slave->next = NULL;
1279         slave->prev = NULL;
1280         bond->slave_cnt--;
1281 }
1282
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 static inline int slave_enable_netpoll(struct slave *slave)
1285 {
1286         struct netpoll *np;
1287         int err = 0;
1288
1289         np = kzalloc(sizeof(*np), GFP_KERNEL);
1290         err = -ENOMEM;
1291         if (!np)
1292                 goto out;
1293
1294         np->dev = slave->dev;
1295         err = __netpoll_setup(np);
1296         if (err) {
1297                 kfree(np);
1298                 goto out;
1299         }
1300         slave->np = np;
1301 out:
1302         return err;
1303 }
1304 static inline void slave_disable_netpoll(struct slave *slave)
1305 {
1306         struct netpoll *np = slave->np;
1307
1308         if (!np)
1309                 return;
1310
1311         slave->np = NULL;
1312         synchronize_rcu_bh();
1313         __netpoll_cleanup(np);
1314         kfree(np);
1315 }
1316 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1317 {
1318         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1319                 return false;
1320         if (!slave_dev->netdev_ops->ndo_poll_controller)
1321                 return false;
1322         return true;
1323 }
1324
1325 static void bond_poll_controller(struct net_device *bond_dev)
1326 {
1327 }
1328
1329 static void __bond_netpoll_cleanup(struct bonding *bond)
1330 {
1331         struct slave *slave;
1332         int i;
1333
1334         bond_for_each_slave(bond, slave, i)
1335                 if (IS_UP(slave->dev))
1336                         slave_disable_netpoll(slave);
1337 }
1338 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1339 {
1340         struct bonding *bond = netdev_priv(bond_dev);
1341
1342         read_lock(&bond->lock);
1343         __bond_netpoll_cleanup(bond);
1344         read_unlock(&bond->lock);
1345 }
1346
1347 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1348 {
1349         struct bonding *bond = netdev_priv(dev);
1350         struct slave *slave;
1351         int i, err = 0;
1352
1353         read_lock(&bond->lock);
1354         bond_for_each_slave(bond, slave, i) {
1355                 if (!IS_UP(slave->dev))
1356                         continue;
1357                 err = slave_enable_netpoll(slave);
1358                 if (err) {
1359                         __bond_netpoll_cleanup(bond);
1360                         break;
1361                 }
1362         }
1363         read_unlock(&bond->lock);
1364         return err;
1365 }
1366
1367 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1368 {
1369         return bond->dev->npinfo;
1370 }
1371
1372 #else
1373 static inline int slave_enable_netpoll(struct slave *slave)
1374 {
1375         return 0;
1376 }
1377 static inline void slave_disable_netpoll(struct slave *slave)
1378 {
1379 }
1380 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1381 {
1382 }
1383 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1384 {
1385         return 0;
1386 }
1387 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1388 {
1389         return NULL;
1390 }
1391 #endif
1392
1393 /*---------------------------------- IOCTL ----------------------------------*/
1394
1395 static int bond_sethwaddr(struct net_device *bond_dev,
1396                           struct net_device *slave_dev)
1397 {
1398         pr_debug("bond_dev=%p\n", bond_dev);
1399         pr_debug("slave_dev=%p\n", slave_dev);
1400         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1401         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1402         return 0;
1403 }
1404
1405 #define BOND_VLAN_FEATURES \
1406         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1407          NETIF_F_HW_VLAN_FILTER)
1408
1409 /*
1410  * Compute the common dev->feature set available to all slaves.  Some
1411  * feature bits are managed elsewhere, so preserve those feature bits
1412  * on the master device.
1413  */
1414 static int bond_compute_features(struct bonding *bond)
1415 {
1416         struct slave *slave;
1417         struct net_device *bond_dev = bond->dev;
1418         u32 features = bond_dev->features;
1419         u32 vlan_features = 0;
1420         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1421                                                 bond_dev->hard_header_len);
1422         int i;
1423
1424         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1425         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1426
1427         if (!bond->first_slave)
1428                 goto done;
1429
1430         features &= ~NETIF_F_ONE_FOR_ALL;
1431
1432         vlan_features = bond->first_slave->dev->vlan_features;
1433         bond_for_each_slave(bond, slave, i) {
1434                 features = netdev_increment_features(features,
1435                                                      slave->dev->features,
1436                                                      NETIF_F_ONE_FOR_ALL);
1437                 vlan_features = netdev_increment_features(vlan_features,
1438                                                         slave->dev->vlan_features,
1439                                                         NETIF_F_ONE_FOR_ALL);
1440                 if (slave->dev->hard_header_len > max_hard_header_len)
1441                         max_hard_header_len = slave->dev->hard_header_len;
1442         }
1443
1444 done:
1445         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1446         bond_dev->features = netdev_fix_features(bond_dev, features);
1447         bond_dev->vlan_features = netdev_fix_features(bond_dev, vlan_features);
1448         bond_dev->hard_header_len = max_hard_header_len;
1449
1450         return 0;
1451 }
1452
1453 static void bond_setup_by_slave(struct net_device *bond_dev,
1454                                 struct net_device *slave_dev)
1455 {
1456         struct bonding *bond = netdev_priv(bond_dev);
1457
1458         bond_dev->header_ops        = slave_dev->header_ops;
1459
1460         bond_dev->type              = slave_dev->type;
1461         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1462         bond_dev->addr_len          = slave_dev->addr_len;
1463
1464         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1465                 slave_dev->addr_len);
1466         bond->setup_by_slave = 1;
1467 }
1468
1469 /* enslave device <slave> to bond device <master> */
1470 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1471 {
1472         struct bonding *bond = netdev_priv(bond_dev);
1473         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1474         struct slave *new_slave = NULL;
1475         struct netdev_hw_addr *ha;
1476         struct sockaddr addr;
1477         int link_reporting;
1478         int old_features = bond_dev->features;
1479         int res = 0;
1480
1481         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1482                 slave_ops->ndo_do_ioctl == NULL) {
1483                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1484                            bond_dev->name, slave_dev->name);
1485         }
1486
1487         /* bond must be initialized by bond_open() before enslaving */
1488         if (!(bond_dev->flags & IFF_UP)) {
1489                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1490                            bond_dev->name);
1491         }
1492
1493         /* already enslaved */
1494         if (slave_dev->flags & IFF_SLAVE) {
1495                 pr_debug("Error, Device was already enslaved\n");
1496                 return -EBUSY;
1497         }
1498
1499         /* vlan challenged mutual exclusion */
1500         /* no need to lock since we're protected by rtnl_lock */
1501         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1502                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1503                 if (bond->vlgrp) {
1504                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1505                                bond_dev->name, slave_dev->name, bond_dev->name);
1506                         return -EPERM;
1507                 } else {
1508                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1509                                    bond_dev->name, slave_dev->name,
1510                                    slave_dev->name, bond_dev->name);
1511                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1512                 }
1513         } else {
1514                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1515                 if (bond->slave_cnt == 0) {
1516                         /* First slave, and it is not VLAN challenged,
1517                          * so remove the block of adding VLANs over the bond.
1518                          */
1519                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1520                 }
1521         }
1522
1523         /*
1524          * Old ifenslave binaries are no longer supported.  These can
1525          * be identified with moderate accuracy by the state of the slave:
1526          * the current ifenslave will set the interface down prior to
1527          * enslaving it; the old ifenslave will not.
1528          */
1529         if ((slave_dev->flags & IFF_UP)) {
1530                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1531                        slave_dev->name);
1532                 res = -EPERM;
1533                 goto err_undo_flags;
1534         }
1535
1536         /* set bonding device ether type by slave - bonding netdevices are
1537          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1538          * there is a need to override some of the type dependent attribs/funcs.
1539          *
1540          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1541          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1542          */
1543         if (bond->slave_cnt == 0) {
1544                 if (bond_dev->type != slave_dev->type) {
1545                         pr_debug("%s: change device type from %d to %d\n",
1546                                  bond_dev->name,
1547                                  bond_dev->type, slave_dev->type);
1548
1549                         res = netdev_bonding_change(bond_dev,
1550                                                     NETDEV_PRE_TYPE_CHANGE);
1551                         res = notifier_to_errno(res);
1552                         if (res) {
1553                                 pr_err("%s: refused to change device type\n",
1554                                        bond_dev->name);
1555                                 res = -EBUSY;
1556                                 goto err_undo_flags;
1557                         }
1558
1559                         /* Flush unicast and multicast addresses */
1560                         dev_uc_flush(bond_dev);
1561                         dev_mc_flush(bond_dev);
1562
1563                         if (slave_dev->type != ARPHRD_ETHER)
1564                                 bond_setup_by_slave(bond_dev, slave_dev);
1565                         else
1566                                 ether_setup(bond_dev);
1567
1568                         netdev_bonding_change(bond_dev,
1569                                               NETDEV_POST_TYPE_CHANGE);
1570                 }
1571         } else if (bond_dev->type != slave_dev->type) {
1572                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1573                        slave_dev->name,
1574                        slave_dev->type, bond_dev->type);
1575                 res = -EINVAL;
1576                 goto err_undo_flags;
1577         }
1578
1579         if (slave_ops->ndo_set_mac_address == NULL) {
1580                 if (bond->slave_cnt == 0) {
1581                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1582                                    bond_dev->name);
1583                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1584                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1585                         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",
1586                                bond_dev->name);
1587                         res = -EOPNOTSUPP;
1588                         goto err_undo_flags;
1589                 }
1590         }
1591
1592         /* If this is the first slave, then we need to set the master's hardware
1593          * address to be the same as the slave's. */
1594         if (is_zero_ether_addr(bond->dev->dev_addr))
1595                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1596                        slave_dev->addr_len);
1597
1598
1599         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1600         if (!new_slave) {
1601                 res = -ENOMEM;
1602                 goto err_undo_flags;
1603         }
1604
1605         /*
1606          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1607          * is set via sysfs or module option if desired.
1608          */
1609         new_slave->queue_id = 0;
1610
1611         /* Save slave's original mtu and then set it to match the bond */
1612         new_slave->original_mtu = slave_dev->mtu;
1613         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1614         if (res) {
1615                 pr_debug("Error %d calling dev_set_mtu\n", res);
1616                 goto err_free;
1617         }
1618
1619         /*
1620          * Save slave's original ("permanent") mac address for modes
1621          * that need it, and for restoring it upon release, and then
1622          * set it to the master's address
1623          */
1624         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1625
1626         if (!bond->params.fail_over_mac) {
1627                 /*
1628                  * Set slave to master's mac address.  The application already
1629                  * set the master's mac address to that of the first slave
1630                  */
1631                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1632                 addr.sa_family = slave_dev->type;
1633                 res = dev_set_mac_address(slave_dev, &addr);
1634                 if (res) {
1635                         pr_debug("Error %d calling set_mac_address\n", res);
1636                         goto err_restore_mtu;
1637                 }
1638         }
1639
1640         res = netdev_set_bond_master(slave_dev, bond_dev);
1641         if (res) {
1642                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1643                 goto err_restore_mac;
1644         }
1645         /* open the slave since the application closed it */
1646         res = dev_open(slave_dev);
1647         if (res) {
1648                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1649                 goto err_unset_master;
1650         }
1651
1652         new_slave->dev = slave_dev;
1653         slave_dev->priv_flags |= IFF_BONDING;
1654
1655         if (bond_is_lb(bond)) {
1656                 /* bond_alb_init_slave() must be called before all other stages since
1657                  * it might fail and we do not want to have to undo everything
1658                  */
1659                 res = bond_alb_init_slave(bond, new_slave);
1660                 if (res)
1661                         goto err_close;
1662         }
1663
1664         /* If the mode USES_PRIMARY, then the new slave gets the
1665          * master's promisc (and mc) settings only if it becomes the
1666          * curr_active_slave, and that is taken care of later when calling
1667          * bond_change_active()
1668          */
1669         if (!USES_PRIMARY(bond->params.mode)) {
1670                 /* set promiscuity level to new slave */
1671                 if (bond_dev->flags & IFF_PROMISC) {
1672                         res = dev_set_promiscuity(slave_dev, 1);
1673                         if (res)
1674                                 goto err_close;
1675                 }
1676
1677                 /* set allmulti level to new slave */
1678                 if (bond_dev->flags & IFF_ALLMULTI) {
1679                         res = dev_set_allmulti(slave_dev, 1);
1680                         if (res)
1681                                 goto err_close;
1682                 }
1683
1684                 netif_addr_lock_bh(bond_dev);
1685                 /* upload master's mc_list to new slave */
1686                 netdev_for_each_mc_addr(ha, bond_dev)
1687                         dev_mc_add(slave_dev, ha->addr);
1688                 netif_addr_unlock_bh(bond_dev);
1689         }
1690
1691         if (bond->params.mode == BOND_MODE_8023AD) {
1692                 /* add lacpdu mc addr to mc list */
1693                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1694
1695                 dev_mc_add(slave_dev, lacpdu_multicast);
1696         }
1697
1698         bond_add_vlans_on_slave(bond, slave_dev);
1699
1700         write_lock_bh(&bond->lock);
1701
1702         bond_attach_slave(bond, new_slave);
1703
1704         new_slave->delay = 0;
1705         new_slave->link_failure_count = 0;
1706
1707         bond_compute_features(bond);
1708
1709         write_unlock_bh(&bond->lock);
1710
1711         read_lock(&bond->lock);
1712
1713         new_slave->last_arp_rx = jiffies;
1714
1715         if (bond->params.miimon && !bond->params.use_carrier) {
1716                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1717
1718                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1719                         /*
1720                          * miimon is set but a bonded network driver
1721                          * does not support ETHTOOL/MII and
1722                          * arp_interval is not set.  Note: if
1723                          * use_carrier is enabled, we will never go
1724                          * here (because netif_carrier is always
1725                          * supported); thus, we don't need to change
1726                          * the messages for netif_carrier.
