drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <linux/preempt.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include <net/netns/generic.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval = BOND_LINK_ARP_INTERV;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *fail_over_mac;
 107 static int all_slaves_active = 0;
 108 static struct bond_params bonding_defaults;
 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 110
 111 module_param(max_bonds, int, 0);
 112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 113 module_param(tx_queues, int, 0);
 114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 117                                "failover event (alias of num_unsol_na)");
 118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 120                                "failover event (alias of num_grat_arp)");
 121 module_param(miimon, int, 0);
 122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 123 module_param(updelay, int, 0);
 124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 125 module_param(downdelay, int, 0);
 126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 127                             "in milliseconds");
 128 module_param(use_carrier, int, 0);
 129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 130                               "0 for off, 1 for on (default)");
 131 module_param(mode, charp, 0);
 132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 133                        "1 for active-backup, 2 for balance-xor, "
 134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 135                        "6 for balance-alb");
 136 module_param(primary, charp, 0);
 137 MODULE_PARM_DESC(primary, "Primary network device to use");
 138 module_param(primary_reselect, charp, 0);
 139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 140                                    "once it comes up; "
 141                                    "0 for always (default), "
 142                                    "1 for only if speed of primary is "
 143                                    "better, "
 144                                    "2 for only on active slave "
 145                                    "failure");
 146 module_param(lacp_rate, charp, 0);
 147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 148                             "0 for slow, 1 for fast");
 149 module_param(ad_select, charp, 0);
 150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 151                             "0 for stable (default), 1 for bandwidth, "
 152                             "2 for count");
 153 module_param(xmit_hash_policy, charp, 0);
 154 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 155                                    "0 for layer 2 (default), 1 for layer 3+4, "
 156                                    "2 for layer 2+3");
 157 module_param(arp_interval, int, 0);
 158 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 159 module_param_array(arp_ip_target, charp, NULL, 0);
 160 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 161 module_param(arp_validate, charp, 0);
 162 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 163                                "0 for none (default), 1 for active, "
 164                                "2 for backup, 3 for all");
 165 module_param(fail_over_mac, charp, 0);
 166 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 167                                 "the same MAC; 0 for none (default), "
 168                                 "1 for active, 2 for follow");
 169 module_param(all_slaves_active, int, 0);
 170 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 171                                      "by setting active flag for all slaves; "
 172                                      "0 for never (default), 1 for always.");
 173 module_param(resend_igmp, int, 0);
 174 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 175                               "link failure");
 176
 177 /*----------------------------- Global variables ----------------------------*/
 178
 179 #ifdef CONFIG_NET_POLL_CONTROLLER
 180 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 181 #endif
 182
 183 int bond_net_id __read_mostly;
 184
 185 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 186 static int arp_ip_count;
 187 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 188 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 189 static int lacp_fast;
 190
 191 const struct bond_parm_tbl bond_lacp_tbl[] = {
 192 {       "slow",         AD_LACP_SLOW},
 193 {       "fast",         AD_LACP_FAST},
 194 {       NULL,           -1},
 195 };
 196
 197 const struct bond_parm_tbl bond_mode_tbl[] = {
 198 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 199 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 200 {       "balance-xor",          BOND_MODE_XOR},
 201 {       "broadcast",            BOND_MODE_BROADCAST},
 202 {       "802.3ad",              BOND_MODE_8023AD},
 203 {       "balance-tlb",          BOND_MODE_TLB},
 204 {       "balance-alb",          BOND_MODE_ALB},
 205 {       NULL,                   -1},
 206 };
 207
 208 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 209 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 210 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 211 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl arp_validate_tbl[] = {
 216 {       "none",                 BOND_ARP_VALIDATE_NONE},
 217 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 218 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 219 {       "all",                  BOND_ARP_VALIDATE_ALL},
 220 {       NULL,                   -1},
 221 };
 222
 223 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 224 {       "none",                 BOND_FOM_NONE},
 225 {       "active",               BOND_FOM_ACTIVE},
 226 {       "follow",               BOND_FOM_FOLLOW},
 227 {       NULL,                   -1},
 228 };
 229
 230 const struct bond_parm_tbl pri_reselect_tbl[] = {
 231 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 232 {       "better",               BOND_PRI_RESELECT_BETTER},
 233 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 234 {       NULL,                   -1},
 235 };
 236
 237 struct bond_parm_tbl ad_select_tbl[] = {
 238 {       "stable",       BOND_AD_STABLE},
 239 {       "bandwidth",    BOND_AD_BANDWIDTH},
 240 {       "count",        BOND_AD_COUNT},
 241 {       NULL,           -1},
 242 };
 243
 244 /*-------------------------- Forward declarations ---------------------------*/
 245
 246 static int bond_init(struct net_device *bond_dev);
 247 static void bond_uninit(struct net_device *bond_dev);
 248
 249 /*---------------------------- General routines -----------------------------*/
 250
 251 const char *bond_mode_name(int mode)
 252 {
 253         static const char *names[] = {
 254                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 255                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 256                 [BOND_MODE_XOR] = "load balancing (xor)",
 257                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 258                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 259                 [BOND_MODE_TLB] = "transmit load balancing",
 260                 [BOND_MODE_ALB] = "adaptive load balancing",
 261         };
 262
 263         if (mode < 0 || mode > BOND_MODE_ALB)
 264                 return "unknown";
 265
 266         return names[mode];
 267 }
 268
 269 /*---------------------------------- VLAN -----------------------------------*/
 270
 271 /**
 272  * bond_add_vlan - add a new vlan id on bond
 273  * @bond: bond that got the notification
 274  * @vlan_id: the vlan id to add
 275  *
 276  * Returns -ENOMEM if allocation failed.
 277  */
 278 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 279 {
 280         struct vlan_entry *vlan;
 281
 282         pr_debug("bond: %s, vlan id %d\n",
 283                  (bond ? bond->dev->name : "None"), vlan_id);
 284
 285         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 286         if (!vlan)
 287                 return -ENOMEM;
 288
 289         INIT_LIST_HEAD(&vlan->vlan_list);
 290         vlan->vlan_id = vlan_id;
 291
 292         write_lock_bh(&bond->lock);
 293
 294         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 295
 296         write_unlock_bh(&bond->lock);
 297
 298         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 299
 300         return 0;
 301 }
 302
 303 /**
 304  * bond_del_vlan - delete a vlan id from bond
 305  * @bond: bond that got the notification
 306  * @vlan_id: the vlan id to delete
 307  *
 308  * returns -ENODEV if @vlan_id was not found in @bond.
 309  */
 310 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 311 {
 312         struct vlan_entry *vlan;
 313         int res = -ENODEV;
 314
 315         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 316
 317         block_netpoll_tx();
 318         write_lock_bh(&bond->lock);
 319
 320         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 321                 if (vlan->vlan_id == vlan_id) {
 322                         list_del(&vlan->vlan_list);
 323
 324                         if (bond_is_lb(bond))
 325                                 bond_alb_clear_vlan(bond, vlan_id);
 326
 327                         pr_debug("removed VLAN ID %d from bond %s\n",
 328                                  vlan_id, bond->dev->name);
 329
 330                         kfree(vlan);
 331
 332                         if (list_empty(&bond->vlan_list) &&
 333                             (bond->slave_cnt == 0)) {
 334                                 /* Last VLAN removed and no slaves, so
 335                                  * restore block on adding VLANs. This will
 336                                  * be removed once new slaves that are not
 337                                  * VLAN challenged will be added.
 338                                  */
 339                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 340                         }
 341
 342                         res = 0;
 343                         goto out;
 344                 }
 345         }
 346
 347         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 348                  vlan_id, bond->dev->name);
 349
 350 out:
 351         write_unlock_bh(&bond->lock);
 352         unblock_netpoll_tx();
 353         return res;
 354 }
 355
 356 /**
 357  * bond_next_vlan - safely skip to the next item in the vlans list.
 358  * @bond: the bond we're working on
 359  * @curr: item we're advancing from
 360  *
 361  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 362  * or @curr->next otherwise (even if it is @curr itself again).
 363  *
 364  * Caller must hold bond->lock
 365  */
 366 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 367 {
 368         struct vlan_entry *next, *last;
 369
 370         if (list_empty(&bond->vlan_list))
 371                 return NULL;
 372
 373         if (!curr) {
 374                 next = list_entry(bond->vlan_list.next,
 375                                   struct vlan_entry, vlan_list);
 376         } else {
 377                 last = list_entry(bond->vlan_list.prev,
 378                                   struct vlan_entry, vlan_list);
 379                 if (last == curr) {
 380                         next = list_entry(bond->vlan_list.next,
 381                                           struct vlan_entry, vlan_list);
 382                 } else {
 383                         next = list_entry(curr->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 }
 386         }
 387
 388         return next;
 389 }
 390
 391 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 392
 393 /**
 394  * bond_dev_queue_xmit - Prepare skb for xmit.
 395  *
 396  * @bond: bond device that got this skb for tx.
 397  * @skb: hw accel VLAN tagged skb to transmit
 398  * @slave_dev: slave that is supposed to xmit this skbuff
 399  */
 400 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 401                         struct net_device *slave_dev)
 402 {
 403         skb->dev = slave_dev;
 404         skb->priority = 1;
 405
 406         skb->queue_mapping = bond_queue_mapping(skb);
 407
 408         if (unlikely(netpoll_tx_running(slave_dev)))
 409                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 410         else
 411                 dev_queue_xmit(skb);
 412
 413         return 0;
 414 }
 415
 416 /*
 417  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 418  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 419  * lock because:
 420  * a. This operation is performed in IOCTL context,
 421  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 422  * c. Holding a lock with BH disabled while directly calling a base driver
 423  *    entry point is generally a BAD idea.
 424  *
 425  * The design of synchronization/protection for this operation in the 8021q
 426  * module is good for one or more VLAN devices over a single physical device
 427  * and cannot be extended for a teaming solution like bonding, so there is a
 428  * potential race condition here where a net device from the vlan group might
 429  * be referenced (either by a base driver or the 8021q code) while it is being
 430  * removed from the system. However, it turns out we're not making matters
 431  * worse, and if it works for regular VLAN usage it will work here too.
 432 */
 433
 434 /**
 435  * bond_vlan_rx_register - Propagates registration to slaves
 436  * @bond_dev: bonding net device that got called
 437  * @grp: vlan group being registered
 438  */
 439 static void bond_vlan_rx_register(struct net_device *bond_dev,
 440                                   struct vlan_group *grp)
 441 {
 442         struct bonding *bond = netdev_priv(bond_dev);
 443         struct slave *slave;
 444         int i;
 445
 446         write_lock_bh(&bond->lock);
 447         bond->vlgrp = grp;
 448         write_unlock_bh(&bond->lock);
 449
 450         bond_for_each_slave(bond, slave, i) {
 451                 struct net_device *slave_dev = slave->dev;
 452                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 453
 454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 455                     slave_ops->ndo_vlan_rx_register) {
 456                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 457                 }
 458         }
 459 }
 460
 461 /**
 462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being added
 465  */
 466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i) {
 473                 struct net_device *slave_dev = slave->dev;
 474                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 475
 476                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 477                     slave_ops->ndo_vlan_rx_add_vid) {
 478                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 479                 }
 480         }
 481
 482         res = bond_add_vlan(bond, vid);
 483         if (res) {
 484                 pr_err("%s: Error: Failed to add vlan id %d\n",
 485                        bond_dev->name, vid);
 486         }
 487 }
 488
 489 /**
 490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 491  * @bond_dev: bonding net device that got called
 492  * @vid: vlan id being removed
 493  */
 494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 495 {
 496         struct bonding *bond = netdev_priv(bond_dev);
 497         struct slave *slave;
 498         struct net_device *vlan_dev;
 499         int i, res;
 500
 501         bond_for_each_slave(bond, slave, i) {
 502                 struct net_device *slave_dev = slave->dev;
 503                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 504
 505                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 506                     slave_ops->ndo_vlan_rx_kill_vid) {
 507                         /* Save and then restore vlan_dev in the grp array,
 508                          * since the slave's driver might clear it.
 509                          */
 510                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 511                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 512                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 513                 }
 514         }
 515
 516         res = bond_del_vlan(bond, vid);
 517         if (res) {
 518                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 519                        bond_dev->name, vid);
 520         }
 521 }
 522
 523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 524 {
 525         struct vlan_entry *vlan;
 526         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 527
 528         if (!bond->vlgrp)
 529                 return;
 530
 531         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 532             slave_ops->ndo_vlan_rx_register)
 533                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 534
 535         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 536             !(slave_ops->ndo_vlan_rx_add_vid))
 537                 return;
 538
 539         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 540                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 541 }
 542
 543 static void bond_del_vlans_from_slave(struct bonding *bond,
 544                                       struct net_device *slave_dev)
 545 {
 546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 547         struct vlan_entry *vlan;
 548         struct net_device *vlan_dev;
 549
 550         if (!bond->vlgrp)
 551                 return;
 552
 553         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 554             !(slave_ops->ndo_vlan_rx_kill_vid))
 555                 goto unreg;
 556
 557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 558                 if (!vlan->vlan_id)
 559                         continue;
 560                 /* Save and then restore vlan_dev in the grp array,
 561                  * since the slave's driver might clear it.
 562                  */
 563                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 564                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 565                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 566         }
 567
 568 unreg:
 569         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 570             slave_ops->ndo_vlan_rx_register)
 571                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 572 }
 573
 574 /*------------------------------- Link status -------------------------------*/
 575
 576 /*
 577  * Set the carrier state for the master according to the state of its
 578  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 579  * do special 802.3ad magic.
 580  *
 581  * Returns zero if carrier state does not change, nonzero if it does.
 582  */
 583 static int bond_set_carrier(struct bonding *bond)
 584 {
 585         struct slave *slave;
 586         int i;
 587
 588         if (bond->slave_cnt == 0)
 589                 goto down;
 590
 591         if (bond->params.mode == BOND_MODE_8023AD)
 592                 return bond_3ad_set_carrier(bond);
 593
 594         bond_for_each_slave(bond, slave, i) {
 595                 if (slave->link == BOND_LINK_UP) {
 596                         if (!netif_carrier_ok(bond->dev)) {
 597                                 netif_carrier_on(bond->dev);
 598                                 return 1;
 599                         }
 600                         return 0;
 601                 }
 602         }
 603
 604 down:
 605         if (netif_carrier_ok(bond->dev)) {
 606                 netif_carrier_off(bond->dev);
 607                 return 1;
 608         }
 609         return 0;
 610 }
 611
 612 /*
 613  * Get link speed and duplex from the slave's base driver
 614  * using ethtool. If for some reason the call fails or the
 615  * values are invalid, fake speed and duplex to 100/Full
 616  * and return error.
 617  */
 618 static int bond_update_speed_duplex(struct slave *slave)
 619 {
 620         struct net_device *slave_dev = slave->dev;
 621         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 622         u32 slave_speed;
 623         int res;
 624
 625         /* Fake speed and duplex */
 626         slave->speed = SPEED_100;
 627         slave->duplex = DUPLEX_FULL;
 628
 629         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 630                 return -1;
 631
 632         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 633         if (res < 0)
 634                 return -1;
 635
 636         slave_speed = ethtool_cmd_speed(&etool);
 637         switch (slave_speed) {
 638         case SPEED_10:
 639         case SPEED_100:
 640         case SPEED_1000:
 641         case SPEED_10000:
 642                 break;
 643         default:
 644                 return -1;
 645         }
 646
 647         switch (etool.duplex) {
 648         case DUPLEX_FULL:
 649         case DUPLEX_HALF:
 650                 break;
 651         default:
 652                 return -1;
 653         }
 654
 655         slave->speed = slave_speed;
 656         slave->duplex = etool.duplex;
 657
 658         return 0;
 659 }
 660
 661 /*
 662  * if <dev> supports MII link status reporting, check its link status.
 663  *
 664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 665  * depending upon the setting of the use_carrier parameter.
 666  *
 667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 668  * can't tell and just pretend it is), or 0, meaning that the link is
 669  * down.
 670  *
 671  * If reporting is non-zero, instead of faking link up, return -1 if
 672  * both ETHTOOL and MII ioctls fail (meaning the device does not
 673  * support them).  If use_carrier is set, return whatever it says.
 674  * It'd be nice if there was a good way to tell if a driver supports
 675  * netif_carrier, but there really isn't.
 676  */
 677 static int bond_check_dev_link(struct bonding *bond,
 678                                struct net_device *slave_dev, int reporting)
 679 {
 680         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 681         int (*ioctl)(struct net_device *, struct ifreq *, int);
 682         struct ifreq ifr;
 683         struct mii_ioctl_data *mii;
 684
 685         if (!reporting && !netif_running(slave_dev))
 686                 return 0;
 687
 688         if (bond->params.use_carrier)
 689                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 690
 691         /* Try to get link status using Ethtool first. */
 692         if (slave_dev->ethtool_ops) {
 693                 if (slave_dev->ethtool_ops->get_link) {
 694                         u32 link;
 695
 696                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 697
 698                         return link ? BMSR_LSTATUS : 0;
 699                 }
 700         }
 701
 702         /* Ethtool can't be used, fallback to MII ioctls. */
 703         ioctl = slave_ops->ndo_do_ioctl;
 704         if (ioctl) {
 705                 /* TODO: set pointer to correct ioctl on a per team member */
 706                 /*       bases to make this more efficient. that is, once  */
 707                 /*       we determine the correct ioctl, we will always    */
 708                 /*       call it and not the others for that team          */
 709                 /*       member.                                           */
 710
 711                 /*
 712                  * We cannot assume that SIOCGMIIPHY will also read a
 713                  * register; not all network drivers (e.g., e100)
 714                  * support that.
 715                  */
 716
 717                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 718                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 719                 mii = if_mii(&ifr);
 720                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 721                         mii->reg_num = MII_BMSR;
 722                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 723                                 return mii->val_out & BMSR_LSTATUS;
 724                 }
 725         }
 726
 727         /*
 728          * If reporting, report that either there's no dev->do_ioctl,
 729          * or both SIOCGMIIREG and get_link failed (meaning that we
 730          * cannot report link status).  If not reporting, pretend
 731          * we're ok.
