drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <linux/preempt.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include <net/netns/generic.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval = BOND_LINK_ARP_INTERV;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *fail_over_mac;
 107 static int all_slaves_active = 0;
 108 static struct bond_params bonding_defaults;
 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 110
 111 module_param(max_bonds, int, 0);
 112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 113 module_param(tx_queues, int, 0);
 114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on failover event (alias of num_unsol_na)");
 117 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 118 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on failover event (alias of num_grat_arp)");
 119 module_param(miimon, int, 0);
 120 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 121 module_param(updelay, int, 0);
 122 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 123 module_param(downdelay, int, 0);
 124 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 125                             "in milliseconds");
 126 module_param(use_carrier, int, 0);
 127 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 128                               "0 for off, 1 for on (default)");
 129 module_param(mode, charp, 0);
 130 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 131                        "1 for active-backup, 2 for balance-xor, "
 132                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 133                        "6 for balance-alb");
 134 module_param(primary, charp, 0);
 135 MODULE_PARM_DESC(primary, "Primary network device to use");
 136 module_param(primary_reselect, charp, 0);
 137 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 138                                    "once it comes up; "
 139                                    "0 for always (default), "
 140                                    "1 for only if speed of primary is "
 141                                    "better, "
 142                                    "2 for only on active slave "
 143                                    "failure");
 144 module_param(lacp_rate, charp, 0);
 145 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 146                             "(slow/fast)");
 147 module_param(ad_select, charp, 0);
 148 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 149 module_param(xmit_hash_policy, charp, 0);
 150 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 151                                    ", 1 for layer 3+4");
 152 module_param(arp_interval, int, 0);
 153 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 154 module_param_array(arp_ip_target, charp, NULL, 0);
 155 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 156 module_param(arp_validate, charp, 0);
 157 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 158 module_param(fail_over_mac, charp, 0);
 159 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 160 module_param(all_slaves_active, int, 0);
 161 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 162                                      "by setting active flag for all slaves.  "
 163                                      "0 for never (default), 1 for always.");
 164 module_param(resend_igmp, int, 0);
 165 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
 166
 167 /*----------------------------- Global variables ----------------------------*/
 168
 169 #ifdef CONFIG_NET_POLL_CONTROLLER
 170 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 171 #endif
 172
 173 int bond_net_id __read_mostly;
 174
 175 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 176 static int arp_ip_count;
 177 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 178 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 179 static int lacp_fast;
 180
 181 const struct bond_parm_tbl bond_lacp_tbl[] = {
 182 {       "slow",         AD_LACP_SLOW},
 183 {       "fast",         AD_LACP_FAST},
 184 {       NULL,           -1},
 185 };
 186
 187 const struct bond_parm_tbl bond_mode_tbl[] = {
 188 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 189 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 190 {       "balance-xor",          BOND_MODE_XOR},
 191 {       "broadcast",            BOND_MODE_BROADCAST},
 192 {       "802.3ad",              BOND_MODE_8023AD},
 193 {       "balance-tlb",          BOND_MODE_TLB},
 194 {       "balance-alb",          BOND_MODE_ALB},
 195 {       NULL,                   -1},
 196 };
 197
 198 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 199 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 200 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 201 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 202 {       NULL,                   -1},
 203 };
 204
 205 const struct bond_parm_tbl arp_validate_tbl[] = {
 206 {       "none",                 BOND_ARP_VALIDATE_NONE},
 207 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 208 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 209 {       "all",                  BOND_ARP_VALIDATE_ALL},
 210 {       NULL,                   -1},
 211 };
 212
 213 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 214 {       "none",                 BOND_FOM_NONE},
 215 {       "active",               BOND_FOM_ACTIVE},
 216 {       "follow",               BOND_FOM_FOLLOW},
 217 {       NULL,                   -1},
 218 };
 219
 220 const struct bond_parm_tbl pri_reselect_tbl[] = {
 221 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 222 {       "better",               BOND_PRI_RESELECT_BETTER},
 223 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 224 {       NULL,                   -1},
 225 };
 226
 227 struct bond_parm_tbl ad_select_tbl[] = {
 228 {       "stable",       BOND_AD_STABLE},
 229 {       "bandwidth",    BOND_AD_BANDWIDTH},
 230 {       "count",        BOND_AD_COUNT},
 231 {       NULL,           -1},
 232 };
 233
 234 /*-------------------------- Forward declarations ---------------------------*/
 235
 236 static int bond_init(struct net_device *bond_dev);
 237 static void bond_uninit(struct net_device *bond_dev);
 238
 239 /*---------------------------- General routines -----------------------------*/
 240
 241 const char *bond_mode_name(int mode)
 242 {
 243         static const char *names[] = {
 244                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 245                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 246                 [BOND_MODE_XOR] = "load balancing (xor)",
 247                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 248                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 249                 [BOND_MODE_TLB] = "transmit load balancing",
 250                 [BOND_MODE_ALB] = "adaptive load balancing",
 251         };
 252
 253         if (mode < 0 || mode > BOND_MODE_ALB)
 254                 return "unknown";
 255
 256         return names[mode];
 257 }
 258
 259 /*---------------------------------- VLAN -----------------------------------*/
 260
 261 /**
 262  * bond_add_vlan - add a new vlan id on bond
 263  * @bond: bond that got the notification
 264  * @vlan_id: the vlan id to add
 265  *
 266  * Returns -ENOMEM if allocation failed.
 267  */
 268 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 269 {
 270         struct vlan_entry *vlan;
 271
 272         pr_debug("bond: %s, vlan id %d\n",
 273                  (bond ? bond->dev->name : "None"), vlan_id);
 274
 275         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 276         if (!vlan)
 277                 return -ENOMEM;
 278
 279         INIT_LIST_HEAD(&vlan->vlan_list);
 280         vlan->vlan_id = vlan_id;
 281
 282         write_lock_bh(&bond->lock);
 283
 284         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 285
 286         write_unlock_bh(&bond->lock);
 287
 288         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 289
 290         return 0;
 291 }
 292
 293 /**
 294  * bond_del_vlan - delete a vlan id from bond
 295  * @bond: bond that got the notification
 296  * @vlan_id: the vlan id to delete
 297  *
 298  * returns -ENODEV if @vlan_id was not found in @bond.
 299  */
 300 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 301 {
 302         struct vlan_entry *vlan;
 303         int res = -ENODEV;
 304
 305         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 306
 307         block_netpoll_tx();
 308         write_lock_bh(&bond->lock);
 309
 310         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 311                 if (vlan->vlan_id == vlan_id) {
 312                         list_del(&vlan->vlan_list);
 313
 314                         if (bond_is_lb(bond))
 315                                 bond_alb_clear_vlan(bond, vlan_id);
 316
 317                         pr_debug("removed VLAN ID %d from bond %s\n",
 318                                  vlan_id, bond->dev->name);
 319
 320                         kfree(vlan);
 321
 322                         if (list_empty(&bond->vlan_list) &&
 323                             (bond->slave_cnt == 0)) {
 324                                 /* Last VLAN removed and no slaves, so
 325                                  * restore block on adding VLANs. This will
 326                                  * be removed once new slaves that are not
 327                                  * VLAN challenged will be added.
 328                                  */
 329                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 330                         }
 331
 332                         res = 0;
 333                         goto out;
 334                 }
 335         }
 336
 337         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 338                  vlan_id, bond->dev->name);
 339
 340 out:
 341         write_unlock_bh(&bond->lock);
 342         unblock_netpoll_tx();
 343         return res;
 344 }
 345
 346 /**
 347  * bond_next_vlan - safely skip to the next item in the vlans list.
 348  * @bond: the bond we're working on
 349  * @curr: item we're advancing from
 350  *
 351  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 352  * or @curr->next otherwise (even if it is @curr itself again).
 353  *
 354  * Caller must hold bond->lock
 355  */
 356 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 357 {
 358         struct vlan_entry *next, *last;
 359
 360         if (list_empty(&bond->vlan_list))
 361                 return NULL;
 362
 363         if (!curr) {
 364                 next = list_entry(bond->vlan_list.next,
 365                                   struct vlan_entry, vlan_list);
 366         } else {
 367                 last = list_entry(bond->vlan_list.prev,
 368                                   struct vlan_entry, vlan_list);
 369                 if (last == curr) {
 370                         next = list_entry(bond->vlan_list.next,
 371                                           struct vlan_entry, vlan_list);
 372                 } else {
 373                         next = list_entry(curr->vlan_list.next,
 374                                           struct vlan_entry, vlan_list);
 375                 }
 376         }
 377
 378         return next;
 379 }
 380
 381 /**
 382  * bond_dev_queue_xmit - Prepare skb for xmit.
 383  *
 384  * @bond: bond device that got this skb for tx.
 385  * @skb: hw accel VLAN tagged skb to transmit
 386  * @slave_dev: slave that is supposed to xmit this skbuff
 387  */
 388 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 389                         struct net_device *slave_dev)
 390 {
 391         skb->dev = slave_dev;
 392         skb->priority = 1;
 393         if (unlikely(netpoll_tx_running(slave_dev)))
 394                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 395         else
 396                 dev_queue_xmit(skb);
 397
 398         return 0;
 399 }
 400
 401 /*
 402  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 403  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 404  * lock because:
 405  * a. This operation is performed in IOCTL context,
 406  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 407  * c. Holding a lock with BH disabled while directly calling a base driver
 408  *    entry point is generally a BAD idea.
 409  *
 410  * The design of synchronization/protection for this operation in the 8021q
 411  * module is good for one or more VLAN devices over a single physical device
 412  * and cannot be extended for a teaming solution like bonding, so there is a
 413  * potential race condition here where a net device from the vlan group might
 414  * be referenced (either by a base driver or the 8021q code) while it is being
 415  * removed from the system. However, it turns out we're not making matters
 416  * worse, and if it works for regular VLAN usage it will work here too.
 417 */
 418
 419 /**
 420  * bond_vlan_rx_register - Propagates registration to slaves
 421  * @bond_dev: bonding net device that got called
 422  * @grp: vlan group being registered
 423  */
 424 static void bond_vlan_rx_register(struct net_device *bond_dev,
 425                                   struct vlan_group *grp)
 426 {
 427         struct bonding *bond = netdev_priv(bond_dev);
 428         struct slave *slave;
 429         int i;
 430
 431         write_lock_bh(&bond->lock);
 432         bond->vlgrp = grp;
 433         write_unlock_bh(&bond->lock);
 434
 435         bond_for_each_slave(bond, slave, i) {
 436                 struct net_device *slave_dev = slave->dev;
 437                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 438
 439                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 440                     slave_ops->ndo_vlan_rx_register) {
 441                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 442                 }
 443         }
 444 }
 445
 446 /**
 447  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 448  * @bond_dev: bonding net device that got called
 449  * @vid: vlan id being added
 450  */
 451 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 452 {
 453         struct bonding *bond = netdev_priv(bond_dev);
 454         struct slave *slave;
 455         int i, res;
 456
 457         bond_for_each_slave(bond, slave, i) {
 458                 struct net_device *slave_dev = slave->dev;
 459                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 460
 461                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 462                     slave_ops->ndo_vlan_rx_add_vid) {
 463                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 464                 }
 465         }
 466
 467         res = bond_add_vlan(bond, vid);
 468         if (res) {
 469                 pr_err("%s: Error: Failed to add vlan id %d\n",
 470                        bond_dev->name, vid);
 471         }
 472 }
 473
 474 /**
 475  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 476  * @bond_dev: bonding net device that got called
 477  * @vid: vlan id being removed
 478  */
 479 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 480 {
 481         struct bonding *bond = netdev_priv(bond_dev);
 482         struct slave *slave;
 483         struct net_device *vlan_dev;
 484         int i, res;
 485
 486         bond_for_each_slave(bond, slave, i) {
 487                 struct net_device *slave_dev = slave->dev;
 488                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 489
 490                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 491                     slave_ops->ndo_vlan_rx_kill_vid) {
 492                         /* Save and then restore vlan_dev in the grp array,
 493                          * since the slave's driver might clear it.
 494                          */
 495                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 496                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 497                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 498                 }
 499         }
 500
 501         res = bond_del_vlan(bond, vid);
 502         if (res) {
 503                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 504                        bond_dev->name, vid);
 505         }
 506 }
 507
 508 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 509 {
 510         struct vlan_entry *vlan;
 511         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 512
 513         if (!bond->vlgrp)
 514                 return;
 515
 516         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 517             slave_ops->ndo_vlan_rx_register)
 518                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 519
 520         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 521             !(slave_ops->ndo_vlan_rx_add_vid))
 522                 return;
 523
 524         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 525                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 526 }
 527
 528 static void bond_del_vlans_from_slave(struct bonding *bond,
 529                                       struct net_device *slave_dev)
 530 {
 531         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 532         struct vlan_entry *vlan;
 533         struct net_device *vlan_dev;
 534
 535         if (!bond->vlgrp)
 536                 return;
 537
 538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 539             !(slave_ops->ndo_vlan_rx_kill_vid))
 540                 goto unreg;
 541
 542         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 543                 if (!vlan->vlan_id)
 544                         continue;
 545                 /* Save and then restore vlan_dev in the grp array,
 546                  * since the slave's driver might clear it.
 547                  */
 548                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 549                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 550                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 551         }
 552
 553 unreg:
 554         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 555             slave_ops->ndo_vlan_rx_register)
 556                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 557 }
 558
 559 /*------------------------------- Link status -------------------------------*/
 560
 561 /*
 562  * Set the carrier state for the master according to the state of its
 563  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 564  * do special 802.3ad magic.
 565  *
 566  * Returns zero if carrier state does not change, nonzero if it does.
 567  */
 568 static int bond_set_carrier(struct bonding *bond)
 569 {
 570         struct slave *slave;
 571         int i;
 572
 573         if (bond->slave_cnt == 0)
 574                 goto down;
 575
 576         if (bond->params.mode == BOND_MODE_8023AD)
 577                 return bond_3ad_set_carrier(bond);
 578
 579         bond_for_each_slave(bond, slave, i) {
 580                 if (slave->link == BOND_LINK_UP) {
 581                         if (!netif_carrier_ok(bond->dev)) {
 582                                 netif_carrier_on(bond->dev);
 583                                 return 1;
 584                         }
 585                         return 0;
 586                 }
 587         }
 588
 589 down:
 590         if (netif_carrier_ok(bond->dev)) {
 591                 netif_carrier_off(bond->dev);
 592                 return 1;
 593         }
 594         return 0;
 595 }
 596
 597 /*
 598  * Get link speed and duplex from the slave's base driver
 599  * using ethtool. If for some reason the call fails or the
 600  * values are invalid, fake speed and duplex to 100/Full
 601  * and return error.
 602  */
 603 static int bond_update_speed_duplex(struct slave *slave)
 604 {
 605         struct net_device *slave_dev = slave->dev;
 606         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 607         u32 slave_speed;
 608         int res;
 609
 610         /* Fake speed and duplex */
 611         slave->speed = SPEED_100;
 612         slave->duplex = DUPLEX_FULL;
 613
 614         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 615                 return -1;
 616
 617         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 618         if (res < 0)
 619                 return -1;
 620
 621         slave_speed = ethtool_cmd_speed(&etool);
 622         switch (slave_speed) {
 623         case SPEED_10:
 624         case SPEED_100:
 625         case SPEED_1000:
 626         case SPEED_10000:
 627                 break;
 628         default:
 629                 return -1;
 630         }
 631
 632         switch (etool.duplex) {
 633         case DUPLEX_FULL:
 634         case DUPLEX_HALF:
 635                 break;
 636         default:
 637                 return -1;
 638         }
 639
 640         slave->speed = slave_speed;
 641         slave->duplex = etool.duplex;
 642
 643         return 0;
 644 }
 645
 646 /*
 647  * if <dev> supports MII link status reporting, check its link status.
 648  *
 649  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 650  * depending upon the setting of the use_carrier parameter.
 651  *
 652  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 653  * can't tell and just pretend it is), or 0, meaning that the link is
 654  * down.
 655  *
 656  * If reporting is non-zero, instead of faking link up, return -1 if
 657  * both ETHTOOL and MII ioctls fail (meaning the device does not
 658  * support them).  If use_carrier is set, return whatever it says.
 659  * It'd be nice if there was a good way to tell if a driver supports
 660  * netif_carrier, but there really isn't.
 661  */
 662 static int bond_check_dev_link(struct bonding *bond,
 663                                struct net_device *slave_dev, int reporting)
 664 {
 665         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 666         int (*ioctl)(struct net_device *, struct ifreq *, int);
 667         struct ifreq ifr;
 668         struct mii_ioctl_data *mii;
 669
 670         if (!reporting && !netif_running(slave_dev))
 671                 return 0;
 672
 673         if (bond->params.use_carrier)
 674                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 675
 676         /* Try to get link status using Ethtool first. */
 677         if (slave_dev->ethtool_ops) {
 678                 if (slave_dev->ethtool_ops->get_link) {
 679                         u32 link;
 680
 681                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 682
 683                         return link ? BMSR_LSTATUS : 0;
 684                 }
 685         }
 686
 687         /* Ethtool can't be used, fallback to MII ioctls. */
 688         ioctl = slave_ops->ndo_do_ioctl;
 689         if (ioctl) {
 690                 /* TODO: set pointer to correct ioctl on a per team member */
 691                 /*       bases to make this more efficient. that is, once  */
 692                 /*       we determine the correct ioctl, we will always    */
 693                 /*       call it and not the others for that team          */
 694                 /*       member.                                           */
 695
 696                 /*
 697                  * We cannot assume that SIOCGMIIPHY will also read a
 698                  * register; not all network drivers (e.g., e100)
 699                  * support that.
 700                  */
 701
 702                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 703                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 704                 mii = if_mii(&ifr);
 705                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 706                         mii->reg_num = MII_BMSR;
 707                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 708                                 return mii->val_out & BMSR_LSTATUS;
 709                 }
 710         }
 711
 712         /*
 713          * If reporting, report that either there's no dev->do_ioctl,
 714          * or both SIOCGMIIREG and get_link failed (meaning that we
 715          * cannot report link status).  If not reporting, pretend
 716          * we're ok.