1727                          */
1728                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1729                                bond_dev->name, slave_dev->name);
1730                 } else if (link_reporting == -1) {
1731                         /* unable get link status using mii/ethtool */
1732                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1733                                    bond_dev->name, slave_dev->name);
1734                 }
1735         }
1736
1737         /* check for initial state */
1738         if (!bond->params.miimon ||
1739             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1740                 if (bond->params.updelay) {
1741                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1742                         new_slave->link  = BOND_LINK_BACK;
1743                         new_slave->delay = bond->params.updelay;
1744                 } else {
1745                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1746                         new_slave->link  = BOND_LINK_UP;
1747                 }
1748                 new_slave->jiffies = jiffies;
1749         } else {
1750                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1751                 new_slave->link  = BOND_LINK_DOWN;
1752         }
1753
1754         if (bond_update_speed_duplex(new_slave) &&
1755             (new_slave->link != BOND_LINK_DOWN)) {
1756                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1757                            bond_dev->name, new_slave->dev->name);
1758
1759                 if (bond->params.mode == BOND_MODE_8023AD) {
1760                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1761                                    bond_dev->name);
1762                 }
1763         }
1764
1765         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1766                 /* if there is a primary slave, remember it */
1767                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1768                         bond->primary_slave = new_slave;
1769                         bond->force_primary = true;
1770                 }
1771         }
1772
1773         write_lock_bh(&bond->curr_slave_lock);
1774
1775         switch (bond->params.mode) {
1776         case BOND_MODE_ACTIVEBACKUP:
1777                 bond_set_slave_inactive_flags(new_slave);
1778                 bond_select_active_slave(bond);
1779                 break;
1780         case BOND_MODE_8023AD:
1781                 /* in 802.3ad mode, the internal mechanism
1782                  * will activate the slaves in the selected
1783                  * aggregator
1784                  */
1785                 bond_set_slave_inactive_flags(new_slave);
1786                 /* if this is the first slave */
1787                 if (bond->slave_cnt == 1) {
1788                         SLAVE_AD_INFO(new_slave).id = 1;
1789                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1790                          * can be called only after the mac address of the bond is set
1791                          */
1792                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1793                                             bond->params.lacp_fast);
1794                 } else {
1795                         SLAVE_AD_INFO(new_slave).id =
1796                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1797                 }
1798
1799                 bond_3ad_bind_slave(new_slave);
1800                 break;
1801         case BOND_MODE_TLB:
1802         case BOND_MODE_ALB:
1803                 new_slave->state = BOND_STATE_ACTIVE;
1804                 bond_set_slave_inactive_flags(new_slave);
1805                 bond_select_active_slave(bond);
1806                 break;
1807         default:
1808                 pr_debug("This slave is always active in trunk mode\n");
1809
1810                 /* always active in trunk mode */
1811                 new_slave->state = BOND_STATE_ACTIVE;
1812
1813                 /* In trunking mode there is little meaning to curr_active_slave
1814                  * anyway (it holds no special properties of the bond device),
1815                  * so we can change it without calling change_active_interface()
1816                  */
1817                 if (!bond->curr_active_slave)
1818                         bond->curr_active_slave = new_slave;
1819
1820                 break;
1821         } /* switch(bond_mode) */
1822
1823         write_unlock_bh(&bond->curr_slave_lock);
1824
1825         bond_set_carrier(bond);
1826
1827 #ifdef CONFIG_NET_POLL_CONTROLLER
1828         slave_dev->npinfo = bond_netpoll_info(bond);
1829         if (slave_dev->npinfo) {
1830                 if (slave_enable_netpoll(new_slave)) {
1831                         read_unlock(&bond->lock);
1832                         pr_info("Error, %s: master_dev is using netpoll, "
1833                                  "but new slave device does not support netpoll.\n",
1834                                  bond_dev->name);
1835                         res = -EBUSY;
1836                         goto err_close;
1837                 }
1838         }
1839 #endif
1840
1841         read_unlock(&bond->lock);
1842
1843         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1844         if (res)
1845                 goto err_close;
1846
1847         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1848                 bond_dev->name, slave_dev->name,
1849                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1850                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1851
1852         /* enslave is successful */
1853         return 0;
1854
1855 /* Undo stages on error */
1856 err_close:
1857         dev_close(slave_dev);
1858
1859 err_unset_master:
1860         netdev_set_bond_master(slave_dev, NULL);
1861
1862 err_restore_mac:
1863         if (!bond->params.fail_over_mac) {
1864                 /* XXX TODO - fom follow mode needs to change master's
1865                  * MAC if this slave's MAC is in use by the bond, or at
1866                  * least print a warning.
1867                  */
1868                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1869                 addr.sa_family = slave_dev->type;
1870                 dev_set_mac_address(slave_dev, &addr);
1871         }
1872
1873 err_restore_mtu:
1874         dev_set_mtu(slave_dev, new_slave->original_mtu);
1875
1876 err_free:
1877         kfree(new_slave);
1878
1879 err_undo_flags:
1880         bond_dev->features = old_features;
1881
1882         return res;
1883 }
1884
1885 /*
1886  * Try to release the slave device <slave> from the bond device <master>
1887  * It is legal to access curr_active_slave without a lock because all the function
1888  * is write-locked.
1889  *
1890  * The rules for slave state should be:
1891  *   for Active/Backup:
1892  *     Active stays on all backups go down
1893  *   for Bonded connections:
1894  *     The first up interface should be left on and all others downed.
1895  */
1896 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1897 {
1898         struct bonding *bond = netdev_priv(bond_dev);
1899         struct slave *slave, *oldcurrent;
1900         struct sockaddr addr;
1901
1902         /* slave is not a slave or master is not master of this slave */
1903         if (!(slave_dev->flags & IFF_SLAVE) ||
1904             (slave_dev->master != bond_dev)) {
1905                 pr_err("%s: Error: cannot release %s.\n",
1906                        bond_dev->name, slave_dev->name);
1907                 return -EINVAL;
1908         }
1909
1910         block_netpoll_tx();
1911         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1912         write_lock_bh(&bond->lock);
1913
1914         slave = bond_get_slave_by_dev(bond, slave_dev);
1915         if (!slave) {
1916                 /* not a slave of this bond */
1917                 pr_info("%s: %s not enslaved\n",
1918                         bond_dev->name, slave_dev->name);
1919                 write_unlock_bh(&bond->lock);
1920                 unblock_netpoll_tx();
1921                 return -EINVAL;
1922         }
1923
1924         if (!bond->params.fail_over_mac) {
1925                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1926                     bond->slave_cnt > 1)
1927                         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",
1928                                    bond_dev->name, slave_dev->name,
1929                                    slave->perm_hwaddr,
1930                                    bond_dev->name, slave_dev->name);
1931         }
1932
1933         /* Inform AD package of unbinding of slave. */
1934         if (bond->params.mode == BOND_MODE_8023AD) {
1935                 /* must be called before the slave is
1936                  * detached from the list
1937                  */
1938                 bond_3ad_unbind_slave(slave);
1939         }
1940
1941         pr_info("%s: releasing %s interface %s\n",
1942                 bond_dev->name,
1943                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1944                 slave_dev->name);
1945
1946         oldcurrent = bond->curr_active_slave;
1947
1948         bond->current_arp_slave = NULL;
1949
1950         /* release the slave from its bond */
1951         bond_detach_slave(bond, slave);
1952
1953         bond_compute_features(bond);
1954
1955         if (bond->primary_slave == slave)
1956                 bond->primary_slave = NULL;
1957
1958         if (oldcurrent == slave)
1959                 bond_change_active_slave(bond, NULL);
1960
1961         if (bond_is_lb(bond)) {
1962                 /* Must be called only after the slave has been
1963                  * detached from the list and the curr_active_slave
1964                  * has been cleared (if our_slave == old_current),
1965                  * but before a new active slave is selected.
1966                  */
1967                 write_unlock_bh(&bond->lock);
1968                 bond_alb_deinit_slave(bond, slave);
1969                 write_lock_bh(&bond->lock);
1970         }
1971
1972         if (oldcurrent == slave) {
1973                 /*
1974                  * Note that we hold RTNL over this sequence, so there
1975                  * is no concern that another slave add/remove event
1976                  * will interfere.
1977                  */
1978                 write_unlock_bh(&bond->lock);
1979                 read_lock(&bond->lock);
1980                 write_lock_bh(&bond->curr_slave_lock);
1981
1982                 bond_select_active_slave(bond);
1983
1984                 write_unlock_bh(&bond->curr_slave_lock);
1985                 read_unlock(&bond->lock);
1986                 write_lock_bh(&bond->lock);
1987         }
1988
1989         if (bond->slave_cnt == 0) {
1990                 bond_set_carrier(bond);
1991
1992                 /* if the last slave was removed, zero the mac address
1993                  * of the master so it will be set by the application
1994                  * to the mac address of the first slave
1995                  */
1996                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1997
1998                 if (!bond->vlgrp) {
1999                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2000                 } else {
2001                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2002                                    bond_dev->name, bond_dev->name);
2003                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2004                                    bond_dev->name);
2005                 }
2006         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2007                    !bond_has_challenged_slaves(bond)) {
2008                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2009                         bond_dev->name, slave_dev->name, bond_dev->name);
2010                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2011         }
2012
2013         write_unlock_bh(&bond->lock);
2014         unblock_netpoll_tx();
2015
2016         /* must do this from outside any spinlocks */
2017         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2018
2019         bond_del_vlans_from_slave(bond, slave_dev);
2020
2021         /* If the mode USES_PRIMARY, then we should only remove its
2022          * promisc and mc settings if it was the curr_active_slave, but that was
2023          * already taken care of above when we detached the slave
2024          */
2025         if (!USES_PRIMARY(bond->params.mode)) {
2026                 /* unset promiscuity level from slave */
2027                 if (bond_dev->flags & IFF_PROMISC)
2028                         dev_set_promiscuity(slave_dev, -1);
2029
2030                 /* unset allmulti level from slave */
2031                 if (bond_dev->flags & IFF_ALLMULTI)
2032                         dev_set_allmulti(slave_dev, -1);
2033
2034                 /* flush master's mc_list from slave */
2035                 netif_addr_lock_bh(bond_dev);
2036                 bond_mc_list_flush(bond_dev, slave_dev);
2037                 netif_addr_unlock_bh(bond_dev);
2038         }
2039
2040         netdev_set_bond_master(slave_dev, NULL);
2041
2042         slave_disable_netpoll(slave);
2043
2044         /* close slave before restoring its mac address */
2045         dev_close(slave_dev);
2046
2047         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2048                 /* restore original ("permanent") mac address */
2049                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2050                 addr.sa_family = slave_dev->type;
2051                 dev_set_mac_address(slave_dev, &addr);
2052         }
2053
2054         dev_set_mtu(slave_dev, slave->original_mtu);
2055
2056         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2057                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2058                                    IFF_SLAVE_NEEDARP);
2059
2060         kfree(slave);
2061
2062         return 0;  /* deletion OK */
2063 }
2064
2065 /*
2066 * First release a slave and than destroy the bond if no more slaves are left.
2067 * Must be under rtnl_lock when this function is called.
2068 */
2069 static int  bond_release_and_destroy(struct net_device *bond_dev,
2070                                      struct net_device *slave_dev)
2071 {
2072         struct bonding *bond = netdev_priv(bond_dev);
2073         int ret;
2074
2075         ret = bond_release(bond_dev, slave_dev);
2076         if ((ret == 0) && (bond->slave_cnt == 0)) {
2077                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2078                 pr_info("%s: destroying bond %s.\n",
2079                         bond_dev->name, bond_dev->name);
2080                 unregister_netdevice(bond_dev);
2081         }
2082         return ret;
2083 }
2084
2085 /*
2086  * This function releases all slaves.
2087  */
2088 static int bond_release_all(struct net_device *bond_dev)
2089 {
2090         struct bonding *bond = netdev_priv(bond_dev);
2091         struct slave *slave;
2092         struct net_device *slave_dev;
2093         struct sockaddr addr;
2094
2095         write_lock_bh(&bond->lock);
2096
2097         netif_carrier_off(bond_dev);
2098
2099         if (bond->slave_cnt == 0)
2100                 goto out;
2101
2102         bond->current_arp_slave = NULL;
2103         bond->primary_slave = NULL;
2104         bond_change_active_slave(bond, NULL);
2105
2106         while ((slave = bond->first_slave) != NULL) {
2107                 /* Inform AD package of unbinding of slave
2108                  * before slave is detached from the list.
2109                  */
2110                 if (bond->params.mode == BOND_MODE_8023AD)
2111                         bond_3ad_unbind_slave(slave);
2112
2113                 slave_dev = slave->dev;
2114                 bond_detach_slave(bond, slave);
2115
2116                 /* now that the slave is detached, unlock and perform
2117                  * all the undo steps that should not be called from
2118                  * within a lock.
2119                  */
2120                 write_unlock_bh(&bond->lock);
2121
2122                 if (bond_is_lb(bond)) {
2123                         /* must be called only after the slave
2124                          * has been detached from the list
2125                          */
2126                         bond_alb_deinit_slave(bond, slave);
2127                 }
2128
2129                 bond_compute_features(bond);
2130
2131                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2132                 bond_del_vlans_from_slave(bond, slave_dev);
2133
2134                 /* If the mode USES_PRIMARY, then we should only remove its
2135                  * promisc and mc settings if it was the curr_active_slave, but that was
2136                  * already taken care of above when we detached the slave
2137                  */
2138                 if (!USES_PRIMARY(bond->params.mode)) {
2139                         /* unset promiscuity level from slave */
2140                         if (bond_dev->flags & IFF_PROMISC)
2141                                 dev_set_promiscuity(slave_dev, -1);
2142
2143                         /* unset allmulti level from slave */
2144                         if (bond_dev->flags & IFF_ALLMULTI)
2145                                 dev_set_allmulti(slave_dev, -1);
2146
2147                         /* flush master's mc_list from slave */
2148                         netif_addr_lock_bh(bond_dev);
2149                         bond_mc_list_flush(bond_dev, slave_dev);
2150                         netif_addr_unlock_bh(bond_dev);
2151                 }
2152
2153                 netdev_set_bond_master(slave_dev, NULL);
2154
2155                 slave_disable_netpoll(slave);
2156
2157                 /* close slave before restoring its mac address */
2158                 dev_close(slave_dev);
2159
2160                 if (!bond->params.fail_over_mac) {
2161                         /* restore original ("permanent") mac address*/
2162                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2163                         addr.sa_family = slave_dev->type;
2164                         dev_set_mac_address(slave_dev, &addr);
2165                 }
2166
2167                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2168                                            IFF_SLAVE_INACTIVE);
2169
2170                 kfree(slave);
2171
2172                 /* re-acquire the lock before getting the next slave */
2173                 write_lock_bh(&bond->lock);
2174         }
2175
2176         /* zero the mac address of the master so it will be
2177          * set by the application to the mac address of the
2178          * first slave
2179          */
2180         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2181
2182         if (!bond->vlgrp) {
2183                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2184         } else {
2185                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2186                            bond_dev->name, bond_dev->name);
2187                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2188                            bond_dev->name);
2189         }
2190
2191         pr_info("%s: released all slaves\n", bond_dev->name);
2192
2193 out:
2194         write_unlock_bh(&bond->lock);
2195         return 0;
2196 }
2197
2198 /*
2199  * This function changes the active slave to slave <slave_dev>.
2200  * It returns -EINVAL in the following cases.
2201  *  - <slave_dev> is not found in the list.
2202  *  - There is not active slave now.
2203  *  - <slave_dev> is already active.
2204  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2205  *  - <slave_dev> is not running.
2206  * In these cases, this function does nothing.
2207  * In the other cases, current_slave pointer is changed and 0 is returned.
2208  */
2209 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2210 {
2211         struct bonding *bond = netdev_priv(bond_dev);
2212         struct slave *old_active = NULL;
2213         struct slave *new_active = NULL;
2214         int res = 0;
2215
2216         if (!USES_PRIMARY(bond->params.mode))
2217                 return -EINVAL;
2218
2219         /* Verify that master_dev is indeed the master of slave_dev */
2220         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2221                 return -EINVAL;
2222
2223         read_lock(&bond->lock);
2224
2225         read_lock(&bond->curr_slave_lock);
2226         old_active = bond->curr_active_slave;
2227         read_unlock(&bond->curr_slave_lock);
2228
2229         new_active = bond_get_slave_by_dev(bond, slave_dev);
2230
2231         /*
2232          * Changing to the current active: do nothing; return success.
2233          */
2234         if (new_active && (new_active == old_active)) {
2235                 read_unlock(&bond->lock);
2236                 return 0;
2237         }
2238
2239         if ((new_active) &&
2240             (old_active) &&
2241             (new_active->link == BOND_LINK_UP) &&
2242             IS_UP(new_active->dev)) {
2243                 block_netpoll_tx();
2244                 write_lock_bh(&bond->curr_slave_lock);
2245                 bond_change_active_slave(bond, new_active);
2246                 write_unlock_bh(&bond->curr_slave_lock);
2247                 unblock_netpoll_tx();
2248         } else
2249                 res = -EINVAL;
2250
2251         read_unlock(&bond->lock);
2252
2253         return res;
2254 }
2255
2256 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2257 {
2258         struct bonding *bond = netdev_priv(bond_dev);
2259
2260         info->bond_mode = bond->params.mode;
2261         info->miimon = bond->params.miimon;
2262
2263         read_lock(&bond->lock);
2264         info->num_slaves = bond->slave_cnt;
2265         read_unlock(&bond->lock);
2266
2267         return 0;
2268 }
2269
2270 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2271 {
2272         struct bonding *bond = netdev_priv(bond_dev);
2273         struct slave *slave;
2274         int i, res = -ENODEV;
2275
2276         read_lock(&bond->lock);
2277
2278         bond_for_each_slave(bond, slave, i) {
2279                 if (i == (int)info->slave_id) {
2280                         res = 0;
2281                         strcpy(info->slave_name, slave->dev->name);
2282                         info->link = slave->link;
2283                         info->state = slave->state;
2284                         info->link_failure_count = slave->link_failure_count;
2285                         break;
2286                 }
2287         }
2288
2289         read_unlock(&bond->lock);
2290
2291         return res;
2292 }
2293
2294 /*-------------------------------- Monitoring -------------------------------*/
2295
2296
2297 static int bond_miimon_inspect(struct bonding *bond)
2298 {
2299         struct slave *slave;
2300         int i, link_state, commit = 0;
2301         bool ignore_updelay;
2302
2303         ignore_updelay = !bond->curr_active_slave ? true : false;
2304
2305         bond_for_each_slave(bond, slave, i) {
2306                 slave->new_link = BOND_LINK_NOCHANGE;
2307
2308                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2309
2310                 switch (slave->link) {
2311                 case BOND_LINK_UP:
2312                         if (link_state)
2313                                 continue;
2314
2315                         slave->link = BOND_LINK_FAIL;
2316                         slave->delay = bond->params.downdelay;
2317                         if (slave->delay) {
2318                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2319                                         bond->dev->name,
2320                                         (bond->params.mode ==
2321                                          BOND_MODE_ACTIVEBACKUP) ?
2322                                         ((slave->state == BOND_STATE_ACTIVE) ?