 732          */
 733         return reporting ? -1 : BMSR_LSTATUS;
 734 }
 735
 736 /*----------------------------- Multicast list ------------------------------*/
 737
 738 /*
 739  * Push the promiscuity flag down to appropriate slaves
 740  */
 741 static int bond_set_promiscuity(struct bonding *bond, int inc)
 742 {
 743         int err = 0;
 744         if (USES_PRIMARY(bond->params.mode)) {
 745                 /* write lock already acquired */
 746                 if (bond->curr_active_slave) {
 747                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 748                                                   inc);
 749                 }
 750         } else {
 751                 struct slave *slave;
 752                 int i;
 753                 bond_for_each_slave(bond, slave, i) {
 754                         err = dev_set_promiscuity(slave->dev, inc);
 755                         if (err)
 756                                 return err;
 757                 }
 758         }
 759         return err;
 760 }
 761
 762 /*
 763  * Push the allmulti flag down to all slaves
 764  */
 765 static int bond_set_allmulti(struct bonding *bond, int inc)
 766 {
 767         int err = 0;
 768         if (USES_PRIMARY(bond->params.mode)) {
 769                 /* write lock already acquired */
 770                 if (bond->curr_active_slave) {
 771                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 772                                                inc);
 773                 }
 774         } else {
 775                 struct slave *slave;
 776                 int i;
 777                 bond_for_each_slave(bond, slave, i) {
 778                         err = dev_set_allmulti(slave->dev, inc);
 779                         if (err)
 780                                 return err;
 781                 }
 782         }
 783         return err;
 784 }
 785
 786 /*
 787  * Add a Multicast address to slaves
 788  * according to mode
 789  */
 790 static void bond_mc_add(struct bonding *bond, void *addr)
 791 {
 792         if (USES_PRIMARY(bond->params.mode)) {
 793                 /* write lock already acquired */
 794                 if (bond->curr_active_slave)
 795                         dev_mc_add(bond->curr_active_slave->dev, addr);
 796         } else {
 797                 struct slave *slave;
 798                 int i;
 799
 800                 bond_for_each_slave(bond, slave, i)
 801                         dev_mc_add(slave->dev, addr);
 802         }
 803 }
 804
 805 /*
 806  * Remove a multicast address from slave
 807  * according to mode
 808  */
 809 static void bond_mc_del(struct bonding *bond, void *addr)
 810 {
 811         if (USES_PRIMARY(bond->params.mode)) {
 812                 /* write lock already acquired */
 813                 if (bond->curr_active_slave)
 814                         dev_mc_del(bond->curr_active_slave->dev, addr);
 815         } else {
 816                 struct slave *slave;
 817                 int i;
 818                 bond_for_each_slave(bond, slave, i) {
 819                         dev_mc_del(slave->dev, addr);
 820                 }
 821         }
 822 }
 823
 824
 825 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 826 {
 827         struct in_device *in_dev;
 828
 829         rcu_read_lock();
 830         in_dev = __in_dev_get_rcu(dev);
 831         if (in_dev)
 832                 ip_mc_rejoin_groups(in_dev);
 833         rcu_read_unlock();
 834 }
 835
 836 /*
 837  * Retrieve the list of registered multicast addresses for the bonding
 838  * device and retransmit an IGMP JOIN request to the current active
 839  * slave.
 840  */
 841 static void bond_resend_igmp_join_requests(struct bonding *bond)
 842 {
 843         struct net_device *vlan_dev;
 844         struct vlan_entry *vlan;
 845
 846         read_lock(&bond->lock);
 847
 848         /* rejoin all groups on bond device */
 849         __bond_resend_igmp_join_requests(bond->dev);
 850
 851         /* rejoin all groups on vlan devices */
 852         if (bond->vlgrp) {
 853                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 854                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 855                                                          vlan->vlan_id);
 856                         if (vlan_dev)
 857                                 __bond_resend_igmp_join_requests(vlan_dev);
 858                 }
 859         }
 860
 861         if (--bond->igmp_retrans > 0)
 862                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 863
 864         read_unlock(&bond->lock);
 865 }
 866
 867 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 868 {
 869         struct bonding *bond = container_of(work, struct bonding,
 870                                             mcast_work.work);
 871         bond_resend_igmp_join_requests(bond);
 872 }
 873
 874 /*
 875  * flush all members of flush->mc_list from device dev->mc_list
 876  */
 877 static void bond_mc_list_flush(struct net_device *bond_dev,
 878                                struct net_device *slave_dev)
 879 {
 880         struct bonding *bond = netdev_priv(bond_dev);
 881         struct netdev_hw_addr *ha;
 882
 883         netdev_for_each_mc_addr(ha, bond_dev)
 884                 dev_mc_del(slave_dev, ha->addr);
 885
 886         if (bond->params.mode == BOND_MODE_8023AD) {
 887                 /* del lacpdu mc addr from mc list */
 888                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 889
 890                 dev_mc_del(slave_dev, lacpdu_multicast);
 891         }
 892 }
 893
 894 /*--------------------------- Active slave change ---------------------------*/
 895
 896 /*
 897  * Update the mc list and multicast-related flags for the new and
 898  * old active slaves (if any) according to the multicast mode, and
 899  * promiscuous flags unconditionally.
 900  */
 901 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 902                          struct slave *old_active)
 903 {
 904         struct netdev_hw_addr *ha;
 905
 906         if (!USES_PRIMARY(bond->params.mode))
 907                 /* nothing to do -  mc list is already up-to-date on
 908                  * all slaves
 909                  */
 910                 return;
 911
 912         if (old_active) {
 913                 if (bond->dev->flags & IFF_PROMISC)
 914                         dev_set_promiscuity(old_active->dev, -1);
 915
 916                 if (bond->dev->flags & IFF_ALLMULTI)
 917                         dev_set_allmulti(old_active->dev, -1);
 918
 919                 netdev_for_each_mc_addr(ha, bond->dev)
 920                         dev_mc_del(old_active->dev, ha->addr);
 921         }
 922
 923         if (new_active) {
 924                 /* FIXME: Signal errors upstream. */
 925                 if (bond->dev->flags & IFF_PROMISC)
 926                         dev_set_promiscuity(new_active->dev, 1);
 927
 928                 if (bond->dev->flags & IFF_ALLMULTI)
 929                         dev_set_allmulti(new_active->dev, 1);
 930
 931                 netdev_for_each_mc_addr(ha, bond->dev)
 932                         dev_mc_add(new_active->dev, ha->addr);
 933         }
 934 }
 935
 936 /*
 937  * bond_do_fail_over_mac
 938  *
 939  * Perform special MAC address swapping for fail_over_mac settings
 940  *
 941  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 942  */
 943 static void bond_do_fail_over_mac(struct bonding *bond,
 944                                   struct slave *new_active,
 945                                   struct slave *old_active)
 946         __releases(&bond->curr_slave_lock)
 947         __releases(&bond->lock)
 948         __acquires(&bond->lock)
 949         __acquires(&bond->curr_slave_lock)
 950 {
 951         u8 tmp_mac[ETH_ALEN];
 952         struct sockaddr saddr;
 953         int rv;
 954
 955         switch (bond->params.fail_over_mac) {
 956         case BOND_FOM_ACTIVE:
 957                 if (new_active)
 958                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 959                                new_active->dev->addr_len);
 960                 break;
 961         case BOND_FOM_FOLLOW:
 962                 /*
 963                  * if new_active && old_active, swap them
 964                  * if just old_active, do nothing (going to no active slave)
 965                  * if just new_active, set new_active to bond's MAC
 966                  */
 967                 if (!new_active)
 968                         return;
 969
 970                 write_unlock_bh(&bond->curr_slave_lock);
 971                 read_unlock(&bond->lock);
 972
 973                 if (old_active) {
 974                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 975                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 976                                ETH_ALEN);
 977                         saddr.sa_family = new_active->dev->type;
 978                 } else {
 979                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 980                         saddr.sa_family = bond->dev->type;
 981                 }
 982
 983                 rv = dev_set_mac_address(new_active->dev, &saddr);
 984                 if (rv) {
 985                         pr_err("%s: Error %d setting MAC of slave %s\n",
 986                                bond->dev->name, -rv, new_active->dev->name);
 987                         goto out;
 988                 }
 989
 990                 if (!old_active)
 991                         goto out;
 992
 993                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 994                 saddr.sa_family = old_active->dev->type;
 995
 996                 rv = dev_set_mac_address(old_active->dev, &saddr);
 997                 if (rv)
 998                         pr_err("%s: Error %d setting MAC of slave %s\n",
 999                                bond->dev->name, -rv, new_active->dev->name);
1000 out:
1001                 read_lock(&bond->lock);
1002                 write_lock_bh(&bond->curr_slave_lock);
1003                 break;
1004         default:
1005                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1006                        bond->dev->name, bond->params.fail_over_mac);
1007                 break;
1008         }
1009
1010 }
1011
1012 static bool bond_should_change_active(struct bonding *bond)
1013 {
1014         struct slave *prim = bond->primary_slave;
1015         struct slave *curr = bond->curr_active_slave;
1016
1017         if (!prim || !curr || curr->link != BOND_LINK_UP)
1018                 return true;
1019         if (bond->force_primary) {
1020                 bond->force_primary = false;
1021                 return true;
1022         }
1023         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1024             (prim->speed < curr->speed ||
1025              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1026                 return false;
1027         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1028                 return false;
1029         return true;
1030 }
1031
1032 /**
1033  * find_best_interface - select the best available slave to be the active one
1034  * @bond: our bonding struct
1035  *
1036  * Warning: Caller must hold curr_slave_lock for writing.
1037  */
1038 static struct slave *bond_find_best_slave(struct bonding *bond)
1039 {
1040         struct slave *new_active, *old_active;
1041         struct slave *bestslave = NULL;
1042         int mintime = bond->params.updelay;
1043         int i;
1044
1045         new_active = bond->curr_active_slave;
1046
1047         if (!new_active) { /* there were no active slaves left */
1048                 if (bond->slave_cnt > 0)   /* found one slave */
1049                         new_active = bond->first_slave;
1050                 else
1051                         return NULL; /* still no slave, return NULL */
1052         }
1053
1054         if ((bond->primary_slave) &&
1055             bond->primary_slave->link == BOND_LINK_UP &&
1056             bond_should_change_active(bond)) {
1057                 new_active = bond->primary_slave;
1058         }
1059
1060         /* remember where to stop iterating over the slaves */
1061         old_active = new_active;
1062
1063         bond_for_each_slave_from(bond, new_active, i, old_active) {
1064                 if (new_active->link == BOND_LINK_UP) {
1065                         return new_active;
1066                 } else if (new_active->link == BOND_LINK_BACK &&
1067                            IS_UP(new_active->dev)) {
1068                         /* link up, but waiting for stabilization */
1069                         if (new_active->delay < mintime) {
1070                                 mintime = new_active->delay;
1071                                 bestslave = new_active;
1072                         }
1073                 }
1074         }
1075
1076         return bestslave;
1077 }
1078
1079 static bool bond_should_notify_peers(struct bonding *bond)
1080 {
1081         struct slave *slave = bond->curr_active_slave;
1082
1083         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1084                  bond->dev->name, slave ? slave->dev->name : "NULL");
1085
1086         if (!slave || !bond->send_peer_notif ||
1087             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1088                 return false;
1089
1090         bond->send_peer_notif--;
1091         return true;
1092 }
1093
1094 /**
1095  * change_active_interface - change the active slave into the specified one
1096  * @bond: our bonding struct
1097  * @new: the new slave to make the active one
1098  *
1099  * Set the new slave to the bond's settings and unset them on the old
1100  * curr_active_slave.
1101  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1102  *
1103  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1104  * because it is apparently the best available slave we have, even though its
1105  * updelay hasn't timed out yet.
1106  *
1107  * If new_active is not NULL, caller must hold bond->lock for read and
1108  * curr_slave_lock for write_bh.
1109  */
1110 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1111 {
1112         struct slave *old_active = bond->curr_active_slave;
1113
1114         if (old_active == new_active)
1115                 return;
1116
1117         if (new_active) {
1118                 new_active->jiffies = jiffies;
1119
1120                 if (new_active->link == BOND_LINK_BACK) {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1123                                         bond->dev->name, new_active->dev->name,
1124                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1125                         }
1126
1127                         new_active->delay = 0;
1128                         new_active->link = BOND_LINK_UP;
1129
1130                         if (bond->params.mode == BOND_MODE_8023AD)
1131                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1132
1133                         if (bond_is_lb(bond))
1134                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1135                 } else {
1136                         if (USES_PRIMARY(bond->params.mode)) {
1137                                 pr_info("%s: making interface %s the new active one.\n",
1138                                         bond->dev->name, new_active->dev->name);
1139                         }
1140                 }
1141         }
1142
1143         if (USES_PRIMARY(bond->params.mode))
1144                 bond_mc_swap(bond, new_active, old_active);
1145
1146         if (bond_is_lb(bond)) {
1147                 bond_alb_handle_active_change(bond, new_active);
1148                 if (old_active)
1149                         bond_set_slave_inactive_flags(old_active);
1150                 if (new_active)
1151                         bond_set_slave_active_flags(new_active);
1152         } else {
1153                 bond->curr_active_slave = new_active;
1154         }
1155
1156         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1157                 if (old_active)
1158                         bond_set_slave_inactive_flags(old_active);
1159
1160                 if (new_active) {
1161                         bool should_notify_peers = false;
1162
1163                         bond_set_slave_active_flags(new_active);
1164
1165                         if (bond->params.fail_over_mac)
1166                                 bond_do_fail_over_mac(bond, new_active,
1167                                                       old_active);
1168
1169                         if (netif_running(bond->dev)) {
1170                                 bond->send_peer_notif =
1171                                         bond->params.num_peer_notif;
1172                                 should_notify_peers =
1173                                         bond_should_notify_peers(bond);
1174                         }
1175
1176                         write_unlock_bh(&bond->curr_slave_lock);
1177                         read_unlock(&bond->lock);
1178
1179                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1180                         if (should_notify_peers)
1181                                 netdev_bonding_change(bond->dev,
1182                                                       NETDEV_NOTIFY_PEERS);
1183
1184                         read_lock(&bond->lock);
1185                         write_lock_bh(&bond->curr_slave_lock);
1186                 }
1187         }
1188
1189         /* resend IGMP joins since active slave has changed or
1190          * all were sent on curr_active_slave.
1191          * resend only if bond is brought up with the affected
1192          * bonding modes and the retransmission is enabled */
1193         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1194             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1195              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1196                 bond->igmp_retrans = bond->params.resend_igmp;
1197                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1198         }
1199 }
1200
1201 /**
1202  * bond_select_active_slave - select a new active slave, if needed
1203  * @bond: our bonding struct
1204  *
1205  * This functions should be called when one of the following occurs:
1206  * - The old curr_active_slave has been released or lost its link.
1207  * - The primary_slave has got its link back.
1208  * - A slave has got its link back and there's no old curr_active_slave.
1209  *
1210  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1211  */
1212 void bond_select_active_slave(struct bonding *bond)
1213 {
1214         struct slave *best_slave;
1215         int rv;
1216
1217         best_slave = bond_find_best_slave(bond);
1218         if (best_slave != bond->curr_active_slave) {
1219                 bond_change_active_slave(bond, best_slave);
1220                 rv = bond_set_carrier(bond);
1221                 if (!rv)
1222                         return;
1223
1224                 if (netif_carrier_ok(bond->dev)) {
1225                         pr_info("%s: first active interface up!\n",
1226                                 bond->dev->name);
1227                 } else {
1228                         pr_info("%s: now running without any active interface !\n",
1229                                 bond->dev->name);
1230                 }
1231         }
1232 }
1233
1234 /*--------------------------- slave list handling ---------------------------*/
1235
1236 /*
1237  * This function attaches the slave to the end of list.
1238  *
1239  * bond->lock held for writing by caller.
1240  */
1241 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1242 {
1243         if (bond->first_slave == NULL) { /* attaching the first slave */
1244                 new_slave->next = new_slave;
1245                 new_slave->prev = new_slave;
1246                 bond->first_slave = new_slave;
1247         } else {
1248                 new_slave->next = bond->first_slave;
1249                 new_slave->prev = bond->first_slave->prev;
1250                 new_slave->next->prev = new_slave;
1251                 new_slave->prev->next = new_slave;
1252         }
1253
1254         bond->slave_cnt++;
1255 }
1256
1257 /*
1258  * This function detaches the slave from the list.
1259  * WARNING: no check is made to verify if the slave effectively
1260  * belongs to <bond>.
1261  * Nothing is freed on return, structures are just unchained.
1262  * If any slave pointer in bond was pointing to <slave>,
1263  * it should be changed by the calling function.
1264  *
1265  * bond->lock held for writing by caller.