 717          */
 718         return reporting ? -1 : BMSR_LSTATUS;
 719 }
 720
 721 /*----------------------------- Multicast list ------------------------------*/
 722
 723 /*
 724  * Push the promiscuity flag down to appropriate slaves
 725  */
 726 static int bond_set_promiscuity(struct bonding *bond, int inc)
 727 {
 728         int err = 0;
 729         if (USES_PRIMARY(bond->params.mode)) {
 730                 /* write lock already acquired */
 731                 if (bond->curr_active_slave) {
 732                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 733                                                   inc);
 734                 }
 735         } else {
 736                 struct slave *slave;
 737                 int i;
 738                 bond_for_each_slave(bond, slave, i) {
 739                         err = dev_set_promiscuity(slave->dev, inc);
 740                         if (err)
 741                                 return err;
 742                 }
 743         }
 744         return err;
 745 }
 746
 747 /*
 748  * Push the allmulti flag down to all slaves
 749  */
 750 static int bond_set_allmulti(struct bonding *bond, int inc)
 751 {
 752         int err = 0;
 753         if (USES_PRIMARY(bond->params.mode)) {
 754                 /* write lock already acquired */
 755                 if (bond->curr_active_slave) {
 756                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 757                                                inc);
 758                 }
 759         } else {
 760                 struct slave *slave;
 761                 int i;
 762                 bond_for_each_slave(bond, slave, i) {
 763                         err = dev_set_allmulti(slave->dev, inc);
 764                         if (err)
 765                                 return err;
 766                 }
 767         }
 768         return err;
 769 }
 770
 771 /*
 772  * Add a Multicast address to slaves
 773  * according to mode
 774  */
 775 static void bond_mc_add(struct bonding *bond, void *addr)
 776 {
 777         if (USES_PRIMARY(bond->params.mode)) {
 778                 /* write lock already acquired */
 779                 if (bond->curr_active_slave)
 780                         dev_mc_add(bond->curr_active_slave->dev, addr);
 781         } else {
 782                 struct slave *slave;
 783                 int i;
 784
 785                 bond_for_each_slave(bond, slave, i)
 786                         dev_mc_add(slave->dev, addr);
 787         }
 788 }
 789
 790 /*
 791  * Remove a multicast address from slave
 792  * according to mode
 793  */
 794 static void bond_mc_del(struct bonding *bond, void *addr)
 795 {
 796         if (USES_PRIMARY(bond->params.mode)) {
 797                 /* write lock already acquired */
 798                 if (bond->curr_active_slave)
 799                         dev_mc_del(bond->curr_active_slave->dev, addr);
 800         } else {
 801                 struct slave *slave;
 802                 int i;
 803                 bond_for_each_slave(bond, slave, i) {
 804                         dev_mc_del(slave->dev, addr);
 805                 }
 806         }
 807 }
 808
 809
 810 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 811 {
 812         struct in_device *in_dev;
 813
 814         rcu_read_lock();
 815         in_dev = __in_dev_get_rcu(dev);
 816         if (in_dev)
 817                 ip_mc_rejoin_groups(in_dev);
 818         rcu_read_unlock();
 819 }
 820
 821 /*
 822  * Retrieve the list of registered multicast addresses for the bonding
 823  * device and retransmit an IGMP JOIN request to the current active
 824  * slave.
 825  */
 826 static void bond_resend_igmp_join_requests(struct bonding *bond)
 827 {
 828         struct net_device *vlan_dev;
 829         struct vlan_entry *vlan;
 830
 831         read_lock(&bond->lock);
 832
 833         /* rejoin all groups on bond device */
 834         __bond_resend_igmp_join_requests(bond->dev);
 835
 836         /* rejoin all groups on vlan devices */
 837         if (bond->vlgrp) {
 838                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 839                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 840                                                          vlan->vlan_id);
 841                         if (vlan_dev)
 842                                 __bond_resend_igmp_join_requests(vlan_dev);
 843                 }
 844         }
 845
 846         if (--bond->igmp_retrans > 0)
 847                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 848
 849         read_unlock(&bond->lock);
 850 }
 851
 852 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 853 {
 854         struct bonding *bond = container_of(work, struct bonding,
 855                                                         mcast_work.work);
 856         bond_resend_igmp_join_requests(bond);
 857 }
 858
 859 /*
 860  * flush all members of flush->mc_list from device dev->mc_list
 861  */
 862 static void bond_mc_list_flush(struct net_device *bond_dev,
 863                                struct net_device *slave_dev)
 864 {
 865         struct bonding *bond = netdev_priv(bond_dev);
 866         struct netdev_hw_addr *ha;
 867
 868         netdev_for_each_mc_addr(ha, bond_dev)
 869                 dev_mc_del(slave_dev, ha->addr);
 870
 871         if (bond->params.mode == BOND_MODE_8023AD) {
 872                 /* del lacpdu mc addr from mc list */
 873                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 874
 875                 dev_mc_del(slave_dev, lacpdu_multicast);
 876         }
 877 }
 878
 879 /*--------------------------- Active slave change ---------------------------*/
 880
 881 /*
 882  * Update the mc list and multicast-related flags for the new and
 883  * old active slaves (if any) according to the multicast mode, and
 884  * promiscuous flags unconditionally.
 885  */
 886 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 887                          struct slave *old_active)
 888 {
 889         struct netdev_hw_addr *ha;
 890
 891         if (!USES_PRIMARY(bond->params.mode))
 892                 /* nothing to do -  mc list is already up-to-date on
 893                  * all slaves
 894                  */
 895                 return;
 896
 897         if (old_active) {
 898                 if (bond->dev->flags & IFF_PROMISC)
 899                         dev_set_promiscuity(old_active->dev, -1);
 900
 901                 if (bond->dev->flags & IFF_ALLMULTI)
 902                         dev_set_allmulti(old_active->dev, -1);
 903
 904                 netdev_for_each_mc_addr(ha, bond->dev)
 905                         dev_mc_del(old_active->dev, ha->addr);
 906         }
 907
 908         if (new_active) {
 909                 /* FIXME: Signal errors upstream. */
 910                 if (bond->dev->flags & IFF_PROMISC)
 911                         dev_set_promiscuity(new_active->dev, 1);
 912
 913                 if (bond->dev->flags & IFF_ALLMULTI)
 914                         dev_set_allmulti(new_active->dev, 1);
 915
 916                 netdev_for_each_mc_addr(ha, bond->dev)
 917                         dev_mc_add(new_active->dev, ha->addr);
 918         }
 919 }
 920
 921 /*
 922  * bond_do_fail_over_mac
 923  *
 924  * Perform special MAC address swapping for fail_over_mac settings
 925  *
 926  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 927  */
 928 static void bond_do_fail_over_mac(struct bonding *bond,
 929                                   struct slave *new_active,
 930                                   struct slave *old_active)
 931         __releases(&bond->curr_slave_lock)
 932         __releases(&bond->lock)
 933         __acquires(&bond->lock)
 934         __acquires(&bond->curr_slave_lock)
 935 {
 936         u8 tmp_mac[ETH_ALEN];
 937         struct sockaddr saddr;
 938         int rv;
 939
 940         switch (bond->params.fail_over_mac) {
 941         case BOND_FOM_ACTIVE:
 942                 if (new_active)
 943                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 944                                new_active->dev->addr_len);
 945                 break;
 946         case BOND_FOM_FOLLOW:
 947                 /*
 948                  * if new_active && old_active, swap them
 949                  * if just old_active, do nothing (going to no active slave)
 950                  * if just new_active, set new_active to bond's MAC
 951                  */
 952                 if (!new_active)
 953                         return;
 954
 955                 write_unlock_bh(&bond->curr_slave_lock);
 956                 read_unlock(&bond->lock);
 957
 958                 if (old_active) {
 959                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 960                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 961                                ETH_ALEN);
 962                         saddr.sa_family = new_active->dev->type;
 963                 } else {
 964                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 965                         saddr.sa_family = bond->dev->type;
 966                 }
 967
 968                 rv = dev_set_mac_address(new_active->dev, &saddr);
 969                 if (rv) {
 970                         pr_err("%s: Error %d setting MAC of slave %s\n",
 971                                bond->dev->name, -rv, new_active->dev->name);
 972                         goto out;
 973                 }
 974
 975                 if (!old_active)
 976                         goto out;
 977
 978                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 979                 saddr.sa_family = old_active->dev->type;
 980
 981                 rv = dev_set_mac_address(old_active->dev, &saddr);
 982                 if (rv)
 983                         pr_err("%s: Error %d setting MAC of slave %s\n",
 984                                bond->dev->name, -rv, new_active->dev->name);
 985 out:
 986                 read_lock(&bond->lock);
 987                 write_lock_bh(&bond->curr_slave_lock);
 988                 break;
 989         default:
 990                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 991                        bond->dev->name, bond->params.fail_over_mac);
 992                 break;
 993         }
 994
 995 }
 996
 997 static bool bond_should_change_active(struct bonding *bond)
 998 {
 999         struct slave *prim = bond->primary_slave;
1000         struct slave *curr = bond->curr_active_slave;
1001
1002         if (!prim || !curr || curr->link != BOND_LINK_UP)
1003                 return true;
1004         if (bond->force_primary) {
1005                 bond->force_primary = false;
1006                 return true;
1007         }
1008         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1009             (prim->speed < curr->speed ||
1010              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1011                 return false;
1012         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1013                 return false;
1014         return true;
1015 }
1016
1017 /**
1018  * find_best_interface - select the best available slave to be the active one
1019  * @bond: our bonding struct
1020  *
1021  * Warning: Caller must hold curr_slave_lock for writing.
1022  */
1023 static struct slave *bond_find_best_slave(struct bonding *bond)
1024 {
1025         struct slave *new_active, *old_active;
1026         struct slave *bestslave = NULL;
1027         int mintime = bond->params.updelay;
1028         int i;
1029
1030         new_active = bond->curr_active_slave;
1031
1032         if (!new_active) { /* there were no active slaves left */
1033                 if (bond->slave_cnt > 0)   /* found one slave */
1034                         new_active = bond->first_slave;
1035                 else
1036                         return NULL; /* still no slave, return NULL */
1037         }
1038
1039         if ((bond->primary_slave) &&
1040             bond->primary_slave->link == BOND_LINK_UP &&
1041             bond_should_change_active(bond)) {
1042                 new_active = bond->primary_slave;
1043         }
1044
1045         /* remember where to stop iterating over the slaves */
1046         old_active = new_active;
1047
1048         bond_for_each_slave_from(bond, new_active, i, old_active) {
1049                 if (new_active->link == BOND_LINK_UP) {
1050                         return new_active;
1051                 } else if (new_active->link == BOND_LINK_BACK &&
1052                            IS_UP(new_active->dev)) {
1053                         /* link up, but waiting for stabilization */
1054                         if (new_active->delay < mintime) {
1055                                 mintime = new_active->delay;
1056                                 bestslave = new_active;
1057                         }
1058                 }
1059         }
1060
1061         return bestslave;
1062 }
1063
1064 static bool bond_should_notify_peers(struct bonding *bond)
1065 {
1066         struct slave *slave = bond->curr_active_slave;
1067
1068         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1069                  bond->dev->name, slave ? slave->dev->name : "NULL");
1070
1071         if (!slave || !bond->send_peer_notif ||
1072             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1073                 return false;
1074
1075         bond->send_peer_notif--;
1076         return true;
1077 }
1078
1079 /**
1080  * change_active_interface - change the active slave into the specified one
1081  * @bond: our bonding struct
1082  * @new: the new slave to make the active one
1083  *
1084  * Set the new slave to the bond's settings and unset them on the old
1085  * curr_active_slave.
1086  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1087  *
1088  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1089  * because it is apparently the best available slave we have, even though its
1090  * updelay hasn't timed out yet.
1091  *
1092  * If new_active is not NULL, caller must hold bond->lock for read and
1093  * curr_slave_lock for write_bh.
1094  */
1095 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1096 {
1097         struct slave *old_active = bond->curr_active_slave;
1098
1099         if (old_active == new_active)
1100                 return;
1101
1102         if (new_active) {
1103                 new_active->jiffies = jiffies;
1104
1105                 if (new_active->link == BOND_LINK_BACK) {
1106                         if (USES_PRIMARY(bond->params.mode)) {
1107                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1108                                         bond->dev->name, new_active->dev->name,
1109                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1110                         }
1111
1112                         new_active->delay = 0;
1113                         new_active->link = BOND_LINK_UP;
1114
1115                         if (bond->params.mode == BOND_MODE_8023AD)
1116                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1117
1118                         if (bond_is_lb(bond))
1119                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1120                 } else {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one.\n",
1123                                         bond->dev->name, new_active->dev->name);
1124                         }
1125                 }
1126         }
1127
1128         if (USES_PRIMARY(bond->params.mode))
1129                 bond_mc_swap(bond, new_active, old_active);
1130
1131         if (bond_is_lb(bond)) {
1132                 bond_alb_handle_active_change(bond, new_active);
1133                 if (old_active)
1134                         bond_set_slave_inactive_flags(old_active);
1135                 if (new_active)
1136                         bond_set_slave_active_flags(new_active);
1137         } else {
1138                 bond->curr_active_slave = new_active;
1139         }
1140
1141         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1142                 if (old_active)
1143                         bond_set_slave_inactive_flags(old_active);
1144
1145                 if (new_active) {
1146                         bool should_notify_peers = false;
1147
1148                         bond_set_slave_active_flags(new_active);
1149
1150                         if (bond->params.fail_over_mac)
1151                                 bond_do_fail_over_mac(bond, new_active,
1152                                                       old_active);
1153
1154                         if (netif_running(bond->dev)) {
1155                                 bond->send_peer_notif =
1156                                         bond->params.num_peer_notif;
1157                                 should_notify_peers =
1158                                         bond_should_notify_peers(bond);
1159                         }
1160
1161                         write_unlock_bh(&bond->curr_slave_lock);
1162                         read_unlock(&bond->lock);
1163
1164                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1165                         if (should_notify_peers)
1166                                 netdev_bonding_change(bond->dev,
1167                                                       NETDEV_NOTIFY_PEERS);
1168
1169                         read_lock(&bond->lock);
1170                         write_lock_bh(&bond->curr_slave_lock);
1171                 }
1172         }
1173
1174         /* resend IGMP joins since active slave has changed or
1175          * all were sent on curr_active_slave */
1176         if (((USES_PRIMARY(bond->params.mode) && new_active) ||
1177              bond->params.mode == BOND_MODE_ROUNDROBIN) &&
1178             netif_running(bond->dev)) {
1179                 bond->igmp_retrans = bond->params.resend_igmp;
1180                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1181         }
1182 }
1183
1184 /**
1185  * bond_select_active_slave - select a new active slave, if needed
1186  * @bond: our bonding struct
1187  *
1188  * This functions should be called when one of the following occurs:
1189  * - The old curr_active_slave has been released or lost its link.
1190  * - The primary_slave has got its link back.
1191  * - A slave has got its link back and there's no old curr_active_slave.
1192  *
1193  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1194  */
1195 void bond_select_active_slave(struct bonding *bond)
1196 {
1197         struct slave *best_slave;
1198         int rv;
1199
1200         best_slave = bond_find_best_slave(bond);
1201         if (best_slave != bond->curr_active_slave) {
1202                 bond_change_active_slave(bond, best_slave);
1203                 rv = bond_set_carrier(bond);
1204                 if (!rv)
1205                         return;
1206
1207                 if (netif_carrier_ok(bond->dev)) {
1208                         pr_info("%s: first active interface up!\n",
1209                                 bond->dev->name);
1210                 } else {
1211                         pr_info("%s: now running without any active interface !\n",
1212                                 bond->dev->name);
1213                 }
1214         }
1215 }
1216
1217 /*--------------------------- slave list handling ---------------------------*/
1218
1219 /*
1220  * This function attaches the slave to the end of list.
1221  *
1222  * bond->lock held for writing by caller.
1223  */
1224 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1225 {
1226         if (bond->first_slave == NULL) { /* attaching the first slave */
1227                 new_slave->next = new_slave;
1228                 new_slave->prev = new_slave;
1229                 bond->first_slave = new_slave;
1230         } else {
1231                 new_slave->next = bond->first_slave;
1232                 new_slave->prev = bond->first_slave->prev;
1233                 new_slave->next->prev = new_slave;
1234                 new_slave->prev->next = new_slave;
1235         }
1236
1237         bond->slave_cnt++;
1238 }
1239
1240 /*
1241  * This function detaches the slave from the list.
1242  * WARNING: no check is made to verify if the slave effectively
1243  * belongs to <bond>.
1244  * Nothing is freed on return, structures are just unchained.
1245  * If any slave pointer in bond was pointing to <slave>,
1246  * it should be changed by the calling function.
1247  *
1248  * bond->lock held for writing by caller.