2323                                          "active " : "backup ") : "",
2324                                         slave->dev->name,
2325                                         bond->params.downdelay * bond->params.miimon);
2326                         }
2327                         /*FALLTHRU*/
2328                 case BOND_LINK_FAIL:
2329                         if (link_state) {
2330                                 /*
2331                                  * recovered before downdelay expired
2332                                  */
2333                                 slave->link = BOND_LINK_UP;
2334                                 slave->jiffies = jiffies;
2335                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2336                                         bond->dev->name,
2337                                         (bond->params.downdelay - slave->delay) *
2338                                         bond->params.miimon,
2339                                         slave->dev->name);
2340                                 continue;
2341                         }
2342
2343                         if (slave->delay <= 0) {
2344                                 slave->new_link = BOND_LINK_DOWN;
2345                                 commit++;
2346                                 continue;
2347                         }
2348
2349                         slave->delay--;
2350                         break;
2351
2352                 case BOND_LINK_DOWN:
2353                         if (!link_state)
2354                                 continue;
2355
2356                         slave->link = BOND_LINK_BACK;
2357                         slave->delay = bond->params.updelay;
2358
2359                         if (slave->delay) {
2360                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2361                                         bond->dev->name, slave->dev->name,
2362                                         ignore_updelay ? 0 :
2363                                         bond->params.updelay *
2364                                         bond->params.miimon);
2365                         }
2366                         /*FALLTHRU*/
2367                 case BOND_LINK_BACK:
2368                         if (!link_state) {
2369                                 slave->link = BOND_LINK_DOWN;
2370                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2371                                         bond->dev->name,
2372                                         (bond->params.updelay - slave->delay) *
2373                                         bond->params.miimon,
2374                                         slave->dev->name);
2375
2376                                 continue;
2377                         }
2378
2379                         if (ignore_updelay)
2380                                 slave->delay = 0;
2381
2382                         if (slave->delay <= 0) {
2383                                 slave->new_link = BOND_LINK_UP;
2384                                 commit++;
2385                                 ignore_updelay = false;
2386                                 continue;
2387                         }
2388
2389                         slave->delay--;
2390                         break;
2391                 }
2392         }
2393
2394         return commit;
2395 }
2396
2397 static void bond_miimon_commit(struct bonding *bond)
2398 {
2399         struct slave *slave;
2400         int i;
2401
2402         bond_for_each_slave(bond, slave, i) {
2403                 switch (slave->new_link) {
2404                 case BOND_LINK_NOCHANGE:
2405                         continue;
2406
2407                 case BOND_LINK_UP:
2408                         slave->link = BOND_LINK_UP;
2409                         slave->jiffies = jiffies;
2410
2411                         if (bond->params.mode == BOND_MODE_8023AD) {
2412                                 /* prevent it from being the active one */
2413                                 slave->state = BOND_STATE_BACKUP;
2414                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2415                                 /* make it immediately active */
2416                                 slave->state = BOND_STATE_ACTIVE;
2417                         } else if (slave != bond->primary_slave) {
2418                                 /* prevent it from being the active one */
2419                                 slave->state = BOND_STATE_BACKUP;
2420                         }
2421
2422                         bond_update_speed_duplex(slave);
2423
2424                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2425                                 bond->dev->name, slave->dev->name,
2426                                 slave->speed, slave->duplex ? "full" : "half");
2427
2428                         /* notify ad that the link status has changed */
2429                         if (bond->params.mode == BOND_MODE_8023AD)
2430                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2431
2432                         if (bond_is_lb(bond))
2433                                 bond_alb_handle_link_change(bond, slave,
2434                                                             BOND_LINK_UP);
2435
2436                         if (!bond->curr_active_slave ||
2437                             (slave == bond->primary_slave))
2438                                 goto do_failover;
2439
2440                         continue;
2441
2442                 case BOND_LINK_DOWN:
2443                         if (slave->link_failure_count < UINT_MAX)
2444                                 slave->link_failure_count++;
2445
2446                         slave->link = BOND_LINK_DOWN;
2447
2448                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2449                             bond->params.mode == BOND_MODE_8023AD)
2450                                 bond_set_slave_inactive_flags(slave);
2451
2452                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2453                                 bond->dev->name, slave->dev->name);
2454
2455                         if (bond->params.mode == BOND_MODE_8023AD)
2456                                 bond_3ad_handle_link_change(slave,
2457                                                             BOND_LINK_DOWN);
2458
2459                         if (bond_is_lb(bond))
2460                                 bond_alb_handle_link_change(bond, slave,
2461                                                             BOND_LINK_DOWN);
2462
2463                         if (slave == bond->curr_active_slave)
2464                                 goto do_failover;
2465
2466                         continue;
2467
2468                 default:
2469                         pr_err("%s: invalid new link %d on slave %s\n",
2470                                bond->dev->name, slave->new_link,
2471                                slave->dev->name);
2472                         slave->new_link = BOND_LINK_NOCHANGE;
2473
2474                         continue;
2475                 }
2476
2477 do_failover:
2478                 ASSERT_RTNL();
2479                 block_netpoll_tx();
2480                 write_lock_bh(&bond->curr_slave_lock);
2481                 bond_select_active_slave(bond);
2482                 write_unlock_bh(&bond->curr_slave_lock);
2483                 unblock_netpoll_tx();
2484         }
2485
2486         bond_set_carrier(bond);
2487 }
2488
2489 /*
2490  * bond_mii_monitor
2491  *
2492  * Really a wrapper that splits the mii monitor into two phases: an
2493  * inspection, then (if inspection indicates something needs to be done)
2494  * an acquisition of appropriate locks followed by a commit phase to
2495  * implement whatever link state changes are indicated.
2496  */
2497 void bond_mii_monitor(struct work_struct *work)
2498 {
2499         struct bonding *bond = container_of(work, struct bonding,
2500                                             mii_work.work);
2501
2502         read_lock(&bond->lock);
2503         if (bond->kill_timers)
2504                 goto out;
2505
2506         if (bond->slave_cnt == 0)
2507                 goto re_arm;
2508
2509         if (bond->send_grat_arp) {
2510                 read_lock(&bond->curr_slave_lock);
2511                 bond_send_gratuitous_arp(bond);
2512                 read_unlock(&bond->curr_slave_lock);
2513         }
2514
2515         if (bond->send_unsol_na) {
2516                 read_lock(&bond->curr_slave_lock);
2517                 bond_send_unsolicited_na(bond);
2518                 read_unlock(&bond->curr_slave_lock);
2519         }
2520
2521         if (bond_miimon_inspect(bond)) {
2522                 read_unlock(&bond->lock);
2523                 rtnl_lock();
2524                 read_lock(&bond->lock);
2525
2526                 bond_miimon_commit(bond);
2527
2528                 read_unlock(&bond->lock);
2529                 rtnl_unlock();  /* might sleep, hold no other locks */
2530                 read_lock(&bond->lock);
2531         }
2532
2533 re_arm:
2534         if (bond->params.miimon)
2535                 queue_delayed_work(bond->wq, &bond->mii_work,
2536                                    msecs_to_jiffies(bond->params.miimon));
2537 out:
2538         read_unlock(&bond->lock);
2539 }
2540
2541 static __be32 bond_glean_dev_ip(struct net_device *dev)
2542 {
2543         struct in_device *idev;
2544         struct in_ifaddr *ifa;
2545         __be32 addr = 0;
2546
2547         if (!dev)
2548                 return 0;
2549
2550         rcu_read_lock();
2551         idev = __in_dev_get_rcu(dev);
2552         if (!idev)
2553                 goto out;
2554
2555         ifa = idev->ifa_list;
2556         if (!ifa)
2557                 goto out;
2558
2559         addr = ifa->ifa_local;
2560 out:
2561         rcu_read_unlock();
2562         return addr;
2563 }
2564
2565 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2566 {
2567         struct vlan_entry *vlan;
2568
2569         if (ip == bond->master_ip)
2570                 return 1;
2571
2572         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2573                 if (ip == vlan->vlan_ip)
2574                         return 1;
2575         }
2576
2577         return 0;
2578 }
2579
2580 /*
2581  * We go to the (large) trouble of VLAN tagging ARP frames because
2582  * switches in VLAN mode (especially if ports are configured as
2583  * "native" to a VLAN) might not pass non-tagged frames.
2584  */
2585 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2586 {
2587         struct sk_buff *skb;
2588
2589         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2590                  slave_dev->name, dest_ip, src_ip, vlan_id);
2591
2592         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2593                          NULL, slave_dev->dev_addr, NULL);
2594
2595         if (!skb) {
2596                 pr_err("ARP packet allocation failed\n");
2597                 return;
2598         }
2599         if (vlan_id) {
2600                 skb = vlan_put_tag(skb, vlan_id);
2601                 if (!skb) {
2602                         pr_err("failed to insert VLAN tag\n");
2603                         return;
2604                 }
2605         }
2606         arp_xmit(skb);
2607 }
2608
2609
2610 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2611 {
2612         int i, vlan_id, rv;
2613         __be32 *targets = bond->params.arp_targets;
2614         struct vlan_entry *vlan;
2615         struct net_device *vlan_dev;
2616         struct flowi fl;
2617         struct rtable *rt;
2618
2619         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2620                 if (!targets[i])
2621                         break;
2622                 pr_debug("basa: target %x\n", targets[i]);
2623                 if (!bond->vlgrp) {
2624                         pr_debug("basa: empty vlan: arp_send\n");
2625                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2626                                       bond->master_ip, 0);
2627                         continue;
2628                 }
2629
2630                 /*
2631                  * If VLANs are configured, we do a route lookup to
2632                  * determine which VLAN interface would be used, so we
2633                  * can tag the ARP with the proper VLAN tag.
2634                  */
2635                 memset(&fl, 0, sizeof(fl));
2636                 fl.fl4_dst = targets[i];
2637                 fl.fl4_tos = RTO_ONLINK;
2638
2639                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2640                 if (rv) {
2641                         if (net_ratelimit()) {
2642                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2643                                            bond->dev->name, &fl.fl4_dst);
2644                         }
2645                         continue;
2646                 }
2647
2648                 /*
2649                  * This target is not on a VLAN
2650                  */
2651                 if (rt->dst.dev == bond->dev) {
2652                         ip_rt_put(rt);
2653                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2654                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2655                                       bond->master_ip, 0);
2656                         continue;
2657                 }
2658
2659                 vlan_id = 0;
2660                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2661                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2662                         if (vlan_dev == rt->dst.dev) {
2663                                 vlan_id = vlan->vlan_id;
2664                                 pr_debug("basa: vlan match on %s %d\n",
2665                                        vlan_dev->name, vlan_id);
2666                                 break;
2667                         }
2668                 }
2669
2670                 if (vlan_id) {
2671                         ip_rt_put(rt);
2672                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2673                                       vlan->vlan_ip, vlan_id);
2674                         continue;
2675                 }
2676
2677                 if (net_ratelimit()) {
2678                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2679                                    bond->dev->name, &fl.fl4_dst,
2680                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2681                 }
2682                 ip_rt_put(rt);
2683         }
2684 }
2685
2686 /*
2687  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2688  * for each VLAN above us.
2689  *
2690  * Caller must hold curr_slave_lock for read or better
2691  */
2692 static void bond_send_gratuitous_arp(struct bonding *bond)
2693 {
2694         struct slave *slave = bond->curr_active_slave;
2695         struct vlan_entry *vlan;
2696         struct net_device *vlan_dev;
2697
2698         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2699                  bond->dev->name, slave ? slave->dev->name : "NULL");
2700
2701         if (!slave || !bond->send_grat_arp ||
2702             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2703                 return;
2704
2705         bond->send_grat_arp--;
2706
2707         if (bond->master_ip) {
2708                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2709                                 bond->master_ip, 0);
2710         }
2711
2712         if (!bond->vlgrp)
2713                 return;
2714
2715         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2716                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2717                 if (vlan->vlan_ip) {
2718                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2719                                       vlan->vlan_ip, vlan->vlan_id);
2720                 }
2721         }
2722 }
2723
2724 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2725 {
2726         int i;
2727         __be32 *targets = bond->params.arp_targets;
2728
2729         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2730                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2731                          &sip, &tip, i, &targets[i],
2732                          bond_has_this_ip(bond, tip));
2733                 if (sip == targets[i]) {
2734                         if (bond_has_this_ip(bond, tip))
2735                                 slave->last_arp_rx = jiffies;
2736                         return;
2737                 }
2738         }
2739 }
2740
2741 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2742 {
2743         struct arphdr *arp;
2744         struct slave *slave;
2745         struct bonding *bond;
2746         unsigned char *arp_ptr;
2747         __be32 sip, tip;
2748
2749         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2750                 /*
2751                  * When using VLANS and bonding, dev and oriv_dev may be
2752                  * incorrect if the physical interface supports VLAN
2753                  * acceleration.  With this change ARP validation now
2754                  * works for hosts only reachable on the VLAN interface.
2755                  */
2756                 dev = vlan_dev_real_dev(dev);
2757                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2758         }
2759
2760         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2761                 goto out;
2762
2763         bond = netdev_priv(dev);
2764         read_lock(&bond->lock);
2765
2766         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2767                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2768                  orig_dev ? orig_dev->name : "NULL");
2769
2770         slave = bond_get_slave_by_dev(bond, orig_dev);
2771         if (!slave || !slave_do_arp_validate(bond, slave))
2772                 goto out_unlock;
2773
2774         skb = skb_share_check(skb, GFP_ATOMIC);
2775         if (!skb)
2776                 goto out_unlock;
2777
2778         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2779                 goto out_unlock;
2780
2781         arp = arp_hdr(skb);
2782         if (arp->ar_hln != dev->addr_len ||
2783             skb->pkt_type == PACKET_OTHERHOST ||
2784             skb->pkt_type == PACKET_LOOPBACK ||
2785             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2786             arp->ar_pro != htons(ETH_P_IP) ||
2787             arp->ar_pln != 4)
2788                 goto out_unlock;
2789
2790         arp_ptr = (unsigned char *)(arp + 1);
2791         arp_ptr += dev->addr_len;
2792         memcpy(&sip, arp_ptr, 4);
2793         arp_ptr += 4 + dev->addr_len;
2794         memcpy(&tip, arp_ptr, 4);
2795
2796         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2797                  bond->dev->name, slave->dev->name, slave->state,
2798                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2799                  &sip, &tip);
2800
2801         /*
2802          * Backup slaves won't see the ARP reply, but do come through
2803          * here for each ARP probe (so we swap the sip/tip to validate
2804          * the probe).  In a "redundant switch, common router" type of
2805          * configuration, the ARP probe will (hopefully) travel from
2806          * the active, through one switch, the router, then the other
2807          * switch before reaching the backup.