1266  */
1267 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1268 {
1269         if (slave->next)
1270                 slave->next->prev = slave->prev;
1271
1272         if (slave->prev)
1273                 slave->prev->next = slave->next;
1274
1275         if (bond->first_slave == slave) { /* slave is the first slave */
1276                 if (bond->slave_cnt > 1) { /* there are more slave */
1277                         bond->first_slave = slave->next;
1278                 } else {
1279                         bond->first_slave = NULL; /* slave was the last one */
1280                 }
1281         }
1282
1283         slave->next = NULL;
1284         slave->prev = NULL;
1285         bond->slave_cnt--;
1286 }
1287
1288 #ifdef CONFIG_NET_POLL_CONTROLLER
1289 static inline int slave_enable_netpoll(struct slave *slave)
1290 {
1291         struct netpoll *np;
1292         int err = 0;
1293
1294         np = kzalloc(sizeof(*np), GFP_KERNEL);
1295         err = -ENOMEM;
1296         if (!np)
1297                 goto out;
1298
1299         np->dev = slave->dev;
1300         err = __netpoll_setup(np);
1301         if (err) {
1302                 kfree(np);
1303                 goto out;
1304         }
1305         slave->np = np;
1306 out:
1307         return err;
1308 }
1309 static inline void slave_disable_netpoll(struct slave *slave)
1310 {
1311         struct netpoll *np = slave->np;
1312
1313         if (!np)
1314                 return;
1315
1316         slave->np = NULL;
1317         synchronize_rcu_bh();
1318         __netpoll_cleanup(np);
1319         kfree(np);
1320 }
1321 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1322 {
1323         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1324                 return false;
1325         if (!slave_dev->netdev_ops->ndo_poll_controller)
1326                 return false;
1327         return true;
1328 }
1329
1330 static void bond_poll_controller(struct net_device *bond_dev)
1331 {
1332 }
1333
1334 static void __bond_netpoll_cleanup(struct bonding *bond)
1335 {
1336         struct slave *slave;
1337         int i;
1338
1339         bond_for_each_slave(bond, slave, i)
1340                 if (IS_UP(slave->dev))
1341                         slave_disable_netpoll(slave);
1342 }
1343 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1344 {
1345         struct bonding *bond = netdev_priv(bond_dev);
1346
1347         read_lock(&bond->lock);
1348         __bond_netpoll_cleanup(bond);
1349         read_unlock(&bond->lock);
1350 }
1351
1352 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1353 {
1354         struct bonding *bond = netdev_priv(dev);
1355         struct slave *slave;
1356         int i, err = 0;
1357
1358         read_lock(&bond->lock);
1359         bond_for_each_slave(bond, slave, i) {
1360                 err = slave_enable_netpoll(slave);
1361                 if (err) {
1362                         __bond_netpoll_cleanup(bond);
1363                         break;
1364                 }
1365         }
1366         read_unlock(&bond->lock);
1367         return err;
1368 }
1369
1370 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1371 {
1372         return bond->dev->npinfo;
1373 }
1374
1375 #else
1376 static inline int slave_enable_netpoll(struct slave *slave)
1377 {
1378         return 0;
1379 }
1380 static inline void slave_disable_netpoll(struct slave *slave)
1381 {
1382 }
1383 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1384 {
1385 }
1386 #endif
1387
1388 /*---------------------------------- IOCTL ----------------------------------*/
1389
1390 static int bond_sethwaddr(struct net_device *bond_dev,
1391                           struct net_device *slave_dev)
1392 {
1393         pr_debug("bond_dev=%p\n", bond_dev);
1394         pr_debug("slave_dev=%p\n", slave_dev);
1395         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1396         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1397         return 0;
1398 }
1399
1400 static u32 bond_fix_features(struct net_device *dev, u32 features)
1401 {
1402         struct slave *slave;
1403         struct bonding *bond = netdev_priv(dev);
1404         u32 mask;
1405         int i;
1406
1407         read_lock(&bond->lock);
1408
1409         if (!bond->first_slave) {
1410                 /* Disable adding VLANs to empty bond. But why? --mq */
1411                 features |= NETIF_F_VLAN_CHALLENGED;
1412                 goto out;
1413         }
1414
1415         mask = features;
1416         features &= ~NETIF_F_ONE_FOR_ALL;
1417         features |= NETIF_F_ALL_FOR_ALL;
1418
1419         bond_for_each_slave(bond, slave, i) {
1420                 features = netdev_increment_features(features,
1421                                                      slave->dev->features,
1422                                                      mask);
1423         }
1424
1425 out:
1426         read_unlock(&bond->lock);
1427         return features;
1428 }
1429
1430 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_TX_OFFLOADS | \
1431                                  NETIF_F_SOFT_FEATURES | \
1432                                  NETIF_F_LRO)
1433
1434 static void bond_compute_features(struct bonding *bond)
1435 {
1436         struct slave *slave;
1437         struct net_device *bond_dev = bond->dev;
1438         u32 vlan_features = BOND_VLAN_FEATURES;
1439         unsigned short max_hard_header_len = ETH_HLEN;
1440         int i;
1441
1442         read_lock(&bond->lock);
1443
1444         if (!bond->first_slave)
1445                 goto done;
1446
1447         bond_for_each_slave(bond, slave, i) {
1448                 vlan_features = netdev_increment_features(vlan_features,
1449                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1450
1451                 if (slave->dev->hard_header_len > max_hard_header_len)
1452                         max_hard_header_len = slave->dev->hard_header_len;
1453         }
1454
1455 done:
1456         bond_dev->vlan_features = vlan_features;
1457         bond_dev->hard_header_len = max_hard_header_len;
1458
1459         read_unlock(&bond->lock);
1460
1461         netdev_change_features(bond_dev);
1462 }
1463
1464 static void bond_setup_by_slave(struct net_device *bond_dev,
1465                                 struct net_device *slave_dev)
1466 {
1467         struct bonding *bond = netdev_priv(bond_dev);
1468
1469         bond_dev->header_ops        = slave_dev->header_ops;
1470
1471         bond_dev->type              = slave_dev->type;
1472         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1473         bond_dev->addr_len          = slave_dev->addr_len;
1474
1475         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1476                 slave_dev->addr_len);
1477         bond->setup_by_slave = 1;
1478 }
1479
1480 /* On bonding slaves other than the currently active slave, suppress
1481  * duplicates except for alb non-mcast/bcast.
1482  */
1483 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1484                                             struct slave *slave,
1485                                             struct bonding *bond)
1486 {
1487         if (bond_is_slave_inactive(slave)) {
1488                 if (bond->params.mode == BOND_MODE_ALB &&
1489                     skb->pkt_type != PACKET_BROADCAST &&
1490                     skb->pkt_type != PACKET_MULTICAST)
1491                         return false;
1492                 return true;
1493         }
1494         return false;
1495 }
1496
1497 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1498 {
1499         struct sk_buff *skb = *pskb;
1500         struct slave *slave;
1501         struct bonding *bond;
1502
1503         skb = skb_share_check(skb, GFP_ATOMIC);
1504         if (unlikely(!skb))
1505                 return RX_HANDLER_CONSUMED;
1506
1507         *pskb = skb;
1508
1509         slave = bond_slave_get_rcu(skb->dev);
1510         bond = slave->bond;
1511
1512         if (bond->params.arp_interval)
1513                 slave->dev->last_rx = jiffies;
1514
1515         if (bond->recv_probe) {
1516                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1517
1518                 if (likely(nskb)) {
1519                         bond->recv_probe(nskb, bond, slave);
1520                         dev_kfree_skb(nskb);
1521                 }
1522         }
1523
1524         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1525                 return RX_HANDLER_EXACT;
1526         }
1527
1528         skb->dev = bond->dev;
1529
1530         if (bond->params.mode == BOND_MODE_ALB &&
1531             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1532             skb->pkt_type == PACKET_HOST) {
1533
1534                 if (unlikely(skb_cow_head(skb,
1535                                           skb->data - skb_mac_header(skb)))) {
1536                         kfree_skb(skb);
1537                         return RX_HANDLER_CONSUMED;
1538                 }
1539                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1540         }
1541
1542         return RX_HANDLER_ANOTHER;
1543 }
1544
1545 /* enslave device <slave> to bond device <master> */
1546 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1547 {
1548         struct bonding *bond = netdev_priv(bond_dev);
1549         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1550         struct slave *new_slave = NULL;
1551         struct netdev_hw_addr *ha;
1552         struct sockaddr addr;
1553         int link_reporting;
1554         int res = 0;
1555
1556         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1557                 slave_ops->ndo_do_ioctl == NULL) {
1558                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1559                            bond_dev->name, slave_dev->name);
1560         }
1561
1562         /* already enslaved */
1563         if (slave_dev->flags & IFF_SLAVE) {
1564                 pr_debug("Error, Device was already enslaved\n");
1565                 return -EBUSY;
1566         }
1567
1568         /* vlan challenged mutual exclusion */
1569         /* no need to lock since we're protected by rtnl_lock */
1570         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1571                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1572                 if (bond->vlgrp) {
1573                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1574                                bond_dev->name, slave_dev->name, bond_dev->name);
1575                         return -EPERM;
1576                 } else {
1577                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1578                                    bond_dev->name, slave_dev->name,
1579                                    slave_dev->name, bond_dev->name);
1580                 }
1581         } else {
1582                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1583         }
1584
1585         /*
1586          * Old ifenslave binaries are no longer supported.  These can
1587          * be identified with moderate accuracy by the state of the slave:
1588          * the current ifenslave will set the interface down prior to
1589          * enslaving it; the old ifenslave will not.
1590          */
1591         if ((slave_dev->flags & IFF_UP)) {
1592                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1593                        slave_dev->name);
1594                 res = -EPERM;
1595                 goto err_undo_flags;
1596         }
1597
1598         /* set bonding device ether type by slave - bonding netdevices are
1599          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1600          * there is a need to override some of the type dependent attribs/funcs.
1601          *
1602          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1603          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1604          */
1605         if (bond->slave_cnt == 0) {
1606                 if (bond_dev->type != slave_dev->type) {
1607                         pr_debug("%s: change device type from %d to %d\n",
1608                                  bond_dev->name,
1609                                  bond_dev->type, slave_dev->type);
1610
1611                         res = netdev_bonding_change(bond_dev,
1612                                                     NETDEV_PRE_TYPE_CHANGE);
1613                         res = notifier_to_errno(res);
1614                         if (res) {
1615                                 pr_err("%s: refused to change device type\n",
1616                                        bond_dev->name);
1617                                 res = -EBUSY;
1618                                 goto err_undo_flags;
1619                         }
1620
1621                         /* Flush unicast and multicast addresses */
1622                         dev_uc_flush(bond_dev);
1623                         dev_mc_flush(bond_dev);
1624
1625                         if (slave_dev->type != ARPHRD_ETHER)
1626                                 bond_setup_by_slave(bond_dev, slave_dev);
1627                         else
1628                                 ether_setup(bond_dev);
1629
1630                         netdev_bonding_change(bond_dev,
1631                                               NETDEV_POST_TYPE_CHANGE);
1632                 }
1633         } else if (bond_dev->type != slave_dev->type) {
1634                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1635                        slave_dev->name,
1636                        slave_dev->type, bond_dev->type);
1637                 res = -EINVAL;
1638                 goto err_undo_flags;
1639         }
1640
1641         if (slave_ops->ndo_set_mac_address == NULL) {
1642                 if (bond->slave_cnt == 0) {
1643                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1644                                    bond_dev->name);
1645                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1646                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1647                         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",
1648                                bond_dev->name);
1649                         res = -EOPNOTSUPP;
1650                         goto err_undo_flags;
1651                 }
1652         }
1653
1654         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1655
1656         /* If this is the first slave, then we need to set the master's hardware
1657          * address to be the same as the slave's. */
1658         if (is_zero_ether_addr(bond->dev->dev_addr))
1659                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1660                        slave_dev->addr_len);
1661
1662
1663         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1664         if (!new_slave) {
1665                 res = -ENOMEM;
1666                 goto err_undo_flags;
1667         }
1668
1669         /*
1670          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1671          * is set via sysfs or module option if desired.
1672          */
1673         new_slave->queue_id = 0;
1674
1675         /* Save slave's original mtu and then set it to match the bond */
1676         new_slave->original_mtu = slave_dev->mtu;
1677         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1678         if (res) {
1679                 pr_debug("Error %d calling dev_set_mtu\n", res);
1680                 goto err_free;
1681         }
1682
1683         /*
1684          * Save slave's original ("permanent") mac address for modes
1685          * that need it, and for restoring it upon release, and then
1686          * set it to the master's address
1687          */
1688         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1689
1690         if (!bond->params.fail_over_mac) {
1691                 /*
1692                  * Set slave to master's mac address.  The application already
1693                  * set the master's mac address to that of the first slave
1694                  */
1695                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1696                 addr.sa_family = slave_dev->type;
1697                 res = dev_set_mac_address(slave_dev, &addr);
1698                 if (res) {
1699                         pr_debug("Error %d calling set_mac_address\n", res);
1700                         goto err_restore_mtu;
1701                 }
1702         }
1703
1704         res = netdev_set_bond_master(slave_dev, bond_dev);
1705         if (res) {
1706                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1707                 goto err_restore_mac;
1708         }
1709
1710         /* open the slave since the application closed it */
1711         res = dev_open(slave_dev);
1712         if (res) {
1713                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1714                 goto err_unset_master;
1715         }
1716
1717         new_slave->bond = bond;
1718         new_slave->dev = slave_dev;
1719         slave_dev->priv_flags |= IFF_BONDING;
1720
1721         if (bond_is_lb(bond)) {
1722                 /* bond_alb_init_slave() must be called before all other stages since
1723                  * it might fail and we do not want to have to undo everything
1724                  */
1725                 res = bond_alb_init_slave(bond, new_slave);
1726                 if (res)
1727                         goto err_close;
1728         }
1729
1730         /* If the mode USES_PRIMARY, then the new slave gets the
1731          * master's promisc (and mc) settings only if it becomes the
1732          * curr_active_slave, and that is taken care of later when calling
1733          * bond_change_active()
1734          */
1735         if (!USES_PRIMARY(bond->params.mode)) {
1736                 /* set promiscuity level to new slave */
1737                 if (bond_dev->flags & IFF_PROMISC) {
1738                         res = dev_set_promiscuity(slave_dev, 1);
1739                         if (res)
1740                                 goto err_close;
1741                 }
1742
1743                 /* set allmulti level to new slave */
1744                 if (bond_dev->flags & IFF_ALLMULTI) {
1745                         res = dev_set_allmulti(slave_dev, 1);
1746                         if (res)
1747                                 goto err_close;
1748                 }
1749
1750                 netif_addr_lock_bh(bond_dev);
1751                 /* upload master's mc_list to new slave */
1752                 netdev_for_each_mc_addr(ha, bond_dev)
1753                         dev_mc_add(slave_dev, ha->addr);
1754                 netif_addr_unlock_bh(bond_dev);
1755         }
1756
1757         if (bond->params.mode == BOND_MODE_8023AD) {
1758                 /* add lacpdu mc addr to mc list */
1759                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1760
1761                 dev_mc_add(slave_dev, lacpdu_multicast);
1762         }
1763
1764         bond_add_vlans_on_slave(bond, slave_dev);
1765
1766         write_lock_bh(&bond->lock);
1767
1768         bond_attach_slave(bond, new_slave);
1769
1770         new_slave->delay = 0;
1771         new_slave->link_failure_count = 0;
1772
1773         write_unlock_bh(&bond->lock);
1774
1775         bond_compute_features(bond);
1776
1777         read_lock(&bond->lock);
1778
1779         new_slave->last_arp_rx = jiffies;
1780
1781         if (bond->params.miimon && !bond->params.use_carrier) {
1782                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1783
1784                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1785                         /*
1786                          * miimon is set but a bonded network driver
1787                          * does not support ETHTOOL/MII and
1788                          * arp_interval is not set.  Note: if
1789                          * use_carrier is enabled, we will never go
1790                          * here (because netif_carrier is always
1791                          * supported); thus, we don't need to change
1792                          * the messages for netif_carrier.
1793                          */
1794                         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",
1795                                bond_dev->name, slave_dev->name);
1796                 } else if (link_reporting == -1) {
1797                         /* unable get link status using mii/ethtool */
1798                         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",
1799                                    bond_dev->name, slave_dev->name);
1800                 }
1801         }
1802
1803         /* check for initial state */
1804         if (!bond->params.miimon ||
1805             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1806                 if (bond->params.updelay) {
1807                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1808                         new_slave->link  = BOND_LINK_BACK;
1809                         new_slave->delay = bond->params.updelay;
1810                 } else {
1811                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1812                         new_slave->link  = BOND_LINK_UP;
1813                 }
1814                 new_slave->jiffies = jiffies;
1815         } else {
1816                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1817                 new_slave->link  = BOND_LINK_DOWN;
1818         }
1819
1820         if (bond_update_speed_duplex(new_slave) &&
1821             (new_slave->link != BOND_LINK_DOWN)) {
1822                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1823                            bond_dev->name, new_slave->dev->name);
1824
1825                 if (bond->params.mode == BOND_MODE_8023AD) {
1826                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1827                                    bond_dev->name);
1828                 }
1829         }
1830
1831         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1832                 /* if there is a primary slave, remember it */
1833                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1834                         bond->primary_slave = new_slave;
1835                         bond->force_primary = true;
1836                 }
1837         }
1838
1839         write_lock_bh(&bond->curr_slave_lock);
1840
1841         switch (bond->params.mode) {
1842         case BOND_MODE_ACTIVEBACKUP:
1843                 bond_set_slave_inactive_flags(new_slave);
1844                 bond_select_active_slave(bond);
1845                 break;
1846         case BOND_MODE_8023AD:
1847                 /* in 802.3ad mode, the internal mechanism
1848                  * will activate the slaves in the selected
1849                  * aggregator
1850                  */
1851                 bond_set_slave_inactive_flags(new_slave);
1852                 /* if this is the first slave */
1853                 if (bond->slave_cnt == 1) {
1854                         SLAVE_AD_INFO(new_slave).id = 1;
1855                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1856                          * can be called only after the mac address of the bond is set
1857                          */
1858                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1859                                             bond->params.lacp_fast);
1860                 } else {
1861                         SLAVE_AD_INFO(new_slave).id =
1862                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1863                 }
1864
1865                 bond_3ad_bind_slave(new_slave);
1866                 break;
1867         case BOND_MODE_TLB:
1868         case BOND_MODE_ALB:
1869                 bond_set_active_slave(new_slave);
1870                 bond_set_slave_inactive_flags(new_slave);
1871                 bond_select_active_slave(bond);
1872                 break;
1873         default:
1874                 pr_debug("This slave is always active in trunk mode\n");
1875
1876                 /* always active in trunk mode */
1877                 bond_set_active_slave(new_slave);
1878
1879                 /* In trunking mode there is little meaning to curr_active_slave
1880                  * anyway (it holds no special properties of the bond device),
1881                  * so we can change it without calling change_active_interface()
1882                  */
1883                 if (!bond->curr_active_slave)
1884                         bond->curr_active_slave = new_slave;
1885
1886                 break;
1887         } /* switch(bond_mode) */
1888
1889         write_unlock_bh(&bond->curr_slave_lock);
1890
1891         bond_set_carrier(bond);
1892
1893 #ifdef CONFIG_NET_POLL_CONTROLLER
1894         slave_dev->npinfo = bond_netpoll_info(bond);
1895         if (slave_dev->npinfo) {
1896                 if (slave_enable_netpoll(new_slave)) {
1897                         read_unlock(&bond->lock);
1898                         pr_info("Error, %s: master_dev is using netpoll, "
1899                                  "but new slave device does not support netpoll.\n",
1900                                  bond_dev->name);
1901                         res = -EBUSY;
1902                         goto err_close;
1903                 }
1904         }
1905 #endif
1906
1907         read_unlock(&bond->lock);
1908
1909         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1910         if (res)
1911                 goto err_close;
1912
1913         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1914                                          new_slave);
1915         if (res) {
1916                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1917                 goto err_dest_symlinks;
1918         }
1919
1920         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1921                 bond_dev->name, slave_dev->name,
1922                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1923                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1924
1925         /* enslave is successful */
1926         return 0;
1927
1928 /* Undo stages on error */
1929 err_dest_symlinks:
1930         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1931
1932 err_close:
1933         dev_close(slave_dev);
1934
1935 err_unset_master:
1936         netdev_set_bond_master(slave_dev, NULL);
1937
1938 err_restore_mac:
1939         if (!bond->params.fail_over_mac) {
1940                 /* XXX TODO - fom follow mode needs to change master's
1941                  * MAC if this slave's MAC is in use by the bond, or at
1942                  * least print a warning.
1943                  */
1944                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1945                 addr.sa_family = slave_dev->type;
1946                 dev_set_mac_address(slave_dev, &addr);
1947         }
1948
1949 err_restore_mtu:
1950         dev_set_mtu(slave_dev, new_slave->original_mtu);
1951
1952 err_free:
1953         kfree(new_slave);
1954
1955 err_undo_flags:
1956         bond_compute_features(bond);
1957
1958         return res;
1959 }
1960
1961 /*
1962  * Try to release the slave device <slave> from the bond device <master>
1963  * It is legal to access curr_active_slave without a lock because all the function
1964  * is write-locked.
1965  *
1966  * The rules for slave state should be:
1967  *   for Active/Backup:
1968  *     Active stays on all backups go down
1969  *   for Bonded connections:
1970  *     The first up interface should be left on and all others downed.