1249  */
1250 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1251 {
1252         if (slave->next)
1253                 slave->next->prev = slave->prev;
1254
1255         if (slave->prev)
1256                 slave->prev->next = slave->next;
1257
1258         if (bond->first_slave == slave) { /* slave is the first slave */
1259                 if (bond->slave_cnt > 1) { /* there are more slave */
1260                         bond->first_slave = slave->next;
1261                 } else {
1262                         bond->first_slave = NULL; /* slave was the last one */
1263                 }
1264         }
1265
1266         slave->next = NULL;
1267         slave->prev = NULL;
1268         bond->slave_cnt--;
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 static inline int slave_enable_netpoll(struct slave *slave)
1273 {
1274         struct netpoll *np;
1275         int err = 0;
1276
1277         np = kzalloc(sizeof(*np), GFP_KERNEL);
1278         err = -ENOMEM;
1279         if (!np)
1280                 goto out;
1281
1282         np->dev = slave->dev;
1283         err = __netpoll_setup(np);
1284         if (err) {
1285                 kfree(np);
1286                 goto out;
1287         }
1288         slave->np = np;
1289 out:
1290         return err;
1291 }
1292 static inline void slave_disable_netpoll(struct slave *slave)
1293 {
1294         struct netpoll *np = slave->np;
1295
1296         if (!np)
1297                 return;
1298
1299         slave->np = NULL;
1300         synchronize_rcu_bh();
1301         __netpoll_cleanup(np);
1302         kfree(np);
1303 }
1304 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1305 {
1306         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1307                 return false;
1308         if (!slave_dev->netdev_ops->ndo_poll_controller)
1309                 return false;
1310         return true;
1311 }
1312
1313 static void bond_poll_controller(struct net_device *bond_dev)
1314 {
1315 }
1316
1317 static void __bond_netpoll_cleanup(struct bonding *bond)
1318 {
1319         struct slave *slave;
1320         int i;
1321
1322         bond_for_each_slave(bond, slave, i)
1323                 if (IS_UP(slave->dev))
1324                         slave_disable_netpoll(slave);
1325 }
1326 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1327 {
1328         struct bonding *bond = netdev_priv(bond_dev);
1329
1330         read_lock(&bond->lock);
1331         __bond_netpoll_cleanup(bond);
1332         read_unlock(&bond->lock);
1333 }
1334
1335 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1336 {
1337         struct bonding *bond = netdev_priv(dev);
1338         struct slave *slave;
1339         int i, err = 0;
1340
1341         read_lock(&bond->lock);
1342         bond_for_each_slave(bond, slave, i) {
1343                 err = slave_enable_netpoll(slave);
1344                 if (err) {
1345                         __bond_netpoll_cleanup(bond);
1346                         break;
1347                 }
1348         }
1349         read_unlock(&bond->lock);
1350         return err;
1351 }
1352
1353 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1354 {
1355         return bond->dev->npinfo;
1356 }
1357
1358 #else
1359 static inline int slave_enable_netpoll(struct slave *slave)
1360 {
1361         return 0;
1362 }
1363 static inline void slave_disable_netpoll(struct slave *slave)
1364 {
1365 }
1366 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1367 {
1368 }
1369 #endif
1370
1371 /*---------------------------------- IOCTL ----------------------------------*/
1372
1373 static int bond_sethwaddr(struct net_device *bond_dev,
1374                           struct net_device *slave_dev)
1375 {
1376         pr_debug("bond_dev=%p\n", bond_dev);
1377         pr_debug("slave_dev=%p\n", slave_dev);
1378         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1379         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1380         return 0;
1381 }
1382
1383 static u32 bond_fix_features(struct net_device *dev, u32 features)
1384 {
1385         struct slave *slave;
1386         struct bonding *bond = netdev_priv(dev);
1387         u32 mask;
1388         int i;
1389
1390         read_lock(&bond->lock);
1391
1392         if (!bond->first_slave) {
1393                 /* Disable adding VLANs to empty bond. But why? --mq */
1394                 features |= NETIF_F_VLAN_CHALLENGED;
1395                 goto out;
1396         }
1397
1398         mask = features;
1399         features &= ~NETIF_F_ONE_FOR_ALL;
1400         features |= NETIF_F_ALL_FOR_ALL;
1401
1402         bond_for_each_slave(bond, slave, i) {
1403                 features = netdev_increment_features(features,
1404                                                      slave->dev->features,
1405                                                      mask);
1406         }
1407
1408 out:
1409         read_unlock(&bond->lock);
1410         return features;
1411 }
1412
1413 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_TX_OFFLOADS | \
1414                                  NETIF_F_SOFT_FEATURES | \
1415                                  NETIF_F_LRO)
1416
1417 static void bond_compute_features(struct bonding *bond)
1418 {
1419         struct slave *slave;
1420         struct net_device *bond_dev = bond->dev;
1421         u32 vlan_features = BOND_VLAN_FEATURES;
1422         unsigned short max_hard_header_len = ETH_HLEN;
1423         int i;
1424
1425         read_lock(&bond->lock);
1426
1427         if (!bond->first_slave)
1428                 goto done;
1429
1430         bond_for_each_slave(bond, slave, i) {
1431                 vlan_features = netdev_increment_features(vlan_features,
1432                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1433
1434                 if (slave->dev->hard_header_len > max_hard_header_len)
1435                         max_hard_header_len = slave->dev->hard_header_len;
1436         }
1437
1438 done:
1439         bond_dev->vlan_features = vlan_features;
1440         bond_dev->hard_header_len = max_hard_header_len;
1441
1442         read_unlock(&bond->lock);
1443
1444         netdev_change_features(bond_dev);
1445 }
1446
1447 static void bond_setup_by_slave(struct net_device *bond_dev,
1448                                 struct net_device *slave_dev)
1449 {
1450         struct bonding *bond = netdev_priv(bond_dev);
1451
1452         bond_dev->header_ops        = slave_dev->header_ops;
1453
1454         bond_dev->type              = slave_dev->type;
1455         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1456         bond_dev->addr_len          = slave_dev->addr_len;
1457
1458         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1459                 slave_dev->addr_len);
1460         bond->setup_by_slave = 1;
1461 }
1462
1463 /* On bonding slaves other than the currently active slave, suppress
1464  * duplicates except for alb non-mcast/bcast.
1465  */
1466 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1467                                             struct slave *slave,
1468                                             struct bonding *bond)
1469 {
1470         if (bond_is_slave_inactive(slave)) {
1471                 if (bond->params.mode == BOND_MODE_ALB &&
1472                     skb->pkt_type != PACKET_BROADCAST &&
1473                     skb->pkt_type != PACKET_MULTICAST)
1474                         return false;
1475                 return true;
1476         }
1477         return false;
1478 }
1479
1480 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1481 {
1482         struct sk_buff *skb = *pskb;
1483         struct slave *slave;
1484         struct bonding *bond;
1485
1486         skb = skb_share_check(skb, GFP_ATOMIC);
1487         if (unlikely(!skb))
1488                 return RX_HANDLER_CONSUMED;
1489
1490         *pskb = skb;
1491
1492         slave = bond_slave_get_rcu(skb->dev);
1493         bond = slave->bond;
1494
1495         if (bond->params.arp_interval)
1496                 slave->dev->last_rx = jiffies;
1497
1498         if (bond->recv_probe) {
1499                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1500
1501                 if (likely(nskb)) {
1502                         bond->recv_probe(nskb, bond, slave);
1503                         dev_kfree_skb(nskb);
1504                 }
1505         }
1506
1507         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1508                 return RX_HANDLER_EXACT;
1509         }
1510
1511         skb->dev = bond->dev;
1512
1513         if (bond->params.mode == BOND_MODE_ALB &&
1514             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1515             skb->pkt_type == PACKET_HOST) {
1516
1517                 if (unlikely(skb_cow_head(skb,
1518                                           skb->data - skb_mac_header(skb)))) {
1519                         kfree_skb(skb);
1520                         return RX_HANDLER_CONSUMED;
1521                 }
1522                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1523         }
1524
1525         return RX_HANDLER_ANOTHER;
1526 }
1527
1528 /* enslave device <slave> to bond device <master> */
1529 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1530 {
1531         struct bonding *bond = netdev_priv(bond_dev);
1532         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1533         struct slave *new_slave = NULL;
1534         struct netdev_hw_addr *ha;
1535         struct sockaddr addr;
1536         int link_reporting;
1537         int res = 0;
1538
1539         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1540                 slave_ops->ndo_do_ioctl == NULL) {
1541                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1542                            bond_dev->name, slave_dev->name);
1543         }
1544
1545         /* bond must be initialized by bond_open() before enslaving */
1546         if (!(bond_dev->flags & IFF_UP)) {
1547                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1548                            bond_dev->name);
1549         }
1550
1551         /* already enslaved */
1552         if (slave_dev->flags & IFF_SLAVE) {
1553                 pr_debug("Error, Device was already enslaved\n");
1554                 return -EBUSY;
1555         }
1556
1557         /* vlan challenged mutual exclusion */
1558         /* no need to lock since we're protected by rtnl_lock */
1559         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1560                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1561                 if (bond->vlgrp) {
1562                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1563                                bond_dev->name, slave_dev->name, bond_dev->name);
1564                         return -EPERM;
1565                 } else {
1566                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1567                                    bond_dev->name, slave_dev->name,
1568                                    slave_dev->name, bond_dev->name);
1569                 }
1570         } else {
1571                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1572         }
1573
1574         /*
1575          * Old ifenslave binaries are no longer supported.  These can
1576          * be identified with moderate accuracy by the state of the slave:
1577          * the current ifenslave will set the interface down prior to
1578          * enslaving it; the old ifenslave will not.
1579          */
1580         if ((slave_dev->flags & IFF_UP)) {
1581                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1582                        slave_dev->name);
1583                 res = -EPERM;
1584                 goto err_undo_flags;
1585         }
1586
1587         /* set bonding device ether type by slave - bonding netdevices are
1588          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1589          * there is a need to override some of the type dependent attribs/funcs.
1590          *
1591          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1592          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1593          */
1594         if (bond->slave_cnt == 0) {
1595                 if (bond_dev->type != slave_dev->type) {
1596                         pr_debug("%s: change device type from %d to %d\n",
1597                                  bond_dev->name,
1598                                  bond_dev->type, slave_dev->type);
1599
1600                         res = netdev_bonding_change(bond_dev,
1601                                                     NETDEV_PRE_TYPE_CHANGE);
1602                         res = notifier_to_errno(res);
1603                         if (res) {
1604                                 pr_err("%s: refused to change device type\n",
1605                                        bond_dev->name);
1606                                 res = -EBUSY;
1607                                 goto err_undo_flags;
1608                         }
1609
1610                         /* Flush unicast and multicast addresses */
1611                         dev_uc_flush(bond_dev);
1612                         dev_mc_flush(bond_dev);
1613
1614                         if (slave_dev->type != ARPHRD_ETHER)
1615                                 bond_setup_by_slave(bond_dev, slave_dev);
1616                         else
1617                                 ether_setup(bond_dev);
1618
1619                         netdev_bonding_change(bond_dev,
1620                                               NETDEV_POST_TYPE_CHANGE);
1621                 }
1622         } else if (bond_dev->type != slave_dev->type) {
1623                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1624                        slave_dev->name,
1625                        slave_dev->type, bond_dev->type);
1626                 res = -EINVAL;
1627                 goto err_undo_flags;
1628         }
1629
1630         if (slave_ops->ndo_set_mac_address == NULL) {
1631                 if (bond->slave_cnt == 0) {
1632                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1633                                    bond_dev->name);
1634                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1635                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1636                         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",
1637                                bond_dev->name);
1638                         res = -EOPNOTSUPP;
1639                         goto err_undo_flags;
1640                 }
1641         }
1642
1643         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1644
1645         /* If this is the first slave, then we need to set the master's hardware
1646          * address to be the same as the slave's. */
1647         if (is_zero_ether_addr(bond->dev->dev_addr))
1648                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1649                        slave_dev->addr_len);
1650
1651
1652         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1653         if (!new_slave) {
1654                 res = -ENOMEM;
1655                 goto err_undo_flags;
1656         }
1657
1658         /*
1659          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1660          * is set via sysfs or module option if desired.
1661          */
1662         new_slave->queue_id = 0;
1663
1664         /* Save slave's original mtu and then set it to match the bond */
1665         new_slave->original_mtu = slave_dev->mtu;
1666         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1667         if (res) {
1668                 pr_debug("Error %d calling dev_set_mtu\n", res);
1669                 goto err_free;
1670         }
1671
1672         /*
1673          * Save slave's original ("permanent") mac address for modes
1674          * that need it, and for restoring it upon release, and then
1675          * set it to the master's address
1676          */
1677         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1678
1679         if (!bond->params.fail_over_mac) {
1680                 /*
1681                  * Set slave to master's mac address.  The application already
1682                  * set the master's mac address to that of the first slave
1683                  */
1684                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1685                 addr.sa_family = slave_dev->type;
1686                 res = dev_set_mac_address(slave_dev, &addr);
1687                 if (res) {
1688                         pr_debug("Error %d calling set_mac_address\n", res);
1689                         goto err_restore_mtu;
1690                 }
1691         }
1692
1693         res = netdev_set_bond_master(slave_dev, bond_dev);
1694         if (res) {
1695                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1696                 goto err_restore_mac;
1697         }
1698
1699         /* open the slave since the application closed it */
1700         res = dev_open(slave_dev);
1701         if (res) {
1702                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1703                 goto err_unset_master;
1704         }
1705
1706         new_slave->bond = bond;
1707         new_slave->dev = slave_dev;
1708         slave_dev->priv_flags |= IFF_BONDING;
1709
1710         if (bond_is_lb(bond)) {
1711                 /* bond_alb_init_slave() must be called before all other stages since
1712                  * it might fail and we do not want to have to undo everything
1713                  */
1714                 res = bond_alb_init_slave(bond, new_slave);
1715                 if (res)
1716                         goto err_close;
1717         }
1718
1719         /* If the mode USES_PRIMARY, then the new slave gets the
1720          * master's promisc (and mc) settings only if it becomes the
1721          * curr_active_slave, and that is taken care of later when calling
1722          * bond_change_active()
1723          */
1724         if (!USES_PRIMARY(bond->params.mode)) {
1725                 /* set promiscuity level to new slave */
1726                 if (bond_dev->flags & IFF_PROMISC) {
1727                         res = dev_set_promiscuity(slave_dev, 1);
1728                         if (res)
1729                                 goto err_close;
1730                 }
1731
1732                 /* set allmulti level to new slave */
1733                 if (bond_dev->flags & IFF_ALLMULTI) {
1734                         res = dev_set_allmulti(slave_dev, 1);
1735                         if (res)
1736                                 goto err_close;
1737                 }
1738
1739                 netif_addr_lock_bh(bond_dev);
1740                 /* upload master's mc_list to new slave */
1741                 netdev_for_each_mc_addr(ha, bond_dev)
1742                         dev_mc_add(slave_dev, ha->addr);
1743                 netif_addr_unlock_bh(bond_dev);
1744         }
1745
1746         if (bond->params.mode == BOND_MODE_8023AD) {
1747                 /* add lacpdu mc addr to mc list */
1748                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1749
1750                 dev_mc_add(slave_dev, lacpdu_multicast);
1751         }
1752
1753         bond_add_vlans_on_slave(bond, slave_dev);
1754
1755         write_lock_bh(&bond->lock);
1756
1757         bond_attach_slave(bond, new_slave);
1758
1759         new_slave->delay = 0;
1760         new_slave->link_failure_count = 0;
1761
1762         write_unlock_bh(&bond->lock);
1763
1764         bond_compute_features(bond);
1765
1766         read_lock(&bond->lock);
1767
1768         new_slave->last_arp_rx = jiffies;
1769
1770         if (bond->params.miimon && !bond->params.use_carrier) {
1771                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1772
1773                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1774                         /*
1775                          * miimon is set but a bonded network driver
1776                          * does not support ETHTOOL/MII and
1777                          * arp_interval is not set.  Note: if
1778                          * use_carrier is enabled, we will never go
1779                          * here (because netif_carrier is always
1780                          * supported); thus, we don't need to change
1781                          * the messages for netif_carrier.
1782                          */
1783                         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",
1784                                bond_dev->name, slave_dev->name);
1785                 } else if (link_reporting == -1) {
1786                         /* unable get link status using mii/ethtool */
1787                         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",
1788                                    bond_dev->name, slave_dev->name);
1789                 }
1790         }
1791
1792         /* check for initial state */
1793         if (!bond->params.miimon ||
1794             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1795                 if (bond->params.updelay) {
1796                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1797                         new_slave->link  = BOND_LINK_BACK;
1798                         new_slave->delay = bond->params.updelay;
1799                 } else {
1800                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1801                         new_slave->link  = BOND_LINK_UP;
1802                 }
1803                 new_slave->jiffies = jiffies;
1804         } else {
1805                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1806                 new_slave->link  = BOND_LINK_DOWN;
1807         }
1808
1809         if (bond_update_speed_duplex(new_slave) &&
1810             (new_slave->link != BOND_LINK_DOWN)) {
1811                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1812                            bond_dev->name, new_slave->dev->name);
1813
1814                 if (bond->params.mode == BOND_MODE_8023AD) {
1815                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1816                                    bond_dev->name);
1817                 }
1818         }
1819
1820         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1821                 /* if there is a primary slave, remember it */
1822                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1823                         bond->primary_slave = new_slave;
1824                         bond->force_primary = true;
1825                 }
1826         }
1827
1828         write_lock_bh(&bond->curr_slave_lock);
1829
1830         switch (bond->params.mode) {
1831         case BOND_MODE_ACTIVEBACKUP:
1832                 bond_set_slave_inactive_flags(new_slave);
1833                 bond_select_active_slave(bond);
1834                 break;
1835         case BOND_MODE_8023AD:
1836                 /* in 802.3ad mode, the internal mechanism
1837                  * will activate the slaves in the selected
1838                  * aggregator
1839                  */
1840                 bond_set_slave_inactive_flags(new_slave);
1841                 /* if this is the first slave */
1842                 if (bond->slave_cnt == 1) {
1843                         SLAVE_AD_INFO(new_slave).id = 1;
1844                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1845                          * can be called only after the mac address of the bond is set
1846                          */
1847                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1848                                             bond->params.lacp_fast);
1849                 } else {
1850                         SLAVE_AD_INFO(new_slave).id =
1851                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1852                 }
1853
1854                 bond_3ad_bind_slave(new_slave);
1855                 break;
1856         case BOND_MODE_TLB:
1857         case BOND_MODE_ALB:
1858                 bond_set_active_slave(new_slave);
1859                 bond_set_slave_inactive_flags(new_slave);
1860                 bond_select_active_slave(bond);
1861                 break;
1862         default:
1863                 pr_debug("This slave is always active in trunk mode\n");
1864
1865                 /* always active in trunk mode */
1866                 bond_set_active_slave(new_slave);
1867
1868                 /* In trunking mode there is little meaning to curr_active_slave
1869                  * anyway (it holds no special properties of the bond device),
1870                  * so we can change it without calling change_active_interface()
1871                  */
1872                 if (!bond->curr_active_slave)
1873                         bond->curr_active_slave = new_slave;
1874
1875                 break;
1876         } /* switch(bond_mode) */
1877
1878         write_unlock_bh(&bond->curr_slave_lock);
1879
1880         bond_set_carrier(bond);
1881
1882 #ifdef CONFIG_NET_POLL_CONTROLLER
1883         slave_dev->npinfo = bond_netpoll_info(bond);
1884         if (slave_dev->npinfo) {
1885                 if (slave_enable_netpoll(new_slave)) {
1886                         read_unlock(&bond->lock);
1887                         pr_info("Error, %s: master_dev is using netpoll, "
1888                                  "but new slave device does not support netpoll.\n",
1889                                  bond_dev->name);
1890                         res = -EBUSY;
1891                         goto err_close;
1892                 }
1893         }
1894 #endif
1895
1896         read_unlock(&bond->lock);
1897
1898         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1899         if (res)
1900                 goto err_close;
1901
1902         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1903                                          new_slave);
1904         if (res) {
1905                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1906                 goto err_dest_symlinks;
1907         }
1908
1909         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1910                 bond_dev->name, slave_dev->name,
1911                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1912                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1913
1914         /* enslave is successful */
1915         return 0;
1916
1917 /* Undo stages on error */
1918 err_dest_symlinks:
1919         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1920
1921 err_close:
1922         dev_close(slave_dev);
1923
1924 err_unset_master:
1925         netdev_set_bond_master(slave_dev, NULL);
1926
1927 err_restore_mac:
1928         if (!bond->params.fail_over_mac) {
1929                 /* XXX TODO - fom follow mode needs to change master's
1930                  * MAC if this slave's MAC is in use by the bond, or at
1931                  * least print a warning.
1932                  */
1933                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1934                 addr.sa_family = slave_dev->type;
1935                 dev_set_mac_address(slave_dev, &addr);
1936         }
1937
1938 err_restore_mtu:
1939         dev_set_mtu(slave_dev, new_slave->original_mtu);
1940
1941 err_free:
1942         kfree(new_slave);
1943
1944 err_undo_flags:
1945         bond_compute_features(bond);
1946
1947         return res;
1948 }
1949
1950 /*
1951  * Try to release the slave device <slave> from the bond device <master>
1952  * It is legal to access curr_active_slave without a lock because all the function
1953  * is write-locked.
1954  *
1955  * The rules for slave state should be:
1956  *   for Active/Backup:
1957  *     Active stays on all backups go down
1958  *   for Bonded connections:
1959  *     The first up interface should be left on and all others downed.