2808          */
2809         if (slave->state == BOND_STATE_ACTIVE)
2810                 bond_validate_arp(bond, slave, sip, tip);
2811         else
2812                 bond_validate_arp(bond, slave, tip, sip);
2813
2814 out_unlock:
2815         read_unlock(&bond->lock);
2816 out:
2817         dev_kfree_skb(skb);
2818         return NET_RX_SUCCESS;
2819 }
2820
2821 /*
2822  * this function is called regularly to monitor each slave's link
2823  * ensuring that traffic is being sent and received when arp monitoring
2824  * is used in load-balancing mode. if the adapter has been dormant, then an
2825  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2826  * arp monitoring in active backup mode.
2827  */
2828 void bond_loadbalance_arp_mon(struct work_struct *work)
2829 {
2830         struct bonding *bond = container_of(work, struct bonding,
2831                                             arp_work.work);
2832         struct slave *slave, *oldcurrent;
2833         int do_failover = 0;
2834         int delta_in_ticks;
2835         int i;
2836
2837         read_lock(&bond->lock);
2838
2839         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2840
2841         if (bond->kill_timers)
2842                 goto out;
2843
2844         if (bond->slave_cnt == 0)
2845                 goto re_arm;
2846
2847         read_lock(&bond->curr_slave_lock);
2848         oldcurrent = bond->curr_active_slave;
2849         read_unlock(&bond->curr_slave_lock);
2850
2851         /* see if any of the previous devices are up now (i.e. they have
2852          * xmt and rcv traffic). the curr_active_slave does not come into
2853          * the picture unless it is null. also, slave->jiffies is not needed
2854          * here because we send an arp on each slave and give a slave as
2855          * long as it needs to get the tx/rx within the delta.
2856          * TODO: what about up/down delay in arp mode? it wasn't here before
2857          *       so it can wait
2858          */
2859         bond_for_each_slave(bond, slave, i) {
2860                 unsigned long trans_start = dev_trans_start(slave->dev);
2861
2862                 if (slave->link != BOND_LINK_UP) {
2863                         if (time_in_range(jiffies,
2864                                 trans_start - delta_in_ticks,
2865                                 trans_start + delta_in_ticks) &&
2866                             time_in_range(jiffies,
2867                                 slave->dev->last_rx - delta_in_ticks,
2868                                 slave->dev->last_rx + delta_in_ticks)) {
2869
2870                                 slave->link  = BOND_LINK_UP;
2871                                 slave->state = BOND_STATE_ACTIVE;
2872
2873                                 /* primary_slave has no meaning in round-robin
2874                                  * mode. the window of a slave being up and
2875                                  * curr_active_slave being null after enslaving
2876                                  * is closed.
2877                                  */
2878                                 if (!oldcurrent) {
2879                                         pr_info("%s: link status definitely up for interface %s, ",
2880                                                 bond->dev->name,
2881                                                 slave->dev->name);
2882                                         do_failover = 1;
2883                                 } else {
2884                                         pr_info("%s: interface %s is now up\n",
2885                                                 bond->dev->name,
2886                                                 slave->dev->name);
2887                                 }
2888                         }
2889                 } else {
2890                         /* slave->link == BOND_LINK_UP */
2891
2892                         /* not all switches will respond to an arp request
2893                          * when the source ip is 0, so don't take the link down
2894                          * if we don't know our ip yet
2895                          */
2896                         if (!time_in_range(jiffies,
2897                                 trans_start - delta_in_ticks,
2898                                 trans_start + 2 * delta_in_ticks) ||
2899                             !time_in_range(jiffies,
2900                                 slave->dev->last_rx - delta_in_ticks,
2901                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2902
2903                                 slave->link  = BOND_LINK_DOWN;
2904                                 slave->state = BOND_STATE_BACKUP;
2905
2906                                 if (slave->link_failure_count < UINT_MAX)
2907                                         slave->link_failure_count++;
2908
2909                                 pr_info("%s: interface %s is now down.\n",
2910                                         bond->dev->name,
2911                                         slave->dev->name);
2912
2913                                 if (slave == oldcurrent)
2914                                         do_failover = 1;
2915                         }
2916                 }
2917
2918                 /* note: if switch is in round-robin mode, all links
2919                  * must tx arp to ensure all links rx an arp - otherwise
2920                  * links may oscillate or not come up at all; if switch is
2921                  * in something like xor mode, there is nothing we can
2922                  * do - all replies will be rx'ed on same link causing slaves
2923                  * to be unstable during low/no traffic periods
2924                  */
2925                 if (IS_UP(slave->dev))
2926                         bond_arp_send_all(bond, slave);
2927         }
2928
2929         if (do_failover) {
2930                 block_netpoll_tx();
2931                 write_lock_bh(&bond->curr_slave_lock);
2932
2933                 bond_select_active_slave(bond);
2934
2935                 write_unlock_bh(&bond->curr_slave_lock);
2936                 unblock_netpoll_tx();
2937         }
2938
2939 re_arm:
2940         if (bond->params.arp_interval)
2941                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2942 out:
2943         read_unlock(&bond->lock);
2944 }
2945
2946 /*
2947  * Called to inspect slaves for active-backup mode ARP monitor link state
2948  * changes.  Sets new_link in slaves to specify what action should take
2949  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2950  * to link states must be committed.
2951  *
2952  * Called with bond->lock held for read.
2953  */
2954 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2955 {
2956         struct slave *slave;
2957         int i, commit = 0;
2958         unsigned long trans_start;
2959
2960         bond_for_each_slave(bond, slave, i) {
2961                 slave->new_link = BOND_LINK_NOCHANGE;
2962
2963                 if (slave->link != BOND_LINK_UP) {
2964                         if (time_in_range(jiffies,
2965                                 slave_last_rx(bond, slave) - delta_in_ticks,
2966                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2967
2968                                 slave->new_link = BOND_LINK_UP;
2969                                 commit++;
2970                         }
2971
2972                         continue;
2973                 }
2974
2975                 /*
2976                  * Give slaves 2*delta after being enslaved or made
2977                  * active.  This avoids bouncing, as the last receive
2978                  * times need a full ARP monitor cycle to be updated.
2979                  */
2980                 if (time_in_range(jiffies,
2981                                   slave->jiffies - delta_in_ticks,
2982                                   slave->jiffies + 2 * delta_in_ticks))
2983                         continue;
2984
2985                 /*
2986                  * Backup slave is down if:
2987                  * - No current_arp_slave AND
2988                  * - more than 3*delta since last receive AND
2989                  * - the bond has an IP address
2990                  *
2991                  * Note: a non-null current_arp_slave indicates
2992                  * the curr_active_slave went down and we are
2993                  * searching for a new one; under this condition
2994                  * we only take the curr_active_slave down - this
2995                  * gives each slave a chance to tx/rx traffic
2996                  * before being taken out
2997                  */
2998                 if (slave->state == BOND_STATE_BACKUP &&
2999                     !bond->current_arp_slave &&
3000                     !time_in_range(jiffies,
3001                         slave_last_rx(bond, slave) - delta_in_ticks,
3002                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3003
3004                         slave->new_link = BOND_LINK_DOWN;
3005                         commit++;
3006                 }
3007
3008                 /*
3009                  * Active slave is down if:
3010                  * - more than 2*delta since transmitting OR
3011                  * - (more than 2*delta since receive AND
3012                  *    the bond has an IP address)
3013                  */
3014                 trans_start = dev_trans_start(slave->dev);
3015                 if ((slave->state == BOND_STATE_ACTIVE) &&
3016                     (!time_in_range(jiffies,
3017                         trans_start - delta_in_ticks,
3018                         trans_start + 2 * delta_in_ticks) ||
3019                      !time_in_range(jiffies,
3020                         slave_last_rx(bond, slave) - delta_in_ticks,
3021                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3022
3023                         slave->new_link = BOND_LINK_DOWN;
3024                         commit++;
3025                 }
3026         }
3027
3028         return commit;
3029 }
3030
3031 /*
3032  * Called to commit link state changes noted by inspection step of
3033  * active-backup mode ARP monitor.
3034  *
3035  * Called with RTNL and bond->lock for read.
3036  */
3037 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3038 {
3039         struct slave *slave;
3040         int i;
3041         unsigned long trans_start;
3042
3043         bond_for_each_slave(bond, slave, i) {
3044                 switch (slave->new_link) {
3045                 case BOND_LINK_NOCHANGE:
3046                         continue;
3047
3048                 case BOND_LINK_UP:
3049                         trans_start = dev_trans_start(slave->dev);
3050                         if ((!bond->curr_active_slave &&
3051                              time_in_range(jiffies,
3052                                            trans_start - delta_in_ticks,
3053                                            trans_start + delta_in_ticks)) ||
3054                             bond->curr_active_slave != slave) {
3055                                 slave->link = BOND_LINK_UP;
3056                                 bond->current_arp_slave = NULL;
3057
3058                                 pr_info("%s: link status definitely up for interface %s.\n",
3059                                         bond->dev->name, slave->dev->name);
3060
3061                                 if (!bond->curr_active_slave ||
3062                                     (slave == bond->primary_slave))
3063                                         goto do_failover;
3064
3065                         }
3066
3067                         continue;
3068
3069                 case BOND_LINK_DOWN:
3070                         if (slave->link_failure_count < UINT_MAX)
3071                                 slave->link_failure_count++;
3072
3073                         slave->link = BOND_LINK_DOWN;
3074                         bond_set_slave_inactive_flags(slave);
3075
3076                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3077                                 bond->dev->name, slave->dev->name);
3078
3079                         if (slave == bond->curr_active_slave) {
3080                                 bond->current_arp_slave = NULL;
3081                                 goto do_failover;
3082                         }
3083
3084                         continue;
3085
3086                 default:
3087                         pr_err("%s: impossible: new_link %d on slave %s\n",
3088                                bond->dev->name, slave->new_link,
3089                                slave->dev->name);
3090                         continue;
3091                 }
3092
3093 do_failover:
3094                 ASSERT_RTNL();
3095                 block_netpoll_tx();
3096                 write_lock_bh(&bond->curr_slave_lock);
3097                 bond_select_active_slave(bond);
3098                 write_unlock_bh(&bond->curr_slave_lock);
3099                 unblock_netpoll_tx();
3100         }
3101
3102         bond_set_carrier(bond);
3103 }
3104
3105 /*
3106  * Send ARP probes for active-backup mode ARP monitor.
3107  *
3108  * Called with bond->lock held for read.
3109  */
3110 static void bond_ab_arp_probe(struct bonding *bond)
3111 {
3112         struct slave *slave;
3113         int i;
3114
3115         read_lock(&bond->curr_slave_lock);
3116
3117         if (bond->current_arp_slave && bond->curr_active_slave)
3118                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3119                         bond->current_arp_slave->dev->name,
3120                         bond->curr_active_slave->dev->name);
3121
3122         if (bond->curr_active_slave) {
3123                 bond_arp_send_all(bond, bond->curr_active_slave);
3124                 read_unlock(&bond->curr_slave_lock);
3125                 return;
3126         }
3127
3128         read_unlock(&bond->curr_slave_lock);
3129
3130         /* if we don't have a curr_active_slave, search for the next available
3131          * backup slave from the current_arp_slave and make it the candidate
3132          * for becoming the curr_active_slave
3133          */
3134
3135         if (!bond->current_arp_slave) {
3136                 bond->current_arp_slave = bond->first_slave;
3137                 if (!bond->current_arp_slave)
3138                         return;
3139         }
3140
3141         bond_set_slave_inactive_flags(bond->current_arp_slave);
3142
3143         /* search for next candidate */
3144         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3145                 if (IS_UP(slave->dev)) {
3146                         slave->link = BOND_LINK_BACK;
3147                         bond_set_slave_active_flags(slave);
3148                         bond_arp_send_all(bond, slave);
3149                         slave->jiffies = jiffies;
3150                         bond->current_arp_slave = slave;
3151                         break;
3152                 }
3153
3154                 /* if the link state is up at this point, we
3155                  * mark it down - this can happen if we have
3156                  * simultaneous link failures and
3157                  * reselect_active_interface doesn't make this
3158                  * one the current slave so it is still marked
3159                  * up when it is actually down
3160                  */
3161                 if (slave->link == BOND_LINK_UP) {
3162                         slave->link = BOND_LINK_DOWN;
3163                         if (slave->link_failure_count < UINT_MAX)
3164                                 slave->link_failure_count++;
3165
3166                         bond_set_slave_inactive_flags(slave);
3167
3168                         pr_info("%s: backup interface %s is now down.\n",
3169                                 bond->dev->name, slave->dev->name);
3170                 }
3171         }
3172 }
3173
3174 void bond_activebackup_arp_mon(struct work_struct *work)
3175 {
3176         struct bonding *bond = container_of(work, struct bonding,
3177                                             arp_work.work);
3178         int delta_in_ticks;
3179
3180         read_lock(&bond->lock);
3181
3182         if (bond->kill_timers)
3183                 goto out;
3184
3185         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3186
3187         if (bond->slave_cnt == 0)
3188                 goto re_arm;
3189
3190         if (bond->send_grat_arp) {
3191                 read_lock(&bond->curr_slave_lock);
3192                 bond_send_gratuitous_arp(bond);
3193                 read_unlock(&bond->curr_slave_lock);
3194         }
3195
3196         if (bond->send_unsol_na) {
3197                 read_lock(&bond->curr_slave_lock);
3198                 bond_send_unsolicited_na(bond);
3199                 read_unlock(&bond->curr_slave_lock);
3200         }
3201
3202         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3203                 read_unlock(&bond->lock);
3204                 rtnl_lock();
3205                 read_lock(&bond->lock);
3206
3207                 bond_ab_arp_commit(bond, delta_in_ticks);
3208
3209                 read_unlock(&bond->lock);
3210                 rtnl_unlock();
3211                 read_lock(&bond->lock);
3212         }
3213
3214         bond_ab_arp_probe(bond);
3215
3216 re_arm:
3217         if (bond->params.arp_interval)
3218                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3219 out:
3220         read_unlock(&bond->lock);
3221 }
3222
3223 /*------------------------------ proc/seq_file-------------------------------*/
3224
3225 #ifdef CONFIG_PROC_FS
3226
3227 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3228         __acquires(RCU)
3229         __acquires(&bond->lock)
3230 {
3231         struct bonding *bond = seq->private;
3232         loff_t off = 0;
3233         struct slave *slave;
3234         int i;
3235
3236         /* make sure the bond won't be taken away */
3237         rcu_read_lock();
3238         read_lock(&bond->lock);
3239
3240         if (*pos == 0)
3241                 return SEQ_START_TOKEN;
3242
3243         bond_for_each_slave(bond, slave, i) {
3244                 if (++off == *pos)
3245                         return slave;
3246         }
3247
3248         return NULL;
3249 }
3250
3251 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3252 {
3253         struct bonding *bond = seq->private;
3254         struct slave *slave = v;
3255
3256         ++*pos;
3257         if (v == SEQ_START_TOKEN)
3258                 return bond->first_slave;
3259
3260         slave = slave->next;
3261
3262         return (slave == bond->first_slave) ? NULL : slave;
3263 }
3264
3265 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3266         __releases(&bond->lock)
3267         __releases(RCU)
3268 {
3269         struct bonding *bond = seq->private;
3270
3271         read_unlock(&bond->lock);
3272         rcu_read_unlock();
3273 }
3274
3275 static void bond_info_show_master(struct seq_file *seq)
3276 {
3277         struct bonding *bond = seq->private;
3278         struct slave *curr;
3279         int i;
3280
3281         read_lock(&bond->curr_slave_lock);
3282         curr = bond->curr_active_slave;
3283         read_unlock(&bond->curr_slave_lock);
3284
3285         seq_printf(seq, "Bonding Mode: %s",
3286                    bond_mode_name(bond->params.mode));
3287
3288         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3289             bond->params.fail_over_mac)
3290                 seq_printf(seq, " (fail_over_mac %s)",
3291                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3292
3293         seq_printf(seq, "\n");
3294
3295         if (bond->params.mode == BOND_MODE_XOR ||
3296                 bond->params.mode == BOND_MODE_8023AD) {
3297                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3298                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3299                         bond->params.xmit_policy);
3300         }
3301
3302         if (USES_PRIMARY(bond->params.mode)) {
3303                 seq_printf(seq, "Primary Slave: %s",
3304                            (bond->primary_slave) ?