1971  */
1972 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1973 {
1974         struct bonding *bond = netdev_priv(bond_dev);
1975         struct slave *slave, *oldcurrent;
1976         struct sockaddr addr;
1977         u32 old_features = bond_dev->features;
1978
1979         /* slave is not a slave or master is not master of this slave */
1980         if (!(slave_dev->flags & IFF_SLAVE) ||
1981             (slave_dev->master != bond_dev)) {
1982                 pr_err("%s: Error: cannot release %s.\n",
1983                        bond_dev->name, slave_dev->name);
1984                 return -EINVAL;
1985         }
1986
1987         block_netpoll_tx();
1988         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1989         write_lock_bh(&bond->lock);
1990
1991         slave = bond_get_slave_by_dev(bond, slave_dev);
1992         if (!slave) {
1993                 /* not a slave of this bond */
1994                 pr_info("%s: %s not enslaved\n",
1995                         bond_dev->name, slave_dev->name);
1996                 write_unlock_bh(&bond->lock);
1997                 unblock_netpoll_tx();
1998                 return -EINVAL;
1999         }
2000
2001         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2002          * for this slave anymore.
2003          */
2004         netdev_rx_handler_unregister(slave_dev);
2005         write_unlock_bh(&bond->lock);
2006         synchronize_net();
2007         write_lock_bh(&bond->lock);
2008
2009         if (!bond->params.fail_over_mac) {
2010                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
2011                     bond->slave_cnt > 1)
2012                         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",
2013                                    bond_dev->name, slave_dev->name,
2014                                    slave->perm_hwaddr,
2015                                    bond_dev->name, slave_dev->name);
2016         }
2017
2018         /* Inform AD package of unbinding of slave. */
2019         if (bond->params.mode == BOND_MODE_8023AD) {
2020                 /* must be called before the slave is
2021                  * detached from the list
2022                  */
2023                 bond_3ad_unbind_slave(slave);
2024         }
2025
2026         pr_info("%s: releasing %s interface %s\n",
2027                 bond_dev->name,
2028                 bond_is_active_slave(slave) ? "active" : "backup",
2029                 slave_dev->name);
2030
2031         oldcurrent = bond->curr_active_slave;
2032
2033         bond->current_arp_slave = NULL;
2034
2035         /* release the slave from its bond */
2036         bond_detach_slave(bond, slave);
2037
2038         if (bond->primary_slave == slave)
2039                 bond->primary_slave = NULL;
2040
2041         if (oldcurrent == slave)
2042                 bond_change_active_slave(bond, NULL);
2043
2044         if (bond_is_lb(bond)) {
2045                 /* Must be called only after the slave has been
2046                  * detached from the list and the curr_active_slave
2047                  * has been cleared (if our_slave == old_current),
2048                  * but before a new active slave is selected.
2049                  */
2050                 write_unlock_bh(&bond->lock);
2051                 bond_alb_deinit_slave(bond, slave);
2052                 write_lock_bh(&bond->lock);
2053         }
2054
2055         if (oldcurrent == slave) {
2056                 /*
2057                  * Note that we hold RTNL over this sequence, so there
2058                  * is no concern that another slave add/remove event
2059                  * will interfere.
2060                  */
2061                 write_unlock_bh(&bond->lock);
2062                 read_lock(&bond->lock);
2063                 write_lock_bh(&bond->curr_slave_lock);
2064
2065                 bond_select_active_slave(bond);
2066
2067                 write_unlock_bh(&bond->curr_slave_lock);
2068                 read_unlock(&bond->lock);
2069                 write_lock_bh(&bond->lock);
2070         }
2071
2072         if (bond->slave_cnt == 0) {
2073                 bond_set_carrier(bond);
2074
2075                 /* if the last slave was removed, zero the mac address
2076                  * of the master so it will be set by the application
2077                  * to the mac address of the first slave
2078                  */
2079                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2080
2081                 if (bond->vlgrp) {
2082                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2083                                    bond_dev->name, bond_dev->name);
2084                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2085                                    bond_dev->name);
2086                 }
2087         }
2088
2089         write_unlock_bh(&bond->lock);
2090         unblock_netpoll_tx();
2091
2092         bond_compute_features(bond);
2093         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2094             (old_features & NETIF_F_VLAN_CHALLENGED))
2095                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2096                         bond_dev->name, slave_dev->name, bond_dev->name);
2097
2098         /* must do this from outside any spinlocks */
2099         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2100
2101         bond_del_vlans_from_slave(bond, slave_dev);
2102
2103         /* If the mode USES_PRIMARY, then we should only remove its
2104          * promisc and mc settings if it was the curr_active_slave, but that was
2105          * already taken care of above when we detached the slave
2106          */
2107         if (!USES_PRIMARY(bond->params.mode)) {
2108                 /* unset promiscuity level from slave */
2109                 if (bond_dev->flags & IFF_PROMISC)
2110                         dev_set_promiscuity(slave_dev, -1);
2111
2112                 /* unset allmulti level from slave */
2113                 if (bond_dev->flags & IFF_ALLMULTI)
2114                         dev_set_allmulti(slave_dev, -1);
2115
2116                 /* flush master's mc_list from slave */
2117                 netif_addr_lock_bh(bond_dev);
2118                 bond_mc_list_flush(bond_dev, slave_dev);
2119                 netif_addr_unlock_bh(bond_dev);
2120         }
2121
2122         netdev_set_bond_master(slave_dev, NULL);
2123
2124         slave_disable_netpoll(slave);
2125
2126         /* close slave before restoring its mac address */
2127         dev_close(slave_dev);
2128
2129         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2130                 /* restore original ("permanent") mac address */
2131                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2132                 addr.sa_family = slave_dev->type;
2133                 dev_set_mac_address(slave_dev, &addr);
2134         }
2135
2136         dev_set_mtu(slave_dev, slave->original_mtu);
2137
2138         slave_dev->priv_flags &= ~IFF_BONDING;
2139
2140         kfree(slave);
2141
2142         return 0;  /* deletion OK */
2143 }
2144
2145 /*
2146 * First release a slave and then destroy the bond if no more slaves are left.
2147 * Must be under rtnl_lock when this function is called.
2148 */
2149 static int  bond_release_and_destroy(struct net_device *bond_dev,
2150                                      struct net_device *slave_dev)
2151 {
2152         struct bonding *bond = netdev_priv(bond_dev);
2153         int ret;
2154
2155         ret = bond_release(bond_dev, slave_dev);
2156         if ((ret == 0) && (bond->slave_cnt == 0)) {
2157                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2158                 pr_info("%s: destroying bond %s.\n",
2159                         bond_dev->name, bond_dev->name);
2160                 unregister_netdevice(bond_dev);
2161         }
2162         return ret;
2163 }
2164
2165 /*
2166  * This function releases all slaves.
2167  */
2168 static int bond_release_all(struct net_device *bond_dev)
2169 {
2170         struct bonding *bond = netdev_priv(bond_dev);
2171         struct slave *slave;
2172         struct net_device *slave_dev;
2173         struct sockaddr addr;
2174
2175         write_lock_bh(&bond->lock);
2176
2177         netif_carrier_off(bond_dev);
2178
2179         if (bond->slave_cnt == 0)
2180                 goto out;
2181
2182         bond->current_arp_slave = NULL;
2183         bond->primary_slave = NULL;
2184         bond_change_active_slave(bond, NULL);
2185
2186         while ((slave = bond->first_slave) != NULL) {
2187                 /* Inform AD package of unbinding of slave
2188                  * before slave is detached from the list.
2189                  */
2190                 if (bond->params.mode == BOND_MODE_8023AD)
2191                         bond_3ad_unbind_slave(slave);
2192
2193                 slave_dev = slave->dev;
2194                 bond_detach_slave(bond, slave);
2195
2196                 /* now that the slave is detached, unlock and perform
2197                  * all the undo steps that should not be called from
2198                  * within a lock.
2199                  */
2200                 write_unlock_bh(&bond->lock);
2201
2202                 /* unregister rx_handler early so bond_handle_frame wouldn't
2203                  * be called for this slave anymore.
2204                  */
2205                 netdev_rx_handler_unregister(slave_dev);
2206                 synchronize_net();
2207
2208                 if (bond_is_lb(bond)) {
2209                         /* must be called only after the slave
2210                          * has been detached from the list
2211                          */
2212                         bond_alb_deinit_slave(bond, slave);
2213                 }
2214
2215                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2216                 bond_del_vlans_from_slave(bond, slave_dev);
2217
2218                 /* If the mode USES_PRIMARY, then we should only remove its
2219                  * promisc and mc settings if it was the curr_active_slave, but that was
2220                  * already taken care of above when we detached the slave
2221                  */
2222                 if (!USES_PRIMARY(bond->params.mode)) {
2223                         /* unset promiscuity level from slave */
2224                         if (bond_dev->flags & IFF_PROMISC)
2225                                 dev_set_promiscuity(slave_dev, -1);
2226
2227                         /* unset allmulti level from slave */
2228                         if (bond_dev->flags & IFF_ALLMULTI)
2229                                 dev_set_allmulti(slave_dev, -1);
2230
2231                         /* flush master's mc_list from slave */
2232                         netif_addr_lock_bh(bond_dev);
2233                         bond_mc_list_flush(bond_dev, slave_dev);
2234                         netif_addr_unlock_bh(bond_dev);
2235                 }
2236
2237                 netdev_set_bond_master(slave_dev, NULL);
2238
2239                 slave_disable_netpoll(slave);
2240
2241                 /* close slave before restoring its mac address */
2242                 dev_close(slave_dev);
2243
2244                 if (!bond->params.fail_over_mac) {
2245                         /* restore original ("permanent") mac address*/
2246                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2247                         addr.sa_family = slave_dev->type;
2248                         dev_set_mac_address(slave_dev, &addr);
2249                 }
2250
2251                 kfree(slave);
2252
2253                 /* re-acquire the lock before getting the next slave */
2254                 write_lock_bh(&bond->lock);
2255         }
2256
2257         /* zero the mac address of the master so it will be
2258          * set by the application to the mac address of the
2259          * first slave
2260          */
2261         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2262
2263         if (bond->vlgrp) {
2264                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2265                            bond_dev->name, bond_dev->name);
2266                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2267                            bond_dev->name);
2268         }
2269
2270         pr_info("%s: released all slaves\n", bond_dev->name);
2271
2272 out:
2273         write_unlock_bh(&bond->lock);
2274
2275         bond_compute_features(bond);
2276
2277         return 0;
2278 }
2279
2280 /*
2281  * This function changes the active slave to slave <slave_dev>.
2282  * It returns -EINVAL in the following cases.
2283  *  - <slave_dev> is not found in the list.
2284  *  - There is not active slave now.
2285  *  - <slave_dev> is already active.
2286  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2287  *  - <slave_dev> is not running.
2288  * In these cases, this function does nothing.
2289  * In the other cases, current_slave pointer is changed and 0 is returned.
2290  */
2291 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2292 {
2293         struct bonding *bond = netdev_priv(bond_dev);
2294         struct slave *old_active = NULL;
2295         struct slave *new_active = NULL;
2296         int res = 0;
2297
2298         if (!USES_PRIMARY(bond->params.mode))
2299                 return -EINVAL;
2300
2301         /* Verify that master_dev is indeed the master of slave_dev */
2302         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2303                 return -EINVAL;
2304
2305         read_lock(&bond->lock);
2306
2307         read_lock(&bond->curr_slave_lock);
2308         old_active = bond->curr_active_slave;
2309         read_unlock(&bond->curr_slave_lock);
2310
2311         new_active = bond_get_slave_by_dev(bond, slave_dev);
2312
2313         /*
2314          * Changing to the current active: do nothing; return success.
2315          */
2316         if (new_active && (new_active == old_active)) {
2317                 read_unlock(&bond->lock);
2318                 return 0;
2319         }
2320
2321         if ((new_active) &&
2322             (old_active) &&
2323             (new_active->link == BOND_LINK_UP) &&
2324             IS_UP(new_active->dev)) {
2325                 block_netpoll_tx();
2326                 write_lock_bh(&bond->curr_slave_lock);
2327                 bond_change_active_slave(bond, new_active);
2328                 write_unlock_bh(&bond->curr_slave_lock);
2329                 unblock_netpoll_tx();
2330         } else
2331                 res = -EINVAL;
2332
2333         read_unlock(&bond->lock);
2334
2335         return res;
2336 }
2337
2338 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2339 {
2340         struct bonding *bond = netdev_priv(bond_dev);
2341
2342         info->bond_mode = bond->params.mode;
2343         info->miimon = bond->params.miimon;
2344
2345         read_lock(&bond->lock);
2346         info->num_slaves = bond->slave_cnt;
2347         read_unlock(&bond->lock);
2348
2349         return 0;
2350 }
2351
2352 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2353 {
2354         struct bonding *bond = netdev_priv(bond_dev);
2355         struct slave *slave;
2356         int i, res = -ENODEV;
2357
2358         read_lock(&bond->lock);
2359
2360         bond_for_each_slave(bond, slave, i) {
2361                 if (i == (int)info->slave_id) {
2362                         res = 0;
2363                         strcpy(info->slave_name, slave->dev->name);
2364                         info->link = slave->link;
2365                         info->state = bond_slave_state(slave);
2366                         info->link_failure_count = slave->link_failure_count;
2367                         break;
2368                 }
2369         }
2370
2371         read_unlock(&bond->lock);
2372
2373         return res;
2374 }
2375
2376 /*-------------------------------- Monitoring -------------------------------*/
2377
2378
2379 static int bond_miimon_inspect(struct bonding *bond)
2380 {
2381         struct slave *slave;
2382         int i, link_state, commit = 0;
2383         bool ignore_updelay;
2384
2385         ignore_updelay = !bond->curr_active_slave ? true : false;
2386
2387         bond_for_each_slave(bond, slave, i) {
2388                 slave->new_link = BOND_LINK_NOCHANGE;
2389
2390                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2391
2392                 switch (slave->link) {
2393                 case BOND_LINK_UP:
2394                         if (link_state)
2395                                 continue;
2396
2397                         slave->link = BOND_LINK_FAIL;
2398                         slave->delay = bond->params.downdelay;
2399                         if (slave->delay) {
2400                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2401                                         bond->dev->name,
2402                                         (bond->params.mode ==
2403                                          BOND_MODE_ACTIVEBACKUP) ?
2404                                         (bond_is_active_slave(slave) ?
2405                                          "active " : "backup ") : "",
2406                                         slave->dev->name,
2407                                         bond->params.downdelay * bond->params.miimon);
2408                         }
2409                         /*FALLTHRU*/
2410                 case BOND_LINK_FAIL:
2411                         if (link_state) {
2412                                 /*
2413                                  * recovered before downdelay expired
2414                                  */
2415                                 slave->link = BOND_LINK_UP;
2416                                 slave->jiffies = jiffies;
2417                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2418                                         bond->dev->name,
2419                                         (bond->params.downdelay - slave->delay) *
2420                                         bond->params.miimon,
2421                                         slave->dev->name);
2422                                 continue;
2423                         }
2424
2425                         if (slave->delay <= 0) {
2426                                 slave->new_link = BOND_LINK_DOWN;
2427                                 commit++;
2428                                 continue;
2429                         }
2430
2431                         slave->delay--;
2432                         break;
2433
2434                 case BOND_LINK_DOWN:
2435                         if (!link_state)
2436                                 continue;
2437
2438                         slave->link = BOND_LINK_BACK;
2439                         slave->delay = bond->params.updelay;
2440
2441                         if (slave->delay) {
2442                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2443                                         bond->dev->name, slave->dev->name,
2444                                         ignore_updelay ? 0 :
2445                                         bond->params.updelay *
2446                                         bond->params.miimon);
2447                         }
2448                         /*FALLTHRU*/
2449                 case BOND_LINK_BACK:
2450                         if (!link_state) {
2451                                 slave->link = BOND_LINK_DOWN;
2452                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2453                                         bond->dev->name,
2454                                         (bond->params.updelay - slave->delay) *
2455                                         bond->params.miimon,
2456                                         slave->dev->name);
2457
2458                                 continue;
2459                         }
2460
2461                         if (ignore_updelay)
2462                                 slave->delay = 0;
2463
2464                         if (slave->delay <= 0) {
2465                                 slave->new_link = BOND_LINK_UP;
2466                                 commit++;
2467                                 ignore_updelay = false;
2468                                 continue;
2469                         }
2470
2471                         slave->delay--;
2472                         break;
2473                 }
2474         }
2475
2476         return commit;
2477 }
2478
2479 static void bond_miimon_commit(struct bonding *bond)
2480 {
2481         struct slave *slave;
2482         int i;
2483
2484         bond_for_each_slave(bond, slave, i) {
2485                 switch (slave->new_link) {
2486                 case BOND_LINK_NOCHANGE:
2487                         continue;
2488
2489                 case BOND_LINK_UP:
2490                         slave->link = BOND_LINK_UP;
2491                         slave->jiffies = jiffies;
2492
2493                         if (bond->params.mode == BOND_MODE_8023AD) {
2494                                 /* prevent it from being the active one */
2495                                 bond_set_backup_slave(slave);
2496                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2497                                 /* make it immediately active */
2498                                 bond_set_active_slave(slave);
2499                         } else if (slave != bond->primary_slave) {
2500                                 /* prevent it from being the active one */
2501                                 bond_set_backup_slave(slave);
2502                         }
2503
2504                         bond_update_speed_duplex(slave);
2505
2506                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2507                                 bond->dev->name, slave->dev->name,
2508                                 slave->speed, slave->duplex ? "full" : "half");
2509
2510                         /* notify ad that the link status has changed */
2511                         if (bond->params.mode == BOND_MODE_8023AD)
2512                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2513
2514                         if (bond_is_lb(bond))
2515                                 bond_alb_handle_link_change(bond, slave,
2516                                                             BOND_LINK_UP);
2517
2518                         if (!bond->curr_active_slave ||
2519                             (slave == bond->primary_slave))
2520                                 goto do_failover;
2521
2522                         continue;
2523
2524                 case BOND_LINK_DOWN:
2525                         if (slave->link_failure_count < UINT_MAX)
2526                                 slave->link_failure_count++;
2527
2528                         slave->link = BOND_LINK_DOWN;
2529
2530                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2531                             bond->params.mode == BOND_MODE_8023AD)
2532                                 bond_set_slave_inactive_flags(slave);
2533
2534                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2535                                 bond->dev->name, slave->dev->name);
2536
2537                         if (bond->params.mode == BOND_MODE_8023AD)
2538                                 bond_3ad_handle_link_change(slave,
2539                                                             BOND_LINK_DOWN);
2540
2541                         if (bond_is_lb(bond))
2542                                 bond_alb_handle_link_change(bond, slave,
2543                                                             BOND_LINK_DOWN);
2544
2545                         if (slave == bond->curr_active_slave)
2546                                 goto do_failover;
2547
2548                         continue;
2549
2550                 default:
2551                         pr_err("%s: invalid new link %d on slave %s\n",
2552                                bond->dev->name, slave->new_link,
2553                                slave->dev->name);
2554                         slave->new_link = BOND_LINK_NOCHANGE;
2555
2556                         continue;
2557                 }
2558
2559 do_failover:
2560                 ASSERT_RTNL();
2561                 block_netpoll_tx();
2562                 write_lock_bh(&bond->curr_slave_lock);
2563                 bond_select_active_slave(bond);
2564                 write_unlock_bh(&bond->curr_slave_lock);
2565                 unblock_netpoll_tx();
2566         }
2567
2568         bond_set_carrier(bond);
2569 }
2570
2571 /*
2572  * bond_mii_monitor
2573  *
2574  * Really a wrapper that splits the mii monitor into two phases: an
2575  * inspection, then (if inspection indicates something needs to be done)
2576  * an acquisition of appropriate locks followed by a commit phase to
2577  * implement whatever link state changes are indicated.