1960  */
1961 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1962 {
1963         struct bonding *bond = netdev_priv(bond_dev);
1964         struct slave *slave, *oldcurrent;
1965         struct sockaddr addr;
1966         u32 old_features = bond_dev->features;
1967
1968         /* slave is not a slave or master is not master of this slave */
1969         if (!(slave_dev->flags & IFF_SLAVE) ||
1970             (slave_dev->master != bond_dev)) {
1971                 pr_err("%s: Error: cannot release %s.\n",
1972                        bond_dev->name, slave_dev->name);
1973                 return -EINVAL;
1974         }
1975
1976         block_netpoll_tx();
1977         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1978         write_lock_bh(&bond->lock);
1979
1980         slave = bond_get_slave_by_dev(bond, slave_dev);
1981         if (!slave) {
1982                 /* not a slave of this bond */
1983                 pr_info("%s: %s not enslaved\n",
1984                         bond_dev->name, slave_dev->name);
1985                 write_unlock_bh(&bond->lock);
1986                 unblock_netpoll_tx();
1987                 return -EINVAL;
1988         }
1989
1990         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1991          * for this slave anymore.
1992          */
1993         netdev_rx_handler_unregister(slave_dev);
1994         write_unlock_bh(&bond->lock);
1995         synchronize_net();
1996         write_lock_bh(&bond->lock);
1997
1998         if (!bond->params.fail_over_mac) {
1999                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
2000                     bond->slave_cnt > 1)
2001                         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",
2002                                    bond_dev->name, slave_dev->name,
2003                                    slave->perm_hwaddr,
2004                                    bond_dev->name, slave_dev->name);
2005         }
2006
2007         /* Inform AD package of unbinding of slave. */
2008         if (bond->params.mode == BOND_MODE_8023AD) {
2009                 /* must be called before the slave is
2010                  * detached from the list
2011                  */
2012                 bond_3ad_unbind_slave(slave);
2013         }
2014
2015         pr_info("%s: releasing %s interface %s\n",
2016                 bond_dev->name,
2017                 bond_is_active_slave(slave) ? "active" : "backup",
2018                 slave_dev->name);
2019
2020         oldcurrent = bond->curr_active_slave;
2021
2022         bond->current_arp_slave = NULL;
2023
2024         /* release the slave from its bond */
2025         bond_detach_slave(bond, slave);
2026
2027         if (bond->primary_slave == slave)
2028                 bond->primary_slave = NULL;
2029
2030         if (oldcurrent == slave)
2031                 bond_change_active_slave(bond, NULL);
2032
2033         if (bond_is_lb(bond)) {
2034                 /* Must be called only after the slave has been
2035                  * detached from the list and the curr_active_slave
2036                  * has been cleared (if our_slave == old_current),
2037                  * but before a new active slave is selected.
2038                  */
2039                 write_unlock_bh(&bond->lock);
2040                 bond_alb_deinit_slave(bond, slave);
2041                 write_lock_bh(&bond->lock);
2042         }
2043
2044         if (oldcurrent == slave) {
2045                 /*
2046                  * Note that we hold RTNL over this sequence, so there
2047                  * is no concern that another slave add/remove event
2048                  * will interfere.
2049                  */
2050                 write_unlock_bh(&bond->lock);
2051                 read_lock(&bond->lock);
2052                 write_lock_bh(&bond->curr_slave_lock);
2053
2054                 bond_select_active_slave(bond);
2055
2056                 write_unlock_bh(&bond->curr_slave_lock);
2057                 read_unlock(&bond->lock);
2058                 write_lock_bh(&bond->lock);
2059         }
2060
2061         if (bond->slave_cnt == 0) {
2062                 bond_set_carrier(bond);
2063
2064                 /* if the last slave was removed, zero the mac address
2065                  * of the master so it will be set by the application
2066                  * to the mac address of the first slave
2067                  */
2068                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2069
2070                 if (bond->vlgrp) {
2071                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2072                                    bond_dev->name, bond_dev->name);
2073                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2074                                    bond_dev->name);
2075                 }
2076         }
2077
2078         write_unlock_bh(&bond->lock);
2079         unblock_netpoll_tx();
2080
2081         bond_compute_features(bond);
2082         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2083             (old_features & NETIF_F_VLAN_CHALLENGED))
2084                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2085                         bond_dev->name, slave_dev->name, bond_dev->name);
2086
2087         /* must do this from outside any spinlocks */
2088         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2089
2090         bond_del_vlans_from_slave(bond, slave_dev);
2091
2092         /* If the mode USES_PRIMARY, then we should only remove its
2093          * promisc and mc settings if it was the curr_active_slave, but that was
2094          * already taken care of above when we detached the slave
2095          */
2096         if (!USES_PRIMARY(bond->params.mode)) {
2097                 /* unset promiscuity level from slave */
2098                 if (bond_dev->flags & IFF_PROMISC)
2099                         dev_set_promiscuity(slave_dev, -1);
2100
2101                 /* unset allmulti level from slave */
2102                 if (bond_dev->flags & IFF_ALLMULTI)
2103                         dev_set_allmulti(slave_dev, -1);
2104
2105                 /* flush master's mc_list from slave */
2106                 netif_addr_lock_bh(bond_dev);
2107                 bond_mc_list_flush(bond_dev, slave_dev);
2108                 netif_addr_unlock_bh(bond_dev);
2109         }
2110
2111         netdev_set_bond_master(slave_dev, NULL);
2112
2113         slave_disable_netpoll(slave);
2114
2115         /* close slave before restoring its mac address */
2116         dev_close(slave_dev);
2117
2118         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2119                 /* restore original ("permanent") mac address */
2120                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2121                 addr.sa_family = slave_dev->type;
2122                 dev_set_mac_address(slave_dev, &addr);
2123         }
2124
2125         dev_set_mtu(slave_dev, slave->original_mtu);
2126
2127         slave_dev->priv_flags &= ~IFF_BONDING;
2128
2129         kfree(slave);
2130
2131         return 0;  /* deletion OK */
2132 }
2133
2134 /*
2135 * First release a slave and then destroy the bond if no more slaves are left.
2136 * Must be under rtnl_lock when this function is called.
2137 */
2138 static int  bond_release_and_destroy(struct net_device *bond_dev,
2139                                      struct net_device *slave_dev)
2140 {
2141         struct bonding *bond = netdev_priv(bond_dev);
2142         int ret;
2143
2144         ret = bond_release(bond_dev, slave_dev);
2145         if ((ret == 0) && (bond->slave_cnt == 0)) {
2146                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2147                 pr_info("%s: destroying bond %s.\n",
2148                         bond_dev->name, bond_dev->name);
2149                 unregister_netdevice(bond_dev);
2150         }
2151         return ret;
2152 }
2153
2154 /*
2155  * This function releases all slaves.
2156  */
2157 static int bond_release_all(struct net_device *bond_dev)
2158 {
2159         struct bonding *bond = netdev_priv(bond_dev);
2160         struct slave *slave;
2161         struct net_device *slave_dev;
2162         struct sockaddr addr;
2163
2164         write_lock_bh(&bond->lock);
2165
2166         netif_carrier_off(bond_dev);
2167
2168         if (bond->slave_cnt == 0)
2169                 goto out;
2170
2171         bond->current_arp_slave = NULL;
2172         bond->primary_slave = NULL;
2173         bond_change_active_slave(bond, NULL);
2174
2175         while ((slave = bond->first_slave) != NULL) {
2176                 /* Inform AD package of unbinding of slave
2177                  * before slave is detached from the list.
2178                  */
2179                 if (bond->params.mode == BOND_MODE_8023AD)
2180                         bond_3ad_unbind_slave(slave);
2181
2182                 slave_dev = slave->dev;
2183                 bond_detach_slave(bond, slave);
2184
2185                 /* now that the slave is detached, unlock and perform
2186                  * all the undo steps that should not be called from
2187                  * within a lock.
2188                  */
2189                 write_unlock_bh(&bond->lock);
2190
2191                 /* unregister rx_handler early so bond_handle_frame wouldn't
2192                  * be called for this slave anymore.
2193                  */
2194                 netdev_rx_handler_unregister(slave_dev);
2195                 synchronize_net();
2196
2197                 if (bond_is_lb(bond)) {
2198                         /* must be called only after the slave
2199                          * has been detached from the list
2200                          */
2201                         bond_alb_deinit_slave(bond, slave);
2202                 }
2203
2204                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2205                 bond_del_vlans_from_slave(bond, slave_dev);
2206
2207                 /* If the mode USES_PRIMARY, then we should only remove its
2208                  * promisc and mc settings if it was the curr_active_slave, but that was
2209                  * already taken care of above when we detached the slave
2210                  */
2211                 if (!USES_PRIMARY(bond->params.mode)) {
2212                         /* unset promiscuity level from slave */
2213                         if (bond_dev->flags & IFF_PROMISC)
2214                                 dev_set_promiscuity(slave_dev, -1);
2215
2216                         /* unset allmulti level from slave */
2217                         if (bond_dev->flags & IFF_ALLMULTI)
2218                                 dev_set_allmulti(slave_dev, -1);
2219
2220                         /* flush master's mc_list from slave */
2221                         netif_addr_lock_bh(bond_dev);
2222                         bond_mc_list_flush(bond_dev, slave_dev);
2223                         netif_addr_unlock_bh(bond_dev);
2224                 }
2225
2226                 netdev_set_bond_master(slave_dev, NULL);
2227
2228                 slave_disable_netpoll(slave);
2229
2230                 /* close slave before restoring its mac address */
2231                 dev_close(slave_dev);
2232
2233                 if (!bond->params.fail_over_mac) {
2234                         /* restore original ("permanent") mac address*/
2235                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2236                         addr.sa_family = slave_dev->type;
2237                         dev_set_mac_address(slave_dev, &addr);
2238                 }
2239
2240                 kfree(slave);
2241
2242                 /* re-acquire the lock before getting the next slave */
2243                 write_lock_bh(&bond->lock);
2244         }
2245
2246         /* zero the mac address of the master so it will be
2247          * set by the application to the mac address of the
2248          * first slave
2249          */
2250         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2251
2252         if (bond->vlgrp) {
2253                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2254                            bond_dev->name, bond_dev->name);
2255                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2256                            bond_dev->name);
2257         }
2258
2259         pr_info("%s: released all slaves\n", bond_dev->name);
2260
2261 out:
2262         write_unlock_bh(&bond->lock);
2263
2264         bond_compute_features(bond);
2265
2266         return 0;
2267 }
2268
2269 /*
2270  * This function changes the active slave to slave <slave_dev>.
2271  * It returns -EINVAL in the following cases.
2272  *  - <slave_dev> is not found in the list.
2273  *  - There is not active slave now.
2274  *  - <slave_dev> is already active.
2275  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2276  *  - <slave_dev> is not running.
2277  * In these cases, this function does nothing.
2278  * In the other cases, current_slave pointer is changed and 0 is returned.
2279  */
2280 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2281 {
2282         struct bonding *bond = netdev_priv(bond_dev);
2283         struct slave *old_active = NULL;
2284         struct slave *new_active = NULL;
2285         int res = 0;
2286
2287         if (!USES_PRIMARY(bond->params.mode))
2288                 return -EINVAL;
2289
2290         /* Verify that master_dev is indeed the master of slave_dev */
2291         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2292                 return -EINVAL;
2293
2294         read_lock(&bond->lock);
2295
2296         read_lock(&bond->curr_slave_lock);
2297         old_active = bond->curr_active_slave;
2298         read_unlock(&bond->curr_slave_lock);
2299
2300         new_active = bond_get_slave_by_dev(bond, slave_dev);
2301
2302         /*
2303          * Changing to the current active: do nothing; return success.
2304          */
2305         if (new_active && (new_active == old_active)) {
2306                 read_unlock(&bond->lock);
2307                 return 0;
2308         }
2309
2310         if ((new_active) &&
2311             (old_active) &&
2312             (new_active->link == BOND_LINK_UP) &&
2313             IS_UP(new_active->dev)) {
2314                 block_netpoll_tx();
2315                 write_lock_bh(&bond->curr_slave_lock);
2316                 bond_change_active_slave(bond, new_active);
2317                 write_unlock_bh(&bond->curr_slave_lock);
2318                 unblock_netpoll_tx();
2319         } else
2320                 res = -EINVAL;
2321
2322         read_unlock(&bond->lock);
2323
2324         return res;
2325 }
2326
2327 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2328 {
2329         struct bonding *bond = netdev_priv(bond_dev);
2330
2331         info->bond_mode = bond->params.mode;
2332         info->miimon = bond->params.miimon;
2333
2334         read_lock(&bond->lock);
2335         info->num_slaves = bond->slave_cnt;
2336         read_unlock(&bond->lock);
2337
2338         return 0;
2339 }
2340
2341 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2342 {
2343         struct bonding *bond = netdev_priv(bond_dev);
2344         struct slave *slave;
2345         int i, res = -ENODEV;
2346
2347         read_lock(&bond->lock);
2348
2349         bond_for_each_slave(bond, slave, i) {
2350                 if (i == (int)info->slave_id) {
2351                         res = 0;
2352                         strcpy(info->slave_name, slave->dev->name);
2353                         info->link = slave->link;
2354                         info->state = bond_slave_state(slave);
2355                         info->link_failure_count = slave->link_failure_count;
2356                         break;
2357                 }
2358         }
2359
2360         read_unlock(&bond->lock);
2361
2362         return res;
2363 }
2364
2365 /*-------------------------------- Monitoring -------------------------------*/
2366
2367
2368 static int bond_miimon_inspect(struct bonding *bond)
2369 {
2370         struct slave *slave;
2371         int i, link_state, commit = 0;
2372         bool ignore_updelay;
2373
2374         ignore_updelay = !bond->curr_active_slave ? true : false;
2375
2376         bond_for_each_slave(bond, slave, i) {
2377                 slave->new_link = BOND_LINK_NOCHANGE;
2378
2379                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2380
2381                 switch (slave->link) {
2382                 case BOND_LINK_UP:
2383                         if (link_state)
2384                                 continue;
2385
2386                         slave->link = BOND_LINK_FAIL;
2387                         slave->delay = bond->params.downdelay;
2388                         if (slave->delay) {
2389                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2390                                         bond->dev->name,
2391                                         (bond->params.mode ==
2392                                          BOND_MODE_ACTIVEBACKUP) ?
2393                                         (bond_is_active_slave(slave) ?
2394                                          "active " : "backup ") : "",
2395                                         slave->dev->name,
2396                                         bond->params.downdelay * bond->params.miimon);
2397                         }
2398                         /*FALLTHRU*/
2399                 case BOND_LINK_FAIL:
2400                         if (link_state) {
2401                                 /*
2402                                  * recovered before downdelay expired
2403                                  */
2404                                 slave->link = BOND_LINK_UP;
2405                                 slave->jiffies = jiffies;
2406                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2407                                         bond->dev->name,
2408                                         (bond->params.downdelay - slave->delay) *
2409                                         bond->params.miimon,
2410                                         slave->dev->name);
2411                                 continue;
2412                         }
2413
2414                         if (slave->delay <= 0) {
2415                                 slave->new_link = BOND_LINK_DOWN;
2416                                 commit++;
2417                                 continue;
2418                         }
2419
2420                         slave->delay--;
2421                         break;
2422
2423                 case BOND_LINK_DOWN:
2424                         if (!link_state)
2425                                 continue;
2426
2427                         slave->link = BOND_LINK_BACK;
2428                         slave->delay = bond->params.updelay;
2429
2430                         if (slave->delay) {
2431                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2432                                         bond->dev->name, slave->dev->name,
2433                                         ignore_updelay ? 0 :
2434                                         bond->params.updelay *
2435                                         bond->params.miimon);
2436                         }
2437                         /*FALLTHRU*/
2438                 case BOND_LINK_BACK:
2439                         if (!link_state) {
2440                                 slave->link = BOND_LINK_DOWN;
2441                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2442                                         bond->dev->name,
2443                                         (bond->params.updelay - slave->delay) *
2444                                         bond->params.miimon,
2445                                         slave->dev->name);
2446
2447                                 continue;
2448                         }
2449
2450                         if (ignore_updelay)
2451                                 slave->delay = 0;
2452
2453                         if (slave->delay <= 0) {
2454                                 slave->new_link = BOND_LINK_UP;
2455                                 commit++;
2456                                 ignore_updelay = false;
2457                                 continue;
2458                         }
2459
2460                         slave->delay--;
2461                         break;
2462                 }
2463         }
2464
2465         return commit;
2466 }
2467
2468 static void bond_miimon_commit(struct bonding *bond)
2469 {
2470         struct slave *slave;
2471         int i;
2472
2473         bond_for_each_slave(bond, slave, i) {
2474                 switch (slave->new_link) {
2475                 case BOND_LINK_NOCHANGE:
2476                         continue;
2477
2478                 case BOND_LINK_UP:
2479                         slave->link = BOND_LINK_UP;
2480                         slave->jiffies = jiffies;
2481
2482                         if (bond->params.mode == BOND_MODE_8023AD) {
2483                                 /* prevent it from being the active one */
2484                                 bond_set_backup_slave(slave);
2485                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2486                                 /* make it immediately active */
2487                                 bond_set_active_slave(slave);
2488                         } else if (slave != bond->primary_slave) {
2489                                 /* prevent it from being the active one */
2490                                 bond_set_backup_slave(slave);
2491                         }
2492
2493                         bond_update_speed_duplex(slave);
2494
2495                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2496                                 bond->dev->name, slave->dev->name,
2497                                 slave->speed, slave->duplex ? "full" : "half");
2498
2499                         /* notify ad that the link status has changed */
2500                         if (bond->params.mode == BOND_MODE_8023AD)
2501                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2502
2503                         if (bond_is_lb(bond))
2504                                 bond_alb_handle_link_change(bond, slave,
2505                                                             BOND_LINK_UP);
2506
2507                         if (!bond->curr_active_slave ||
2508                             (slave == bond->primary_slave))
2509                                 goto do_failover;
2510
2511                         continue;
2512
2513                 case BOND_LINK_DOWN:
2514                         if (slave->link_failure_count < UINT_MAX)
2515                                 slave->link_failure_count++;
2516
2517                         slave->link = BOND_LINK_DOWN;
2518
2519                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2520                             bond->params.mode == BOND_MODE_8023AD)
2521                                 bond_set_slave_inactive_flags(slave);
2522
2523                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2524                                 bond->dev->name, slave->dev->name);
2525
2526                         if (bond->params.mode == BOND_MODE_8023AD)
2527                                 bond_3ad_handle_link_change(slave,
2528                                                             BOND_LINK_DOWN);
2529
2530                         if (bond_is_lb(bond))
2531                                 bond_alb_handle_link_change(bond, slave,
2532                                                             BOND_LINK_DOWN);
2533
2534                         if (slave == bond->curr_active_slave)
2535                                 goto do_failover;
2536
2537                         continue;
2538
2539                 default:
2540                         pr_err("%s: invalid new link %d on slave %s\n",
2541                                bond->dev->name, slave->new_link,
2542                                slave->dev->name);
2543                         slave->new_link = BOND_LINK_NOCHANGE;
2544
2545                         continue;
2546                 }
2547
2548 do_failover:
2549                 ASSERT_RTNL();
2550                 block_netpoll_tx();
2551                 write_lock_bh(&bond->curr_slave_lock);
2552                 bond_select_active_slave(bond);
2553                 write_unlock_bh(&bond->curr_slave_lock);
2554                 unblock_netpoll_tx();
2555         }
2556
2557         bond_set_carrier(bond);
2558 }
2559
2560 /*
2561  * bond_mii_monitor
2562  *
2563  * Really a wrapper that splits the mii monitor into two phases: an
2564  * inspection, then (if inspection indicates something needs to be done)
2565  * an acquisition of appropriate locks followed by a commit phase to
2566  * implement whatever link state changes are indicated.