3305                            bond->primary_slave->dev->name : "None");
3306                 if (bond->primary_slave)
3307                         seq_printf(seq, " (primary_reselect %s)",
3308                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3309
3310                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3311                            (curr) ? curr->dev->name : "None");
3312         }
3313
3314         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3315                    "up" : "down");
3316         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3317         seq_printf(seq, "Up Delay (ms): %d\n",
3318                    bond->params.updelay * bond->params.miimon);
3319         seq_printf(seq, "Down Delay (ms): %d\n",
3320                    bond->params.downdelay * bond->params.miimon);
3321
3322
3323         /* ARP information */
3324         if (bond->params.arp_interval > 0) {
3325                 int printed = 0;
3326                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3327                                 bond->params.arp_interval);
3328
3329                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3330
3331                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3332                         if (!bond->params.arp_targets[i])
3333                                 break;
3334                         if (printed)
3335                                 seq_printf(seq, ",");
3336                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3337                         printed = 1;
3338                 }
3339                 seq_printf(seq, "\n");
3340         }
3341
3342         if (bond->params.mode == BOND_MODE_8023AD) {
3343                 struct ad_info ad_info;
3344
3345                 seq_puts(seq, "\n802.3ad info\n");
3346                 seq_printf(seq, "LACP rate: %s\n",
3347                            (bond->params.lacp_fast) ? "fast" : "slow");
3348                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3349                            ad_select_tbl[bond->params.ad_select].modename);
3350
3351                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3352                         seq_printf(seq, "bond %s has no active aggregator\n",
3353                                    bond->dev->name);
3354                 } else {
3355                         seq_printf(seq, "Active Aggregator Info:\n");
3356
3357                         seq_printf(seq, "\tAggregator ID: %d\n",
3358                                    ad_info.aggregator_id);
3359                         seq_printf(seq, "\tNumber of ports: %d\n",
3360                                    ad_info.ports);
3361                         seq_printf(seq, "\tActor Key: %d\n",
3362                                    ad_info.actor_key);
3363                         seq_printf(seq, "\tPartner Key: %d\n",
3364                                    ad_info.partner_key);
3365                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3366                                    ad_info.partner_system);
3367                 }
3368         }
3369 }
3370
3371 static void bond_info_show_slave(struct seq_file *seq,
3372                                  const struct slave *slave)
3373 {
3374         struct bonding *bond = seq->private;
3375
3376         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3377         seq_printf(seq, "MII Status: %s\n",
3378                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3379         seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3380         seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3381         seq_printf(seq, "Link Failure Count: %u\n",
3382                    slave->link_failure_count);
3383
3384         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3385
3386         if (bond->params.mode == BOND_MODE_8023AD) {
3387                 const struct aggregator *agg
3388                         = SLAVE_AD_INFO(slave).port.aggregator;
3389
3390                 if (agg)
3391                         seq_printf(seq, "Aggregator ID: %d\n",
3392                                    agg->aggregator_identifier);
3393                 else
3394                         seq_puts(seq, "Aggregator ID: N/A\n");
3395         }
3396         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3397 }
3398
3399 static int bond_info_seq_show(struct seq_file *seq, void *v)
3400 {
3401         if (v == SEQ_START_TOKEN) {
3402                 seq_printf(seq, "%s\n", version);
3403                 bond_info_show_master(seq);
3404         } else
3405                 bond_info_show_slave(seq, v);
3406
3407         return 0;
3408 }
3409
3410 static const struct seq_operations bond_info_seq_ops = {
3411         .start = bond_info_seq_start,
3412         .next  = bond_info_seq_next,
3413         .stop  = bond_info_seq_stop,
3414         .show  = bond_info_seq_show,
3415 };
3416
3417 static int bond_info_open(struct inode *inode, struct file *file)
3418 {
3419         struct seq_file *seq;
3420         struct proc_dir_entry *proc;
3421         int res;
3422
3423         res = seq_open(file, &bond_info_seq_ops);
3424         if (!res) {
3425                 /* recover the pointer buried in proc_dir_entry data */
3426                 seq = file->private_data;
3427                 proc = PDE(inode);
3428                 seq->private = proc->data;
3429         }
3430
3431         return res;
3432 }
3433
3434 static const struct file_operations bond_info_fops = {
3435         .owner   = THIS_MODULE,
3436         .open    = bond_info_open,
3437         .read    = seq_read,
3438         .llseek  = seq_lseek,
3439         .release = seq_release,
3440 };
3441
3442 static void bond_create_proc_entry(struct bonding *bond)
3443 {
3444         struct net_device *bond_dev = bond->dev;
3445         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3446
3447         if (bn->proc_dir) {
3448                 bond->proc_entry = proc_create_data(bond_dev->name,
3449                                                     S_IRUGO, bn->proc_dir,
3450                                                     &bond_info_fops, bond);
3451                 if (bond->proc_entry == NULL)
3452                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3453                                    DRV_NAME, bond_dev->name);
3454                 else
3455                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3456         }
3457 }
3458
3459 static void bond_remove_proc_entry(struct bonding *bond)
3460 {
3461         struct net_device *bond_dev = bond->dev;
3462         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3463
3464         if (bn->proc_dir && bond->proc_entry) {
3465                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3466                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3467                 bond->proc_entry = NULL;
3468         }
3469 }
3470
3471 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3472  * Caller must hold rtnl_lock.
3473  */
3474 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3475 {
3476         if (!bn->proc_dir) {
3477                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3478                 if (!bn->proc_dir)
3479                         pr_warning("Warning: cannot create /proc/net/%s\n",
3480                                    DRV_NAME);
3481         }
3482 }
3483
3484 /* Destroy the bonding directory under /proc/net, if empty.
3485  * Caller must hold rtnl_lock.
3486  */
3487 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3488 {
3489         if (bn->proc_dir) {
3490                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3491                 bn->proc_dir = NULL;
3492         }
3493 }
3494
3495 #else /* !CONFIG_PROC_FS */
3496
3497 static void bond_create_proc_entry(struct bonding *bond)
3498 {
3499 }
3500
3501 static void bond_remove_proc_entry(struct bonding *bond)
3502 {
3503 }
3504
3505 static inline void bond_create_proc_dir(struct bond_net *bn)
3506 {
3507 }
3508
3509 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3510 {
3511 }
3512
3513 #endif /* CONFIG_PROC_FS */
3514
3515
3516 /*-------------------------- netdev event handling --------------------------*/
3517
3518 /*
3519  * Change device name
3520  */
3521 static int bond_event_changename(struct bonding *bond)
3522 {
3523         bond_remove_proc_entry(bond);
3524         bond_create_proc_entry(bond);
3525
3526         bond_debug_reregister(bond);
3527
3528         return NOTIFY_DONE;
3529 }
3530
3531 static int bond_master_netdev_event(unsigned long event,
3532                                     struct net_device *bond_dev)
3533 {
3534         struct bonding *event_bond = netdev_priv(bond_dev);
3535
3536         switch (event) {
3537         case NETDEV_CHANGENAME:
3538                 return bond_event_changename(event_bond);
3539         default:
3540                 break;
3541         }
3542
3543         return NOTIFY_DONE;
3544 }
3545
3546 static int bond_slave_netdev_event(unsigned long event,
3547                                    struct net_device *slave_dev)
3548 {
3549         struct net_device *bond_dev = slave_dev->master;
3550         struct bonding *bond = netdev_priv(bond_dev);
3551
3552         switch (event) {
3553         case NETDEV_UNREGISTER:
3554                 if (bond_dev) {
3555                         if (bond->setup_by_slave)
3556                                 bond_release_and_destroy(bond_dev, slave_dev);
3557                         else
3558                                 bond_release(bond_dev, slave_dev);
3559                 }
3560                 break;
3561         case NETDEV_CHANGE:
3562                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3563                         struct slave *slave;
3564
3565                         slave = bond_get_slave_by_dev(bond, slave_dev);
3566                         if (slave) {
3567                                 u16 old_speed = slave->speed;
3568                                 u16 old_duplex = slave->duplex;
3569
3570                                 bond_update_speed_duplex(slave);
3571
3572                                 if (bond_is_lb(bond))
3573                                         break;
3574
3575                                 if (old_speed != slave->speed)
3576                                         bond_3ad_adapter_speed_changed(slave);
3577                                 if (old_duplex != slave->duplex)
3578                                         bond_3ad_adapter_duplex_changed(slave);
3579                         }
3580                 }
3581
3582                 break;
3583         case NETDEV_DOWN:
3584                 /*
3585                  * ... Or is it this?
3586                  */
3587                 break;
3588         case NETDEV_CHANGEMTU:
3589                 /*
3590                  * TODO: Should slaves be allowed to
3591                  * independently alter their MTU?  For
3592                  * an active-backup bond, slaves need
3593                  * not be the same type of device, so
3594                  * MTUs may vary.  For other modes,
3595                  * slaves arguably should have the
3596                  * same MTUs. To do this, we'd need to
3597                  * take over the slave's change_mtu
3598                  * function for the duration of their
3599                  * servitude.
3600                  */
3601                 break;
3602         case NETDEV_CHANGENAME:
3603                 /*
3604                  * TODO: handle changing the primary's name
3605                  */
3606                 break;
3607         case NETDEV_FEAT_CHANGE:
3608                 bond_compute_features(bond);
3609                 break;
3610         default:
3611                 break;
3612         }
3613
3614         return NOTIFY_DONE;
3615 }
3616
3617 /*
3618  * bond_netdev_event: handle netdev notifier chain events.
3619  *
3620  * This function receives events for the netdev chain.  The caller (an
3621  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3622  * locks for us to safely manipulate the slave devices (RTNL lock,
3623  * dev_probe_lock).
3624  */
3625 static int bond_netdev_event(struct notifier_block *this,
3626                              unsigned long event, void *ptr)
3627 {
3628         struct net_device *event_dev = (struct net_device *)ptr;
3629
3630         pr_debug("event_dev: %s, event: %lx\n",
3631                  event_dev ? event_dev->name : "None",
3632                  event);
3633
3634         if (!(event_dev->priv_flags & IFF_BONDING))
3635                 return NOTIFY_DONE;
3636
3637         if (event_dev->flags & IFF_MASTER) {
3638                 pr_debug("IFF_MASTER\n");
3639                 return bond_master_netdev_event(event, event_dev);
3640         }
3641
3642         if (event_dev->flags & IFF_SLAVE) {
3643                 pr_debug("IFF_SLAVE\n");
3644                 return bond_slave_netdev_event(event, event_dev);
3645         }
3646
3647         return NOTIFY_DONE;
3648 }
3649
3650 /*
3651  * bond_inetaddr_event: handle inetaddr notifier chain events.
3652  *
3653  * We keep track of device IPs primarily to use as source addresses in
3654  * ARP monitor probes (rather than spewing out broadcasts all the time).
3655  *
3656  * We track one IP for the main device (if it has one), plus one per VLAN.
3657  */
3658 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3659 {
3660         struct in_ifaddr *ifa = ptr;
3661         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3662         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3663         struct bonding *bond;
3664         struct vlan_entry *vlan;
3665
3666         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3667                 if (bond->dev == event_dev) {
3668                         switch (event) {
3669                         case NETDEV_UP:
3670                                 bond->master_ip = ifa->ifa_local;
3671                                 return NOTIFY_OK;
3672                         case NETDEV_DOWN:
3673                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3674                                 return NOTIFY_OK;
3675                         default:
3676                                 return NOTIFY_DONE;
3677                         }
3678                 }
3679
3680                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3681                         if (!bond->vlgrp)
3682                                 continue;
3683                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3684                         if (vlan_dev == event_dev) {
3685                                 switch (event) {
3686                                 case NETDEV_UP:
3687                                         vlan->vlan_ip = ifa->ifa_local;
3688                                         return NOTIFY_OK;
3689                                 case NETDEV_DOWN:
3690                                         vlan->vlan_ip =
3691                                                 bond_glean_dev_ip(vlan_dev);
3692                                         return NOTIFY_OK;
3693                                 default:
3694                                         return NOTIFY_DONE;
3695                                 }
3696                         }
3697                 }
3698         }
3699         return NOTIFY_DONE;
3700 }
3701
3702 static struct notifier_block bond_netdev_notifier = {
3703         .notifier_call = bond_netdev_event,
3704 };
3705
3706 static struct notifier_block bond_inetaddr_notifier = {
3707         .notifier_call = bond_inetaddr_event,
3708 };
3709
3710 /*-------------------------- Packet type handling ---------------------------*/
3711
3712 /* register to receive lacpdus on a bond */
3713 static void bond_register_lacpdu(struct bonding *bond)
3714 {
3715         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3716
3717         /* initialize packet type */
3718         pk_type->type = PKT_TYPE_LACPDU;
3719         pk_type->dev = bond->dev;
3720         pk_type->func = bond_3ad_lacpdu_recv;
3721
3722         dev_add_pack(pk_type);
3723 }
3724
3725 /* unregister to receive lacpdus on a bond */
3726 static void bond_unregister_lacpdu(struct bonding *bond)
3727 {
3728         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3729 }
3730
3731 void bond_register_arp(struct bonding *bond)
3732 {
3733         struct packet_type *pt = &bond->arp_mon_pt;
3734
3735         if (pt->type)
3736                 return;
3737
3738         pt->type = htons(ETH_P_ARP);
3739         pt->dev = bond->dev;
3740         pt->func = bond_arp_rcv;
3741         dev_add_pack(pt);
3742 }
3743
3744 void bond_unregister_arp(struct bonding *bond)
3745 {
3746         struct packet_type *pt = &bond->arp_mon_pt;
3747
3748         dev_remove_pack(pt);
3749         pt->type = 0;
3750 }
3751
3752 /*---------------------------- Hashing Policies -----------------------------*/
3753
3754 /*
3755  * Hash for the output device based upon layer 2 and layer 3 data. If
3756  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3757  */
3758 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3759 {
3760         struct ethhdr *data = (struct ethhdr *)skb->data;
3761         struct iphdr *iph = ip_hdr(skb);
3762
3763         if (skb->protocol == htons(ETH_P_IP)) {
3764                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3765                         (data->h_dest[5] ^ data->h_source[5])) % count;
3766         }
3767
3768         return (data->h_dest[5] ^ data->h_source[5]) % count;
3769 }
3770
3771 /*
3772  * Hash for the output device based upon layer 3 and layer 4 data. If
3773  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3774  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3775  */
3776 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3777 {
3778         struct ethhdr *data = (struct ethhdr *)skb->data;
3779         struct iphdr *iph = ip_hdr(skb);
3780         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3781         int layer4_xor = 0;
3782
3783         if (skb->protocol == htons(ETH_P_IP)) {
3784                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3785                     (iph->protocol == IPPROTO_TCP ||
3786                      iph->protocol == IPPROTO_UDP)) {
3787                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3788                 }
3789                 return (layer4_xor ^
3790                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3791
3792         }
3793
3794         return (data->h_dest[5] ^ data->h_source[5]) % count;
3795 }
3796
3797 /*
3798  * Hash for the output device based upon layer 2 data
3799  */
3800 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3801 {
3802         struct ethhdr *data = (struct ethhdr *)skb->data;
3803
3804         return (data->h_dest[5] ^ data->h_source[5]) % count;
3805 }
3806
3807 /*-------------------------- Device entry points ----------------------------*/
3808
3809 static int bond_open(struct net_device *bond_dev)
3810 {
3811         struct bonding *bond = netdev_priv(bond_dev);
3812
3813         bond->kill_timers = 0;
3814
3815         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3816
3817         if (bond_is_lb(bond)) {
3818                 /* bond_alb_initialize must be called before the timer
3819                  * is started.