2578  */
2579 void bond_mii_monitor(struct work_struct *work)
2580 {
2581         struct bonding *bond = container_of(work, struct bonding,
2582                                             mii_work.work);
2583         bool should_notify_peers = false;
2584
2585         read_lock(&bond->lock);
2586         if (bond->kill_timers)
2587                 goto out;
2588
2589         if (bond->slave_cnt == 0)
2590                 goto re_arm;
2591
2592         should_notify_peers = bond_should_notify_peers(bond);
2593
2594         if (bond_miimon_inspect(bond)) {
2595                 read_unlock(&bond->lock);
2596                 rtnl_lock();
2597                 read_lock(&bond->lock);
2598
2599                 bond_miimon_commit(bond);
2600
2601                 read_unlock(&bond->lock);
2602                 rtnl_unlock();  /* might sleep, hold no other locks */
2603                 read_lock(&bond->lock);
2604         }
2605
2606 re_arm:
2607         if (bond->params.miimon)
2608                 queue_delayed_work(bond->wq, &bond->mii_work,
2609                                    msecs_to_jiffies(bond->params.miimon));
2610 out:
2611         read_unlock(&bond->lock);
2612
2613         if (should_notify_peers) {
2614                 rtnl_lock();
2615                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2616                 rtnl_unlock();
2617         }
2618 }
2619
2620 static __be32 bond_glean_dev_ip(struct net_device *dev)
2621 {
2622         struct in_device *idev;
2623         struct in_ifaddr *ifa;
2624         __be32 addr = 0;
2625
2626         if (!dev)
2627                 return 0;
2628
2629         rcu_read_lock();
2630         idev = __in_dev_get_rcu(dev);
2631         if (!idev)
2632                 goto out;
2633
2634         ifa = idev->ifa_list;
2635         if (!ifa)
2636                 goto out;
2637
2638         addr = ifa->ifa_local;
2639 out:
2640         rcu_read_unlock();
2641         return addr;
2642 }
2643
2644 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2645 {
2646         struct vlan_entry *vlan;
2647
2648         if (ip == bond->master_ip)
2649                 return 1;
2650
2651         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2652                 if (ip == vlan->vlan_ip)
2653                         return 1;
2654         }
2655
2656         return 0;
2657 }
2658
2659 /*
2660  * We go to the (large) trouble of VLAN tagging ARP frames because
2661  * switches in VLAN mode (especially if ports are configured as
2662  * "native" to a VLAN) might not pass non-tagged frames.
2663  */
2664 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2665 {
2666         struct sk_buff *skb;
2667
2668         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2669                  slave_dev->name, dest_ip, src_ip, vlan_id);
2670
2671         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2672                          NULL, slave_dev->dev_addr, NULL);
2673
2674         if (!skb) {
2675                 pr_err("ARP packet allocation failed\n");
2676                 return;
2677         }
2678         if (vlan_id) {
2679                 skb = vlan_put_tag(skb, vlan_id);
2680                 if (!skb) {
2681                         pr_err("failed to insert VLAN tag\n");
2682                         return;
2683                 }
2684         }
2685         arp_xmit(skb);
2686 }
2687
2688
2689 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2690 {
2691         int i, vlan_id;
2692         __be32 *targets = bond->params.arp_targets;
2693         struct vlan_entry *vlan;
2694         struct net_device *vlan_dev;
2695         struct rtable *rt;
2696
2697         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2698                 if (!targets[i])
2699                         break;
2700                 pr_debug("basa: target %x\n", targets[i]);
2701                 if (!bond->vlgrp) {
2702                         pr_debug("basa: empty vlan: arp_send\n");
2703                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2704                                       bond->master_ip, 0);
2705                         continue;
2706                 }
2707
2708                 /*
2709                  * If VLANs are configured, we do a route lookup to
2710                  * determine which VLAN interface would be used, so we
2711                  * can tag the ARP with the proper VLAN tag.
2712                  */
2713                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2714                                      RTO_ONLINK, 0);
2715                 if (IS_ERR(rt)) {
2716                         if (net_ratelimit()) {
2717                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2718                                            bond->dev->name, &targets[i]);
2719                         }
2720                         continue;
2721                 }
2722
2723                 /*
2724                  * This target is not on a VLAN
2725                  */
2726                 if (rt->dst.dev == bond->dev) {
2727                         ip_rt_put(rt);
2728                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2729                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2730                                       bond->master_ip, 0);
2731                         continue;
2732                 }
2733
2734                 vlan_id = 0;
2735                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2736                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2737                         if (vlan_dev == rt->dst.dev) {
2738                                 vlan_id = vlan->vlan_id;
2739                                 pr_debug("basa: vlan match on %s %d\n",
2740                                        vlan_dev->name, vlan_id);
2741                                 break;
2742                         }
2743                 }
2744
2745                 if (vlan_id) {
2746                         ip_rt_put(rt);
2747                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2748                                       vlan->vlan_ip, vlan_id);
2749                         continue;
2750                 }
2751
2752                 if (net_ratelimit()) {
2753                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2754                                    bond->dev->name, &targets[i],
2755                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2756                 }
2757                 ip_rt_put(rt);
2758         }
2759 }
2760
2761 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2762 {
2763         int i;
2764         __be32 *targets = bond->params.arp_targets;
2765
2766         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2767                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2768                          &sip, &tip, i, &targets[i],
2769                          bond_has_this_ip(bond, tip));
2770                 if (sip == targets[i]) {
2771                         if (bond_has_this_ip(bond, tip))
2772                                 slave->last_arp_rx = jiffies;
2773                         return;
2774                 }
2775         }
2776 }
2777
2778 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2779                          struct slave *slave)
2780 {
2781         struct arphdr *arp;
2782         unsigned char *arp_ptr;
2783         __be32 sip, tip;
2784
2785         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2786                 return;
2787
2788         read_lock(&bond->lock);
2789
2790         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2791                  bond->dev->name, skb->dev->name);
2792
2793         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2794                 goto out_unlock;
2795
2796         arp = arp_hdr(skb);
2797         if (arp->ar_hln != bond->dev->addr_len ||
2798             skb->pkt_type == PACKET_OTHERHOST ||
2799             skb->pkt_type == PACKET_LOOPBACK ||
2800             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2801             arp->ar_pro != htons(ETH_P_IP) ||
2802             arp->ar_pln != 4)
2803                 goto out_unlock;
2804
2805         arp_ptr = (unsigned char *)(arp + 1);
2806         arp_ptr += bond->dev->addr_len;
2807         memcpy(&sip, arp_ptr, 4);
2808         arp_ptr += 4 + bond->dev->addr_len;
2809         memcpy(&tip, arp_ptr, 4);
2810
2811         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2812                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2813                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2814                  &sip, &tip);
2815
2816         /*
2817          * Backup slaves won't see the ARP reply, but do come through
2818          * here for each ARP probe (so we swap the sip/tip to validate
2819          * the probe).  In a "redundant switch, common router" type of
2820          * configuration, the ARP probe will (hopefully) travel from
2821          * the active, through one switch, the router, then the other
2822          * switch before reaching the backup.
2823          */
2824         if (bond_is_active_slave(slave))
2825                 bond_validate_arp(bond, slave, sip, tip);
2826         else
2827                 bond_validate_arp(bond, slave, tip, sip);
2828
2829 out_unlock:
2830         read_unlock(&bond->lock);
2831 }
2832
2833 /*
2834  * this function is called regularly to monitor each slave's link
2835  * ensuring that traffic is being sent and received when arp monitoring
2836  * is used in load-balancing mode. if the adapter has been dormant, then an
2837  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2838  * arp monitoring in active backup mode.
2839  */
2840 void bond_loadbalance_arp_mon(struct work_struct *work)
2841 {
2842         struct bonding *bond = container_of(work, struct bonding,
2843                                             arp_work.work);
2844         struct slave *slave, *oldcurrent;
2845         int do_failover = 0;
2846         int delta_in_ticks;
2847         int i;
2848
2849         read_lock(&bond->lock);
2850
2851         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2852
2853         if (bond->kill_timers)
2854                 goto out;
2855
2856         if (bond->slave_cnt == 0)
2857                 goto re_arm;
2858
2859         read_lock(&bond->curr_slave_lock);
2860         oldcurrent = bond->curr_active_slave;
2861         read_unlock(&bond->curr_slave_lock);
2862
2863         /* see if any of the previous devices are up now (i.e. they have
2864          * xmt and rcv traffic). the curr_active_slave does not come into
2865          * the picture unless it is null. also, slave->jiffies is not needed
2866          * here because we send an arp on each slave and give a slave as
2867          * long as it needs to get the tx/rx within the delta.
2868          * TODO: what about up/down delay in arp mode? it wasn't here before
2869          *       so it can wait
2870          */
2871         bond_for_each_slave(bond, slave, i) {
2872                 unsigned long trans_start = dev_trans_start(slave->dev);
2873
2874                 if (slave->link != BOND_LINK_UP) {
2875                         if (time_in_range(jiffies,
2876                                 trans_start - delta_in_ticks,
2877                                 trans_start + delta_in_ticks) &&
2878                             time_in_range(jiffies,
2879                                 slave->dev->last_rx - delta_in_ticks,
2880                                 slave->dev->last_rx + delta_in_ticks)) {
2881
2882                                 slave->link  = BOND_LINK_UP;
2883                                 bond_set_active_slave(slave);
2884
2885                                 /* primary_slave has no meaning in round-robin
2886                                  * mode. the window of a slave being up and
2887                                  * curr_active_slave being null after enslaving
2888                                  * is closed.
2889                                  */
2890                                 if (!oldcurrent) {
2891                                         pr_info("%s: link status definitely up for interface %s, ",
2892                                                 bond->dev->name,
2893                                                 slave->dev->name);
2894                                         do_failover = 1;
2895                                 } else {
2896                                         pr_info("%s: interface %s is now up\n",
2897                                                 bond->dev->name,
2898                                                 slave->dev->name);
2899                                 }
2900                         }
2901                 } else {
2902                         /* slave->link == BOND_LINK_UP */
2903
2904                         /* not all switches will respond to an arp request
2905                          * when the source ip is 0, so don't take the link down
2906                          * if we don't know our ip yet
2907                          */
2908                         if (!time_in_range(jiffies,
2909                                 trans_start - delta_in_ticks,
2910                                 trans_start + 2 * delta_in_ticks) ||
2911                             !time_in_range(jiffies,
2912                                 slave->dev->last_rx - delta_in_ticks,
2913                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2914
2915                                 slave->link  = BOND_LINK_DOWN;
2916                                 bond_set_backup_slave(slave);
2917
2918                                 if (slave->link_failure_count < UINT_MAX)
2919                                         slave->link_failure_count++;
2920
2921                                 pr_info("%s: interface %s is now down.\n",
2922                                         bond->dev->name,
2923                                         slave->dev->name);
2924
2925                                 if (slave == oldcurrent)
2926                                         do_failover = 1;
2927                         }
2928                 }
2929
2930                 /* note: if switch is in round-robin mode, all links
2931                  * must tx arp to ensure all links rx an arp - otherwise
2932                  * links may oscillate or not come up at all; if switch is
2933                  * in something like xor mode, there is nothing we can
2934                  * do - all replies will be rx'ed on same link causing slaves
2935                  * to be unstable during low/no traffic periods
2936                  */
2937                 if (IS_UP(slave->dev))
2938                         bond_arp_send_all(bond, slave);
2939         }
2940
2941         if (do_failover) {
2942                 block_netpoll_tx();
2943                 write_lock_bh(&bond->curr_slave_lock);
2944
2945                 bond_select_active_slave(bond);
2946
2947                 write_unlock_bh(&bond->curr_slave_lock);
2948                 unblock_netpoll_tx();
2949         }
2950
2951 re_arm:
2952         if (bond->params.arp_interval)
2953                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2954 out:
2955         read_unlock(&bond->lock);
2956 }
2957
2958 /*
2959  * Called to inspect slaves for active-backup mode ARP monitor link state
2960  * changes.  Sets new_link in slaves to specify what action should take
2961  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2962  * to link states must be committed.
2963  *
2964  * Called with bond->lock held for read.
2965  */
2966 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2967 {
2968         struct slave *slave;
2969         int i, commit = 0;
2970         unsigned long trans_start;
2971
2972         bond_for_each_slave(bond, slave, i) {
2973                 slave->new_link = BOND_LINK_NOCHANGE;
2974
2975                 if (slave->link != BOND_LINK_UP) {
2976                         if (time_in_range(jiffies,
2977                                 slave_last_rx(bond, slave) - delta_in_ticks,
2978                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2979
2980                                 slave->new_link = BOND_LINK_UP;
2981                                 commit++;
2982                         }
2983
2984                         continue;
2985                 }
2986
2987                 /*
2988                  * Give slaves 2*delta after being enslaved or made
2989                  * active.  This avoids bouncing, as the last receive
2990                  * times need a full ARP monitor cycle to be updated.
2991                  */
2992                 if (time_in_range(jiffies,
2993                                   slave->jiffies - delta_in_ticks,
2994                                   slave->jiffies + 2 * delta_in_ticks))
2995                         continue;
2996
2997                 /*
2998                  * Backup slave is down if:
2999                  * - No current_arp_slave AND
3000                  * - more than 3*delta since last receive AND
3001                  * - the bond has an IP address
3002                  *
3003                  * Note: a non-null current_arp_slave indicates
3004                  * the curr_active_slave went down and we are
3005                  * searching for a new one; under this condition
3006                  * we only take the curr_active_slave down - this
3007                  * gives each slave a chance to tx/rx traffic
3008                  * before being taken out
3009                  */
3010                 if (!bond_is_active_slave(slave) &&
3011                     !bond->current_arp_slave &&
3012                     !time_in_range(jiffies,
3013                         slave_last_rx(bond, slave) - delta_in_ticks,
3014                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3015
3016                         slave->new_link = BOND_LINK_DOWN;
3017                         commit++;
3018                 }
3019
3020                 /*
3021                  * Active slave is down if:
3022                  * - more than 2*delta since transmitting OR
3023                  * - (more than 2*delta since receive AND
3024                  *    the bond has an IP address)
3025                  */
3026                 trans_start = dev_trans_start(slave->dev);
3027                 if (bond_is_active_slave(slave) &&
3028                     (!time_in_range(jiffies,
3029                         trans_start - delta_in_ticks,
3030                         trans_start + 2 * delta_in_ticks) ||
3031                      !time_in_range(jiffies,
3032                         slave_last_rx(bond, slave) - delta_in_ticks,
3033                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3034
3035                         slave->new_link = BOND_LINK_DOWN;
3036                         commit++;
3037                 }
3038         }
3039
3040         return commit;
3041 }
3042
3043 /*
3044  * Called to commit link state changes noted by inspection step of
3045  * active-backup mode ARP monitor.
3046  *
3047  * Called with RTNL and bond->lock for read.
3048  */
3049 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3050 {
3051         struct slave *slave;
3052         int i;
3053         unsigned long trans_start;
3054
3055         bond_for_each_slave(bond, slave, i) {
3056                 switch (slave->new_link) {
3057                 case BOND_LINK_NOCHANGE:
3058                         continue;
3059
3060                 case BOND_LINK_UP:
3061                         trans_start = dev_trans_start(slave->dev);
3062                         if ((!bond->curr_active_slave &&
3063                              time_in_range(jiffies,
3064                                            trans_start - delta_in_ticks,
3065                                            trans_start + delta_in_ticks)) ||
3066                             bond->curr_active_slave != slave) {
3067                                 slave->link = BOND_LINK_UP;
3068                                 bond->current_arp_slave = NULL;
3069
3070                                 pr_info("%s: link status definitely up for interface %s.\n",
3071                                         bond->dev->name, slave->dev->name);
3072
3073                                 if (!bond->curr_active_slave ||
3074                                     (slave == bond->primary_slave))
3075                                         goto do_failover;
3076
3077                         }
3078
3079                         continue;
3080
3081                 case BOND_LINK_DOWN:
3082                         if (slave->link_failure_count < UINT_MAX)
3083                                 slave->link_failure_count++;
3084
3085                         slave->link = BOND_LINK_DOWN;
3086                         bond_set_slave_inactive_flags(slave);
3087
3088                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3089                                 bond->dev->name, slave->dev->name);
3090
3091                         if (slave == bond->curr_active_slave) {
3092                                 bond->current_arp_slave = NULL;
3093                                 goto do_failover;
3094                         }
3095
3096                         continue;
3097
3098                 default:
3099                         pr_err("%s: impossible: new_link %d on slave %s\n",
3100                                bond->dev->name, slave->new_link,
3101                                slave->dev->name);
3102                         continue;
3103                 }
3104
3105 do_failover:
3106                 ASSERT_RTNL();
3107                 block_netpoll_tx();
3108                 write_lock_bh(&bond->curr_slave_lock);
3109                 bond_select_active_slave(bond);
3110                 write_unlock_bh(&bond->curr_slave_lock);
3111                 unblock_netpoll_tx();
3112         }
3113
3114         bond_set_carrier(bond);
3115 }
3116
3117 /*
3118  * Send ARP probes for active-backup mode ARP monitor.
3119  *
3120  * Called with bond->lock held for read.