2567  */
2568 void bond_mii_monitor(struct work_struct *work)
2569 {
2570         struct bonding *bond = container_of(work, struct bonding,
2571                                             mii_work.work);
2572         bool should_notify_peers = false;
2573
2574         read_lock(&bond->lock);
2575         if (bond->kill_timers)
2576                 goto out;
2577
2578         if (bond->slave_cnt == 0)
2579                 goto re_arm;
2580
2581         should_notify_peers = bond_should_notify_peers(bond);
2582
2583         if (bond_miimon_inspect(bond)) {
2584                 read_unlock(&bond->lock);
2585                 rtnl_lock();
2586                 read_lock(&bond->lock);
2587
2588                 bond_miimon_commit(bond);
2589
2590                 read_unlock(&bond->lock);
2591                 rtnl_unlock();  /* might sleep, hold no other locks */
2592                 read_lock(&bond->lock);
2593         }
2594
2595 re_arm:
2596         if (bond->params.miimon)
2597                 queue_delayed_work(bond->wq, &bond->mii_work,
2598                                    msecs_to_jiffies(bond->params.miimon));
2599 out:
2600         read_unlock(&bond->lock);
2601
2602         if (should_notify_peers) {
2603                 rtnl_lock();
2604                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2605                 rtnl_unlock();
2606         }
2607 }
2608
2609 static __be32 bond_glean_dev_ip(struct net_device *dev)
2610 {
2611         struct in_device *idev;
2612         struct in_ifaddr *ifa;
2613         __be32 addr = 0;
2614
2615         if (!dev)
2616                 return 0;
2617
2618         rcu_read_lock();
2619         idev = __in_dev_get_rcu(dev);
2620         if (!idev)
2621                 goto out;
2622
2623         ifa = idev->ifa_list;
2624         if (!ifa)
2625                 goto out;
2626
2627         addr = ifa->ifa_local;
2628 out:
2629         rcu_read_unlock();
2630         return addr;
2631 }
2632
2633 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2634 {
2635         struct vlan_entry *vlan;
2636
2637         if (ip == bond->master_ip)
2638                 return 1;
2639
2640         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2641                 if (ip == vlan->vlan_ip)
2642                         return 1;
2643         }
2644
2645         return 0;
2646 }
2647
2648 /*
2649  * We go to the (large) trouble of VLAN tagging ARP frames because
2650  * switches in VLAN mode (especially if ports are configured as
2651  * "native" to a VLAN) might not pass non-tagged frames.
2652  */
2653 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2654 {
2655         struct sk_buff *skb;
2656
2657         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2658                  slave_dev->name, dest_ip, src_ip, vlan_id);
2659
2660         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2661                          NULL, slave_dev->dev_addr, NULL);
2662
2663         if (!skb) {
2664                 pr_err("ARP packet allocation failed\n");
2665                 return;
2666         }
2667         if (vlan_id) {
2668                 skb = vlan_put_tag(skb, vlan_id);
2669                 if (!skb) {
2670                         pr_err("failed to insert VLAN tag\n");
2671                         return;
2672                 }
2673         }
2674         arp_xmit(skb);
2675 }
2676
2677
2678 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2679 {
2680         int i, vlan_id;
2681         __be32 *targets = bond->params.arp_targets;
2682         struct vlan_entry *vlan;
2683         struct net_device *vlan_dev;
2684         struct rtable *rt;
2685
2686         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2687                 if (!targets[i])
2688                         break;
2689                 pr_debug("basa: target %x\n", targets[i]);
2690                 if (!bond->vlgrp) {
2691                         pr_debug("basa: empty vlan: arp_send\n");
2692                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2693                                       bond->master_ip, 0);
2694                         continue;
2695                 }
2696
2697                 /*
2698                  * If VLANs are configured, we do a route lookup to
2699                  * determine which VLAN interface would be used, so we
2700                  * can tag the ARP with the proper VLAN tag.
2701                  */
2702                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2703                                      RTO_ONLINK, 0);
2704                 if (IS_ERR(rt)) {
2705                         if (net_ratelimit()) {
2706                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2707                                            bond->dev->name, &targets[i]);
2708                         }
2709                         continue;
2710                 }
2711
2712                 /*
2713                  * This target is not on a VLAN
2714                  */
2715                 if (rt->dst.dev == bond->dev) {
2716                         ip_rt_put(rt);
2717                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2718                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2719                                       bond->master_ip, 0);
2720                         continue;
2721                 }
2722
2723                 vlan_id = 0;
2724                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2725                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2726                         if (vlan_dev == rt->dst.dev) {
2727                                 vlan_id = vlan->vlan_id;
2728                                 pr_debug("basa: vlan match on %s %d\n",
2729                                        vlan_dev->name, vlan_id);
2730                                 break;
2731                         }
2732                 }
2733
2734                 if (vlan_id) {
2735                         ip_rt_put(rt);
2736                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2737                                       vlan->vlan_ip, vlan_id);
2738                         continue;
2739                 }
2740
2741                 if (net_ratelimit()) {
2742                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2743                                    bond->dev->name, &targets[i],
2744                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2745                 }
2746                 ip_rt_put(rt);
2747         }
2748 }
2749
2750 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2751 {
2752         int i;
2753         __be32 *targets = bond->params.arp_targets;
2754
2755         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2756                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2757                          &sip, &tip, i, &targets[i],
2758                          bond_has_this_ip(bond, tip));
2759                 if (sip == targets[i]) {
2760                         if (bond_has_this_ip(bond, tip))
2761                                 slave->last_arp_rx = jiffies;
2762                         return;
2763                 }
2764         }
2765 }
2766
2767 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2768                          struct slave *slave)
2769 {
2770         struct arphdr *arp;
2771         unsigned char *arp_ptr;
2772         __be32 sip, tip;
2773
2774         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2775                 return;
2776
2777         read_lock(&bond->lock);
2778
2779         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2780                  bond->dev->name, skb->dev->name);
2781
2782         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2783                 goto out_unlock;
2784
2785         arp = arp_hdr(skb);
2786         if (arp->ar_hln != bond->dev->addr_len ||
2787             skb->pkt_type == PACKET_OTHERHOST ||
2788             skb->pkt_type == PACKET_LOOPBACK ||
2789             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2790             arp->ar_pro != htons(ETH_P_IP) ||
2791             arp->ar_pln != 4)
2792                 goto out_unlock;
2793
2794         arp_ptr = (unsigned char *)(arp + 1);
2795         arp_ptr += bond->dev->addr_len;
2796         memcpy(&sip, arp_ptr, 4);
2797         arp_ptr += 4 + bond->dev->addr_len;
2798         memcpy(&tip, arp_ptr, 4);
2799
2800         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2801                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2802                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2803                  &sip, &tip);
2804
2805         /*
2806          * Backup slaves won't see the ARP reply, but do come through
2807          * here for each ARP probe (so we swap the sip/tip to validate
2808          * the probe).  In a "redundant switch, common router" type of
2809          * configuration, the ARP probe will (hopefully) travel from
2810          * the active, through one switch, the router, then the other
2811          * switch before reaching the backup.
2812          */
2813         if (bond_is_active_slave(slave))
2814                 bond_validate_arp(bond, slave, sip, tip);
2815         else
2816                 bond_validate_arp(bond, slave, tip, sip);
2817
2818 out_unlock:
2819         read_unlock(&bond->lock);
2820 }
2821
2822 /*
2823  * this function is called regularly to monitor each slave's link
2824  * ensuring that traffic is being sent and received when arp monitoring
2825  * is used in load-balancing mode. if the adapter has been dormant, then an
2826  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2827  * arp monitoring in active backup mode.
2828  */
2829 void bond_loadbalance_arp_mon(struct work_struct *work)
2830 {
2831         struct bonding *bond = container_of(work, struct bonding,
2832                                             arp_work.work);
2833         struct slave *slave, *oldcurrent;
2834         int do_failover = 0;
2835         int delta_in_ticks;
2836         int i;
2837
2838         read_lock(&bond->lock);
2839
2840         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2841
2842         if (bond->kill_timers)
2843                 goto out;
2844
2845         if (bond->slave_cnt == 0)
2846                 goto re_arm;
2847
2848         read_lock(&bond->curr_slave_lock);
2849         oldcurrent = bond->curr_active_slave;
2850         read_unlock(&bond->curr_slave_lock);
2851
2852         /* see if any of the previous devices are up now (i.e. they have
2853          * xmt and rcv traffic). the curr_active_slave does not come into
2854          * the picture unless it is null. also, slave->jiffies is not needed
2855          * here because we send an arp on each slave and give a slave as
2856          * long as it needs to get the tx/rx within the delta.
2857          * TODO: what about up/down delay in arp mode? it wasn't here before
2858          *       so it can wait
2859          */
2860         bond_for_each_slave(bond, slave, i) {
2861                 unsigned long trans_start = dev_trans_start(slave->dev);
2862
2863                 if (slave->link != BOND_LINK_UP) {
2864                         if (time_in_range(jiffies,
2865                                 trans_start - delta_in_ticks,
2866                                 trans_start + delta_in_ticks) &&
2867                             time_in_range(jiffies,
2868                                 slave->dev->last_rx - delta_in_ticks,
2869                                 slave->dev->last_rx + delta_in_ticks)) {
2870
2871                                 slave->link  = BOND_LINK_UP;
2872                                 bond_set_active_slave(slave);
2873
2874                                 /* primary_slave has no meaning in round-robin
2875                                  * mode. the window of a slave being up and
2876                                  * curr_active_slave being null after enslaving
2877                                  * is closed.
2878                                  */
2879                                 if (!oldcurrent) {
2880                                         pr_info("%s: link status definitely up for interface %s, ",
2881                                                 bond->dev->name,
2882                                                 slave->dev->name);
2883                                         do_failover = 1;
2884                                 } else {
2885                                         pr_info("%s: interface %s is now up\n",
2886                                                 bond->dev->name,
2887                                                 slave->dev->name);
2888                                 }
2889                         }
2890                 } else {
2891                         /* slave->link == BOND_LINK_UP */
2892
2893                         /* not all switches will respond to an arp request
2894                          * when the source ip is 0, so don't take the link down
2895                          * if we don't know our ip yet
2896                          */
2897                         if (!time_in_range(jiffies,
2898                                 trans_start - delta_in_ticks,
2899                                 trans_start + 2 * delta_in_ticks) ||
2900                             !time_in_range(jiffies,
2901                                 slave->dev->last_rx - delta_in_ticks,
2902                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2903
2904                                 slave->link  = BOND_LINK_DOWN;
2905                                 bond_set_backup_slave(slave);
2906
2907                                 if (slave->link_failure_count < UINT_MAX)
2908                                         slave->link_failure_count++;
2909
2910                                 pr_info("%s: interface %s is now down.\n",
2911                                         bond->dev->name,
2912                                         slave->dev->name);
2913
2914                                 if (slave == oldcurrent)
2915                                         do_failover = 1;
2916                         }
2917                 }
2918
2919                 /* note: if switch is in round-robin mode, all links
2920                  * must tx arp to ensure all links rx an arp - otherwise
2921                  * links may oscillate or not come up at all; if switch is
2922                  * in something like xor mode, there is nothing we can
2923                  * do - all replies will be rx'ed on same link causing slaves
2924                  * to be unstable during low/no traffic periods
2925                  */
2926                 if (IS_UP(slave->dev))
2927                         bond_arp_send_all(bond, slave);
2928         }
2929
2930         if (do_failover) {
2931                 block_netpoll_tx();
2932                 write_lock_bh(&bond->curr_slave_lock);
2933
2934                 bond_select_active_slave(bond);
2935
2936                 write_unlock_bh(&bond->curr_slave_lock);
2937                 unblock_netpoll_tx();
2938         }
2939
2940 re_arm:
2941         if (bond->params.arp_interval)
2942                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2943 out:
2944         read_unlock(&bond->lock);
2945 }
2946
2947 /*
2948  * Called to inspect slaves for active-backup mode ARP monitor link state
2949  * changes.  Sets new_link in slaves to specify what action should take
2950  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2951  * to link states must be committed.
2952  *
2953  * Called with bond->lock held for read.
2954  */
2955 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2956 {
2957         struct slave *slave;
2958         int i, commit = 0;
2959         unsigned long trans_start;
2960
2961         bond_for_each_slave(bond, slave, i) {
2962                 slave->new_link = BOND_LINK_NOCHANGE;
2963
2964                 if (slave->link != BOND_LINK_UP) {
2965                         if (time_in_range(jiffies,
2966                                 slave_last_rx(bond, slave) - delta_in_ticks,
2967                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2968
2969                                 slave->new_link = BOND_LINK_UP;
2970                                 commit++;
2971                         }
2972
2973                         continue;
2974                 }
2975
2976                 /*
2977                  * Give slaves 2*delta after being enslaved or made
2978                  * active.  This avoids bouncing, as the last receive
2979                  * times need a full ARP monitor cycle to be updated.
2980                  */
2981                 if (time_in_range(jiffies,
2982                                   slave->jiffies - delta_in_ticks,
2983                                   slave->jiffies + 2 * delta_in_ticks))
2984                         continue;
2985
2986                 /*
2987                  * Backup slave is down if:
2988                  * - No current_arp_slave AND
2989                  * - more than 3*delta since last receive AND
2990                  * - the bond has an IP address
2991                  *
2992                  * Note: a non-null current_arp_slave indicates
2993                  * the curr_active_slave went down and we are
2994                  * searching for a new one; under this condition
2995                  * we only take the curr_active_slave down - this
2996                  * gives each slave a chance to tx/rx traffic
2997                  * before being taken out
2998                  */
2999                 if (!bond_is_active_slave(slave) &&
3000                     !bond->current_arp_slave &&
3001                     !time_in_range(jiffies,
3002                         slave_last_rx(bond, slave) - delta_in_ticks,
3003                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3004
3005                         slave->new_link = BOND_LINK_DOWN;
3006                         commit++;
3007                 }
3008
3009                 /*
3010                  * Active slave is down if:
3011                  * - more than 2*delta since transmitting OR
3012                  * - (more than 2*delta since receive AND
3013                  *    the bond has an IP address)
3014                  */
3015                 trans_start = dev_trans_start(slave->dev);
3016                 if (bond_is_active_slave(slave) &&
3017                     (!time_in_range(jiffies,
3018                         trans_start - delta_in_ticks,
3019                         trans_start + 2 * delta_in_ticks) ||
3020                      !time_in_range(jiffies,
3021                         slave_last_rx(bond, slave) - delta_in_ticks,
3022                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3023
3024                         slave->new_link = BOND_LINK_DOWN;
3025                         commit++;
3026                 }
3027         }
3028
3029         return commit;
3030 }
3031
3032 /*
3033  * Called to commit link state changes noted by inspection step of
3034  * active-backup mode ARP monitor.
3035  *
3036  * Called with RTNL and bond->lock for read.
3037  */
3038 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3039 {
3040         struct slave *slave;
3041         int i;
3042         unsigned long trans_start;
3043
3044         bond_for_each_slave(bond, slave, i) {
3045                 switch (slave->new_link) {
3046                 case BOND_LINK_NOCHANGE:
3047                         continue;
3048
3049                 case BOND_LINK_UP:
3050                         trans_start = dev_trans_start(slave->dev);
3051                         if ((!bond->curr_active_slave &&
3052                              time_in_range(jiffies,
3053                                            trans_start - delta_in_ticks,
3054                                            trans_start + delta_in_ticks)) ||
3055                             bond->curr_active_slave != slave) {
3056                                 slave->link = BOND_LINK_UP;
3057                                 bond->current_arp_slave = NULL;
3058
3059                                 pr_info("%s: link status definitely up for interface %s.\n",
3060                                         bond->dev->name, slave->dev->name);
3061
3062                                 if (!bond->curr_active_slave ||
3063                                     (slave == bond->primary_slave))
3064                                         goto do_failover;
3065
3066                         }
3067
3068                         continue;
3069
3070                 case BOND_LINK_DOWN:
3071                         if (slave->link_failure_count < UINT_MAX)
3072                                 slave->link_failure_count++;
3073
3074                         slave->link = BOND_LINK_DOWN;
3075                         bond_set_slave_inactive_flags(slave);
3076
3077                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3078                                 bond->dev->name, slave->dev->name);
3079
3080                         if (slave == bond->curr_active_slave) {
3081                                 bond->current_arp_slave = NULL;
3082                                 goto do_failover;
3083                         }
3084
3085                         continue;
3086
3087                 default:
3088                         pr_err("%s: impossible: new_link %d on slave %s\n",
3089                                bond->dev->name, slave->new_link,
3090                                slave->dev->name);
3091                         continue;
3092                 }
3093
3094 do_failover:
3095                 ASSERT_RTNL();
3096                 block_netpoll_tx();
3097                 write_lock_bh(&bond->curr_slave_lock);
3098                 bond_select_active_slave(bond);
3099                 write_unlock_bh(&bond->curr_slave_lock);
3100                 unblock_netpoll_tx();
3101         }
3102
3103         bond_set_carrier(bond);
3104 }
3105
3106 /*
3107  * Send ARP probes for active-backup mode ARP monitor.
3108  *
3109  * Called with bond->lock held for read.