3820                  */
3821                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3822                         /* something went wrong - fail the open operation */
3823                         return -ENOMEM;
3824                 }
3825
3826                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3827                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3828         }
3829
3830         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3831                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3832                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3833         }
3834
3835         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3836                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3837                         INIT_DELAYED_WORK(&bond->arp_work,
3838                                           bond_activebackup_arp_mon);
3839                 else
3840                         INIT_DELAYED_WORK(&bond->arp_work,
3841                                           bond_loadbalance_arp_mon);
3842
3843                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3844                 if (bond->params.arp_validate)
3845                         bond_register_arp(bond);
3846         }
3847
3848         if (bond->params.mode == BOND_MODE_8023AD) {
3849                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3850                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3851                 /* register to receive LACPDUs */
3852                 bond_register_lacpdu(bond);
3853                 bond_3ad_initiate_agg_selection(bond, 1);
3854         }
3855
3856         return 0;
3857 }
3858
3859 static int bond_close(struct net_device *bond_dev)
3860 {
3861         struct bonding *bond = netdev_priv(bond_dev);
3862
3863         if (bond->params.mode == BOND_MODE_8023AD) {
3864                 /* Unregister the receive of LACPDUs */
3865                 bond_unregister_lacpdu(bond);
3866         }
3867
3868         if (bond->params.arp_validate)
3869                 bond_unregister_arp(bond);
3870
3871         write_lock_bh(&bond->lock);
3872
3873         bond->send_grat_arp = 0;
3874         bond->send_unsol_na = 0;
3875
3876         /* signal timers not to re-arm */
3877         bond->kill_timers = 1;
3878
3879         write_unlock_bh(&bond->lock);
3880
3881         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3882                 cancel_delayed_work(&bond->mii_work);
3883         }
3884
3885         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3886                 cancel_delayed_work(&bond->arp_work);
3887         }
3888
3889         switch (bond->params.mode) {
3890         case BOND_MODE_8023AD:
3891                 cancel_delayed_work(&bond->ad_work);
3892                 break;
3893         case BOND_MODE_TLB:
3894         case BOND_MODE_ALB:
3895                 cancel_delayed_work(&bond->alb_work);
3896                 break;
3897         default:
3898                 break;
3899         }
3900
3901         if (delayed_work_pending(&bond->mcast_work))
3902                 cancel_delayed_work(&bond->mcast_work);
3903
3904         if (bond_is_lb(bond)) {
3905                 /* Must be called only after all
3906                  * slaves have been released
3907                  */
3908                 bond_alb_deinitialize(bond);
3909         }
3910
3911         return 0;
3912 }
3913
3914 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3915                                                 struct rtnl_link_stats64 *stats)
3916 {
3917         struct bonding *bond = netdev_priv(bond_dev);
3918         struct rtnl_link_stats64 temp;
3919         struct slave *slave;
3920         int i;
3921
3922         memset(stats, 0, sizeof(*stats));
3923
3924         read_lock_bh(&bond->lock);
3925
3926         bond_for_each_slave(bond, slave, i) {
3927                 const struct rtnl_link_stats64 *sstats =
3928                         dev_get_stats(slave->dev, &temp);
3929
3930                 stats->rx_packets += sstats->rx_packets;
3931                 stats->rx_bytes += sstats->rx_bytes;
3932                 stats->rx_errors += sstats->rx_errors;
3933                 stats->rx_dropped += sstats->rx_dropped;
3934
3935                 stats->tx_packets += sstats->tx_packets;
3936                 stats->tx_bytes += sstats->tx_bytes;
3937                 stats->tx_errors += sstats->tx_errors;
3938                 stats->tx_dropped += sstats->tx_dropped;
3939
3940                 stats->multicast += sstats->multicast;
3941                 stats->collisions += sstats->collisions;
3942
3943                 stats->rx_length_errors += sstats->rx_length_errors;
3944                 stats->rx_over_errors += sstats->rx_over_errors;
3945                 stats->rx_crc_errors += sstats->rx_crc_errors;
3946                 stats->rx_frame_errors += sstats->rx_frame_errors;
3947                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3948                 stats->rx_missed_errors += sstats->rx_missed_errors;
3949
3950                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3951                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3952                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3953                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3954                 stats->tx_window_errors += sstats->tx_window_errors;
3955         }
3956
3957         read_unlock_bh(&bond->lock);
3958
3959         return stats;
3960 }
3961
3962 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3963 {
3964         struct net_device *slave_dev = NULL;
3965         struct ifbond k_binfo;
3966         struct ifbond __user *u_binfo = NULL;
3967         struct ifslave k_sinfo;
3968         struct ifslave __user *u_sinfo = NULL;
3969         struct mii_ioctl_data *mii = NULL;
3970         int res = 0;
3971
3972         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3973
3974         switch (cmd) {
3975         case SIOCGMIIPHY:
3976                 mii = if_mii(ifr);
3977                 if (!mii)
3978                         return -EINVAL;
3979
3980                 mii->phy_id = 0;
3981                 /* Fall Through */
3982         case SIOCGMIIREG:
3983                 /*
3984                  * We do this again just in case we were called by SIOCGMIIREG
3985                  * instead of SIOCGMIIPHY.
3986                  */
3987                 mii = if_mii(ifr);
3988                 if (!mii)
3989                         return -EINVAL;
3990
3991
3992                 if (mii->reg_num == 1) {
3993                         struct bonding *bond = netdev_priv(bond_dev);
3994                         mii->val_out = 0;
3995                         read_lock(&bond->lock);
3996                         read_lock(&bond->curr_slave_lock);
3997                         if (netif_carrier_ok(bond->dev))
3998                                 mii->val_out = BMSR_LSTATUS;
3999
4000                         read_unlock(&bond->curr_slave_lock);
4001                         read_unlock(&bond->lock);
4002                 }
4003
4004                 return 0;
4005         case BOND_INFO_QUERY_OLD:
4006         case SIOCBONDINFOQUERY:
4007                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4008
4009                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4010                         return -EFAULT;
4011
4012                 res = bond_info_query(bond_dev, &k_binfo);
4013                 if (res == 0 &&
4014                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4015                         return -EFAULT;
4016
4017                 return res;
4018         case BOND_SLAVE_INFO_QUERY_OLD:
4019         case SIOCBONDSLAVEINFOQUERY:
4020                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4021
4022                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4023                         return -EFAULT;
4024
4025                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4026                 if (res == 0 &&
4027                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4028                         return -EFAULT;
4029
4030                 return res;
4031         default:
4032                 /* Go on */
4033                 break;
4034         }
4035
4036         if (!capable(CAP_NET_ADMIN))
4037                 return -EPERM;
4038
4039         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4040
4041         pr_debug("slave_dev=%p:\n", slave_dev);
4042
4043         if (!slave_dev)
4044                 res = -ENODEV;
4045         else {
4046                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4047                 switch (cmd) {
4048                 case BOND_ENSLAVE_OLD:
4049                 case SIOCBONDENSLAVE:
4050                         res = bond_enslave(bond_dev, slave_dev);
4051                         break;
4052                 case BOND_RELEASE_OLD:
4053                 case SIOCBONDRELEASE:
4054                         res = bond_release(bond_dev, slave_dev);
4055                         break;
4056                 case BOND_SETHWADDR_OLD:
4057                 case SIOCBONDSETHWADDR:
4058                         res = bond_sethwaddr(bond_dev, slave_dev);
4059                         break;
4060                 case BOND_CHANGE_ACTIVE_OLD:
4061                 case SIOCBONDCHANGEACTIVE:
4062                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4063                         break;
4064                 default:
4065                         res = -EOPNOTSUPP;
4066                 }
4067
4068                 dev_put(slave_dev);
4069         }
4070
4071         return res;
4072 }
4073
4074 static bool bond_addr_in_mc_list(unsigned char *addr,
4075                                  struct netdev_hw_addr_list *list,
4076                                  int addrlen)
4077 {
4078         struct netdev_hw_addr *ha;
4079
4080         netdev_hw_addr_list_for_each(ha, list)
4081                 if (!memcmp(ha->addr, addr, addrlen))
4082                         return true;
4083
4084         return false;
4085 }
4086
4087 static void bond_set_multicast_list(struct net_device *bond_dev)
4088 {
4089         struct bonding *bond = netdev_priv(bond_dev);
4090         struct netdev_hw_addr *ha;
4091         bool found;
4092
4093         /*
4094          * Do promisc before checking multicast_mode
4095          */
4096         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4097                 /*
4098                  * FIXME: Need to handle the error when one of the multi-slaves
4099                  * encounters error.
4100                  */
4101                 bond_set_promiscuity(bond, 1);
4102
4103
4104         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4105                 bond_set_promiscuity(bond, -1);
4106
4107
4108         /* set allmulti flag to slaves */
4109         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4110                 /*
4111                  * FIXME: Need to handle the error when one of the multi-slaves
4112                  * encounters error.
4113                  */
4114                 bond_set_allmulti(bond, 1);
4115
4116
4117         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4118                 bond_set_allmulti(bond, -1);
4119
4120
4121         read_lock(&bond->lock);
4122
4123         bond->flags = bond_dev->flags;
4124
4125         /* looking for addresses to add to slaves' mc list */
4126         netdev_for_each_mc_addr(ha, bond_dev) {
4127                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4128                                              bond_dev->addr_len);
4129                 if (!found)
4130                         bond_mc_add(bond, ha->addr);
4131         }
4132
4133         /* looking for addresses to delete from slaves' list */
4134         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4135                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4136                                              bond_dev->addr_len);
4137                 if (!found)
4138                         bond_mc_del(bond, ha->addr);
4139         }
4140
4141         /* save master's multicast list */
4142         __hw_addr_flush(&bond->mc_list);
4143         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4144                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4145
4146         read_unlock(&bond->lock);
4147 }
4148
4149 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4150 {
4151         struct bonding *bond = netdev_priv(dev);
4152         struct slave *slave = bond->first_slave;
4153
4154         if (slave) {
4155                 const struct net_device_ops *slave_ops
4156                         = slave->dev->netdev_ops;
4157                 if (slave_ops->ndo_neigh_setup)
4158                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4159         }
4160         return 0;
4161 }
4162
4163 /*
4164  * Change the MTU of all of a master's slaves to match the master
4165  */
4166 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4167 {
4168         struct bonding *bond = netdev_priv(bond_dev);
4169         struct slave *slave, *stop_at;
4170         int res = 0;
4171         int i;
4172
4173         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4174                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4175
4176         /* Can't hold bond->lock with bh disabled here since
4177          * some base drivers panic. On the other hand we can't
4178          * hold bond->lock without bh disabled because we'll
4179          * deadlock. The only solution is to rely on the fact
4180          * that we're under rtnl_lock here, and the slaves
4181          * list won't change. This doesn't solve the problem
4182          * of setting the slave's MTU while it is
4183          * transmitting, but the assumption is that the base
4184          * driver can handle that.
4185          *
4186          * TODO: figure out a way to safely iterate the slaves
4187          * list, but without holding a lock around the actual
4188          * call to the base driver.
4189          */
4190
4191         bond_for_each_slave(bond, slave, i) {
4192                 pr_debug("s %p s->p %p c_m %p\n",
4193                          slave,
4194                          slave->prev,
4195                          slave->dev->netdev_ops->ndo_change_mtu);
4196
4197                 res = dev_set_mtu(slave->dev, new_mtu);
4198
4199                 if (res) {
4200                         /* If we failed to set the slave's mtu to the new value
4201                          * we must abort the operation even in ACTIVE_BACKUP
4202                          * mode, because if we allow the backup slaves to have
4203                          * different mtu values than the active slave we'll
4204                          * need to change their mtu when doing a failover. That
4205                          * means changing their mtu from timer context, which
4206                          * is probably not a good idea.
4207                          */
4208                         pr_debug("err %d %s\n", res, slave->dev->name);
4209                         goto unwind;
4210                 }
4211         }
4212
4213         bond_dev->mtu = new_mtu;
4214
4215         return 0;
4216
4217 unwind:
4218         /* unwind from head to the slave that failed */
4219         stop_at = slave;
4220         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4221                 int tmp_res;
4222
4223                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4224                 if (tmp_res) {
4225                         pr_debug("unwind err %d dev %s\n",
4226                                  tmp_res, slave->dev->name);
4227                 }
4228         }
4229
4230         return res;
4231 }
4232
4233 /*
4234  * Change HW address
4235  *
4236  * Note that many devices must be down to change the HW address, and
4237  * downing the master releases all slaves.  We can make bonds full of
4238  * bonding devices to test this, however.
4239  */
4240 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4241 {
4242         struct bonding *bond = netdev_priv(bond_dev);
4243         struct sockaddr *sa = addr, tmp_sa;
4244         struct slave *slave, *stop_at;
4245         int res = 0;
4246         int i;
4247
4248         if (bond->params.mode == BOND_MODE_ALB)
4249                 return bond_alb_set_mac_address(bond_dev, addr);
4250
4251
4252         pr_debug("bond=%p, name=%s\n",
4253                  bond, bond_dev ? bond_dev->name : "None");
4254
4255         /*
4256          * If fail_over_mac is set to active, do nothing and return
4257          * success.  Returning an error causes ifenslave to fail.
4258          */
4259         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4260                 return 0;
4261
4262         if (!is_valid_ether_addr(sa->sa_data))
4263                 return -EADDRNOTAVAIL;
4264
4265         /* Can't hold bond->lock with bh disabled here since
4266          * some base drivers panic. On the other hand we can't
4267          * hold bond->lock without bh disabled because we'll
4268          * deadlock. The only solution is to rely on the fact
4269          * that we're under rtnl_lock here, and the slaves
4270          * list won't change. This doesn't solve the problem
4271          * of setting the slave's hw address while it is
4272          * transmitting, but the assumption is that the base
4273          * driver can handle that.
4274          *
4275          * TODO: figure out a way to safely iterate the slaves
4276          * list, but without holding a lock around the actual
4277          * call to the base driver.
4278          */
4279
4280         bond_for_each_slave(bond, slave, i) {
4281                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4282                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4283
4284                 if (slave_ops->ndo_set_mac_address == NULL) {
4285                         res = -EOPNOTSUPP;
4286                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4287                         goto unwind;
4288                 }
4289
4290                 res = dev_set_mac_address(slave->dev, addr);
4291                 if (res) {
4292                         /* TODO: consider downing the slave
4293                          * and retry ?
4294                          * User should expect communications
4295                          * breakage anyway until ARP finish
4296                          * updating, so...
4297                          */
4298                         pr_debug("err %d %s\n", res, slave->dev->name);
4299                         goto unwind;
4300                 }
4301         }
4302
4303         /* success */
4304         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4305         return 0;
4306
4307 unwind:
4308         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4309         tmp_sa.sa_family = bond_dev->type;
4310
4311         /* unwind from head to the slave that failed */
4312         stop_at = slave;
4313         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4314                 int tmp_res;
4315
4316                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4317                 if (tmp_res) {
4318                         pr_debug("unwind err %d dev %s\n",
4319                                  tmp_res, slave->dev->name);
4320                 }
4321         }
4322
4323         return res;
4324 }
4325
4326 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4327 {
4328         struct bonding *bond = netdev_priv(bond_dev);
4329         struct slave *slave, *start_at;
4330         int i, slave_no, res = 1;
4331         struct iphdr *iph = ip_hdr(skb);
4332
4333         read_lock(&bond->lock);
4334
4335         if (!BOND_IS_OK(bond))
4336                 goto out;
4337         /*
4338          * Start with the curr_active_slave that joined the bond as the
4339          * default for sending IGMP traffic.  For failover purposes one
4340          * needs to maintain some consistency for the interface that will
4341          * send the join/membership reports.  The curr_active_slave found
4342          * will send all of this type of traffic.