3121  */
3122 static void bond_ab_arp_probe(struct bonding *bond)
3123 {
3124         struct slave *slave;
3125         int i;
3126
3127         read_lock(&bond->curr_slave_lock);
3128
3129         if (bond->current_arp_slave && bond->curr_active_slave)
3130                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3131                         bond->current_arp_slave->dev->name,
3132                         bond->curr_active_slave->dev->name);
3133
3134         if (bond->curr_active_slave) {
3135                 bond_arp_send_all(bond, bond->curr_active_slave);
3136                 read_unlock(&bond->curr_slave_lock);
3137                 return;
3138         }
3139
3140         read_unlock(&bond->curr_slave_lock);
3141
3142         /* if we don't have a curr_active_slave, search for the next available
3143          * backup slave from the current_arp_slave and make it the candidate
3144          * for becoming the curr_active_slave
3145          */
3146
3147         if (!bond->current_arp_slave) {
3148                 bond->current_arp_slave = bond->first_slave;
3149                 if (!bond->current_arp_slave)
3150                         return;
3151         }
3152
3153         bond_set_slave_inactive_flags(bond->current_arp_slave);
3154
3155         /* search for next candidate */
3156         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3157                 if (IS_UP(slave->dev)) {
3158                         slave->link = BOND_LINK_BACK;
3159                         bond_set_slave_active_flags(slave);
3160                         bond_arp_send_all(bond, slave);
3161                         slave->jiffies = jiffies;
3162                         bond->current_arp_slave = slave;
3163                         break;
3164                 }
3165
3166                 /* if the link state is up at this point, we
3167                  * mark it down - this can happen if we have
3168                  * simultaneous link failures and
3169                  * reselect_active_interface doesn't make this
3170                  * one the current slave so it is still marked
3171                  * up when it is actually down
3172                  */
3173                 if (slave->link == BOND_LINK_UP) {
3174                         slave->link = BOND_LINK_DOWN;
3175                         if (slave->link_failure_count < UINT_MAX)
3176                                 slave->link_failure_count++;
3177
3178                         bond_set_slave_inactive_flags(slave);
3179
3180                         pr_info("%s: backup interface %s is now down.\n",
3181                                 bond->dev->name, slave->dev->name);
3182                 }
3183         }
3184 }
3185
3186 void bond_activebackup_arp_mon(struct work_struct *work)
3187 {
3188         struct bonding *bond = container_of(work, struct bonding,
3189                                             arp_work.work);
3190         bool should_notify_peers = false;
3191         int delta_in_ticks;
3192
3193         read_lock(&bond->lock);
3194
3195         if (bond->kill_timers)
3196                 goto out;
3197
3198         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3199
3200         if (bond->slave_cnt == 0)
3201                 goto re_arm;
3202
3203         should_notify_peers = bond_should_notify_peers(bond);
3204
3205         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3206                 read_unlock(&bond->lock);
3207                 rtnl_lock();
3208                 read_lock(&bond->lock);
3209
3210                 bond_ab_arp_commit(bond, delta_in_ticks);
3211
3212                 read_unlock(&bond->lock);
3213                 rtnl_unlock();
3214                 read_lock(&bond->lock);
3215         }
3216
3217         bond_ab_arp_probe(bond);
3218
3219 re_arm:
3220         if (bond->params.arp_interval)
3221                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3222 out:
3223         read_unlock(&bond->lock);
3224
3225         if (should_notify_peers) {
3226                 rtnl_lock();
3227                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3228                 rtnl_unlock();
3229         }
3230 }
3231
3232 /*-------------------------- netdev event handling --------------------------*/
3233
3234 /*
3235  * Change device name
3236  */
3237 static int bond_event_changename(struct bonding *bond)
3238 {
3239         bond_remove_proc_entry(bond);
3240         bond_create_proc_entry(bond);
3241
3242         bond_debug_reregister(bond);
3243
3244         return NOTIFY_DONE;
3245 }
3246
3247 static int bond_master_netdev_event(unsigned long event,
3248                                     struct net_device *bond_dev)
3249 {
3250         struct bonding *event_bond = netdev_priv(bond_dev);
3251
3252         switch (event) {
3253         case NETDEV_CHANGENAME:
3254                 return bond_event_changename(event_bond);
3255         default:
3256                 break;
3257         }
3258
3259         return NOTIFY_DONE;
3260 }
3261
3262 static int bond_slave_netdev_event(unsigned long event,
3263                                    struct net_device *slave_dev)
3264 {
3265         struct net_device *bond_dev = slave_dev->master;
3266         struct bonding *bond = netdev_priv(bond_dev);
3267
3268         switch (event) {
3269         case NETDEV_UNREGISTER:
3270                 if (bond_dev) {
3271                         if (bond->setup_by_slave)
3272                                 bond_release_and_destroy(bond_dev, slave_dev);
3273                         else
3274                                 bond_release(bond_dev, slave_dev);
3275                 }
3276                 break;
3277         case NETDEV_CHANGE:
3278                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3279                         struct slave *slave;
3280
3281                         slave = bond_get_slave_by_dev(bond, slave_dev);
3282                         if (slave) {
3283                                 u32 old_speed = slave->speed;
3284                                 u8  old_duplex = slave->duplex;
3285
3286                                 bond_update_speed_duplex(slave);
3287
3288                                 if (bond_is_lb(bond))
3289                                         break;
3290
3291                                 if (old_speed != slave->speed)
3292                                         bond_3ad_adapter_speed_changed(slave);
3293                                 if (old_duplex != slave->duplex)
3294                                         bond_3ad_adapter_duplex_changed(slave);
3295                         }
3296                 }
3297
3298                 break;
3299         case NETDEV_DOWN:
3300                 /*
3301                  * ... Or is it this?
3302                  */
3303                 break;
3304         case NETDEV_CHANGEMTU:
3305                 /*
3306                  * TODO: Should slaves be allowed to
3307                  * independently alter their MTU?  For
3308                  * an active-backup bond, slaves need
3309                  * not be the same type of device, so
3310                  * MTUs may vary.  For other modes,
3311                  * slaves arguably should have the
3312                  * same MTUs. To do this, we'd need to
3313                  * take over the slave's change_mtu
3314                  * function for the duration of their
3315                  * servitude.
3316                  */
3317                 break;
3318         case NETDEV_CHANGENAME:
3319                 /*
3320                  * TODO: handle changing the primary's name
3321                  */
3322                 break;
3323         case NETDEV_FEAT_CHANGE:
3324                 bond_compute_features(bond);
3325                 break;
3326         default:
3327                 break;
3328         }
3329
3330         return NOTIFY_DONE;
3331 }
3332
3333 /*
3334  * bond_netdev_event: handle netdev notifier chain events.
3335  *
3336  * This function receives events for the netdev chain.  The caller (an
3337  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3338  * locks for us to safely manipulate the slave devices (RTNL lock,
3339  * dev_probe_lock).
3340  */
3341 static int bond_netdev_event(struct notifier_block *this,
3342                              unsigned long event, void *ptr)
3343 {
3344         struct net_device *event_dev = (struct net_device *)ptr;
3345
3346         pr_debug("event_dev: %s, event: %lx\n",
3347                  event_dev ? event_dev->name : "None",
3348                  event);
3349
3350         if (!(event_dev->priv_flags & IFF_BONDING))
3351                 return NOTIFY_DONE;
3352
3353         if (event_dev->flags & IFF_MASTER) {
3354                 pr_debug("IFF_MASTER\n");
3355                 return bond_master_netdev_event(event, event_dev);
3356         }
3357
3358         if (event_dev->flags & IFF_SLAVE) {
3359                 pr_debug("IFF_SLAVE\n");
3360                 return bond_slave_netdev_event(event, event_dev);
3361         }
3362
3363         return NOTIFY_DONE;
3364 }
3365
3366 /*
3367  * bond_inetaddr_event: handle inetaddr notifier chain events.
3368  *
3369  * We keep track of device IPs primarily to use as source addresses in
3370  * ARP monitor probes (rather than spewing out broadcasts all the time).
3371  *
3372  * We track one IP for the main device (if it has one), plus one per VLAN.
3373  */
3374 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3375 {
3376         struct in_ifaddr *ifa = ptr;
3377         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3378         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3379         struct bonding *bond;
3380         struct vlan_entry *vlan;
3381
3382         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3383                 if (bond->dev == event_dev) {
3384                         switch (event) {
3385                         case NETDEV_UP:
3386                                 bond->master_ip = ifa->ifa_local;
3387                                 return NOTIFY_OK;
3388                         case NETDEV_DOWN:
3389                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3390                                 return NOTIFY_OK;
3391                         default:
3392                                 return NOTIFY_DONE;
3393                         }
3394                 }
3395
3396                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3397                         if (!bond->vlgrp)
3398                                 continue;
3399                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3400                         if (vlan_dev == event_dev) {
3401                                 switch (event) {
3402                                 case NETDEV_UP:
3403                                         vlan->vlan_ip = ifa->ifa_local;
3404                                         return NOTIFY_OK;
3405                                 case NETDEV_DOWN:
3406                                         vlan->vlan_ip =
3407                                                 bond_glean_dev_ip(vlan_dev);
3408                                         return NOTIFY_OK;
3409                                 default:
3410                                         return NOTIFY_DONE;
3411                                 }
3412                         }
3413                 }
3414         }
3415         return NOTIFY_DONE;
3416 }
3417
3418 static struct notifier_block bond_netdev_notifier = {
3419         .notifier_call = bond_netdev_event,
3420 };
3421
3422 static struct notifier_block bond_inetaddr_notifier = {
3423         .notifier_call = bond_inetaddr_event,
3424 };
3425
3426 /*---------------------------- Hashing Policies -----------------------------*/
3427
3428 /*
3429  * Hash for the output device based upon layer 2 and layer 3 data. If
3430  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3431  */
3432 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3433 {
3434         struct ethhdr *data = (struct ethhdr *)skb->data;
3435         struct iphdr *iph = ip_hdr(skb);
3436
3437         if (skb->protocol == htons(ETH_P_IP)) {
3438                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3439                         (data->h_dest[5] ^ data->h_source[5])) % count;
3440         }
3441
3442         return (data->h_dest[5] ^ data->h_source[5]) % count;
3443 }
3444
3445 /*
3446  * Hash for the output device based upon layer 3 and layer 4 data. If
3447  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3448  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3449  */
3450 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3451 {
3452         struct ethhdr *data = (struct ethhdr *)skb->data;
3453         struct iphdr *iph = ip_hdr(skb);
3454         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3455         int layer4_xor = 0;
3456
3457         if (skb->protocol == htons(ETH_P_IP)) {
3458                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3459                     (iph->protocol == IPPROTO_TCP ||
3460                      iph->protocol == IPPROTO_UDP)) {
3461                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3462                 }
3463                 return (layer4_xor ^
3464                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3465
3466         }
3467
3468         return (data->h_dest[5] ^ data->h_source[5]) % count;
3469 }
3470
3471 /*
3472  * Hash for the output device based upon layer 2 data
3473  */
3474 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3475 {
3476         struct ethhdr *data = (struct ethhdr *)skb->data;
3477
3478         return (data->h_dest[5] ^ data->h_source[5]) % count;
3479 }
3480
3481 /*-------------------------- Device entry points ----------------------------*/
3482
3483 static int bond_open(struct net_device *bond_dev)
3484 {
3485         struct bonding *bond = netdev_priv(bond_dev);
3486
3487         bond->kill_timers = 0;
3488
3489         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3490
3491         if (bond_is_lb(bond)) {
3492                 /* bond_alb_initialize must be called before the timer
3493                  * is started.
3494                  */
3495                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3496                         /* something went wrong - fail the open operation */
3497                         return -ENOMEM;
3498                 }
3499
3500                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3501                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3502         }
3503
3504         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3505                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3506                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3507         }
3508
3509         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3510                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3511                         INIT_DELAYED_WORK(&bond->arp_work,
3512                                           bond_activebackup_arp_mon);
3513                 else
3514                         INIT_DELAYED_WORK(&bond->arp_work,
3515                                           bond_loadbalance_arp_mon);
3516
3517                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3518                 if (bond->params.arp_validate)
3519                         bond->recv_probe = bond_arp_rcv;
3520         }
3521
3522         if (bond->params.mode == BOND_MODE_8023AD) {
3523                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3524                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3525                 /* register to receive LACPDUs */
3526                 bond->recv_probe = bond_3ad_lacpdu_recv;
3527                 bond_3ad_initiate_agg_selection(bond, 1);
3528         }
3529
3530         return 0;
3531 }
3532
3533 static int bond_close(struct net_device *bond_dev)
3534 {
3535         struct bonding *bond = netdev_priv(bond_dev);
3536
3537         write_lock_bh(&bond->lock);
3538
3539         bond->send_peer_notif = 0;
3540
3541         /* signal timers not to re-arm */
3542         bond->kill_timers = 1;
3543
3544         write_unlock_bh(&bond->lock);
3545
3546         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3547                 cancel_delayed_work(&bond->mii_work);
3548         }
3549
3550         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3551                 cancel_delayed_work(&bond->arp_work);
3552         }
3553
3554         switch (bond->params.mode) {
3555         case BOND_MODE_8023AD:
3556                 cancel_delayed_work(&bond->ad_work);
3557                 break;
3558         case BOND_MODE_TLB:
3559         case BOND_MODE_ALB:
3560                 cancel_delayed_work(&bond->alb_work);
3561                 break;
3562         default:
3563                 break;
3564         }
3565
3566         if (delayed_work_pending(&bond->mcast_work))
3567                 cancel_delayed_work(&bond->mcast_work);
3568
3569         if (bond_is_lb(bond)) {
3570                 /* Must be called only after all
3571                  * slaves have been released
3572                  */
3573                 bond_alb_deinitialize(bond);
3574         }
3575         bond->recv_probe = NULL;
3576
3577         return 0;
3578 }
3579
3580 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3581                                                 struct rtnl_link_stats64 *stats)
3582 {
3583         struct bonding *bond = netdev_priv(bond_dev);
3584         struct rtnl_link_stats64 temp;
3585         struct slave *slave;
3586         int i;
3587
3588         memset(stats, 0, sizeof(*stats));
3589
3590         read_lock_bh(&bond->lock);
3591
3592         bond_for_each_slave(bond, slave, i) {
3593                 const struct rtnl_link_stats64 *sstats =
3594                         dev_get_stats(slave->dev, &temp);
3595
3596                 stats->rx_packets += sstats->rx_packets;
3597                 stats->rx_bytes += sstats->rx_bytes;
3598                 stats->rx_errors += sstats->rx_errors;
3599                 stats->rx_dropped += sstats->rx_dropped;
3600
3601                 stats->tx_packets += sstats->tx_packets;
3602                 stats->tx_bytes += sstats->tx_bytes;
3603                 stats->tx_errors += sstats->tx_errors;
3604                 stats->tx_dropped += sstats->tx_dropped;
3605
3606                 stats->multicast += sstats->multicast;
3607                 stats->collisions += sstats->collisions;
3608
3609                 stats->rx_length_errors += sstats->rx_length_errors;
3610                 stats->rx_over_errors += sstats->rx_over_errors;
3611                 stats->rx_crc_errors += sstats->rx_crc_errors;
3612                 stats->rx_frame_errors += sstats->rx_frame_errors;
3613                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3614                 stats->rx_missed_errors += sstats->rx_missed_errors;
3615
3616                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3617                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3618                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3619                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3620                 stats->tx_window_errors += sstats->tx_window_errors;
3621         }
3622
3623         read_unlock_bh(&bond->lock);
3624
3625         return stats;
3626 }
3627
3628 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3629 {
3630         struct net_device *slave_dev = NULL;
3631         struct ifbond k_binfo;
3632         struct ifbond __user *u_binfo = NULL;
3633         struct ifslave k_sinfo;
3634         struct ifslave __user *u_sinfo = NULL;
3635         struct mii_ioctl_data *mii = NULL;
3636         int res = 0;
3637
3638         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3639
3640         switch (cmd) {
3641         case SIOCGMIIPHY:
3642                 mii = if_mii(ifr);
3643                 if (!mii)
3644                         return -EINVAL;
3645
3646                 mii->phy_id = 0;
3647                 /* Fall Through */
3648         case SIOCGMIIREG:
3649                 /*
3650                  * We do this again just in case we were called by SIOCGMIIREG
3651                  * instead of SIOCGMIIPHY.
3652                  */
3653                 mii = if_mii(ifr);
3654                 if (!mii)
3655                         return -EINVAL;
3656
3657
3658                 if (mii->reg_num == 1) {
3659                         struct bonding *bond = netdev_priv(bond_dev);
3660                         mii->val_out = 0;
3661                         read_lock(&bond->lock);
3662                         read_lock(&bond->curr_slave_lock);
3663                         if (netif_carrier_ok(bond->dev))
3664                                 mii->val_out = BMSR_LSTATUS;
3665
3666                         read_unlock(&bond->curr_slave_lock);
3667                         read_unlock(&bond->lock);
3668                 }
3669
3670                 return 0;
3671         case BOND_INFO_QUERY_OLD:
3672         case SIOCBONDINFOQUERY:
3673                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3674
3675                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3676                         return -EFAULT;
3677
3678                 res = bond_info_query(bond_dev, &k_binfo);
3679                 if (res == 0 &&
3680                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3681                         return -EFAULT;
3682
3683                 return res;
3684         case BOND_SLAVE_INFO_QUERY_OLD:
3685         case SIOCBONDSLAVEINFOQUERY:
3686                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3687
3688                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3689                         return -EFAULT;
3690
3691                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3692                 if (res == 0 &&
3693                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3694                         return -EFAULT;
3695
3696                 return res;
3697         default:
3698                 /* Go on */
3699                 break;
3700         }
3701
3702         if (!capable(CAP_NET_ADMIN))
3703                 return -EPERM;
3704
3705         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3706
3707         pr_debug("slave_dev=%p:\n", slave_dev);
3708
3709         if (!slave_dev)
3710                 res = -ENODEV;
3711         else {
3712                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3713                 switch (cmd) {
3714                 case BOND_ENSLAVE_OLD:
3715                 case SIOCBONDENSLAVE:
3716                         res = bond_enslave(bond_dev, slave_dev);
3717                         break;
3718                 case BOND_RELEASE_OLD:
3719                 case SIOCBONDRELEASE:
3720                         res = bond_release(bond_dev, slave_dev);
3721                         break;
3722                 case BOND_SETHWADDR_OLD:
3723                 case SIOCBONDSETHWADDR:
3724                         res = bond_sethwaddr(bond_dev, slave_dev);
3725                         break;
3726                 case BOND_CHANGE_ACTIVE_OLD:
3727                 case SIOCBONDCHANGEACTIVE:
3728                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3729                         break;
3730                 default:
3731                         res = -EOPNOTSUPP;
3732                 }
3733
3734                 dev_put(slave_dev);
3735         }
3736
3737         return res;
3738 }
3739
3740 static bool bond_addr_in_mc_list(unsigned char *addr,
3741                                  struct netdev_hw_addr_list *list,
3742                                  int addrlen)
3743 {
3744         struct netdev_hw_addr *ha;
3745
3746         netdev_hw_addr_list_for_each(ha, list)
3747                 if (!memcmp(ha->addr, addr, addrlen))
3748                         return true;
3749
3750         return false;
3751 }
3752
3753 static void bond_set_multicast_list(struct net_device *bond_dev)
3754 {
3755         struct bonding *bond = netdev_priv(bond_dev);
3756         struct netdev_hw_addr *ha;
3757         bool found;
3758
3759         /*
3760          * Do promisc before checking multicast_mode
3761          */
3762         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3763                 /*
3764                  * FIXME: Need to handle the error when one of the multi-slaves
3765                  * encounters error.
3766                  */
3767                 bond_set_promiscuity(bond, 1);
3768
3769
3770         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3771                 bond_set_promiscuity(bond, -1);
3772
3773
3774         /* set allmulti flag to slaves */
3775         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3776                 /*
3777                  * FIXME: Need to handle the error when one of the multi-slaves
3778                  * encounters error.