3110  */
3111 static void bond_ab_arp_probe(struct bonding *bond)
3112 {
3113         struct slave *slave;
3114         int i;
3115
3116         read_lock(&bond->curr_slave_lock);
3117
3118         if (bond->current_arp_slave && bond->curr_active_slave)
3119                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3120                         bond->current_arp_slave->dev->name,
3121                         bond->curr_active_slave->dev->name);
3122
3123         if (bond->curr_active_slave) {
3124                 bond_arp_send_all(bond, bond->curr_active_slave);
3125                 read_unlock(&bond->curr_slave_lock);
3126                 return;
3127         }
3128
3129         read_unlock(&bond->curr_slave_lock);
3130
3131         /* if we don't have a curr_active_slave, search for the next available
3132          * backup slave from the current_arp_slave and make it the candidate
3133          * for becoming the curr_active_slave
3134          */
3135
3136         if (!bond->current_arp_slave) {
3137                 bond->current_arp_slave = bond->first_slave;
3138                 if (!bond->current_arp_slave)
3139                         return;
3140         }
3141
3142         bond_set_slave_inactive_flags(bond->current_arp_slave);
3143
3144         /* search for next candidate */
3145         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3146                 if (IS_UP(slave->dev)) {
3147                         slave->link = BOND_LINK_BACK;
3148                         bond_set_slave_active_flags(slave);
3149                         bond_arp_send_all(bond, slave);
3150                         slave->jiffies = jiffies;
3151                         bond->current_arp_slave = slave;
3152                         break;
3153                 }
3154
3155                 /* if the link state is up at this point, we
3156                  * mark it down - this can happen if we have
3157                  * simultaneous link failures and
3158                  * reselect_active_interface doesn't make this
3159                  * one the current slave so it is still marked
3160                  * up when it is actually down
3161                  */
3162                 if (slave->link == BOND_LINK_UP) {
3163                         slave->link = BOND_LINK_DOWN;
3164                         if (slave->link_failure_count < UINT_MAX)
3165                                 slave->link_failure_count++;
3166
3167                         bond_set_slave_inactive_flags(slave);
3168
3169                         pr_info("%s: backup interface %s is now down.\n",
3170                                 bond->dev->name, slave->dev->name);
3171                 }
3172         }
3173 }
3174
3175 void bond_activebackup_arp_mon(struct work_struct *work)
3176 {
3177         struct bonding *bond = container_of(work, struct bonding,
3178                                             arp_work.work);
3179         bool should_notify_peers = false;
3180         int delta_in_ticks;
3181
3182         read_lock(&bond->lock);
3183
3184         if (bond->kill_timers)
3185                 goto out;
3186
3187         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3188
3189         if (bond->slave_cnt == 0)
3190                 goto re_arm;
3191
3192         should_notify_peers = bond_should_notify_peers(bond);
3193
3194         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3195                 read_unlock(&bond->lock);
3196                 rtnl_lock();
3197                 read_lock(&bond->lock);
3198
3199                 bond_ab_arp_commit(bond, delta_in_ticks);
3200
3201                 read_unlock(&bond->lock);
3202                 rtnl_unlock();
3203                 read_lock(&bond->lock);
3204         }
3205
3206         bond_ab_arp_probe(bond);
3207
3208 re_arm:
3209         if (bond->params.arp_interval)
3210                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3211 out:
3212         read_unlock(&bond->lock);
3213
3214         if (should_notify_peers) {
3215                 rtnl_lock();
3216                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3217                 rtnl_unlock();
3218         }
3219 }
3220
3221 /*-------------------------- netdev event handling --------------------------*/
3222
3223 /*
3224  * Change device name
3225  */
3226 static int bond_event_changename(struct bonding *bond)
3227 {
3228         bond_remove_proc_entry(bond);
3229         bond_create_proc_entry(bond);
3230
3231         bond_debug_reregister(bond);
3232
3233         return NOTIFY_DONE;
3234 }
3235
3236 static int bond_master_netdev_event(unsigned long event,
3237                                     struct net_device *bond_dev)
3238 {
3239         struct bonding *event_bond = netdev_priv(bond_dev);
3240
3241         switch (event) {
3242         case NETDEV_CHANGENAME:
3243                 return bond_event_changename(event_bond);
3244         default:
3245                 break;
3246         }
3247
3248         return NOTIFY_DONE;
3249 }
3250
3251 static int bond_slave_netdev_event(unsigned long event,
3252                                    struct net_device *slave_dev)
3253 {
3254         struct net_device *bond_dev = slave_dev->master;
3255         struct bonding *bond = netdev_priv(bond_dev);
3256
3257         switch (event) {
3258         case NETDEV_UNREGISTER:
3259                 if (bond_dev) {
3260                         if (bond->setup_by_slave)
3261                                 bond_release_and_destroy(bond_dev, slave_dev);
3262                         else
3263                                 bond_release(bond_dev, slave_dev);
3264                 }
3265                 break;
3266         case NETDEV_CHANGE:
3267                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3268                         struct slave *slave;
3269
3270                         slave = bond_get_slave_by_dev(bond, slave_dev);
3271                         if (slave) {
3272                                 u32 old_speed = slave->speed;
3273                                 u8  old_duplex = slave->duplex;
3274
3275                                 bond_update_speed_duplex(slave);
3276
3277                                 if (bond_is_lb(bond))
3278                                         break;
3279
3280                                 if (old_speed != slave->speed)
3281                                         bond_3ad_adapter_speed_changed(slave);
3282                                 if (old_duplex != slave->duplex)
3283                                         bond_3ad_adapter_duplex_changed(slave);
3284                         }
3285                 }
3286
3287                 break;
3288         case NETDEV_DOWN:
3289                 /*
3290                  * ... Or is it this?
3291                  */
3292                 break;
3293         case NETDEV_CHANGEMTU:
3294                 /*
3295                  * TODO: Should slaves be allowed to
3296                  * independently alter their MTU?  For
3297                  * an active-backup bond, slaves need
3298                  * not be the same type of device, so
3299                  * MTUs may vary.  For other modes,
3300                  * slaves arguably should have the
3301                  * same MTUs. To do this, we'd need to
3302                  * take over the slave's change_mtu
3303                  * function for the duration of their
3304                  * servitude.
3305                  */
3306                 break;
3307         case NETDEV_CHANGENAME:
3308                 /*
3309                  * TODO: handle changing the primary's name
3310                  */
3311                 break;
3312         case NETDEV_FEAT_CHANGE:
3313                 bond_compute_features(bond);
3314                 break;
3315         default:
3316                 break;
3317         }
3318
3319         return NOTIFY_DONE;
3320 }
3321
3322 /*
3323  * bond_netdev_event: handle netdev notifier chain events.
3324  *
3325  * This function receives events for the netdev chain.  The caller (an
3326  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3327  * locks for us to safely manipulate the slave devices (RTNL lock,
3328  * dev_probe_lock).
3329  */
3330 static int bond_netdev_event(struct notifier_block *this,
3331                              unsigned long event, void *ptr)
3332 {
3333         struct net_device *event_dev = (struct net_device *)ptr;
3334
3335         pr_debug("event_dev: %s, event: %lx\n",
3336                  event_dev ? event_dev->name : "None",
3337                  event);
3338
3339         if (!(event_dev->priv_flags & IFF_BONDING))
3340                 return NOTIFY_DONE;
3341
3342         if (event_dev->flags & IFF_MASTER) {
3343                 pr_debug("IFF_MASTER\n");
3344                 return bond_master_netdev_event(event, event_dev);
3345         }
3346
3347         if (event_dev->flags & IFF_SLAVE) {
3348                 pr_debug("IFF_SLAVE\n");
3349                 return bond_slave_netdev_event(event, event_dev);
3350         }
3351
3352         return NOTIFY_DONE;
3353 }
3354
3355 /*
3356  * bond_inetaddr_event: handle inetaddr notifier chain events.
3357  *
3358  * We keep track of device IPs primarily to use as source addresses in
3359  * ARP monitor probes (rather than spewing out broadcasts all the time).
3360  *
3361  * We track one IP for the main device (if it has one), plus one per VLAN.
3362  */
3363 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3364 {
3365         struct in_ifaddr *ifa = ptr;
3366         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3367         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3368         struct bonding *bond;
3369         struct vlan_entry *vlan;
3370
3371         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3372                 if (bond->dev == event_dev) {
3373                         switch (event) {
3374                         case NETDEV_UP:
3375                                 bond->master_ip = ifa->ifa_local;
3376                                 return NOTIFY_OK;
3377                         case NETDEV_DOWN:
3378                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3379                                 return NOTIFY_OK;
3380                         default:
3381                                 return NOTIFY_DONE;
3382                         }
3383                 }
3384
3385                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3386                         if (!bond->vlgrp)
3387                                 continue;
3388                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3389                         if (vlan_dev == event_dev) {
3390                                 switch (event) {
3391                                 case NETDEV_UP:
3392                                         vlan->vlan_ip = ifa->ifa_local;
3393                                         return NOTIFY_OK;
3394                                 case NETDEV_DOWN:
3395                                         vlan->vlan_ip =
3396                                                 bond_glean_dev_ip(vlan_dev);
3397                                         return NOTIFY_OK;
3398                                 default:
3399                                         return NOTIFY_DONE;
3400                                 }
3401                         }
3402                 }
3403         }
3404         return NOTIFY_DONE;
3405 }
3406
3407 static struct notifier_block bond_netdev_notifier = {
3408         .notifier_call = bond_netdev_event,
3409 };
3410
3411 static struct notifier_block bond_inetaddr_notifier = {
3412         .notifier_call = bond_inetaddr_event,
3413 };
3414
3415 /*---------------------------- Hashing Policies -----------------------------*/
3416
3417 /*
3418  * Hash for the output device based upon layer 2 and layer 3 data. If
3419  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3420  */
3421 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3422 {
3423         struct ethhdr *data = (struct ethhdr *)skb->data;
3424         struct iphdr *iph = ip_hdr(skb);
3425
3426         if (skb->protocol == htons(ETH_P_IP)) {
3427                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3428                         (data->h_dest[5] ^ data->h_source[5])) % count;
3429         }
3430
3431         return (data->h_dest[5] ^ data->h_source[5]) % count;
3432 }
3433
3434 /*
3435  * Hash for the output device based upon layer 3 and layer 4 data. If
3436  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3437  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3438  */
3439 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3440 {
3441         struct ethhdr *data = (struct ethhdr *)skb->data;
3442         struct iphdr *iph = ip_hdr(skb);
3443         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3444         int layer4_xor = 0;
3445
3446         if (skb->protocol == htons(ETH_P_IP)) {
3447                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3448                     (iph->protocol == IPPROTO_TCP ||
3449                      iph->protocol == IPPROTO_UDP)) {
3450                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3451                 }
3452                 return (layer4_xor ^
3453                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3454
3455         }
3456
3457         return (data->h_dest[5] ^ data->h_source[5]) % count;
3458 }
3459
3460 /*
3461  * Hash for the output device based upon layer 2 data
3462  */
3463 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3464 {
3465         struct ethhdr *data = (struct ethhdr *)skb->data;
3466
3467         return (data->h_dest[5] ^ data->h_source[5]) % count;
3468 }
3469
3470 /*-------------------------- Device entry points ----------------------------*/
3471
3472 static int bond_open(struct net_device *bond_dev)
3473 {
3474         struct bonding *bond = netdev_priv(bond_dev);
3475
3476         bond->kill_timers = 0;
3477
3478         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3479
3480         if (bond_is_lb(bond)) {
3481                 /* bond_alb_initialize must be called before the timer
3482                  * is started.
3483                  */
3484                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3485                         /* something went wrong - fail the open operation */
3486                         return -ENOMEM;
3487                 }
3488
3489                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3490                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3491         }
3492
3493         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3494                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3495                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3496         }
3497
3498         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3499                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3500                         INIT_DELAYED_WORK(&bond->arp_work,
3501                                           bond_activebackup_arp_mon);
3502                 else
3503                         INIT_DELAYED_WORK(&bond->arp_work,
3504                                           bond_loadbalance_arp_mon);
3505
3506                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3507                 if (bond->params.arp_validate)
3508                         bond->recv_probe = bond_arp_rcv;
3509         }
3510
3511         if (bond->params.mode == BOND_MODE_8023AD) {
3512                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3513                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3514                 /* register to receive LACPDUs */
3515                 bond->recv_probe = bond_3ad_lacpdu_recv;
3516                 bond_3ad_initiate_agg_selection(bond, 1);
3517         }
3518
3519         return 0;
3520 }
3521
3522 static int bond_close(struct net_device *bond_dev)
3523 {
3524         struct bonding *bond = netdev_priv(bond_dev);
3525
3526         write_lock_bh(&bond->lock);
3527
3528         bond->send_peer_notif = 0;
3529
3530         /* signal timers not to re-arm */
3531         bond->kill_timers = 1;
3532
3533         write_unlock_bh(&bond->lock);
3534
3535         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3536                 cancel_delayed_work(&bond->mii_work);
3537         }
3538
3539         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3540                 cancel_delayed_work(&bond->arp_work);
3541         }
3542
3543         switch (bond->params.mode) {
3544         case BOND_MODE_8023AD:
3545                 cancel_delayed_work(&bond->ad_work);
3546                 break;
3547         case BOND_MODE_TLB:
3548         case BOND_MODE_ALB:
3549                 cancel_delayed_work(&bond->alb_work);
3550                 break;
3551         default:
3552                 break;
3553         }
3554
3555         if (delayed_work_pending(&bond->mcast_work))
3556                 cancel_delayed_work(&bond->mcast_work);
3557
3558         if (bond_is_lb(bond)) {
3559                 /* Must be called only after all
3560                  * slaves have been released
3561                  */
3562                 bond_alb_deinitialize(bond);
3563         }
3564         bond->recv_probe = NULL;
3565
3566         return 0;
3567 }
3568
3569 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3570                                                 struct rtnl_link_stats64 *stats)
3571 {
3572         struct bonding *bond = netdev_priv(bond_dev);
3573         struct rtnl_link_stats64 temp;
3574         struct slave *slave;
3575         int i;
3576
3577         memset(stats, 0, sizeof(*stats));
3578
3579         read_lock_bh(&bond->lock);
3580
3581         bond_for_each_slave(bond, slave, i) {
3582                 const struct rtnl_link_stats64 *sstats =
3583                         dev_get_stats(slave->dev, &temp);
3584
3585                 stats->rx_packets += sstats->rx_packets;
3586                 stats->rx_bytes += sstats->rx_bytes;
3587                 stats->rx_errors += sstats->rx_errors;
3588                 stats->rx_dropped += sstats->rx_dropped;
3589
3590                 stats->tx_packets += sstats->tx_packets;
3591                 stats->tx_bytes += sstats->tx_bytes;
3592                 stats->tx_errors += sstats->tx_errors;
3593                 stats->tx_dropped += sstats->tx_dropped;
3594
3595                 stats->multicast += sstats->multicast;
3596                 stats->collisions += sstats->collisions;
3597
3598                 stats->rx_length_errors += sstats->rx_length_errors;
3599                 stats->rx_over_errors += sstats->rx_over_errors;
3600                 stats->rx_crc_errors += sstats->rx_crc_errors;
3601                 stats->rx_frame_errors += sstats->rx_frame_errors;
3602                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3603                 stats->rx_missed_errors += sstats->rx_missed_errors;
3604
3605                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3606                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3607                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3608                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3609                 stats->tx_window_errors += sstats->tx_window_errors;
3610         }
3611
3612         read_unlock_bh(&bond->lock);
3613
3614         return stats;
3615 }
3616
3617 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3618 {
3619         struct net_device *slave_dev = NULL;
3620         struct ifbond k_binfo;
3621         struct ifbond __user *u_binfo = NULL;
3622         struct ifslave k_sinfo;
3623         struct ifslave __user *u_sinfo = NULL;
3624         struct mii_ioctl_data *mii = NULL;
3625         int res = 0;
3626
3627         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3628
3629         switch (cmd) {
3630         case SIOCGMIIPHY:
3631                 mii = if_mii(ifr);
3632                 if (!mii)
3633                         return -EINVAL;
3634
3635                 mii->phy_id = 0;
3636                 /* Fall Through */
3637         case SIOCGMIIREG:
3638                 /*
3639                  * We do this again just in case we were called by SIOCGMIIREG
3640                  * instead of SIOCGMIIPHY.
3641                  */
3642                 mii = if_mii(ifr);
3643                 if (!mii)
3644                         return -EINVAL;
3645
3646
3647                 if (mii->reg_num == 1) {
3648                         struct bonding *bond = netdev_priv(bond_dev);
3649                         mii->val_out = 0;
3650                         read_lock(&bond->lock);
3651                         read_lock(&bond->curr_slave_lock);
3652                         if (netif_carrier_ok(bond->dev))
3653                                 mii->val_out = BMSR_LSTATUS;
3654
3655                         read_unlock(&bond->curr_slave_lock);
3656                         read_unlock(&bond->lock);
3657                 }
3658
3659                 return 0;
3660         case BOND_INFO_QUERY_OLD:
3661         case SIOCBONDINFOQUERY:
3662                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3663
3664                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3665                         return -EFAULT;
3666
3667                 res = bond_info_query(bond_dev, &k_binfo);
3668                 if (res == 0 &&
3669                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3670                         return -EFAULT;
3671
3672                 return res;
3673         case BOND_SLAVE_INFO_QUERY_OLD:
3674         case SIOCBONDSLAVEINFOQUERY:
3675                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3676
3677                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3678                         return -EFAULT;
3679
3680                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3681                 if (res == 0 &&
3682                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3683                         return -EFAULT;
3684
3685                 return res;
3686         default:
3687                 /* Go on */
3688                 break;
3689         }
3690
3691         if (!capable(CAP_NET_ADMIN))
3692                 return -EPERM;
3693
3694         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3695
3696         pr_debug("slave_dev=%p:\n", slave_dev);
3697
3698         if (!slave_dev)
3699                 res = -ENODEV;
3700         else {
3701                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3702                 switch (cmd) {
3703                 case BOND_ENSLAVE_OLD:
3704                 case SIOCBONDENSLAVE:
3705                         res = bond_enslave(bond_dev, slave_dev);
3706                         break;
3707                 case BOND_RELEASE_OLD:
3708                 case SIOCBONDRELEASE:
3709                         res = bond_release(bond_dev, slave_dev);
3710                         break;
3711                 case BOND_SETHWADDR_OLD:
3712                 case SIOCBONDSETHWADDR:
3713                         res = bond_sethwaddr(bond_dev, slave_dev);
3714                         break;
3715                 case BOND_CHANGE_ACTIVE_OLD:
3716                 case SIOCBONDCHANGEACTIVE:
3717                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3718                         break;
3719                 default:
3720                         res = -EOPNOTSUPP;
3721                 }
3722
3723                 dev_put(slave_dev);
3724         }
3725
3726         return res;
3727 }
3728
3729 static bool bond_addr_in_mc_list(unsigned char *addr,
3730                                  struct netdev_hw_addr_list *list,
3731                                  int addrlen)
3732 {
3733         struct netdev_hw_addr *ha;
3734
3735         netdev_hw_addr_list_for_each(ha, list)
3736                 if (!memcmp(ha->addr, addr, addrlen))
3737                         return true;
3738
3739         return false;
3740 }
3741
3742 static void bond_set_multicast_list(struct net_device *bond_dev)
3743 {
3744         struct bonding *bond = netdev_priv(bond_dev);
3745         struct netdev_hw_addr *ha;
3746         bool found;
3747
3748         /*
3749          * Do promisc before checking multicast_mode
3750          */
3751         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3752                 /*
3753                  * FIXME: Need to handle the error when one of the multi-slaves
3754                  * encounters error.
3755                  */
3756                 bond_set_promiscuity(bond, 1);
3757
3758
3759         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3760                 bond_set_promiscuity(bond, -1);
3761
3762
3763         /* set allmulti flag to slaves */
3764         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3765                 /*
3766                  * FIXME: Need to handle the error when one of the multi-slaves
3767                  * encounters error.