4343          */
4344         if ((iph->protocol == IPPROTO_IGMP) &&
4345             (skb->protocol == htons(ETH_P_IP))) {
4346
4347                 read_lock(&bond->curr_slave_lock);
4348                 slave = bond->curr_active_slave;
4349                 read_unlock(&bond->curr_slave_lock);
4350
4351                 if (!slave)
4352                         goto out;
4353         } else {
4354                 /*
4355                  * Concurrent TX may collide on rr_tx_counter; we accept
4356                  * that as being rare enough not to justify using an
4357                  * atomic op here.
4358                  */
4359                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4360
4361                 bond_for_each_slave(bond, slave, i) {
4362                         slave_no--;
4363                         if (slave_no < 0)
4364                                 break;
4365                 }
4366         }
4367
4368         start_at = slave;
4369         bond_for_each_slave_from(bond, slave, i, start_at) {
4370                 if (IS_UP(slave->dev) &&
4371                     (slave->link == BOND_LINK_UP) &&
4372                     (slave->state == BOND_STATE_ACTIVE)) {
4373                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4374                         break;
4375                 }
4376         }
4377
4378 out:
4379         if (res) {
4380                 /* no suitable interface, frame not sent */
4381                 dev_kfree_skb(skb);
4382         }
4383         read_unlock(&bond->lock);
4384         return NETDEV_TX_OK;
4385 }
4386
4387
4388 /*
4389  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4390  * the bond has a usable interface.
4391  */
4392 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4393 {
4394         struct bonding *bond = netdev_priv(bond_dev);
4395         int res = 1;
4396
4397         read_lock(&bond->lock);
4398         read_lock(&bond->curr_slave_lock);
4399
4400         if (!BOND_IS_OK(bond))
4401                 goto out;
4402
4403         if (!bond->curr_active_slave)
4404                 goto out;
4405
4406         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4407
4408 out:
4409         if (res)
4410                 /* no suitable interface, frame not sent */
4411                 dev_kfree_skb(skb);
4412
4413         read_unlock(&bond->curr_slave_lock);
4414         read_unlock(&bond->lock);
4415         return NETDEV_TX_OK;
4416 }
4417
4418 /*
4419  * In bond_xmit_xor() , we determine the output device by using a pre-
4420  * determined xmit_hash_policy(), If the selected device is not enabled,
4421  * find the next active slave.
4422  */
4423 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4424 {
4425         struct bonding *bond = netdev_priv(bond_dev);
4426         struct slave *slave, *start_at;
4427         int slave_no;
4428         int i;
4429         int res = 1;
4430
4431         read_lock(&bond->lock);
4432
4433         if (!BOND_IS_OK(bond))
4434                 goto out;
4435
4436         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4437
4438         bond_for_each_slave(bond, slave, i) {
4439                 slave_no--;
4440                 if (slave_no < 0)
4441                         break;
4442         }
4443
4444         start_at = slave;
4445
4446         bond_for_each_slave_from(bond, slave, i, start_at) {
4447                 if (IS_UP(slave->dev) &&
4448                     (slave->link == BOND_LINK_UP) &&
4449                     (slave->state == BOND_STATE_ACTIVE)) {
4450                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4451                         break;
4452                 }
4453         }
4454
4455 out:
4456         if (res) {
4457                 /* no suitable interface, frame not sent */
4458                 dev_kfree_skb(skb);
4459         }
4460         read_unlock(&bond->lock);
4461         return NETDEV_TX_OK;
4462 }
4463
4464 /*
4465  * in broadcast mode, we send everything to all usable interfaces.
4466  */
4467 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4468 {
4469         struct bonding *bond = netdev_priv(bond_dev);
4470         struct slave *slave, *start_at;
4471         struct net_device *tx_dev = NULL;
4472         int i;
4473         int res = 1;
4474
4475         read_lock(&bond->lock);
4476
4477         if (!BOND_IS_OK(bond))
4478                 goto out;
4479
4480         read_lock(&bond->curr_slave_lock);
4481         start_at = bond->curr_active_slave;
4482         read_unlock(&bond->curr_slave_lock);
4483
4484         if (!start_at)
4485                 goto out;
4486
4487         bond_for_each_slave_from(bond, slave, i, start_at) {
4488                 if (IS_UP(slave->dev) &&
4489                     (slave->link == BOND_LINK_UP) &&
4490                     (slave->state == BOND_STATE_ACTIVE)) {
4491                         if (tx_dev) {
4492                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4493                                 if (!skb2) {
4494                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4495                                                bond_dev->name);
4496                                         continue;
4497                                 }
4498
4499                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4500                                 if (res) {
4501                                         dev_kfree_skb(skb2);
4502                                         continue;
4503                                 }
4504                         }
4505                         tx_dev = slave->dev;
4506                 }
4507         }
4508
4509         if (tx_dev)
4510                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4511
4512 out:
4513         if (res)
4514                 /* no suitable interface, frame not sent */
4515                 dev_kfree_skb(skb);
4516
4517         /* frame sent to all suitable interfaces */
4518         read_unlock(&bond->lock);
4519         return NETDEV_TX_OK;
4520 }
4521
4522 /*------------------------- Device initialization ---------------------------*/
4523
4524 static void bond_set_xmit_hash_policy(struct bonding *bond)
4525 {
4526         switch (bond->params.xmit_policy) {
4527         case BOND_XMIT_POLICY_LAYER23:
4528                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4529                 break;
4530         case BOND_XMIT_POLICY_LAYER34:
4531                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4532                 break;
4533         case BOND_XMIT_POLICY_LAYER2:
4534         default:
4535                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4536                 break;
4537         }
4538 }
4539
4540 /*
4541  * Lookup the slave that corresponds to a qid
4542  */
4543 static inline int bond_slave_override(struct bonding *bond,
4544                                       struct sk_buff *skb)
4545 {
4546         int i, res = 1;
4547         struct slave *slave = NULL;
4548         struct slave *check_slave;
4549
4550         read_lock(&bond->lock);
4551
4552         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4553                 goto out;
4554
4555         /* Find out if any slaves have the same mapping as this skb. */
4556         bond_for_each_slave(bond, check_slave, i) {
4557                 if (check_slave->queue_id == skb->queue_mapping) {
4558                         slave = check_slave;
4559                         break;
4560                 }
4561         }
4562
4563         /* If the slave isn't UP, use default transmit policy. */
4564         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4565             (slave->link == BOND_LINK_UP)) {
4566                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4567         }
4568
4569 out:
4570         read_unlock(&bond->lock);
4571         return res;
4572 }
4573
4574 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4575 {
4576         /*
4577          * This helper function exists to help dev_pick_tx get the correct
4578          * destination queue.  Using a helper function skips the a call to
4579          * skb_tx_hash and will put the skbs in the queue we expect on their
4580          * way down to the bonding driver.
4581          */
4582         return skb->queue_mapping;
4583 }
4584
4585 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4586 {
4587         struct bonding *bond = netdev_priv(dev);
4588
4589         /*
4590          * If we risk deadlock from transmitting this in the
4591          * netpoll path, tell netpoll to queue the frame for later tx
4592          */
4593         if (is_netpoll_tx_blocked(dev))
4594                 return NETDEV_TX_BUSY;
4595
4596         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4597                 if (!bond_slave_override(bond, skb))
4598                         return NETDEV_TX_OK;
4599         }
4600
4601         switch (bond->params.mode) {
4602         case BOND_MODE_ROUNDROBIN:
4603                 return bond_xmit_roundrobin(skb, dev);
4604         case BOND_MODE_ACTIVEBACKUP:
4605                 return bond_xmit_activebackup(skb, dev);
4606         case BOND_MODE_XOR:
4607                 return bond_xmit_xor(skb, dev);
4608         case BOND_MODE_BROADCAST:
4609                 return bond_xmit_broadcast(skb, dev);
4610         case BOND_MODE_8023AD:
4611                 return bond_3ad_xmit_xor(skb, dev);
4612         case BOND_MODE_ALB:
4613         case BOND_MODE_TLB:
4614                 return bond_alb_xmit(skb, dev);
4615         default:
4616                 /* Should never happen, mode already checked */
4617                 pr_err("%s: Error: Unknown bonding mode %d\n",
4618                        dev->name, bond->params.mode);
4619                 WARN_ON_ONCE(1);
4620                 dev_kfree_skb(skb);
4621                 return NETDEV_TX_OK;
4622         }
4623 }
4624
4625
4626 /*
4627  * set bond mode specific net device operations
4628  */
4629 void bond_set_mode_ops(struct bonding *bond, int mode)
4630 {
4631         struct net_device *bond_dev = bond->dev;
4632
4633         switch (mode) {
4634         case BOND_MODE_ROUNDROBIN:
4635                 break;
4636         case BOND_MODE_ACTIVEBACKUP:
4637                 break;
4638         case BOND_MODE_XOR:
4639                 bond_set_xmit_hash_policy(bond);
4640                 break;
4641         case BOND_MODE_BROADCAST:
4642                 break;
4643         case BOND_MODE_8023AD:
4644                 bond_set_master_3ad_flags(bond);
4645                 bond_set_xmit_hash_policy(bond);
4646                 break;
4647         case BOND_MODE_ALB:
4648                 bond_set_master_alb_flags(bond);
4649                 /* FALLTHRU */
4650         case BOND_MODE_TLB:
4651                 break;
4652         default:
4653                 /* Should never happen, mode already checked */
4654                 pr_err("%s: Error: Unknown bonding mode %d\n",
4655                        bond_dev->name, mode);
4656                 break;
4657         }
4658 }
4659
4660 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4661                                     struct ethtool_drvinfo *drvinfo)
4662 {
4663         strncpy(drvinfo->driver, DRV_NAME, 32);
4664         strncpy(drvinfo->version, DRV_VERSION, 32);
4665         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4666 }
4667
4668 static const struct ethtool_ops bond_ethtool_ops = {
4669         .get_drvinfo            = bond_ethtool_get_drvinfo,
4670         .get_link               = ethtool_op_get_link,
4671         .get_tx_csum            = ethtool_op_get_tx_csum,
4672         .get_sg                 = ethtool_op_get_sg,
4673         .get_tso                = ethtool_op_get_tso,
4674         .get_ufo                = ethtool_op_get_ufo,
4675         .get_flags              = ethtool_op_get_flags,
4676 };
4677
4678 static const struct net_device_ops bond_netdev_ops = {
4679         .ndo_init               = bond_init,
4680         .ndo_uninit             = bond_uninit,
4681         .ndo_open               = bond_open,
4682         .ndo_stop               = bond_close,
4683         .ndo_start_xmit         = bond_start_xmit,
4684         .ndo_select_queue       = bond_select_queue,
4685         .ndo_get_stats64        = bond_get_stats,
4686         .ndo_do_ioctl           = bond_do_ioctl,
4687         .ndo_set_multicast_list = bond_set_multicast_list,
4688         .ndo_change_mtu         = bond_change_mtu,
4689         .ndo_set_mac_address    = bond_set_mac_address,
4690         .ndo_neigh_setup        = bond_neigh_setup,
4691         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4692         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4693         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4694 #ifdef CONFIG_NET_POLL_CONTROLLER
4695         .ndo_netpoll_setup      = bond_netpoll_setup,
4696         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4697         .ndo_poll_controller    = bond_poll_controller,
4698 #endif
4699         .ndo_add_slave          = bond_enslave,
4700         .ndo_del_slave          = bond_release,
4701 };
4702
4703 static void bond_destructor(struct net_device *bond_dev)
4704 {
4705         struct bonding *bond = netdev_priv(bond_dev);
4706         if (bond->wq)
4707                 destroy_workqueue(bond->wq);
4708         free_netdev(bond_dev);
4709 }
4710
4711 static void bond_setup(struct net_device *bond_dev)
4712 {
4713         struct bonding *bond = netdev_priv(bond_dev);
4714
4715         /* initialize rwlocks */
4716         rwlock_init(&bond->lock);
4717         rwlock_init(&bond->curr_slave_lock);
4718
4719         bond->params = bonding_defaults;
4720
4721         /* Initialize pointers */
4722         bond->dev = bond_dev;
4723         INIT_LIST_HEAD(&bond->vlan_list);
4724
4725         /* Initialize the device entry points */
4726         ether_setup(bond_dev);
4727         bond_dev->netdev_ops = &bond_netdev_ops;
4728         bond_dev->ethtool_ops = &bond_ethtool_ops;
4729         bond_set_mode_ops(bond, bond->params.mode);
4730
4731         bond_dev->destructor = bond_destructor;
4732
4733         /* Initialize the device options */
4734         bond_dev->tx_queue_len = 0;
4735         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4736         bond_dev->priv_flags |= IFF_BONDING;
4737         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4738
4739         if (bond->params.arp_interval)
4740                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4741
4742         /* At first, we block adding VLANs. That's the only way to
4743          * prevent problems that occur when adding VLANs over an
4744          * empty bond. The block will be removed once non-challenged
4745          * slaves are enslaved.
4746          */
4747         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4748
4749         /* don't acquire bond device's netif_tx_lock when
4750          * transmitting */
4751         bond_dev->features |= NETIF_F_LLTX;
4752
4753         /* By default, we declare the bond to be fully
4754          * VLAN hardware accelerated capable. Special
4755          * care is taken in the various xmit functions
4756          * when there are slaves that are not hw accel
4757          * capable
4758          */
4759         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4760                                NETIF_F_HW_VLAN_RX |
4761                                NETIF_F_HW_VLAN_FILTER);
4762
4763         /* By default, we enable GRO on bonding devices.
4764          * Actual support requires lowlevel drivers are GRO ready.
4765          */
4766         bond_dev->features |= NETIF_F_GRO;
4767 }
4768
4769 static void bond_work_cancel_all(struct bonding *bond)
4770 {
4771         write_lock_bh(&bond->lock);
4772         bond->kill_timers = 1;
4773         write_unlock_bh(&bond->lock);
4774
4775         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4776                 cancel_delayed_work(&bond->mii_work);
4777
4778         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4779                 cancel_delayed_work(&bond->arp_work);
4780
4781         if (bond->params.mode == BOND_MODE_ALB &&
4782             delayed_work_pending(&bond->alb_work))
4783                 cancel_delayed_work(&bond->alb_work);
4784
4785         if (bond->params.mode == BOND_MODE_8023AD &&
4786             delayed_work_pending(&bond->ad_work))
4787                 cancel_delayed_work(&bond->ad_work);
4788
4789         if (delayed_work_pending(&bond->mcast_work))
4790                 cancel_delayed_work(&bond->mcast_work);
4791 }
4792
4793 /*
4794 * Destroy a bonding device.
4795 * Must be under rtnl_lock when this function is called.
4796 */
4797 static void bond_uninit(struct net_device *bond_dev)
4798 {
4799         struct bonding *bond = netdev_priv(bond_dev);
4800         struct vlan_entry *vlan, *tmp;
4801
4802         bond_netpoll_cleanup(bond_dev);
4803
4804         /* Release the bonded slaves */
4805         bond_release_all(bond_dev);
4806
4807         list_del(&bond->bond_list);
4808
4809         bond_work_cancel_all(bond);
4810
4811         bond_remove_proc_entry(bond);
4812
4813         bond_debug_unregister(bond);
4814
4815         __hw_addr_flush(&bond->mc_list);
4816
4817         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4818                 list_del(&vlan->vlan_list);
4819                 kfree(vlan);
4820         }
4821 }
4822
4823 /*------------------------- Module initialization ---------------------------*/
4824
4825 /*
4826  * Convert string input module parms.  Accept either the
4827  * number of the mode or its string name.  A bit complicated because
4828  * some mode names are substrings of other names, and calls from sysfs
4829  * may have whitespace in the name (trailing newlines, for example).