3779                  */
3780                 bond_set_allmulti(bond, 1);
3781
3782
3783         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3784                 bond_set_allmulti(bond, -1);
3785
3786
3787         read_lock(&bond->lock);
3788
3789         bond->flags = bond_dev->flags;
3790
3791         /* looking for addresses to add to slaves' mc list */
3792         netdev_for_each_mc_addr(ha, bond_dev) {
3793                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3794                                              bond_dev->addr_len);
3795                 if (!found)
3796                         bond_mc_add(bond, ha->addr);
3797         }
3798
3799         /* looking for addresses to delete from slaves' list */
3800         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3801                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3802                                              bond_dev->addr_len);
3803                 if (!found)
3804                         bond_mc_del(bond, ha->addr);
3805         }
3806
3807         /* save master's multicast list */
3808         __hw_addr_flush(&bond->mc_list);
3809         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3810                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3811
3812         read_unlock(&bond->lock);
3813 }
3814
3815 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3816 {
3817         struct bonding *bond = netdev_priv(dev);
3818         struct slave *slave = bond->first_slave;
3819
3820         if (slave) {
3821                 const struct net_device_ops *slave_ops
3822                         = slave->dev->netdev_ops;
3823                 if (slave_ops->ndo_neigh_setup)
3824                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3825         }
3826         return 0;
3827 }
3828
3829 /*
3830  * Change the MTU of all of a master's slaves to match the master
3831  */
3832 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3833 {
3834         struct bonding *bond = netdev_priv(bond_dev);
3835         struct slave *slave, *stop_at;
3836         int res = 0;
3837         int i;
3838
3839         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3840                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3841
3842         /* Can't hold bond->lock with bh disabled here since
3843          * some base drivers panic. On the other hand we can't
3844          * hold bond->lock without bh disabled because we'll
3845          * deadlock. The only solution is to rely on the fact
3846          * that we're under rtnl_lock here, and the slaves
3847          * list won't change. This doesn't solve the problem
3848          * of setting the slave's MTU while it is
3849          * transmitting, but the assumption is that the base
3850          * driver can handle that.
3851          *
3852          * TODO: figure out a way to safely iterate the slaves
3853          * list, but without holding a lock around the actual
3854          * call to the base driver.
3855          */
3856
3857         bond_for_each_slave(bond, slave, i) {
3858                 pr_debug("s %p s->p %p c_m %p\n",
3859                          slave,
3860                          slave->prev,
3861                          slave->dev->netdev_ops->ndo_change_mtu);
3862
3863                 res = dev_set_mtu(slave->dev, new_mtu);
3864
3865                 if (res) {
3866                         /* If we failed to set the slave's mtu to the new value
3867                          * we must abort the operation even in ACTIVE_BACKUP
3868                          * mode, because if we allow the backup slaves to have
3869                          * different mtu values than the active slave we'll
3870                          * need to change their mtu when doing a failover. That
3871                          * means changing their mtu from timer context, which
3872                          * is probably not a good idea.
3873                          */
3874                         pr_debug("err %d %s\n", res, slave->dev->name);
3875                         goto unwind;
3876                 }
3877         }
3878
3879         bond_dev->mtu = new_mtu;
3880
3881         return 0;
3882
3883 unwind:
3884         /* unwind from head to the slave that failed */
3885         stop_at = slave;
3886         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3887                 int tmp_res;
3888
3889                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3890                 if (tmp_res) {
3891                         pr_debug("unwind err %d dev %s\n",
3892                                  tmp_res, slave->dev->name);
3893                 }
3894         }
3895
3896         return res;
3897 }
3898
3899 /*
3900  * Change HW address
3901  *
3902  * Note that many devices must be down to change the HW address, and
3903  * downing the master releases all slaves.  We can make bonds full of
3904  * bonding devices to test this, however.
3905  */
3906 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3907 {
3908         struct bonding *bond = netdev_priv(bond_dev);
3909         struct sockaddr *sa = addr, tmp_sa;
3910         struct slave *slave, *stop_at;
3911         int res = 0;
3912         int i;
3913
3914         if (bond->params.mode == BOND_MODE_ALB)
3915                 return bond_alb_set_mac_address(bond_dev, addr);
3916
3917
3918         pr_debug("bond=%p, name=%s\n",
3919                  bond, bond_dev ? bond_dev->name : "None");
3920
3921         /*
3922          * If fail_over_mac is set to active, do nothing and return
3923          * success.  Returning an error causes ifenslave to fail.
3924          */
3925         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3926                 return 0;
3927
3928         if (!is_valid_ether_addr(sa->sa_data))
3929                 return -EADDRNOTAVAIL;
3930
3931         /* Can't hold bond->lock with bh disabled here since
3932          * some base drivers panic. On the other hand we can't
3933          * hold bond->lock without bh disabled because we'll
3934          * deadlock. The only solution is to rely on the fact
3935          * that we're under rtnl_lock here, and the slaves
3936          * list won't change. This doesn't solve the problem
3937          * of setting the slave's hw address while it is
3938          * transmitting, but the assumption is that the base
3939          * driver can handle that.
3940          *
3941          * TODO: figure out a way to safely iterate the slaves
3942          * list, but without holding a lock around the actual
3943          * call to the base driver.
3944          */
3945
3946         bond_for_each_slave(bond, slave, i) {
3947                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3948                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3949
3950                 if (slave_ops->ndo_set_mac_address == NULL) {
3951                         res = -EOPNOTSUPP;
3952                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3953                         goto unwind;
3954                 }
3955
3956                 res = dev_set_mac_address(slave->dev, addr);
3957                 if (res) {
3958                         /* TODO: consider downing the slave
3959                          * and retry ?
3960                          * User should expect communications
3961                          * breakage anyway until ARP finish
3962                          * updating, so...
3963                          */
3964                         pr_debug("err %d %s\n", res, slave->dev->name);
3965                         goto unwind;
3966                 }
3967         }
3968
3969         /* success */
3970         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3971         return 0;
3972
3973 unwind:
3974         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3975         tmp_sa.sa_family = bond_dev->type;
3976
3977         /* unwind from head to the slave that failed */
3978         stop_at = slave;
3979         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3980                 int tmp_res;
3981
3982                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3983                 if (tmp_res) {
3984                         pr_debug("unwind err %d dev %s\n",
3985                                  tmp_res, slave->dev->name);
3986                 }
3987         }
3988
3989         return res;
3990 }
3991
3992 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3993 {
3994         struct bonding *bond = netdev_priv(bond_dev);
3995         struct slave *slave, *start_at;
3996         int i, slave_no, res = 1;
3997         struct iphdr *iph = ip_hdr(skb);
3998
3999         /*
4000          * Start with the curr_active_slave that joined the bond as the
4001          * default for sending IGMP traffic.  For failover purposes one
4002          * needs to maintain some consistency for the interface that will
4003          * send the join/membership reports.  The curr_active_slave found
4004          * will send all of this type of traffic.
4005          */
4006         if ((iph->protocol == IPPROTO_IGMP) &&
4007             (skb->protocol == htons(ETH_P_IP))) {
4008
4009                 read_lock(&bond->curr_slave_lock);
4010                 slave = bond->curr_active_slave;
4011                 read_unlock(&bond->curr_slave_lock);
4012
4013                 if (!slave)
4014                         goto out;
4015         } else {
4016                 /*
4017                  * Concurrent TX may collide on rr_tx_counter; we accept
4018                  * that as being rare enough not to justify using an
4019                  * atomic op here.
4020                  */
4021                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4022
4023                 bond_for_each_slave(bond, slave, i) {
4024                         slave_no--;
4025                         if (slave_no < 0)
4026                                 break;
4027                 }
4028         }
4029
4030         start_at = slave;
4031         bond_for_each_slave_from(bond, slave, i, start_at) {
4032                 if (IS_UP(slave->dev) &&
4033                     (slave->link == BOND_LINK_UP) &&
4034                     bond_is_active_slave(slave)) {
4035                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4036                         break;
4037                 }
4038         }
4039
4040 out:
4041         if (res) {
4042                 /* no suitable interface, frame not sent */
4043                 dev_kfree_skb(skb);
4044         }
4045
4046         return NETDEV_TX_OK;
4047 }
4048
4049
4050 /*
4051  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4052  * the bond has a usable interface.
4053  */
4054 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4055 {
4056         struct bonding *bond = netdev_priv(bond_dev);
4057         int res = 1;
4058
4059         read_lock(&bond->curr_slave_lock);
4060
4061         if (bond->curr_active_slave)
4062                 res = bond_dev_queue_xmit(bond, skb,
4063                         bond->curr_active_slave->dev);
4064
4065         if (res)
4066                 /* no suitable interface, frame not sent */
4067                 dev_kfree_skb(skb);
4068
4069         read_unlock(&bond->curr_slave_lock);
4070
4071         return NETDEV_TX_OK;
4072 }
4073
4074 /*
4075  * In bond_xmit_xor() , we determine the output device by using a pre-
4076  * determined xmit_hash_policy(), If the selected device is not enabled,
4077  * find the next active slave.
4078  */
4079 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4080 {
4081         struct bonding *bond = netdev_priv(bond_dev);
4082         struct slave *slave, *start_at;
4083         int slave_no;
4084         int i;
4085         int res = 1;
4086
4087         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4088
4089         bond_for_each_slave(bond, slave, i) {
4090                 slave_no--;
4091                 if (slave_no < 0)
4092                         break;
4093         }
4094
4095         start_at = slave;
4096
4097         bond_for_each_slave_from(bond, slave, i, start_at) {
4098                 if (IS_UP(slave->dev) &&
4099                     (slave->link == BOND_LINK_UP) &&
4100                     bond_is_active_slave(slave)) {
4101                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4102                         break;
4103                 }
4104         }
4105
4106         if (res) {
4107                 /* no suitable interface, frame not sent */
4108                 dev_kfree_skb(skb);
4109         }
4110
4111         return NETDEV_TX_OK;
4112 }
4113
4114 /*
4115  * in broadcast mode, we send everything to all usable interfaces.
4116  */
4117 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4118 {
4119         struct bonding *bond = netdev_priv(bond_dev);
4120         struct slave *slave, *start_at;
4121         struct net_device *tx_dev = NULL;
4122         int i;
4123         int res = 1;
4124
4125         read_lock(&bond->curr_slave_lock);
4126         start_at = bond->curr_active_slave;
4127         read_unlock(&bond->curr_slave_lock);
4128
4129         if (!start_at)
4130                 goto out;
4131
4132         bond_for_each_slave_from(bond, slave, i, start_at) {
4133                 if (IS_UP(slave->dev) &&
4134                     (slave->link == BOND_LINK_UP) &&
4135                     bond_is_active_slave(slave)) {
4136                         if (tx_dev) {
4137                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4138                                 if (!skb2) {
4139                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4140                                                bond_dev->name);
4141                                         continue;
4142                                 }
4143
4144                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4145                                 if (res) {
4146                                         dev_kfree_skb(skb2);
4147                                         continue;
4148                                 }
4149                         }
4150                         tx_dev = slave->dev;
4151                 }
4152         }
4153
4154         if (tx_dev)
4155                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4156
4157 out:
4158         if (res)
4159                 /* no suitable interface, frame not sent */
4160                 dev_kfree_skb(skb);
4161
4162         /* frame sent to all suitable interfaces */
4163         return NETDEV_TX_OK;
4164 }
4165
4166 /*------------------------- Device initialization ---------------------------*/
4167
4168 static void bond_set_xmit_hash_policy(struct bonding *bond)
4169 {
4170         switch (bond->params.xmit_policy) {
4171         case BOND_XMIT_POLICY_LAYER23:
4172                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4173                 break;
4174         case BOND_XMIT_POLICY_LAYER34:
4175                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4176                 break;
4177         case BOND_XMIT_POLICY_LAYER2:
4178         default:
4179                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4180                 break;
4181         }
4182 }
4183
4184 /*
4185  * Lookup the slave that corresponds to a qid
4186  */
4187 static inline int bond_slave_override(struct bonding *bond,
4188                                       struct sk_buff *skb)
4189 {
4190         int i, res = 1;
4191         struct slave *slave = NULL;
4192         struct slave *check_slave;
4193
4194         if (!skb->queue_mapping)
4195                 return 1;
4196
4197         /* Find out if any slaves have the same mapping as this skb. */
4198         bond_for_each_slave(bond, check_slave, i) {
4199                 if (check_slave->queue_id == skb->queue_mapping) {
4200                         slave = check_slave;
4201                         break;
4202                 }
4203         }
4204
4205         /* If the slave isn't UP, use default transmit policy. */
4206         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4207             (slave->link == BOND_LINK_UP)) {
4208                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4209         }
4210
4211         return res;
4212 }
4213
4214
4215 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4216 {
4217         /*
4218          * This helper function exists to help dev_pick_tx get the correct
4219          * destination queue.  Using a helper function skips a call to
4220          * skb_tx_hash and will put the skbs in the queue we expect on their
4221          * way down to the bonding driver.
4222          */
4223         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4224
4225         /*
4226          * Save the original txq to restore before passing to the driver
4227          */
4228         bond_queue_mapping(skb) = skb->queue_mapping;
4229
4230         if (unlikely(txq >= dev->real_num_tx_queues)) {
4231                 do {
4232                         txq -= dev->real_num_tx_queues;
4233                 } while (txq >= dev->real_num_tx_queues);
4234         }
4235         return txq;
4236 }
4237
4238 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4239 {
4240         struct bonding *bond = netdev_priv(dev);
4241
4242         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4243                 if (!bond_slave_override(bond, skb))
4244                         return NETDEV_TX_OK;
4245         }
4246
4247         switch (bond->params.mode) {
4248         case BOND_MODE_ROUNDROBIN:
4249                 return bond_xmit_roundrobin(skb, dev);
4250         case BOND_MODE_ACTIVEBACKUP:
4251                 return bond_xmit_activebackup(skb, dev);
4252         case BOND_MODE_XOR:
4253                 return bond_xmit_xor(skb, dev);
4254         case BOND_MODE_BROADCAST:
4255                 return bond_xmit_broadcast(skb, dev);
4256         case BOND_MODE_8023AD:
4257                 return bond_3ad_xmit_xor(skb, dev);
4258         case BOND_MODE_ALB:
4259         case BOND_MODE_TLB:
4260                 return bond_alb_xmit(skb, dev);
4261         default:
4262                 /* Should never happen, mode already checked */
4263                 pr_err("%s: Error: Unknown bonding mode %d\n",
4264                        dev->name, bond->params.mode);
4265                 WARN_ON_ONCE(1);
4266                 dev_kfree_skb(skb);
4267                 return NETDEV_TX_OK;
4268         }
4269 }
4270
4271 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4272 {
4273         struct bonding *bond = netdev_priv(dev);
4274         netdev_tx_t ret = NETDEV_TX_OK;
4275
4276         /*
4277          * If we risk deadlock from transmitting this in the
4278          * netpoll path, tell netpoll to queue the frame for later tx
4279          */
4280         if (is_netpoll_tx_blocked(dev))
4281                 return NETDEV_TX_BUSY;
4282
4283         read_lock(&bond->lock);
4284
4285         if (bond->slave_cnt)
4286                 ret = __bond_start_xmit(skb, dev);
4287         else
4288                 dev_kfree_skb(skb);
4289
4290         read_unlock(&bond->lock);
4291
4292         return ret;
4293 }
4294
4295 /*
4296  * set bond mode specific net device operations
4297  */
4298 void bond_set_mode_ops(struct bonding *bond, int mode)
4299 {
4300         struct net_device *bond_dev = bond->dev;
4301
4302         switch (mode) {
4303         case BOND_MODE_ROUNDROBIN:
4304                 break;
4305         case BOND_MODE_ACTIVEBACKUP:
4306                 break;
4307         case BOND_MODE_XOR:
4308                 bond_set_xmit_hash_policy(bond);
4309                 break;
4310         case BOND_MODE_BROADCAST:
4311                 break;
4312         case BOND_MODE_8023AD:
4313                 bond_set_xmit_hash_policy(bond);
4314                 break;
4315         case BOND_MODE_ALB:
4316                 /* FALLTHRU */
4317         case BOND_MODE_TLB:
4318                 break;
4319         default:
4320                 /* Should never happen, mode already checked */
4321                 pr_err("%s: Error: Unknown bonding mode %d\n",
4322                        bond_dev->name, mode);
4323                 break;
4324         }
4325 }
4326
4327 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4328                                     struct ethtool_drvinfo *drvinfo)
4329 {
4330         strncpy(drvinfo->driver, DRV_NAME, 32);
4331         strncpy(drvinfo->version, DRV_VERSION, 32);
4332         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4333 }
4334
4335 static const struct ethtool_ops bond_ethtool_ops = {
4336         .get_drvinfo            = bond_ethtool_get_drvinfo,
4337         .get_link               = ethtool_op_get_link,
4338 };
4339
4340 static const struct net_device_ops bond_netdev_ops = {
4341         .ndo_init               = bond_init,
4342         .ndo_uninit             = bond_uninit,
4343         .ndo_open               = bond_open,
4344         .ndo_stop               = bond_close,
4345         .ndo_start_xmit         = bond_start_xmit,
4346         .ndo_select_queue       = bond_select_queue,
4347         .ndo_get_stats64        = bond_get_stats,
4348         .ndo_do_ioctl           = bond_do_ioctl,
4349         .ndo_set_multicast_list = bond_set_multicast_list,
4350         .ndo_change_mtu         = bond_change_mtu,
4351         .ndo_set_mac_address    = bond_set_mac_address,
4352         .ndo_neigh_setup        = bond_neigh_setup,
4353         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4354         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4355         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4356 #ifdef CONFIG_NET_POLL_CONTROLLER
4357         .ndo_netpoll_setup      = bond_netpoll_setup,
4358         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4359         .ndo_poll_controller    = bond_poll_controller,
4360 #endif
4361         .ndo_add_slave          = bond_enslave,
4362         .ndo_del_slave          = bond_release,
4363         .ndo_fix_features       = bond_fix_features,
4364 };
4365
4366 static void bond_destructor(struct net_device *bond_dev)
4367 {
4368         struct bonding *bond = netdev_priv(bond_dev);
4369         if (bond->wq)
4370                 destroy_workqueue(bond->wq);
4371         free_netdev(bond_dev);
4372 }
4373
4374 static void bond_setup(struct net_device *bond_dev)
4375 {
4376         struct bonding *bond = netdev_priv(bond_dev);
4377
4378         /* initialize rwlocks */
4379         rwlock_init(&bond->lock);
4380         rwlock_init(&bond->curr_slave_lock);
4381
4382         bond->params = bonding_defaults;
4383
4384         /* Initialize pointers */
4385         bond->dev = bond_dev;
4386         INIT_LIST_HEAD(&bond->vlan_list);
4387
4388         /* Initialize the device entry points */
4389         ether_setup(bond_dev);
4390         bond_dev->netdev_ops = &bond_netdev_ops;
4391         bond_dev->ethtool_ops = &bond_ethtool_ops;
4392         bond_set_mode_ops(bond, bond->params.mode);
4393
4394         bond_dev->destructor = bond_destructor;
4395
4396         /* Initialize the device options */
4397         bond_dev->tx_queue_len = 0;
4398         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4399         bond_dev->priv_flags |= IFF_BONDING;
4400         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4401
4402         /* At first, we block adding VLANs. That's the only way to
4403          * prevent problems that occur when adding VLANs over an
4404          * empty bond. The block will be removed once non-challenged
4405          * slaves are enslaved.