3768                  */
3769                 bond_set_allmulti(bond, 1);
3770
3771
3772         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3773                 bond_set_allmulti(bond, -1);
3774
3775
3776         read_lock(&bond->lock);
3777
3778         bond->flags = bond_dev->flags;
3779
3780         /* looking for addresses to add to slaves' mc list */
3781         netdev_for_each_mc_addr(ha, bond_dev) {
3782                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3783                                              bond_dev->addr_len);
3784                 if (!found)
3785                         bond_mc_add(bond, ha->addr);
3786         }
3787
3788         /* looking for addresses to delete from slaves' list */
3789         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3790                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3791                                              bond_dev->addr_len);
3792                 if (!found)
3793                         bond_mc_del(bond, ha->addr);
3794         }
3795
3796         /* save master's multicast list */
3797         __hw_addr_flush(&bond->mc_list);
3798         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3799                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3800
3801         read_unlock(&bond->lock);
3802 }
3803
3804 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3805 {
3806         struct bonding *bond = netdev_priv(dev);
3807         struct slave *slave = bond->first_slave;
3808
3809         if (slave) {
3810                 const struct net_device_ops *slave_ops
3811                         = slave->dev->netdev_ops;
3812                 if (slave_ops->ndo_neigh_setup)
3813                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3814         }
3815         return 0;
3816 }
3817
3818 /*
3819  * Change the MTU of all of a master's slaves to match the master
3820  */
3821 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3822 {
3823         struct bonding *bond = netdev_priv(bond_dev);
3824         struct slave *slave, *stop_at;
3825         int res = 0;
3826         int i;
3827
3828         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3829                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3830
3831         /* Can't hold bond->lock with bh disabled here since
3832          * some base drivers panic. On the other hand we can't
3833          * hold bond->lock without bh disabled because we'll
3834          * deadlock. The only solution is to rely on the fact
3835          * that we're under rtnl_lock here, and the slaves
3836          * list won't change. This doesn't solve the problem
3837          * of setting the slave's MTU while it is
3838          * transmitting, but the assumption is that the base
3839          * driver can handle that.
3840          *
3841          * TODO: figure out a way to safely iterate the slaves
3842          * list, but without holding a lock around the actual
3843          * call to the base driver.
3844          */
3845
3846         bond_for_each_slave(bond, slave, i) {
3847                 pr_debug("s %p s->p %p c_m %p\n",
3848                          slave,
3849                          slave->prev,
3850                          slave->dev->netdev_ops->ndo_change_mtu);
3851
3852                 res = dev_set_mtu(slave->dev, new_mtu);
3853
3854                 if (res) {
3855                         /* If we failed to set the slave's mtu to the new value
3856                          * we must abort the operation even in ACTIVE_BACKUP
3857                          * mode, because if we allow the backup slaves to have
3858                          * different mtu values than the active slave we'll
3859                          * need to change their mtu when doing a failover. That
3860                          * means changing their mtu from timer context, which
3861                          * is probably not a good idea.
3862                          */
3863                         pr_debug("err %d %s\n", res, slave->dev->name);
3864                         goto unwind;
3865                 }
3866         }
3867
3868         bond_dev->mtu = new_mtu;
3869
3870         return 0;
3871
3872 unwind:
3873         /* unwind from head to the slave that failed */
3874         stop_at = slave;
3875         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3876                 int tmp_res;
3877
3878                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3879                 if (tmp_res) {
3880                         pr_debug("unwind err %d dev %s\n",
3881                                  tmp_res, slave->dev->name);
3882                 }
3883         }
3884
3885         return res;
3886 }
3887
3888 /*
3889  * Change HW address
3890  *
3891  * Note that many devices must be down to change the HW address, and
3892  * downing the master releases all slaves.  We can make bonds full of
3893  * bonding devices to test this, however.
3894  */
3895 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3896 {
3897         struct bonding *bond = netdev_priv(bond_dev);
3898         struct sockaddr *sa = addr, tmp_sa;
3899         struct slave *slave, *stop_at;
3900         int res = 0;
3901         int i;
3902
3903         if (bond->params.mode == BOND_MODE_ALB)
3904                 return bond_alb_set_mac_address(bond_dev, addr);
3905
3906
3907         pr_debug("bond=%p, name=%s\n",
3908                  bond, bond_dev ? bond_dev->name : "None");
3909
3910         /*
3911          * If fail_over_mac is set to active, do nothing and return
3912          * success.  Returning an error causes ifenslave to fail.
3913          */
3914         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3915                 return 0;
3916
3917         if (!is_valid_ether_addr(sa->sa_data))
3918                 return -EADDRNOTAVAIL;
3919
3920         /* Can't hold bond->lock with bh disabled here since
3921          * some base drivers panic. On the other hand we can't
3922          * hold bond->lock without bh disabled because we'll
3923          * deadlock. The only solution is to rely on the fact
3924          * that we're under rtnl_lock here, and the slaves
3925          * list won't change. This doesn't solve the problem
3926          * of setting the slave's hw address while it is
3927          * transmitting, but the assumption is that the base
3928          * driver can handle that.
3929          *
3930          * TODO: figure out a way to safely iterate the slaves
3931          * list, but without holding a lock around the actual
3932          * call to the base driver.
3933          */
3934
3935         bond_for_each_slave(bond, slave, i) {
3936                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3937                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3938
3939                 if (slave_ops->ndo_set_mac_address == NULL) {
3940                         res = -EOPNOTSUPP;
3941                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3942                         goto unwind;
3943                 }
3944
3945                 res = dev_set_mac_address(slave->dev, addr);
3946                 if (res) {
3947                         /* TODO: consider downing the slave
3948                          * and retry ?
3949                          * User should expect communications
3950                          * breakage anyway until ARP finish
3951                          * updating, so...
3952                          */
3953                         pr_debug("err %d %s\n", res, slave->dev->name);
3954                         goto unwind;
3955                 }
3956         }
3957
3958         /* success */
3959         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3960         return 0;
3961
3962 unwind:
3963         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3964         tmp_sa.sa_family = bond_dev->type;
3965
3966         /* unwind from head to the slave that failed */
3967         stop_at = slave;
3968         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3969                 int tmp_res;
3970
3971                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3972                 if (tmp_res) {
3973                         pr_debug("unwind err %d dev %s\n",
3974                                  tmp_res, slave->dev->name);
3975                 }
3976         }
3977
3978         return res;
3979 }
3980
3981 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3982 {
3983         struct bonding *bond = netdev_priv(bond_dev);
3984         struct slave *slave, *start_at;
3985         int i, slave_no, res = 1;
3986         struct iphdr *iph = ip_hdr(skb);
3987
3988         /*
3989          * Start with the curr_active_slave that joined the bond as the
3990          * default for sending IGMP traffic.  For failover purposes one
3991          * needs to maintain some consistency for the interface that will
3992          * send the join/membership reports.  The curr_active_slave found
3993          * will send all of this type of traffic.
3994          */
3995         if ((iph->protocol == IPPROTO_IGMP) &&
3996             (skb->protocol == htons(ETH_P_IP))) {
3997
3998                 read_lock(&bond->curr_slave_lock);
3999                 slave = bond->curr_active_slave;
4000                 read_unlock(&bond->curr_slave_lock);
4001
4002                 if (!slave)
4003                         goto out;
4004         } else {
4005                 /*
4006                  * Concurrent TX may collide on rr_tx_counter; we accept
4007                  * that as being rare enough not to justify using an
4008                  * atomic op here.
4009                  */
4010                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4011
4012                 bond_for_each_slave(bond, slave, i) {
4013                         slave_no--;
4014                         if (slave_no < 0)
4015                                 break;
4016                 }
4017         }
4018
4019         start_at = slave;
4020         bond_for_each_slave_from(bond, slave, i, start_at) {
4021                 if (IS_UP(slave->dev) &&
4022                     (slave->link == BOND_LINK_UP) &&
4023                     bond_is_active_slave(slave)) {
4024                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4025                         break;
4026                 }
4027         }
4028
4029 out:
4030         if (res) {
4031                 /* no suitable interface, frame not sent */
4032                 dev_kfree_skb(skb);
4033         }
4034
4035         return NETDEV_TX_OK;
4036 }
4037
4038
4039 /*
4040  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4041  * the bond has a usable interface.
4042  */
4043 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4044 {
4045         struct bonding *bond = netdev_priv(bond_dev);
4046         int res = 1;
4047
4048         read_lock(&bond->curr_slave_lock);
4049
4050         if (bond->curr_active_slave)
4051                 res = bond_dev_queue_xmit(bond, skb,
4052                         bond->curr_active_slave->dev);
4053
4054         if (res)
4055                 /* no suitable interface, frame not sent */
4056                 dev_kfree_skb(skb);
4057
4058         read_unlock(&bond->curr_slave_lock);
4059
4060         return NETDEV_TX_OK;
4061 }
4062
4063 /*
4064  * In bond_xmit_xor() , we determine the output device by using a pre-
4065  * determined xmit_hash_policy(), If the selected device is not enabled,
4066  * find the next active slave.
4067  */
4068 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4069 {
4070         struct bonding *bond = netdev_priv(bond_dev);
4071         struct slave *slave, *start_at;
4072         int slave_no;
4073         int i;
4074         int res = 1;
4075
4076         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4077
4078         bond_for_each_slave(bond, slave, i) {
4079                 slave_no--;
4080                 if (slave_no < 0)
4081                         break;
4082         }
4083
4084         start_at = slave;
4085
4086         bond_for_each_slave_from(bond, slave, i, start_at) {
4087                 if (IS_UP(slave->dev) &&
4088                     (slave->link == BOND_LINK_UP) &&
4089                     bond_is_active_slave(slave)) {
4090                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4091                         break;
4092                 }
4093         }
4094
4095         if (res) {
4096                 /* no suitable interface, frame not sent */
4097                 dev_kfree_skb(skb);
4098         }
4099
4100         return NETDEV_TX_OK;
4101 }
4102
4103 /*
4104  * in broadcast mode, we send everything to all usable interfaces.
4105  */
4106 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4107 {
4108         struct bonding *bond = netdev_priv(bond_dev);
4109         struct slave *slave, *start_at;
4110         struct net_device *tx_dev = NULL;
4111         int i;
4112         int res = 1;
4113
4114         read_lock(&bond->curr_slave_lock);
4115         start_at = bond->curr_active_slave;
4116         read_unlock(&bond->curr_slave_lock);
4117
4118         if (!start_at)
4119                 goto out;
4120
4121         bond_for_each_slave_from(bond, slave, i, start_at) {
4122                 if (IS_UP(slave->dev) &&
4123                     (slave->link == BOND_LINK_UP) &&
4124                     bond_is_active_slave(slave)) {
4125                         if (tx_dev) {
4126                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4127                                 if (!skb2) {
4128                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4129                                                bond_dev->name);
4130                                         continue;
4131                                 }
4132
4133                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4134                                 if (res) {
4135                                         dev_kfree_skb(skb2);
4136                                         continue;
4137                                 }
4138                         }
4139                         tx_dev = slave->dev;
4140                 }
4141         }
4142
4143         if (tx_dev)
4144                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4145
4146 out:
4147         if (res)
4148                 /* no suitable interface, frame not sent */
4149                 dev_kfree_skb(skb);
4150
4151         /* frame sent to all suitable interfaces */
4152         return NETDEV_TX_OK;
4153 }
4154
4155 /*------------------------- Device initialization ---------------------------*/
4156
4157 static void bond_set_xmit_hash_policy(struct bonding *bond)
4158 {
4159         switch (bond->params.xmit_policy) {
4160         case BOND_XMIT_POLICY_LAYER23:
4161                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4162                 break;
4163         case BOND_XMIT_POLICY_LAYER34:
4164                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4165                 break;
4166         case BOND_XMIT_POLICY_LAYER2:
4167         default:
4168                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4169                 break;
4170         }
4171 }
4172
4173 /*
4174  * Lookup the slave that corresponds to a qid
4175  */
4176 static inline int bond_slave_override(struct bonding *bond,
4177                                       struct sk_buff *skb)
4178 {
4179         int i, res = 1;
4180         struct slave *slave = NULL;
4181         struct slave *check_slave;
4182
4183         if (!skb->queue_mapping)
4184                 return 1;
4185
4186         /* Find out if any slaves have the same mapping as this skb. */
4187         bond_for_each_slave(bond, check_slave, i) {
4188                 if (check_slave->queue_id == skb->queue_mapping) {
4189                         slave = check_slave;
4190                         break;
4191                 }
4192         }
4193
4194         /* If the slave isn't UP, use default transmit policy. */
4195         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4196             (slave->link == BOND_LINK_UP)) {
4197                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4198         }
4199
4200         return res;
4201 }
4202
4203 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4204 {
4205         /*
4206          * This helper function exists to help dev_pick_tx get the correct
4207          * destination queue.  Using a helper function skips a call to
4208          * skb_tx_hash and will put the skbs in the queue we expect on their
4209          * way down to the bonding driver.
4210          */
4211         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4212
4213         if (unlikely(txq >= dev->real_num_tx_queues)) {
4214                 do {
4215                         txq -= dev->real_num_tx_queues;
4216                 } while (txq >= dev->real_num_tx_queues);
4217         }
4218         return txq;
4219 }
4220
4221 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4222 {
4223         struct bonding *bond = netdev_priv(dev);
4224
4225         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4226                 if (!bond_slave_override(bond, skb))
4227                         return NETDEV_TX_OK;
4228         }
4229
4230         switch (bond->params.mode) {
4231         case BOND_MODE_ROUNDROBIN:
4232                 return bond_xmit_roundrobin(skb, dev);
4233         case BOND_MODE_ACTIVEBACKUP:
4234                 return bond_xmit_activebackup(skb, dev);
4235         case BOND_MODE_XOR:
4236                 return bond_xmit_xor(skb, dev);
4237         case BOND_MODE_BROADCAST:
4238                 return bond_xmit_broadcast(skb, dev);
4239         case BOND_MODE_8023AD:
4240                 return bond_3ad_xmit_xor(skb, dev);
4241         case BOND_MODE_ALB:
4242         case BOND_MODE_TLB:
4243                 return bond_alb_xmit(skb, dev);
4244         default:
4245                 /* Should never happen, mode already checked */
4246                 pr_err("%s: Error: Unknown bonding mode %d\n",
4247                        dev->name, bond->params.mode);
4248                 WARN_ON_ONCE(1);
4249                 dev_kfree_skb(skb);
4250                 return NETDEV_TX_OK;
4251         }
4252 }
4253
4254 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4255 {
4256         struct bonding *bond = netdev_priv(dev);
4257         netdev_tx_t ret = NETDEV_TX_OK;
4258
4259         /*
4260          * If we risk deadlock from transmitting this in the
4261          * netpoll path, tell netpoll to queue the frame for later tx
4262          */
4263         if (is_netpoll_tx_blocked(dev))
4264                 return NETDEV_TX_BUSY;
4265
4266         read_lock(&bond->lock);
4267
4268         if (bond->slave_cnt)
4269                 ret = __bond_start_xmit(skb, dev);
4270         else
4271                 dev_kfree_skb(skb);
4272
4273         read_unlock(&bond->lock);
4274
4275         return ret;
4276 }
4277
4278 /*
4279  * set bond mode specific net device operations
4280  */
4281 void bond_set_mode_ops(struct bonding *bond, int mode)
4282 {
4283         struct net_device *bond_dev = bond->dev;
4284
4285         switch (mode) {
4286         case BOND_MODE_ROUNDROBIN:
4287                 break;
4288         case BOND_MODE_ACTIVEBACKUP:
4289                 break;
4290         case BOND_MODE_XOR:
4291                 bond_set_xmit_hash_policy(bond);
4292                 break;
4293         case BOND_MODE_BROADCAST:
4294                 break;
4295         case BOND_MODE_8023AD:
4296                 bond_set_xmit_hash_policy(bond);
4297                 break;
4298         case BOND_MODE_ALB:
4299                 /* FALLTHRU */
4300         case BOND_MODE_TLB:
4301                 break;
4302         default:
4303                 /* Should never happen, mode already checked */
4304                 pr_err("%s: Error: Unknown bonding mode %d\n",
4305                        bond_dev->name, mode);
4306                 break;
4307         }
4308 }
4309
4310 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4311                                     struct ethtool_drvinfo *drvinfo)
4312 {
4313         strncpy(drvinfo->driver, DRV_NAME, 32);
4314         strncpy(drvinfo->version, DRV_VERSION, 32);
4315         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4316 }
4317
4318 static const struct ethtool_ops bond_ethtool_ops = {
4319         .get_drvinfo            = bond_ethtool_get_drvinfo,
4320         .get_link               = ethtool_op_get_link,
4321 };
4322
4323 static const struct net_device_ops bond_netdev_ops = {
4324         .ndo_init               = bond_init,
4325         .ndo_uninit             = bond_uninit,
4326         .ndo_open               = bond_open,
4327         .ndo_stop               = bond_close,
4328         .ndo_start_xmit         = bond_start_xmit,
4329         .ndo_select_queue       = bond_select_queue,
4330         .ndo_get_stats64        = bond_get_stats,
4331         .ndo_do_ioctl           = bond_do_ioctl,
4332         .ndo_set_multicast_list = bond_set_multicast_list,
4333         .ndo_change_mtu         = bond_change_mtu,
4334         .ndo_set_mac_address    = bond_set_mac_address,
4335         .ndo_neigh_setup        = bond_neigh_setup,
4336         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4337         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4338         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4339 #ifdef CONFIG_NET_POLL_CONTROLLER
4340         .ndo_netpoll_setup      = bond_netpoll_setup,
4341         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4342         .ndo_poll_controller    = bond_poll_controller,
4343 #endif
4344         .ndo_add_slave          = bond_enslave,
4345         .ndo_del_slave          = bond_release,
4346         .ndo_fix_features       = bond_fix_features,
4347 };
4348
4349 static void bond_destructor(struct net_device *bond_dev)
4350 {
4351         struct bonding *bond = netdev_priv(bond_dev);
4352         if (bond->wq)
4353                 destroy_workqueue(bond->wq);
4354         free_netdev(bond_dev);
4355 }
4356
4357 static void bond_setup(struct net_device *bond_dev)
4358 {
4359         struct bonding *bond = netdev_priv(bond_dev);
4360
4361         /* initialize rwlocks */
4362         rwlock_init(&bond->lock);
4363         rwlock_init(&bond->curr_slave_lock);
4364
4365         bond->params = bonding_defaults;
4366
4367         /* Initialize pointers */
4368         bond->dev = bond_dev;
4369         INIT_LIST_HEAD(&bond->vlan_list);
4370
4371         /* Initialize the device entry points */
4372         ether_setup(bond_dev);
4373         bond_dev->netdev_ops = &bond_netdev_ops;
4374         bond_dev->ethtool_ops = &bond_ethtool_ops;
4375         bond_set_mode_ops(bond, bond->params.mode);
4376
4377         bond_dev->destructor = bond_destructor;
4378
4379         /* Initialize the device options */
4380         bond_dev->tx_queue_len = 0;
4381         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4382         bond_dev->priv_flags |= IFF_BONDING;
4383         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4384
4385         /* At first, we block adding VLANs. That's the only way to
4386          * prevent problems that occur when adding VLANs over an
4387          * empty bond. The block will be removed once non-challenged
4388          * slaves are enslaved.