4830  */
4831 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4832 {
4833         int modeint = -1, i, rv;
4834         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4835
4836         for (p = (char *)buf; *p; p++)
4837                 if (!(isdigit(*p) || isspace(*p)))
4838                         break;
4839
4840         if (*p)
4841                 rv = sscanf(buf, "%20s", modestr);
4842         else
4843                 rv = sscanf(buf, "%d", &modeint);
4844
4845         if (!rv)
4846                 return -1;
4847
4848         for (i = 0; tbl[i].modename; i++) {
4849                 if (modeint == tbl[i].mode)
4850                         return tbl[i].mode;
4851                 if (strcmp(modestr, tbl[i].modename) == 0)
4852                         return tbl[i].mode;
4853         }
4854
4855         return -1;
4856 }
4857
4858 static int bond_check_params(struct bond_params *params)
4859 {
4860         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4861
4862         /*
4863          * Convert string parameters.
4864          */
4865         if (mode) {
4866                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4867                 if (bond_mode == -1) {
4868                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4869                                mode == NULL ? "NULL" : mode);
4870                         return -EINVAL;
4871                 }
4872         }
4873
4874         if (xmit_hash_policy) {
4875                 if ((bond_mode != BOND_MODE_XOR) &&
4876                     (bond_mode != BOND_MODE_8023AD)) {
4877                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4878                                bond_mode_name(bond_mode));
4879                 } else {
4880                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4881                                                         xmit_hashtype_tbl);
4882                         if (xmit_hashtype == -1) {
4883                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4884                                        xmit_hash_policy == NULL ? "NULL" :
4885                                        xmit_hash_policy);
4886                                 return -EINVAL;
4887                         }
4888                 }
4889         }
4890
4891         if (lacp_rate) {
4892                 if (bond_mode != BOND_MODE_8023AD) {
4893                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4894                                 bond_mode_name(bond_mode));
4895                 } else {
4896                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4897                         if (lacp_fast == -1) {
4898                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4899                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4900                                 return -EINVAL;
4901                         }
4902                 }
4903         }
4904
4905         if (ad_select) {
4906                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4907                 if (params->ad_select == -1) {
4908                         pr_err("Error: Invalid ad_select \"%s\"\n",
4909                                ad_select == NULL ? "NULL" : ad_select);
4910                         return -EINVAL;
4911                 }
4912
4913                 if (bond_mode != BOND_MODE_8023AD) {
4914                         pr_warning("ad_select param only affects 802.3ad mode\n");
4915                 }
4916         } else {
4917                 params->ad_select = BOND_AD_STABLE;
4918         }
4919
4920         if (max_bonds < 0) {
4921                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4922                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4923                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4924         }
4925
4926         if (miimon < 0) {
4927                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4928                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4929                 miimon = BOND_LINK_MON_INTERV;
4930         }
4931
4932         if (updelay < 0) {
4933                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4934                            updelay, INT_MAX);
4935                 updelay = 0;
4936         }
4937
4938         if (downdelay < 0) {
4939                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4940                            downdelay, INT_MAX);
4941                 downdelay = 0;
4942         }
4943
4944         if ((use_carrier != 0) && (use_carrier != 1)) {
4945                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4946                            use_carrier);
4947                 use_carrier = 1;
4948         }
4949
4950         if (num_grat_arp < 0 || num_grat_arp > 255) {
4951                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4952                            num_grat_arp);
4953                 num_grat_arp = 1;
4954         }
4955
4956         if (num_unsol_na < 0 || num_unsol_na > 255) {
4957                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4958                            num_unsol_na);
4959                 num_unsol_na = 1;
4960         }
4961
4962         /* reset values for 802.3ad */
4963         if (bond_mode == BOND_MODE_8023AD) {
4964                 if (!miimon) {
4965                         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");
4966                         pr_warning("Forcing miimon to 100msec\n");
4967                         miimon = 100;
4968                 }
4969         }
4970
4971         if (tx_queues < 1 || tx_queues > 255) {
4972                 pr_warning("Warning: tx_queues (%d) should be between "
4973                            "1 and 255, resetting to %d\n",
4974                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4975                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4976         }
4977
4978         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4979                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4980                            "not of valid value (0/1), so it was set to "
4981                            "0\n", all_slaves_active);
4982                 all_slaves_active = 0;
4983         }
4984
4985         if (resend_igmp < 0 || resend_igmp > 255) {
4986                 pr_warning("Warning: resend_igmp (%d) should be between "
4987                            "0 and 255, resetting to %d\n",
4988                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4989                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4990         }
4991
4992         /* reset values for TLB/ALB */
4993         if ((bond_mode == BOND_MODE_TLB) ||
4994             (bond_mode == BOND_MODE_ALB)) {
4995                 if (!miimon) {
4996                         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");
4997                         pr_warning("Forcing miimon to 100msec\n");
4998                         miimon = 100;
4999                 }
5000         }
5001
5002         if (bond_mode == BOND_MODE_ALB) {
5003                 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",
5004                           updelay);
5005         }
5006
5007         if (!miimon) {
5008                 if (updelay || downdelay) {
5009                         /* just warn the user the up/down delay will have
5010                          * no effect since miimon is zero...
5011                          */
5012                         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",
5013                                    updelay, downdelay);
5014                 }
5015         } else {
5016                 /* don't allow arp monitoring */
5017                 if (arp_interval) {
5018                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5019                                    miimon, arp_interval);
5020                         arp_interval = 0;
5021                 }
5022
5023                 if ((updelay % miimon) != 0) {
5024                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5025                                    updelay, miimon,
5026                                    (updelay / miimon) * miimon);
5027                 }
5028
5029                 updelay /= miimon;
5030
5031                 if ((downdelay % miimon) != 0) {
5032                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5033                                    downdelay, miimon,
5034                                    (downdelay / miimon) * miimon);
5035                 }
5036
5037                 downdelay /= miimon;
5038         }
5039
5040         if (arp_interval < 0) {
5041                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5042                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5043                 arp_interval = BOND_LINK_ARP_INTERV;
5044         }
5045
5046         for (arp_ip_count = 0;
5047              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5048              arp_ip_count++) {
5049                 /* not complete check, but should be good enough to
5050                    catch mistakes */
5051                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5052                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5053                                    arp_ip_target[arp_ip_count]);
5054                         arp_interval = 0;
5055                 } else {
5056                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5057                         arp_target[arp_ip_count] = ip;
5058                 }
5059         }
5060
5061         if (arp_interval && !arp_ip_count) {
5062                 /* don't allow arping if no arp_ip_target given... */
5063                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5064                            arp_interval);
5065                 arp_interval = 0;
5066         }
5067
5068         if (arp_validate) {
5069                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5070                         pr_err("arp_validate only supported in active-backup mode\n");
5071                         return -EINVAL;
5072                 }
5073                 if (!arp_interval) {
5074                         pr_err("arp_validate requires arp_interval\n");
5075                         return -EINVAL;
5076                 }
5077
5078                 arp_validate_value = bond_parse_parm(arp_validate,
5079                                                      arp_validate_tbl);
5080                 if (arp_validate_value == -1) {
5081                         pr_err("Error: invalid arp_validate \"%s\"\n",
5082                                arp_validate == NULL ? "NULL" : arp_validate);
5083                         return -EINVAL;
5084                 }
5085         } else
5086                 arp_validate_value = 0;
5087
5088         if (miimon) {
5089                 pr_info("MII link monitoring set to %d ms\n", miimon);
5090         } else if (arp_interval) {
5091                 int i;
5092
5093                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5094                         arp_interval,
5095                         arp_validate_tbl[arp_validate_value].modename,
5096                         arp_ip_count);
5097
5098                 for (i = 0; i < arp_ip_count; i++)
5099                         pr_info(" %s", arp_ip_target[i]);
5100
5101                 pr_info("\n");
5102
5103         } else if (max_bonds) {
5104                 /* miimon and arp_interval not set, we need one so things
5105                  * work as expected, see bonding.txt for details
5106                  */
5107                 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");
5108         }
5109
5110         if (primary && !USES_PRIMARY(bond_mode)) {
5111                 /* currently, using a primary only makes sense
5112                  * in active backup, TLB or ALB modes
5113                  */
5114                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5115                            primary, bond_mode_name(bond_mode));
5116                 primary = NULL;
5117         }
5118
5119         if (primary && primary_reselect) {
5120                 primary_reselect_value = bond_parse_parm(primary_reselect,
5121                                                          pri_reselect_tbl);
5122                 if (primary_reselect_value == -1) {
5123                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5124                                primary_reselect ==
5125                                         NULL ? "NULL" : primary_reselect);
5126                         return -EINVAL;
5127                 }
5128         } else {
5129                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5130         }
5131
5132         if (fail_over_mac) {
5133                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5134                                                       fail_over_mac_tbl);
5135                 if (fail_over_mac_value == -1) {
5136                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5137                                arp_validate == NULL ? "NULL" : arp_validate);
5138                         return -EINVAL;
5139                 }
5140
5141                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5142                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5143         } else {
5144                 fail_over_mac_value = BOND_FOM_NONE;
5145         }
5146
5147         /* fill params struct with the proper values */
5148         params->mode = bond_mode;
5149         params->xmit_policy = xmit_hashtype;
5150         params->miimon = miimon;
5151         params->num_grat_arp = num_grat_arp;
5152         params->num_unsol_na = num_unsol_na;
5153         params->arp_interval = arp_interval;
5154         params->arp_validate = arp_validate_value;
5155         params->updelay = updelay;
5156         params->downdelay = downdelay;
5157         params->use_carrier = use_carrier;
5158         params->lacp_fast = lacp_fast;
5159         params->primary[0] = 0;
5160         params->primary_reselect = primary_reselect_value;
5161         params->fail_over_mac = fail_over_mac_value;
5162         params->tx_queues = tx_queues;
5163         params->all_slaves_active = all_slaves_active;
5164         params->resend_igmp = resend_igmp;
5165
5166         if (primary) {
5167                 strncpy(params->primary, primary, IFNAMSIZ);
5168                 params->primary[IFNAMSIZ - 1] = 0;
5169         }
5170
5171         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5172
5173         return 0;
5174 }
5175
5176 static struct lock_class_key bonding_netdev_xmit_lock_key;
5177 static struct lock_class_key bonding_netdev_addr_lock_key;
5178
5179 static void bond_set_lockdep_class_one(struct net_device *dev,
5180                                        struct netdev_queue *txq,
5181                                        void *_unused)
5182 {
5183         lockdep_set_class(&txq->_xmit_lock,
5184                           &bonding_netdev_xmit_lock_key);
5185 }
5186
5187 static void bond_set_lockdep_class(struct net_device *dev)
5188 {
5189         lockdep_set_class(&dev->addr_list_lock,
5190                           &bonding_netdev_addr_lock_key);
5191         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5192 }
5193
5194 /*
5195  * Called from registration process
5196  */
5197 static int bond_init(struct net_device *bond_dev)
5198 {
5199         struct bonding *bond = netdev_priv(bond_dev);
5200         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5201
5202         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5203
5204         bond->wq = create_singlethread_workqueue(bond_dev->name);
5205         if (!bond->wq)
5206                 return -ENOMEM;
5207
5208         bond_set_lockdep_class(bond_dev);
5209
5210         netif_carrier_off(bond_dev);
5211
5212         bond_create_proc_entry(bond);
5213         list_add_tail(&bond->bond_list, &bn->dev_list);
5214
5215         bond_prepare_sysfs_group(bond);
5216
5217         bond_debug_register(bond);
5218
5219         __hw_addr_init(&bond->mc_list);
5220         return 0;
5221 }
5222
5223 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5224 {
5225         if (tb[IFLA_ADDRESS]) {
5226                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5227                         return -EINVAL;
5228                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5229                         return -EADDRNOTAVAIL;
5230         }
5231         return 0;
5232 }
5233
5234 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5235         .kind           = "bond",
5236         .priv_size      = sizeof(struct bonding),
5237         .setup          = bond_setup,
5238         .validate       = bond_validate,
5239 };
5240
5241 /* Create a new bond based on the specified name and bonding parameters.
5242  * If name is NULL, obtain a suitable "bond%d" name for us.
5243  * Caller must NOT hold rtnl_lock; we need to release it here before we
5244  * set up our sysfs entries.
5245  */
5246 int bond_create(struct net *net, const char *name)
5247 {
5248         struct net_device *bond_dev;
5249         int res;
5250
5251         rtnl_lock();
5252
5253         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5254                                 bond_setup, tx_queues);
5255         if (!bond_dev) {
5256                 pr_err("%s: eek! can't alloc netdev!\n", name);
5257                 rtnl_unlock();
5258                 return -ENOMEM;
5259         }
5260
5261         dev_net_set(bond_dev, net);
5262         bond_dev->rtnl_link_ops = &bond_link_ops;
5263
5264         if (!name) {
5265                 res = dev_alloc_name(bond_dev, "bond%d");
5266                 if (res < 0)
5267                         goto out;
5268         } else {
5269                 /*
5270                  * If we're given a name to register
5271                  * we need to ensure that its not already
5272                  * registered
5273                  */
5274                 res = -EEXIST;
5275                 if (__dev_get_by_name(net, name) != NULL)
5276                         goto out;
5277         }
5278
5279         res = register_netdevice(bond_dev);
5280
5281 out:
5282         rtnl_unlock();
5283         if (res < 0)
5284                 bond_destructor(bond_dev);
5285         return res;
5286 }
5287
5288 static int __net_init bond_net_init(struct net *net)
5289 {
5290         struct bond_net *bn = net_generic(net, bond_net_id);
5291
5292         bn->net = net;
5293         INIT_LIST_HEAD(&bn->dev_list);
5294
5295         bond_create_proc_dir(bn);
5296
5297         return 0;
5298 }
5299
5300 static void __net_exit bond_net_exit(struct net *net)
5301 {
5302         struct bond_net *bn = net_generic(net, bond_net_id);
5303
5304         bond_destroy_proc_dir(bn);
5305 }
5306
5307 static struct pernet_operations bond_net_ops = {
5308         .init = bond_net_init,
5309         .exit = bond_net_exit,
5310         .id   = &bond_net_id,
5311         .size = sizeof(struct bond_net),
5312 };
5313
5314 static int __init bonding_init(void)
5315 {
5316         int i;
5317         int res;
5318
5319         pr_info("%s", version);
5320
5321         res = bond_check_params(&bonding_defaults);
5322         if (res)
5323                 goto out;
5324
5325         res = register_pernet_subsys(&bond_net_ops);
5326         if (res)
5327                 goto out;
5328
5329         res = rtnl_link_register(&bond_link_ops);
5330         if (res)
5331                 goto err_link;
5332
5333         bond_create_debugfs();
5334
5335         for (i = 0; i < max_bonds; i++) {
5336                 res = bond_create(&init_net, NULL);
5337                 if (res)
5338                         goto err;
5339         }
5340
5341         res = bond_create_sysfs();
5342         if (res)
5343                 goto err;
5344
5345         register_netdevice_notifier(&bond_netdev_notifier);
5346         register_inetaddr_notifier(&bond_inetaddr_notifier);
5347         bond_register_ipv6_notifier();
5348 out:
5349         return res;
5350 err:
5351         rtnl_link_unregister(&bond_link_ops);
5352 err_link:
5353         unregister_pernet_subsys(&bond_net_ops);
5354         goto out;
5355
5356 }
5357
5358 static void __exit bonding_exit(void)
5359 {
5360         unregister_netdevice_notifier(&bond_netdev_notifier);
5361         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5362         bond_unregister_ipv6_notifier();
5363
5364         bond_destroy_sysfs();
5365         bond_destroy_debugfs();
5366
5367         rtnl_link_unregister(&bond_link_ops);
5368         unregister_pernet_subsys(&bond_net_ops);
5369
5370 #ifdef CONFIG_NET_POLL_CONTROLLER
5371         /*
5372          * Make sure we don't have an imbalance on our netpoll blocking
5373          */
5374         WARN_ON(atomic_read(&netpoll_block_tx));
5375 #endif
5376 }
5377
5378 module_init(bonding_init);
5379 module_exit(bonding_exit);
5380 MODULE_LICENSE("GPL");
5381 MODULE_VERSION(DRV_VERSION);
5382 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5383 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5384 MODULE_ALIAS_RTNL_LINK("bond");