4406          */
4407         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4408
4409         /* don't acquire bond device's netif_tx_lock when
4410          * transmitting */
4411         bond_dev->features |= NETIF_F_LLTX;
4412
4413         /* By default, we declare the bond to be fully
4414          * VLAN hardware accelerated capable. Special
4415          * care is taken in the various xmit functions
4416          * when there are slaves that are not hw accel
4417          * capable
4418          */
4419
4420         bond_dev->hw_features = BOND_VLAN_FEATURES |
4421                                 NETIF_F_HW_VLAN_TX |
4422                                 NETIF_F_HW_VLAN_RX |
4423                                 NETIF_F_HW_VLAN_FILTER;
4424
4425         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4426         bond_dev->features |= bond_dev->hw_features;
4427 }
4428
4429 static void bond_work_cancel_all(struct bonding *bond)
4430 {
4431         write_lock_bh(&bond->lock);
4432         bond->kill_timers = 1;
4433         write_unlock_bh(&bond->lock);
4434
4435         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4436                 cancel_delayed_work(&bond->mii_work);
4437
4438         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4439                 cancel_delayed_work(&bond->arp_work);
4440
4441         if (bond->params.mode == BOND_MODE_ALB &&
4442             delayed_work_pending(&bond->alb_work))
4443                 cancel_delayed_work(&bond->alb_work);
4444
4445         if (bond->params.mode == BOND_MODE_8023AD &&
4446             delayed_work_pending(&bond->ad_work))
4447                 cancel_delayed_work(&bond->ad_work);
4448
4449         if (delayed_work_pending(&bond->mcast_work))
4450                 cancel_delayed_work(&bond->mcast_work);
4451 }
4452
4453 /*
4454 * Destroy a bonding device.
4455 * Must be under rtnl_lock when this function is called.
4456 */
4457 static void bond_uninit(struct net_device *bond_dev)
4458 {
4459         struct bonding *bond = netdev_priv(bond_dev);
4460         struct vlan_entry *vlan, *tmp;
4461
4462         bond_netpoll_cleanup(bond_dev);
4463
4464         /* Release the bonded slaves */
4465         bond_release_all(bond_dev);
4466
4467         list_del(&bond->bond_list);
4468
4469         bond_work_cancel_all(bond);
4470
4471         bond_remove_proc_entry(bond);
4472
4473         bond_debug_unregister(bond);
4474
4475         __hw_addr_flush(&bond->mc_list);
4476
4477         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4478                 list_del(&vlan->vlan_list);
4479                 kfree(vlan);
4480         }
4481 }
4482
4483 /*------------------------- Module initialization ---------------------------*/
4484
4485 /*
4486  * Convert string input module parms.  Accept either the
4487  * number of the mode or its string name.  A bit complicated because
4488  * some mode names are substrings of other names, and calls from sysfs
4489  * may have whitespace in the name (trailing newlines, for example).
4490  */
4491 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4492 {
4493         int modeint = -1, i, rv;
4494         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4495
4496         for (p = (char *)buf; *p; p++)
4497                 if (!(isdigit(*p) || isspace(*p)))
4498                         break;
4499
4500         if (*p)
4501                 rv = sscanf(buf, "%20s", modestr);
4502         else
4503                 rv = sscanf(buf, "%d", &modeint);
4504
4505         if (!rv)
4506                 return -1;
4507
4508         for (i = 0; tbl[i].modename; i++) {
4509                 if (modeint == tbl[i].mode)
4510                         return tbl[i].mode;
4511                 if (strcmp(modestr, tbl[i].modename) == 0)
4512                         return tbl[i].mode;
4513         }
4514
4515         return -1;
4516 }
4517
4518 static int bond_check_params(struct bond_params *params)
4519 {
4520         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4521
4522         /*
4523          * Convert string parameters.
4524          */
4525         if (mode) {
4526                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4527                 if (bond_mode == -1) {
4528                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4529                                mode == NULL ? "NULL" : mode);
4530                         return -EINVAL;
4531                 }
4532         }
4533
4534         if (xmit_hash_policy) {
4535                 if ((bond_mode != BOND_MODE_XOR) &&
4536                     (bond_mode != BOND_MODE_8023AD)) {
4537                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4538                                bond_mode_name(bond_mode));
4539                 } else {
4540                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4541                                                         xmit_hashtype_tbl);
4542                         if (xmit_hashtype == -1) {
4543                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4544                                        xmit_hash_policy == NULL ? "NULL" :
4545                                        xmit_hash_policy);
4546                                 return -EINVAL;
4547                         }
4548                 }
4549         }
4550
4551         if (lacp_rate) {
4552                 if (bond_mode != BOND_MODE_8023AD) {
4553                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4554                                 bond_mode_name(bond_mode));
4555                 } else {
4556                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4557                         if (lacp_fast == -1) {
4558                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4559                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4560                                 return -EINVAL;
4561                         }
4562                 }
4563         }
4564
4565         if (ad_select) {
4566                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4567                 if (params->ad_select == -1) {
4568                         pr_err("Error: Invalid ad_select \"%s\"\n",
4569                                ad_select == NULL ? "NULL" : ad_select);
4570                         return -EINVAL;
4571                 }
4572
4573                 if (bond_mode != BOND_MODE_8023AD) {
4574                         pr_warning("ad_select param only affects 802.3ad mode\n");
4575                 }
4576         } else {
4577                 params->ad_select = BOND_AD_STABLE;
4578         }
4579
4580         if (max_bonds < 0) {
4581                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4582                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4583                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4584         }
4585
4586         if (miimon < 0) {
4587                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4588                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4589                 miimon = BOND_LINK_MON_INTERV;
4590         }
4591
4592         if (updelay < 0) {
4593                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4594                            updelay, INT_MAX);
4595                 updelay = 0;
4596         }
4597
4598         if (downdelay < 0) {
4599                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4600                            downdelay, INT_MAX);
4601                 downdelay = 0;
4602         }
4603
4604         if ((use_carrier != 0) && (use_carrier != 1)) {
4605                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4606                            use_carrier);
4607                 use_carrier = 1;
4608         }
4609
4610         if (num_peer_notif < 0 || num_peer_notif > 255) {
4611                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4612                            num_peer_notif);
4613                 num_peer_notif = 1;
4614         }
4615
4616         /* reset values for 802.3ad */
4617         if (bond_mode == BOND_MODE_8023AD) {
4618                 if (!miimon) {
4619                         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");
4620                         pr_warning("Forcing miimon to 100msec\n");
4621                         miimon = 100;
4622                 }
4623         }
4624
4625         if (tx_queues < 1 || tx_queues > 255) {
4626                 pr_warning("Warning: tx_queues (%d) should be between "
4627                            "1 and 255, resetting to %d\n",
4628                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4629                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4630         }
4631
4632         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4633                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4634                            "not of valid value (0/1), so it was set to "
4635                            "0\n", all_slaves_active);
4636                 all_slaves_active = 0;
4637         }
4638
4639         if (resend_igmp < 0 || resend_igmp > 255) {
4640                 pr_warning("Warning: resend_igmp (%d) should be between "
4641                            "0 and 255, resetting to %d\n",
4642                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4643                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4644         }
4645
4646         /* reset values for TLB/ALB */
4647         if ((bond_mode == BOND_MODE_TLB) ||
4648             (bond_mode == BOND_MODE_ALB)) {
4649                 if (!miimon) {
4650                         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");
4651                         pr_warning("Forcing miimon to 100msec\n");
4652                         miimon = 100;
4653                 }
4654         }
4655
4656         if (bond_mode == BOND_MODE_ALB) {
4657                 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",
4658                           updelay);
4659         }
4660
4661         if (!miimon) {
4662                 if (updelay || downdelay) {
4663                         /* just warn the user the up/down delay will have
4664                          * no effect since miimon is zero...
4665                          */
4666                         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",
4667                                    updelay, downdelay);
4668                 }
4669         } else {
4670                 /* don't allow arp monitoring */
4671                 if (arp_interval) {
4672                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4673                                    miimon, arp_interval);
4674                         arp_interval = 0;
4675                 }
4676
4677                 if ((updelay % miimon) != 0) {
4678                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4679                                    updelay, miimon,
4680                                    (updelay / miimon) * miimon);
4681                 }
4682
4683                 updelay /= miimon;
4684
4685                 if ((downdelay % miimon) != 0) {
4686                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4687                                    downdelay, miimon,
4688                                    (downdelay / miimon) * miimon);
4689                 }
4690
4691                 downdelay /= miimon;
4692         }
4693
4694         if (arp_interval < 0) {
4695                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4696                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4697                 arp_interval = BOND_LINK_ARP_INTERV;
4698         }
4699
4700         for (arp_ip_count = 0;
4701              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4702              arp_ip_count++) {
4703                 /* not complete check, but should be good enough to
4704                    catch mistakes */
4705                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4706                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4707                                    arp_ip_target[arp_ip_count]);
4708                         arp_interval = 0;
4709                 } else {
4710                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4711                         arp_target[arp_ip_count] = ip;
4712                 }
4713         }
4714
4715         if (arp_interval && !arp_ip_count) {
4716                 /* don't allow arping if no arp_ip_target given... */
4717                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4718                            arp_interval);
4719                 arp_interval = 0;
4720         }
4721
4722         if (arp_validate) {
4723                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4724                         pr_err("arp_validate only supported in active-backup mode\n");
4725                         return -EINVAL;
4726                 }
4727                 if (!arp_interval) {
4728                         pr_err("arp_validate requires arp_interval\n");
4729                         return -EINVAL;
4730                 }
4731
4732                 arp_validate_value = bond_parse_parm(arp_validate,
4733                                                      arp_validate_tbl);
4734                 if (arp_validate_value == -1) {
4735                         pr_err("Error: invalid arp_validate \"%s\"\n",
4736                                arp_validate == NULL ? "NULL" : arp_validate);
4737                         return -EINVAL;
4738                 }
4739         } else
4740                 arp_validate_value = 0;
4741
4742         if (miimon) {
4743                 pr_info("MII link monitoring set to %d ms\n", miimon);
4744         } else if (arp_interval) {
4745                 int i;
4746
4747                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4748                         arp_interval,
4749                         arp_validate_tbl[arp_validate_value].modename,
4750                         arp_ip_count);
4751
4752                 for (i = 0; i < arp_ip_count; i++)
4753                         pr_info(" %s", arp_ip_target[i]);
4754
4755                 pr_info("\n");
4756
4757         } else if (max_bonds) {
4758                 /* miimon and arp_interval not set, we need one so things
4759                  * work as expected, see bonding.txt for details
4760                  */
4761                 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");
4762         }
4763
4764         if (primary && !USES_PRIMARY(bond_mode)) {
4765                 /* currently, using a primary only makes sense
4766                  * in active backup, TLB or ALB modes
4767                  */
4768                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4769                            primary, bond_mode_name(bond_mode));
4770                 primary = NULL;
4771         }
4772
4773         if (primary && primary_reselect) {
4774                 primary_reselect_value = bond_parse_parm(primary_reselect,
4775                                                          pri_reselect_tbl);
4776                 if (primary_reselect_value == -1) {
4777                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4778                                primary_reselect ==
4779                                         NULL ? "NULL" : primary_reselect);
4780                         return -EINVAL;
4781                 }
4782         } else {
4783                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4784         }
4785
4786         if (fail_over_mac) {
4787                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4788                                                       fail_over_mac_tbl);
4789                 if (fail_over_mac_value == -1) {
4790                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4791                                arp_validate == NULL ? "NULL" : arp_validate);
4792                         return -EINVAL;
4793                 }
4794
4795                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4796                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4797         } else {
4798                 fail_over_mac_value = BOND_FOM_NONE;
4799         }
4800
4801         /* fill params struct with the proper values */
4802         params->mode = bond_mode;
4803         params->xmit_policy = xmit_hashtype;
4804         params->miimon = miimon;
4805         params->num_peer_notif = num_peer_notif;
4806         params->arp_interval = arp_interval;
4807         params->arp_validate = arp_validate_value;
4808         params->updelay = updelay;
4809         params->downdelay = downdelay;
4810         params->use_carrier = use_carrier;
4811         params->lacp_fast = lacp_fast;
4812         params->primary[0] = 0;
4813         params->primary_reselect = primary_reselect_value;
4814         params->fail_over_mac = fail_over_mac_value;
4815         params->tx_queues = tx_queues;
4816         params->all_slaves_active = all_slaves_active;
4817         params->resend_igmp = resend_igmp;
4818
4819         if (primary) {
4820                 strncpy(params->primary, primary, IFNAMSIZ);
4821                 params->primary[IFNAMSIZ - 1] = 0;
4822         }
4823
4824         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4825
4826         return 0;
4827 }
4828
4829 static struct lock_class_key bonding_netdev_xmit_lock_key;
4830 static struct lock_class_key bonding_netdev_addr_lock_key;
4831
4832 static void bond_set_lockdep_class_one(struct net_device *dev,
4833                                        struct netdev_queue *txq,
4834                                        void *_unused)
4835 {
4836         lockdep_set_class(&txq->_xmit_lock,
4837                           &bonding_netdev_xmit_lock_key);
4838 }
4839
4840 static void bond_set_lockdep_class(struct net_device *dev)
4841 {
4842         lockdep_set_class(&dev->addr_list_lock,
4843                           &bonding_netdev_addr_lock_key);
4844         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4845 }
4846
4847 /*
4848  * Called from registration process
4849  */
4850 static int bond_init(struct net_device *bond_dev)
4851 {
4852         struct bonding *bond = netdev_priv(bond_dev);
4853         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4854         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4855
4856         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4857
4858         /*
4859          * Initialize locks that may be required during
4860          * en/deslave operations.  All of the bond_open work
4861          * (of which this is part) should really be moved to
4862          * a phase prior to dev_open
4863          */
4864         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4865         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4866
4867         bond->wq = create_singlethread_workqueue(bond_dev->name);
4868         if (!bond->wq)
4869                 return -ENOMEM;
4870
4871         bond_set_lockdep_class(bond_dev);
4872
4873         bond_create_proc_entry(bond);
4874         list_add_tail(&bond->bond_list, &bn->dev_list);
4875
4876         bond_prepare_sysfs_group(bond);
4877
4878         bond_debug_register(bond);
4879
4880         __hw_addr_init(&bond->mc_list);
4881         return 0;
4882 }
4883
4884 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4885 {
4886         if (tb[IFLA_ADDRESS]) {
4887                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4888                         return -EINVAL;
4889                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4890                         return -EADDRNOTAVAIL;
4891         }
4892         return 0;
4893 }
4894
4895 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4896         .kind           = "bond",
4897         .priv_size      = sizeof(struct bonding),
4898         .setup          = bond_setup,
4899         .validate       = bond_validate,
4900 };
4901
4902 /* Create a new bond based on the specified name and bonding parameters.
4903  * If name is NULL, obtain a suitable "bond%d" name for us.
4904  * Caller must NOT hold rtnl_lock; we need to release it here before we
4905  * set up our sysfs entries.
4906  */
4907 int bond_create(struct net *net, const char *name)
4908 {
4909         struct net_device *bond_dev;
4910         int res;
4911
4912         rtnl_lock();
4913
4914         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4915                                    name ? name : "bond%d",
4916                                    bond_setup, tx_queues);
4917         if (!bond_dev) {
4918                 pr_err("%s: eek! can't alloc netdev!\n", name);
4919                 rtnl_unlock();
4920                 return -ENOMEM;
4921         }
4922
4923         dev_net_set(bond_dev, net);
4924         bond_dev->rtnl_link_ops = &bond_link_ops;
4925
4926         res = register_netdevice(bond_dev);
4927
4928         netif_carrier_off(bond_dev);
4929
4930         rtnl_unlock();
4931         if (res < 0)
4932                 bond_destructor(bond_dev);
4933         return res;
4934 }
4935
4936 static int __net_init bond_net_init(struct net *net)
4937 {
4938         struct bond_net *bn = net_generic(net, bond_net_id);
4939
4940         bn->net = net;
4941         INIT_LIST_HEAD(&bn->dev_list);
4942
4943         bond_create_proc_dir(bn);
4944
4945         return 0;
4946 }
4947
4948 static void __net_exit bond_net_exit(struct net *net)
4949 {
4950         struct bond_net *bn = net_generic(net, bond_net_id);
4951
4952         bond_destroy_proc_dir(bn);
4953 }
4954
4955 static struct pernet_operations bond_net_ops = {
4956         .init = bond_net_init,
4957         .exit = bond_net_exit,
4958         .id   = &bond_net_id,
4959         .size = sizeof(struct bond_net),
4960 };
4961
4962 static int __init bonding_init(void)
4963 {
4964         int i;
4965         int res;
4966
4967         pr_info("%s", bond_version);
4968
4969         res = bond_check_params(&bonding_defaults);
4970         if (res)
4971                 goto out;
4972
4973         res = register_pernet_subsys(&bond_net_ops);
4974         if (res)
4975                 goto out;
4976
4977         res = rtnl_link_register(&bond_link_ops);
4978         if (res)
4979                 goto err_link;
4980
4981         bond_create_debugfs();
4982
4983         for (i = 0; i < max_bonds; i++) {
4984                 res = bond_create(&init_net, NULL);
4985                 if (res)
4986                         goto err;
4987         }
4988
4989         res = bond_create_sysfs();
4990         if (res)
4991                 goto err;
4992
4993         register_netdevice_notifier(&bond_netdev_notifier);
4994         register_inetaddr_notifier(&bond_inetaddr_notifier);
4995 out:
4996         return res;
4997 err:
4998         rtnl_link_unregister(&bond_link_ops);
4999 err_link:
5000         unregister_pernet_subsys(&bond_net_ops);
5001         goto out;
5002
5003 }
5004
5005 static void __exit bonding_exit(void)
5006 {
5007         unregister_netdevice_notifier(&bond_netdev_notifier);
5008         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5009
5010         bond_destroy_sysfs();
5011         bond_destroy_debugfs();
5012
5013         rtnl_link_unregister(&bond_link_ops);
5014         unregister_pernet_subsys(&bond_net_ops);
5015
5016 #ifdef CONFIG_NET_POLL_CONTROLLER
5017         /*
5018          * Make sure we don't have an imbalance on our netpoll blocking
5019          */
5020         WARN_ON(atomic_read(&netpoll_block_tx));
5021 #endif
5022 }
5023
5024 module_init(bonding_init);
5025 module_exit(bonding_exit);
5026 MODULE_LICENSE("GPL");
5027 MODULE_VERSION(DRV_VERSION);
5028 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5029 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5030 MODULE_ALIAS_RTNL_LINK("bond");