Pileus Git - ~andy/linux/blob - drivers/net/bonding/bond_main.c

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