4389          */
4390         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4391
4392         /* don't acquire bond device's netif_tx_lock when
4393          * transmitting */
4394         bond_dev->features |= NETIF_F_LLTX;
4395
4396         /* By default, we declare the bond to be fully
4397          * VLAN hardware accelerated capable. Special
4398          * care is taken in the various xmit functions
4399          * when there are slaves that are not hw accel
4400          * capable
4401          */
4402
4403         bond_dev->hw_features = BOND_VLAN_FEATURES |
4404                                 NETIF_F_HW_VLAN_TX |
4405                                 NETIF_F_HW_VLAN_RX |
4406                                 NETIF_F_HW_VLAN_FILTER;
4407
4408         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4409         bond_dev->features |= bond_dev->hw_features;
4410 }
4411
4412 static void bond_work_cancel_all(struct bonding *bond)
4413 {
4414         write_lock_bh(&bond->lock);
4415         bond->kill_timers = 1;
4416         write_unlock_bh(&bond->lock);
4417
4418         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4419                 cancel_delayed_work(&bond->mii_work);
4420
4421         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4422                 cancel_delayed_work(&bond->arp_work);
4423
4424         if (bond->params.mode == BOND_MODE_ALB &&
4425             delayed_work_pending(&bond->alb_work))
4426                 cancel_delayed_work(&bond->alb_work);
4427
4428         if (bond->params.mode == BOND_MODE_8023AD &&
4429             delayed_work_pending(&bond->ad_work))
4430                 cancel_delayed_work(&bond->ad_work);
4431
4432         if (delayed_work_pending(&bond->mcast_work))
4433                 cancel_delayed_work(&bond->mcast_work);
4434 }
4435
4436 /*
4437 * Destroy a bonding device.
4438 * Must be under rtnl_lock when this function is called.
4439 */
4440 static void bond_uninit(struct net_device *bond_dev)
4441 {
4442         struct bonding *bond = netdev_priv(bond_dev);
4443         struct vlan_entry *vlan, *tmp;
4444
4445         bond_netpoll_cleanup(bond_dev);
4446
4447         /* Release the bonded slaves */
4448         bond_release_all(bond_dev);
4449
4450         list_del(&bond->bond_list);
4451
4452         bond_work_cancel_all(bond);
4453
4454         bond_remove_proc_entry(bond);
4455
4456         bond_debug_unregister(bond);
4457
4458         __hw_addr_flush(&bond->mc_list);
4459
4460         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4461                 list_del(&vlan->vlan_list);
4462                 kfree(vlan);
4463         }
4464 }
4465
4466 /*------------------------- Module initialization ---------------------------*/
4467
4468 /*
4469  * Convert string input module parms.  Accept either the
4470  * number of the mode or its string name.  A bit complicated because
4471  * some mode names are substrings of other names, and calls from sysfs
4472  * may have whitespace in the name (trailing newlines, for example).
4473  */
4474 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4475 {
4476         int modeint = -1, i, rv;
4477         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4478
4479         for (p = (char *)buf; *p; p++)
4480                 if (!(isdigit(*p) || isspace(*p)))
4481                         break;
4482
4483         if (*p)
4484                 rv = sscanf(buf, "%20s", modestr);
4485         else
4486                 rv = sscanf(buf, "%d", &modeint);
4487
4488         if (!rv)
4489                 return -1;
4490
4491         for (i = 0; tbl[i].modename; i++) {
4492                 if (modeint == tbl[i].mode)
4493                         return tbl[i].mode;
4494                 if (strcmp(modestr, tbl[i].modename) == 0)
4495                         return tbl[i].mode;
4496         }
4497
4498         return -1;
4499 }
4500
4501 static int bond_check_params(struct bond_params *params)
4502 {
4503         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4504
4505         /*
4506          * Convert string parameters.
4507          */
4508         if (mode) {
4509                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4510                 if (bond_mode == -1) {
4511                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4512                                mode == NULL ? "NULL" : mode);
4513                         return -EINVAL;
4514                 }
4515         }
4516
4517         if (xmit_hash_policy) {
4518                 if ((bond_mode != BOND_MODE_XOR) &&
4519                     (bond_mode != BOND_MODE_8023AD)) {
4520                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4521                                bond_mode_name(bond_mode));
4522                 } else {
4523                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4524                                                         xmit_hashtype_tbl);
4525                         if (xmit_hashtype == -1) {
4526                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4527                                        xmit_hash_policy == NULL ? "NULL" :
4528                                        xmit_hash_policy);
4529                                 return -EINVAL;
4530                         }
4531                 }
4532         }
4533
4534         if (lacp_rate) {
4535                 if (bond_mode != BOND_MODE_8023AD) {
4536                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4537                                 bond_mode_name(bond_mode));
4538                 } else {
4539                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4540                         if (lacp_fast == -1) {
4541                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4542                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4543                                 return -EINVAL;
4544                         }
4545                 }
4546         }
4547
4548         if (ad_select) {
4549                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4550                 if (params->ad_select == -1) {
4551                         pr_err("Error: Invalid ad_select \"%s\"\n",
4552                                ad_select == NULL ? "NULL" : ad_select);
4553                         return -EINVAL;
4554                 }
4555
4556                 if (bond_mode != BOND_MODE_8023AD) {
4557                         pr_warning("ad_select param only affects 802.3ad mode\n");
4558                 }
4559         } else {
4560                 params->ad_select = BOND_AD_STABLE;
4561         }
4562
4563         if (max_bonds < 0) {
4564                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4565                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4566                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4567         }
4568
4569         if (miimon < 0) {
4570                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4571                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4572                 miimon = BOND_LINK_MON_INTERV;
4573         }
4574
4575         if (updelay < 0) {
4576                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4577                            updelay, INT_MAX);
4578                 updelay = 0;
4579         }
4580
4581         if (downdelay < 0) {
4582                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4583                            downdelay, INT_MAX);
4584                 downdelay = 0;
4585         }
4586
4587         if ((use_carrier != 0) && (use_carrier != 1)) {
4588                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4589                            use_carrier);
4590                 use_carrier = 1;
4591         }
4592
4593         if (num_peer_notif < 0 || num_peer_notif > 255) {
4594                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4595                            num_peer_notif);
4596                 num_peer_notif = 1;
4597         }
4598
4599         /* reset values for 802.3ad */
4600         if (bond_mode == BOND_MODE_8023AD) {
4601                 if (!miimon) {
4602                         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");
4603                         pr_warning("Forcing miimon to 100msec\n");
4604                         miimon = 100;
4605                 }
4606         }
4607
4608         if (tx_queues < 1 || tx_queues > 255) {
4609                 pr_warning("Warning: tx_queues (%d) should be between "
4610                            "1 and 255, resetting to %d\n",
4611                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4612                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4613         }
4614
4615         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4616                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4617                            "not of valid value (0/1), so it was set to "
4618                            "0\n", all_slaves_active);
4619                 all_slaves_active = 0;
4620         }
4621
4622         if (resend_igmp < 0 || resend_igmp > 255) {
4623                 pr_warning("Warning: resend_igmp (%d) should be between "
4624                            "0 and 255, resetting to %d\n",
4625                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4626                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4627         }
4628
4629         /* reset values for TLB/ALB */
4630         if ((bond_mode == BOND_MODE_TLB) ||
4631             (bond_mode == BOND_MODE_ALB)) {
4632                 if (!miimon) {
4633                         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");
4634                         pr_warning("Forcing miimon to 100msec\n");
4635                         miimon = 100;
4636                 }
4637         }
4638
4639         if (bond_mode == BOND_MODE_ALB) {
4640                 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",
4641                           updelay);
4642         }
4643
4644         if (!miimon) {
4645                 if (updelay || downdelay) {
4646                         /* just warn the user the up/down delay will have
4647                          * no effect since miimon is zero...
4648                          */
4649                         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",
4650                                    updelay, downdelay);
4651                 }
4652         } else {
4653                 /* don't allow arp monitoring */
4654                 if (arp_interval) {
4655                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4656                                    miimon, arp_interval);
4657                         arp_interval = 0;
4658                 }
4659
4660                 if ((updelay % miimon) != 0) {
4661                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4662                                    updelay, miimon,
4663                                    (updelay / miimon) * miimon);
4664                 }
4665
4666                 updelay /= miimon;
4667
4668                 if ((downdelay % miimon) != 0) {
4669                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4670                                    downdelay, miimon,
4671                                    (downdelay / miimon) * miimon);
4672                 }
4673
4674                 downdelay /= miimon;
4675         }
4676
4677         if (arp_interval < 0) {
4678                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4679                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4680                 arp_interval = BOND_LINK_ARP_INTERV;
4681         }
4682
4683         for (arp_ip_count = 0;
4684              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4685              arp_ip_count++) {
4686                 /* not complete check, but should be good enough to
4687                    catch mistakes */
4688                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4689                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4690                                    arp_ip_target[arp_ip_count]);
4691                         arp_interval = 0;
4692                 } else {
4693                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4694                         arp_target[arp_ip_count] = ip;
4695                 }
4696         }
4697
4698         if (arp_interval && !arp_ip_count) {
4699                 /* don't allow arping if no arp_ip_target given... */
4700                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4701                            arp_interval);
4702                 arp_interval = 0;
4703         }
4704
4705         if (arp_validate) {
4706                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4707                         pr_err("arp_validate only supported in active-backup mode\n");
4708                         return -EINVAL;
4709                 }
4710                 if (!arp_interval) {
4711                         pr_err("arp_validate requires arp_interval\n");
4712                         return -EINVAL;
4713                 }
4714
4715                 arp_validate_value = bond_parse_parm(arp_validate,
4716                                                      arp_validate_tbl);
4717                 if (arp_validate_value == -1) {
4718                         pr_err("Error: invalid arp_validate \"%s\"\n",
4719                                arp_validate == NULL ? "NULL" : arp_validate);
4720                         return -EINVAL;
4721                 }
4722         } else
4723                 arp_validate_value = 0;
4724
4725         if (miimon) {
4726                 pr_info("MII link monitoring set to %d ms\n", miimon);
4727         } else if (arp_interval) {
4728                 int i;
4729
4730                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4731                         arp_interval,
4732                         arp_validate_tbl[arp_validate_value].modename,
4733                         arp_ip_count);
4734
4735                 for (i = 0; i < arp_ip_count; i++)
4736                         pr_info(" %s", arp_ip_target[i]);
4737
4738                 pr_info("\n");
4739
4740         } else if (max_bonds) {
4741                 /* miimon and arp_interval not set, we need one so things
4742                  * work as expected, see bonding.txt for details
4743                  */
4744                 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");
4745         }
4746
4747         if (primary && !USES_PRIMARY(bond_mode)) {
4748                 /* currently, using a primary only makes sense
4749                  * in active backup, TLB or ALB modes
4750                  */
4751                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4752                            primary, bond_mode_name(bond_mode));
4753                 primary = NULL;
4754         }
4755
4756         if (primary && primary_reselect) {
4757                 primary_reselect_value = bond_parse_parm(primary_reselect,
4758                                                          pri_reselect_tbl);
4759                 if (primary_reselect_value == -1) {
4760                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4761                                primary_reselect ==
4762                                         NULL ? "NULL" : primary_reselect);
4763                         return -EINVAL;
4764                 }
4765         } else {
4766                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4767         }
4768
4769         if (fail_over_mac) {
4770                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4771                                                       fail_over_mac_tbl);
4772                 if (fail_over_mac_value == -1) {
4773                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4774                                arp_validate == NULL ? "NULL" : arp_validate);
4775                         return -EINVAL;
4776                 }
4777
4778                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4779                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4780         } else {
4781                 fail_over_mac_value = BOND_FOM_NONE;
4782         }
4783
4784         /* fill params struct with the proper values */
4785         params->mode = bond_mode;
4786         params->xmit_policy = xmit_hashtype;
4787         params->miimon = miimon;
4788         params->num_peer_notif = num_peer_notif;
4789         params->arp_interval = arp_interval;
4790         params->arp_validate = arp_validate_value;
4791         params->updelay = updelay;
4792         params->downdelay = downdelay;
4793         params->use_carrier = use_carrier;
4794         params->lacp_fast = lacp_fast;
4795         params->primary[0] = 0;
4796         params->primary_reselect = primary_reselect_value;
4797         params->fail_over_mac = fail_over_mac_value;
4798         params->tx_queues = tx_queues;
4799         params->all_slaves_active = all_slaves_active;
4800         params->resend_igmp = resend_igmp;
4801
4802         if (primary) {
4803                 strncpy(params->primary, primary, IFNAMSIZ);
4804                 params->primary[IFNAMSIZ - 1] = 0;
4805         }
4806
4807         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4808
4809         return 0;
4810 }
4811
4812 static struct lock_class_key bonding_netdev_xmit_lock_key;
4813 static struct lock_class_key bonding_netdev_addr_lock_key;
4814
4815 static void bond_set_lockdep_class_one(struct net_device *dev,
4816                                        struct netdev_queue *txq,
4817                                        void *_unused)
4818 {
4819         lockdep_set_class(&txq->_xmit_lock,
4820                           &bonding_netdev_xmit_lock_key);
4821 }
4822
4823 static void bond_set_lockdep_class(struct net_device *dev)
4824 {
4825         lockdep_set_class(&dev->addr_list_lock,
4826                           &bonding_netdev_addr_lock_key);
4827         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4828 }
4829
4830 /*
4831  * Called from registration process
4832  */
4833 static int bond_init(struct net_device *bond_dev)
4834 {
4835         struct bonding *bond = netdev_priv(bond_dev);
4836         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4837
4838         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4839
4840         bond->wq = create_singlethread_workqueue(bond_dev->name);
4841         if (!bond->wq)
4842                 return -ENOMEM;
4843
4844         bond_set_lockdep_class(bond_dev);
4845
4846         bond_create_proc_entry(bond);
4847         list_add_tail(&bond->bond_list, &bn->dev_list);
4848
4849         bond_prepare_sysfs_group(bond);
4850
4851         bond_debug_register(bond);
4852
4853         __hw_addr_init(&bond->mc_list);
4854         return 0;
4855 }
4856
4857 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4858 {
4859         if (tb[IFLA_ADDRESS]) {
4860                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4861                         return -EINVAL;
4862                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4863                         return -EADDRNOTAVAIL;
4864         }
4865         return 0;
4866 }
4867
4868 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4869         .kind           = "bond",
4870         .priv_size      = sizeof(struct bonding),
4871         .setup          = bond_setup,
4872         .validate       = bond_validate,
4873 };
4874
4875 /* Create a new bond based on the specified name and bonding parameters.
4876  * If name is NULL, obtain a suitable "bond%d" name for us.
4877  * Caller must NOT hold rtnl_lock; we need to release it here before we
4878  * set up our sysfs entries.
4879  */
4880 int bond_create(struct net *net, const char *name)
4881 {
4882         struct net_device *bond_dev;
4883         int res;
4884
4885         rtnl_lock();
4886
4887         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4888                                    name ? name : "bond%d",
4889                                    bond_setup, tx_queues);
4890         if (!bond_dev) {
4891                 pr_err("%s: eek! can't alloc netdev!\n", name);
4892                 rtnl_unlock();
4893                 return -ENOMEM;
4894         }
4895
4896         dev_net_set(bond_dev, net);
4897         bond_dev->rtnl_link_ops = &bond_link_ops;
4898
4899         res = register_netdevice(bond_dev);
4900
4901         netif_carrier_off(bond_dev);
4902
4903         rtnl_unlock();
4904         if (res < 0)
4905                 bond_destructor(bond_dev);
4906         return res;
4907 }
4908
4909 static int __net_init bond_net_init(struct net *net)
4910 {
4911         struct bond_net *bn = net_generic(net, bond_net_id);
4912
4913         bn->net = net;
4914         INIT_LIST_HEAD(&bn->dev_list);
4915
4916         bond_create_proc_dir(bn);
4917
4918         return 0;
4919 }
4920
4921 static void __net_exit bond_net_exit(struct net *net)
4922 {
4923         struct bond_net *bn = net_generic(net, bond_net_id);
4924
4925         bond_destroy_proc_dir(bn);
4926 }
4927
4928 static struct pernet_operations bond_net_ops = {
4929         .init = bond_net_init,
4930         .exit = bond_net_exit,
4931         .id   = &bond_net_id,
4932         .size = sizeof(struct bond_net),
4933 };
4934
4935 static int __init bonding_init(void)
4936 {
4937         int i;
4938         int res;
4939
4940         pr_info("%s", bond_version);
4941
4942         res = bond_check_params(&bonding_defaults);
4943         if (res)
4944                 goto out;
4945
4946         res = register_pernet_subsys(&bond_net_ops);
4947         if (res)
4948                 goto out;
4949
4950         res = rtnl_link_register(&bond_link_ops);
4951         if (res)
4952                 goto err_link;
4953
4954         bond_create_debugfs();
4955
4956         for (i = 0; i < max_bonds; i++) {
4957                 res = bond_create(&init_net, NULL);
4958                 if (res)
4959                         goto err;
4960         }
4961
4962         res = bond_create_sysfs();
4963         if (res)
4964                 goto err;
4965
4966         register_netdevice_notifier(&bond_netdev_notifier);
4967         register_inetaddr_notifier(&bond_inetaddr_notifier);
4968 out:
4969         return res;
4970 err:
4971         rtnl_link_unregister(&bond_link_ops);
4972 err_link:
4973         unregister_pernet_subsys(&bond_net_ops);
4974         goto out;
4975
4976 }
4977
4978 static void __exit bonding_exit(void)
4979 {
4980         unregister_netdevice_notifier(&bond_netdev_notifier);
4981         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4982
4983         bond_destroy_sysfs();
4984         bond_destroy_debugfs();
4985
4986         rtnl_link_unregister(&bond_link_ops);
4987         unregister_pernet_subsys(&bond_net_ops);
4988
4989 #ifdef CONFIG_NET_POLL_CONTROLLER
4990         /*
4991          * Make sure we don't have an imbalance on our netpoll blocking
4992          */
4993         WARN_ON(atomic_read(&netpoll_block_tx));
4994 #endif
4995 }
4996
4997 module_init(bonding_init);
4998 module_exit(bonding_exit);
4999 MODULE_LICENSE("GPL");
5000 MODULE_VERSION(DRV_VERSION);
5001 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5002 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5003 MODULE_ALIAS_RTNL_LINK("bond");